xref: /kvm-unit-tests/x86/vmx_tests.c (revision dca3f4c041143c8e8dc70c6890a19a5730310230)
1 /*
2  * All test cases of nested virtualization should be in this file
3  *
4  * Author : Arthur Chunqi Li <yzt356@gmail.com>
5  */
6 
7 #include <asm/debugreg.h>
8 
9 #include "vmx.h"
10 #include "msr.h"
11 #include "processor.h"
12 #include "pmu.h"
13 #include "vm.h"
14 #include "pci.h"
15 #include "fwcfg.h"
16 #include "isr.h"
17 #include "desc.h"
18 #include "apic.h"
19 #include "vmalloc.h"
20 #include "alloc_page.h"
21 #include "smp.h"
22 #include "delay.h"
23 #include "access.h"
24 #include "x86/usermode.h"
25 
26 /*
27  * vmcs.GUEST_PENDING_DEBUG has the same format as DR6, although some bits that
28  * are legal in DR6 are reserved in vmcs.GUEST_PENDING_DEBUG.  And if any data
29  * or I/O breakpoint matches *and* was enabled, bit 12 is also set.
30  */
31 #define PENDING_DBG_TRAP	BIT(12)
32 
33 #define VPID_CAP_INVVPID_TYPES_SHIFT 40
34 
35 u64 ia32_pat;
36 u64 ia32_efer;
37 void *io_bitmap_a, *io_bitmap_b;
38 u16 ioport;
39 
40 unsigned long *pml4;
41 u64 eptp;
42 void *data_page1, *data_page2;
43 
44 phys_addr_t pci_physaddr;
45 
46 void *pml_log;
47 #define PML_INDEX 512
48 
ffs(unsigned x)49 static inline unsigned ffs(unsigned x)
50 {
51 	int pos = -1;
52 
53 	__asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0"
54 			     : "+r"(pos) : "rm"(x) : "eax");
55 	return pos + 1;
56 }
57 
vmcall(void)58 static inline void vmcall(void)
59 {
60 	asm volatile("vmcall");
61 }
62 
get_vapic_page(void)63 static u32 *get_vapic_page(void)
64 {
65 	return (u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR));
66 }
67 
get_pi_desc(void)68 static u64 *get_pi_desc(void)
69 {
70 	return (u64 *)phys_to_virt(vmcs_read(POSTED_INTR_DESC_ADDR));
71 }
72 
basic_guest_main(void)73 static void basic_guest_main(void)
74 {
75 	report_pass("Basic VMX test");
76 }
77 
basic_exit_handler(union exit_reason exit_reason)78 static int basic_exit_handler(union exit_reason exit_reason)
79 {
80 	report_fail("Basic VMX test");
81 	print_vmexit_info(exit_reason);
82 	return VMX_TEST_EXIT;
83 }
84 
vmenter_main(void)85 static void vmenter_main(void)
86 {
87 	u64 rax;
88 	u64 rsp, resume_rsp;
89 
90 	report_pass("test vmlaunch");
91 
92 	asm volatile(
93 		"mov %%rsp, %0\n\t"
94 		"mov %3, %%rax\n\t"
95 		"vmcall\n\t"
96 		"mov %%rax, %1\n\t"
97 		"mov %%rsp, %2\n\t"
98 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
99 		: "g"(0xABCD));
100 	report((rax == 0xFFFF) && (rsp == resume_rsp), "test vmresume");
101 }
102 
vmenter_exit_handler(union exit_reason exit_reason)103 static int vmenter_exit_handler(union exit_reason exit_reason)
104 {
105 	u64 guest_rip = vmcs_read(GUEST_RIP);
106 
107 	switch (exit_reason.basic) {
108 	case VMX_VMCALL:
109 		if (regs.rax != 0xABCD) {
110 			report_fail("test vmresume");
111 			return VMX_TEST_VMEXIT;
112 		}
113 		regs.rax = 0xFFFF;
114 		vmcs_write(GUEST_RIP, guest_rip + 3);
115 		return VMX_TEST_RESUME;
116 	default:
117 		report_fail("test vmresume");
118 		print_vmexit_info(exit_reason);
119 	}
120 	return VMX_TEST_VMEXIT;
121 }
122 
123 u32 preempt_scale;
124 volatile unsigned long long tsc_val;
125 volatile u32 preempt_val;
126 u64 saved_rip;
127 
preemption_timer_init(struct vmcs * vmcs)128 static int preemption_timer_init(struct vmcs *vmcs)
129 {
130 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
131 		printf("\tPreemption timer is not supported\n");
132 		return VMX_TEST_EXIT;
133 	}
134 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
135 	preempt_val = 10000000;
136 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
137 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
138 
139 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
140 		printf("\tSave preemption value is not supported\n");
141 
142 	return VMX_TEST_START;
143 }
144 
preemption_timer_main(void)145 static void preemption_timer_main(void)
146 {
147 	tsc_val = rdtsc();
148 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
149 		vmx_set_test_stage(0);
150 		vmcall();
151 		if (vmx_get_test_stage() == 1)
152 			vmcall();
153 	}
154 	vmx_set_test_stage(1);
155 	while (vmx_get_test_stage() == 1) {
156 		if (((rdtsc() - tsc_val) >> preempt_scale)
157 				> 10 * preempt_val) {
158 			vmx_set_test_stage(2);
159 			vmcall();
160 		}
161 	}
162 	tsc_val = rdtsc();
163 	asm volatile ("hlt");
164 	vmcall();
165 	vmx_set_test_stage(5);
166 	vmcall();
167 }
168 
preemption_timer_exit_handler(union exit_reason exit_reason)169 static int preemption_timer_exit_handler(union exit_reason exit_reason)
170 {
171 	bool guest_halted;
172 	u64 guest_rip;
173 	u32 insn_len;
174 	u32 ctrl_exit;
175 
176 	guest_rip = vmcs_read(GUEST_RIP);
177 	insn_len = vmcs_read(EXI_INST_LEN);
178 	switch (exit_reason.basic) {
179 	case VMX_PREEMPT:
180 		switch (vmx_get_test_stage()) {
181 		case 1:
182 		case 2:
183 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val,
184 			       "busy-wait for preemption timer");
185 			vmx_set_test_stage(3);
186 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
187 			return VMX_TEST_RESUME;
188 		case 3:
189 			guest_halted =
190 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
191 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val
192 			        && guest_halted,
193 			       "preemption timer during hlt");
194 			vmx_set_test_stage(4);
195 			vmcs_write(PIN_CONTROLS,
196 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
197 			vmcs_write(EXI_CONTROLS,
198 				   vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT);
199 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
200 			return VMX_TEST_RESUME;
201 		case 4:
202 			report(saved_rip == guest_rip,
203 			       "preemption timer with 0 value");
204 			break;
205 		default:
206 			report_fail("Invalid stage.");
207 			print_vmexit_info(exit_reason);
208 			break;
209 		}
210 		break;
211 	case VMX_VMCALL:
212 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
213 		switch (vmx_get_test_stage()) {
214 		case 0:
215 			report(vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val,
216 			       "Keep preemption value");
217 			vmx_set_test_stage(1);
218 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
219 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
220 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
221 			vmcs_write(EXI_CONTROLS, ctrl_exit);
222 			return VMX_TEST_RESUME;
223 		case 1:
224 			report(vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val,
225 			       "Save preemption value");
226 			return VMX_TEST_RESUME;
227 		case 2:
228 			report_fail("busy-wait for preemption timer");
229 			vmx_set_test_stage(3);
230 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
231 			return VMX_TEST_RESUME;
232 		case 3:
233 			report_fail("preemption timer during hlt");
234 			vmx_set_test_stage(4);
235 			/* fall through */
236 		case 4:
237 			vmcs_write(PIN_CONTROLS,
238 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
239 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
240 			saved_rip = guest_rip + insn_len;
241 			return VMX_TEST_RESUME;
242 		case 5:
243 			report_fail("preemption timer with 0 value (vmcall stage 5)");
244 			break;
245 		default:
246 			// Should not reach here
247 			report_fail("unexpected stage, %d",
248 				    vmx_get_test_stage());
249 			print_vmexit_info(exit_reason);
250 			return VMX_TEST_VMEXIT;
251 		}
252 		break;
253 	default:
254 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
255 		print_vmexit_info(exit_reason);
256 	}
257 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
258 	return VMX_TEST_VMEXIT;
259 }
260 
msr_bmp_init(void)261 static void msr_bmp_init(void)
262 {
263 	void *msr_bitmap;
264 	u32 ctrl_cpu0;
265 
266 	msr_bitmap = alloc_page();
267 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
268 	ctrl_cpu0 |= CPU_MSR_BITMAP;
269 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
270 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
271 }
272 
get_msr_bitmap(void)273 static void *get_msr_bitmap(void)
274 {
275 	void *msr_bitmap;
276 
277 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
278 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
279 	} else {
280 		msr_bitmap = alloc_page();
281 		memset(msr_bitmap, 0xff, PAGE_SIZE);
282 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
283 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
284 	}
285 
286 	return msr_bitmap;
287 }
288 
disable_intercept_for_x2apic_msrs(void)289 static void disable_intercept_for_x2apic_msrs(void)
290 {
291 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
292 	u32 msr;
293 
294 	for (msr = APIC_BASE_MSR;
295 		 msr < (APIC_BASE_MSR+0xff);
296 		 msr += BITS_PER_LONG) {
297 		unsigned int word = msr / BITS_PER_LONG;
298 
299 		msr_bitmap[word] = 0;
300 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
301 	}
302 }
303 
test_ctrl_pat_init(struct vmcs * vmcs)304 static int test_ctrl_pat_init(struct vmcs *vmcs)
305 {
306 	u64 ctrl_ent;
307 	u64 ctrl_exi;
308 
309 	msr_bmp_init();
310 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
311 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
312 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
313 		printf("\tSave/load PAT is not supported\n");
314 		return 1;
315 	}
316 
317 	ctrl_ent = vmcs_read(ENT_CONTROLS);
318 	ctrl_exi = vmcs_read(EXI_CONTROLS);
319 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
320 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
321 	vmcs_write(ENT_CONTROLS, ctrl_ent);
322 	vmcs_write(EXI_CONTROLS, ctrl_exi);
323 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
324 	vmcs_write(GUEST_PAT, 0x0);
325 	vmcs_write(HOST_PAT, ia32_pat);
326 	return VMX_TEST_START;
327 }
328 
test_ctrl_pat_main(void)329 static void test_ctrl_pat_main(void)
330 {
331 	u64 guest_ia32_pat;
332 
333 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
334 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
335 		printf("\tENT_LOAD_PAT is not supported.\n");
336 	else {
337 		if (guest_ia32_pat != 0) {
338 			report_fail("Entry load PAT");
339 			return;
340 		}
341 	}
342 	wrmsr(MSR_IA32_CR_PAT, 0x6);
343 	vmcall();
344 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
345 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
346 		report(guest_ia32_pat == ia32_pat, "Entry load PAT");
347 }
348 
test_ctrl_pat_exit_handler(union exit_reason exit_reason)349 static int test_ctrl_pat_exit_handler(union exit_reason exit_reason)
350 {
351 	u64 guest_rip;
352 	u64 guest_pat;
353 
354 	guest_rip = vmcs_read(GUEST_RIP);
355 	switch (exit_reason.basic) {
356 	case VMX_VMCALL:
357 		guest_pat = vmcs_read(GUEST_PAT);
358 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
359 			printf("\tEXI_SAVE_PAT is not supported\n");
360 			vmcs_write(GUEST_PAT, 0x6);
361 		} else {
362 			report(guest_pat == 0x6, "Exit save PAT");
363 		}
364 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
365 			printf("\tEXI_LOAD_PAT is not supported\n");
366 		else
367 			report(rdmsr(MSR_IA32_CR_PAT) == ia32_pat,
368 			       "Exit load PAT");
369 		vmcs_write(GUEST_PAT, ia32_pat);
370 		vmcs_write(GUEST_RIP, guest_rip + 3);
371 		return VMX_TEST_RESUME;
372 	default:
373 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
374 		break;
375 	}
376 	return VMX_TEST_VMEXIT;
377 }
378 
test_ctrl_efer_init(struct vmcs * vmcs)379 static int test_ctrl_efer_init(struct vmcs *vmcs)
380 {
381 	u64 ctrl_ent;
382 	u64 ctrl_exi;
383 
384 	msr_bmp_init();
385 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
386 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
387 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
388 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
389 	ia32_efer = rdmsr(MSR_EFER);
390 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
391 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
392 	return VMX_TEST_START;
393 }
394 
test_ctrl_efer_main(void)395 static void test_ctrl_efer_main(void)
396 {
397 	u64 guest_ia32_efer;
398 
399 	guest_ia32_efer = rdmsr(MSR_EFER);
400 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
401 		printf("\tENT_LOAD_EFER is not supported.\n");
402 	else {
403 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
404 			report_fail("Entry load EFER");
405 			return;
406 		}
407 	}
408 	wrmsr(MSR_EFER, ia32_efer);
409 	vmcall();
410 	guest_ia32_efer = rdmsr(MSR_EFER);
411 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
412 		report(guest_ia32_efer == ia32_efer, "Entry load EFER");
413 }
414 
test_ctrl_efer_exit_handler(union exit_reason exit_reason)415 static int test_ctrl_efer_exit_handler(union exit_reason exit_reason)
416 {
417 	u64 guest_rip;
418 	u64 guest_efer;
419 
420 	guest_rip = vmcs_read(GUEST_RIP);
421 	switch (exit_reason.basic) {
422 	case VMX_VMCALL:
423 		guest_efer = vmcs_read(GUEST_EFER);
424 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
425 			printf("\tEXI_SAVE_EFER is not supported\n");
426 			vmcs_write(GUEST_EFER, ia32_efer);
427 		} else {
428 			report(guest_efer == ia32_efer, "Exit save EFER");
429 		}
430 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
431 			printf("\tEXI_LOAD_EFER is not supported\n");
432 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
433 		} else {
434 			report(rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX),
435 			       "Exit load EFER");
436 		}
437 		vmcs_write(GUEST_PAT, ia32_efer);
438 		vmcs_write(GUEST_RIP, guest_rip + 3);
439 		return VMX_TEST_RESUME;
440 	default:
441 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
442 		break;
443 	}
444 	return VMX_TEST_VMEXIT;
445 }
446 
447 u32 guest_cr0, guest_cr4;
448 
cr_shadowing_main(void)449 static void cr_shadowing_main(void)
450 {
451 	u32 cr0, cr4, tmp;
452 
453 	// Test read through
454 	vmx_set_test_stage(0);
455 	guest_cr0 = read_cr0();
456 	if (vmx_get_test_stage() == 1)
457 		report_fail("Read through CR0");
458 	else
459 		vmcall();
460 	vmx_set_test_stage(1);
461 	guest_cr4 = read_cr4();
462 	if (vmx_get_test_stage() == 2)
463 		report_fail("Read through CR4");
464 	else
465 		vmcall();
466 	// Test write through
467 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
468 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
469 	vmx_set_test_stage(2);
470 	write_cr0(guest_cr0);
471 	if (vmx_get_test_stage() == 3)
472 		report_fail("Write through CR0");
473 	else
474 		vmcall();
475 	vmx_set_test_stage(3);
476 	write_cr4(guest_cr4);
477 	if (vmx_get_test_stage() == 4)
478 		report_fail("Write through CR4");
479 	else
480 		vmcall();
481 	// Test read shadow
482 	vmx_set_test_stage(4);
483 	vmcall();
484 	cr0 = read_cr0();
485 	if (vmx_get_test_stage() != 5)
486 		report(cr0 == guest_cr0, "Read shadowing CR0");
487 	vmx_set_test_stage(5);
488 	cr4 = read_cr4();
489 	if (vmx_get_test_stage() != 6)
490 		report(cr4 == guest_cr4, "Read shadowing CR4");
491 	// Test write shadow (same value with shadow)
492 	vmx_set_test_stage(6);
493 	write_cr0(guest_cr0);
494 	if (vmx_get_test_stage() == 7)
495 		report_fail("Write shadowing CR0 (same value with shadow)");
496 	else
497 		vmcall();
498 	vmx_set_test_stage(7);
499 	write_cr4(guest_cr4);
500 	if (vmx_get_test_stage() == 8)
501 		report_fail("Write shadowing CR4 (same value with shadow)");
502 	else
503 		vmcall();
504 	// Test write shadow (different value)
505 	vmx_set_test_stage(8);
506 	tmp = guest_cr0 ^ X86_CR0_TS;
507 	asm volatile("mov %0, %%rsi\n\t"
508 		"mov %%rsi, %%cr0\n\t"
509 		::"m"(tmp)
510 		:"rsi", "memory", "cc");
511 	report(vmx_get_test_stage() == 9,
512 	       "Write shadowing different X86_CR0_TS");
513 	vmx_set_test_stage(9);
514 	tmp = guest_cr0 ^ X86_CR0_MP;
515 	asm volatile("mov %0, %%rsi\n\t"
516 		"mov %%rsi, %%cr0\n\t"
517 		::"m"(tmp)
518 		:"rsi", "memory", "cc");
519 	report(vmx_get_test_stage() == 10,
520 	       "Write shadowing different X86_CR0_MP");
521 	vmx_set_test_stage(10);
522 	tmp = guest_cr4 ^ X86_CR4_TSD;
523 	asm volatile("mov %0, %%rsi\n\t"
524 		"mov %%rsi, %%cr4\n\t"
525 		::"m"(tmp)
526 		:"rsi", "memory", "cc");
527 	report(vmx_get_test_stage() == 11,
528 	       "Write shadowing different X86_CR4_TSD");
529 	vmx_set_test_stage(11);
530 	tmp = guest_cr4 ^ X86_CR4_DE;
531 	asm volatile("mov %0, %%rsi\n\t"
532 		"mov %%rsi, %%cr4\n\t"
533 		::"m"(tmp)
534 		:"rsi", "memory", "cc");
535 	report(vmx_get_test_stage() == 12,
536 	       "Write shadowing different X86_CR4_DE");
537 }
538 
cr_shadowing_exit_handler(union exit_reason exit_reason)539 static int cr_shadowing_exit_handler(union exit_reason exit_reason)
540 {
541 	u64 guest_rip;
542 	u32 insn_len;
543 	u32 exit_qual;
544 
545 	guest_rip = vmcs_read(GUEST_RIP);
546 	insn_len = vmcs_read(EXI_INST_LEN);
547 	exit_qual = vmcs_read(EXI_QUALIFICATION);
548 	switch (exit_reason.basic) {
549 	case VMX_VMCALL:
550 		switch (vmx_get_test_stage()) {
551 		case 0:
552 			report(guest_cr0 == vmcs_read(GUEST_CR0),
553 			       "Read through CR0");
554 			break;
555 		case 1:
556 			report(guest_cr4 == vmcs_read(GUEST_CR4),
557 			       "Read through CR4");
558 			break;
559 		case 2:
560 			report(guest_cr0 == vmcs_read(GUEST_CR0),
561 			       "Write through CR0");
562 			break;
563 		case 3:
564 			report(guest_cr4 == vmcs_read(GUEST_CR4),
565 			       "Write through CR4");
566 			break;
567 		case 4:
568 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
569 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
570 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
571 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
572 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
573 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
574 			break;
575 		case 6:
576 			report(guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)),
577 			       "Write shadowing CR0 (same value)");
578 			break;
579 		case 7:
580 			report(guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)),
581 			       "Write shadowing CR4 (same value)");
582 			break;
583 		default:
584 			// Should not reach here
585 			report_fail("unexpected stage, %d",
586 				    vmx_get_test_stage());
587 			print_vmexit_info(exit_reason);
588 			return VMX_TEST_VMEXIT;
589 		}
590 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
591 		return VMX_TEST_RESUME;
592 	case VMX_CR:
593 		switch (vmx_get_test_stage()) {
594 		case 4:
595 			report_fail("Read shadowing CR0");
596 			vmx_inc_test_stage();
597 			break;
598 		case 5:
599 			report_fail("Read shadowing CR4");
600 			vmx_inc_test_stage();
601 			break;
602 		case 6:
603 			report_fail("Write shadowing CR0 (same value)");
604 			vmx_inc_test_stage();
605 			break;
606 		case 7:
607 			report_fail("Write shadowing CR4 (same value)");
608 			vmx_inc_test_stage();
609 			break;
610 		case 8:
611 		case 9:
612 			// 0x600 encodes "mov %esi, %cr0"
613 			if (exit_qual == 0x600)
614 				vmx_inc_test_stage();
615 			break;
616 		case 10:
617 		case 11:
618 			// 0x604 encodes "mov %esi, %cr4"
619 			if (exit_qual == 0x604)
620 				vmx_inc_test_stage();
621 			break;
622 		default:
623 			// Should not reach here
624 			report_fail("unexpected stage, %d",
625 				    vmx_get_test_stage());
626 			print_vmexit_info(exit_reason);
627 			return VMX_TEST_VMEXIT;
628 		}
629 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
630 		return VMX_TEST_RESUME;
631 	default:
632 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
633 		print_vmexit_info(exit_reason);
634 	}
635 	return VMX_TEST_VMEXIT;
636 }
637 
iobmp_init(struct vmcs * vmcs)638 static int iobmp_init(struct vmcs *vmcs)
639 {
640 	u32 ctrl_cpu0;
641 
642 	io_bitmap_a = alloc_page();
643 	io_bitmap_b = alloc_page();
644 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
645 	ctrl_cpu0 |= CPU_IO_BITMAP;
646 	ctrl_cpu0 &= (~CPU_IO);
647 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
648 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
649 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
650 	return VMX_TEST_START;
651 }
652 
iobmp_main(void)653 static void iobmp_main(void)
654 {
655 	// stage 0, test IO pass
656 	vmx_set_test_stage(0);
657 	inb(0x5000);
658 	outb(0x0, 0x5000);
659 	report(vmx_get_test_stage() == 0, "I/O bitmap - I/O pass");
660 	// test IO width, in/out
661 	((u8 *)io_bitmap_a)[0] = 0xFF;
662 	vmx_set_test_stage(2);
663 	inb(0x0);
664 	report(vmx_get_test_stage() == 3, "I/O bitmap - trap in");
665 	vmx_set_test_stage(3);
666 	outw(0x0, 0x0);
667 	report(vmx_get_test_stage() == 4, "I/O bitmap - trap out");
668 	vmx_set_test_stage(4);
669 	inl(0x0);
670 	report(vmx_get_test_stage() == 5, "I/O bitmap - I/O width, long");
671 	// test low/high IO port
672 	vmx_set_test_stage(5);
673 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
674 	inb(0x5000);
675 	report(vmx_get_test_stage() == 6, "I/O bitmap - I/O port, low part");
676 	vmx_set_test_stage(6);
677 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
678 	inb(0x9000);
679 	report(vmx_get_test_stage() == 7, "I/O bitmap - I/O port, high part");
680 	// test partial pass
681 	vmx_set_test_stage(7);
682 	inl(0x4FFF);
683 	report(vmx_get_test_stage() == 8, "I/O bitmap - partial pass");
684 	// test overrun
685 	vmx_set_test_stage(8);
686 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
687 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
688 	inl(0xFFFF);
689 	report(vmx_get_test_stage() == 9, "I/O bitmap - overrun");
690 	vmx_set_test_stage(9);
691 	vmcall();
692 	outb(0x0, 0x0);
693 	report(vmx_get_test_stage() == 9,
694 	       "I/O bitmap - ignore unconditional exiting");
695 	vmx_set_test_stage(10);
696 	vmcall();
697 	outb(0x0, 0x0);
698 	report(vmx_get_test_stage() == 11,
699 	       "I/O bitmap - unconditional exiting");
700 }
701 
iobmp_exit_handler(union exit_reason exit_reason)702 static int iobmp_exit_handler(union exit_reason exit_reason)
703 {
704 	u64 guest_rip;
705 	ulong exit_qual;
706 	u32 insn_len, ctrl_cpu0;
707 
708 	guest_rip = vmcs_read(GUEST_RIP);
709 	exit_qual = vmcs_read(EXI_QUALIFICATION);
710 	insn_len = vmcs_read(EXI_INST_LEN);
711 	switch (exit_reason.basic) {
712 	case VMX_IO:
713 		switch (vmx_get_test_stage()) {
714 		case 0:
715 		case 1:
716 			vmx_inc_test_stage();
717 			break;
718 		case 2:
719 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE,
720 			       "I/O bitmap - I/O width, byte");
721 			report(exit_qual & VMX_IO_IN,
722 			       "I/O bitmap - I/O direction, in");
723 			vmx_inc_test_stage();
724 			break;
725 		case 3:
726 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD,
727 			       "I/O bitmap - I/O width, word");
728 			report(!(exit_qual & VMX_IO_IN),
729 			       "I/O bitmap - I/O direction, out");
730 			vmx_inc_test_stage();
731 			break;
732 		case 4:
733 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG,
734 			       "I/O bitmap - I/O width, long");
735 			vmx_inc_test_stage();
736 			break;
737 		case 5:
738 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
739 				vmx_inc_test_stage();
740 			break;
741 		case 6:
742 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
743 				vmx_inc_test_stage();
744 			break;
745 		case 7:
746 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
747 				vmx_inc_test_stage();
748 			break;
749 		case 8:
750 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
751 				vmx_inc_test_stage();
752 			break;
753 		case 9:
754 		case 10:
755 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
756 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
757 			vmx_inc_test_stage();
758 			break;
759 		default:
760 			// Should not reach here
761 			report_fail("unexpected stage, %d",
762 				    vmx_get_test_stage());
763 			print_vmexit_info(exit_reason);
764 			return VMX_TEST_VMEXIT;
765 		}
766 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
767 		return VMX_TEST_RESUME;
768 	case VMX_VMCALL:
769 		switch (vmx_get_test_stage()) {
770 		case 9:
771 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
772 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
773 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
774 			break;
775 		case 10:
776 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
777 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
778 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
779 			break;
780 		default:
781 			// Should not reach here
782 			report_fail("unexpected stage, %d",
783 				    vmx_get_test_stage());
784 			print_vmexit_info(exit_reason);
785 			return VMX_TEST_VMEXIT;
786 		}
787 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
788 		return VMX_TEST_RESUME;
789 	default:
790 		printf("guest_rip = %#lx\n", guest_rip);
791 		printf("\tERROR : Unknown exit reason, 0x%x\n", exit_reason.full);
792 		break;
793 	}
794 	return VMX_TEST_VMEXIT;
795 }
796 
797 #define INSN_CPU0		0
798 #define INSN_CPU1		1
799 #define INSN_ALWAYS_TRAP	2
800 
801 #define FIELD_EXIT_QUAL		(1 << 0)
802 #define FIELD_INSN_INFO		(1 << 1)
803 
804 asm(
805 	"insn_hlt: hlt;ret\n\t"
806 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
807 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
808 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
809 	"insn_rdtsc: rdtsc;ret\n\t"
810 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
811 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
812 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
813 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
814 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
815 	"insn_pause: pause;ret\n\t"
816 	"insn_wbinvd: wbinvd;ret\n\t"
817 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
818 	"insn_invd: invd;ret\n\t"
819 	"insn_sgdt: sgdt gdt_descr;ret\n\t"
820 	"insn_lgdt: lgdt gdt_descr;ret\n\t"
821 	"insn_sidt: sidt idt_descr;ret\n\t"
822 	"insn_lidt: lidt idt_descr;ret\n\t"
823 	"insn_sldt: sldt %ax;ret\n\t"
824 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
825 	"insn_str: str %ax;ret\n\t"
826 	"insn_rdrand: rdrand %rax;ret\n\t"
827 	"insn_rdseed: rdseed %rax;ret\n\t"
828 );
829 extern void insn_hlt(void);
830 extern void insn_invlpg(void);
831 extern void insn_mwait(void);
832 extern void insn_rdpmc(void);
833 extern void insn_rdtsc(void);
834 extern void insn_cr3_load(void);
835 extern void insn_cr3_store(void);
836 extern void insn_cr8_load(void);
837 extern void insn_cr8_store(void);
838 extern void insn_monitor(void);
839 extern void insn_pause(void);
840 extern void insn_wbinvd(void);
841 extern void insn_sgdt(void);
842 extern void insn_lgdt(void);
843 extern void insn_sidt(void);
844 extern void insn_lidt(void);
845 extern void insn_sldt(void);
846 extern void insn_lldt(void);
847 extern void insn_str(void);
848 extern void insn_cpuid(void);
849 extern void insn_invd(void);
850 extern void insn_rdrand(void);
851 extern void insn_rdseed(void);
852 
853 u32 cur_insn;
854 u64 cr3;
855 
856 typedef bool (*supported_fn)(void);
857 
this_cpu_has_mwait(void)858 static bool this_cpu_has_mwait(void)
859 {
860 	return this_cpu_has(X86_FEATURE_MWAIT);
861 }
862 
863 struct insn_table {
864 	const char *name;
865 	u32 flag;
866 	void (*insn_func)(void);
867 	u32 type;
868 	u32 reason;
869 	ulong exit_qual;
870 	u32 insn_info;
871 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
872 	// which field need to be tested, reason is always tested
873 	u32 test_field;
874 	const supported_fn supported_fn;
875 	u8 disabled;
876 };
877 
878 /*
879  * Add more test cases of instruction intercept here. Elements in this
880  * table is:
881  *	name/control flag/insn function/type/exit reason/exit qulification/
882  *	instruction info/field to test
883  * The last field defines which fields (exit_qual and insn_info) need to be
884  * tested in exit handler. If set to 0, only "reason" is checked.
885  */
886 static struct insn_table insn_table[] = {
887 	// Flags for Primary Processor-Based VM-Execution Controls
888 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
889 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
890 		0x12345678, 0, FIELD_EXIT_QUAL},
891 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0, this_cpu_has_mwait},
892 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0, this_cpu_has_pmu},
893 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
894 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
895 		FIELD_EXIT_QUAL},
896 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
897 		FIELD_EXIT_QUAL},
898 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
899 		FIELD_EXIT_QUAL},
900 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
901 		FIELD_EXIT_QUAL},
902 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0, this_cpu_has_mwait},
903 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
904 	// Flags for Secondary Processor-Based VM-Execution Controls
905 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
906 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
907 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
908 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
909 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
910 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
911 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
912 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
913 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
914 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
915 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
916 	// Instructions always trap
917 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
918 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
919 	// Instructions never trap
920 	{NULL},
921 };
922 
insn_intercept_init(struct vmcs * vmcs)923 static int insn_intercept_init(struct vmcs *vmcs)
924 {
925 	u32 ctrl_cpu, cur_insn;
926 
927 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
928 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
929 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
930 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
931 	cr3 = read_cr3();
932 
933 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
934 		if (insn_table[cur_insn].supported_fn == NULL)
935 			continue;
936 		insn_table[cur_insn].disabled = !insn_table[cur_insn].supported_fn();
937 	}
938 	return VMX_TEST_START;
939 }
940 
insn_intercept_main(void)941 static void insn_intercept_main(void)
942 {
943 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
944 		vmx_set_test_stage(cur_insn * 2);
945 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
946 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
947 		    (insn_table[cur_insn].type == INSN_CPU1 &&
948 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
949 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
950 			       insn_table[cur_insn].type - INSN_CPU0,
951 			       insn_table[cur_insn].name);
952 			continue;
953 		}
954 
955 		if (insn_table[cur_insn].disabled) {
956 			printf("\tFeature required for %s is not supported.\n",
957 			       insn_table[cur_insn].name);
958 			continue;
959 		}
960 
961 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
962 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
963 		    (insn_table[cur_insn].type == INSN_CPU1 &&
964 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
965 			/* skip hlt, it stalls the guest and is tested below */
966 			if (insn_table[cur_insn].insn_func != insn_hlt)
967 				insn_table[cur_insn].insn_func();
968 			report(vmx_get_test_stage() == cur_insn * 2,
969 					"execute %s",
970 					insn_table[cur_insn].name);
971 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
972 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
973 			       insn_table[cur_insn].type - INSN_CPU0,
974 			       insn_table[cur_insn].name);
975 
976 		vmcall();
977 
978 		insn_table[cur_insn].insn_func();
979 		report(vmx_get_test_stage() == cur_insn * 2 + 1,
980 				"intercept %s",
981 				insn_table[cur_insn].name);
982 
983 		vmx_set_test_stage(cur_insn * 2 + 1);
984 		vmcall();
985 	}
986 }
987 
insn_intercept_exit_handler(union exit_reason exit_reason)988 static int insn_intercept_exit_handler(union exit_reason exit_reason)
989 {
990 	u64 guest_rip;
991 	ulong exit_qual;
992 	u32 insn_len;
993 	u32 insn_info;
994 	bool pass;
995 
996 	guest_rip = vmcs_read(GUEST_RIP);
997 	exit_qual = vmcs_read(EXI_QUALIFICATION);
998 	insn_len = vmcs_read(EXI_INST_LEN);
999 	insn_info = vmcs_read(EXI_INST_INFO);
1000 
1001 	if (exit_reason.basic == VMX_VMCALL) {
1002 		u32 val = 0;
1003 
1004 		if (insn_table[cur_insn].type == INSN_CPU0)
1005 			val = vmcs_read(CPU_EXEC_CTRL0);
1006 		else if (insn_table[cur_insn].type == INSN_CPU1)
1007 			val = vmcs_read(CPU_EXEC_CTRL1);
1008 
1009 		if (vmx_get_test_stage() & 1)
1010 			val &= ~insn_table[cur_insn].flag;
1011 		else
1012 			val |= insn_table[cur_insn].flag;
1013 
1014 		if (insn_table[cur_insn].type == INSN_CPU0)
1015 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
1016 		else if (insn_table[cur_insn].type == INSN_CPU1)
1017 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
1018 	} else {
1019 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
1020 			insn_table[cur_insn].reason == exit_reason.full;
1021 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
1022 		    insn_table[cur_insn].exit_qual != exit_qual)
1023 			pass = false;
1024 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
1025 		    insn_table[cur_insn].insn_info != insn_info)
1026 			pass = false;
1027 		if (pass)
1028 			vmx_inc_test_stage();
1029 	}
1030 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1031 	return VMX_TEST_RESUME;
1032 }
1033 
1034 /**
1035  * __setup_ept - Setup the VMCS fields to enable Extended Page Tables (EPT)
1036  * @hpa:	Host physical address of the top-level, a.k.a. root, EPT table
1037  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1038  *
1039  * Returns 0 on success, 1 on failure.
1040  *
1041  * Note that @hpa doesn't need to point at actual memory if VM-Launch is
1042  * expected to fail, e.g. setup_dummy_ept() arbitrarily passes '0' to satisfy
1043  * the various EPTP consistency checks, but doesn't ensure backing for HPA '0'.
1044  */
__setup_ept(u64 hpa,bool enable_ad)1045 static int __setup_ept(u64 hpa, bool enable_ad)
1046 {
1047 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1048 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1049 		printf("\tEPT is not supported\n");
1050 		return 1;
1051 	}
1052 	if (!is_ept_memtype_supported(EPT_MEM_TYPE_WB)) {
1053 		printf("\tWB memtype for EPT walks not supported\n");
1054 		return 1;
1055 	}
1056 
1057 	if (!is_4_level_ept_supported()) {
1058 		/* Support for 4-level EPT is mandatory. */
1059 		report(false, "4-level EPT support check");
1060 		printf("\tPWL4 is not supported\n");
1061 		return 1;
1062 	}
1063 
1064 	eptp = EPT_MEM_TYPE_WB;
1065 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1066 	eptp |= hpa;
1067 	if (enable_ad)
1068 		eptp |= EPTP_AD_FLAG;
1069 
1070 	vmcs_write(EPTP, eptp);
1071 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY);
1072 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT);
1073 
1074 	return 0;
1075 }
1076 
1077 /**
1078  * setup_ept - Enable Extended Page Tables (EPT) and setup an identity map
1079  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1080  *
1081  * Returns 0 on success, 1 on failure.
1082  *
1083  * This is the "real" function for setting up EPT tables, i.e. use this for
1084  * tests that need to run code in the guest with EPT enabled.
1085  */
setup_ept(bool enable_ad)1086 static int setup_ept(bool enable_ad)
1087 {
1088 	unsigned long end_of_memory;
1089 
1090 	pml4 = alloc_page();
1091 
1092 	if (__setup_ept(virt_to_phys(pml4), enable_ad))
1093 		return 1;
1094 
1095 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1096 	if (end_of_memory < (1ul << 32))
1097 		end_of_memory = (1ul << 32);
1098 	/* Cannot use large EPT pages if we need to track EPT
1099 	 * accessed/dirty bits at 4K granularity.
1100 	 */
1101 	setup_ept_range(pml4, 0, end_of_memory, 0,
1102 			!enable_ad && ept_2m_supported(),
1103 			EPT_WA | EPT_RA | EPT_EA);
1104 	return 0;
1105 }
1106 
1107 /**
1108  * setup_dummy_ept - Enable Extended Page Tables (EPT) with a dummy root HPA
1109  *
1110  * Setup EPT using a semi-arbitrary dummy root HPA.  This function is intended
1111  * for use by tests that need EPT enabled to verify dependent VMCS controls
1112  * but never expect to fully enter the guest, i.e. don't need setup the actual
1113  * EPT tables.
1114  */
setup_dummy_ept(void)1115 static void setup_dummy_ept(void)
1116 {
1117 	if (__setup_ept(0, false))
1118 		report_abort("EPT setup unexpectedly failed");
1119 }
1120 
enable_unrestricted_guest(bool need_valid_ept)1121 static int enable_unrestricted_guest(bool need_valid_ept)
1122 {
1123 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1124 	    !(ctrl_cpu_rev[1].clr & CPU_URG) ||
1125 	    !(ctrl_cpu_rev[1].clr & CPU_EPT))
1126 		return 1;
1127 
1128 	if (need_valid_ept)
1129 		setup_ept(false);
1130 	else
1131 		setup_dummy_ept();
1132 
1133 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
1134 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
1135 
1136 	return 0;
1137 }
1138 
ept_enable_ad_bits(void)1139 static void ept_enable_ad_bits(void)
1140 {
1141 	eptp |= EPTP_AD_FLAG;
1142 	vmcs_write(EPTP, eptp);
1143 }
1144 
ept_disable_ad_bits(void)1145 static void ept_disable_ad_bits(void)
1146 {
1147 	eptp &= ~EPTP_AD_FLAG;
1148 	vmcs_write(EPTP, eptp);
1149 }
1150 
ept_ad_enabled(void)1151 static int ept_ad_enabled(void)
1152 {
1153 	return eptp & EPTP_AD_FLAG;
1154 }
1155 
ept_enable_ad_bits_or_skip_test(void)1156 static void ept_enable_ad_bits_or_skip_test(void)
1157 {
1158 	if (!ept_ad_bits_supported())
1159 		test_skip("EPT AD bits not supported.");
1160 	ept_enable_ad_bits();
1161 }
1162 
1163 static int apic_version;
1164 
ept_init_common(bool have_ad)1165 static int ept_init_common(bool have_ad)
1166 {
1167 	int ret;
1168 	struct pci_dev pcidev;
1169 
1170 	/* INVEPT is required by the EPT violation handler. */
1171 	if (!is_invept_type_supported(INVEPT_SINGLE))
1172 		return VMX_TEST_EXIT;
1173 
1174 	if (setup_ept(have_ad))
1175 		return VMX_TEST_EXIT;
1176 
1177 	data_page1 = alloc_page();
1178 	data_page2 = alloc_page();
1179 	*((u32 *)data_page1) = MAGIC_VAL_1;
1180 	*((u32 *)data_page2) = MAGIC_VAL_2;
1181 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1182 			EPT_RA | EPT_WA | EPT_EA);
1183 
1184 	apic_version = apic_read(APIC_LVR);
1185 
1186 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1187 	if (ret != PCIDEVADDR_INVALID) {
1188 		pci_dev_init(&pcidev, ret);
1189 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1190 	}
1191 
1192 	return VMX_TEST_START;
1193 }
1194 
ept_init(struct vmcs * vmcs)1195 static int ept_init(struct vmcs *vmcs)
1196 {
1197 	return ept_init_common(false);
1198 }
1199 
ept_common(void)1200 static void ept_common(void)
1201 {
1202 	vmx_set_test_stage(0);
1203 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1204 			*((u32 *)data_page1) != MAGIC_VAL_1)
1205 		report_fail("EPT basic framework - read");
1206 	else {
1207 		*((u32 *)data_page2) = MAGIC_VAL_3;
1208 		vmcall();
1209 		if (vmx_get_test_stage() == 1) {
1210 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1211 					*((u32 *)data_page2) == MAGIC_VAL_2)
1212 				report_pass("EPT basic framework");
1213 			else
1214 				report_pass("EPT basic framework - remap");
1215 		}
1216 	}
1217 	// Test EPT Misconfigurations
1218 	vmx_set_test_stage(1);
1219 	vmcall();
1220 	*((u32 *)data_page1) = MAGIC_VAL_1;
1221 	if (vmx_get_test_stage() != 2) {
1222 		report_fail("EPT misconfigurations");
1223 		goto t1;
1224 	}
1225 	vmx_set_test_stage(2);
1226 	vmcall();
1227 	*((u32 *)data_page1) = MAGIC_VAL_1;
1228 	report(vmx_get_test_stage() == 3, "EPT misconfigurations");
1229 t1:
1230 	// Test EPT violation
1231 	vmx_set_test_stage(3);
1232 	vmcall();
1233 	*((u32 *)data_page1) = MAGIC_VAL_1;
1234 	report(vmx_get_test_stage() == 4, "EPT violation - page permission");
1235 	// Violation caused by EPT paging structure
1236 	vmx_set_test_stage(4);
1237 	vmcall();
1238 	*((u32 *)data_page1) = MAGIC_VAL_2;
1239 	report(vmx_get_test_stage() == 5, "EPT violation - paging structure");
1240 
1241 	// MMIO Read/Write
1242 	vmx_set_test_stage(5);
1243 	vmcall();
1244 
1245 	*(u32 volatile *)pci_physaddr;
1246 	report(vmx_get_test_stage() == 6, "MMIO EPT violation - read");
1247 
1248 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1249 	report(vmx_get_test_stage() == 7, "MMIO EPT violation - write");
1250 }
1251 
ept_main(void)1252 static void ept_main(void)
1253 {
1254 	ept_common();
1255 
1256 	// Test EPT access to L1 MMIO
1257 	vmx_set_test_stage(7);
1258 	report(*((u32 *)0xfee00030UL) == apic_version, "EPT - MMIO access");
1259 
1260 	// Test invalid operand for INVEPT
1261 	vmcall();
1262 	report(vmx_get_test_stage() == 8, "EPT - unsupported INVEPT");
1263 }
1264 
invept_test(int type,u64 eptp)1265 static bool invept_test(int type, u64 eptp)
1266 {
1267 	bool ret, supported;
1268 
1269 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1270 	ret = __invept(type, eptp);
1271 
1272 	if (ret == !supported)
1273 		return false;
1274 
1275 	if (!supported)
1276 		printf("WARNING: unsupported invept passed!\n");
1277 	else
1278 		printf("WARNING: invept failed!\n");
1279 
1280 	return true;
1281 }
1282 
pml_exit_handler(union exit_reason exit_reason)1283 static int pml_exit_handler(union exit_reason exit_reason)
1284 {
1285 	u16 index, count;
1286 	u64 *pmlbuf = pml_log;
1287 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1288 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1289 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1290 
1291 	switch (exit_reason.basic) {
1292 	case VMX_VMCALL:
1293 		switch (vmx_get_test_stage()) {
1294 		case 0:
1295 			index = vmcs_read(GUEST_PML_INDEX);
1296 			for (count = index + 1; count < PML_INDEX; count++) {
1297 				if (pmlbuf[count] == (u64)data_page2) {
1298 					vmx_inc_test_stage();
1299 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1300 					break;
1301 				}
1302 			}
1303 			break;
1304 		case 1:
1305 			index = vmcs_read(GUEST_PML_INDEX);
1306 			/* Keep clearing the dirty bit till a overflow */
1307 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1308 			break;
1309 		default:
1310 			report_fail("unexpected stage, %d.",
1311 			       vmx_get_test_stage());
1312 			print_vmexit_info(exit_reason);
1313 			return VMX_TEST_VMEXIT;
1314 		}
1315 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1316 		return VMX_TEST_RESUME;
1317 	case VMX_PML_FULL:
1318 		vmx_inc_test_stage();
1319 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1320 		return VMX_TEST_RESUME;
1321 	default:
1322 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1323 		print_vmexit_info(exit_reason);
1324 	}
1325 	return VMX_TEST_VMEXIT;
1326 }
1327 
ept_exit_handler_common(union exit_reason exit_reason,bool have_ad)1328 static int ept_exit_handler_common(union exit_reason exit_reason, bool have_ad)
1329 {
1330 	u64 guest_rip;
1331 	u64 guest_cr3;
1332 	u32 insn_len;
1333 	u32 exit_qual;
1334 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte,
1335 			     guest_pte_addr;
1336 
1337 	guest_rip = vmcs_read(GUEST_RIP);
1338 	guest_cr3 = vmcs_read(GUEST_CR3);
1339 	insn_len = vmcs_read(EXI_INST_LEN);
1340 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1341 	pteval_t *ptep;
1342 	switch (exit_reason.basic) {
1343 	case VMX_VMCALL:
1344 		switch (vmx_get_test_stage()) {
1345 		case 0:
1346 			check_ept_ad(pml4, guest_cr3,
1347 				     (unsigned long)data_page1,
1348 				     have_ad ? EPT_ACCESS_FLAG : 0,
1349 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1350 			check_ept_ad(pml4, guest_cr3,
1351 				     (unsigned long)data_page2,
1352 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1353 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1354 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1355 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1356 			if (have_ad)
1357 				invept(INVEPT_SINGLE, eptp);
1358 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1359 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1360 				vmx_inc_test_stage();
1361 				install_ept(pml4, (unsigned long)data_page2,
1362 						(unsigned long)data_page2,
1363 						EPT_RA | EPT_WA | EPT_EA);
1364 			} else
1365 				report_fail("EPT basic framework - write");
1366 			break;
1367 		case 1:
1368 			install_ept(pml4, (unsigned long)data_page1,
1369  				(unsigned long)data_page1, EPT_WA);
1370 			invept(INVEPT_SINGLE, eptp);
1371 			break;
1372 		case 2:
1373 			install_ept(pml4, (unsigned long)data_page1,
1374  				(unsigned long)data_page1,
1375  				EPT_RA | EPT_WA | EPT_EA |
1376  				(2 << EPT_MEM_TYPE_SHIFT));
1377 			invept(INVEPT_SINGLE, eptp);
1378 			break;
1379 		case 3:
1380 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1381 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1382 						1, &data_page1_pte));
1383 			set_ept_pte(pml4, (unsigned long)data_page1,
1384 				1, data_page1_pte & ~EPT_PRESENT);
1385 			invept(INVEPT_SINGLE, eptp);
1386 			break;
1387 		case 4:
1388 			ptep = get_pte_level((pgd_t *)guest_cr3, data_page1, /*level=*/2);
1389 			guest_pte_addr = virt_to_phys(ptep) & PAGE_MASK;
1390 
1391 			TEST_ASSERT(get_ept_pte(pml4, guest_pte_addr, 2, &data_page1_pte_pte));
1392 			set_ept_pte(pml4, guest_pte_addr, 2,
1393 				data_page1_pte_pte & ~EPT_PRESENT);
1394 			invept(INVEPT_SINGLE, eptp);
1395 			break;
1396 		case 5:
1397 			install_ept(pml4, (unsigned long)pci_physaddr,
1398 				(unsigned long)pci_physaddr, 0);
1399 			invept(INVEPT_SINGLE, eptp);
1400 			break;
1401 		case 7:
1402 			if (!invept_test(0, eptp))
1403 				vmx_inc_test_stage();
1404 			break;
1405 		// Should not reach here
1406 		default:
1407 			report_fail("ERROR - unexpected stage, %d.",
1408 			       vmx_get_test_stage());
1409 			print_vmexit_info(exit_reason);
1410 			return VMX_TEST_VMEXIT;
1411 		}
1412 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1413 		return VMX_TEST_RESUME;
1414 	case VMX_EPT_MISCONFIG:
1415 		switch (vmx_get_test_stage()) {
1416 		case 1:
1417 		case 2:
1418 			vmx_inc_test_stage();
1419 			install_ept(pml4, (unsigned long)data_page1,
1420  				(unsigned long)data_page1,
1421  				EPT_RA | EPT_WA | EPT_EA);
1422 			invept(INVEPT_SINGLE, eptp);
1423 			break;
1424 		// Should not reach here
1425 		default:
1426 			report_fail("ERROR - unexpected stage, %d.",
1427 			       vmx_get_test_stage());
1428 			print_vmexit_info(exit_reason);
1429 			return VMX_TEST_VMEXIT;
1430 		}
1431 		return VMX_TEST_RESUME;
1432 	case VMX_EPT_VIOLATION:
1433 		/*
1434 		 * Exit-qualifications are masked not to account for advanced
1435 		 * VM-exit information. Once KVM supports this feature, this
1436 		 * masking should be removed.
1437 		 */
1438 		exit_qual &= ~EPT_VLT_GUEST_MASK;
1439 
1440 		switch(vmx_get_test_stage()) {
1441 		case 3:
1442 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1443 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1444 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1445 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1446 					EPT_VLT_PADDR))
1447 				vmx_inc_test_stage();
1448 			set_ept_pte(pml4, (unsigned long)data_page1,
1449 				1, data_page1_pte | (EPT_PRESENT));
1450 			invept(INVEPT_SINGLE, eptp);
1451 			break;
1452 		case 4:
1453 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1454 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1455 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1456 			if (exit_qual == (EPT_VLT_RD |
1457 					  (have_ad ? EPT_VLT_WR : 0) |
1458 					  EPT_VLT_LADDR_VLD))
1459 				vmx_inc_test_stage();
1460 			set_ept_pte(pml4, guest_pte_addr, 2,
1461 				data_page1_pte_pte | (EPT_PRESENT));
1462 			invept(INVEPT_SINGLE, eptp);
1463 			break;
1464 		case 5:
1465 			if (exit_qual & EPT_VLT_RD)
1466 				vmx_inc_test_stage();
1467 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1468 						1, &memaddr_pte));
1469 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1470 			invept(INVEPT_SINGLE, eptp);
1471 			break;
1472 		case 6:
1473 			if (exit_qual & EPT_VLT_WR)
1474 				vmx_inc_test_stage();
1475 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1476 						1, &memaddr_pte));
1477 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1478 			invept(INVEPT_SINGLE, eptp);
1479 			break;
1480 		default:
1481 			// Should not reach here
1482 			report_fail("ERROR : unexpected stage, %d",
1483 			       vmx_get_test_stage());
1484 			print_vmexit_info(exit_reason);
1485 			return VMX_TEST_VMEXIT;
1486 		}
1487 		return VMX_TEST_RESUME;
1488 	default:
1489 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1490 		print_vmexit_info(exit_reason);
1491 	}
1492 	return VMX_TEST_VMEXIT;
1493 }
1494 
ept_exit_handler(union exit_reason exit_reason)1495 static int ept_exit_handler(union exit_reason exit_reason)
1496 {
1497 	return ept_exit_handler_common(exit_reason, false);
1498 }
1499 
eptad_init(struct vmcs * vmcs)1500 static int eptad_init(struct vmcs *vmcs)
1501 {
1502 	int r = ept_init_common(true);
1503 
1504 	if (r == VMX_TEST_EXIT)
1505 		return r;
1506 
1507 	if (!ept_ad_bits_supported()) {
1508 		printf("\tEPT A/D bits are not supported");
1509 		return VMX_TEST_EXIT;
1510 	}
1511 
1512 	return r;
1513 }
1514 
pml_init(struct vmcs * vmcs)1515 static int pml_init(struct vmcs *vmcs)
1516 {
1517 	u32 ctrl_cpu;
1518 	int r = eptad_init(vmcs);
1519 
1520 	if (r == VMX_TEST_EXIT)
1521 		return r;
1522 
1523 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1524 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1525 		printf("\tPML is not supported");
1526 		return VMX_TEST_EXIT;
1527 	}
1528 
1529 	pml_log = alloc_page();
1530 	vmcs_write(PMLADDR, (u64)pml_log);
1531 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1532 
1533 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1534 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1535 
1536 	return VMX_TEST_START;
1537 }
1538 
pml_main(void)1539 static void pml_main(void)
1540 {
1541 	int count = 0;
1542 
1543 	vmx_set_test_stage(0);
1544 	*((u32 *)data_page2) = 0x1;
1545 	vmcall();
1546 	report(vmx_get_test_stage() == 1, "PML - Dirty GPA Logging");
1547 
1548 	while (vmx_get_test_stage() == 1) {
1549 		vmcall();
1550 		*((u32 *)data_page2) = 0x1;
1551 		if (count++ > PML_INDEX)
1552 			break;
1553 	}
1554 	report(vmx_get_test_stage() == 2, "PML Full Event");
1555 }
1556 
eptad_main(void)1557 static void eptad_main(void)
1558 {
1559 	ept_common();
1560 }
1561 
eptad_exit_handler(union exit_reason exit_reason)1562 static int eptad_exit_handler(union exit_reason exit_reason)
1563 {
1564 	return ept_exit_handler_common(exit_reason, true);
1565 }
1566 
1567 #define TIMER_VECTOR	222
1568 
1569 static volatile bool timer_fired;
1570 
timer_isr(isr_regs_t * regs)1571 static void timer_isr(isr_regs_t *regs)
1572 {
1573 	timer_fired = true;
1574 	apic_write(APIC_EOI, 0);
1575 }
1576 
interrupt_init(struct vmcs * vmcs)1577 static int interrupt_init(struct vmcs *vmcs)
1578 {
1579 	msr_bmp_init();
1580 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1581 	handle_irq(TIMER_VECTOR, timer_isr);
1582 	return VMX_TEST_START;
1583 }
1584 
interrupt_main(void)1585 static void interrupt_main(void)
1586 {
1587 	long long start, loops;
1588 
1589 	vmx_set_test_stage(0);
1590 
1591 	apic_write(APIC_LVTT, TIMER_VECTOR);
1592 	sti();
1593 
1594 	apic_write(APIC_TMICT, 1);
1595 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1596 		asm volatile ("nop");
1597 	report(timer_fired, "direct interrupt while running guest");
1598 
1599 	apic_write(APIC_TMICT, 0);
1600 	cli();
1601 	vmcall();
1602 	timer_fired = false;
1603 	apic_write(APIC_TMICT, 1);
1604 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1605 		asm volatile ("nop");
1606 	report(timer_fired, "intercepted interrupt while running guest");
1607 
1608 	sti();
1609 	apic_write(APIC_TMICT, 0);
1610 	cli();
1611 	vmcall();
1612 	timer_fired = false;
1613 	start = rdtsc();
1614 	apic_write(APIC_TMICT, 1000000);
1615 
1616 	safe_halt();
1617 
1618 	report(rdtsc() - start > 1000000 && timer_fired,
1619 	       "direct interrupt + hlt");
1620 
1621 	apic_write(APIC_TMICT, 0);
1622 	cli();
1623 	vmcall();
1624 	timer_fired = false;
1625 	start = rdtsc();
1626 	apic_write(APIC_TMICT, 1000000);
1627 
1628 	safe_halt();
1629 
1630 	report(rdtsc() - start > 10000 && timer_fired,
1631 	       "intercepted interrupt + hlt");
1632 
1633 	apic_write(APIC_TMICT, 0);
1634 	cli();
1635 	vmcall();
1636 	timer_fired = false;
1637 	start = rdtsc();
1638 	apic_write(APIC_TMICT, 1000000);
1639 
1640 	sti_nop();
1641 	vmcall();
1642 
1643 	report(rdtsc() - start > 10000 && timer_fired,
1644 	       "direct interrupt + activity state hlt");
1645 
1646 	apic_write(APIC_TMICT, 0);
1647 	cli();
1648 	vmcall();
1649 	timer_fired = false;
1650 	start = rdtsc();
1651 	apic_write(APIC_TMICT, 1000000);
1652 
1653 	sti_nop();
1654 	vmcall();
1655 
1656 	report(rdtsc() - start > 10000 && timer_fired,
1657 	       "intercepted interrupt + activity state hlt");
1658 
1659 	apic_write(APIC_TMICT, 0);
1660 	cli();
1661 	vmx_set_test_stage(7);
1662 	vmcall();
1663 	timer_fired = false;
1664 	apic_write(APIC_TMICT, 1);
1665 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1666 		asm volatile ("nop");
1667 	report(timer_fired,
1668 	       "running a guest with interrupt acknowledgement set");
1669 
1670 	apic_write(APIC_TMICT, 0);
1671 	sti();
1672 	timer_fired = false;
1673 	vmcall();
1674 	report(timer_fired, "Inject an event to a halted guest");
1675 }
1676 
interrupt_exit_handler(union exit_reason exit_reason)1677 static int interrupt_exit_handler(union exit_reason exit_reason)
1678 {
1679 	u64 guest_rip = vmcs_read(GUEST_RIP);
1680 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1681 
1682 	switch (exit_reason.basic) {
1683 	case VMX_VMCALL:
1684 		switch (vmx_get_test_stage()) {
1685 		case 0:
1686 		case 2:
1687 		case 5:
1688 			vmcs_write(PIN_CONTROLS,
1689 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1690 			break;
1691 		case 7:
1692 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1693 			vmcs_write(PIN_CONTROLS,
1694 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1695 			break;
1696 		case 1:
1697 		case 3:
1698 			vmcs_write(PIN_CONTROLS,
1699 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1700 			break;
1701 		case 4:
1702 		case 6:
1703 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1704 			break;
1705 
1706 		case 8:
1707 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1708 			vmcs_write(ENT_INTR_INFO,
1709 				   TIMER_VECTOR |
1710 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1711 				   INTR_INFO_VALID_MASK);
1712 			break;
1713 		}
1714 		vmx_inc_test_stage();
1715 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1716 		return VMX_TEST_RESUME;
1717 	case VMX_EXTINT:
1718 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1719 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1720 			handle_external_interrupt(vector);
1721 		} else {
1722 			sti_nop_cli();
1723 		}
1724 		if (vmx_get_test_stage() >= 2)
1725 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1726 		return VMX_TEST_RESUME;
1727 	default:
1728 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1729 		print_vmexit_info(exit_reason);
1730 	}
1731 
1732 	return VMX_TEST_VMEXIT;
1733 }
1734 
1735 
1736 static volatile int nmi_fired;
1737 
1738 #define NMI_DELAY 100000000ULL
1739 
nmi_isr(isr_regs_t * regs)1740 static void nmi_isr(isr_regs_t *regs)
1741 {
1742 	nmi_fired = true;
1743 }
1744 
nmi_hlt_init(struct vmcs * vmcs)1745 static int nmi_hlt_init(struct vmcs *vmcs)
1746 {
1747 	msr_bmp_init();
1748 	handle_irq(NMI_VECTOR, nmi_isr);
1749 	vmcs_write(PIN_CONTROLS,
1750 		   vmcs_read(PIN_CONTROLS) & ~PIN_NMI);
1751 	vmcs_write(PIN_CONTROLS,
1752 		   vmcs_read(PIN_CONTROLS) & ~PIN_VIRT_NMI);
1753 	return VMX_TEST_START;
1754 }
1755 
nmi_message_thread(void * data)1756 static void nmi_message_thread(void *data)
1757 {
1758     while (vmx_get_test_stage() != 1)
1759         pause();
1760 
1761     delay(NMI_DELAY);
1762     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1763 
1764     while (vmx_get_test_stage() != 2)
1765         pause();
1766 
1767     delay(NMI_DELAY);
1768     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1769 }
1770 
nmi_hlt_main(void)1771 static void nmi_hlt_main(void)
1772 {
1773     long long start;
1774 
1775     if (cpu_count() < 2) {
1776         report_skip("%s : CPU count < 2", __func__);
1777         vmx_set_test_stage(-1);
1778         return;
1779     }
1780 
1781     vmx_set_test_stage(0);
1782     on_cpu_async(1, nmi_message_thread, NULL);
1783     start = rdtsc();
1784     vmx_set_test_stage(1);
1785     asm volatile ("hlt");
1786     report((rdtsc() - start > NMI_DELAY) && nmi_fired,
1787             "direct NMI + hlt");
1788     if (!nmi_fired)
1789         vmx_set_test_stage(-1);
1790     nmi_fired = false;
1791 
1792     vmcall();
1793 
1794     start = rdtsc();
1795     vmx_set_test_stage(2);
1796     asm volatile ("hlt");
1797     report((rdtsc() - start > NMI_DELAY) && !nmi_fired,
1798             "intercepted NMI + hlt");
1799     if (nmi_fired) {
1800         report(!nmi_fired, "intercepted NMI was dispatched");
1801         vmx_set_test_stage(-1);
1802         return;
1803     }
1804     vmx_set_test_stage(3);
1805 }
1806 
nmi_hlt_exit_handler(union exit_reason exit_reason)1807 static int nmi_hlt_exit_handler(union exit_reason exit_reason)
1808 {
1809     u64 guest_rip = vmcs_read(GUEST_RIP);
1810     u32 insn_len = vmcs_read(EXI_INST_LEN);
1811 
1812     switch (vmx_get_test_stage()) {
1813     case 1:
1814         if (exit_reason.basic != VMX_VMCALL) {
1815             report_fail("VMEXIT not due to vmcall. Exit reason 0x%x",
1816                         exit_reason.full);
1817             print_vmexit_info(exit_reason);
1818             return VMX_TEST_VMEXIT;
1819         }
1820 
1821         vmcs_write(PIN_CONTROLS,
1822                vmcs_read(PIN_CONTROLS) | PIN_NMI);
1823         vmcs_write(PIN_CONTROLS,
1824                vmcs_read(PIN_CONTROLS) | PIN_VIRT_NMI);
1825         vmcs_write(GUEST_RIP, guest_rip + insn_len);
1826         break;
1827 
1828     case 2:
1829         if (exit_reason.basic != VMX_EXC_NMI) {
1830             report_fail("VMEXIT not due to NMI intercept. Exit reason 0x%x",
1831                         exit_reason.full);
1832             print_vmexit_info(exit_reason);
1833             return VMX_TEST_VMEXIT;
1834         }
1835         report_pass("NMI intercept while running guest");
1836         vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1837         break;
1838 
1839     case 3:
1840         break;
1841 
1842     default:
1843         return VMX_TEST_VMEXIT;
1844     }
1845 
1846     if (vmx_get_test_stage() == 3)
1847         return VMX_TEST_VMEXIT;
1848 
1849     return VMX_TEST_RESUME;
1850 }
1851 
1852 
dbgctls_init(struct vmcs * vmcs)1853 static int dbgctls_init(struct vmcs *vmcs)
1854 {
1855 	u64 dr7 = 0x402;
1856 	u64 zero = 0;
1857 
1858 	msr_bmp_init();
1859 	asm volatile(
1860 		"mov %0,%%dr0\n\t"
1861 		"mov %0,%%dr1\n\t"
1862 		"mov %0,%%dr2\n\t"
1863 		"mov %1,%%dr7\n\t"
1864 		: : "r" (zero), "r" (dr7));
1865 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1866 	vmcs_write(GUEST_DR7, 0x404);
1867 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1868 
1869 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1870 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1871 
1872 	return VMX_TEST_START;
1873 }
1874 
dbgctls_main(void)1875 static void dbgctls_main(void)
1876 {
1877 	u64 dr7, debugctl;
1878 
1879 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1880 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1881 	/* Commented out: KVM does not support DEBUGCTL so far */
1882 	(void)debugctl;
1883 	report(dr7 == 0x404, "Load debug controls" /* && debugctl == 0x2 */);
1884 
1885 	dr7 = 0x408;
1886 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1887 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1888 
1889 	vmx_set_test_stage(0);
1890 	vmcall();
1891 	report(vmx_get_test_stage() == 1, "Save debug controls");
1892 
1893 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1894 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1895 		printf("\tDebug controls are always loaded/saved\n");
1896 		return;
1897 	}
1898 	vmx_set_test_stage(2);
1899 	vmcall();
1900 
1901 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1902 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1903 	/* Commented out: KVM does not support DEBUGCTL so far */
1904 	(void)debugctl;
1905 	report(dr7 == 0x402,
1906 	       "Guest=host debug controls" /* && debugctl == 0x1 */);
1907 
1908 	dr7 = 0x408;
1909 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1910 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1911 
1912 	vmx_set_test_stage(3);
1913 	vmcall();
1914 	report(vmx_get_test_stage() == 4, "Don't save debug controls");
1915 }
1916 
dbgctls_exit_handler(union exit_reason exit_reason)1917 static int dbgctls_exit_handler(union exit_reason exit_reason)
1918 {
1919 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1920 	u64 guest_rip = vmcs_read(GUEST_RIP);
1921 	u64 dr7, debugctl;
1922 
1923 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1924 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1925 
1926 	switch (exit_reason.basic) {
1927 	case VMX_VMCALL:
1928 		switch (vmx_get_test_stage()) {
1929 		case 0:
1930 			if (dr7 == 0x400 && debugctl == 0 &&
1931 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1932 			    Commented out: KVM does not support DEBUGCTL so far
1933 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1934 				vmx_inc_test_stage();
1935 			break;
1936 		case 2:
1937 			dr7 = 0x402;
1938 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1939 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1940 			vmcs_write(GUEST_DR7, 0x404);
1941 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1942 
1943 			vmcs_write(ENT_CONTROLS,
1944 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1945 			vmcs_write(EXI_CONTROLS,
1946 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1947 			break;
1948 		case 3:
1949 			if (dr7 == 0x400 && debugctl == 0 &&
1950 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1951 			    Commented out: KVM does not support DEBUGCTL so far
1952 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1953 				vmx_inc_test_stage();
1954 			break;
1955 		}
1956 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1957 		return VMX_TEST_RESUME;
1958 	default:
1959 		report_fail("Unknown exit reason, %d", exit_reason.full);
1960 		print_vmexit_info(exit_reason);
1961 	}
1962 	return VMX_TEST_VMEXIT;
1963 }
1964 
1965 struct vmx_msr_entry {
1966 	u32 index;
1967 	u32 reserved;
1968 	u64 value;
1969 } __attribute__((packed));
1970 
1971 #define MSR_MAGIC 0x31415926
1972 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1973 
msr_switch_init(struct vmcs * vmcs)1974 static int msr_switch_init(struct vmcs *vmcs)
1975 {
1976 	msr_bmp_init();
1977 	exit_msr_store = alloc_page();
1978 	exit_msr_load = alloc_page();
1979 	entry_msr_load = alloc_page();
1980 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1981 	entry_msr_load[0].value = MSR_MAGIC;
1982 
1983 	vmx_set_test_stage(1);
1984 	vmcs_write(ENT_MSR_LD_CNT, 1);
1985 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1986 	vmcs_write(EXI_MSR_ST_CNT, 1);
1987 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1988 	vmcs_write(EXI_MSR_LD_CNT, 1);
1989 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1990 	return VMX_TEST_START;
1991 }
1992 
msr_switch_main(void)1993 static void msr_switch_main(void)
1994 {
1995 	if (vmx_get_test_stage() == 1) {
1996 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC,
1997 		       "VM entry MSR load");
1998 		vmx_set_test_stage(2);
1999 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
2000 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
2001 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
2002 		exit_msr_load[0].value = MSR_MAGIC + 2;
2003 	}
2004 	vmcall();
2005 }
2006 
msr_switch_exit_handler(union exit_reason exit_reason)2007 static int msr_switch_exit_handler(union exit_reason exit_reason)
2008 {
2009 	if (exit_reason.basic == VMX_VMCALL && vmx_get_test_stage() == 2) {
2010 		report(exit_msr_store[0].value == MSR_MAGIC + 1,
2011 		       "VM exit MSR store");
2012 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2,
2013 		       "VM exit MSR load");
2014 		vmx_set_test_stage(3);
2015 		entry_msr_load[0].index = MSR_FS_BASE;
2016 		return VMX_TEST_RESUME;
2017 	}
2018 	printf("ERROR %s: unexpected stage=%u or reason=0x%x\n",
2019 		__func__, vmx_get_test_stage(), exit_reason.full);
2020 	return VMX_TEST_EXIT;
2021 }
2022 
msr_switch_entry_failure(struct vmentry_result * result)2023 static int msr_switch_entry_failure(struct vmentry_result *result)
2024 {
2025 	if (result->vm_fail) {
2026 		printf("ERROR %s: VM-Fail on %s\n", __func__, result->instr);
2027 		return VMX_TEST_EXIT;
2028 	}
2029 
2030 	if (result->exit_reason.failed_vmentry &&
2031 	    result->exit_reason.basic == VMX_FAIL_MSR &&
2032 	    vmx_get_test_stage() == 3) {
2033 		report(vmcs_read(EXI_QUALIFICATION) == 1,
2034 		       "VM entry MSR load: try to load FS_BASE");
2035 		return VMX_TEST_VMEXIT;
2036 	}
2037 	printf("ERROR %s: unexpected stage=%u or reason=%x\n",
2038 		__func__, vmx_get_test_stage(), result->exit_reason.full);
2039 	return VMX_TEST_EXIT;
2040 }
2041 
vmmcall_init(struct vmcs * vmcs)2042 static int vmmcall_init(struct vmcs *vmcs)
2043 {
2044 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
2045 	return VMX_TEST_START;
2046 }
2047 
vmmcall_main(void)2048 static void vmmcall_main(void)
2049 {
2050 	asm volatile(
2051 		"mov $0xABCD, %%rax\n\t"
2052 		"vmmcall\n\t"
2053 		::: "rax");
2054 
2055 	report_fail("VMMCALL");
2056 }
2057 
vmmcall_exit_handler(union exit_reason exit_reason)2058 static int vmmcall_exit_handler(union exit_reason exit_reason)
2059 {
2060 	switch (exit_reason.basic) {
2061 	case VMX_VMCALL:
2062 		printf("here\n");
2063 		report_fail("VMMCALL triggers #UD");
2064 		break;
2065 	case VMX_EXC_NMI:
2066 		report((vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR,
2067 		       "VMMCALL triggers #UD");
2068 		break;
2069 	default:
2070 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2071 		print_vmexit_info(exit_reason);
2072 	}
2073 
2074 	return VMX_TEST_VMEXIT;
2075 }
2076 
disable_rdtscp_init(struct vmcs * vmcs)2077 static int disable_rdtscp_init(struct vmcs *vmcs)
2078 {
2079 	u32 ctrl_cpu1;
2080 
2081 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
2082 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
2083 		ctrl_cpu1 &= ~CPU_RDTSCP;
2084 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
2085 	}
2086 
2087 	return VMX_TEST_START;
2088 }
2089 
disable_rdtscp_ud_handler(struct ex_regs * regs)2090 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
2091 {
2092 	switch (vmx_get_test_stage()) {
2093 	case 0:
2094 		report_pass("RDTSCP triggers #UD");
2095 		vmx_inc_test_stage();
2096 		regs->rip += 3;
2097 		break;
2098 	case 2:
2099 		report_pass("RDPID triggers #UD");
2100 		vmx_inc_test_stage();
2101 		regs->rip += 4;
2102 		break;
2103 	}
2104 	return;
2105 
2106 }
2107 
disable_rdtscp_main(void)2108 static void disable_rdtscp_main(void)
2109 {
2110 	/* Test that #UD is properly injected in L2.  */
2111 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2112 
2113 	vmx_set_test_stage(0);
2114 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2115 	vmcall();
2116 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2117 
2118 	handle_exception(UD_VECTOR, 0);
2119 	vmcall();
2120 }
2121 
disable_rdtscp_exit_handler(union exit_reason exit_reason)2122 static int disable_rdtscp_exit_handler(union exit_reason exit_reason)
2123 {
2124 	switch (exit_reason.basic) {
2125 	case VMX_VMCALL:
2126 		switch (vmx_get_test_stage()) {
2127 		case 0:
2128 			report_fail("RDTSCP triggers #UD");
2129 			vmx_inc_test_stage();
2130 			/* fallthrough */
2131 		case 1:
2132 			vmx_inc_test_stage();
2133 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2134 			return VMX_TEST_RESUME;
2135 		case 2:
2136 			report_fail("RDPID triggers #UD");
2137 			break;
2138 		}
2139 		break;
2140 
2141 	default:
2142 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2143 		print_vmexit_info(exit_reason);
2144 	}
2145 	return VMX_TEST_VMEXIT;
2146 }
2147 
exit_monitor_from_l2_main(void)2148 static void exit_monitor_from_l2_main(void)
2149 {
2150 	printf("Calling exit(0) from l2...\n");
2151 	exit(0);
2152 }
2153 
exit_monitor_from_l2_handler(union exit_reason exit_reason)2154 static int exit_monitor_from_l2_handler(union exit_reason exit_reason)
2155 {
2156 	report_fail("The guest should have killed the VMM");
2157 	return VMX_TEST_EXIT;
2158 }
2159 
assert_exit_reason(u64 expected)2160 static void assert_exit_reason(u64 expected)
2161 {
2162 	u64 actual = vmcs_read(EXI_REASON);
2163 
2164 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2165 			   exit_reason_description(expected),
2166 			   exit_reason_description(actual));
2167 }
2168 
skip_exit_insn(void)2169 static void skip_exit_insn(void)
2170 {
2171 	u64 guest_rip = vmcs_read(GUEST_RIP);
2172 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2173 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2174 }
2175 
skip_exit_vmcall(void)2176 static void skip_exit_vmcall(void)
2177 {
2178 	assert_exit_reason(VMX_VMCALL);
2179 	skip_exit_insn();
2180 }
2181 
v2_null_test_guest(void)2182 static void v2_null_test_guest(void)
2183 {
2184 }
2185 
v2_null_test(void)2186 static void v2_null_test(void)
2187 {
2188 	test_set_guest(v2_null_test_guest);
2189 	enter_guest();
2190 	report_pass(__func__);
2191 }
2192 
v2_multiple_entries_test_guest(void)2193 static void v2_multiple_entries_test_guest(void)
2194 {
2195 	vmx_set_test_stage(1);
2196 	vmcall();
2197 	vmx_set_test_stage(2);
2198 }
2199 
v2_multiple_entries_test(void)2200 static void v2_multiple_entries_test(void)
2201 {
2202 	test_set_guest(v2_multiple_entries_test_guest);
2203 	enter_guest();
2204 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2205 	skip_exit_vmcall();
2206 	enter_guest();
2207 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2208 	report_pass(__func__);
2209 }
2210 
2211 static int fixture_test_data = 1;
2212 
fixture_test_teardown(void * data)2213 static void fixture_test_teardown(void *data)
2214 {
2215 	*((int *) data) = 1;
2216 }
2217 
fixture_test_guest(void)2218 static void fixture_test_guest(void)
2219 {
2220 	fixture_test_data++;
2221 }
2222 
2223 
fixture_test_setup(void)2224 static void fixture_test_setup(void)
2225 {
2226 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2227 			   "fixture_test_teardown didn't run?!");
2228 	fixture_test_data = 2;
2229 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2230 	test_set_guest(fixture_test_guest);
2231 }
2232 
fixture_test_case1(void)2233 static void fixture_test_case1(void)
2234 {
2235 	fixture_test_setup();
2236 	TEST_ASSERT_EQ(2, fixture_test_data);
2237 	enter_guest();
2238 	TEST_ASSERT_EQ(3, fixture_test_data);
2239 	report_pass(__func__);
2240 }
2241 
fixture_test_case2(void)2242 static void fixture_test_case2(void)
2243 {
2244 	fixture_test_setup();
2245 	TEST_ASSERT_EQ(2, fixture_test_data);
2246 	enter_guest();
2247 	TEST_ASSERT_EQ(3, fixture_test_data);
2248 	report_pass(__func__);
2249 }
2250 
2251 enum ept_access_op {
2252 	OP_READ,
2253 	OP_WRITE,
2254 	OP_EXEC,
2255 	OP_FLUSH_TLB,
2256 	OP_EXIT,
2257 };
2258 
2259 static struct ept_access_test_data {
2260 	unsigned long gpa;
2261 	unsigned long *gva;
2262 	unsigned long hpa;
2263 	unsigned long *hva;
2264 	enum ept_access_op op;
2265 } ept_access_test_data;
2266 
2267 extern unsigned char ret42_start;
2268 extern unsigned char ret42_end;
2269 
2270 /* Returns 42. */
2271 asm(
2272 	".align 64\n"
2273 	"ret42_start:\n"
2274 	"mov $42, %eax\n"
2275 	"ret\n"
2276 	"ret42_end:\n"
2277 );
2278 
2279 static void
diagnose_ept_violation_qual(u64 expected,u64 actual)2280 diagnose_ept_violation_qual(u64 expected, u64 actual)
2281 {
2282 
2283 #define DIAGNOSE(flag)							\
2284 do {									\
2285 	if ((expected & flag) != (actual & flag))			\
2286 		printf(#flag " %sexpected\n",				\
2287 		       (expected & flag) ? "" : "un");			\
2288 } while (0)
2289 
2290 	DIAGNOSE(EPT_VLT_RD);
2291 	DIAGNOSE(EPT_VLT_WR);
2292 	DIAGNOSE(EPT_VLT_FETCH);
2293 	DIAGNOSE(EPT_VLT_PERM_RD);
2294 	DIAGNOSE(EPT_VLT_PERM_WR);
2295 	DIAGNOSE(EPT_VLT_PERM_EX);
2296 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2297 	DIAGNOSE(EPT_VLT_PADDR);
2298 
2299 #undef DIAGNOSE
2300 }
2301 
do_ept_access_op(enum ept_access_op op)2302 static void do_ept_access_op(enum ept_access_op op)
2303 {
2304 	ept_access_test_data.op = op;
2305 	enter_guest();
2306 }
2307 
2308 /*
2309  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2310  * needed by tests that modify guest PTEs.
2311  */
ept_access_test_guest_flush_tlb(void)2312 static void ept_access_test_guest_flush_tlb(void)
2313 {
2314 	do_ept_access_op(OP_FLUSH_TLB);
2315 	skip_exit_vmcall();
2316 }
2317 
2318 /*
2319  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2320  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2321  * a huge page.
2322  */
ept_twiddle(unsigned long gpa,bool mkhuge,int level,unsigned long clear,unsigned long set)2323 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2324 				 unsigned long clear, unsigned long set)
2325 {
2326 	struct ept_access_test_data *data = &ept_access_test_data;
2327 	unsigned long orig_pte;
2328 	unsigned long pte;
2329 
2330 	/* Screw with the mapping at the requested level. */
2331 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2332 	pte = orig_pte;
2333 	if (mkhuge)
2334 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2335 	else
2336 		pte = orig_pte;
2337 	pte = (pte & ~clear) | set;
2338 	set_ept_pte(pml4, gpa, level, pte);
2339 	invept(INVEPT_SINGLE, eptp);
2340 
2341 	return orig_pte;
2342 }
2343 
ept_untwiddle(unsigned long gpa,int level,unsigned long orig_pte)2344 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2345 {
2346 	set_ept_pte(pml4, gpa, level, orig_pte);
2347 	invept(INVEPT_SINGLE, eptp);
2348 }
2349 
do_ept_violation(bool leaf,enum ept_access_op op,u64 expected_qual,u64 expected_paddr)2350 static void do_ept_violation(bool leaf, enum ept_access_op op,
2351 			     u64 expected_qual, u64 expected_paddr)
2352 {
2353 	u64 qual;
2354 
2355 	/* Try the access and observe the violation. */
2356 	do_ept_access_op(op);
2357 
2358 	assert_exit_reason(VMX_EPT_VIOLATION);
2359 
2360 	qual = vmcs_read(EXI_QUALIFICATION);
2361 
2362 	/* Mask undefined bits (which may later be defined in certain cases). */
2363 	qual &= ~(EPT_VLT_GUEST_USER | EPT_VLT_GUEST_RW | EPT_VLT_GUEST_EX |
2364 		 EPT_VLT_PERM_USER_EX);
2365 
2366 	diagnose_ept_violation_qual(expected_qual, qual);
2367 	TEST_EXPECT_EQ(expected_qual, qual);
2368 
2369 	#if 0
2370 	/* Disable for now otherwise every test will fail */
2371 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2372 		       (unsigned long) (
2373 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2374 	#endif
2375 	/*
2376 	 * TODO: tests that probe expected_paddr in pages other than the one at
2377 	 * the beginning of the 1g region.
2378 	 */
2379 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2380 }
2381 
2382 static void
ept_violation_at_level_mkhuge(bool mkhuge,int level,unsigned long clear,unsigned long set,enum ept_access_op op,u64 expected_qual)2383 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2384 			      unsigned long set, enum ept_access_op op,
2385 			      u64 expected_qual)
2386 {
2387 	struct ept_access_test_data *data = &ept_access_test_data;
2388 	unsigned long orig_pte;
2389 
2390 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2391 
2392 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2393 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2394 					 data->gpa);
2395 
2396 	/* Fix the violation and resume the op loop. */
2397 	ept_untwiddle(data->gpa, level, orig_pte);
2398 	enter_guest();
2399 	skip_exit_vmcall();
2400 }
2401 
2402 static void
ept_violation_at_level(int level,unsigned long clear,unsigned long set,enum ept_access_op op,u64 expected_qual)2403 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2404 		       enum ept_access_op op, u64 expected_qual)
2405 {
2406 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2407 				      expected_qual);
2408 	if (ept_huge_pages_supported(level))
2409 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2410 					      expected_qual);
2411 }
2412 
ept_violation(unsigned long clear,unsigned long set,enum ept_access_op op,u64 expected_qual)2413 static void ept_violation(unsigned long clear, unsigned long set,
2414 			  enum ept_access_op op, u64 expected_qual)
2415 {
2416 	ept_violation_at_level(1, clear, set, op, expected_qual);
2417 	ept_violation_at_level(2, clear, set, op, expected_qual);
2418 	ept_violation_at_level(3, clear, set, op, expected_qual);
2419 	ept_violation_at_level(4, clear, set, op, expected_qual);
2420 }
2421 
ept_access_violation(unsigned long access,enum ept_access_op op,u64 expected_qual)2422 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2423 				       u64 expected_qual)
2424 {
2425 	ept_violation(EPT_PRESENT, access, op,
2426 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2427 }
2428 
2429 /*
2430  * For translations that don't involve a GVA, that is physical address (paddr)
2431  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2432  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2433  * translations don't involve GVAs, such as when the hardware does the guest
2434  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2435  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2436  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2437  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2438  *
2439  * Note that paddr violations can also be triggered by loading PAE page tables
2440  * with wonky addresses. We don't test that yet.
2441  *
2442  * This function modifies the EPT entry that maps the GPA that the guest page
2443  * table entry mapping ept_access_test_data.gva resides on.
2444  *
2445  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2446  *
2447  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2448  *
2449  *	@op		Guest operation to perform with
2450  *			ept_access_test_data.gva.
2451  *
2452  *	@expect_violation
2453  *			Is a violation expected during the paddr access?
2454  *
2455  *	@expected_qual	Expected qualification for the EPT violation.
2456  *			EPT_VLT_PADDR should be clear.
2457  */
ept_access_paddr(unsigned long ept_access,unsigned long pte_ad,enum ept_access_op op,bool expect_violation,u64 expected_qual)2458 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2459 			     enum ept_access_op op, bool expect_violation,
2460 			     u64 expected_qual)
2461 {
2462 	struct ept_access_test_data *data = &ept_access_test_data;
2463 	unsigned long *ptep;
2464 	unsigned long gpa;
2465 	unsigned long orig_epte;
2466 	unsigned long epte;
2467 	int i;
2468 
2469 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2470 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2471 	TEST_ASSERT(ptep);
2472 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2473 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2474 	ept_access_test_guest_flush_tlb();
2475 
2476 	/*
2477 	 * Now modify the access bits on the EPT entry for the GPA that the
2478 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2479 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2480 	 * constructed our test such that those other 511 PTEs aren't used by
2481 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2482 	 * PTE we're modifying is at the beginning of a 4K page and the
2483 	 * following 511 entries are also under our control (and not touched by
2484 	 * the guest).
2485 	 */
2486 	gpa = virt_to_phys(ptep);
2487 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2488 	/*
2489 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2490 	 * otherwise our level=1 twiddling below will fail. We use the
2491 	 * identity map (gpa = gpa) since page tables are shared with the host.
2492 	 */
2493 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2494 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2495 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2496 
2497 	if (expect_violation) {
2498 		do_ept_violation(/*leaf=*/true, op,
2499 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2500 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2501 		do_ept_access_op(op);
2502 	} else {
2503 		do_ept_access_op(op);
2504 		if (ept_ad_enabled()) {
2505 			for (i = EPT_PAGE_LEVEL; i > 0; i--) {
2506 				TEST_ASSERT(get_ept_pte(pml4, gpa, i, &epte));
2507 				TEST_ASSERT(epte & EPT_ACCESS_FLAG);
2508 				if (i == 1)
2509 					TEST_ASSERT(epte & EPT_DIRTY_FLAG);
2510 				else
2511 					TEST_ASSERT_EQ(epte & EPT_DIRTY_FLAG, 0);
2512 			}
2513 		}
2514 
2515 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2516 	}
2517 
2518 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2519 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2520 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2521 
2522 	skip_exit_vmcall();
2523 }
2524 
ept_access_allowed_paddr(unsigned long ept_access,unsigned long pte_ad,enum ept_access_op op)2525 static void ept_access_allowed_paddr(unsigned long ept_access,
2526 				     unsigned long pte_ad,
2527 				     enum ept_access_op op)
2528 {
2529 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2530 			 /*expected_qual=*/-1);
2531 }
2532 
ept_access_violation_paddr(unsigned long ept_access,unsigned long pte_ad,enum ept_access_op op,u64 expected_qual)2533 static void ept_access_violation_paddr(unsigned long ept_access,
2534 				       unsigned long pte_ad,
2535 				       enum ept_access_op op,
2536 				       u64 expected_qual)
2537 {
2538 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2539 			 expected_qual);
2540 }
2541 
2542 
ept_allowed_at_level_mkhuge(bool mkhuge,int level,unsigned long clear,unsigned long set,enum ept_access_op op)2543 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2544 					unsigned long clear,
2545 					unsigned long set,
2546 					enum ept_access_op op)
2547 {
2548 	struct ept_access_test_data *data = &ept_access_test_data;
2549 	unsigned long orig_pte;
2550 
2551 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2552 
2553 	/* No violation. Should proceed to vmcall. */
2554 	do_ept_access_op(op);
2555 	skip_exit_vmcall();
2556 
2557 	ept_untwiddle(data->gpa, level, orig_pte);
2558 }
2559 
ept_allowed_at_level(int level,unsigned long clear,unsigned long set,enum ept_access_op op)2560 static void ept_allowed_at_level(int level, unsigned long clear,
2561 				 unsigned long set, enum ept_access_op op)
2562 {
2563 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2564 	if (ept_huge_pages_supported(level))
2565 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2566 }
2567 
ept_allowed(unsigned long clear,unsigned long set,enum ept_access_op op)2568 static void ept_allowed(unsigned long clear, unsigned long set,
2569 			enum ept_access_op op)
2570 {
2571 	ept_allowed_at_level(1, clear, set, op);
2572 	ept_allowed_at_level(2, clear, set, op);
2573 	ept_allowed_at_level(3, clear, set, op);
2574 	ept_allowed_at_level(4, clear, set, op);
2575 }
2576 
ept_ignored_bit(int bit)2577 static void ept_ignored_bit(int bit)
2578 {
2579 	/* Set the bit. */
2580 	ept_allowed(0, 1ul << bit, OP_READ);
2581 	ept_allowed(0, 1ul << bit, OP_WRITE);
2582 	ept_allowed(0, 1ul << bit, OP_EXEC);
2583 
2584 	/* Clear the bit. */
2585 	ept_allowed(1ul << bit, 0, OP_READ);
2586 	ept_allowed(1ul << bit, 0, OP_WRITE);
2587 	ept_allowed(1ul << bit, 0, OP_EXEC);
2588 }
2589 
ept_access_allowed(unsigned long access,enum ept_access_op op)2590 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2591 {
2592 	ept_allowed(EPT_PRESENT, access, op);
2593 }
2594 
2595 
ept_misconfig_at_level_mkhuge_op(bool mkhuge,int level,unsigned long clear,unsigned long set,enum ept_access_op op)2596 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2597 					     unsigned long clear,
2598 					     unsigned long set,
2599 					     enum ept_access_op op)
2600 {
2601 	struct ept_access_test_data *data = &ept_access_test_data;
2602 	unsigned long orig_pte;
2603 
2604 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2605 
2606 	do_ept_access_op(op);
2607 	assert_exit_reason(VMX_EPT_MISCONFIG);
2608 
2609 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2610 	#if 0
2611 	/* broken: */
2612 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2613 	#endif
2614 	#if 0
2615 	/*
2616 	 * broken:
2617 	 * According to description of exit qual for EPT violation,
2618 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2619 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2620 	 * to be set for msiconfig.
2621 	 */
2622 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2623 		       (unsigned long) (
2624 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2625 	#endif
2626 
2627 	/* Fix the violation and resume the op loop. */
2628 	ept_untwiddle(data->gpa, level, orig_pte);
2629 	enter_guest();
2630 	skip_exit_vmcall();
2631 }
2632 
ept_misconfig_at_level_mkhuge(bool mkhuge,int level,unsigned long clear,unsigned long set)2633 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2634 					  unsigned long clear,
2635 					  unsigned long set)
2636 {
2637 	/* The op shouldn't matter (read, write, exec), so try them all! */
2638 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2639 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2640 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2641 }
2642 
ept_misconfig_at_level(int level,unsigned long clear,unsigned long set)2643 static void ept_misconfig_at_level(int level, unsigned long clear,
2644 				   unsigned long set)
2645 {
2646 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2647 	if (ept_huge_pages_supported(level))
2648 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2649 }
2650 
ept_misconfig(unsigned long clear,unsigned long set)2651 static void ept_misconfig(unsigned long clear, unsigned long set)
2652 {
2653 	ept_misconfig_at_level(1, clear, set);
2654 	ept_misconfig_at_level(2, clear, set);
2655 	ept_misconfig_at_level(3, clear, set);
2656 	ept_misconfig_at_level(4, clear, set);
2657 }
2658 
ept_access_misconfig(unsigned long access)2659 static void ept_access_misconfig(unsigned long access)
2660 {
2661 	ept_misconfig(EPT_PRESENT, access);
2662 }
2663 
ept_reserved_bit_at_level_nohuge(int level,int bit)2664 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2665 {
2666 	/* Setting the bit causes a misconfig. */
2667 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2668 
2669 	/* Making the entry non-present turns reserved bits into ignored. */
2670 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2671 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2672 }
2673 
ept_reserved_bit_at_level_huge(int level,int bit)2674 static void ept_reserved_bit_at_level_huge(int level, int bit)
2675 {
2676 	/* Setting the bit causes a misconfig. */
2677 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2678 
2679 	/* Making the entry non-present turns reserved bits into ignored. */
2680 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2681 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2682 }
2683 
ept_reserved_bit_at_level(int level,int bit)2684 static void ept_reserved_bit_at_level(int level, int bit)
2685 {
2686 	/* Setting the bit causes a misconfig. */
2687 	ept_misconfig_at_level(level, 0, 1ul << bit);
2688 
2689 	/* Making the entry non-present turns reserved bits into ignored. */
2690 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2691 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2692 }
2693 
ept_reserved_bit(int bit)2694 static void ept_reserved_bit(int bit)
2695 {
2696 	ept_reserved_bit_at_level(1, bit);
2697 	ept_reserved_bit_at_level(2, bit);
2698 	ept_reserved_bit_at_level(3, bit);
2699 	ept_reserved_bit_at_level(4, bit);
2700 }
2701 
2702 #define PAGE_2M_ORDER 9
2703 #define PAGE_1G_ORDER 18
2704 
get_1g_page(void)2705 static void *get_1g_page(void)
2706 {
2707 	static void *alloc;
2708 
2709 	if (!alloc)
2710 		alloc = alloc_pages(PAGE_1G_ORDER);
2711 	return alloc;
2712 }
2713 
ept_access_test_teardown(void * unused)2714 static void ept_access_test_teardown(void *unused)
2715 {
2716 	/* Exit the guest cleanly. */
2717 	do_ept_access_op(OP_EXIT);
2718 }
2719 
ept_access_test_guest(void)2720 static void ept_access_test_guest(void)
2721 {
2722 	struct ept_access_test_data *data = &ept_access_test_data;
2723 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2724 
2725 	while (true) {
2726 		switch (data->op) {
2727 		case OP_READ:
2728 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2729 			break;
2730 		case OP_WRITE:
2731 			*data->gva = MAGIC_VAL_2;
2732 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2733 			*data->gva = MAGIC_VAL_1;
2734 			break;
2735 		case OP_EXEC:
2736 			TEST_ASSERT_EQ(42, code());
2737 			break;
2738 		case OP_FLUSH_TLB:
2739 			write_cr3(read_cr3());
2740 			break;
2741 		case OP_EXIT:
2742 			return;
2743 		default:
2744 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2745 		}
2746 		vmcall();
2747 	}
2748 }
2749 
ept_access_test_setup(void)2750 static void ept_access_test_setup(void)
2751 {
2752 	struct ept_access_test_data *data = &ept_access_test_data;
2753 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2754 	unsigned long size = npages * PAGE_SIZE;
2755 	unsigned long *page_table = current_page_table();
2756 	unsigned long pte;
2757 
2758 	if (setup_ept(false))
2759 		test_skip("EPT not supported");
2760 
2761 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2762 	if (cpuid_maxphyaddr() < 40)
2763 		test_skip("Test needs MAXPHYADDR >= 40");
2764 
2765 	test_set_guest(ept_access_test_guest);
2766 	test_add_teardown(ept_access_test_teardown, NULL);
2767 
2768 	data->hva = get_1g_page();
2769 	TEST_ASSERT(data->hva);
2770 	data->hpa = virt_to_phys(data->hva);
2771 
2772 	data->gpa = 1ul << 39;
2773 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2774 				   size);
2775 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2776 	install_pages(page_table, data->gpa, size, data->gva);
2777 
2778 	/*
2779 	 * Make sure nothing's mapped here so the tests that screw with the
2780 	 * pml4 entry don't inadvertently break something.
2781 	 */
2782 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2783 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2784 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2785 
2786 	data->hva[0] = MAGIC_VAL_1;
2787 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2788 }
2789 
ept_access_test_not_present(void)2790 static void ept_access_test_not_present(void)
2791 {
2792 	ept_access_test_setup();
2793 	/* --- */
2794 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2795 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2796 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2797 }
2798 
ept_access_test_read_only(void)2799 static void ept_access_test_read_only(void)
2800 {
2801 	ept_access_test_setup();
2802 
2803 	/* r-- */
2804 	ept_access_allowed(EPT_RA, OP_READ);
2805 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2806 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2807 }
2808 
ept_access_test_write_only(void)2809 static void ept_access_test_write_only(void)
2810 {
2811 	ept_access_test_setup();
2812 	/* -w- */
2813 	ept_access_misconfig(EPT_WA);
2814 }
2815 
ept_access_test_read_write(void)2816 static void ept_access_test_read_write(void)
2817 {
2818 	ept_access_test_setup();
2819 	/* rw- */
2820 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2821 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2822 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2823 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2824 }
2825 
2826 
ept_access_test_execute_only(void)2827 static void ept_access_test_execute_only(void)
2828 {
2829 	ept_access_test_setup();
2830 	/* --x */
2831 	if (ept_execute_only_supported()) {
2832 		ept_access_violation(EPT_EA, OP_READ,
2833 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2834 		ept_access_violation(EPT_EA, OP_WRITE,
2835 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2836 		ept_access_allowed(EPT_EA, OP_EXEC);
2837 	} else {
2838 		ept_access_misconfig(EPT_EA);
2839 	}
2840 }
2841 
ept_access_test_read_execute(void)2842 static void ept_access_test_read_execute(void)
2843 {
2844 	ept_access_test_setup();
2845 	/* r-x */
2846 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2847 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2848 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2849 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2850 }
2851 
ept_access_test_write_execute(void)2852 static void ept_access_test_write_execute(void)
2853 {
2854 	ept_access_test_setup();
2855 	/* -wx */
2856 	ept_access_misconfig(EPT_WA | EPT_EA);
2857 }
2858 
ept_access_test_read_write_execute(void)2859 static void ept_access_test_read_write_execute(void)
2860 {
2861 	ept_access_test_setup();
2862 	/* rwx */
2863 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2864 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2865 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2866 }
2867 
ept_access_test_reserved_bits(void)2868 static void ept_access_test_reserved_bits(void)
2869 {
2870 	int i;
2871 	int maxphyaddr;
2872 
2873 	ept_access_test_setup();
2874 
2875 	/* Reserved bits above maxphyaddr. */
2876 	maxphyaddr = cpuid_maxphyaddr();
2877 	for (i = maxphyaddr; i <= 51; i++) {
2878 		report_prefix_pushf("reserved_bit=%d", i);
2879 		ept_reserved_bit(i);
2880 		report_prefix_pop();
2881 	}
2882 
2883 	/* Level-specific reserved bits. */
2884 	ept_reserved_bit_at_level_nohuge(2, 3);
2885 	ept_reserved_bit_at_level_nohuge(2, 4);
2886 	ept_reserved_bit_at_level_nohuge(2, 5);
2887 	ept_reserved_bit_at_level_nohuge(2, 6);
2888 	/* 2M alignment. */
2889 	for (i = 12; i < 20; i++) {
2890 		report_prefix_pushf("reserved_bit=%d", i);
2891 		ept_reserved_bit_at_level_huge(2, i);
2892 		report_prefix_pop();
2893 	}
2894 	ept_reserved_bit_at_level_nohuge(3, 3);
2895 	ept_reserved_bit_at_level_nohuge(3, 4);
2896 	ept_reserved_bit_at_level_nohuge(3, 5);
2897 	ept_reserved_bit_at_level_nohuge(3, 6);
2898 	/* 1G alignment. */
2899 	for (i = 12; i < 29; i++) {
2900 		report_prefix_pushf("reserved_bit=%d", i);
2901 		ept_reserved_bit_at_level_huge(3, i);
2902 		report_prefix_pop();
2903 	}
2904 	ept_reserved_bit_at_level(4, 3);
2905 	ept_reserved_bit_at_level(4, 4);
2906 	ept_reserved_bit_at_level(4, 5);
2907 	ept_reserved_bit_at_level(4, 6);
2908 	ept_reserved_bit_at_level(4, 7);
2909 }
2910 
ept_access_test_ignored_bits(void)2911 static void ept_access_test_ignored_bits(void)
2912 {
2913 	ept_access_test_setup();
2914 	/*
2915 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2916 	 * far as translation is concerned even if AD bits are enabled in the
2917 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2918 	 * control is 0.
2919 	 */
2920 	ept_ignored_bit(8);
2921 	ept_ignored_bit(9);
2922 	ept_ignored_bit(10);
2923 	ept_ignored_bit(11);
2924 	ept_ignored_bit(52);
2925 	ept_ignored_bit(53);
2926 	ept_ignored_bit(54);
2927 	ept_ignored_bit(55);
2928 	ept_ignored_bit(56);
2929 	ept_ignored_bit(57);
2930 	ept_ignored_bit(58);
2931 	ept_ignored_bit(59);
2932 	ept_ignored_bit(60);
2933 	ept_ignored_bit(61);
2934 	ept_ignored_bit(62);
2935 	ept_ignored_bit(63);
2936 }
2937 
ept_access_test_paddr_not_present_ad_disabled(void)2938 static void ept_access_test_paddr_not_present_ad_disabled(void)
2939 {
2940 	ept_access_test_setup();
2941 	ept_disable_ad_bits();
2942 
2943 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
2944 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
2945 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
2946 }
2947 
ept_access_test_paddr_not_present_ad_enabled(void)2948 static void ept_access_test_paddr_not_present_ad_enabled(void)
2949 {
2950 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
2951 
2952 	ept_access_test_setup();
2953 	ept_enable_ad_bits_or_skip_test();
2954 
2955 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
2956 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
2957 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
2958 }
2959 
ept_access_test_paddr_read_only_ad_disabled(void)2960 static void ept_access_test_paddr_read_only_ad_disabled(void)
2961 {
2962 	/*
2963 	 * When EPT AD bits are disabled, all accesses to guest paging
2964 	 * structures are reported separately as a read and (after
2965 	 * translation of the GPA to host physical address) a read+write
2966 	 * if the A/D bits have to be set.
2967 	 */
2968 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2969 
2970 	ept_access_test_setup();
2971 	ept_disable_ad_bits();
2972 
2973 	/* Can't update A bit, so all accesses fail. */
2974 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2975 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
2976 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
2977 	/* AD bits disabled, so only writes try to update the D bit. */
2978 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
2979 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
2980 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
2981 	/* Both A and D already set, so read-only is OK. */
2982 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
2983 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
2984 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
2985 }
2986 
ept_access_test_paddr_read_only_ad_enabled(void)2987 static void ept_access_test_paddr_read_only_ad_enabled(void)
2988 {
2989 	/*
2990 	 * When EPT AD bits are enabled, all accesses to guest paging
2991 	 * structures are considered writes as far as EPT translation
2992 	 * is concerned.
2993 	 */
2994 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2995 
2996 	ept_access_test_setup();
2997 	ept_enable_ad_bits_or_skip_test();
2998 
2999 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3000 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3001 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3002 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
3003 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3004 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
3005 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
3006 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
3007 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
3008 }
3009 
ept_access_test_paddr_read_write(void)3010 static void ept_access_test_paddr_read_write(void)
3011 {
3012 	ept_access_test_setup();
3013 	/* Read-write access to paging structure. */
3014 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
3015 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
3016 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
3017 }
3018 
ept_access_test_paddr_read_write_execute(void)3019 static void ept_access_test_paddr_read_write_execute(void)
3020 {
3021 	ept_access_test_setup();
3022 	/* RWX access to paging structure. */
3023 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
3024 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
3025 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
3026 }
3027 
ept_access_test_paddr_read_execute_ad_disabled(void)3028 static void ept_access_test_paddr_read_execute_ad_disabled(void)
3029 {
3030   	/*
3031 	 * When EPT AD bits are disabled, all accesses to guest paging
3032 	 * structures are reported separately as a read and (after
3033 	 * translation of the GPA to host physical address) a read+write
3034 	 * if the A/D bits have to be set.
3035 	 */
3036 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3037 
3038 	ept_access_test_setup();
3039 	ept_disable_ad_bits();
3040 
3041 	/* Can't update A bit, so all accesses fail. */
3042 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3043 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3044 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3045 	/* AD bits disabled, so only writes try to update the D bit. */
3046 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
3047 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3048 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3049 	/* Both A and D already set, so read-only is OK. */
3050 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3051 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3052 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3053 }
3054 
ept_access_test_paddr_read_execute_ad_enabled(void)3055 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3056 {
3057 	/*
3058 	 * When EPT AD bits are enabled, all accesses to guest paging
3059 	 * structures are considered writes as far as EPT translation
3060 	 * is concerned.
3061 	 */
3062 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3063 
3064 	ept_access_test_setup();
3065 	ept_enable_ad_bits_or_skip_test();
3066 
3067 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3068 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3069 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3070 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3071 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3072 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3073 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3074 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3075 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3076 }
3077 
ept_access_test_paddr_not_present_page_fault(void)3078 static void ept_access_test_paddr_not_present_page_fault(void)
3079 {
3080 	ept_access_test_setup();
3081 	/*
3082 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3083 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3084 	 * Thus guest page fault before host takes EPT violation for trying to
3085 	 * update A bit.
3086 	 */
3087 }
3088 
ept_access_test_force_2m_page(void)3089 static void ept_access_test_force_2m_page(void)
3090 {
3091 	ept_access_test_setup();
3092 
3093 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3094 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3095 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3096 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3097 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3098 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3099 }
3100 
invvpid_valid(u64 type,u64 vpid,u64 gla)3101 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3102 {
3103 	if (!is_invvpid_type_supported(type))
3104 		return false;
3105 
3106 	if (vpid >> 16)
3107 		return false;
3108 
3109 	if (type != INVVPID_ALL && !vpid)
3110 		return false;
3111 
3112 	if (type == INVVPID_ADDR && !is_canonical(gla))
3113 		return false;
3114 
3115 	return true;
3116 }
3117 
try_invvpid(u64 type,u64 vpid,u64 gla)3118 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3119 {
3120 	int rc;
3121 	bool valid = invvpid_valid(type, vpid, gla);
3122 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3123 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3124 	/*
3125 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3126 	 * that we can tell if it is updated by INVVPID.
3127 	 */
3128 	vmcs_read(~0);
3129 	rc = __invvpid(type, vpid, gla);
3130 	report(!rc == valid, "INVVPID type %ld VPID %lx GLA %lx %s", type,
3131 	       vpid, gla,
3132 	       valid ? "passes" : "fails");
3133 	report(vmcs_read(VMX_INST_ERROR) == expected,
3134 	       "After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3135 	       rc ? "failed" : "successful",
3136 	       expected, vmcs_read(VMX_INST_ERROR));
3137 }
3138 
get_first_supported_invvpid_type(void)3139 static inline unsigned long get_first_supported_invvpid_type(void)
3140 {
3141 	u64 type = ffs(ept_vpid.val >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3142 
3143 	__TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3144 	return type;
3145 }
3146 
ds_invvpid(void * data)3147 static void ds_invvpid(void *data)
3148 {
3149 	asm volatile("invvpid %0, %1"
3150 		     :
3151 		     : "m"(*(struct invvpid_operand *)data),
3152 		       "r"(get_first_supported_invvpid_type()));
3153 }
3154 
3155 /*
3156  * The SS override is ignored in 64-bit mode, so we use an addressing
3157  * mode with %rsp as the base register to generate an implicit SS
3158  * reference.
3159  */
ss_invvpid(void * data)3160 static void ss_invvpid(void *data)
3161 {
3162 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3163 		     : "+r"(data)
3164 		     : "r"(get_first_supported_invvpid_type()));
3165 }
3166 
invvpid_test_gp(void)3167 static void invvpid_test_gp(void)
3168 {
3169 	bool fault;
3170 
3171 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3172 				   (void *)NONCANONICAL);
3173 	report(fault, "INVVPID with non-canonical DS operand raises #GP");
3174 }
3175 
invvpid_test_ss(void)3176 static void invvpid_test_ss(void)
3177 {
3178 	bool fault;
3179 
3180 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3181 				   (void *)NONCANONICAL);
3182 	report(fault, "INVVPID with non-canonical SS operand raises #SS");
3183 }
3184 
invvpid_test_pf(void)3185 static void invvpid_test_pf(void)
3186 {
3187 	void *vpage = alloc_vpage();
3188 	bool fault;
3189 
3190 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3191 	report(fault, "INVVPID with unmapped operand raises #PF");
3192 }
3193 
try_compat_invvpid(void * unused)3194 static void try_compat_invvpid(void *unused)
3195 {
3196 	struct far_pointer32 fp = {
3197 		.offset = (uintptr_t)&&invvpid,
3198 		.selector = KERNEL_CS32,
3199 	};
3200 	uintptr_t rsp;
3201 
3202 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
3203 
3204 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3205 			"Code address too high.");
3206 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3207 
3208 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3209 	return;
3210 invvpid:
3211 	asm volatile (".code32;"
3212 		      "invvpid (%eax), %eax;"
3213 		      "lret;"
3214 		      ".code64");
3215 	__builtin_unreachable();
3216 }
3217 
invvpid_test_compatibility_mode(void)3218 static void invvpid_test_compatibility_mode(void)
3219 {
3220 	bool fault;
3221 
3222 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3223 	report(fault, "Compatibility mode INVVPID raises #UD");
3224 }
3225 
invvpid_test_not_in_vmx_operation(void)3226 static void invvpid_test_not_in_vmx_operation(void)
3227 {
3228 	bool fault;
3229 
3230 	TEST_ASSERT(!vmx_off());
3231 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3232 	report(fault, "INVVPID outside of VMX operation raises #UD");
3233 	TEST_ASSERT(!vmx_on());
3234 }
3235 
3236 /* LAM doesn't apply to the linear address inside the descriptor of invvpid */
invvpid_test_lam(void)3237 static void invvpid_test_lam(void)
3238 {
3239 	void *vaddr;
3240 	struct invvpid_operand *operand;
3241 	u64 lam_mask;
3242 	bool fault;
3243 
3244 	if (!this_cpu_has(X86_FEATURE_LAM)) {
3245 		report_skip("LAM is not supported, skip INVVPID with LAM");
3246 		return;
3247 	}
3248 
3249 	write_cr4(read_cr4() | X86_CR4_LAM_SUP);
3250 	lam_mask = is_la57_enabled() ? LAM57_MASK : LAM48_MASK;
3251 
3252 	vaddr = alloc_vpage();
3253 	install_page(current_page_table(), virt_to_phys(alloc_page()), vaddr);
3254 	/*
3255 	 * Since the stack memory address in KUT doesn't follow kernel address
3256 	 * space partition rule, reuse the memory address for descriptor and
3257 	 * the target address in the descriptor of invvpid.
3258 	 */
3259 	operand = (struct invvpid_operand *)vaddr;
3260 	operand->vpid = 0xffff;
3261 	operand->gla = (u64)vaddr;
3262 	operand = (struct invvpid_operand *)get_non_canonical((u64)operand, lam_mask);
3263 	fault = test_for_exception(GP_VECTOR, ds_invvpid, operand);
3264 	report(!fault, "Expected INVVPID with tagged operand when LAM is enabled to succeed");
3265 
3266 	/*
3267 	 * Verify that LAM doesn't apply to the address inside the descriptor
3268 	 * even when LAM is enabled. i.e., the address in the descriptor should
3269 	 * be canonical.
3270 	 */
3271 	try_invvpid(INVVPID_ADDR, 0xffff, (u64)operand);
3272 
3273 	write_cr4(read_cr4() & ~X86_CR4_LAM_SUP);
3274 }
3275 
3276 /*
3277  * This does not test real-address mode, virtual-8086 mode, protected mode,
3278  * or CPL > 0.
3279  */
invvpid_test(void)3280 static void invvpid_test(void)
3281 {
3282 	int i;
3283 	unsigned types = 0;
3284 	unsigned type;
3285 
3286 	if (!is_vpid_supported())
3287 		test_skip("VPID not supported");
3288 
3289 	if (!is_invvpid_supported())
3290 		test_skip("INVVPID not supported.\n");
3291 
3292 	if (is_invvpid_type_supported(INVVPID_ADDR))
3293 		types |= 1u << INVVPID_ADDR;
3294 	if (is_invvpid_type_supported(INVVPID_CONTEXT_GLOBAL))
3295 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3296 	if (is_invvpid_type_supported(INVVPID_ALL))
3297 		types |= 1u << INVVPID_ALL;
3298 	if (is_invvpid_type_supported(INVVPID_CONTEXT_LOCAL))
3299 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3300 
3301 	if (!types)
3302 		test_skip("No INVVPID types supported.\n");
3303 
3304 	for (i = -127; i < 128; i++)
3305 		try_invvpid(i, 0xffff, 0);
3306 
3307 	/*
3308 	 * VPID must not be more than 16 bits.
3309 	 */
3310 	for (i = 0; i < 64; i++)
3311 		for (type = 0; type < 4; type++)
3312 			if (types & (1u << type))
3313 				try_invvpid(type, 1ul << i, 0);
3314 
3315 	/*
3316 	 * VPID must not be zero, except for "all contexts."
3317 	 */
3318 	for (type = 0; type < 4; type++)
3319 		if (types & (1u << type))
3320 			try_invvpid(type, 0, 0);
3321 
3322 	/*
3323 	 * The gla operand is only validated for single-address INVVPID.
3324 	 */
3325 	if (types & (1u << INVVPID_ADDR)) {
3326 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3327 		invvpid_test_lam();
3328 	}
3329 
3330 	invvpid_test_gp();
3331 	invvpid_test_ss();
3332 	invvpid_test_pf();
3333 	invvpid_test_compatibility_mode();
3334 	invvpid_test_not_in_vmx_operation();
3335 }
3336 
test_assert_vmlaunch_inst_error(u32 expected_error)3337 static void test_assert_vmlaunch_inst_error(u32 expected_error)
3338 {
3339 	u32 vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3340 
3341 	report(vmx_inst_err == expected_error,
3342 	       "VMX inst error is %d (actual %d)", expected_error, vmx_inst_err);
3343 }
3344 
3345 /*
3346  * This version is wildly unsafe and should _only_ be used to test VM-Fail
3347  * scenarios involving HOST_RIP.
3348  */
test_vmx_vmlaunch_must_fail(u32 expected_error)3349 static void test_vmx_vmlaunch_must_fail(u32 expected_error)
3350 {
3351 	/* Read the function name. */
3352 	TEST_ASSERT(expected_error);
3353 
3354 	/*
3355 	 * Don't bother with any prep work, if VMLAUNCH passes the VM-Fail
3356 	 * consistency checks and generates a VM-Exit, then the test is doomed
3357 	 * no matter what as it will jump to a garbage RIP.
3358 	 */
3359 	__asm__ __volatile__ ("vmlaunch");
3360 	test_assert_vmlaunch_inst_error(expected_error);
3361 }
3362 
3363 /*
3364  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3365  * at least as far as the guest-state checks. Returns false if the
3366  * VMLAUNCH fails early and execution falls through to the next
3367  * instruction.
3368  */
vmlaunch(void)3369 static bool vmlaunch(void)
3370 {
3371 	u32 exit_reason;
3372 
3373 	/*
3374 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3375 	 * unsupported VMCS component"). The caller can then check
3376 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3377 	 */
3378 	vmcs_write(~0u, 0);
3379 
3380 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3381 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3382 				   :
3383 				   : "r" ((u64)HOST_RSP)
3384 				   : "cc", "memory"
3385 				   : success);
3386 	return false;
3387 success:
3388 	exit_reason = vmcs_read(EXI_REASON);
3389 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3390 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3391 	return true;
3392 }
3393 
3394 /*
3395  * Try to launch the current VMCS.
3396  */
test_vmx_vmlaunch(u32 xerror)3397 static void test_vmx_vmlaunch(u32 xerror)
3398 {
3399 	bool success = vmlaunch();
3400 
3401 	report(success == !xerror, "vmlaunch %s",
3402 	       !xerror ? "succeeds" : "fails");
3403 	if (!success && xerror)
3404 		test_assert_vmlaunch_inst_error(xerror);
3405 }
3406 
3407 /*
3408  * Try to launch the current VMCS, and expect one of two possible
3409  * errors (or success) codes.
3410  */
test_vmx_vmlaunch2(u32 xerror1,u32 xerror2)3411 static void test_vmx_vmlaunch2(u32 xerror1, u32 xerror2)
3412 {
3413 	bool success = vmlaunch();
3414 	u32 vmx_inst_err;
3415 
3416 	if (!xerror1 == !xerror2)
3417 		report(success == !xerror1, "vmlaunch %s",
3418 		       !xerror1 ? "succeeds" : "fails");
3419 
3420 	if (!success && (xerror1 || xerror2)) {
3421 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3422 		report(vmx_inst_err == xerror1 || vmx_inst_err == xerror2,
3423 		       "VMX inst error is %d or %d (actual %d)", xerror1,
3424 		       xerror2, vmx_inst_err);
3425 	}
3426 }
3427 
test_vmx_invalid_controls(void)3428 static void test_vmx_invalid_controls(void)
3429 {
3430 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3431 }
3432 
test_vmx_valid_controls(void)3433 static void test_vmx_valid_controls(void)
3434 {
3435 	test_vmx_vmlaunch(0);
3436 }
3437 
3438 /*
3439  * Test a particular value of a VM-execution control bit, if the value
3440  * is required or if the value is zero.
3441  */
test_rsvd_ctl_bit_value(const char * name,union vmx_ctrl_msr msr,enum Encoding encoding,unsigned bit,unsigned val)3442 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3443 				    enum Encoding encoding, unsigned bit,
3444 				    unsigned val)
3445 {
3446 	u32 mask = 1u << bit;
3447 	bool expected;
3448 	u32 controls;
3449 
3450 	if (msr.set & mask)
3451 		TEST_ASSERT(msr.clr & mask);
3452 
3453 	/*
3454 	 * We can't arbitrarily turn on a control bit, because it may
3455 	 * introduce dependencies on other VMCS fields. So, we only
3456 	 * test turning on bits that have a required setting.
3457 	 */
3458 	if (val && (msr.clr & mask) && !(msr.set & mask))
3459 		return;
3460 
3461 	report_prefix_pushf("%s %s bit %d",
3462 			    val ? "Set" : "Clear", name, bit);
3463 
3464 	controls = vmcs_read(encoding);
3465 	if (val) {
3466 		vmcs_write(encoding, msr.set | mask);
3467 		expected = (msr.clr & mask);
3468 	} else {
3469 		vmcs_write(encoding, msr.set & ~mask);
3470 		expected = !(msr.set & mask);
3471 	}
3472 	if (expected)
3473 		test_vmx_valid_controls();
3474 	else
3475 		test_vmx_invalid_controls();
3476 	vmcs_write(encoding, controls);
3477 	report_prefix_pop();
3478 }
3479 
3480 /*
3481  * Test reserved values of a VM-execution control bit, based on the
3482  * allowed bit settings from the corresponding VMX capability MSR.
3483  */
test_rsvd_ctl_bit(const char * name,union vmx_ctrl_msr msr,enum Encoding encoding,unsigned bit)3484 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3485 			      enum Encoding encoding, unsigned bit)
3486 {
3487 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3488 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3489 }
3490 
3491 /*
3492  * Reserved bits in the pin-based VM-execution controls must be set
3493  * properly. Software may consult the VMX capability MSRs to determine
3494  * the proper settings.
3495  * [Intel SDM]
3496  */
test_pin_based_ctls(void)3497 static void test_pin_based_ctls(void)
3498 {
3499 	unsigned bit;
3500 
3501 	printf("%s: %lx\n", basic_msr.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3502 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3503 	for (bit = 0; bit < 32; bit++)
3504 		test_rsvd_ctl_bit("pin-based controls",
3505 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3506 }
3507 
3508 /*
3509  * Reserved bits in the primary processor-based VM-execution controls
3510  * must be set properly. Software may consult the VMX capability MSRs
3511  * to determine the proper settings.
3512  * [Intel SDM]
3513  */
test_primary_processor_based_ctls(void)3514 static void test_primary_processor_based_ctls(void)
3515 {
3516 	unsigned bit;
3517 
3518 	printf("\n%s: %lx\n", basic_msr.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3519 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3520 	for (bit = 0; bit < 32; bit++)
3521 		test_rsvd_ctl_bit("primary processor-based controls",
3522 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3523 }
3524 
3525 /*
3526  * If the "activate secondary controls" primary processor-based
3527  * VM-execution control is 1, reserved bits in the secondary
3528  * processor-based VM-execution controls must be cleared. Software may
3529  * consult the VMX capability MSRs to determine which bits are
3530  * reserved.
3531  * If the "activate secondary controls" primary processor-based
3532  * VM-execution control is 0 (or if the processor does not support the
3533  * 1-setting of that control), no checks are performed on the
3534  * secondary processor-based VM-execution controls.
3535  * [Intel SDM]
3536  */
test_secondary_processor_based_ctls(void)3537 static void test_secondary_processor_based_ctls(void)
3538 {
3539 	u32 primary;
3540 	u32 secondary;
3541 	unsigned bit;
3542 
3543 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3544 		return;
3545 
3546 	primary = vmcs_read(CPU_EXEC_CTRL0);
3547 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3548 
3549 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3550 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3551 	for (bit = 0; bit < 32; bit++)
3552 		test_rsvd_ctl_bit("secondary processor-based controls",
3553 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3554 
3555 	/*
3556 	 * When the "activate secondary controls" VM-execution control
3557 	 * is clear, there are no checks on the secondary controls.
3558 	 */
3559 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3560 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3561 	report(vmlaunch(),
3562 	       "Secondary processor-based controls ignored");
3563 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3564 	vmcs_write(CPU_EXEC_CTRL0, primary);
3565 }
3566 
try_cr3_target_count(unsigned i,unsigned max)3567 static void try_cr3_target_count(unsigned i, unsigned max)
3568 {
3569 	report_prefix_pushf("CR3 target count 0x%x", i);
3570 	vmcs_write(CR3_TARGET_COUNT, i);
3571 	if (i <= max)
3572 		test_vmx_valid_controls();
3573 	else
3574 		test_vmx_invalid_controls();
3575 	report_prefix_pop();
3576 }
3577 
3578 /*
3579  * The CR3-target count must not be greater than 4. Future processors
3580  * may support a different number of CR3-target values. Software
3581  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3582  * number of values supported.
3583  * [Intel SDM]
3584  */
test_cr3_targets(void)3585 static void test_cr3_targets(void)
3586 {
3587 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3588 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3589 	unsigned i;
3590 
3591 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3592 	TEST_ASSERT(supported_targets <= 256);
3593 
3594 	try_cr3_target_count(-1u, supported_targets);
3595 	try_cr3_target_count(0x80000000, supported_targets);
3596 	try_cr3_target_count(0x7fffffff, supported_targets);
3597 	for (i = 0; i <= supported_targets + 1; i++)
3598 		try_cr3_target_count(i, supported_targets);
3599 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3600 
3601 	/* VMWRITE to nonexistent target fields should fail. */
3602 	for (i = supported_targets; i < 256; i++)
3603 		TEST_ASSERT(vmcs_write(CR3_TARGET_0 + i*2, 0));
3604 }
3605 
3606 /*
3607  * Test a particular address setting in the VMCS
3608  */
test_vmcs_addr(const char * name,enum Encoding encoding,u64 align,bool ignored,bool skip_beyond_mapped_ram,u64 addr)3609 static void test_vmcs_addr(const char *name,
3610 			   enum Encoding encoding,
3611 			   u64 align,
3612 			   bool ignored,
3613 			   bool skip_beyond_mapped_ram,
3614 			   u64 addr)
3615 {
3616 	report_prefix_pushf("%s = %lx", name, addr);
3617 	vmcs_write(encoding, addr);
3618 	if (skip_beyond_mapped_ram &&
3619 	    addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3620 	    addr < (1ul << cpuid_maxphyaddr()))
3621 		printf("Skipping physical address beyond mapped RAM\n");
3622 	else if (ignored || (IS_ALIGNED(addr, align) &&
3623 	    addr < (1ul << cpuid_maxphyaddr())))
3624 		test_vmx_valid_controls();
3625 	else
3626 		test_vmx_invalid_controls();
3627 	report_prefix_pop();
3628 }
3629 
3630 /*
3631  * Test interesting values for a VMCS address
3632  */
test_vmcs_addr_values(const char * name,enum Encoding encoding,u64 align,bool ignored,bool skip_beyond_mapped_ram,u32 bit_start,u32 bit_end)3633 static void test_vmcs_addr_values(const char *name,
3634 				  enum Encoding encoding,
3635 				  u64 align,
3636 				  bool ignored,
3637 				  bool skip_beyond_mapped_ram,
3638 				  u32 bit_start, u32 bit_end)
3639 {
3640 	unsigned i;
3641 	u64 orig_val = vmcs_read(encoding);
3642 
3643 	for (i = bit_start; i <= bit_end; i++)
3644 		test_vmcs_addr(name, encoding, align, ignored,
3645 			       skip_beyond_mapped_ram, 1ul << i);
3646 
3647 	test_vmcs_addr(name, encoding, align, ignored,
3648 		       skip_beyond_mapped_ram, PAGE_SIZE - 1);
3649 	test_vmcs_addr(name, encoding, align, ignored,
3650 		       skip_beyond_mapped_ram, PAGE_SIZE);
3651 	test_vmcs_addr(name, encoding, align, ignored,
3652 		       skip_beyond_mapped_ram,
3653 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3654 	test_vmcs_addr(name, encoding, align, ignored,
3655 		       skip_beyond_mapped_ram, -1ul);
3656 
3657 	vmcs_write(encoding, orig_val);
3658 }
3659 
3660 /*
3661  * Test a physical address reference in the VMCS, when the corresponding
3662  * feature is enabled and when the corresponding feature is disabled.
3663  */
test_vmcs_addr_reference(u32 control_bit,enum Encoding field,const char * field_name,const char * control_name,u64 align,bool skip_beyond_mapped_ram,bool control_primary)3664 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3665 				     const char *field_name,
3666 				     const char *control_name, u64 align,
3667 				     bool skip_beyond_mapped_ram,
3668 				     bool control_primary)
3669 {
3670 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3671 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3672 	u64 page_addr;
3673 
3674 	if (control_primary) {
3675 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3676 			return;
3677 	} else {
3678 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3679 			return;
3680 	}
3681 
3682 	page_addr = vmcs_read(field);
3683 
3684 	report_prefix_pushf("%s enabled", control_name);
3685 	if (control_primary) {
3686 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3687 	} else {
3688 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3689 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3690 	}
3691 
3692 	test_vmcs_addr_values(field_name, field, align, false,
3693 			      skip_beyond_mapped_ram, 0, 63);
3694 	report_prefix_pop();
3695 
3696 	report_prefix_pushf("%s disabled", control_name);
3697 	if (control_primary) {
3698 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3699 	} else {
3700 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3701 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3702 	}
3703 
3704 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3705 	report_prefix_pop();
3706 
3707 	vmcs_write(field, page_addr);
3708 	vmcs_write(CPU_EXEC_CTRL0, primary);
3709 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3710 }
3711 
3712 /*
3713  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3714  * each I/O-bitmap address must be 0. Neither address should set any
3715  * bits beyond the processor's physical-address width.
3716  * [Intel SDM]
3717  */
test_io_bitmaps(void)3718 static void test_io_bitmaps(void)
3719 {
3720 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3721 				 "I/O bitmap A", "Use I/O bitmaps",
3722 				 PAGE_SIZE, false, true);
3723 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3724 				 "I/O bitmap B", "Use I/O bitmaps",
3725 				 PAGE_SIZE, false, true);
3726 }
3727 
3728 /*
3729  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3730  * the MSR-bitmap address must be 0. The address should not set any
3731  * bits beyond the processor's physical-address width.
3732  * [Intel SDM]
3733  */
test_msr_bitmap(void)3734 static void test_msr_bitmap(void)
3735 {
3736 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3737 				 "MSR bitmap", "Use MSR bitmaps",
3738 				 PAGE_SIZE, false, true);
3739 }
3740 
3741 /*
3742  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3743  * address must satisfy the following checks:
3744  * - Bits 11:0 of the address must be 0.
3745  * - The address should not set any bits beyond the processor's
3746  *   physical-address width.
3747  * [Intel SDM]
3748  */
test_apic_virt_addr(void)3749 static void test_apic_virt_addr(void)
3750 {
3751 	/*
3752 	 * Ensure the processor will never use the virtual-APIC page, since
3753 	 * we will point it to invalid RAM.  Otherwise KVM is puzzled about
3754 	 * what we're trying to achieve and fails vmentry.
3755 	 */
3756 	u32 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
3757 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0 | CPU_CR8_LOAD | CPU_CR8_STORE);
3758 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3759 				 "virtual-APIC address", "Use TPR shadow",
3760 				 PAGE_SIZE, false, true);
3761 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
3762 }
3763 
3764 /*
3765  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3766  * APIC-access address must satisfy the following checks:
3767  *  - Bits 11:0 of the address must be 0.
3768  *  - The address should not set any bits beyond the processor's
3769  *    physical-address width.
3770  * [Intel SDM]
3771  */
test_apic_access_addr(void)3772 static void test_apic_access_addr(void)
3773 {
3774 	void *apic_access_page = alloc_page();
3775 
3776 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3777 
3778 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3779 				 "APIC-access address",
3780 				 "virtualize APIC-accesses", PAGE_SIZE,
3781 				 true, false);
3782 }
3783 
set_bit_pattern(u8 mask,u32 * secondary)3784 static bool set_bit_pattern(u8 mask, u32 *secondary)
3785 {
3786 	u8 i;
3787 	bool flag = false;
3788 	u32 test_bits[3] = {
3789 		CPU_VIRT_X2APIC,
3790 		CPU_APIC_REG_VIRT,
3791 		CPU_VINTD
3792 	};
3793 
3794         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3795 		if ((mask & (1u << i)) &&
3796 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3797 			*secondary |= test_bits[i];
3798 			flag = true;
3799 		}
3800 	}
3801 
3802 	return (flag);
3803 }
3804 
3805 /*
3806  * If the "use TPR shadow" VM-execution control is 0, the following
3807  * VM-execution controls must also be 0:
3808  * 	- virtualize x2APIC mode
3809  *	- APIC-register virtualization
3810  *	- virtual-interrupt delivery
3811  *    [Intel SDM]
3812  *
3813  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3814  *    "virtualize APIC accesses" VM-execution control must be 0.
3815  *    [Intel SDM]
3816  */
test_apic_virtual_ctls(void)3817 static void test_apic_virtual_ctls(void)
3818 {
3819 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3820 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3821 	u32 primary = saved_primary;
3822 	u32 secondary = saved_secondary;
3823 	bool is_ctrl_valid = false;
3824 	char str[10] = "disabled";
3825 	u8 i = 0, j;
3826 
3827 	/*
3828 	 * First test
3829 	 */
3830 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3831 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3832 		return;
3833 
3834 	primary |= CPU_SECONDARY;
3835 	primary &= ~CPU_TPR_SHADOW;
3836 	vmcs_write(CPU_EXEC_CTRL0, primary);
3837 
3838 	while (1) {
3839 		for (j = 1; j < 8; j++) {
3840 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3841 			if (primary & CPU_TPR_SHADOW) {
3842 				is_ctrl_valid = true;
3843 			} else {
3844 				if (! set_bit_pattern(j, &secondary))
3845 					is_ctrl_valid = true;
3846 				else
3847 					is_ctrl_valid = false;
3848 			}
3849 
3850 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3851 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3852 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3853 			if (is_ctrl_valid)
3854 				test_vmx_valid_controls();
3855 			else
3856 				test_vmx_invalid_controls();
3857 			report_prefix_pop();
3858 		}
3859 
3860 		if (i == 1)
3861 			break;
3862 		i++;
3863 
3864 		primary |= CPU_TPR_SHADOW;
3865 		vmcs_write(CPU_EXEC_CTRL0, primary);
3866 		strcpy(str, "enabled");
3867 	}
3868 
3869 	/*
3870 	 * Second test
3871 	 */
3872 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3873 
3874 	primary = saved_primary;
3875 	secondary = saved_secondary;
3876 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3877 		return;
3878 
3879 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3880 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3881 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3882 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3883 	test_vmx_valid_controls();
3884 	report_prefix_pop();
3885 
3886 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3887 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3888 	test_vmx_valid_controls();
3889 	report_prefix_pop();
3890 
3891 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3892 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3893 	test_vmx_invalid_controls();
3894 	report_prefix_pop();
3895 
3896 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3897 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3898 	test_vmx_valid_controls();
3899 	report_prefix_pop();
3900 
3901 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3902 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3903 }
3904 
3905 /*
3906  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3907  * "external-interrupt exiting" VM-execution control must be 1.
3908  * [Intel SDM]
3909  */
test_virtual_intr_ctls(void)3910 static void test_virtual_intr_ctls(void)
3911 {
3912 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3913 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3914 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3915 	u32 primary = saved_primary;
3916 	u32 secondary = saved_secondary;
3917 	u32 pin = saved_pin;
3918 
3919 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3920 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
3921 		return;
3922 
3923 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3924 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
3925 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3926 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
3927 	test_vmx_valid_controls();
3928 	report_prefix_pop();
3929 
3930 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
3931 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3932 	test_vmx_invalid_controls();
3933 	report_prefix_pop();
3934 
3935 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
3936 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
3937 	test_vmx_valid_controls();
3938 	report_prefix_pop();
3939 
3940 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3941 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3942 	test_vmx_invalid_controls();
3943 	report_prefix_pop();
3944 
3945 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3946 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3947 	vmcs_write(PIN_CONTROLS, saved_pin);
3948 }
3949 
test_pi_desc_addr(u64 addr,bool is_ctrl_valid)3950 static void test_pi_desc_addr(u64 addr, bool is_ctrl_valid)
3951 {
3952 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
3953 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
3954 	if (is_ctrl_valid)
3955 		test_vmx_valid_controls();
3956 	else
3957 		test_vmx_invalid_controls();
3958 	report_prefix_pop();
3959 }
3960 
3961 /*
3962  * If the "process posted interrupts" VM-execution control is 1, the
3963  * following must be true:
3964  *
3965  *	- The "virtual-interrupt delivery" VM-execution control is 1.
3966  *	- The "acknowledge interrupt on exit" VM-exit control is 1.
3967  *	- The posted-interrupt notification vector has a value in the
3968  *	- range 0 - 255 (bits 15:8 are all 0).
3969  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
3970  *	- The posted-interrupt descriptor address does not set any bits
3971  *	  beyond the processor's physical-address width.
3972  * [Intel SDM]
3973  */
test_posted_intr(void)3974 static void test_posted_intr(void)
3975 {
3976 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3977 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3978 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3979 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
3980 	u32 primary = saved_primary;
3981 	u32 secondary = saved_secondary;
3982 	u32 pin = saved_pin;
3983 	u32 exit_ctl = exit_ctl_saved;
3984 	u16 vec;
3985 	int i;
3986 
3987 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
3988 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3989 	    (ctrl_exit_rev.clr & EXI_INTA)))
3990 		return;
3991 
3992 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3993 
3994 	/*
3995 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
3996 	 */
3997 	pin |= PIN_POST_INTR;
3998 	vmcs_write(PIN_CONTROLS, pin);
3999 	secondary &= ~CPU_VINTD;
4000 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4001 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
4002 	test_vmx_invalid_controls();
4003 	report_prefix_pop();
4004 
4005 	secondary |= CPU_VINTD;
4006 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4007 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
4008 	test_vmx_invalid_controls();
4009 	report_prefix_pop();
4010 
4011 	exit_ctl &= ~EXI_INTA;
4012 	vmcs_write(EXI_CONTROLS, exit_ctl);
4013 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
4014 	test_vmx_invalid_controls();
4015 	report_prefix_pop();
4016 
4017 	exit_ctl |= EXI_INTA;
4018 	vmcs_write(EXI_CONTROLS, exit_ctl);
4019 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4020 	test_vmx_valid_controls();
4021 	report_prefix_pop();
4022 
4023 	secondary &= ~CPU_VINTD;
4024 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4025 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
4026 	test_vmx_invalid_controls();
4027 	report_prefix_pop();
4028 
4029 	secondary |= CPU_VINTD;
4030 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4031 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4032 	test_vmx_valid_controls();
4033 	report_prefix_pop();
4034 
4035 	/*
4036 	 * Test posted-interrupt notification vector
4037 	 */
4038 	for (i = 0; i < 8; i++) {
4039 		vec = (1ul << i);
4040 		vmcs_write(PINV, vec);
4041 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4042 		test_vmx_valid_controls();
4043 		report_prefix_pop();
4044 	}
4045 	for (i = 8; i < 16; i++) {
4046 		vec = (1ul << i);
4047 		vmcs_write(PINV, vec);
4048 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4049 		test_vmx_invalid_controls();
4050 		report_prefix_pop();
4051 	}
4052 
4053 	vec &= ~(0xff << 8);
4054 	vmcs_write(PINV, vec);
4055 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4056 	test_vmx_valid_controls();
4057 	report_prefix_pop();
4058 
4059 	/*
4060 	 * Test posted-interrupt descriptor address
4061 	 */
4062 	for (i = 0; i < 6; i++) {
4063 		test_pi_desc_addr(1ul << i, false);
4064 	}
4065 
4066 	test_pi_desc_addr(0xf0, false);
4067 	test_pi_desc_addr(0xff, false);
4068 	test_pi_desc_addr(0x0f, false);
4069 	test_pi_desc_addr(0x8000, true);
4070 	test_pi_desc_addr(0x00, true);
4071 	test_pi_desc_addr(0xc000, true);
4072 
4073 	test_vmcs_addr_values("process-posted interrupts",
4074 			       POSTED_INTR_DESC_ADDR, 64,
4075 			       false, false, 0, 63);
4076 
4077 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4078 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4079 	vmcs_write(PIN_CONTROLS, saved_pin);
4080 }
4081 
test_apic_ctls(void)4082 static void test_apic_ctls(void)
4083 {
4084 	test_apic_virt_addr();
4085 	test_apic_access_addr();
4086 	test_apic_virtual_ctls();
4087 	test_virtual_intr_ctls();
4088 	test_posted_intr();
4089 }
4090 
4091 /*
4092  * If the "enable VPID" VM-execution control is 1, the value of the
4093  * of the VPID VM-execution control field must not be 0000H.
4094  * [Intel SDM]
4095  */
test_vpid(void)4096 static void test_vpid(void)
4097 {
4098 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4099 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4100 	u16 vpid = 0x0000;
4101 	int i;
4102 
4103 	if (!is_vpid_supported()) {
4104 		report_skip("%s : Secondary controls and/or VPID not supported", __func__);
4105 		return;
4106 	}
4107 
4108 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
4109 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
4110 	vmcs_write(VPID, vpid);
4111 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
4112 	test_vmx_valid_controls();
4113 	report_prefix_pop();
4114 
4115 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
4116 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4117 	test_vmx_invalid_controls();
4118 	report_prefix_pop();
4119 
4120 	for (i = 0; i < 16; i++) {
4121 		vpid = (short)1 << i;;
4122 		vmcs_write(VPID, vpid);
4123 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4124 		test_vmx_valid_controls();
4125 		report_prefix_pop();
4126 	}
4127 
4128 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4129 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4130 }
4131 
set_vtpr(unsigned vtpr)4132 static void set_vtpr(unsigned vtpr)
4133 {
4134 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
4135 }
4136 
try_tpr_threshold_and_vtpr(unsigned threshold,unsigned vtpr)4137 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
4138 {
4139 	bool valid = true;
4140 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4141 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4142 
4143 	if ((primary & CPU_TPR_SHADOW) &&
4144 	    (!(primary & CPU_SECONDARY) ||
4145 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
4146 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
4147 
4148 	set_vtpr(vtpr);
4149 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
4150 	    threshold, (vtpr >> 4) & 0xf);
4151 	if (valid)
4152 		test_vmx_valid_controls();
4153 	else
4154 		test_vmx_invalid_controls();
4155 	report_prefix_pop();
4156 }
4157 
test_invalid_event_injection(void)4158 static void test_invalid_event_injection(void)
4159 {
4160 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
4161 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
4162 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
4163 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
4164 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
4165 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4166 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4167 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4168 	u32 cnt;
4169 
4170 	/* Setup */
4171 	report_prefix_push("invalid event injection");
4172 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4173 	vmcs_write(ENT_INST_LEN, 0x00000001);
4174 
4175 	/* The field's interruption type is not set to a reserved value. */
4176 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4177 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4178 			    "RESERVED interruption type invalid [-]",
4179 			    ent_intr_info);
4180 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4181 	test_vmx_invalid_controls();
4182 	report_prefix_pop();
4183 
4184 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4185 			DE_VECTOR;
4186 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4187 			    "RESERVED interruption type invalid [+]",
4188 			    ent_intr_info);
4189 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4190 	test_vmx_valid_controls();
4191 	report_prefix_pop();
4192 
4193 	/* If the interruption type is other event, the vector is 0. */
4194 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4195 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4196 			    "(OTHER EVENT && vector != 0) invalid [-]",
4197 			    ent_intr_info);
4198 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4199 	test_vmx_invalid_controls();
4200 	report_prefix_pop();
4201 
4202 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4203 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4204 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4205 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4206 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4207 	test_vmx_invalid_controls();
4208 	report_prefix_pop();
4209 
4210 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4211 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4212 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4213 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4214 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4215 	test_vmx_valid_controls();
4216 	report_prefix_pop();
4217 
4218 	/*
4219 	 * If the interruption type
4220 	 * is HW exception, the vector is at most 31.
4221 	 */
4222 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4223 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4224 			    "(HW exception && vector > 31) invalid [-]",
4225 			    ent_intr_info);
4226 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4227 	test_vmx_invalid_controls();
4228 	report_prefix_pop();
4229 
4230 	/*
4231 	 * deliver-error-code is 1 iff either
4232 	 * (a) the "unrestricted guest" VM-execution control is 0
4233 	 * (b) CR0.PE is set.
4234 	 */
4235 
4236 	/* Assert that unrestricted guest is disabled or unsupported */
4237 	assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
4238 	       !(secondary_save & CPU_URG));
4239 
4240 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4241 			GP_VECTOR;
4242 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4243 			    "error code <-> (!URG || prot_mode) [-]",
4244 			    ent_intr_info);
4245 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4246 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4247 	if (basic_msr.no_hw_errcode_cc)
4248 		test_vmx_valid_controls();
4249 	else
4250 		test_vmx_invalid_controls();
4251 	report_prefix_pop();
4252 
4253 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4254 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4255 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4256 			    "error code <-> (!URG || prot_mode) [+]",
4257 			    ent_intr_info);
4258 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4259 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4260 	test_vmx_valid_controls();
4261 	report_prefix_pop();
4262 
4263 	if (enable_unrestricted_guest(false))
4264 		goto skip_unrestricted_guest;
4265 
4266 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4267 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4268 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4269 			    "error code <-> (!URG || prot_mode) [-]",
4270 			    ent_intr_info);
4271 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4272 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4273 	test_vmx_invalid_controls();
4274 	report_prefix_pop();
4275 
4276 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4277 			GP_VECTOR;
4278 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4279 			    "error code <-> (!URG || prot_mode) [-]",
4280 			    ent_intr_info);
4281 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4282 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4283 	if (basic_msr.no_hw_errcode_cc)
4284 		test_vmx_valid_controls();
4285 	else
4286 		test_vmx_invalid_controls();
4287 	report_prefix_pop();
4288 
4289 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4290 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4291 
4292 skip_unrestricted_guest:
4293 	vmcs_write(GUEST_CR0, guest_cr0_save);
4294 
4295 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4296 	report_prefix_push("error code <-> HW exception");
4297 	for (cnt = 0; cnt < 8; cnt++) {
4298 		u32 exception_type_mask = cnt << 8;
4299 		u32 deliver_error_code_mask =
4300 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4301 			INTR_INFO_DELIVER_CODE_MASK : 0;
4302 
4303 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4304 				exception_type_mask | GP_VECTOR;
4305 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4306 				    ent_intr_info);
4307 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4308 		if (exception_type_mask == INTR_TYPE_HARD_EXCEPTION &&
4309 		    basic_msr.no_hw_errcode_cc)
4310 			test_vmx_valid_controls();
4311 		else
4312 			test_vmx_invalid_controls();
4313 		report_prefix_pop();
4314 	}
4315 	report_prefix_pop();
4316 
4317 	/*
4318 	 * deliver-error-code is 1 iff the the vector
4319 	 * indicates an exception that would normally deliver an error code
4320 	 */
4321 	report_prefix_push("error code <-> vector delivers error code");
4322 	for (cnt = 0; cnt < 32; cnt++) {
4323 		bool has_error_code = false;
4324 		u32 deliver_error_code_mask;
4325 
4326 		switch (cnt) {
4327 		case DF_VECTOR:
4328 		case TS_VECTOR:
4329 		case NP_VECTOR:
4330 		case SS_VECTOR:
4331 		case GP_VECTOR:
4332 		case PF_VECTOR:
4333 		case AC_VECTOR:
4334 			has_error_code = true;
4335 		case CP_VECTOR:
4336 			/* Some CPUs have error code and some do not, skip */
4337 			continue;
4338 		}
4339 
4340 		/* Negative case */
4341 		deliver_error_code_mask = has_error_code ?
4342 						0 :
4343 						INTR_INFO_DELIVER_CODE_MASK;
4344 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4345 				INTR_TYPE_HARD_EXCEPTION | cnt;
4346 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4347 				    ent_intr_info);
4348 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4349 		if (basic_msr.no_hw_errcode_cc)
4350 			test_vmx_valid_controls();
4351 		else
4352 			test_vmx_invalid_controls();
4353 		report_prefix_pop();
4354 
4355 		/* Positive case */
4356 		deliver_error_code_mask = has_error_code ?
4357 						INTR_INFO_DELIVER_CODE_MASK :
4358 						0;
4359 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4360 				INTR_TYPE_HARD_EXCEPTION | cnt;
4361 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4362 				    ent_intr_info);
4363 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4364 		test_vmx_valid_controls();
4365 		report_prefix_pop();
4366 	}
4367 	report_prefix_pop();
4368 
4369 	/* Reserved bits in the field (30:12) are 0. */
4370 	report_prefix_push("reserved bits clear");
4371 	for (cnt = 12; cnt <= 30; cnt++) {
4372 		ent_intr_info = ent_intr_info_base |
4373 				INTR_INFO_DELIVER_CODE_MASK |
4374 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4375 				(1U << cnt);
4376 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4377 				    ent_intr_info);
4378 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4379 		test_vmx_invalid_controls();
4380 		report_prefix_pop();
4381 	}
4382 	report_prefix_pop();
4383 
4384 	/*
4385 	 * If deliver-error-code is 1
4386 	 * bits 31:16 of the VM-entry exception error-code field are 0.
4387 	 */
4388 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4389 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4390 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4391 			    "VM-entry exception error code[31:16] clear",
4392 			    ent_intr_info);
4393 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4394 	for (cnt = 16; cnt <= 31; cnt++) {
4395 		ent_intr_err = 1U << cnt;
4396 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4397 				    ent_intr_err);
4398 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4399 		test_vmx_invalid_controls();
4400 		report_prefix_pop();
4401 	}
4402 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4403 	report_prefix_pop();
4404 
4405 	/*
4406 	 * If the interruption type is software interrupt, software exception,
4407 	 * or privileged software exception, the VM-entry instruction-length
4408 	 * field is in the range 0 - 15.
4409 	 */
4410 
4411 	for (cnt = 0; cnt < 3; cnt++) {
4412 		switch (cnt) {
4413 		case 0:
4414 			ent_intr_info = ent_intr_info_base |
4415 					INTR_TYPE_SOFT_INTR;
4416 			break;
4417 		case 1:
4418 			ent_intr_info = ent_intr_info_base |
4419 					INTR_TYPE_SOFT_EXCEPTION;
4420 			break;
4421 		case 2:
4422 			ent_intr_info = ent_intr_info_base |
4423 					INTR_TYPE_PRIV_SW_EXCEPTION;
4424 			break;
4425 		}
4426 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4427 				    "VM-entry instruction-length check",
4428 				    ent_intr_info);
4429 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4430 
4431 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4432 		ent_intr_len = -1;
4433 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4434 				    ent_intr_len);
4435 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4436 		test_vmx_invalid_controls();
4437 		report_prefix_pop();
4438 
4439 		/* Instruction length set to 16 should fail */
4440 		ent_intr_len = 0x00000010;
4441 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4442 				    ent_intr_len);
4443 		vmcs_write(ENT_INST_LEN, 0x00000010);
4444 		test_vmx_invalid_controls();
4445 		report_prefix_pop();
4446 
4447 		report_prefix_pop();
4448 	}
4449 
4450 	/* Cleanup */
4451 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4452 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4453 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4454 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4455 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4456 	vmcs_write(GUEST_CR0, guest_cr0_save);
4457 	report_prefix_pop();
4458 }
4459 
4460 /*
4461  * Test interesting vTPR values for a given TPR threshold.
4462  */
test_vtpr_values(unsigned threshold)4463 static void test_vtpr_values(unsigned threshold)
4464 {
4465 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4466 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4467 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4468 }
4469 
try_tpr_threshold(unsigned threshold)4470 static void try_tpr_threshold(unsigned threshold)
4471 {
4472 	bool valid = true;
4473 
4474 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4475 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4476 
4477 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4478 	    (secondary & CPU_VINTD)))
4479 		valid = !(threshold >> 4);
4480 
4481 	set_vtpr(-1);
4482 	vmcs_write(TPR_THRESHOLD, threshold);
4483 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4484 	if (valid)
4485 		test_vmx_valid_controls();
4486 	else
4487 		test_vmx_invalid_controls();
4488 	report_prefix_pop();
4489 
4490 	if (valid)
4491 		test_vtpr_values(threshold);
4492 }
4493 
4494 /*
4495  * Test interesting TPR threshold values.
4496  */
test_tpr_threshold_values(void)4497 static void test_tpr_threshold_values(void)
4498 {
4499 	unsigned i;
4500 
4501 	for (i = 0; i < 0x10; i++)
4502 		try_tpr_threshold(i);
4503 	for (i = 4; i < 32; i++)
4504 		try_tpr_threshold(1u << i);
4505 	try_tpr_threshold(-1u);
4506 	try_tpr_threshold(0x7fffffff);
4507 }
4508 
4509 /*
4510  * This test covers the following two VM entry checks:
4511  *
4512  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4513  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4514  *         31:4 of the TPR threshold VM-execution control field must
4515 	   be 0.
4516  *         [Intel SDM]
4517  *
4518  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4519  *          "virtual-interrupt delivery" VM-execution control is 0
4520  *          and the "virtualize APIC accesses" VM-execution control
4521  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4522  *          control field must not be greater than the value of bits
4523  *          7:4 of VTPR.
4524  *          [Intel SDM]
4525  */
test_tpr_threshold(void)4526 static void test_tpr_threshold(void)
4527 {
4528 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4529 	u64 apic_virt_addr = vmcs_read(APIC_VIRT_ADDR);
4530 	u64 threshold = vmcs_read(TPR_THRESHOLD);
4531 	void *virtual_apic_page;
4532 
4533 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4534 		return;
4535 
4536 	virtual_apic_page = alloc_page();
4537 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4538 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4539 
4540 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4541 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4542 	test_tpr_threshold_values();
4543 	report_prefix_pop();
4544 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4545 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4546 	test_tpr_threshold_values();
4547 	report_prefix_pop();
4548 
4549 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4550 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES))))
4551 		goto out;
4552 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4553 
4554 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4555 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4556 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4557 		test_tpr_threshold_values();
4558 		report_prefix_pop();
4559 
4560 		vmcs_write(CPU_EXEC_CTRL0,
4561 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4562 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4563 		test_tpr_threshold_values();
4564 		report_prefix_pop();
4565 	}
4566 
4567 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4568 		vmcs_write(CPU_EXEC_CTRL0,
4569 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4570 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4571 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4572 		test_tpr_threshold_values();
4573 		report_prefix_pop();
4574 
4575 		vmcs_write(CPU_EXEC_CTRL0,
4576 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4577 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4578 		test_tpr_threshold_values();
4579 		report_prefix_pop();
4580 	}
4581 
4582 	if ((ctrl_cpu_rev[1].clr &
4583 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4584 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4585 		vmcs_write(CPU_EXEC_CTRL0,
4586 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4587 		vmcs_write(CPU_EXEC_CTRL1,
4588 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4589 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4590 		test_tpr_threshold_values();
4591 		report_prefix_pop();
4592 
4593 		vmcs_write(CPU_EXEC_CTRL0,
4594 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4595 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4596 		test_tpr_threshold_values();
4597 		report_prefix_pop();
4598 	}
4599 
4600 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4601 out:
4602 	vmcs_write(TPR_THRESHOLD, threshold);
4603 	vmcs_write(APIC_VIRT_ADDR, apic_virt_addr);
4604 	vmcs_write(CPU_EXEC_CTRL0, primary);
4605 }
4606 
4607 /*
4608  * This test verifies the following two vmentry checks:
4609  *
4610  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4611  *  VM-execution control must be 0.
4612  *  [Intel SDM]
4613  *
4614  *  If the "virtual NMIs" VM-execution control is 0, the "NMI-window
4615  *  exiting" VM-execution control must be 0.
4616  *  [Intel SDM]
4617  */
test_nmi_ctrls(void)4618 static void test_nmi_ctrls(void)
4619 {
4620 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4621 
4622 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4623 	    (PIN_NMI | PIN_VIRT_NMI)) {
4624 		report_skip("%s : NMI exiting and/or Virtual NMIs not supported", __func__);
4625 		return;
4626 	}
4627 
4628 	/* Save the controls so that we can restore them after our tests */
4629 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4630 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4631 
4632 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4633 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4634 
4635 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4636 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4637 	test_vmx_valid_controls();
4638 	report_prefix_pop();
4639 
4640 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4641 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4642 	test_vmx_invalid_controls();
4643 	report_prefix_pop();
4644 
4645 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4646 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4647 	test_vmx_valid_controls();
4648 	report_prefix_pop();
4649 
4650 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4651 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4652 	test_vmx_valid_controls();
4653 	report_prefix_pop();
4654 
4655 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4656 		report_info("NMI-window exiting is not supported, skipping...");
4657 		goto done;
4658 	}
4659 
4660 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4661 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4662 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4663 	test_vmx_invalid_controls();
4664 	report_prefix_pop();
4665 
4666 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4667 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4668 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4669 	test_vmx_valid_controls();
4670 	report_prefix_pop();
4671 
4672 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4673 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4674 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4675 	test_vmx_valid_controls();
4676 	report_prefix_pop();
4677 
4678 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4679 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4680 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4681 	test_vmx_valid_controls();
4682 	report_prefix_pop();
4683 
4684 	/* Restore the controls to their original values */
4685 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4686 done:
4687 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4688 }
4689 
test_eptp_ad_bit(u64 eptp,bool is_ctrl_valid)4690 static void test_eptp_ad_bit(u64 eptp, bool is_ctrl_valid)
4691 {
4692 	vmcs_write(EPTP, eptp);
4693 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4694 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4695 	if (is_ctrl_valid)
4696 		test_vmx_valid_controls();
4697 	else
4698 		test_vmx_invalid_controls();
4699 	report_prefix_pop();
4700 
4701 }
4702 
4703 /*
4704  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4705  *    control field must satisfy the following checks:
4706  *
4707  *     - The EPT memory type (bits 2:0) must be a value supported by the
4708  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4709  *     - Bits 5:3 (1 less than the EPT page-walk length) must indicate a
4710  *	 supported EPT page-walk length.
4711  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4712  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4713  *	 indicating that the processor does not support accessed and dirty
4714  *	 dirty flags for EPT.
4715  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4716  *	 physical-address width) must all be 0.
4717  *
4718  * 2. If the "unrestricted guest" VM-execution control is 1, the
4719  *    "enable EPT" VM-execution control must also be 1.
4720  */
test_ept_eptp(void)4721 static void test_ept_eptp(void)
4722 {
4723 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4724 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4725 	u64 eptp_saved = vmcs_read(EPTP);
4726 	u32 secondary;
4727 	u64 eptp;
4728 	u32 i, maxphysaddr;
4729 	u64 j, resv_bits_mask = 0;
4730 
4731 	if (__setup_ept(0xfed40000, false)) {
4732 		report_skip("%s : EPT not supported", __func__);
4733 		return;
4734 	}
4735 
4736 	test_vmx_valid_controls();
4737 
4738 	setup_dummy_ept();
4739 
4740 	secondary = vmcs_read(CPU_EXEC_CTRL1);
4741 	eptp = vmcs_read(EPTP);
4742 
4743 	for (i = 0; i < 8; i++) {
4744 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4745 		vmcs_write(EPTP, eptp);
4746 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4747 		    eptp & EPT_MEM_TYPE_MASK);
4748 		if (is_ept_memtype_supported(i))
4749 			test_vmx_valid_controls();
4750 		else
4751 			test_vmx_invalid_controls();
4752 		report_prefix_pop();
4753 	}
4754 
4755 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4756 
4757 	/*
4758 	 * Page walk length (bits 5:3).  Note, the value in VMCS.EPTP "is 1
4759 	 * less than the EPT page-walk length".
4760 	 */
4761 	for (i = 0; i < 8; i++) {
4762 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4763 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4764 
4765 		vmcs_write(EPTP, eptp);
4766 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4767 		    eptp & EPTP_PG_WALK_LEN_MASK);
4768 		if (i == 3 || (i == 4 && is_5_level_ept_supported()))
4769 			test_vmx_valid_controls();
4770 		else
4771 			test_vmx_invalid_controls();
4772 		report_prefix_pop();
4773 	}
4774 
4775 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4776 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4777 
4778 	/*
4779 	 * Accessed and dirty flag (bit 6)
4780 	 */
4781 	if (ept_ad_bits_supported()) {
4782 		report_info("Processor supports accessed and dirty flag");
4783 		eptp &= ~EPTP_AD_FLAG;
4784 		test_eptp_ad_bit(eptp, true);
4785 
4786 		eptp |= EPTP_AD_FLAG;
4787 		test_eptp_ad_bit(eptp, true);
4788 	} else {
4789 		report_info("Processor does not supports accessed and dirty flag");
4790 		eptp &= ~EPTP_AD_FLAG;
4791 		test_eptp_ad_bit(eptp, true);
4792 
4793 		eptp |= EPTP_AD_FLAG;
4794 		test_eptp_ad_bit(eptp, false);
4795 
4796 		eptp &= ~EPTP_AD_FLAG;
4797 	}
4798 
4799 	/*
4800 	 * Reserved bits [11:7] and [63:N]
4801 	 */
4802 	for (i = 0; i < 32; i++) {
4803 		eptp = (eptp &
4804 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4805 		    (i << EPTP_RESERV_BITS_SHIFT);
4806 		vmcs_write(EPTP, eptp);
4807 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4808 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4809 		    EPTP_RESERV_BITS_MASK);
4810 		if (i == 0)
4811 			test_vmx_valid_controls();
4812 		else
4813 			test_vmx_invalid_controls();
4814 		report_prefix_pop();
4815 	}
4816 
4817 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4818 
4819 	maxphysaddr = cpuid_maxphyaddr();
4820 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4821 		resv_bits_mask |= 1ul << i;
4822 	}
4823 
4824 	for (j = maxphysaddr - 1; j <= 63; j++) {
4825 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4826 		    (j < maxphysaddr ? 0 : 1ul << j);
4827 		vmcs_write(EPTP, eptp);
4828 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4829 		    (eptp >> maxphysaddr) & resv_bits_mask);
4830 		if (j < maxphysaddr)
4831 			test_vmx_valid_controls();
4832 		else
4833 			test_vmx_invalid_controls();
4834 		report_prefix_pop();
4835 	}
4836 
4837 	secondary &= ~(CPU_EPT | CPU_URG);
4838 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4839 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4840 	test_vmx_valid_controls();
4841 	report_prefix_pop();
4842 
4843 	if (!(ctrl_cpu_rev[1].clr & CPU_URG))
4844 		goto skip_unrestricted_guest;
4845 
4846 	secondary |= CPU_URG;
4847 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4848 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4849 	test_vmx_invalid_controls();
4850 	report_prefix_pop();
4851 
4852 	secondary |= CPU_EPT;
4853 	setup_dummy_ept();
4854 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4855 	test_vmx_valid_controls();
4856 	report_prefix_pop();
4857 
4858 skip_unrestricted_guest:
4859 	secondary &= ~CPU_URG;
4860 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4861 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4862 	test_vmx_valid_controls();
4863 	report_prefix_pop();
4864 
4865 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4866 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4867 	vmcs_write(EPTP, eptp_saved);
4868 }
4869 
4870 /*
4871  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4872  * VM-execution control must also be 1. In addition, the PML address
4873  * must satisfy the following checks:
4874  *
4875  *    * Bits 11:0 of the address must be 0.
4876  *    * The address should not set any bits beyond the processor's
4877  *	physical-address width.
4878  *
4879  *  [Intel SDM]
4880  */
test_pml(void)4881 static void test_pml(void)
4882 {
4883 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4884 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4885 	u32 primary = primary_saved;
4886 	u32 secondary = secondary_saved;
4887 
4888 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4889 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
4890 		report_skip("%s : \"Secondary execution\" or \"enable EPT\" or \"enable PML\" control not supported", __func__);
4891 		return;
4892 	}
4893 
4894 	primary |= CPU_SECONDARY;
4895 	vmcs_write(CPU_EXEC_CTRL0, primary);
4896 	secondary &= ~(CPU_PML | CPU_EPT);
4897 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4898 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
4899 	test_vmx_valid_controls();
4900 	report_prefix_pop();
4901 
4902 	secondary |= CPU_PML;
4903 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4904 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
4905 	test_vmx_invalid_controls();
4906 	report_prefix_pop();
4907 
4908 	secondary |= CPU_EPT;
4909 	setup_dummy_ept();
4910 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
4911 	test_vmx_valid_controls();
4912 	report_prefix_pop();
4913 
4914 	secondary &= ~CPU_PML;
4915 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4916 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
4917 	test_vmx_valid_controls();
4918 	report_prefix_pop();
4919 
4920 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
4921 				 PAGE_SIZE, false, false);
4922 
4923 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4924 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4925 }
4926 
4927  /*
4928  * If the "activate VMX-preemption timer" VM-execution control is 0, the
4929  * the "save VMX-preemption timer value" VM-exit control must also be 0.
4930  *
4931  *  [Intel SDM]
4932  */
test_vmx_preemption_timer(void)4933 static void test_vmx_preemption_timer(void)
4934 {
4935 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4936 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
4937 	u32 pin = saved_pin;
4938 	u32 exit = saved_exit;
4939 
4940 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
4941 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
4942 		report_skip("%s : \"Save-VMX-preemption-timer\" and/or \"Enable-VMX-preemption-timer\" control not supported", __func__);
4943 		return;
4944 	}
4945 
4946 	pin |= PIN_PREEMPT;
4947 	vmcs_write(PIN_CONTROLS, pin);
4948 	exit &= ~EXI_SAVE_PREEMPT;
4949 	vmcs_write(EXI_CONTROLS, exit);
4950 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
4951 	test_vmx_valid_controls();
4952 	report_prefix_pop();
4953 
4954 	exit |= EXI_SAVE_PREEMPT;
4955 	vmcs_write(EXI_CONTROLS, exit);
4956 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
4957 	test_vmx_valid_controls();
4958 	report_prefix_pop();
4959 
4960 	pin &= ~PIN_PREEMPT;
4961 	vmcs_write(PIN_CONTROLS, pin);
4962 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
4963 	test_vmx_invalid_controls();
4964 	report_prefix_pop();
4965 
4966 	exit &= ~EXI_SAVE_PREEMPT;
4967 	vmcs_write(EXI_CONTROLS, exit);
4968 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
4969 	test_vmx_valid_controls();
4970 	report_prefix_pop();
4971 
4972 	vmcs_write(PIN_CONTROLS, saved_pin);
4973 	vmcs_write(EXI_CONTROLS, saved_exit);
4974 }
4975 
4976 extern unsigned char test_mtf1;
4977 extern unsigned char test_mtf2;
4978 extern unsigned char test_mtf3;
4979 extern unsigned char test_mtf4;
4980 
test_mtf_guest(void)4981 static void test_mtf_guest(void)
4982 {
4983 	asm ("vmcall;\n\t"
4984 	     "out %al, $0x80;\n\t"
4985 	     "test_mtf1:\n\t"
4986 	     "vmcall;\n\t"
4987 	     "out %al, $0x80;\n\t"
4988 	     "test_mtf2:\n\t"
4989 	     /*
4990 	      * Prepare for the 'MOV CR3' test. Attempt to induce a
4991 	      * general-protection fault by moving a non-canonical address into
4992 	      * CR3. The 'MOV CR3' instruction does not take an imm64 operand,
4993 	      * so we must MOV the desired value into a register first.
4994 	      *
4995 	      * MOV RAX is done before the VMCALL such that MTF is only enabled
4996 	      * for the instruction under test.
4997 	      */
4998 	     "mov $0xaaaaaaaaaaaaaaaa, %rax;\n\t"
4999 	     "vmcall;\n\t"
5000 	     "mov %rax, %cr3;\n\t"
5001 	     "test_mtf3:\n\t"
5002 	     "vmcall;\n\t"
5003 	     /*
5004 	      * ICEBP/INT1 instruction. Though the instruction is now
5005 	      * documented, don't rely on assemblers enumerating the
5006 	      * instruction. Resort to hand assembly.
5007 	      */
5008 	     ".byte 0xf1;\n\t"
5009 	     "vmcall;\n\t"
5010 	     "test_mtf4:\n\t"
5011 	     "mov $0, %eax;\n\t");
5012 }
5013 
test_mtf_gp_handler(struct ex_regs * regs)5014 static void test_mtf_gp_handler(struct ex_regs *regs)
5015 {
5016 	regs->rip = (unsigned long) &test_mtf3;
5017 }
5018 
test_mtf_db_handler(struct ex_regs * regs)5019 static void test_mtf_db_handler(struct ex_regs *regs)
5020 {
5021 }
5022 
enable_mtf(void)5023 static void enable_mtf(void)
5024 {
5025 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5026 
5027 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 | CPU_MTF);
5028 }
5029 
disable_mtf(void)5030 static void disable_mtf(void)
5031 {
5032 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5033 
5034 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 & ~CPU_MTF);
5035 }
5036 
enable_tf(void)5037 static void enable_tf(void)
5038 {
5039 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5040 
5041 	vmcs_write(GUEST_RFLAGS, rflags | X86_EFLAGS_TF);
5042 }
5043 
disable_tf(void)5044 static void disable_tf(void)
5045 {
5046 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5047 
5048 	vmcs_write(GUEST_RFLAGS, rflags & ~X86_EFLAGS_TF);
5049 }
5050 
report_mtf(const char * insn_name,unsigned long exp_rip)5051 static void report_mtf(const char *insn_name, unsigned long exp_rip)
5052 {
5053 	unsigned long rip = vmcs_read(GUEST_RIP);
5054 
5055 	assert_exit_reason(VMX_MTF);
5056 	report(rip == exp_rip, "MTF VM-exit after %s. RIP: 0x%lx (expected 0x%lx)",
5057 	       insn_name, rip, exp_rip);
5058 }
5059 
vmx_mtf_test(void)5060 static void vmx_mtf_test(void)
5061 {
5062 	unsigned long pending_dbg;
5063 	handler old_gp, old_db;
5064 
5065 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5066 		report_skip("%s : \"Monitor trap flag\" exec control not supported", __func__);
5067 		return;
5068 	}
5069 
5070 	test_set_guest(test_mtf_guest);
5071 
5072 	/* Expect an MTF VM-exit after OUT instruction */
5073 	enter_guest();
5074 	skip_exit_vmcall();
5075 
5076 	enable_mtf();
5077 	enter_guest();
5078 	report_mtf("OUT", (unsigned long) &test_mtf1);
5079 	disable_mtf();
5080 
5081 	/*
5082 	 * Concurrent #DB trap and MTF on instruction boundary. Expect MTF
5083 	 * VM-exit with populated 'pending debug exceptions' VMCS field.
5084 	 */
5085 	enter_guest();
5086 	skip_exit_vmcall();
5087 
5088 	enable_mtf();
5089 	enable_tf();
5090 
5091 	enter_guest();
5092 	report_mtf("OUT", (unsigned long) &test_mtf2);
5093 	pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
5094 	report(pending_dbg & DR6_BS,
5095 	       "'pending debug exceptions' field after MTF VM-exit: 0x%lx (expected 0x%lx)",
5096 	       pending_dbg, (unsigned long) DR6_BS);
5097 
5098 	disable_mtf();
5099 	disable_tf();
5100 	vmcs_write(GUEST_PENDING_DEBUG, 0);
5101 
5102 	/*
5103 	 * #GP exception takes priority over MTF. Expect MTF VM-exit with RIP
5104 	 * advanced to first instruction of #GP handler.
5105 	 */
5106 	enter_guest();
5107 	skip_exit_vmcall();
5108 
5109 	old_gp = handle_exception(GP_VECTOR, test_mtf_gp_handler);
5110 
5111 	enable_mtf();
5112 	enter_guest();
5113 	report_mtf("MOV CR3", (unsigned long) get_idt_addr(&boot_idt[GP_VECTOR]));
5114 	disable_mtf();
5115 
5116 	/*
5117 	 * Concurrent MTF and privileged software exception (i.e. ICEBP/INT1).
5118 	 * MTF should follow the delivery of #DB trap, though the SDM doesn't
5119 	 * provide clear indication of the relative priority.
5120 	 */
5121 	enter_guest();
5122 	skip_exit_vmcall();
5123 
5124 	handle_exception(GP_VECTOR, old_gp);
5125 	old_db = handle_exception(DB_VECTOR, test_mtf_db_handler);
5126 
5127 	enable_mtf();
5128 	enter_guest();
5129 	report_mtf("INT1", (unsigned long) get_idt_addr(&boot_idt[DB_VECTOR]));
5130 	disable_mtf();
5131 
5132 	enter_guest();
5133 	skip_exit_vmcall();
5134 	handle_exception(DB_VECTOR, old_db);
5135 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK | INTR_TYPE_OTHER_EVENT);
5136 	enter_guest();
5137 	report_mtf("injected MTF", (unsigned long) &test_mtf4);
5138 	enter_guest();
5139 }
5140 
5141 extern char vmx_mtf_pdpte_guest_begin;
5142 extern char vmx_mtf_pdpte_guest_end;
5143 
5144 asm("vmx_mtf_pdpte_guest_begin:\n\t"
5145     "mov %cr0, %rax\n\t"    /* save CR0 with PG=1                 */
5146     "vmcall\n\t"            /* on return from this CR0.PG=0       */
5147     "mov %rax, %cr0\n\t"    /* restore CR0.PG=1 to enter PAE mode */
5148     "vmcall\n\t"
5149     "retq\n\t"
5150     "vmx_mtf_pdpte_guest_end:");
5151 
vmx_mtf_pdpte_test(void)5152 static void vmx_mtf_pdpte_test(void)
5153 {
5154 	void *test_mtf_pdpte_guest;
5155 	pteval_t *pdpt;
5156 	u32 guest_ar_cs;
5157 	u64 guest_efer;
5158 	pteval_t *pte;
5159 	u64 guest_cr0;
5160 	u64 guest_cr3;
5161 	u64 guest_cr4;
5162 	u64 ent_ctls;
5163 	int i;
5164 
5165 	if (setup_ept(false))
5166 		return;
5167 
5168 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5169 		report_skip("%s : \"Monitor trap flag\" exec control not supported", __func__);
5170 		return;
5171 	}
5172 
5173 	if (!(ctrl_cpu_rev[1].clr & CPU_URG)) {
5174 		report_skip("%s : \"Unrestricted guest\" exec control not supported", __func__);
5175 		return;
5176 	}
5177 
5178 	vmcs_write(EXC_BITMAP, ~0);
5179 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
5180 
5181 	/*
5182 	 * Copy the guest code to an identity-mapped page.
5183 	 */
5184 	test_mtf_pdpte_guest = alloc_page();
5185 	memcpy(test_mtf_pdpte_guest, &vmx_mtf_pdpte_guest_begin,
5186 	       &vmx_mtf_pdpte_guest_end - &vmx_mtf_pdpte_guest_begin);
5187 
5188 	test_set_guest(test_mtf_pdpte_guest);
5189 
5190 	enter_guest();
5191 	skip_exit_vmcall();
5192 
5193 	/*
5194 	 * Put the guest in non-paged 32-bit protected mode, ready to enter
5195 	 * PAE mode when CR0.PG is set. CR4.PAE will already have been set
5196 	 * when the guest started out in long mode.
5197 	 */
5198 	ent_ctls = vmcs_read(ENT_CONTROLS);
5199 	vmcs_write(ENT_CONTROLS, ent_ctls & ~ENT_GUEST_64);
5200 
5201 	guest_efer = vmcs_read(GUEST_EFER);
5202 	vmcs_write(GUEST_EFER, guest_efer & ~(EFER_LMA | EFER_LME));
5203 
5204 	/*
5205 	 * Set CS access rights bits for 32-bit protected mode:
5206 	 * 3:0    B execute/read/accessed
5207 	 * 4      1 code or data
5208 	 * 6:5    0 descriptor privilege level
5209 	 * 7      1 present
5210 	 * 11:8   0 reserved
5211 	 * 12     0 available for use by system software
5212 	 * 13     0 64 bit mode not active
5213 	 * 14     1 default operation size 32-bit segment
5214 	 * 15     1 page granularity: segment limit in 4K units
5215 	 * 16     0 segment usable
5216 	 * 31:17  0 reserved
5217 	 */
5218 	guest_ar_cs = vmcs_read(GUEST_AR_CS);
5219 	vmcs_write(GUEST_AR_CS, 0xc09b);
5220 
5221 	guest_cr0 = vmcs_read(GUEST_CR0);
5222 	vmcs_write(GUEST_CR0, guest_cr0 & ~X86_CR0_PG);
5223 
5224 	guest_cr4 = vmcs_read(GUEST_CR4);
5225 	vmcs_write(GUEST_CR4, guest_cr4 & ~X86_CR4_PCIDE);
5226 
5227 	guest_cr3 = vmcs_read(GUEST_CR3);
5228 
5229 	/*
5230 	 * Turn the 4-level page table into a PAE page table by following the 0th
5231 	 * PML4 entry to a PDPT page, and grab the first four PDPTEs from that
5232 	 * page.
5233 	 *
5234 	 * Why does this work?
5235 	 *
5236 	 * PAE uses 32-bit addressing which implies:
5237 	 * Bits 11:0   page offset
5238 	 * Bits 20:12  entry into 512-entry page table
5239 	 * Bits 29:21  entry into a 512-entry directory table
5240 	 * Bits 31:30  entry into the page directory pointer table.
5241 	 * Bits 63:32  zero
5242 	 *
5243 	 * As only 2 bits are needed to select the PDPTEs for the entire
5244 	 * 32-bit address space, take the first 4 PDPTEs in the level 3 page
5245 	 * directory pointer table. It doesn't matter which of these PDPTEs
5246 	 * are present because they must cover the guest code given that it
5247 	 * has already run successfully.
5248 	 *
5249 	 * Get a pointer to PTE for GVA=0 in the page directory pointer table
5250 	 */
5251 	pte = get_pte_level(
5252             (pgd_t *)phys_to_virt(guest_cr3 & ~X86_CR3_PCID_MASK), 0,
5253             PDPT_LEVEL);
5254 
5255 	/*
5256 	 * Need some memory for the 4-entry PAE page directory pointer
5257 	 * table. Use the end of the identity-mapped page where the guest code
5258 	 * is stored. There is definitely space as the guest code is only a
5259 	 * few bytes.
5260 	 */
5261 	pdpt = test_mtf_pdpte_guest + PAGE_SIZE - 4 * sizeof(pteval_t);
5262 
5263 	/*
5264 	 * Copy the first four PDPTEs into the PAE page table with reserved
5265 	 * bits cleared. Note that permission bits from the PML4E and PDPTE
5266 	 * are not propagated.
5267 	 */
5268 	for (i = 0; i < 4; i++) {
5269 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_RSVD_MASK),
5270 				   "PDPTE has invalid reserved bits");
5271 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_PAGE_SIZE_MASK),
5272 				   "Cannot use 1GB super pages for PAE");
5273 		pdpt[i] = pte[i] & ~(PAE_PDPTE_RSVD_MASK);
5274 	}
5275 	vmcs_write(GUEST_CR3, virt_to_phys(pdpt));
5276 
5277 	enable_mtf();
5278 	enter_guest();
5279 	assert_exit_reason(VMX_MTF);
5280 	disable_mtf();
5281 
5282 	/*
5283 	 * The four PDPTEs should have been loaded into the VMCS when
5284 	 * the guest set CR0.PG to enter PAE mode.
5285 	 */
5286 	for (i = 0; i < 4; i++) {
5287 		u64 pdpte = vmcs_read(GUEST_PDPTE + 2 * i);
5288 
5289 		report(pdpte == pdpt[i], "PDPTE%d is 0x%lx (expected 0x%lx)",
5290 		       i, pdpte, pdpt[i]);
5291 	}
5292 
5293 	/*
5294 	 * Now, try to enter the guest in PAE mode. If the PDPTEs in the
5295 	 * vmcs are wrong, this will fail.
5296 	 */
5297 	enter_guest();
5298 	skip_exit_vmcall();
5299 
5300 	/*
5301 	 * Return guest to 64-bit mode and wrap up.
5302 	 */
5303 	vmcs_write(ENT_CONTROLS, ent_ctls);
5304 	vmcs_write(GUEST_EFER, guest_efer);
5305 	vmcs_write(GUEST_AR_CS, guest_ar_cs);
5306 	vmcs_write(GUEST_CR0, guest_cr0);
5307 	vmcs_write(GUEST_CR4, guest_cr4);
5308 	vmcs_write(GUEST_CR3, guest_cr3);
5309 
5310 	enter_guest();
5311 }
5312 
5313 /*
5314  * Tests for VM-execution control fields
5315  */
test_vm_execution_ctls(void)5316 static void test_vm_execution_ctls(void)
5317 {
5318 	test_pin_based_ctls();
5319 	test_primary_processor_based_ctls();
5320 	test_secondary_processor_based_ctls();
5321 	test_cr3_targets();
5322 	test_io_bitmaps();
5323 	test_msr_bitmap();
5324 	test_apic_ctls();
5325 	test_tpr_threshold();
5326 	test_nmi_ctrls();
5327 	test_pml();
5328 	test_vpid();
5329 	test_ept_eptp();
5330 	test_vmx_preemption_timer();
5331 }
5332 
5333  /*
5334   * The following checks are performed for the VM-entry MSR-load address if
5335   * the VM-entry MSR-load count field is non-zero:
5336   *
5337   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
5338   *      The address should not set any bits beyond the processor's
5339   *      physical-address width.
5340   *
5341   *    - The address of the last byte in the VM-entry MSR-load area
5342   *      should not set any bits beyond the processor's physical-address
5343   *      width. The address of this last byte is VM-entry MSR-load address
5344   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5345   *      uses more bits than the processor's physical-address width.)
5346   *
5347   *
5348   *  [Intel SDM]
5349   */
test_entry_msr_load(void)5350 static void test_entry_msr_load(void)
5351 {
5352 	entry_msr_load = alloc_page();
5353 	u64 tmp;
5354 	u32 entry_msr_ld_cnt = 1;
5355 	int i;
5356 	u32 addr_len = 64;
5357 
5358 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
5359 
5360 	/* Check first 4 bits of VM-entry MSR-load address */
5361 	for (i = 0; i < 4; i++) {
5362 		tmp = (u64)entry_msr_load | 1ull << i;
5363 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
5364 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
5365 				    tmp & 0xf);
5366 		test_vmx_invalid_controls();
5367 		report_prefix_pop();
5368 	}
5369 
5370 	if (basic_msr.val & (1ul << 48))
5371 		addr_len = 32;
5372 
5373 	test_vmcs_addr_values("VM-entry-MSR-load address",
5374 				ENTER_MSR_LD_ADDR, 16, false, false,
5375 				4, addr_len - 1);
5376 
5377 	/*
5378 	 * Check last byte of VM-entry MSR-load address
5379 	 */
5380 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
5381 
5382 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5383 							i < 64; i++) {
5384 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
5385 			1ul << i;
5386 		vmcs_write(ENTER_MSR_LD_ADDR,
5387 			   tmp - (entry_msr_ld_cnt * 16 - 1));
5388 		test_vmx_invalid_controls();
5389 	}
5390 
5391 	vmcs_write(ENT_MSR_LD_CNT, 2);
5392 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5393 	test_vmx_invalid_controls();
5394 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5395 	test_vmx_valid_controls();
5396 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5397 	test_vmx_valid_controls();
5398 }
5399 
5400 static struct vmx_state_area_test_data {
5401 	u32 msr;
5402 	u64 exp;
5403 	bool enabled;
5404 } vmx_state_area_test_data;
5405 
guest_state_test_main(void)5406 static void guest_state_test_main(void)
5407 {
5408 	u64 obs;
5409 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
5410 
5411 	while (1) {
5412 		if (vmx_get_test_stage() == 2)
5413 			break;
5414 
5415 		if (data->enabled) {
5416 			obs = rdmsr(data->msr);
5417 			report(data->exp == obs,
5418 			       "Guest state is 0x%lx (expected 0x%lx)",
5419 			       obs, data->exp);
5420 		}
5421 
5422 		vmcall();
5423 	}
5424 
5425 	asm volatile("fnop");
5426 }
5427 
test_guest_state(const char * test,bool xfail,u64 field,const char * field_name)5428 static void test_guest_state(const char *test, bool xfail, u64 field,
5429 			     const char * field_name)
5430 {
5431 	struct vmentry_result result;
5432 	u8 abort_flags;
5433 
5434 	abort_flags = ABORT_ON_EARLY_VMENTRY_FAIL;
5435 	if (!xfail)
5436 		abort_flags = ABORT_ON_INVALID_GUEST_STATE;
5437 
5438 	__enter_guest(abort_flags, &result);
5439 
5440 	report(result.exit_reason.failed_vmentry == xfail &&
5441 	       ((xfail && result.exit_reason.basic == VMX_FAIL_STATE) ||
5442 	        (!xfail && result.exit_reason.basic == VMX_VMCALL)) &&
5443 		(!xfail || vmcs_read(EXI_QUALIFICATION) == ENTRY_FAIL_DEFAULT),
5444 	        "%s, %s = %lx", test, field_name, field);
5445 
5446 	if (!result.exit_reason.failed_vmentry)
5447 		skip_exit_insn();
5448 }
5449 
5450 /*
5451  * Tests for VM-entry control fields
5452  */
test_vm_entry_ctls(void)5453 static void test_vm_entry_ctls(void)
5454 {
5455 	test_invalid_event_injection();
5456 	test_entry_msr_load();
5457 }
5458 
5459 /*
5460  * The following checks are performed for the VM-exit MSR-store address if
5461  * the VM-exit MSR-store count field is non-zero:
5462  *
5463  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
5464  *      The address should not set any bits beyond the processor's
5465  *      physical-address width.
5466  *
5467  *    - The address of the last byte in the VM-exit MSR-store area
5468  *      should not set any bits beyond the processor's physical-address
5469  *      width. The address of this last byte is VM-exit MSR-store address
5470  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5471  *      uses more bits than the processor's physical-address width.)
5472  *
5473  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
5474  * in the range 63:32.
5475  *
5476  *  [Intel SDM]
5477  */
test_exit_msr_store(void)5478 static void test_exit_msr_store(void)
5479 {
5480 	exit_msr_store = alloc_page();
5481 	u64 tmp;
5482 	u32 exit_msr_st_cnt = 1;
5483 	int i;
5484 	u32 addr_len = 64;
5485 
5486 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
5487 
5488 	/* Check first 4 bits of VM-exit MSR-store address */
5489 	for (i = 0; i < 4; i++) {
5490 		tmp = (u64)exit_msr_store | 1ull << i;
5491 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
5492 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
5493 				    tmp & 0xf);
5494 		test_vmx_invalid_controls();
5495 		report_prefix_pop();
5496 	}
5497 
5498 	if (basic_msr.val & (1ul << 48))
5499 		addr_len = 32;
5500 
5501 	test_vmcs_addr_values("VM-exit-MSR-store address",
5502 				EXIT_MSR_ST_ADDR, 16, false, false,
5503 				4, addr_len - 1);
5504 
5505 	/*
5506 	 * Check last byte of VM-exit MSR-store address
5507 	 */
5508 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
5509 
5510 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5511 							i < 64; i++) {
5512 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
5513 			1ul << i;
5514 		vmcs_write(EXIT_MSR_ST_ADDR,
5515 			   tmp - (exit_msr_st_cnt * 16 - 1));
5516 		test_vmx_invalid_controls();
5517 	}
5518 
5519 	vmcs_write(EXI_MSR_ST_CNT, 2);
5520 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5521 	test_vmx_invalid_controls();
5522 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5523 	test_vmx_valid_controls();
5524 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5525 	test_vmx_valid_controls();
5526 }
5527 
5528 /*
5529  * Tests for VM-exit controls
5530  */
test_vm_exit_ctls(void)5531 static void test_vm_exit_ctls(void)
5532 {
5533 	test_exit_msr_store();
5534 }
5535 
5536 /*
5537  * Check that the virtual CPU checks all of the VMX controls as
5538  * documented in the Intel SDM.
5539  */
vmx_controls_test(void)5540 static void vmx_controls_test(void)
5541 {
5542 	/*
5543 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
5544 	 * fail due to invalid guest state, should we make it that
5545 	 * far.
5546 	 */
5547 	vmcs_write(GUEST_RFLAGS, 0);
5548 
5549 	test_vm_execution_ctls();
5550 	test_vm_exit_ctls();
5551 	test_vm_entry_ctls();
5552 }
5553 
5554 struct apic_reg_virt_config {
5555 	bool apic_register_virtualization;
5556 	bool use_tpr_shadow;
5557 	bool virtualize_apic_accesses;
5558 	bool virtualize_x2apic_mode;
5559 	bool activate_secondary_controls;
5560 };
5561 
5562 struct apic_reg_test {
5563 	const char *name;
5564 	struct apic_reg_virt_config apic_reg_virt_config;
5565 };
5566 
5567 struct apic_reg_virt_expectation {
5568 	enum Reason rd_exit_reason;
5569 	enum Reason wr_exit_reason;
5570 	u32 val;
5571 	u32 (*virt_fn)(u32);
5572 
5573 	/*
5574 	 * If false, accessing the APIC access address from L2 is treated as a
5575 	 * normal memory operation, rather than triggering virtualization.
5576 	 */
5577 	bool virtualize_apic_accesses;
5578 };
5579 
apic_virt_identity(u32 val)5580 static u32 apic_virt_identity(u32 val)
5581 {
5582 	return val;
5583 }
5584 
apic_virt_nibble1(u32 val)5585 static u32 apic_virt_nibble1(u32 val)
5586 {
5587 	return val & 0xf0;
5588 }
5589 
apic_virt_byte3(u32 val)5590 static u32 apic_virt_byte3(u32 val)
5591 {
5592 	return val & (0xff << 24);
5593 }
5594 
apic_reg_virt_exit_expectation(u32 reg,struct apic_reg_virt_config * config,struct apic_reg_virt_expectation * expectation)5595 static bool apic_reg_virt_exit_expectation(
5596 	u32 reg, struct apic_reg_virt_config *config,
5597 	struct apic_reg_virt_expectation *expectation)
5598 {
5599 	/* Good configs, where some L2 APIC accesses are virtualized. */
5600 	bool virtualize_apic_accesses_only =
5601 		config->virtualize_apic_accesses &&
5602 		!config->use_tpr_shadow &&
5603 		!config->apic_register_virtualization &&
5604 		!config->virtualize_x2apic_mode &&
5605 		config->activate_secondary_controls;
5606 	bool virtualize_apic_accesses_and_use_tpr_shadow =
5607 		config->virtualize_apic_accesses &&
5608 		config->use_tpr_shadow &&
5609 		!config->apic_register_virtualization &&
5610 		!config->virtualize_x2apic_mode &&
5611 		config->activate_secondary_controls;
5612 	bool apic_register_virtualization =
5613 		config->virtualize_apic_accesses &&
5614 		config->use_tpr_shadow &&
5615 		config->apic_register_virtualization &&
5616 		!config->virtualize_x2apic_mode &&
5617 		config->activate_secondary_controls;
5618 
5619 	expectation->val = MAGIC_VAL_1;
5620 	expectation->virt_fn = apic_virt_identity;
5621 	expectation->virtualize_apic_accesses =
5622 		config->virtualize_apic_accesses &&
5623 		config->activate_secondary_controls;
5624 	if (virtualize_apic_accesses_only) {
5625 		expectation->rd_exit_reason = VMX_APIC_ACCESS;
5626 		expectation->wr_exit_reason = VMX_APIC_ACCESS;
5627 	} else if (virtualize_apic_accesses_and_use_tpr_shadow) {
5628 		switch (reg) {
5629 		case APIC_TASKPRI:
5630 			expectation->rd_exit_reason = VMX_VMCALL;
5631 			expectation->wr_exit_reason = VMX_VMCALL;
5632 			expectation->virt_fn = apic_virt_nibble1;
5633 			break;
5634 		default:
5635 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5636 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5637 		}
5638 	} else if (apic_register_virtualization) {
5639 		expectation->rd_exit_reason = VMX_VMCALL;
5640 
5641 		switch (reg) {
5642 		case APIC_ID:
5643 		case APIC_EOI:
5644 		case APIC_LDR:
5645 		case APIC_DFR:
5646 		case APIC_SPIV:
5647 		case APIC_ESR:
5648 		case APIC_ICR:
5649 		case APIC_LVTT:
5650 		case APIC_LVTTHMR:
5651 		case APIC_LVTPC:
5652 		case APIC_LVT0:
5653 		case APIC_LVT1:
5654 		case APIC_LVTERR:
5655 		case APIC_TMICT:
5656 		case APIC_TDCR:
5657 			expectation->wr_exit_reason = VMX_APIC_WRITE;
5658 			break;
5659 		case APIC_LVR:
5660 		case APIC_ISR ... APIC_ISR + 0x70:
5661 		case APIC_TMR ... APIC_TMR + 0x70:
5662 		case APIC_IRR ... APIC_IRR + 0x70:
5663 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5664 			break;
5665 		case APIC_TASKPRI:
5666 			expectation->wr_exit_reason = VMX_VMCALL;
5667 			expectation->virt_fn = apic_virt_nibble1;
5668 			break;
5669 		case APIC_ICR2:
5670 			expectation->wr_exit_reason = VMX_VMCALL;
5671 			expectation->virt_fn = apic_virt_byte3;
5672 			break;
5673 		default:
5674 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5675 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5676 		}
5677 	} else if (!expectation->virtualize_apic_accesses) {
5678 		/*
5679 		 * No APIC registers are directly virtualized. This includes
5680 		 * VTPR, which can be virtualized through MOV to/from CR8 via
5681 		 * the use TPR shadow control, but not through directly
5682 		 * accessing VTPR.
5683 		 */
5684 		expectation->rd_exit_reason = VMX_VMCALL;
5685 		expectation->wr_exit_reason = VMX_VMCALL;
5686 	} else {
5687 		printf("Cannot parse APIC register virtualization config:\n"
5688 		       "\tvirtualize_apic_accesses: %d\n"
5689 		       "\tuse_tpr_shadow: %d\n"
5690 		       "\tapic_register_virtualization: %d\n"
5691 		       "\tvirtualize_x2apic_mode: %d\n"
5692 		       "\tactivate_secondary_controls: %d\n",
5693 		       config->virtualize_apic_accesses,
5694 		       config->use_tpr_shadow,
5695 		       config->apic_register_virtualization,
5696 		       config->virtualize_x2apic_mode,
5697 		       config->activate_secondary_controls);
5698 
5699 		return false;
5700 	}
5701 
5702 	return true;
5703 }
5704 
5705 struct apic_reg_test apic_reg_tests[] = {
5706 	/* Good configs, where some L2 APIC accesses are virtualized. */
5707 	{
5708 		.name = "Virtualize APIC accesses",
5709 		.apic_reg_virt_config = {
5710 			.virtualize_apic_accesses = true,
5711 			.use_tpr_shadow = false,
5712 			.apic_register_virtualization = false,
5713 			.virtualize_x2apic_mode = false,
5714 			.activate_secondary_controls = true,
5715 		},
5716 	},
5717 	{
5718 		.name = "Virtualize APIC accesses + Use TPR shadow",
5719 		.apic_reg_virt_config = {
5720 			.virtualize_apic_accesses = true,
5721 			.use_tpr_shadow = true,
5722 			.apic_register_virtualization = false,
5723 			.virtualize_x2apic_mode = false,
5724 			.activate_secondary_controls = true,
5725 		},
5726 	},
5727 	{
5728 		.name = "APIC-register virtualization",
5729 		.apic_reg_virt_config = {
5730 			.virtualize_apic_accesses = true,
5731 			.use_tpr_shadow = true,
5732 			.apic_register_virtualization = true,
5733 			.virtualize_x2apic_mode = false,
5734 			.activate_secondary_controls = true,
5735 		},
5736 	},
5737 
5738 	/*
5739 	 * Test that the secondary processor-based VM-execution controls are
5740 	 * correctly ignored when "activate secondary controls" is disabled.
5741 	 */
5742 	{
5743 		.name = "Activate secondary controls off",
5744 		.apic_reg_virt_config = {
5745 			.virtualize_apic_accesses = true,
5746 			.use_tpr_shadow = false,
5747 			.apic_register_virtualization = true,
5748 			.virtualize_x2apic_mode = true,
5749 			.activate_secondary_controls = false,
5750 		},
5751 	},
5752 	{
5753 		.name = "Activate secondary controls off + Use TPR shadow",
5754 		.apic_reg_virt_config = {
5755 			.virtualize_apic_accesses = true,
5756 			.use_tpr_shadow = true,
5757 			.apic_register_virtualization = true,
5758 			.virtualize_x2apic_mode = true,
5759 			.activate_secondary_controls = false,
5760 		},
5761 	},
5762 
5763 	/*
5764 	 * Test that the APIC access address is treated like an arbitrary memory
5765 	 * address when "virtualize APIC accesses" is disabled.
5766 	 */
5767 	{
5768 		.name = "Virtualize APIC accesses off + Use TPR shadow",
5769 		.apic_reg_virt_config = {
5770 			.virtualize_apic_accesses = false,
5771 			.use_tpr_shadow = true,
5772 			.apic_register_virtualization = true,
5773 			.virtualize_x2apic_mode = true,
5774 			.activate_secondary_controls = true,
5775 		},
5776 	},
5777 
5778 	/*
5779 	 * Test that VM entry fails due to invalid controls when
5780 	 * "APIC-register virtualization" is enabled while "use TPR shadow" is
5781 	 * disabled.
5782 	 */
5783 	{
5784 		.name = "APIC-register virtualization + Use TPR shadow off",
5785 		.apic_reg_virt_config = {
5786 			.virtualize_apic_accesses = true,
5787 			.use_tpr_shadow = false,
5788 			.apic_register_virtualization = true,
5789 			.virtualize_x2apic_mode = false,
5790 			.activate_secondary_controls = true,
5791 		},
5792 	},
5793 
5794 	/*
5795 	 * Test that VM entry fails due to invalid controls when
5796 	 * "Virtualize x2APIC mode" is enabled while "use TPR shadow" is
5797 	 * disabled.
5798 	 */
5799 	{
5800 		.name = "Virtualize x2APIC mode + Use TPR shadow off",
5801 		.apic_reg_virt_config = {
5802 			.virtualize_apic_accesses = false,
5803 			.use_tpr_shadow = false,
5804 			.apic_register_virtualization = false,
5805 			.virtualize_x2apic_mode = true,
5806 			.activate_secondary_controls = true,
5807 		},
5808 	},
5809 	{
5810 		.name = "Virtualize x2APIC mode + Use TPR shadow off v2",
5811 		.apic_reg_virt_config = {
5812 			.virtualize_apic_accesses = false,
5813 			.use_tpr_shadow = false,
5814 			.apic_register_virtualization = true,
5815 			.virtualize_x2apic_mode = true,
5816 			.activate_secondary_controls = true,
5817 		},
5818 	},
5819 
5820 	/*
5821 	 * Test that VM entry fails due to invalid controls when
5822 	 * "virtualize x2APIC mode" is enabled while "virtualize APIC accesses"
5823 	 * is enabled.
5824 	 */
5825 	{
5826 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses",
5827 		.apic_reg_virt_config = {
5828 			.virtualize_apic_accesses = true,
5829 			.use_tpr_shadow = true,
5830 			.apic_register_virtualization = false,
5831 			.virtualize_x2apic_mode = true,
5832 			.activate_secondary_controls = true,
5833 		},
5834 	},
5835 	{
5836 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses v2",
5837 		.apic_reg_virt_config = {
5838 			.virtualize_apic_accesses = true,
5839 			.use_tpr_shadow = true,
5840 			.apic_register_virtualization = true,
5841 			.virtualize_x2apic_mode = true,
5842 			.activate_secondary_controls = true,
5843 		},
5844 	},
5845 };
5846 
5847 enum Apic_op {
5848 	APIC_OP_XAPIC_RD,
5849 	APIC_OP_XAPIC_WR,
5850 	TERMINATE,
5851 };
5852 
vmx_xapic_read(u32 * apic_access_address,u32 reg)5853 static u32 vmx_xapic_read(u32 *apic_access_address, u32 reg)
5854 {
5855 	return *(volatile u32 *)((uintptr_t)apic_access_address + reg);
5856 }
5857 
vmx_xapic_write(u32 * apic_access_address,u32 reg,u32 val)5858 static void vmx_xapic_write(u32 *apic_access_address, u32 reg, u32 val)
5859 {
5860 	*(volatile u32 *)((uintptr_t)apic_access_address + reg) = val;
5861 }
5862 
5863 struct apic_reg_virt_guest_args {
5864 	enum Apic_op op;
5865 	u32 *apic_access_address;
5866 	u32 reg;
5867 	u32 val;
5868 	bool check_rd;
5869 	u32 (*virt_fn)(u32);
5870 } apic_reg_virt_guest_args;
5871 
apic_reg_virt_guest(void)5872 static void apic_reg_virt_guest(void)
5873 {
5874 	volatile struct apic_reg_virt_guest_args *args =
5875 		&apic_reg_virt_guest_args;
5876 
5877 	for (;;) {
5878 		enum Apic_op op = args->op;
5879 		u32 *apic_access_address = args->apic_access_address;
5880 		u32 reg = args->reg;
5881 		u32 val = args->val;
5882 		bool check_rd = args->check_rd;
5883 		u32 (*virt_fn)(u32) = args->virt_fn;
5884 
5885 		if (op == TERMINATE)
5886 			break;
5887 
5888 		if (op == APIC_OP_XAPIC_RD) {
5889 			u32 ret = vmx_xapic_read(apic_access_address, reg);
5890 
5891 			if (check_rd) {
5892 				u32 want = virt_fn(val);
5893 				u32 got = virt_fn(ret);
5894 
5895 				report(got == want,
5896 				       "read 0x%x, expected 0x%x.", got, want);
5897 			}
5898 		} else if (op == APIC_OP_XAPIC_WR) {
5899 			vmx_xapic_write(apic_access_address, reg, val);
5900 		}
5901 
5902 		/*
5903 		 * The L1 should always execute a vmcall after it's done testing
5904 		 * an individual APIC operation. This helps to validate that the
5905 		 * L1 and L2 are in sync with each other, as expected.
5906 		 */
5907 		vmcall();
5908 	}
5909 }
5910 
test_xapic_rd(u32 reg,struct apic_reg_virt_expectation * expectation,u32 * apic_access_address,u32 * virtual_apic_page)5911 static void test_xapic_rd(
5912 	u32 reg, struct apic_reg_virt_expectation *expectation,
5913 	u32 *apic_access_address, u32 *virtual_apic_page)
5914 {
5915 	u32 val = expectation->val;
5916 	u32 exit_reason_want = expectation->rd_exit_reason;
5917 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5918 
5919 	report_prefix_pushf("xapic - reading 0x%03x", reg);
5920 
5921 	/* Configure guest to do an xapic read */
5922 	args->op = APIC_OP_XAPIC_RD;
5923 	args->apic_access_address = apic_access_address;
5924 	args->reg = reg;
5925 	args->val = val;
5926 	args->check_rd = exit_reason_want == VMX_VMCALL;
5927 	args->virt_fn = expectation->virt_fn;
5928 
5929 	/* Setup virtual APIC page */
5930 	if (!expectation->virtualize_apic_accesses) {
5931 		apic_access_address[apic_reg_index(reg)] = val;
5932 		virtual_apic_page[apic_reg_index(reg)] = 0;
5933 	} else if (exit_reason_want == VMX_VMCALL) {
5934 		apic_access_address[apic_reg_index(reg)] = 0;
5935 		virtual_apic_page[apic_reg_index(reg)] = val;
5936 	}
5937 
5938 	/* Enter guest */
5939 	enter_guest();
5940 
5941 	/*
5942 	 * Validate the behavior and
5943 	 * pass a magic value back to the guest.
5944 	 */
5945 	if (exit_reason_want == VMX_APIC_ACCESS) {
5946 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5947 
5948 		assert_exit_reason(exit_reason_want);
5949 		report(apic_page_offset == reg,
5950 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
5951 		       apic_page_offset, reg);
5952 		skip_exit_insn();
5953 
5954 		/* Reenter guest so it can consume/check rcx and exit again. */
5955 		enter_guest();
5956 	} else if (exit_reason_want != VMX_VMCALL) {
5957 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
5958 	}
5959 
5960 	skip_exit_vmcall();
5961 	report_prefix_pop();
5962 }
5963 
test_xapic_wr(u32 reg,struct apic_reg_virt_expectation * expectation,u32 * apic_access_address,u32 * virtual_apic_page)5964 static void test_xapic_wr(
5965 	u32 reg, struct apic_reg_virt_expectation *expectation,
5966 	u32 *apic_access_address, u32 *virtual_apic_page)
5967 {
5968 	u32 val = expectation->val;
5969 	u32 exit_reason_want = expectation->wr_exit_reason;
5970 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5971 	bool virtualized =
5972 		expectation->virtualize_apic_accesses &&
5973 		(exit_reason_want == VMX_APIC_WRITE ||
5974 		 exit_reason_want == VMX_VMCALL);
5975 	bool checked = false;
5976 
5977 	report_prefix_pushf("xapic - writing 0x%x to 0x%03x", val, reg);
5978 
5979 	/* Configure guest to do an xapic read */
5980 	args->op = APIC_OP_XAPIC_WR;
5981 	args->apic_access_address = apic_access_address;
5982 	args->reg = reg;
5983 	args->val = val;
5984 
5985 	/* Setup virtual APIC page */
5986 	if (virtualized || !expectation->virtualize_apic_accesses) {
5987 		apic_access_address[apic_reg_index(reg)] = 0;
5988 		virtual_apic_page[apic_reg_index(reg)] = 0;
5989 	}
5990 
5991 	/* Enter guest */
5992 	enter_guest();
5993 
5994 	/*
5995 	 * Validate the behavior and
5996 	 * pass a magic value back to the guest.
5997 	 */
5998 	if (exit_reason_want == VMX_APIC_ACCESS) {
5999 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
6000 
6001 		assert_exit_reason(exit_reason_want);
6002 		report(apic_page_offset == reg,
6003 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
6004 		       apic_page_offset, reg);
6005 		skip_exit_insn();
6006 
6007 		/* Reenter guest so it can consume/check rcx and exit again. */
6008 		enter_guest();
6009 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6010 		assert_exit_reason(exit_reason_want);
6011 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6012 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%x",
6013 		       apic_reg_index(reg),
6014 		       virtual_apic_page[apic_reg_index(reg)], val);
6015 		checked = true;
6016 
6017 		/* Reenter guest so it can consume/check rcx and exit again. */
6018 		enter_guest();
6019 	} else if (exit_reason_want != VMX_VMCALL) {
6020 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6021 	}
6022 
6023 	assert_exit_reason(VMX_VMCALL);
6024 	if (virtualized && !checked) {
6025 		u32 want = expectation->virt_fn(val);
6026 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6027 		got = expectation->virt_fn(got);
6028 
6029 		report(got == want, "exitless write; val is 0x%x, want 0x%x",
6030 		       got, want);
6031 	} else if (!expectation->virtualize_apic_accesses && !checked) {
6032 		u32 got = apic_access_address[apic_reg_index(reg)];
6033 
6034 		report(got == val,
6035 		       "non-virtualized write; val is 0x%x, want 0x%x", got,
6036 		       val);
6037 	} else if (!expectation->virtualize_apic_accesses && checked) {
6038 		report_fail("Non-virtualized write was prematurely checked!");
6039 	}
6040 
6041 	skip_exit_vmcall();
6042 	report_prefix_pop();
6043 }
6044 
6045 enum Config_type {
6046 	CONFIG_TYPE_GOOD,
6047 	CONFIG_TYPE_UNSUPPORTED,
6048 	CONFIG_TYPE_VMENTRY_FAILS_EARLY,
6049 };
6050 
configure_apic_reg_virt_test(struct apic_reg_virt_config * apic_reg_virt_config)6051 static enum Config_type configure_apic_reg_virt_test(
6052 	struct apic_reg_virt_config *apic_reg_virt_config)
6053 {
6054 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6055 	u32 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6056 	/* Configs where L2 entry fails early, due to invalid controls. */
6057 	bool use_tpr_shadow_incorrectly_off =
6058 		!apic_reg_virt_config->use_tpr_shadow &&
6059 		(apic_reg_virt_config->apic_register_virtualization ||
6060 		 apic_reg_virt_config->virtualize_x2apic_mode) &&
6061 		apic_reg_virt_config->activate_secondary_controls;
6062 	bool virtualize_apic_accesses_incorrectly_on =
6063 		apic_reg_virt_config->virtualize_apic_accesses &&
6064 		apic_reg_virt_config->virtualize_x2apic_mode &&
6065 		apic_reg_virt_config->activate_secondary_controls;
6066 	bool vmentry_fails_early =
6067 		use_tpr_shadow_incorrectly_off ||
6068 		virtualize_apic_accesses_incorrectly_on;
6069 
6070 	if (apic_reg_virt_config->activate_secondary_controls) {
6071 		if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
6072 			printf("VM-execution control \"activate secondary controls\" NOT supported.\n");
6073 			return CONFIG_TYPE_UNSUPPORTED;
6074 		}
6075 		cpu_exec_ctrl0 |= CPU_SECONDARY;
6076 	} else {
6077 		cpu_exec_ctrl0 &= ~CPU_SECONDARY;
6078 	}
6079 
6080 	if (apic_reg_virt_config->virtualize_apic_accesses) {
6081 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES)) {
6082 			printf("VM-execution control \"virtualize APIC accesses\" NOT supported.\n");
6083 			return CONFIG_TYPE_UNSUPPORTED;
6084 		}
6085 		cpu_exec_ctrl1 |= CPU_VIRT_APIC_ACCESSES;
6086 	} else {
6087 		cpu_exec_ctrl1 &= ~CPU_VIRT_APIC_ACCESSES;
6088 	}
6089 
6090 	if (apic_reg_virt_config->use_tpr_shadow) {
6091 		if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6092 			printf("VM-execution control \"use TPR shadow\" NOT supported.\n");
6093 			return CONFIG_TYPE_UNSUPPORTED;
6094 		}
6095 		cpu_exec_ctrl0 |= CPU_TPR_SHADOW;
6096 	} else {
6097 		cpu_exec_ctrl0 &= ~CPU_TPR_SHADOW;
6098 	}
6099 
6100 	if (apic_reg_virt_config->apic_register_virtualization) {
6101 		if (!(ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT)) {
6102 			printf("VM-execution control \"APIC-register virtualization\" NOT supported.\n");
6103 			return CONFIG_TYPE_UNSUPPORTED;
6104 		}
6105 		cpu_exec_ctrl1 |= CPU_APIC_REG_VIRT;
6106 	} else {
6107 		cpu_exec_ctrl1 &= ~CPU_APIC_REG_VIRT;
6108 	}
6109 
6110 	if (apic_reg_virt_config->virtualize_x2apic_mode) {
6111 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_X2APIC)) {
6112 			printf("VM-execution control \"virtualize x2APIC mode\" NOT supported.\n");
6113 			return CONFIG_TYPE_UNSUPPORTED;
6114 		}
6115 		cpu_exec_ctrl1 |= CPU_VIRT_X2APIC;
6116 	} else {
6117 		cpu_exec_ctrl1 &= ~CPU_VIRT_X2APIC;
6118 	}
6119 
6120 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6121 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6122 
6123 	if (vmentry_fails_early)
6124 		return CONFIG_TYPE_VMENTRY_FAILS_EARLY;
6125 
6126 	return CONFIG_TYPE_GOOD;
6127 }
6128 
cpu_has_apicv(void)6129 static bool cpu_has_apicv(void)
6130 {
6131 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
6132 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
6133 		(ctrl_pin_rev.clr & PIN_POST_INTR));
6134 }
6135 
6136 /* Validates APIC register access across valid virtualization configurations. */
apic_reg_virt_test(void)6137 static void apic_reg_virt_test(void)
6138 {
6139 	u32 *apic_access_address;
6140 	u32 *virtual_apic_page;
6141 	u64 control;
6142 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6143 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6144 	int i;
6145 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
6146 
6147 	if (!cpu_has_apicv()) {
6148 		report_skip("%s : Not all required APICv bits supported", __func__);
6149 		return;
6150 	}
6151 
6152 	control = cpu_exec_ctrl1;
6153 	control &= ~CPU_VINTD;
6154 	vmcs_write(CPU_EXEC_CTRL1, control);
6155 
6156 	test_set_guest(apic_reg_virt_guest);
6157 
6158 	/*
6159 	 * From the SDM: The 1-setting of the "virtualize APIC accesses"
6160 	 * VM-execution is guaranteed to apply only if translations to the
6161 	 * APIC-access address use a 4-KByte page.
6162 	 */
6163 	apic_access_address = alloc_page();
6164 	force_4k_page(apic_access_address);
6165 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_address));
6166 
6167 	virtual_apic_page = alloc_page();
6168 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6169 
6170 	for (i = 0; i < ARRAY_SIZE(apic_reg_tests); i++) {
6171 		struct apic_reg_test *apic_reg_test = &apic_reg_tests[i];
6172 		struct apic_reg_virt_config *apic_reg_virt_config =
6173 				&apic_reg_test->apic_reg_virt_config;
6174 		enum Config_type config_type;
6175 		u32 reg;
6176 
6177 		printf("--- %s test ---\n", apic_reg_test->name);
6178 		config_type =
6179 			configure_apic_reg_virt_test(apic_reg_virt_config);
6180 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6181 			printf("Skip because of missing features.\n");
6182 			continue;
6183 		}
6184 
6185 		if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6186 			enter_guest_with_bad_controls();
6187 			continue;
6188 		}
6189 
6190 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6191 			struct apic_reg_virt_expectation expectation = {};
6192 			bool ok;
6193 
6194 			ok = apic_reg_virt_exit_expectation(
6195 				reg, apic_reg_virt_config, &expectation);
6196 			if (!ok) {
6197 				report_fail("Malformed test.");
6198 				break;
6199 			}
6200 
6201 			test_xapic_rd(reg, &expectation, apic_access_address,
6202 				      virtual_apic_page);
6203 			test_xapic_wr(reg, &expectation, apic_access_address,
6204 				      virtual_apic_page);
6205 		}
6206 	}
6207 
6208 	/* Terminate the guest */
6209 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6210 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6211 	args->op = TERMINATE;
6212 	enter_guest();
6213 	assert_exit_reason(VMX_VMCALL);
6214 }
6215 
6216 struct virt_x2apic_mode_config {
6217 	struct apic_reg_virt_config apic_reg_virt_config;
6218 	bool virtual_interrupt_delivery;
6219 	bool use_msr_bitmaps;
6220 	bool disable_x2apic_msr_intercepts;
6221 	bool disable_x2apic;
6222 };
6223 
6224 struct virt_x2apic_mode_test_case {
6225 	const char *name;
6226 	struct virt_x2apic_mode_config virt_x2apic_mode_config;
6227 };
6228 
6229 enum Virt_x2apic_mode_behavior_type {
6230 	X2APIC_ACCESS_VIRTUALIZED,
6231 	X2APIC_ACCESS_PASSED_THROUGH,
6232 	X2APIC_ACCESS_TRIGGERS_GP,
6233 };
6234 
6235 struct virt_x2apic_mode_expectation {
6236 	enum Reason rd_exit_reason;
6237 	enum Reason wr_exit_reason;
6238 
6239 	/*
6240 	 * RDMSR and WRMSR handle 64-bit values. However, except for ICR, all of
6241 	 * the x2APIC registers are 32 bits. Notice:
6242 	 *   1. vmx_x2apic_read() clears the upper 32 bits for 32-bit registers.
6243 	 *   2. vmx_x2apic_write() expects the val arg to be well-formed.
6244 	 */
6245 	u64 rd_val;
6246 	u64 wr_val;
6247 
6248 	/*
6249 	 * Compares input to virtualized output;
6250 	 * 1st arg is pointer to return expected virtualization output.
6251 	 */
6252 	u64 (*virt_fn)(u64);
6253 
6254 	enum Virt_x2apic_mode_behavior_type rd_behavior;
6255 	enum Virt_x2apic_mode_behavior_type wr_behavior;
6256 	bool wr_only;
6257 };
6258 
virt_x2apic_mode_identity(u64 val)6259 static u64 virt_x2apic_mode_identity(u64 val)
6260 {
6261 	return val;
6262 }
6263 
virt_x2apic_mode_nibble1(u64 val)6264 static u64 virt_x2apic_mode_nibble1(u64 val)
6265 {
6266 	return val & 0xf0;
6267 }
6268 
virt_x2apic_mode_rd_expectation(u32 reg,bool virt_x2apic_mode_on,bool disable_x2apic,bool apic_register_virtualization,bool virtual_interrupt_delivery,struct virt_x2apic_mode_expectation * expectation)6269 static void virt_x2apic_mode_rd_expectation(
6270 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6271 	bool apic_register_virtualization, bool virtual_interrupt_delivery,
6272 	struct virt_x2apic_mode_expectation *expectation)
6273 {
6274 	enum x2apic_reg_semantics semantics = get_x2apic_reg_semantics(reg);
6275 
6276 	expectation->rd_exit_reason = VMX_VMCALL;
6277 	expectation->virt_fn = virt_x2apic_mode_identity;
6278 	if (virt_x2apic_mode_on && apic_register_virtualization) {
6279 		expectation->rd_val = MAGIC_VAL_1;
6280 		if (reg == APIC_PROCPRI && virtual_interrupt_delivery)
6281 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6282 		else if (reg == APIC_TASKPRI)
6283 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6284 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6285 	} else if (virt_x2apic_mode_on && !apic_register_virtualization &&
6286 		   reg == APIC_TASKPRI) {
6287 		expectation->rd_val = MAGIC_VAL_1;
6288 		expectation->virt_fn = virt_x2apic_mode_nibble1;
6289 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6290 	} else if (!disable_x2apic && (semantics & X2APIC_READABLE)) {
6291 		expectation->rd_val = apic_read(reg);
6292 		expectation->rd_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6293 	} else {
6294 		expectation->rd_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6295 	}
6296 }
6297 
6298 /*
6299  * get_x2apic_wr_val() creates an innocuous write value for an x2APIC register.
6300  *
6301  * For writable registers, get_x2apic_wr_val() deposits the write value into the
6302  * val pointer arg and returns true. For non-writable registers, val is not
6303  * modified and get_x2apic_wr_val() returns false.
6304  */
get_x2apic_wr_val(u32 reg,u64 * val)6305 static bool get_x2apic_wr_val(u32 reg, u64 *val)
6306 {
6307 	switch (reg) {
6308 	case APIC_TASKPRI:
6309 		/* Bits 31:8 are reserved. */
6310 		*val &= 0xff;
6311 		break;
6312 	case APIC_EOI:
6313 	case APIC_ESR:
6314 	case APIC_TMICT:
6315 		/*
6316 		 * EOI, ESR: WRMSR of a non-zero value causes #GP(0).
6317 		 * TMICT: A write of 0 to the initial-count register effectively
6318 		 *        stops the local APIC timer, in both one-shot and
6319 		 *        periodic mode.
6320 		 */
6321 		*val = 0;
6322 		break;
6323 	case APIC_SPIV:
6324 	case APIC_LVTT:
6325 	case APIC_LVTTHMR:
6326 	case APIC_LVTPC:
6327 	case APIC_LVT0:
6328 	case APIC_LVT1:
6329 	case APIC_LVTERR:
6330 	case APIC_TDCR:
6331 		/*
6332 		 * To avoid writing a 1 to a reserved bit or causing some other
6333 		 * unintended side effect, read the current value and use it as
6334 		 * the write value.
6335 		 */
6336 		*val = apic_read(reg);
6337 		break;
6338 	case APIC_CMCI:
6339 		if (!apic_lvt_entry_supported(6))
6340 			return false;
6341 		*val = apic_read(reg);
6342 		break;
6343 	case APIC_ICR:
6344 		*val = 0x40000 | 0xf1;
6345 		break;
6346 	case APIC_SELF_IPI:
6347 		/*
6348 		 * With special processing (i.e., virtualize x2APIC mode +
6349 		 * virtual interrupt delivery), writing zero causes an
6350 		 * APIC-write VM exit. We plan to add a test for enabling
6351 		 * "virtual-interrupt delivery" in VMCS12, and that's where we
6352 		 * will test a self IPI with special processing.
6353 		 */
6354 		*val = 0x0;
6355 		break;
6356 	default:
6357 		return false;
6358 	}
6359 
6360 	return true;
6361 }
6362 
special_processing_applies(u32 reg,u64 * val,bool virt_int_delivery)6363 static bool special_processing_applies(u32 reg, u64 *val,
6364 				       bool virt_int_delivery)
6365 {
6366 	bool special_processing =
6367 		(reg == APIC_TASKPRI) ||
6368 		(virt_int_delivery &&
6369 		 (reg == APIC_EOI || reg == APIC_SELF_IPI));
6370 
6371 	if (special_processing) {
6372 		TEST_ASSERT(get_x2apic_wr_val(reg, val));
6373 		return true;
6374 	}
6375 
6376 	return false;
6377 }
6378 
virt_x2apic_mode_wr_expectation(u32 reg,bool virt_x2apic_mode_on,bool disable_x2apic,bool virt_int_delivery,struct virt_x2apic_mode_expectation * expectation)6379 static void virt_x2apic_mode_wr_expectation(
6380 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6381 	bool virt_int_delivery,
6382 	struct virt_x2apic_mode_expectation *expectation)
6383 {
6384 	expectation->wr_exit_reason = VMX_VMCALL;
6385 	expectation->wr_val = MAGIC_VAL_1;
6386 	expectation->wr_only = false;
6387 
6388 	if (virt_x2apic_mode_on &&
6389 	    special_processing_applies(reg, &expectation->wr_val,
6390 				       virt_int_delivery)) {
6391 		expectation->wr_behavior = X2APIC_ACCESS_VIRTUALIZED;
6392 		if (reg == APIC_SELF_IPI)
6393 			expectation->wr_exit_reason = VMX_APIC_WRITE;
6394 	} else if (!disable_x2apic &&
6395 		   get_x2apic_wr_val(reg, &expectation->wr_val)) {
6396 		expectation->wr_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6397 		if (reg == APIC_EOI || reg == APIC_SELF_IPI)
6398 			expectation->wr_only = true;
6399 		if (reg == APIC_ICR)
6400 			expectation->wr_exit_reason = VMX_EXTINT;
6401 	} else {
6402 		expectation->wr_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6403 		/*
6404 		 * Writing 1 to a reserved bit triggers a #GP.
6405 		 * Thus, set the write value to 0, which seems
6406 		 * the most likely to detect a missed #GP.
6407 		 */
6408 		expectation->wr_val = 0;
6409 	}
6410 }
6411 
virt_x2apic_mode_exit_expectation(u32 reg,struct virt_x2apic_mode_config * config,struct virt_x2apic_mode_expectation * expectation)6412 static void virt_x2apic_mode_exit_expectation(
6413 	u32 reg, struct virt_x2apic_mode_config *config,
6414 	struct virt_x2apic_mode_expectation *expectation)
6415 {
6416 	struct apic_reg_virt_config *base_config =
6417 		&config->apic_reg_virt_config;
6418 	bool virt_x2apic_mode_on =
6419 		base_config->virtualize_x2apic_mode &&
6420 		config->use_msr_bitmaps &&
6421 		config->disable_x2apic_msr_intercepts &&
6422 		base_config->activate_secondary_controls;
6423 
6424 	virt_x2apic_mode_wr_expectation(
6425 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6426 		config->virtual_interrupt_delivery, expectation);
6427 	virt_x2apic_mode_rd_expectation(
6428 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6429 		base_config->apic_register_virtualization,
6430 		config->virtual_interrupt_delivery, expectation);
6431 }
6432 
6433 struct virt_x2apic_mode_test_case virt_x2apic_mode_tests[] = {
6434 	/*
6435 	 * Baseline "virtualize x2APIC mode" configuration:
6436 	 *   - virtualize x2APIC mode
6437 	 *   - virtual-interrupt delivery
6438 	 *   - APIC-register virtualization
6439 	 *   - x2APIC MSR intercepts disabled
6440 	 *
6441 	 * Reads come from virtual APIC page, special processing applies to
6442 	 * VTPR, EOI, and SELF IPI, and all other writes pass through to L1
6443 	 * APIC.
6444 	 */
6445 	{
6446 		.name = "Baseline",
6447 		.virt_x2apic_mode_config = {
6448 			.virtual_interrupt_delivery = true,
6449 			.use_msr_bitmaps = true,
6450 			.disable_x2apic_msr_intercepts = true,
6451 			.disable_x2apic = false,
6452 			.apic_reg_virt_config = {
6453 				.apic_register_virtualization = true,
6454 				.use_tpr_shadow = true,
6455 				.virtualize_apic_accesses = false,
6456 				.virtualize_x2apic_mode = true,
6457 				.activate_secondary_controls = true,
6458 			},
6459 		},
6460 	},
6461 	{
6462 		.name = "Baseline w/ x2apic disabled",
6463 		.virt_x2apic_mode_config = {
6464 			.virtual_interrupt_delivery = true,
6465 			.use_msr_bitmaps = true,
6466 			.disable_x2apic_msr_intercepts = true,
6467 			.disable_x2apic = true,
6468 			.apic_reg_virt_config = {
6469 				.apic_register_virtualization = true,
6470 				.use_tpr_shadow = true,
6471 				.virtualize_apic_accesses = false,
6472 				.virtualize_x2apic_mode = true,
6473 				.activate_secondary_controls = true,
6474 			},
6475 		},
6476 	},
6477 
6478 	/*
6479 	 * Baseline, minus virtual-interrupt delivery. Reads come from virtual
6480 	 * APIC page, special processing applies to VTPR, and all other writes
6481 	 * pass through to L1 APIC.
6482 	 */
6483 	{
6484 		.name = "Baseline - virtual interrupt delivery",
6485 		.virt_x2apic_mode_config = {
6486 			.virtual_interrupt_delivery = false,
6487 			.use_msr_bitmaps = true,
6488 			.disable_x2apic_msr_intercepts = true,
6489 			.disable_x2apic = false,
6490 			.apic_reg_virt_config = {
6491 				.apic_register_virtualization = true,
6492 				.use_tpr_shadow = true,
6493 				.virtualize_apic_accesses = false,
6494 				.virtualize_x2apic_mode = true,
6495 				.activate_secondary_controls = true,
6496 			},
6497 		},
6498 	},
6499 
6500 	/*
6501 	 * Baseline, minus APIC-register virtualization. x2APIC reads pass
6502 	 * through to L1's APIC, unless reading VTPR
6503 	 */
6504 	{
6505 		.name = "Virtualize x2APIC mode, no APIC reg virt",
6506 		.virt_x2apic_mode_config = {
6507 			.virtual_interrupt_delivery = true,
6508 			.use_msr_bitmaps = true,
6509 			.disable_x2apic_msr_intercepts = true,
6510 			.disable_x2apic = false,
6511 			.apic_reg_virt_config = {
6512 				.apic_register_virtualization = false,
6513 				.use_tpr_shadow = true,
6514 				.virtualize_apic_accesses = false,
6515 				.virtualize_x2apic_mode = true,
6516 				.activate_secondary_controls = true,
6517 			},
6518 		},
6519 	},
6520 	{
6521 		.name = "Virtualize x2APIC mode, no APIC reg virt, x2APIC off",
6522 		.virt_x2apic_mode_config = {
6523 			.virtual_interrupt_delivery = true,
6524 			.use_msr_bitmaps = true,
6525 			.disable_x2apic_msr_intercepts = true,
6526 			.disable_x2apic = true,
6527 			.apic_reg_virt_config = {
6528 				.apic_register_virtualization = false,
6529 				.use_tpr_shadow = true,
6530 				.virtualize_apic_accesses = false,
6531 				.virtualize_x2apic_mode = true,
6532 				.activate_secondary_controls = true,
6533 			},
6534 		},
6535 	},
6536 
6537 	/*
6538 	 * Enable "virtualize x2APIC mode" and "APIC-register virtualization",
6539 	 * and disable intercepts for the x2APIC MSRs, but fail to enable
6540 	 * "activate secondary controls" (i.e. L2 gets access to L1's x2APIC
6541 	 * MSRs).
6542 	 */
6543 	{
6544 		.name = "Fail to enable activate secondary controls",
6545 		.virt_x2apic_mode_config = {
6546 			.virtual_interrupt_delivery = true,
6547 			.use_msr_bitmaps = true,
6548 			.disable_x2apic_msr_intercepts = true,
6549 			.disable_x2apic = false,
6550 			.apic_reg_virt_config = {
6551 				.apic_register_virtualization = true,
6552 				.use_tpr_shadow = true,
6553 				.virtualize_apic_accesses = false,
6554 				.virtualize_x2apic_mode = true,
6555 				.activate_secondary_controls = false,
6556 			},
6557 		},
6558 	},
6559 
6560 	/*
6561 	 * Enable "APIC-register virtualization" and enable "activate secondary
6562 	 * controls" and disable intercepts for the x2APIC MSRs, but do not
6563 	 * enable the "virtualize x2APIC mode" VM-execution control (i.e. L2
6564 	 * gets access to L1's x2APIC MSRs).
6565 	 */
6566 	{
6567 		.name = "Fail to enable virtualize x2APIC mode",
6568 		.virt_x2apic_mode_config = {
6569 			.virtual_interrupt_delivery = true,
6570 			.use_msr_bitmaps = true,
6571 			.disable_x2apic_msr_intercepts = true,
6572 			.disable_x2apic = false,
6573 			.apic_reg_virt_config = {
6574 				.apic_register_virtualization = true,
6575 				.use_tpr_shadow = true,
6576 				.virtualize_apic_accesses = false,
6577 				.virtualize_x2apic_mode = false,
6578 				.activate_secondary_controls = true,
6579 			},
6580 		},
6581 	},
6582 
6583 	/*
6584 	 * Disable "Virtualize x2APIC mode", disable x2APIC MSR intercepts, and
6585 	 * enable "APIC-register virtualization" --> L2 gets L1's x2APIC MSRs.
6586 	 */
6587 	{
6588 		.name = "Baseline",
6589 		.virt_x2apic_mode_config = {
6590 			.virtual_interrupt_delivery = true,
6591 			.use_msr_bitmaps = true,
6592 			.disable_x2apic_msr_intercepts = true,
6593 			.disable_x2apic = false,
6594 			.apic_reg_virt_config = {
6595 				.apic_register_virtualization = true,
6596 				.use_tpr_shadow = true,
6597 				.virtualize_apic_accesses = false,
6598 				.virtualize_x2apic_mode = false,
6599 				.activate_secondary_controls = true,
6600 			},
6601 		},
6602 	},
6603 };
6604 
6605 enum X2apic_op {
6606 	X2APIC_OP_RD,
6607 	X2APIC_OP_WR,
6608 	X2APIC_TERMINATE,
6609 };
6610 
vmx_x2apic_read(u32 reg)6611 static u64 vmx_x2apic_read(u32 reg)
6612 {
6613 	u32 msr_addr = x2apic_msr(reg);
6614 	u64 val;
6615 
6616 	val = rdmsr(msr_addr);
6617 
6618 	return val;
6619 }
6620 
vmx_x2apic_write(u32 reg,u64 val)6621 static void vmx_x2apic_write(u32 reg, u64 val)
6622 {
6623 	u32 msr_addr = x2apic_msr(reg);
6624 
6625 	wrmsr(msr_addr, val);
6626 }
6627 
6628 struct virt_x2apic_mode_guest_args {
6629 	enum X2apic_op op;
6630 	u32 reg;
6631 	u64 val;
6632 	bool should_gp;
6633 	u64 (*virt_fn)(u64);
6634 } virt_x2apic_mode_guest_args;
6635 
6636 static volatile bool handle_x2apic_gp_ran;
6637 static volatile u32 handle_x2apic_gp_insn_len;
handle_x2apic_gp(struct ex_regs * regs)6638 static void handle_x2apic_gp(struct ex_regs *regs)
6639 {
6640 	handle_x2apic_gp_ran = true;
6641 	regs->rip += handle_x2apic_gp_insn_len;
6642 }
6643 
setup_x2apic_gp_handler(void)6644 static handler setup_x2apic_gp_handler(void)
6645 {
6646 	handler old_handler;
6647 
6648 	old_handler = handle_exception(GP_VECTOR, handle_x2apic_gp);
6649 	/* RDMSR and WRMSR are both 2 bytes, assuming no prefixes. */
6650 	handle_x2apic_gp_insn_len = 2;
6651 
6652 	return old_handler;
6653 }
6654 
teardown_x2apic_gp_handler(handler old_handler)6655 static void teardown_x2apic_gp_handler(handler old_handler)
6656 {
6657 	handle_exception(GP_VECTOR, old_handler);
6658 
6659 	/*
6660 	 * Defensively reset instruction length, so that if the handler is
6661 	 * incorrectly used, it will loop infinitely, rather than run off into
6662 	 * la la land.
6663 	 */
6664 	handle_x2apic_gp_insn_len = 0;
6665 	handle_x2apic_gp_ran = false;
6666 }
6667 
virt_x2apic_mode_guest(void)6668 static void virt_x2apic_mode_guest(void)
6669 {
6670 	volatile struct virt_x2apic_mode_guest_args *args =
6671 		&virt_x2apic_mode_guest_args;
6672 
6673 	for (;;) {
6674 		enum X2apic_op op = args->op;
6675 		u32 reg = args->reg;
6676 		u64 val = args->val;
6677 		bool should_gp = args->should_gp;
6678 		u64 (*virt_fn)(u64) = args->virt_fn;
6679 		handler old_handler;
6680 
6681 		if (op == X2APIC_TERMINATE)
6682 			break;
6683 
6684 		if (should_gp) {
6685 			TEST_ASSERT(!handle_x2apic_gp_ran);
6686 			old_handler = setup_x2apic_gp_handler();
6687 		}
6688 
6689 		if (op == X2APIC_OP_RD) {
6690 			u64 ret = vmx_x2apic_read(reg);
6691 
6692 			if (!should_gp) {
6693 				u64 want = virt_fn(val);
6694 				u64 got = virt_fn(ret);
6695 
6696 				report(got == want,
6697 				       "APIC read; got 0x%lx, want 0x%lx.",
6698 				       got, want);
6699 			}
6700 		} else if (op == X2APIC_OP_WR) {
6701 			vmx_x2apic_write(reg, val);
6702 		}
6703 
6704 		if (should_gp) {
6705 			report(handle_x2apic_gp_ran,
6706 			       "x2APIC op triggered GP.");
6707 			teardown_x2apic_gp_handler(old_handler);
6708 		}
6709 
6710 		/*
6711 		 * The L1 should always execute a vmcall after it's done testing
6712 		 * an individual APIC operation. This helps to validate that the
6713 		 * L1 and L2 are in sync with each other, as expected.
6714 		 */
6715 		vmcall();
6716 	}
6717 }
6718 
test_x2apic_rd(u32 reg,struct virt_x2apic_mode_expectation * expectation,u32 * virtual_apic_page)6719 static void test_x2apic_rd(
6720 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6721 	u32 *virtual_apic_page)
6722 {
6723 	u64 val = expectation->rd_val;
6724 	u32 exit_reason_want = expectation->rd_exit_reason;
6725 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6726 
6727 	report_prefix_pushf("x2apic - reading 0x%03x", reg);
6728 
6729 	/* Configure guest to do an x2apic read */
6730 	args->op = X2APIC_OP_RD;
6731 	args->reg = reg;
6732 	args->val = val;
6733 	args->should_gp = expectation->rd_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6734 	args->virt_fn = expectation->virt_fn;
6735 
6736 	/* Setup virtual APIC page */
6737 	if (expectation->rd_behavior == X2APIC_ACCESS_VIRTUALIZED)
6738 		virtual_apic_page[apic_reg_index(reg)] = (u32)val;
6739 
6740 	/* Enter guest */
6741 	enter_guest();
6742 
6743 	if (exit_reason_want != VMX_VMCALL) {
6744 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6745 	}
6746 
6747 	skip_exit_vmcall();
6748 	report_prefix_pop();
6749 }
6750 
6751 static volatile bool handle_x2apic_ipi_ran;
handle_x2apic_ipi(isr_regs_t * regs)6752 static void handle_x2apic_ipi(isr_regs_t *regs)
6753 {
6754 	handle_x2apic_ipi_ran = true;
6755 	eoi();
6756 }
6757 
test_x2apic_wr(u32 reg,struct virt_x2apic_mode_expectation * expectation,u32 * virtual_apic_page)6758 static void test_x2apic_wr(
6759 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6760 	u32 *virtual_apic_page)
6761 {
6762 	u64 val = expectation->wr_val;
6763 	u32 exit_reason_want = expectation->wr_exit_reason;
6764 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6765 	int ipi_vector = 0xf1;
6766 	u32 restore_val = 0;
6767 
6768 	report_prefix_pushf("x2apic - writing 0x%lx to 0x%03x", val, reg);
6769 
6770 	/* Configure guest to do an x2apic read */
6771 	args->op = X2APIC_OP_WR;
6772 	args->reg = reg;
6773 	args->val = val;
6774 	args->should_gp = expectation->wr_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6775 
6776 	/* Setup virtual APIC page */
6777 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED)
6778 		virtual_apic_page[apic_reg_index(reg)] = 0;
6779 	if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH && !expectation->wr_only)
6780 		restore_val = apic_read(reg);
6781 
6782 	/* Setup IPI handler */
6783 	handle_x2apic_ipi_ran = false;
6784 	handle_irq(ipi_vector, handle_x2apic_ipi);
6785 
6786 	/* Enter guest */
6787 	enter_guest();
6788 
6789 	/*
6790 	 * Validate the behavior and
6791 	 * pass a magic value back to the guest.
6792 	 */
6793 	if (exit_reason_want == VMX_EXTINT) {
6794 		assert_exit_reason(exit_reason_want);
6795 
6796 		/* Clear the external interrupt. */
6797 		sti_nop_cli();
6798 		report(handle_x2apic_ipi_ran,
6799 		       "Got pending interrupt after IRQ enabled.");
6800 
6801 		enter_guest();
6802 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6803 		assert_exit_reason(exit_reason_want);
6804 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6805 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%lx",
6806 		       apic_reg_index(reg),
6807 		       virtual_apic_page[apic_reg_index(reg)], val);
6808 
6809 		/* Reenter guest so it can consume/check rcx and exit again. */
6810 		enter_guest();
6811 	} else if (exit_reason_want != VMX_VMCALL) {
6812 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6813 	}
6814 
6815 	assert_exit_reason(VMX_VMCALL);
6816 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) {
6817 		u64 want = val;
6818 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6819 
6820 		report(got == want, "x2APIC write; got 0x%x, want 0x%lx", got,
6821 		       want);
6822 	} else if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH) {
6823 		if (!expectation->wr_only) {
6824 			u32 got = apic_read(reg);
6825 			bool ok;
6826 
6827 			/*
6828 			 * When L1's TPR is passed through to L2, the lower
6829 			 * nibble can be lost. For example, if L2 executes
6830 			 * WRMSR(0x808, 0x78), then, L1 might read 0x70.
6831 			 *
6832 			 * Here's how the lower nibble can get lost:
6833 			 *   1. L2 executes WRMSR(0x808, 0x78).
6834 			 *   2. L2 exits to L0 with a WRMSR exit.
6835 			 *   3. L0 emulates WRMSR, by writing L1's TPR.
6836 			 *   4. L0 re-enters L2.
6837 			 *   5. L2 exits to L0 (reason doesn't matter).
6838 			 *   6. L0 reflects L2's exit to L1.
6839 			 *   7. Before entering L1, L0 exits to user-space
6840 			 *      (e.g., to satisfy TPR access reporting).
6841 			 *   8. User-space executes KVM_SET_REGS ioctl, which
6842 			 *      clears the lower nibble of L1's TPR.
6843 			 */
6844 			if (reg == APIC_TASKPRI) {
6845 				got = apic_virt_nibble1(got);
6846 				val = apic_virt_nibble1(val);
6847 			}
6848 
6849 			ok = got == val;
6850 			report(ok,
6851 			       "non-virtualized write; val is 0x%x, want 0x%lx",
6852 			       got, val);
6853 			apic_write(reg, restore_val);
6854 		} else {
6855 			report_pass("non-virtualized and write-only OK");
6856 		}
6857 	}
6858 	skip_exit_insn();
6859 
6860 	report_prefix_pop();
6861 }
6862 
configure_virt_x2apic_mode_test(struct virt_x2apic_mode_config * virt_x2apic_mode_config,u8 * msr_bitmap_page)6863 static enum Config_type configure_virt_x2apic_mode_test(
6864 	struct virt_x2apic_mode_config *virt_x2apic_mode_config,
6865 	u8 *msr_bitmap_page)
6866 {
6867 	int msr;
6868 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6869 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6870 
6871 	/* x2apic-specific VMCS config */
6872 	if (virt_x2apic_mode_config->use_msr_bitmaps) {
6873 		/* virt_x2apic_mode_test() checks for MSR bitmaps support */
6874 		cpu_exec_ctrl0 |= CPU_MSR_BITMAP;
6875 	} else {
6876 		cpu_exec_ctrl0 &= ~CPU_MSR_BITMAP;
6877 	}
6878 
6879 	if (virt_x2apic_mode_config->virtual_interrupt_delivery) {
6880 		if (!(ctrl_cpu_rev[1].clr & CPU_VINTD)) {
6881 			report_skip("%s : \"virtual-interrupt delivery\" exec control not supported", __func__);
6882 			return CONFIG_TYPE_UNSUPPORTED;
6883 		}
6884 		cpu_exec_ctrl1 |= CPU_VINTD;
6885 	} else {
6886 		cpu_exec_ctrl1 &= ~CPU_VINTD;
6887 	}
6888 
6889 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6890 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6891 
6892 	/* x2APIC MSR intercepts are usually off for "Virtualize x2APIC mode" */
6893 	for (msr = 0x800; msr <= 0x8ff; msr++) {
6894 		if (virt_x2apic_mode_config->disable_x2apic_msr_intercepts) {
6895 			clear_bit(msr, msr_bitmap_page + 0x000);
6896 			clear_bit(msr, msr_bitmap_page + 0x800);
6897 		} else {
6898 			set_bit(msr, msr_bitmap_page + 0x000);
6899 			set_bit(msr, msr_bitmap_page + 0x800);
6900 		}
6901 	}
6902 
6903 	/* x2APIC mode can impact virtualization */
6904 	reset_apic();
6905 	if (!virt_x2apic_mode_config->disable_x2apic)
6906 		enable_x2apic();
6907 
6908 	return configure_apic_reg_virt_test(
6909 		&virt_x2apic_mode_config->apic_reg_virt_config);
6910 }
6911 
virt_x2apic_mode_test(void)6912 static void virt_x2apic_mode_test(void)
6913 {
6914 	u32 *virtual_apic_page;
6915 	u8 *msr_bitmap_page;
6916 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6917 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6918 	int i;
6919 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6920 
6921 	if (!cpu_has_apicv()) {
6922 		report_skip("%s : Not all required APICv bits supported", __func__);
6923 		return;
6924 	}
6925 
6926 	/*
6927 	 * This is to exercise an issue in KVM's logic to merge L0's and L1's
6928 	 * MSR bitmaps. Previously, an L1 could get at L0's x2APIC MSRs by
6929 	 * writing the IA32_SPEC_CTRL MSR or the IA32_PRED_CMD MSRs. KVM would
6930 	 * then proceed to manipulate the MSR bitmaps, as if VMCS12 had the
6931 	 * "Virtualize x2APIC mod" control set, even when it didn't.
6932 	 */
6933 	if (this_cpu_has(X86_FEATURE_SPEC_CTRL))
6934 		wrmsr(MSR_IA32_SPEC_CTRL, 1);
6935 
6936 	/*
6937 	 * Check that VMCS12 supports:
6938 	 *   - "Virtual-APIC address", indicated by "use TPR shadow"
6939 	 *   - "MSR-bitmap address", indicated by "use MSR bitmaps"
6940 	 */
6941 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6942 		report_skip("%s : \"Use TPR shadow\" exec control not supported", __func__);
6943 		return;
6944 	} else if (!(ctrl_cpu_rev[0].clr & CPU_MSR_BITMAP)) {
6945 		report_skip("%s : \"Use MSR bitmaps\" exec control not supported", __func__);
6946 		return;
6947 	}
6948 
6949 	test_set_guest(virt_x2apic_mode_guest);
6950 
6951 	virtual_apic_page = alloc_page();
6952 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6953 
6954 	msr_bitmap_page = alloc_page();
6955 	memset(msr_bitmap_page, 0xff, PAGE_SIZE);
6956 	vmcs_write(MSR_BITMAP, virt_to_phys(msr_bitmap_page));
6957 
6958 	for (i = 0; i < ARRAY_SIZE(virt_x2apic_mode_tests); i++) {
6959 		struct virt_x2apic_mode_test_case *virt_x2apic_mode_test_case =
6960 			&virt_x2apic_mode_tests[i];
6961 		struct virt_x2apic_mode_config *virt_x2apic_mode_config =
6962 			&virt_x2apic_mode_test_case->virt_x2apic_mode_config;
6963 		enum Config_type config_type;
6964 		u32 reg;
6965 
6966 		printf("--- %s test ---\n", virt_x2apic_mode_test_case->name);
6967 		config_type =
6968 			configure_virt_x2apic_mode_test(virt_x2apic_mode_config,
6969 							msr_bitmap_page);
6970 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6971 			report_skip("Skip because of missing features.");
6972 			continue;
6973 		} else if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6974 			enter_guest_with_bad_controls();
6975 			continue;
6976 		}
6977 
6978 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6979 			struct virt_x2apic_mode_expectation expectation;
6980 
6981 			virt_x2apic_mode_exit_expectation(
6982 				reg, virt_x2apic_mode_config, &expectation);
6983 
6984 			test_x2apic_rd(reg, &expectation, virtual_apic_page);
6985 			test_x2apic_wr(reg, &expectation, virtual_apic_page);
6986 		}
6987 	}
6988 
6989 
6990 	/* Terminate the guest */
6991 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6992 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6993 	args->op = X2APIC_TERMINATE;
6994 	enter_guest();
6995 	assert_exit_reason(VMX_VMCALL);
6996 }
6997 
test_ctl_reg(const char * cr_name,u64 cr,u64 fixed0,u64 fixed1)6998 static void test_ctl_reg(const char *cr_name, u64 cr, u64 fixed0, u64 fixed1)
6999 {
7000 	u64 val;
7001 	u64 cr_saved = vmcs_read(cr);
7002 	int i;
7003 
7004 	val = fixed0 & fixed1;
7005 	if (cr == HOST_CR4)
7006 		vmcs_write(cr, val | X86_CR4_PAE);
7007 	else
7008 		vmcs_write(cr, val);
7009 	report_prefix_pushf("%s %lx", cr_name, val);
7010 	if (val == fixed0)
7011 		test_vmx_vmlaunch(0);
7012 	else
7013 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7014 	report_prefix_pop();
7015 
7016 	for (i = 0; i < 64; i++) {
7017 
7018 		/* Set a bit when the corresponding bit in fixed1 is 0 */
7019 		if ((fixed1 & (1ull << i)) == 0) {
7020 			if (cr == HOST_CR4 && ((1ull << i) & X86_CR4_SMEP ||
7021 					       (1ull << i) & X86_CR4_SMAP))
7022 				continue;
7023 
7024 			vmcs_write(cr, cr_saved | (1ull << i));
7025 			report_prefix_pushf("%s %llx", cr_name,
7026 						cr_saved | (1ull << i));
7027 			test_vmx_vmlaunch(
7028 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7029 			report_prefix_pop();
7030 		}
7031 
7032 		/* Unset a bit when the corresponding bit in fixed0 is 1 */
7033 		if (fixed0 & (1ull << i)) {
7034 			vmcs_write(cr, cr_saved & ~(1ull << i));
7035 			report_prefix_pushf("%s %llx", cr_name,
7036 						cr_saved & ~(1ull << i));
7037 			test_vmx_vmlaunch(
7038 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7039 			report_prefix_pop();
7040 		}
7041 	}
7042 
7043 	vmcs_write(cr, cr_saved);
7044 }
7045 
7046 /*
7047  * 1. The CR0 field must not set any bit to a value not supported in VMX
7048  *    operation.
7049  * 2. The CR4 field must not set any bit to a value not supported in VMX
7050  *    operation.
7051  * 3. On processors that support Intel 64 architecture, the CR3 field must
7052  *    be such that bits 63:52 and bits in the range 51:32 beyond the
7053  *    processor's physical-address width must be 0.
7054  *
7055  *  [Intel SDM]
7056  */
test_host_ctl_regs(void)7057 static void test_host_ctl_regs(void)
7058 {
7059 	u64 fixed0, fixed1, cr3, cr3_saved;
7060 	int i;
7061 
7062 	/* Test CR0 */
7063 	fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0);
7064 	fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1);
7065 	test_ctl_reg("HOST_CR0", HOST_CR0, fixed0, fixed1);
7066 
7067 	/* Test CR4 */
7068 	fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0);
7069 	fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1) &
7070 		 ~(X86_CR4_SMEP | X86_CR4_SMAP);
7071 	test_ctl_reg("HOST_CR4", HOST_CR4, fixed0, fixed1);
7072 
7073 	/* Test CR3 */
7074 	cr3_saved = vmcs_read(HOST_CR3);
7075 	for (i = cpuid_maxphyaddr(); i < 64; i++) {
7076 		cr3 = cr3_saved | (1ul << i);
7077 		vmcs_write(HOST_CR3, cr3);
7078 		report_prefix_pushf("HOST_CR3 %lx", cr3);
7079 		if (this_cpu_has(X86_FEATURE_LAM) &&
7080 		    ((i == X86_CR3_LAM_U57_BIT) || (i == X86_CR3_LAM_U48_BIT)))
7081 			test_vmx_vmlaunch(0);
7082 		else
7083 			test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7084 		report_prefix_pop();
7085 	}
7086 
7087 	vmcs_write(HOST_CR3, cr3_saved);
7088 }
7089 
test_efer_vmlaunch(u32 fld,bool ok)7090 static void test_efer_vmlaunch(u32 fld, bool ok)
7091 {
7092 	if (fld == HOST_EFER) {
7093 		if (ok)
7094 			test_vmx_vmlaunch(0);
7095 		else
7096 			test_vmx_vmlaunch2(VMXERR_ENTRY_INVALID_CONTROL_FIELD,
7097 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7098 	} else {
7099 		test_guest_state("EFER test", !ok, GUEST_EFER, "GUEST_EFER");
7100 	}
7101 }
7102 
test_efer_one(u32 fld,const char * fld_name,u64 efer,u32 ctrl_fld,u64 ctrl,int i,const char * efer_bit_name)7103 static void test_efer_one(u32 fld, const char * fld_name, u64 efer,
7104 			  u32 ctrl_fld, u64 ctrl,
7105 			  int i, const char *efer_bit_name)
7106 {
7107 	bool ok;
7108 
7109 	ok = true;
7110 	if (ctrl_fld == EXI_CONTROLS && (ctrl & EXI_LOAD_EFER)) {
7111 		if (!!(efer & EFER_LMA) != !!(ctrl & EXI_HOST_64))
7112 			ok = false;
7113 		if (!!(efer & EFER_LME) != !!(ctrl & EXI_HOST_64))
7114 			ok = false;
7115 	}
7116 	if (ctrl_fld == ENT_CONTROLS && (ctrl & ENT_LOAD_EFER)) {
7117 		/* Check LMA too since CR0.PG is set.  */
7118 		if (!!(efer & EFER_LMA) != !!(ctrl & ENT_GUEST_64))
7119 			ok = false;
7120 		if (!!(efer & EFER_LME) != !!(ctrl & ENT_GUEST_64))
7121 			ok = false;
7122 	}
7123 
7124 	/*
7125 	 * Skip the test if it would enter the guest in 32-bit mode.
7126 	 * Perhaps write the test in assembly and make sure it
7127 	 * can be run in either mode?
7128 	 */
7129 	if (fld == GUEST_EFER && ok && !(ctrl & ENT_GUEST_64))
7130 		return;
7131 
7132 	vmcs_write(ctrl_fld, ctrl);
7133 	vmcs_write(fld, efer);
7134 	report_prefix_pushf("%s %s bit turned %s, controls %s",
7135 			    fld_name, efer_bit_name,
7136 			    (i & 1) ? "on" : "off",
7137 			    (i & 2) ? "on" : "off");
7138 
7139 	test_efer_vmlaunch(fld, ok);
7140 	report_prefix_pop();
7141 }
7142 
test_efer_bit(u32 fld,const char * fld_name,u32 ctrl_fld,u64 ctrl_bit,u64 efer_bit,const char * efer_bit_name)7143 static void test_efer_bit(u32 fld, const char * fld_name,
7144 			  u32 ctrl_fld, u64 ctrl_bit, u64 efer_bit,
7145 			  const char *efer_bit_name)
7146 {
7147 	u64 efer_saved = vmcs_read(fld);
7148 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7149 	int i;
7150 
7151 	for (i = 0; i < 4; i++) {
7152 		u64 efer = efer_saved & ~efer_bit;
7153 		u64 ctrl = ctrl_saved & ~ctrl_bit;
7154 
7155 		if (i & 1)
7156 			efer |= efer_bit;
7157 		if (i & 2)
7158 			ctrl |= ctrl_bit;
7159 
7160 		test_efer_one(fld, fld_name, efer, ctrl_fld, ctrl,
7161 			      i, efer_bit_name);
7162 	}
7163 
7164 	vmcs_write(ctrl_fld, ctrl_saved);
7165 	vmcs_write(fld, efer_saved);
7166 }
7167 
test_efer(u32 fld,const char * fld_name,u32 ctrl_fld,u64 ctrl_bit1,u64 ctrl_bit2)7168 static void test_efer(u32 fld, const char * fld_name, u32 ctrl_fld,
7169 		      u64 ctrl_bit1, u64 ctrl_bit2)
7170 {
7171 	u64 efer_saved = vmcs_read(fld);
7172 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7173 	u64 efer_reserved_bits =  ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
7174 	u64 i;
7175 	u64 efer;
7176 
7177 	if (this_cpu_has(X86_FEATURE_NX))
7178 		efer_reserved_bits &= ~EFER_NX;
7179 
7180 	if (!ctrl_bit1) {
7181 		report_skip("%s : \"Load-IA32-EFER\" exit control not supported", __func__);
7182 		goto test_entry_exit_mode;
7183 	}
7184 
7185 	report_prefix_pushf("%s %lx", fld_name, efer_saved);
7186 	test_efer_vmlaunch(fld, true);
7187 	report_prefix_pop();
7188 
7189 	/*
7190 	 * Check reserved bits
7191 	 */
7192 	vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit1);
7193 	for (i = 0; i < 64; i++) {
7194 		if ((1ull << i) & efer_reserved_bits) {
7195 			efer = efer_saved | (1ull << i);
7196 			vmcs_write(fld, efer);
7197 			report_prefix_pushf("%s %lx", fld_name, efer);
7198 			test_efer_vmlaunch(fld, true);
7199 			report_prefix_pop();
7200 		}
7201 	}
7202 
7203 	vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit1);
7204 	for (i = 0; i < 64; i++) {
7205 		if ((1ull << i) & efer_reserved_bits) {
7206 			efer = efer_saved | (1ull << i);
7207 			vmcs_write(fld, efer);
7208 			report_prefix_pushf("%s %lx", fld_name, efer);
7209 			test_efer_vmlaunch(fld, false);
7210 			report_prefix_pop();
7211 		}
7212 	}
7213 
7214 	vmcs_write(ctrl_fld, ctrl_saved);
7215 	vmcs_write(fld, efer_saved);
7216 
7217 	/*
7218 	 * Check LMA and LME bits
7219 	 */
7220 	test_efer_bit(fld, fld_name,
7221 		      ctrl_fld, ctrl_bit1,
7222 		      EFER_LMA,
7223 		      "EFER_LMA");
7224 	test_efer_bit(fld, fld_name,
7225 		      ctrl_fld, ctrl_bit1,
7226 		      EFER_LME,
7227 		      "EFER_LME");
7228 
7229 test_entry_exit_mode:
7230 	test_efer_bit(fld, fld_name,
7231 		      ctrl_fld, ctrl_bit2,
7232 		      EFER_LMA,
7233 		      "EFER_LMA");
7234 	test_efer_bit(fld, fld_name,
7235 		      ctrl_fld, ctrl_bit2,
7236 		      EFER_LME,
7237 		      "EFER_LME");
7238 }
7239 
7240 /*
7241  * If the 'load IA32_EFER' VM-exit control is 1, bits reserved in the
7242  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7243  * the values of the LMA and LME bits in the field must each be that of
7244  * the 'host address-space size' VM-exit control.
7245  *
7246  *  [Intel SDM]
7247  */
test_host_efer(void)7248 static void test_host_efer(void)
7249 {
7250 	test_efer(HOST_EFER, "HOST_EFER", EXI_CONTROLS,
7251 		  ctrl_exit_rev.clr & EXI_LOAD_EFER,
7252 		  EXI_HOST_64);
7253 }
7254 
7255 /*
7256  * If the 'load IA32_EFER' VM-enter control is 1, bits reserved in the
7257  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7258  * the values of the LMA and LME bits in the field must each be that of
7259  * the 'IA32e-mode guest' VM-exit control.
7260  */
test_guest_efer(void)7261 static void test_guest_efer(void)
7262 {
7263 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER)) {
7264 		report_skip("%s : \"Load-IA32-EFER\" entry control not supported", __func__);
7265 		return;
7266 	}
7267 
7268 	vmcs_write(GUEST_EFER, rdmsr(MSR_EFER));
7269 	test_efer(GUEST_EFER, "GUEST_EFER", ENT_CONTROLS,
7270 		  ctrl_enter_rev.clr & ENT_LOAD_EFER,
7271 		  ENT_GUEST_64);
7272 }
7273 
7274 /*
7275  * PAT values higher than 8 are uninteresting since they're likely lumped
7276  * in with "8". We only test values above 8 one bit at a time,
7277  * in order to reduce the number of VM-Entries and keep the runtime reasonable.
7278  */
7279 #define	PAT_VAL_LIMIT	8
7280 
test_pat(u32 field,const char * field_name,u32 ctrl_field,u64 ctrl_bit)7281 static void test_pat(u32 field, const char * field_name, u32 ctrl_field,
7282 		     u64 ctrl_bit)
7283 {
7284 	u64 pat_msr_saved = rdmsr(MSR_IA32_CR_PAT);
7285 	u32 ctrl_saved = vmcs_read(ctrl_field);
7286 	u64 pat_saved = vmcs_read(field);
7287 	u64 i, val;
7288 	u32 j;
7289 	int error;
7290 
7291 	vmcs_clear_bits(ctrl_field, ctrl_bit);
7292 
7293 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7294 		/* Test PAT0..PAT7 fields */
7295 		for (j = 0; j < (i ? 8 : 1); j++) {
7296 			val = i << j * 8;
7297 			vmcs_write(field, val);
7298 			if (field == HOST_PAT) {
7299 				report_prefix_pushf("%s %lx", field_name, val);
7300 				test_vmx_vmlaunch(0);
7301 				report_prefix_pop();
7302 
7303 			} else {	// GUEST_PAT
7304 				test_guest_state("ENT_LOAD_PAT disabled", false,
7305 						 val, "GUEST_PAT");
7306 			}
7307 		}
7308 	}
7309 
7310 	vmcs_set_bits(ctrl_field, ctrl_bit);
7311 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7312 		/* Test PAT0..PAT7 fields */
7313 		for (j = 0; j < (i ? 8 : 1); j++) {
7314 			val = i << j * 8;
7315 			vmcs_write(field, val);
7316 
7317 			if (field == HOST_PAT) {
7318 				report_prefix_pushf("%s %lx", field_name, val);
7319 				if (i == 0x2 || i == 0x3 || i >= 0x8)
7320 					error =
7321 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
7322 				else
7323 					error = 0;
7324 
7325 				test_vmx_vmlaunch(error);
7326 
7327 				if (!error)
7328 					report(rdmsr(MSR_IA32_CR_PAT) == val,
7329 					       "Expected PAT = 0x%lx, got 0x%lx",
7330 						val, rdmsr(MSR_IA32_CR_PAT));
7331 				wrmsr(MSR_IA32_CR_PAT, pat_msr_saved);
7332 
7333 				report_prefix_pop();
7334 
7335 			} else {	// GUEST_PAT
7336 				error = (i == 0x2 || i == 0x3 || i >= 0x8);
7337 				test_guest_state("ENT_LOAD_PAT enabled", !!error,
7338 						 val, "GUEST_PAT");
7339 
7340 				if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
7341 					wrmsr(MSR_IA32_CR_PAT, pat_msr_saved);
7342 			}
7343 
7344 		}
7345 	}
7346 
7347 	vmcs_write(ctrl_field, ctrl_saved);
7348 	vmcs_write(field, pat_saved);
7349 }
7350 
7351 /*
7352  *  If the "load IA32_PAT" VM-exit control is 1, the value of the field
7353  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7354  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7355  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7356  *  6 (WB), or 7 (UC-).
7357  *
7358  *  [Intel SDM]
7359  */
test_load_host_pat(void)7360 static void test_load_host_pat(void)
7361 {
7362 	/*
7363 	 * "load IA32_PAT" VM-exit control
7364 	 */
7365 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) {
7366 		report_skip("%s : \"Load-IA32-PAT\" exit control not supported", __func__);
7367 		return;
7368 	}
7369 
7370 	test_pat(HOST_PAT, "HOST_PAT", EXI_CONTROLS, EXI_LOAD_PAT);
7371 }
7372 
7373 union cpuidA_eax {
7374 	struct {
7375 		unsigned int version_id:8;
7376 		unsigned int num_counters_gp:8;
7377 		unsigned int bit_width:8;
7378 		unsigned int mask_length:8;
7379 	} split;
7380 	unsigned int full;
7381 };
7382 
7383 union cpuidA_ecx {
7384 	unsigned int fixed_counters_mask;
7385 	unsigned int full;
7386 };
7387 
7388 union cpuidA_edx {
7389 	struct {
7390 		unsigned int num_contiguous_fixed_counters:5;
7391 		unsigned int bit_width_fixed:8;
7392 		unsigned int reserved1:2;
7393 		unsigned int anythread_deprecated:1;
7394 		unsigned int reserved2:16;
7395 	} split;
7396 	unsigned int full;
7397 };
7398 
valid_pgc(u64 val)7399 static bool valid_pgc(u64 val)
7400 {
7401 	struct cpuid id;
7402 	union cpuidA_eax eax;
7403 	union cpuidA_ecx ecx;
7404 	union cpuidA_edx edx;
7405 	u64 mask;
7406 
7407 	id = cpuid(0xA);
7408 	eax.full = id.a;
7409 	ecx.full = id.c;
7410 	edx.full = id.d;
7411 
7412 	/* FxCtr[i]_is_supported := ECX[i] || (EDX[4:0] > i); */
7413 	mask = ~(((1ull << eax.split.num_counters_gp) - 1) |
7414 		 (((1ull << edx.split.num_contiguous_fixed_counters) - 1) << 32) |
7415 		 ((u64)ecx.fixed_counters_mask << 32));
7416 
7417 	return !(val & mask);
7418 }
7419 
test_pgc_vmlaunch(u32 xerror,u32 xreason,bool xfail,bool host)7420 static void test_pgc_vmlaunch(u32 xerror, u32 xreason, bool xfail, bool host)
7421 {
7422 	u32 inst_err;
7423 	u64 obs;
7424 	bool success;
7425 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7426 
7427 	if (host) {
7428 		success = vmlaunch();
7429 		obs = rdmsr(data->msr);
7430 		if (!success) {
7431 			inst_err = vmcs_read(VMX_INST_ERROR);
7432 			report(xerror == inst_err, "vmlaunch failed, "
7433 			       "VMX Inst Error is %d (expected %d)",
7434 			       inst_err, xerror);
7435 		} else {
7436 			report(!data->enabled || data->exp == obs,
7437 			       "Host state is 0x%lx (expected 0x%lx)",
7438 			       obs, data->exp);
7439 			report(success != xfail, "vmlaunch succeeded");
7440 		}
7441 	} else {
7442 		test_guest_state("load GUEST_PERF_GLOBAL_CTRL", xfail,
7443 				 GUEST_PERF_GLOBAL_CTRL,
7444 				 "GUEST_PERF_GLOBAL_CTRL");
7445 	}
7446 }
7447 
7448 /*
7449  * test_load_perf_global_ctrl is a generic function for testing the
7450  * "load IA32_PERF_GLOBAL_CTRL" VM-{Entry,Exit} controls. This test function
7451  * tests the provided ctrl_val when disabled and enabled.
7452  *
7453  * @nr: VMCS field number corresponding to the host/guest state field
7454  * @name: Name of the above VMCS field for printing in test report
7455  * @ctrl_nr: VMCS field number corresponding to the VM-{Entry,Exit} control
7456  * @ctrl_val: Bit to set on the ctrl_field
7457  */
test_perf_global_ctrl(u32 nr,const char * name,u32 ctrl_nr,const char * ctrl_name,u64 ctrl_val)7458 static void test_perf_global_ctrl(u32 nr, const char *name, u32 ctrl_nr,
7459 				  const char *ctrl_name, u64 ctrl_val)
7460 {
7461 	u64 ctrl_saved = vmcs_read(ctrl_nr);
7462 	u64 pgc_saved = vmcs_read(nr);
7463 	u64 i, val;
7464 	bool host = nr == HOST_PERF_GLOBAL_CTRL;
7465 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7466 
7467 	data->msr = MSR_CORE_PERF_GLOBAL_CTRL;
7468 	msr_bmp_init();
7469 	vmcs_write(ctrl_nr, ctrl_saved & ~ctrl_val);
7470 	data->enabled = false;
7471 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=0 on %s",
7472 			    ctrl_name);
7473 
7474 	for (i = 0; i < 64; i++) {
7475 		val = 1ull << i;
7476 		vmcs_write(nr, val);
7477 		report_prefix_pushf("%s = 0x%lx", name, val);
7478 		test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7479 		report_prefix_pop();
7480 	}
7481 	report_prefix_pop();
7482 
7483 	vmcs_write(ctrl_nr, ctrl_saved | ctrl_val);
7484 	data->enabled = true;
7485 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=1 on %s",
7486 			    ctrl_name);
7487 	for (i = 0; i < 64; i++) {
7488 		val = 1ull << i;
7489 		data->exp = val;
7490 		vmcs_write(nr, val);
7491 		report_prefix_pushf("%s = 0x%lx", name, val);
7492 		if (valid_pgc(val)) {
7493 			test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7494 		} else {
7495 			if (host)
7496 				test_pgc_vmlaunch(
7497 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
7498 					0,
7499 					true,
7500 					host);
7501 			else
7502 				test_pgc_vmlaunch(
7503 					0,
7504 					VMX_ENTRY_FAILURE | VMX_FAIL_STATE,
7505 					true,
7506 					host);
7507 		}
7508 		report_prefix_pop();
7509 	}
7510 
7511 	data->enabled = false;
7512 	report_prefix_pop();
7513 	vmcs_write(ctrl_nr, ctrl_saved);
7514 	vmcs_write(nr, pgc_saved);
7515 }
7516 
test_load_host_perf_global_ctrl(void)7517 static void test_load_host_perf_global_ctrl(void)
7518 {
7519 	if (!this_cpu_has_perf_global_ctrl()) {
7520 		report_skip("%s : \"IA32_PERF_GLOBAL_CTRL\" MSR not supported", __func__);
7521 		return;
7522 	}
7523 
7524 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PERF)) {
7525 		report_skip("%s : \"Load IA32_PERF_GLOBAL_CTRL\" exit control not supported", __func__);
7526 		return;
7527 	}
7528 
7529 	test_perf_global_ctrl(HOST_PERF_GLOBAL_CTRL, "HOST_PERF_GLOBAL_CTRL",
7530 				   EXI_CONTROLS, "EXI_CONTROLS", EXI_LOAD_PERF);
7531 }
7532 
7533 
test_load_guest_perf_global_ctrl(void)7534 static void test_load_guest_perf_global_ctrl(void)
7535 {
7536 	if (!this_cpu_has_perf_global_ctrl()) {
7537 		report_skip("%s : \"IA32_PERF_GLOBAL_CTRL\" MSR not supported", __func__);
7538 		return;
7539 	}
7540 
7541 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PERF)) {
7542 		report_skip("%s : \"Load IA32_PERF_GLOBAL_CTRL\" entry control not supported", __func__);
7543 		return;
7544 	}
7545 
7546 	test_perf_global_ctrl(GUEST_PERF_GLOBAL_CTRL, "GUEST_PERF_GLOBAL_CTRL",
7547 				   ENT_CONTROLS, "ENT_CONTROLS", ENT_LOAD_PERF);
7548 }
7549 
7550 
7551 /*
7552  * test_vmcs_field - test a value for the given VMCS field
7553  * @field: VMCS field
7554  * @field_name: string name of VMCS field
7555  * @bit_start: starting bit
7556  * @bit_end: ending bit
7557  * @val: value that the bit range must or must not contain
7558  * @valid_val: whether value given in 'val' must be valid or not
7559  * @error: expected VMCS error when vmentry fails for an invalid value
7560  */
test_vmcs_field(u64 field,const char * field_name,u32 bit_start,u32 bit_end,u64 val,bool valid_val,u32 error)7561 static void test_vmcs_field(u64 field, const char *field_name, u32 bit_start,
7562 			    u32 bit_end, u64 val, bool valid_val, u32 error)
7563 {
7564 	u64 field_saved = vmcs_read(field);
7565 	u32 i;
7566 	u64 tmp;
7567 	u32 bit_on;
7568 	u64 mask = ~0ull;
7569 
7570 	mask = (mask >> bit_end) << bit_end;
7571 	mask = mask | ((1 << bit_start) - 1);
7572 	tmp = (field_saved & mask) | (val << bit_start);
7573 
7574 	vmcs_write(field, tmp);
7575 	report_prefix_pushf("%s %lx", field_name, tmp);
7576 	if (valid_val)
7577 		test_vmx_vmlaunch(0);
7578 	else
7579 		test_vmx_vmlaunch(error);
7580 	report_prefix_pop();
7581 
7582 	for (i = bit_start; i <= bit_end; i = i + 2) {
7583 		bit_on = ((1ull < i) & (val << bit_start)) ? 0 : 1;
7584 		if (bit_on)
7585 			tmp = field_saved | (1ull << i);
7586 		else
7587 			tmp = field_saved & ~(1ull << i);
7588 		vmcs_write(field, tmp);
7589 		report_prefix_pushf("%s %lx", field_name, tmp);
7590 		if (valid_val)
7591 			test_vmx_vmlaunch(error);
7592 		else
7593 			test_vmx_vmlaunch(0);
7594 		report_prefix_pop();
7595 	}
7596 
7597 	vmcs_write(field, field_saved);
7598 }
7599 
test_canonical(u64 field,const char * field_name,bool host)7600 static void test_canonical(u64 field, const char * field_name, bool host)
7601 {
7602 	u64 addr_saved = vmcs_read(field);
7603 
7604 	/*
7605 	 * Use the existing value if possible.  Writing a random canonical
7606 	 * value is not an option as doing so would corrupt the field being
7607 	 * tested and likely hose the test.
7608 	 */
7609 	if (is_canonical(addr_saved)) {
7610 		if (host) {
7611 			report_prefix_pushf("%s %lx", field_name, addr_saved);
7612 			test_vmx_vmlaunch(0);
7613 			report_prefix_pop();
7614 		} else {
7615 			test_guest_state("Test canonical address", false,
7616 					 addr_saved, field_name);
7617 		}
7618 	}
7619 
7620 	vmcs_write(field, NONCANONICAL);
7621 
7622 	if (host) {
7623 		report_prefix_pushf("%s %llx", field_name, NONCANONICAL);
7624 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7625 		report_prefix_pop();
7626 	} else {
7627 		test_guest_state("Test non-canonical address", true,
7628 				 NONCANONICAL, field_name);
7629 	}
7630 
7631 	vmcs_write(field, addr_saved);
7632 }
7633 
7634 #define TEST_RPL_TI_FLAGS(reg, name)				\
7635 	test_vmcs_field(reg, name, 0, 2, 0x0, true,		\
7636 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7637 
7638 #define TEST_CS_TR_FLAGS(reg, name)				\
7639 	test_vmcs_field(reg, name, 3, 15, 0x0000, false,	\
7640 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7641 
7642 /*
7643  * 1. In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the
7644  *    RPL (bits 1:0) and the TI flag (bit 2) must be 0.
7645  * 2. The selector fields for CS and TR cannot be 0000H.
7646  * 3. The selector field for SS cannot be 0000H if the "host address-space
7647  *    size" VM-exit control is 0.
7648  * 4. On processors that support Intel 64 architecture, the base-address
7649  *    fields for FS, GS and TR must contain canonical addresses.
7650  */
test_host_segment_regs(void)7651 static void test_host_segment_regs(void)
7652 {
7653 	u16 selector_saved;
7654 
7655 	/*
7656 	 * Test RPL and TI flags
7657 	 */
7658 	TEST_RPL_TI_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7659 	TEST_RPL_TI_FLAGS(HOST_SEL_SS, "HOST_SEL_SS");
7660 	TEST_RPL_TI_FLAGS(HOST_SEL_DS, "HOST_SEL_DS");
7661 	TEST_RPL_TI_FLAGS(HOST_SEL_ES, "HOST_SEL_ES");
7662 	TEST_RPL_TI_FLAGS(HOST_SEL_FS, "HOST_SEL_FS");
7663 	TEST_RPL_TI_FLAGS(HOST_SEL_GS, "HOST_SEL_GS");
7664 	TEST_RPL_TI_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7665 
7666 	/*
7667 	 * Test that CS and TR fields can not be 0x0000
7668 	 */
7669 	TEST_CS_TR_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7670 	TEST_CS_TR_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7671 
7672 	/*
7673 	 * SS field can not be 0x0000 if "host address-space size" VM-exit
7674 	 * control is 0
7675 	 */
7676 	selector_saved = vmcs_read(HOST_SEL_SS);
7677 	vmcs_write(HOST_SEL_SS, 0);
7678 	report_prefix_pushf("HOST_SEL_SS 0");
7679 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7680 		test_vmx_vmlaunch(0);
7681 	} else {
7682 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7683 	}
7684 	report_prefix_pop();
7685 
7686 	vmcs_write(HOST_SEL_SS, selector_saved);
7687 
7688 	/*
7689 	 * Base address for FS, GS and TR must be canonical
7690 	 */
7691 	test_canonical(HOST_BASE_FS, "HOST_BASE_FS", true);
7692 	test_canonical(HOST_BASE_GS, "HOST_BASE_GS", true);
7693 	test_canonical(HOST_BASE_TR, "HOST_BASE_TR", true);
7694 }
7695 
7696 /*
7697  *  On processors that support Intel 64 architecture, the base-address
7698  *  fields for GDTR and IDTR must contain canonical addresses.
7699  */
test_host_desc_tables(void)7700 static void test_host_desc_tables(void)
7701 {
7702 	test_canonical(HOST_BASE_GDTR, "HOST_BASE_GDTR", true);
7703 	test_canonical(HOST_BASE_IDTR, "HOST_BASE_IDTR", true);
7704 }
7705 
7706 /*
7707  * If the "host address-space size" VM-exit control is 0, the following must
7708  * hold:
7709  *    - The "IA-32e mode guest" VM-entry control is 0.
7710  *    - Bit 17 of the CR4 field (corresponding to CR4.PCIDE) is 0.
7711  *    - Bits 63:32 in the RIP field are 0.
7712  *
7713  * If the "host address-space size" VM-exit control is 1, the following must
7714  * hold:
7715  *    - Bit 5 of the CR4 field (corresponding to CR4.PAE) is 1.
7716  *    - The RIP field contains a canonical address.
7717  *
7718  */
test_host_addr_size(void)7719 static void test_host_addr_size(void)
7720 {
7721 	u64 cr4_saved = vmcs_read(HOST_CR4);
7722 	u64 rip_saved = vmcs_read(HOST_RIP);
7723 	u64 entry_ctrl_saved = vmcs_read(ENT_CONTROLS);
7724 
7725 	assert(vmcs_read(EXI_CONTROLS) & EXI_HOST_64);
7726 	assert(cr4_saved & X86_CR4_PAE);
7727 
7728 	vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7729 	report_prefix_pushf("\"IA-32e mode guest\" enabled");
7730 	test_vmx_vmlaunch(0);
7731 	report_prefix_pop();
7732 
7733 	if (this_cpu_has(X86_FEATURE_PCID)) {
7734 		vmcs_write(HOST_CR4, cr4_saved | X86_CR4_PCIDE);
7735 		report_prefix_pushf("\"CR4.PCIDE\" set");
7736 		test_vmx_vmlaunch(0);
7737 		report_prefix_pop();
7738 	}
7739 
7740 	vmcs_write(HOST_CR4, cr4_saved  & ~X86_CR4_PAE);
7741 	report_prefix_pushf("\"CR4.PAE\" unset");
7742 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7743 	vmcs_write(HOST_CR4, cr4_saved);
7744 	report_prefix_pop();
7745 
7746 	vmcs_write(HOST_RIP, NONCANONICAL);
7747 	report_prefix_pushf("HOST_RIP %llx", NONCANONICAL);
7748 	test_vmx_vmlaunch_must_fail(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7749 	report_prefix_pop();
7750 
7751 	vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7752 	vmcs_write(HOST_RIP, rip_saved);
7753 	vmcs_write(HOST_CR4, cr4_saved);
7754 
7755 	/*
7756 	 * Restore host's active CR4 and RIP values by triggering a VM-Exit.
7757 	 * The original CR4 and RIP values in the VMCS are restored between
7758 	 * testcases as needed, but don't guarantee a VM-Exit and so the active
7759 	 * CR4 and RIP may still hold a test value.  Running with the test CR4
7760 	 * and RIP values at some point is unavoidable, and the active values
7761 	 * are unlikely to affect VM-Enter, so the above doesn't force a VM-exit
7762 	 * between testcases.  Note, if VM-Enter is surrounded by CALL+RET then
7763 	 * the active RIP will already be restored, but that's also not
7764 	 * guaranteed, and CR4 needs to be restored regardless.
7765 	 */
7766 	report_prefix_pushf("restore host state");
7767 	test_vmx_vmlaunch(0);
7768 	report_prefix_pop();
7769 }
7770 
7771 /*
7772  * Check that the virtual CPU checks the VMX Host State Area as
7773  * documented in the Intel SDM.
7774  */
vmx_host_state_area_test(void)7775 static void vmx_host_state_area_test(void)
7776 {
7777 	/*
7778 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
7779 	 * fail due to invalid guest state, should we make it that
7780 	 * far.
7781 	 */
7782 	vmcs_write(GUEST_RFLAGS, 0);
7783 
7784 	test_host_ctl_regs();
7785 
7786 	test_canonical(HOST_SYSENTER_ESP, "HOST_SYSENTER_ESP", true);
7787 	test_canonical(HOST_SYSENTER_EIP, "HOST_SYSENTER_EIP", true);
7788 
7789 	test_host_efer();
7790 	test_load_host_pat();
7791 	test_host_segment_regs();
7792 	test_host_desc_tables();
7793 	test_host_addr_size();
7794 	test_load_host_perf_global_ctrl();
7795 }
7796 
7797 /*
7798  * If the "load debug controls" VM-entry control is 1, bits 63:32 in
7799  * the DR7 field must be 0.
7800  *
7801  * [Intel SDM]
7802  */
test_guest_dr7(void)7803 static void test_guest_dr7(void)
7804 {
7805 	u32 ent_saved = vmcs_read(ENT_CONTROLS);
7806 	u64 dr7_saved = vmcs_read(GUEST_DR7);
7807 	u64 val;
7808 	int i;
7809 
7810 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS) {
7811 		vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7812 		for (i = 0; i < 64; i++) {
7813 			val = 1ull << i;
7814 			vmcs_write(GUEST_DR7, val);
7815 			test_guest_state("ENT_LOAD_DBGCTLS disabled", false,
7816 					 val, "GUEST_DR7");
7817 		}
7818 	}
7819 	if (ctrl_enter_rev.clr & ENT_LOAD_DBGCTLS) {
7820 		vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7821 		for (i = 0; i < 64; i++) {
7822 			val = 1ull << i;
7823 			vmcs_write(GUEST_DR7, val);
7824 			test_guest_state("ENT_LOAD_DBGCTLS enabled", i >= 32,
7825 					 val, "GUEST_DR7");
7826 		}
7827 	}
7828 	vmcs_write(GUEST_DR7, dr7_saved);
7829 	vmcs_write(ENT_CONTROLS, ent_saved);
7830 }
7831 
7832 /*
7833  *  If the "load IA32_PAT" VM-entry control is 1, the value of the field
7834  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7835  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7836  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7837  *  6 (WB), or 7 (UC-).
7838  *
7839  *  [Intel SDM]
7840  */
test_load_guest_pat(void)7841 static void test_load_guest_pat(void)
7842 {
7843 	/*
7844 	 * "load IA32_PAT" VM-entry control
7845 	 */
7846 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
7847 		report_skip("%s : \"Load-IA32-PAT\" entry control not supported", __func__);
7848 		return;
7849 	}
7850 
7851 	test_pat(GUEST_PAT, "GUEST_PAT", ENT_CONTROLS, ENT_LOAD_PAT);
7852 }
7853 
7854 #define MSR_IA32_BNDCFGS_RSVD_MASK	0x00000ffc
7855 
7856 /*
7857  * If the "load IA32_BNDCFGS" VM-entry control is 1, the following
7858  * checks are performed on the field for the IA32_BNDCFGS MSR:
7859  *
7860  *   - Bits reserved in the IA32_BNDCFGS MSR must be 0.
7861  *   - The linear address in bits 63:12 must be canonical.
7862  *
7863  *  [Intel SDM]
7864  */
test_load_guest_bndcfgs(void)7865 static void test_load_guest_bndcfgs(void)
7866 {
7867 	u64 bndcfgs_saved = vmcs_read(GUEST_BNDCFGS);
7868 	u64 bndcfgs;
7869 
7870 	if (!(ctrl_enter_rev.clr & ENT_LOAD_BNDCFGS)) {
7871 		report_skip("%s : \"Load-IA32-BNDCFGS\" entry control not supported", __func__);
7872 		return;
7873 	}
7874 
7875 	vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7876 
7877 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7878 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7879 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7880 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7881 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7882 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7883 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7884 
7885 	vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7886 
7887 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7888 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7889 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7890 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7891 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7892 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7893 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7894 
7895 	vmcs_write(GUEST_BNDCFGS, bndcfgs_saved);
7896 }
7897 
7898 #define	GUEST_SEG_UNUSABLE_MASK	(1u << 16)
7899 #define	GUEST_SEG_SEL_TI_MASK	(1u << 2)
7900 
7901 
7902 #define	TEST_SEGMENT_SEL(test, xfail, sel, val)				\
7903 do {									\
7904 	vmcs_write(sel, val);						\
7905 	test_guest_state(test " segment", xfail, val, xstr(sel));	\
7906 } while (0)
7907 
7908 #define	TEST_INVALID_SEG_SEL(sel, val) \
7909 	TEST_SEGMENT_SEL("Invalid: " xstr(val), true, sel, val);
7910 
7911 #define	TEST_VALID_SEG_SEL(sel, val) \
7912 	TEST_SEGMENT_SEL("Valid: " xstr(val), false, sel, val);
7913 
7914 /*
7915  * The following checks are done on the Selector field of the Guest Segment
7916  * Registers:
7917  *    - TR. The TI flag (bit 2) must be 0.
7918  *    - LDTR. If LDTR is usable, the TI flag (bit 2) must be 0.
7919  *    - SS. If the guest will not be virtual-8086 and the "unrestricted
7920  *	guest" VM-execution control is 0, the RPL (bits 1:0) must equal
7921  *	the RPL of the selector field for CS.
7922  *
7923  *  [Intel SDM]
7924  */
test_guest_segment_sel_fields(void)7925 static void test_guest_segment_sel_fields(void)
7926 {
7927 	u16 sel_saved;
7928 	u32 ar_saved;
7929 	u32 cpu_ctrl0_saved;
7930 	u32 cpu_ctrl1_saved;
7931 	u16 cs_rpl_bits;
7932 
7933 	/*
7934 	 * Test for GUEST_SEL_TR
7935 	 */
7936 	sel_saved = vmcs_read(GUEST_SEL_TR);
7937 	TEST_INVALID_SEG_SEL(GUEST_SEL_TR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7938 	vmcs_write(GUEST_SEL_TR, sel_saved);
7939 
7940 	/*
7941 	 * Test for GUEST_SEL_LDTR
7942 	 */
7943 	sel_saved = vmcs_read(GUEST_SEL_LDTR);
7944 	ar_saved = vmcs_read(GUEST_AR_LDTR);
7945 	/* LDTR is set unusable */
7946 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7947 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7948 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7949 	/* LDTR is set usable */
7950 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7951 	TEST_INVALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7952 
7953 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7954 
7955 	vmcs_write(GUEST_AR_LDTR, ar_saved);
7956 	vmcs_write(GUEST_SEL_LDTR, sel_saved);
7957 
7958 	/*
7959 	 * Test for GUEST_SEL_SS
7960 	 */
7961 	cpu_ctrl0_saved = vmcs_read(CPU_EXEC_CTRL0);
7962 	cpu_ctrl1_saved = vmcs_read(CPU_EXEC_CTRL1);
7963 	ar_saved = vmcs_read(GUEST_AR_SS);
7964 	/* Turn off "unrestricted guest" vm-execution control */
7965 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved & ~CPU_URG);
7966 	cs_rpl_bits = vmcs_read(GUEST_SEL_CS) & 0x3;
7967 	sel_saved = vmcs_read(GUEST_SEL_SS);
7968 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7969 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7970 	/* Make SS usable if it's unusable or vice-versa */
7971 	if (ar_saved & GUEST_SEG_UNUSABLE_MASK)
7972 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7973 	else
7974 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7975 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7976 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7977 
7978 	/* Need a valid EPTP as the passing case fully enters the guest. */
7979 	if (enable_unrestricted_guest(true))
7980 		goto skip_ss_tests;
7981 
7982 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7983 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7984 
7985 	/* Make SS usable if it's unusable or vice-versa */
7986 	if (vmcs_read(GUEST_AR_SS) & GUEST_SEG_UNUSABLE_MASK)
7987 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7988 	else
7989 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7990 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7991 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7992 skip_ss_tests:
7993 
7994 	vmcs_write(GUEST_AR_SS, ar_saved);
7995 	vmcs_write(GUEST_SEL_SS, sel_saved);
7996 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrl0_saved);
7997 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved);
7998 }
7999 
8000 #define	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(xfail, seg_base)			\
8001 do {										\
8002 	addr_saved = vmcs_read(seg_base);					\
8003 	for (i = 32; i < 63; i = i + 4) {					\
8004 		addr = addr_saved | 1ull << i;					\
8005 		vmcs_write(seg_base, addr);					\
8006 		test_guest_state("seg.BASE[63:32] != 0, usable = " xstr(xfail),	\
8007 				 xfail, addr, xstr(seg_base));			\
8008 	}									\
8009 	vmcs_write(seg_base, addr_saved);					\
8010 } while (0)
8011 
8012 #define	TEST_SEGMENT_BASE_ADDR_CANONICAL(xfail, seg_base)		  \
8013 do {									  \
8014 	addr_saved = vmcs_read(seg_base);				  \
8015 	vmcs_write(seg_base, NONCANONICAL);				  \
8016 	test_guest_state("seg.BASE non-canonical, usable = " xstr(xfail), \
8017 			 xfail, NONCANONICAL, xstr(seg_base));		  \
8018 	vmcs_write(seg_base, addr_saved);				  \
8019 } while (0)
8020 
8021 /*
8022  * The following checks are done on the Base Address field of the Guest
8023  * Segment Registers on processors that support Intel 64 architecture:
8024  *    - TR, FS, GS : The address must be canonical.
8025  *    - LDTR : If LDTR is usable, the address must be canonical.
8026  *    - CS : Bits 63:32 of the address must be zero.
8027  *    - SS, DS, ES : If the register is usable, bits 63:32 of the address
8028  *	must be zero.
8029  *
8030  *  [Intel SDM]
8031  */
test_guest_segment_base_addr_fields(void)8032 static void test_guest_segment_base_addr_fields(void)
8033 {
8034 	u64 addr_saved;
8035 	u64 addr;
8036 	u32 ar_saved;
8037 	int i;
8038 
8039 	/*
8040 	 * The address of TR, FS, GS and LDTR must be canonical.
8041 	 */
8042 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_TR);
8043 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_FS);
8044 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_GS);
8045 	ar_saved = vmcs_read(GUEST_AR_LDTR);
8046 	/* Make LDTR unusable */
8047 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8048 	TEST_SEGMENT_BASE_ADDR_CANONICAL(false, GUEST_BASE_LDTR);
8049 	/* Make LDTR usable */
8050 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8051 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_LDTR);
8052 
8053 	vmcs_write(GUEST_AR_LDTR, ar_saved);
8054 
8055 	/*
8056 	 * Bits 63:32 in CS, SS, DS and ES base address must be zero
8057 	 */
8058 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_CS);
8059 	ar_saved = vmcs_read(GUEST_AR_SS);
8060 	/* Make SS unusable */
8061 	vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8062 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_SS);
8063 	/* Make SS usable */
8064 	vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8065 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_SS);
8066 	vmcs_write(GUEST_AR_SS, ar_saved);
8067 
8068 	ar_saved = vmcs_read(GUEST_AR_DS);
8069 	/* Make DS unusable */
8070 	vmcs_write(GUEST_AR_DS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8071 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_DS);
8072 	/* Make DS usable */
8073 	vmcs_write(GUEST_AR_DS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8074 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_DS);
8075 	vmcs_write(GUEST_AR_DS, ar_saved);
8076 
8077 	ar_saved = vmcs_read(GUEST_AR_ES);
8078 	/* Make ES unusable */
8079 	vmcs_write(GUEST_AR_ES, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8080 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_ES);
8081 	/* Make ES usable */
8082 	vmcs_write(GUEST_AR_ES, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8083 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_ES);
8084 	vmcs_write(GUEST_AR_ES, ar_saved);
8085 }
8086 
8087 /*
8088  * Check that the virtual CPU checks the VMX Guest State Area as
8089  * documented in the Intel SDM.
8090  */
vmx_guest_state_area_test(void)8091 static void vmx_guest_state_area_test(void)
8092 {
8093 	vmx_set_test_stage(1);
8094 	test_set_guest(guest_state_test_main);
8095 
8096 	/*
8097 	 * The IA32_SYSENTER_ESP field and the IA32_SYSENTER_EIP field
8098 	 * must each contain a canonical address.
8099 	 */
8100 	test_canonical(GUEST_SYSENTER_ESP, "GUEST_SYSENTER_ESP", false);
8101 	test_canonical(GUEST_SYSENTER_EIP, "GUEST_SYSENTER_EIP", false);
8102 
8103 	test_guest_dr7();
8104 	test_load_guest_pat();
8105 	test_guest_efer();
8106 	test_load_guest_perf_global_ctrl();
8107 	test_load_guest_bndcfgs();
8108 
8109 	test_guest_segment_sel_fields();
8110 	test_guest_segment_base_addr_fields();
8111 
8112 	test_canonical(GUEST_BASE_GDTR, "GUEST_BASE_GDTR", false);
8113 	test_canonical(GUEST_BASE_IDTR, "GUEST_BASE_IDTR", false);
8114 
8115 	u32 guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_GDTR);
8116 	int i;
8117 	for (i = 16; i <= 31; i++) {
8118 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8119 		vmcs_write(GUEST_LIMIT_GDTR, tmp);
8120 		test_guest_state("GDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_GDTR");
8121 	}
8122 	vmcs_write(GUEST_LIMIT_GDTR, guest_desc_limit_saved);
8123 
8124 	guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_IDTR);
8125 	for (i = 16; i <= 31; i++) {
8126 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8127 		vmcs_write(GUEST_LIMIT_IDTR, tmp);
8128 		test_guest_state("IDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_IDTR");
8129 	}
8130 	vmcs_write(GUEST_LIMIT_IDTR, guest_desc_limit_saved);
8131 
8132 	/*
8133 	 * Let the guest finish execution
8134 	 */
8135 	vmx_set_test_stage(2);
8136 	enter_guest();
8137 }
8138 
8139 extern void unrestricted_guest_main(void);
8140 asm (".code32\n"
8141 	"unrestricted_guest_main:\n"
8142 	"vmcall\n"
8143 	"nop\n"
8144 	"mov $1, %edi\n"
8145 	"call hypercall\n"
8146 	".code64\n");
8147 
setup_unrestricted_guest(void)8148 static void setup_unrestricted_guest(void)
8149 {
8150 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) & ~(X86_CR0_PG));
8151 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) & ~ENT_GUEST_64);
8152 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) & ~EFER_LMA);
8153 	vmcs_write(GUEST_RIP, virt_to_phys(unrestricted_guest_main));
8154 }
8155 
unsetup_unrestricted_guest(void)8156 static void unsetup_unrestricted_guest(void)
8157 {
8158 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) | X86_CR0_PG);
8159 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_GUEST_64);
8160 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) | EFER_LMA);
8161 	vmcs_write(GUEST_RIP, (u64) phys_to_virt(vmcs_read(GUEST_RIP)));
8162 	vmcs_write(GUEST_RSP, (u64) phys_to_virt(vmcs_read(GUEST_RSP)));
8163 }
8164 
8165 /*
8166  * If "unrestricted guest" secondary VM-execution control is set, guests
8167  * can run in unpaged protected mode.
8168  */
vmentry_unrestricted_guest_test(void)8169 static void vmentry_unrestricted_guest_test(void)
8170 {
8171 	if (enable_unrestricted_guest(true)) {
8172 		report_skip("%s: \"Unrestricted guest\" exec control not supported", __func__);
8173 		return;
8174 	}
8175 
8176 	test_set_guest(unrestricted_guest_main);
8177 	setup_unrestricted_guest();
8178 	test_guest_state("Unrestricted guest test", false, CPU_URG, "CPU_URG");
8179 
8180 	/*
8181 	 * Let the guest finish execution as a regular guest
8182 	 */
8183 	unsetup_unrestricted_guest();
8184 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) & ~CPU_URG);
8185 	enter_guest();
8186 }
8187 
valid_vmcs_for_vmentry(void)8188 static bool valid_vmcs_for_vmentry(void)
8189 {
8190 	struct vmcs *current_vmcs = NULL;
8191 
8192 	if (vmcs_save(&current_vmcs))
8193 		return false;
8194 
8195 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
8196 }
8197 
try_vmentry_in_movss_shadow(void)8198 static void try_vmentry_in_movss_shadow(void)
8199 {
8200 	u32 vm_inst_err;
8201 	u32 flags;
8202 	bool early_failure = false;
8203 	u32 expected_flags = X86_EFLAGS_FIXED;
8204 	bool valid_vmcs = valid_vmcs_for_vmentry();
8205 
8206 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
8207 
8208 	/*
8209 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
8210 	 * unsupported VMCS component").
8211 	 */
8212 	vmcs_write(~0u, 0);
8213 
8214 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
8215 			      "vmwrite %%rsp, %%rdx;"
8216 			      "mov 0f, %%rax;"
8217 			      "mov %[host_rip], %%edx;"
8218 			      "vmwrite %%rax, %%rdx;"
8219 			      "mov $-1, %%ah;"
8220 			      "sahf;"
8221 			      "mov %%ss, %%ax;"
8222 			      "mov %%ax, %%ss;"
8223 			      "vmlaunch;"
8224 			      "mov $1, %[early_failure];"
8225 			      "0: lahf;"
8226 			      "movzbl %%ah, %[flags]"
8227 			      : [early_failure] "+r" (early_failure),
8228 				[flags] "=&a" (flags)
8229 			      : [host_rsp] "i" (HOST_RSP),
8230 				[host_rip] "i" (HOST_RIP)
8231 			      : "rdx", "cc", "memory");
8232 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
8233 
8234 	report(early_failure, "Early VM-entry failure");
8235 	report(flags == expected_flags, "RFLAGS[8:0] is %x (actual %x)",
8236 	       expected_flags, flags);
8237 	if (valid_vmcs)
8238 		report(vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
8239 		       "VM-instruction error is %d (actual %d)",
8240 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
8241 }
8242 
vmentry_movss_shadow_test(void)8243 static void vmentry_movss_shadow_test(void)
8244 {
8245 	struct vmcs *orig_vmcs;
8246 
8247 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
8248 
8249 	/*
8250 	 * Set the launched flag on the current VMCS to verify the correct
8251 	 * error priority, below.
8252 	 */
8253 	test_set_guest(v2_null_test_guest);
8254 	enter_guest();
8255 
8256 	/*
8257 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
8258 	 * fail due to invalid guest state (if we make it that far).
8259 	 */
8260 	vmcs_write(GUEST_RFLAGS, 0);
8261 
8262 	/*
8263 	 * "VM entry with events blocked by MOV SS" takes precedence over
8264 	 * "VMLAUNCH with non-clear VMCS."
8265 	 */
8266 	report_prefix_push("valid current-VMCS");
8267 	try_vmentry_in_movss_shadow();
8268 	report_prefix_pop();
8269 
8270 	/*
8271 	 * VMfailInvalid takes precedence over "VM entry with events
8272 	 * blocked by MOV SS."
8273 	 */
8274 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
8275 	report_prefix_push("no current-VMCS");
8276 	try_vmentry_in_movss_shadow();
8277 	report_prefix_pop();
8278 
8279 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
8280 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8281 }
8282 
vmx_ldtr_test_guest(void)8283 static void vmx_ldtr_test_guest(void)
8284 {
8285 	u16 ldtr = sldt();
8286 
8287 	report(ldtr == NP_SEL, "Expected %x for L2 LDTR selector (got %x)",
8288 	       NP_SEL, ldtr);
8289 }
8290 
8291 /*
8292  * Ensure that the L1 LDTR is set to 0 on VM-exit.
8293  */
vmx_ldtr_test(void)8294 static void vmx_ldtr_test(void)
8295 {
8296 	const u8 ldt_ar = 0x82; /* Present LDT */
8297 	u16 sel = FIRST_SPARE_SEL;
8298 
8299 	/* Set up a non-zero L1 LDTR prior to VM-entry. */
8300 	set_gdt_entry(sel, 0, 0, ldt_ar, 0);
8301 	lldt(sel);
8302 
8303 	test_set_guest(vmx_ldtr_test_guest);
8304 	/*
8305 	 * Set up a different LDTR for L2. The actual GDT contents are
8306 	 * irrelevant, since we stuff the hidden descriptor state
8307 	 * straight into the VMCS rather than reading it from the GDT.
8308 	 */
8309 	vmcs_write(GUEST_SEL_LDTR, NP_SEL);
8310 	vmcs_write(GUEST_AR_LDTR, ldt_ar);
8311 	enter_guest();
8312 
8313 	/*
8314 	 * VM-exit should clear LDTR (and make it unusable, but we
8315 	 * won't verify that here).
8316 	 */
8317 	sel = sldt();
8318 	report(!sel, "Expected 0 for L1 LDTR selector (got %x)", sel);
8319 }
8320 
vmx_single_vmcall_guest(void)8321 static void vmx_single_vmcall_guest(void)
8322 {
8323 	vmcall();
8324 }
8325 
vmx_cr_load_test(void)8326 static void vmx_cr_load_test(void)
8327 {
8328 	unsigned long cr3, cr4, orig_cr3, orig_cr4;
8329 	u32 ctrls[2] = {0};
8330 	pgd_t *pml5;
8331 
8332 	orig_cr4 = read_cr4();
8333 	orig_cr3 = read_cr3();
8334 
8335 	if (!this_cpu_has(X86_FEATURE_PCID)) {
8336 		report_skip("%s : PCID not detected", __func__);
8337 		return;
8338 	}
8339 	if (!this_cpu_has(X86_FEATURE_MCE)) {
8340 		report_skip("%s : MCE not detected", __func__);
8341 		return;
8342 	}
8343 
8344 	TEST_ASSERT(!(orig_cr3 & X86_CR3_PCID_MASK));
8345 
8346 	/* Enable PCID for L1. */
8347 	cr4 = orig_cr4 | X86_CR4_PCIDE;
8348 	cr3 = orig_cr3 | 0x1;
8349 	TEST_ASSERT(!write_cr4_safe(cr4));
8350 	write_cr3(cr3);
8351 
8352 	test_set_guest(vmx_single_vmcall_guest);
8353 	vmcs_write(HOST_CR4, cr4);
8354 	vmcs_write(HOST_CR3, cr3);
8355 	enter_guest();
8356 
8357 	/*
8358 	 * No exception is expected.
8359 	 *
8360 	 * NB. KVM loads the last guest write to CR4 into CR4 read
8361 	 *     shadow. In order to trigger an exit to KVM, we can toggle a
8362 	 *     bit that is owned by KVM. We use CR4.MCE, which shall
8363 	 *     have no side effect because normally no guest MCE (e.g., as the
8364 	 *     result of bad memory) would happen during this test.
8365 	 */
8366 	TEST_ASSERT(!write_cr4_safe(cr4 ^ X86_CR4_MCE));
8367 
8368 	/* Cleanup L1 state. */
8369 	write_cr3(orig_cr3);
8370 	TEST_ASSERT(!write_cr4_safe(orig_cr4));
8371 
8372 	if (!this_cpu_has(X86_FEATURE_LA57))
8373 		goto done;
8374 
8375 	/*
8376 	 * Allocate a full page for PML5 to guarantee alignment, though only
8377 	 * the first entry needs to be filled (the test's virtual addresses
8378 	 * most definitely do not have any of bits 56:48 set).
8379 	 */
8380 	pml5 = alloc_page();
8381 	*pml5 = orig_cr3 | PT_PRESENT_MASK | PT_WRITABLE_MASK;
8382 
8383 	/*
8384 	 * Transition to/from 5-level paging in the host via VM-Exit.  CR4.LA57
8385 	 * can't be toggled while long is active via MOV CR4, but there are no
8386 	 * such restrictions on VM-Exit.
8387 	 */
8388 lol_5level:
8389 	vmcs_write(HOST_CR4, orig_cr4 | X86_CR4_LA57);
8390 	vmcs_write(HOST_CR3, virt_to_phys(pml5));
8391 	enter_guest();
8392 
8393 	/*
8394 	 * VMREAD with a memory operand to verify KVM detects the LA57 change,
8395 	 * e.g. uses the correct guest root level in gva_to_gpa().
8396 	 */
8397 	TEST_ASSERT(vmcs_readm(HOST_CR3) == virt_to_phys(pml5));
8398 	TEST_ASSERT(vmcs_readm(HOST_CR4) == (orig_cr4 | X86_CR4_LA57));
8399 
8400 	vmcs_write(HOST_CR4, orig_cr4);
8401 	vmcs_write(HOST_CR3, orig_cr3);
8402 	enter_guest();
8403 
8404 	TEST_ASSERT(vmcs_readm(HOST_CR3) == orig_cr3);
8405 	TEST_ASSERT(vmcs_readm(HOST_CR4) == orig_cr4);
8406 
8407 	/*
8408 	 * And now do the same LA57 shenanigans with EPT enabled.  KVM uses
8409 	 * two separate MMUs when L1 uses TDP, whereas the above shadow paging
8410 	 * version shares an MMU between L1 and L2.
8411 	 *
8412 	 * If the saved execution controls are non-zero then the EPT version
8413 	 * has already run.  In that case, restore the old controls.  If EPT
8414 	 * setup fails, e.g. EPT isn't supported, fall through and finish up.
8415 	 */
8416 	if (ctrls[0]) {
8417 		vmcs_write(CPU_EXEC_CTRL0, ctrls[0]);
8418 		vmcs_write(CPU_EXEC_CTRL1, ctrls[1]);
8419 	} else if (!setup_ept(false)) {
8420 		ctrls[0] = vmcs_read(CPU_EXEC_CTRL0);
8421 		ctrls[1]  = vmcs_read(CPU_EXEC_CTRL1);
8422 		goto lol_5level;
8423 	}
8424 
8425 	free_page(pml5);
8426 
8427 done:
8428 	skip_exit_vmcall();
8429 	enter_guest();
8430 }
8431 
vmx_cr4_osxsave_test_guest(void)8432 static void vmx_cr4_osxsave_test_guest(void)
8433 {
8434 	write_cr4(read_cr4() & ~X86_CR4_OSXSAVE);
8435 }
8436 
8437 /*
8438  * Ensure that kvm recalculates the L1 guest's CPUID.01H:ECX.OSXSAVE
8439  * after VM-exit from an L2 guest that sets CR4.OSXSAVE to a different
8440  * value than in L1.
8441  */
vmx_cr4_osxsave_test(void)8442 static void vmx_cr4_osxsave_test(void)
8443 {
8444 	if (!this_cpu_has(X86_FEATURE_XSAVE)) {
8445 		report_skip("%s : XSAVE not detected", __func__);
8446 		return;
8447 	}
8448 
8449 	if (!(read_cr4() & X86_CR4_OSXSAVE)) {
8450 		unsigned long cr4 = read_cr4() | X86_CR4_OSXSAVE;
8451 
8452 		write_cr4(cr4);
8453 		vmcs_write(GUEST_CR4, cr4);
8454 		vmcs_write(HOST_CR4, cr4);
8455 	}
8456 
8457 	TEST_ASSERT(this_cpu_has(X86_FEATURE_OSXSAVE));
8458 
8459 	test_set_guest(vmx_cr4_osxsave_test_guest);
8460 	enter_guest();
8461 
8462 	TEST_ASSERT(this_cpu_has(X86_FEATURE_OSXSAVE));
8463 }
8464 
8465 /*
8466  * FNOP with both CR0.TS and CR0.EM clear should not generate #NM, and the L2
8467  * guest should exit normally.
8468  */
vmx_no_nm_test(void)8469 static void vmx_no_nm_test(void)
8470 {
8471 	test_set_guest(fnop);
8472 	vmcs_write(GUEST_CR0, read_cr0() & ~(X86_CR0_TS | X86_CR0_EM));
8473 	enter_guest();
8474 }
8475 
8476 bool vmx_pending_event_ipi_fired;
vmx_pending_event_ipi_isr(isr_regs_t * regs)8477 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
8478 {
8479 	vmx_pending_event_ipi_fired = true;
8480 	eoi();
8481 }
8482 
8483 bool vmx_pending_event_guest_run;
vmx_pending_event_guest(void)8484 static void vmx_pending_event_guest(void)
8485 {
8486 	vmcall();
8487 	vmx_pending_event_guest_run = true;
8488 }
8489 
vmx_pending_event_test_core(bool guest_hlt)8490 static void vmx_pending_event_test_core(bool guest_hlt)
8491 {
8492 	int ipi_vector = 0xf1;
8493 
8494 	vmx_pending_event_ipi_fired = false;
8495 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
8496 
8497 	vmx_pending_event_guest_run = false;
8498 	test_set_guest(vmx_pending_event_guest);
8499 
8500 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
8501 
8502 	enter_guest();
8503 	skip_exit_vmcall();
8504 
8505 	if (guest_hlt)
8506 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8507 
8508 	cli();
8509 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
8510 				   APIC_DM_FIXED | ipi_vector,
8511 				   0);
8512 
8513 	enter_guest();
8514 
8515 	assert_exit_reason(VMX_EXTINT);
8516 	report(!vmx_pending_event_guest_run,
8517 	       "Guest did not run before host received IPI");
8518 
8519 	sti_nop_cli();
8520 	report(vmx_pending_event_ipi_fired,
8521 	       "Got pending interrupt after IRQ enabled");
8522 
8523 	if (guest_hlt)
8524 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8525 
8526 	enter_guest();
8527 	report(vmx_pending_event_guest_run,
8528 	       "Guest finished running when no interrupt");
8529 }
8530 
vmx_pending_event_test(void)8531 static void vmx_pending_event_test(void)
8532 {
8533 	vmx_pending_event_test_core(false);
8534 }
8535 
vmx_pending_event_hlt_test(void)8536 static void vmx_pending_event_hlt_test(void)
8537 {
8538 	vmx_pending_event_test_core(true);
8539 }
8540 
8541 static int vmx_window_test_db_count;
8542 
vmx_window_test_db_handler(struct ex_regs * regs)8543 static void vmx_window_test_db_handler(struct ex_regs *regs)
8544 {
8545 	vmx_window_test_db_count++;
8546 }
8547 
vmx_nmi_window_test_guest(void)8548 static void vmx_nmi_window_test_guest(void)
8549 {
8550 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8551 
8552 	asm volatile("vmcall\n\t"
8553 		     "nop\n\t");
8554 
8555 	handle_exception(DB_VECTOR, NULL);
8556 }
8557 
verify_nmi_window_exit(u64 rip)8558 static void verify_nmi_window_exit(u64 rip)
8559 {
8560 	u32 exit_reason = vmcs_read(EXI_REASON);
8561 
8562 	report(exit_reason == VMX_NMI_WINDOW,
8563 	       "Exit reason (%d) is 'NMI window'", exit_reason);
8564 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8565 	       vmcs_read(GUEST_RIP), rip);
8566 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8567 }
8568 
vmx_nmi_window_test(void)8569 static void vmx_nmi_window_test(void)
8570 {
8571 	u64 nop_addr;
8572 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8573 
8574 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
8575 		report_skip("%s : \"Virtual NMIs\" exec control not supported", __func__);
8576 		return;
8577 	}
8578 
8579 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
8580 		report_skip("%s : \"NMI-window exiting\" exec control not supported", __func__);
8581 		return;
8582 	}
8583 
8584 	vmx_window_test_db_count = 0;
8585 
8586 	report_prefix_push("NMI-window");
8587 	test_set_guest(vmx_nmi_window_test_guest);
8588 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
8589 	enter_guest();
8590 	skip_exit_vmcall();
8591 	nop_addr = vmcs_read(GUEST_RIP);
8592 
8593 	/*
8594 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
8595 	 * RIP will not advance.
8596 	 */
8597 	report_prefix_push("active, no blocking");
8598 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8599 	enter_guest();
8600 	verify_nmi_window_exit(nop_addr);
8601 	report_prefix_pop();
8602 
8603 	/*
8604 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
8605 	 * a VM-exit on the next instruction after the nop. (The nop
8606 	 * is one byte.)
8607 	 */
8608 	report_prefix_push("active, blocking by MOV-SS");
8609 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8610 	enter_guest();
8611 	verify_nmi_window_exit(nop_addr + 1);
8612 	report_prefix_pop();
8613 
8614 	/*
8615 	 * Ask for "NMI-window exiting" (with event injection), and
8616 	 * expect a VM-exit after the event is injected. (RIP should
8617 	 * be at the address specified in the IDT entry for #DB.)
8618 	 */
8619 	report_prefix_push("active, no blocking, injecting #DB");
8620 	vmcs_write(ENT_INTR_INFO,
8621 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8622 	enter_guest();
8623 	verify_nmi_window_exit((u64)db_fault_addr);
8624 	report_prefix_pop();
8625 
8626 	/*
8627 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
8628 	 * a VM-exit after the next IRET (i.e. after the #DB handler
8629 	 * returns). So, RIP should be back at one byte past the nop.
8630 	 */
8631 	report_prefix_push("active, blocking by NMI");
8632 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
8633 	enter_guest();
8634 	verify_nmi_window_exit(nop_addr + 1);
8635 	report(vmx_window_test_db_count == 1,
8636 	       "#DB handler executed once (actual %d times)",
8637 	       vmx_window_test_db_count);
8638 	report_prefix_pop();
8639 
8640 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8641 		report_skip("CPU does not support activity state HLT.");
8642 	} else {
8643 		/*
8644 		 * Ask for "NMI-window exiting" when entering activity
8645 		 * state HLT, and expect an immediate VM-exit. RIP is
8646 		 * still one byte past the nop.
8647 		 */
8648 		report_prefix_push("halted, no blocking");
8649 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8650 		enter_guest();
8651 		verify_nmi_window_exit(nop_addr + 1);
8652 		report_prefix_pop();
8653 
8654 		/*
8655 		 * Ask for "NMI-window exiting" when entering activity
8656 		 * state HLT (with event injection), and expect a
8657 		 * VM-exit after the event is injected. (RIP should be
8658 		 * at the address specified in the IDT entry for #DB.)
8659 		 */
8660 		report_prefix_push("halted, no blocking, injecting #DB");
8661 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8662 		vmcs_write(ENT_INTR_INFO,
8663 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8664 			   DB_VECTOR);
8665 		enter_guest();
8666 		verify_nmi_window_exit((u64)db_fault_addr);
8667 		report_prefix_pop();
8668 	}
8669 
8670 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8671 	enter_guest();
8672 	report_prefix_pop();
8673 }
8674 
vmx_intr_window_test_guest(void)8675 static void vmx_intr_window_test_guest(void)
8676 {
8677 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8678 
8679 	/*
8680 	 * The two consecutive STIs are to ensure that only the first
8681 	 * one has a shadow. Note that NOP and STI are one byte
8682 	 * instructions.
8683 	 */
8684 	asm volatile("vmcall\n\t"
8685 		     "nop\n\t"
8686 		     "sti\n\t"
8687 		     "sti\n\t");
8688 
8689 	handle_exception(DB_VECTOR, NULL);
8690 }
8691 
verify_intr_window_exit(u64 rip)8692 static void verify_intr_window_exit(u64 rip)
8693 {
8694 	u32 exit_reason = vmcs_read(EXI_REASON);
8695 
8696 	report(exit_reason == VMX_INTR_WINDOW,
8697 	       "Exit reason (%d) is 'interrupt window'", exit_reason);
8698 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8699 	       vmcs_read(GUEST_RIP), rip);
8700 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8701 }
8702 
vmx_intr_window_test(void)8703 static void vmx_intr_window_test(void)
8704 {
8705 	u64 vmcall_addr;
8706 	u64 nop_addr;
8707 	unsigned int orig_db_gate_type;
8708 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8709 
8710 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
8711 		report_skip("%s : \"Interrupt-window exiting\" exec control not supported", __func__);
8712 		return;
8713 	}
8714 
8715 	/*
8716 	 * Change the IDT entry for #DB from interrupt gate to trap gate,
8717 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
8718 	 * be disabled after vectoring a #DB.
8719 	 */
8720 	orig_db_gate_type = boot_idt[DB_VECTOR].type;
8721 	boot_idt[DB_VECTOR].type = 15;
8722 
8723 	report_prefix_push("interrupt-window");
8724 	test_set_guest(vmx_intr_window_test_guest);
8725 	enter_guest();
8726 	assert_exit_reason(VMX_VMCALL);
8727 	vmcall_addr = vmcs_read(GUEST_RIP);
8728 
8729 	/*
8730 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
8731 	 * no blocking; expect an immediate VM-exit. Note that we have
8732 	 * not advanced past the vmcall instruction yet, so RIP should
8733 	 * point to the vmcall instruction.
8734 	 */
8735 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
8736 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8737 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
8738 	enter_guest();
8739 	verify_intr_window_exit(vmcall_addr);
8740 	report_prefix_pop();
8741 
8742 	/*
8743 	 * Ask for "interrupt-window exiting" (with event injection)
8744 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
8745 	 * the event is injected. That is, RIP should should be at the
8746 	 * address specified in the IDT entry for #DB.
8747 	 */
8748 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #DB");
8749 	vmcs_write(ENT_INTR_INFO,
8750 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8751 	vmcall_addr = vmcs_read(GUEST_RIP);
8752 	enter_guest();
8753 	verify_intr_window_exit((u64)db_fault_addr);
8754 	report_prefix_pop();
8755 
8756 	/*
8757 	 * Let the L2 guest run through the IRET, back to the VMCALL.
8758 	 * We have to clear the "interrupt-window exiting"
8759 	 * VM-execution control, or it would just keep causing
8760 	 * VM-exits. Then, advance past the VMCALL and set the
8761 	 * "interrupt-window exiting" VM-execution control again.
8762 	 */
8763 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8764 	enter_guest();
8765 	skip_exit_vmcall();
8766 	nop_addr = vmcs_read(GUEST_RIP);
8767 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8768 
8769 	/*
8770 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
8771 	 * RFLAGS.IF set, and expect a VM-exit on the next
8772 	 * instruction. (NOP is one byte.)
8773 	 */
8774 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
8775 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8776 	enter_guest();
8777 	verify_intr_window_exit(nop_addr + 1);
8778 	report_prefix_pop();
8779 
8780 	/*
8781 	 * Back up to the NOP and ask for "interrupt-window exiting"
8782 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
8783 	 * on the next instruction. (NOP is one byte.)
8784 	 */
8785 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
8786 	vmcs_write(GUEST_RIP, nop_addr);
8787 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
8788 	enter_guest();
8789 	verify_intr_window_exit(nop_addr + 1);
8790 	report_prefix_pop();
8791 
8792 	/*
8793 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
8794 	 * and expect a VM-exit on the instruction following the STI
8795 	 * shadow. Only the first STI (which is one byte past the NOP)
8796 	 * should have a shadow. The second STI (which is two bytes
8797 	 * past the NOP) has no shadow. Therefore, the interrupt
8798 	 * window opens at three bytes past the NOP.
8799 	 */
8800 	report_prefix_push("active, RFLAGS.IF = 0");
8801 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8802 	enter_guest();
8803 	verify_intr_window_exit(nop_addr + 3);
8804 	report_prefix_pop();
8805 
8806 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8807 		report_skip("CPU does not support activity state HLT.");
8808 	} else {
8809 		/*
8810 		 * Ask for "interrupt-window exiting" when entering
8811 		 * activity state HLT, and expect an immediate
8812 		 * VM-exit. RIP is still three bytes past the nop.
8813 		 */
8814 		report_prefix_push("halted, no blocking");
8815 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8816 		enter_guest();
8817 		verify_intr_window_exit(nop_addr + 3);
8818 		report_prefix_pop();
8819 
8820 		/*
8821 		 * Ask for "interrupt-window exiting" when entering
8822 		 * activity state HLT (with event injection), and
8823 		 * expect a VM-exit after the event is injected. That
8824 		 * is, RIP should should be at the address specified
8825 		 * in the IDT entry for #DB.
8826 		 */
8827 		report_prefix_push("halted, no blocking, injecting #DB");
8828 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8829 		vmcs_write(ENT_INTR_INFO,
8830 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8831 			   DB_VECTOR);
8832 		enter_guest();
8833 		verify_intr_window_exit((u64)db_fault_addr);
8834 		report_prefix_pop();
8835 	}
8836 
8837 	boot_idt[DB_VECTOR].type = orig_db_gate_type;
8838 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8839 	enter_guest();
8840 	report_prefix_pop();
8841 }
8842 
8843 #define GUEST_TSC_OFFSET (1u << 30)
8844 
8845 static u64 guest_tsc;
8846 
vmx_store_tsc_test_guest(void)8847 static void vmx_store_tsc_test_guest(void)
8848 {
8849 	guest_tsc = rdtsc();
8850 }
8851 
8852 /*
8853  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
8854  * list, the value saved is not subject to the TSC offset that is
8855  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
8856  */
vmx_store_tsc_test(void)8857 static void vmx_store_tsc_test(void)
8858 {
8859 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
8860 	u64 low, high;
8861 
8862 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
8863 		report_skip("%s : \"Use TSC offsetting\" exec control not supported", __func__);
8864 		return;
8865 	}
8866 
8867 	test_set_guest(vmx_store_tsc_test_guest);
8868 
8869 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
8870 	vmcs_write(EXI_MSR_ST_CNT, 1);
8871 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
8872 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
8873 
8874 	low = rdtsc();
8875 	enter_guest();
8876 	high = rdtsc();
8877 
8878 	report(low + GUEST_TSC_OFFSET <= guest_tsc &&
8879 	       guest_tsc <= high + GUEST_TSC_OFFSET,
8880 	       "RDTSC value in the guest (%lu) is in range [%lu, %lu]",
8881 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
8882 	report(low <= msr_entry.value && msr_entry.value <= high,
8883 	       "IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
8884 	       msr_entry.value, low, high);
8885 }
8886 
vmx_preemption_timer_zero_test_db_handler(struct ex_regs * regs)8887 static void vmx_preemption_timer_zero_test_db_handler(struct ex_regs *regs)
8888 {
8889 }
8890 
vmx_preemption_timer_zero_test_guest(void)8891 static void vmx_preemption_timer_zero_test_guest(void)
8892 {
8893 	while (vmx_get_test_stage() < 3)
8894 		vmcall();
8895 }
8896 
vmx_preemption_timer_zero_activate_preemption_timer(void)8897 static void vmx_preemption_timer_zero_activate_preemption_timer(void)
8898 {
8899 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
8900 	vmcs_write(PREEMPT_TIMER_VALUE, 0);
8901 }
8902 
vmx_preemption_timer_zero_advance_past_vmcall(void)8903 static void vmx_preemption_timer_zero_advance_past_vmcall(void)
8904 {
8905 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
8906 	enter_guest();
8907 	skip_exit_vmcall();
8908 }
8909 
vmx_preemption_timer_zero_inject_db(bool intercept_db)8910 static void vmx_preemption_timer_zero_inject_db(bool intercept_db)
8911 {
8912 	vmx_preemption_timer_zero_activate_preemption_timer();
8913 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK |
8914 		   INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8915 	vmcs_write(EXC_BITMAP, intercept_db ? 1 << DB_VECTOR : 0);
8916 	enter_guest();
8917 }
8918 
vmx_preemption_timer_zero_set_pending_dbg(u32 exception_bitmap)8919 static void vmx_preemption_timer_zero_set_pending_dbg(u32 exception_bitmap)
8920 {
8921 	vmx_preemption_timer_zero_activate_preemption_timer();
8922 	vmcs_write(GUEST_PENDING_DEBUG, PENDING_DBG_TRAP | DR6_TRAP1);
8923 	vmcs_write(EXC_BITMAP, exception_bitmap);
8924 	enter_guest();
8925 }
8926 
vmx_preemption_timer_zero_expect_preempt_at_rip(u64 expected_rip)8927 static void vmx_preemption_timer_zero_expect_preempt_at_rip(u64 expected_rip)
8928 {
8929 	u32 reason = (u32)vmcs_read(EXI_REASON);
8930 	u64 guest_rip = vmcs_read(GUEST_RIP);
8931 
8932 	report(reason == VMX_PREEMPT && guest_rip == expected_rip,
8933 	       "Exit reason is 0x%x (expected 0x%x) and guest RIP is %lx (0x%lx expected).",
8934 	       reason, VMX_PREEMPT, guest_rip, expected_rip);
8935 }
8936 
8937 /*
8938  * This test ensures that when the VMX preemption timer is zero at
8939  * VM-entry, a VM-exit occurs after any event injection and after any
8940  * pending debug exceptions are raised, but before execution of any
8941  * guest instructions.
8942  */
vmx_preemption_timer_zero_test(void)8943 static void vmx_preemption_timer_zero_test(void)
8944 {
8945 	u64 db_fault_address = (u64)get_idt_addr(&boot_idt[DB_VECTOR]);
8946 	handler old_db;
8947 	u32 reason;
8948 
8949 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
8950 		report_skip("%s : \"Activate VMX-preemption timer\" pin control not supported", __func__);
8951 		return;
8952 	}
8953 
8954 	/*
8955 	 * Install a custom #DB handler that doesn't abort.
8956 	 */
8957 	old_db = handle_exception(DB_VECTOR,
8958 				  vmx_preemption_timer_zero_test_db_handler);
8959 
8960 	test_set_guest(vmx_preemption_timer_zero_test_guest);
8961 
8962 	/*
8963 	 * VMX-preemption timer should fire after event injection.
8964 	 */
8965 	vmx_set_test_stage(0);
8966 	vmx_preemption_timer_zero_inject_db(0);
8967 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8968 	vmx_preemption_timer_zero_advance_past_vmcall();
8969 
8970 	/*
8971 	 * VMX-preemption timer should fire after event injection.
8972 	 * Exception bitmap is irrelevant, since you can't intercept
8973 	 * an event that you injected.
8974 	 */
8975 	vmx_set_test_stage(1);
8976 	vmx_preemption_timer_zero_inject_db(true);
8977 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8978 	vmx_preemption_timer_zero_advance_past_vmcall();
8979 
8980 	/*
8981 	 * VMX-preemption timer should fire after pending debug exceptions
8982 	 * have delivered a #DB trap.
8983 	 */
8984 	vmx_set_test_stage(2);
8985 	vmx_preemption_timer_zero_set_pending_dbg(0);
8986 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8987 	vmx_preemption_timer_zero_advance_past_vmcall();
8988 
8989 	/*
8990 	 * VMX-preemption timer would fire after pending debug exceptions
8991 	 * have delivered a #DB trap, but in this case, the #DB trap is
8992 	 * intercepted.
8993 	 */
8994 	vmx_set_test_stage(3);
8995 	vmx_preemption_timer_zero_set_pending_dbg(1 << DB_VECTOR);
8996 	reason = (u32)vmcs_read(EXI_REASON);
8997 	report(reason == VMX_EXC_NMI, "Exit reason is 0x%x (expected 0x%x)",
8998 	       reason, VMX_EXC_NMI);
8999 
9000 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9001 	enter_guest();
9002 
9003 	handle_exception(DB_VECTOR, old_db);
9004 }
9005 
9006 static u64 vmx_preemption_timer_tf_test_prev_rip;
9007 
vmx_preemption_timer_tf_test_db_handler(struct ex_regs * regs)9008 static void vmx_preemption_timer_tf_test_db_handler(struct ex_regs *regs)
9009 {
9010 	extern char vmx_preemption_timer_tf_test_endloop;
9011 
9012 	if (vmx_get_test_stage() == 2) {
9013 		/*
9014 		 * Stage 2 means that we're done, one way or another.
9015 		 * Arrange for the iret to drop us out of the wbinvd
9016 		 * loop and stop single-stepping.
9017 		 */
9018 		regs->rip = (u64)&vmx_preemption_timer_tf_test_endloop;
9019 		regs->rflags &= ~X86_EFLAGS_TF;
9020 	} else if (regs->rip == vmx_preemption_timer_tf_test_prev_rip) {
9021 		/*
9022 		 * The RIP should alternate between the wbinvd and the
9023 		 * jmp instruction in the code below. If we ever see
9024 		 * the same instruction twice in a row, that means a
9025 		 * single-step trap has been dropped. Let the
9026 		 * hypervisor know about the failure by executing a
9027 		 * VMCALL.
9028 		 */
9029 		vmcall();
9030 	}
9031 	vmx_preemption_timer_tf_test_prev_rip = regs->rip;
9032 }
9033 
vmx_preemption_timer_tf_test_guest(void)9034 static void vmx_preemption_timer_tf_test_guest(void)
9035 {
9036 	/*
9037 	 * The hypervisor doesn't intercept WBINVD, so the loop below
9038 	 * shouldn't be a problem--it's just two instructions
9039 	 * executing in VMX non-root mode. However, when the
9040 	 * hypervisor is running in a virtual environment, the parent
9041 	 * hypervisor might intercept WBINVD and emulate it. If the
9042 	 * parent hypervisor is broken, the single-step trap after the
9043 	 * WBINVD might be lost.
9044 	 */
9045 	asm volatile("vmcall\n\t"
9046 		     "0: wbinvd\n\t"
9047 		     "1: jmp 0b\n\t"
9048 		     "vmx_preemption_timer_tf_test_endloop:");
9049 }
9050 
9051 /*
9052  * Ensure that the delivery of a "VMX-preemption timer expired"
9053  * VM-exit doesn't disrupt single-stepping in the guest. Note that
9054  * passing this test doesn't ensure correctness, because the test will
9055  * only fail if the VMX-preemtion timer fires at the right time (or
9056  * the wrong time, as it were).
9057  */
vmx_preemption_timer_tf_test(void)9058 static void vmx_preemption_timer_tf_test(void)
9059 {
9060 	handler old_db;
9061 	u32 reason;
9062 	int i;
9063 
9064 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9065 		report_skip("%s : \"Activate VMX-preemption timer\" pin control not supported", __func__);
9066 		return;
9067 	}
9068 
9069 	old_db = handle_exception(DB_VECTOR,
9070 				  vmx_preemption_timer_tf_test_db_handler);
9071 
9072 	test_set_guest(vmx_preemption_timer_tf_test_guest);
9073 
9074 	enter_guest();
9075 	skip_exit_vmcall();
9076 
9077 	vmx_set_test_stage(1);
9078 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9079 	vmcs_write(PREEMPT_TIMER_VALUE, 50000);
9080 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9081 
9082 	/*
9083 	 * The only exit we should see is "VMX-preemption timer
9084 	 * expired."  If we get a VMCALL exit, that means the #DB
9085 	 * handler has detected a missing single-step trap. It doesn't
9086 	 * matter where the guest RIP is when the VMX-preemption timer
9087 	 * expires (whether it's in the WBINVD loop or in the #DB
9088 	 * handler)--a single-step trap should never be discarded.
9089 	 */
9090 	for (i = 0; i < 10000; i++) {
9091 		enter_guest();
9092 		reason = (u32)vmcs_read(EXI_REASON);
9093 		if (reason == VMX_PREEMPT)
9094 			continue;
9095 		TEST_ASSERT(reason == VMX_VMCALL);
9096 		skip_exit_insn();
9097 		break;
9098 	}
9099 
9100 	report(reason == VMX_PREEMPT, "No single-step traps skipped");
9101 
9102 	vmx_set_test_stage(2);
9103 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9104 	enter_guest();
9105 
9106 	handle_exception(DB_VECTOR, old_db);
9107 }
9108 
9109 #define VMX_PREEMPTION_TIMER_EXPIRY_CYCLES 1000000
9110 
9111 static u64 vmx_preemption_timer_expiry_start;
9112 static u64 vmx_preemption_timer_expiry_finish;
9113 
vmx_preemption_timer_expiry_test_guest(void)9114 static void vmx_preemption_timer_expiry_test_guest(void)
9115 {
9116 	vmcall();
9117 	vmx_preemption_timer_expiry_start = fenced_rdtsc();
9118 
9119 	while (vmx_get_test_stage() == 0)
9120 		vmx_preemption_timer_expiry_finish = fenced_rdtsc();
9121 }
9122 
9123 /*
9124  * Test that the VMX-preemption timer is not excessively delayed.
9125  *
9126  * Per the SDM, volume 3, VM-entry starts the VMX-preemption timer
9127  * with the unsigned value in the VMX-preemption timer-value field,
9128  * and the VMX-preemption timer counts down by 1 every time bit X in
9129  * the TSC changes due to a TSC increment (where X is
9130  * IA32_VMX_MISC[4:0]). If the timer counts down to zero in any state
9131  * other than the wait-for-SIPI state, the logical processor
9132  * transitions to the C0 C-state and causes a VM-exit.
9133  *
9134  * The guest code above reads the starting TSC after VM-entry. At this
9135  * point, the VMX-preemption timer has already been activated. Next,
9136  * the guest code reads the current TSC in a loop, storing the value
9137  * read to memory.
9138  *
9139  * If the RDTSC in the loop reads a value past the VMX-preemption
9140  * timer deadline, then the VMX-preemption timer VM-exit must be
9141  * delivered before the next instruction retires. Even if a higher
9142  * priority SMI is delivered first, the VMX-preemption timer VM-exit
9143  * must be delivered before the next instruction retires. Hence, a TSC
9144  * value past the VMX-preemption timer deadline might be read, but it
9145  * cannot be stored. If a TSC value past the deadline *is* stored,
9146  * then the architectural specification has been violated.
9147  */
vmx_preemption_timer_expiry_test(void)9148 static void vmx_preemption_timer_expiry_test(void)
9149 {
9150 	u32 preemption_timer_value;
9151 	union vmx_misc misc;
9152 	u64 tsc_deadline;
9153 	u32 reason;
9154 
9155 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9156 		report_skip("%s : \"Activate VMX-preemption timer\" pin control not supported", __func__);
9157 		return;
9158 	}
9159 
9160 	test_set_guest(vmx_preemption_timer_expiry_test_guest);
9161 
9162 	enter_guest();
9163 	skip_exit_vmcall();
9164 
9165 	misc.val = rdmsr(MSR_IA32_VMX_MISC);
9166 	preemption_timer_value =
9167 		VMX_PREEMPTION_TIMER_EXPIRY_CYCLES >> misc.pt_bit;
9168 
9169 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9170 	vmcs_write(PREEMPT_TIMER_VALUE, preemption_timer_value);
9171 	vmx_set_test_stage(0);
9172 
9173 	enter_guest();
9174 	reason = (u32)vmcs_read(EXI_REASON);
9175 	TEST_ASSERT(reason == VMX_PREEMPT);
9176 
9177 	tsc_deadline = ((vmx_preemption_timer_expiry_start >> misc.pt_bit) <<
9178 			misc.pt_bit) + (preemption_timer_value << misc.pt_bit);
9179 
9180 	report(vmx_preemption_timer_expiry_finish < tsc_deadline,
9181 	       "Last stored guest TSC (%lu) < TSC deadline (%lu)",
9182 	       vmx_preemption_timer_expiry_finish, tsc_deadline);
9183 
9184 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9185 	vmx_set_test_stage(1);
9186 	enter_guest();
9187 }
9188 
vmx_db_test_guest(void)9189 static void vmx_db_test_guest(void)
9190 {
9191 	/*
9192 	 * For a hardware generated single-step #DB.
9193 	 */
9194 	asm volatile("vmcall;"
9195 		     "nop;"
9196 		     ".Lpost_nop:");
9197 	/*
9198 	 * ...in a MOVSS shadow, with pending debug exceptions.
9199 	 */
9200 	asm volatile("vmcall;"
9201 		     "nop;"
9202 		     ".Lpost_movss_nop:");
9203 	/*
9204 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
9205 	 * emulates it in software.)
9206 	 */
9207 	asm volatile("vmcall;"
9208 		     "wbinvd;"
9209 		     ".Lpost_wbinvd:");
9210 	/*
9211 	 * ...in a MOVSS shadow, with pending debug exceptions.
9212 	 */
9213 	asm volatile("vmcall;"
9214 		     "wbinvd;"
9215 		     ".Lpost_movss_wbinvd:");
9216 	/*
9217 	 * For a hardware generated single-step #DB in a transactional region.
9218 	 */
9219 	asm volatile("vmcall;"
9220 		     ".Lxbegin: xbegin .Lskip_rtm;"
9221 		     "xend;"
9222 		     ".Lskip_rtm:");
9223 }
9224 
9225 /*
9226  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
9227  * L2. No #DB is delivered and L2 continues to the next point of
9228  * interest.
9229  */
dismiss_db(void)9230 static void dismiss_db(void)
9231 {
9232 	vmcs_write(GUEST_PENDING_DEBUG, 0);
9233 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
9234 	enter_guest();
9235 }
9236 
9237 /*
9238  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
9239  * Then throw away the #DB exception and resume L2.
9240  */
check_db_exit(bool xfail_qual,bool xfail_dr6,bool xfail_pdbg,void * expected_rip,u64 expected_exit_qual,u64 expected_dr6)9241 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
9242 			  void *expected_rip, u64 expected_exit_qual,
9243 			  u64 expected_dr6)
9244 {
9245 	u32 reason = vmcs_read(EXI_REASON);
9246 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
9247 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
9248 	u64 guest_rip = vmcs_read(GUEST_RIP);
9249 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
9250 	u64 dr6 = read_dr6();
9251 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
9252 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
9253 
9254 	report(reason == VMX_EXC_NMI && intr_info == expected_intr_info,
9255 	       "Expected #DB VM-exit");
9256 	report((u64)expected_rip == guest_rip, "Expected RIP %p (actual %lx)",
9257 	       expected_rip, guest_rip);
9258 	report_xfail(xfail_pdbg, 0 == guest_pending_dbg,
9259 		     "Expected pending debug exceptions 0 (actual %lx)",
9260 		     guest_pending_dbg);
9261 	report_xfail(xfail_qual, expected_exit_qual == exit_qual,
9262 		     "Expected exit qualification %lx (actual %lx)",
9263 		     expected_exit_qual, exit_qual);
9264 	report_xfail(xfail_dr6, expected_dr6 == dr6,
9265 		     "Expected DR6 %lx (actual %lx)", expected_dr6, dr6);
9266 	dismiss_db();
9267 }
9268 
9269 /*
9270  * Assuming the guest has just exited on a VMCALL instruction, skip
9271  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
9272  * pending debug exceptions are non-zero, set the VMCS up as if the
9273  * previous instruction was a MOVSS that generated the indicated
9274  * pending debug exceptions. Then enter L2.
9275  */
single_step_guest(const char * test_name,u64 starting_dr6,u64 pending_debug_exceptions)9276 static void single_step_guest(const char *test_name, u64 starting_dr6,
9277 			      u64 pending_debug_exceptions)
9278 {
9279 	printf("\n%s\n", test_name);
9280 	skip_exit_vmcall();
9281 	write_dr6(starting_dr6);
9282 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9283 	if (pending_debug_exceptions) {
9284 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
9285 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
9286 	}
9287 	enter_guest();
9288 }
9289 
9290 /*
9291  * When L1 intercepts #DB, verify that a single-step trap clears
9292  * pending debug exceptions, populates the exit qualification field
9293  * properly, and that DR6 is not prematurely clobbered. In a
9294  * (simulated) MOVSS shadow, make sure that the pending debug
9295  * exception bits are properly accumulated into the exit qualification
9296  * field.
9297  */
vmx_db_test(void)9298 static void vmx_db_test(void)
9299 {
9300 	/*
9301 	 * We are going to set a few arbitrary bits in DR6 to verify that
9302 	 * (a) DR6 is not modified by an intercepted #DB, and
9303 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
9304          *     the exit qualification field for a subsequent #DB exception.
9305 	 */
9306 	const u64 starting_dr6 = DR6_ACTIVE_LOW | DR6_BS | DR6_TRAP3 | DR6_TRAP1;
9307 	extern char post_nop asm(".Lpost_nop");
9308 	extern char post_movss_nop asm(".Lpost_movss_nop");
9309 	extern char post_wbinvd asm(".Lpost_wbinvd");
9310 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
9311 	extern char xbegin asm(".Lxbegin");
9312 	extern char skip_rtm asm(".Lskip_rtm");
9313 
9314 	/*
9315 	 * L1 wants to intercept #DB exceptions encountered in L2.
9316 	 */
9317 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
9318 
9319 	/*
9320 	 * Start L2 and run it up to the first point of interest.
9321 	 */
9322 	test_set_guest(vmx_db_test_guest);
9323 	enter_guest();
9324 
9325 	/*
9326 	 * Hardware-delivered #DB trap for single-step sets the
9327 	 * standard that L0 has to follow for emulated instructions.
9328 	 */
9329 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
9330 	check_db_exit(false, false, false, &post_nop, DR6_BS, starting_dr6);
9331 
9332 	/*
9333 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
9334 	 * also sets the standard that L0 has to follow for emulated
9335 	 * instructions. Here, we establish the VMCS pending debug
9336 	 * exceptions to indicate that the simulated MOVSS triggered a
9337 	 * data breakpoint as well as the single-step trap.
9338 	 */
9339 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
9340 			  starting_dr6, DR6_BS | PENDING_DBG_TRAP | DR6_TRAP0);
9341 	check_db_exit(false, false, false, &post_movss_nop, DR6_BS | DR6_TRAP0,
9342 		      starting_dr6);
9343 
9344 	/*
9345 	 * L0 synthesized #DB trap for single-step is buggy, because
9346 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
9347 	 * reconstitute the exit qualification from the prematurely
9348 	 * modified DR6, but fails miserably.
9349 	 */
9350 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
9351 	check_db_exit(false, false, false, &post_wbinvd, DR6_BS, starting_dr6);
9352 
9353 	/*
9354 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
9355 	 * even worse, because L0 also leaves the pending debug
9356 	 * exceptions in the VMCS instead of accumulating them into
9357 	 * the exit qualification field for the #DB exception.
9358 	 */
9359 	single_step_guest("Software synthesized single-step in MOVSS shadow",
9360 			  starting_dr6, DR6_BS | PENDING_DBG_TRAP | DR6_TRAP0);
9361 	check_db_exit(true, false, true, &post_movss_wbinvd, DR6_BS | DR6_TRAP0,
9362 		      starting_dr6);
9363 
9364 	/*
9365 	 * Optional RTM test for hardware that supports RTM, to
9366 	 * demonstrate that the current volume 3 of the SDM
9367 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
9368 	 * qualification for debug exceptions is not reserved. It is
9369 	 * set to 1 if a debug exception (#DB) or a breakpoint
9370 	 * exception (#BP) occurs inside an RTM region while advanced
9371 	 * debugging of RTM transactional regions is enabled.
9372 	 */
9373 	if (this_cpu_has(X86_FEATURE_RTM)) {
9374 		vmcs_write(ENT_CONTROLS,
9375 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
9376 		/*
9377 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
9378 		 * in the guest to enable advanced debugging of RTM
9379 		 * transactional regions.
9380 		 */
9381 		vmcs_write(GUEST_DR7, BIT(11));
9382 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
9383 		single_step_guest("Hardware delivered single-step in "
9384 				  "transactional region", starting_dr6, 0);
9385 		check_db_exit(false, false, false, &xbegin, BIT(16),
9386 			      starting_dr6);
9387 	} else {
9388 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
9389 		enter_guest();
9390 	}
9391 }
9392 
enable_vid(void)9393 static void enable_vid(void)
9394 {
9395 	void *virtual_apic_page;
9396 
9397 	assert(cpu_has_apicv());
9398 
9399 	enable_x2apic();
9400 	disable_intercept_for_x2apic_msrs();
9401 
9402 	virtual_apic_page = alloc_page();
9403 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
9404 
9405 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
9406 
9407 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
9408 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
9409 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
9410 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
9411 
9412 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
9413 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
9414 }
9415 
9416 #define	PI_VECTOR	255
9417 
enable_posted_interrupts(void)9418 static void enable_posted_interrupts(void)
9419 {
9420 	void *pi_desc = alloc_page();
9421 
9422 	vmcs_set_bits(PIN_CONTROLS, PIN_POST_INTR);
9423 	vmcs_set_bits(EXI_CONTROLS, EXI_INTA);
9424 	vmcs_write(PINV, PI_VECTOR);
9425 	vmcs_write(POSTED_INTR_DESC_ADDR, (u64)pi_desc);
9426 }
9427 
trigger_ioapic_scan_thread(void * data)9428 static void trigger_ioapic_scan_thread(void *data)
9429 {
9430 	/* Wait until other CPU entered L2 */
9431 	while (vmx_get_test_stage() != 1)
9432 		;
9433 
9434 	/* Trigger ioapic scan */
9435 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
9436 	vmx_set_test_stage(2);
9437 }
9438 
irq_79_handler_guest(isr_regs_t * regs)9439 static void irq_79_handler_guest(isr_regs_t *regs)
9440 {
9441 	eoi();
9442 
9443 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
9444 	vmcall();
9445 }
9446 
9447 /*
9448  * Constant for num of busy-loop iterations after which
9449  * a timer interrupt should have happened in host
9450  */
9451 #define TIMER_INTERRUPT_DELAY 100000000
9452 
vmx_eoi_bitmap_ioapic_scan_test_guest(void)9453 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
9454 {
9455 	handle_irq(0x79, irq_79_handler_guest);
9456 	sti();
9457 
9458 	/* Signal to L1 CPU to trigger ioapic scan */
9459 	vmx_set_test_stage(1);
9460 	/* Wait until L1 CPU to trigger ioapic scan */
9461 	while (vmx_get_test_stage() != 2)
9462 		;
9463 
9464 	/*
9465 	 * Wait for L0 timer interrupt to be raised while we run in L2
9466 	 * such that L0 will process the IOAPIC scan request before
9467 	 * resuming L2
9468 	 */
9469 	delay(TIMER_INTERRUPT_DELAY);
9470 
9471 	asm volatile ("int $0x79");
9472 }
9473 
vmx_eoi_bitmap_ioapic_scan_test(void)9474 static void vmx_eoi_bitmap_ioapic_scan_test(void)
9475 {
9476 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
9477 		report_skip("%s : Not all required APICv bits supported or CPU count < 2", __func__);
9478 		return;
9479 	}
9480 
9481 	enable_vid();
9482 
9483 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
9484 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
9485 
9486 	/*
9487 	 * Launch L2.
9488 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
9489 	 * In case the reason isn't VMX_VMCALL, the assertion inside
9490 	 * skip_exit_vmcall() will fail.
9491 	 */
9492 	enter_guest();
9493 	skip_exit_vmcall();
9494 
9495 	/* Let L2 finish */
9496 	enter_guest();
9497 	report_pass(__func__);
9498 }
9499 
9500 #define HLT_WITH_RVI_VECTOR		(0xf1)
9501 
9502 bool vmx_hlt_with_rvi_guest_isr_fired;
vmx_hlt_with_rvi_guest_isr(isr_regs_t * regs)9503 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
9504 {
9505 	vmx_hlt_with_rvi_guest_isr_fired = true;
9506 	eoi();
9507 }
9508 
vmx_hlt_with_rvi_guest(void)9509 static void vmx_hlt_with_rvi_guest(void)
9510 {
9511 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
9512 
9513 	sti_nop();
9514 	asm volatile ("nop");
9515 
9516 	vmcall();
9517 }
9518 
vmx_hlt_with_rvi_test(void)9519 static void vmx_hlt_with_rvi_test(void)
9520 {
9521 	if (!cpu_has_apicv()) {
9522 		report_skip("%s : Not all required APICv bits supported", __func__);
9523 		return;
9524 	}
9525 
9526 	enable_vid();
9527 
9528 	vmx_hlt_with_rvi_guest_isr_fired = false;
9529 	test_set_guest(vmx_hlt_with_rvi_guest);
9530 
9531 	enter_guest();
9532 	skip_exit_vmcall();
9533 
9534 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
9535 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
9536 	enter_guest();
9537 
9538 	report(vmx_hlt_with_rvi_guest_isr_fired, "Interrupt raised in guest");
9539 }
9540 
set_irq_line_thread(void * data)9541 static void set_irq_line_thread(void *data)
9542 {
9543 	/* Wait until other CPU entered L2 */
9544 	while (vmx_get_test_stage() != 1)
9545 		;
9546 
9547 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
9548 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9549 	vmx_set_test_stage(2);
9550 }
9551 
9552 static bool irq_78_handler_vmcall_before_eoi;
irq_78_handler_guest(isr_regs_t * regs)9553 static void irq_78_handler_guest(isr_regs_t *regs)
9554 {
9555 	set_irq_line(0xf, 0);
9556 	if (irq_78_handler_vmcall_before_eoi)
9557 		vmcall();
9558 	eoi();
9559 	vmcall();
9560 }
9561 
vmx_apic_passthrough_guest(void)9562 static void vmx_apic_passthrough_guest(void)
9563 {
9564 	handle_irq(0x78, irq_78_handler_guest);
9565 	sti();
9566 
9567 	/* If requested, wait for other CPU to trigger ioapic scan */
9568 	if (vmx_get_test_stage() < 1) {
9569 		vmx_set_test_stage(1);
9570 		while (vmx_get_test_stage() != 2)
9571 			;
9572 	}
9573 
9574 	set_irq_line(0xf, 1);
9575 }
9576 
vmx_apic_passthrough(bool set_irq_line_from_thread)9577 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
9578 {
9579 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
9580 		report_skip("%s : CPU count < 2", __func__);
9581 		return;
9582 	}
9583 
9584 	/* Test device is required for generating IRQs */
9585 	if (!test_device_enabled()) {
9586 		report_skip("%s : No test device enabled", __func__);
9587 		return;
9588 	}
9589 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9590 	u64 cpu_ctrl_1 = 0;
9591 
9592 	disable_intercept_for_x2apic_msrs();
9593 
9594 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9595 
9596 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9597 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9598 
9599 	if (set_irq_line_from_thread) {
9600 		irq_78_handler_vmcall_before_eoi = false;
9601 		on_cpu_async(1, set_irq_line_thread, NULL);
9602 	} else {
9603 		irq_78_handler_vmcall_before_eoi = true;
9604 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9605 		vmx_set_test_stage(2);
9606 	}
9607 	test_set_guest(vmx_apic_passthrough_guest);
9608 
9609 	if (irq_78_handler_vmcall_before_eoi) {
9610 		/* Before EOI remote_irr should still be set */
9611 		enter_guest();
9612 		skip_exit_vmcall();
9613 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
9614 			"IOAPIC pass-through: remote_irr=1 before EOI");
9615 	}
9616 
9617 	/* After EOI remote_irr should be cleared */
9618 	enter_guest();
9619 	skip_exit_vmcall();
9620 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
9621 		"IOAPIC pass-through: remote_irr=0 after EOI");
9622 
9623 	/* Let L2 finish */
9624 	enter_guest();
9625 	report_pass(__func__);
9626 }
9627 
vmx_apic_passthrough_test(void)9628 static void vmx_apic_passthrough_test(void)
9629 {
9630 	vmx_apic_passthrough(false);
9631 }
9632 
vmx_apic_passthrough_thread_test(void)9633 static void vmx_apic_passthrough_thread_test(void)
9634 {
9635 	vmx_apic_passthrough(true);
9636 }
9637 
vmx_apic_passthrough_tpr_threshold_guest(void)9638 static void vmx_apic_passthrough_tpr_threshold_guest(void)
9639 {
9640 	cli();
9641 	apic_set_tpr(0);
9642 }
9643 
9644 static bool vmx_apic_passthrough_tpr_threshold_ipi_isr_fired;
vmx_apic_passthrough_tpr_threshold_ipi_isr(isr_regs_t * regs)9645 static void vmx_apic_passthrough_tpr_threshold_ipi_isr(isr_regs_t *regs)
9646 {
9647 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = true;
9648 	eoi();
9649 }
9650 
vmx_apic_passthrough_tpr_threshold_test(void)9651 static void vmx_apic_passthrough_tpr_threshold_test(void)
9652 {
9653 	int ipi_vector = 0xe1;
9654 
9655 	disable_intercept_for_x2apic_msrs();
9656 	vmcs_clear_bits(PIN_CONTROLS, PIN_EXTINT);
9657 
9658 	/* Raise L0 TPR-threshold by queueing vector in LAPIC IRR */
9659 	cli();
9660 	apic_set_tpr((ipi_vector >> 4) + 1);
9661 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
9662 			APIC_DM_FIXED | ipi_vector,
9663 			0);
9664 
9665 	test_set_guest(vmx_apic_passthrough_tpr_threshold_guest);
9666 	enter_guest();
9667 
9668 	report(apic_get_tpr() == 0, "TPR was zero by guest");
9669 
9670 	/* Clean pending self-IPI */
9671 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = false;
9672 	handle_irq(ipi_vector, vmx_apic_passthrough_tpr_threshold_ipi_isr);
9673 	sti_nop();
9674 	report(vmx_apic_passthrough_tpr_threshold_ipi_isr_fired, "self-IPI fired");
9675 
9676 	report_pass(__func__);
9677 }
9678 
9679 static u64 init_signal_test_exit_reason;
9680 static bool init_signal_test_thread_continued;
9681 
init_signal_test_thread(void * data)9682 static void init_signal_test_thread(void *data)
9683 {
9684 	struct vmcs *test_vmcs = data;
9685 
9686 	/* Enter VMX operation (i.e. exec VMXON) */
9687 	u64 *ap_vmxon_region = alloc_page();
9688 	enable_vmx();
9689 	init_vmx(ap_vmxon_region);
9690 	TEST_ASSERT(!__vmxon_safe(ap_vmxon_region));
9691 
9692 	/* Signal CPU have entered VMX operation */
9693 	vmx_set_test_stage(1);
9694 
9695 	/* Wait for BSP CPU to send INIT signal */
9696 	while (vmx_get_test_stage() != 2)
9697 		;
9698 
9699 	/*
9700 	 * Signal that we continue as usual as INIT signal
9701 	 * should be blocked while CPU is in VMX operation
9702 	 */
9703 	vmx_set_test_stage(3);
9704 
9705 	/* Wait for signal to enter VMX non-root mode */
9706 	while (vmx_get_test_stage() != 4)
9707 		;
9708 
9709 	/* Enter VMX non-root mode */
9710 	test_set_guest(v2_null_test_guest);
9711 	make_vmcs_current(test_vmcs);
9712 	enter_guest();
9713 	/* Save exit reason for BSP CPU to compare to expected result */
9714 	init_signal_test_exit_reason = vmcs_read(EXI_REASON);
9715 	/* VMCLEAR test-vmcs so it could be loaded by BSP CPU */
9716 	vmcs_clear(test_vmcs);
9717 	launched = false;
9718 	/* Signal that CPU exited to VMX root mode */
9719 	vmx_set_test_stage(5);
9720 
9721 	/* Wait for BSP CPU to signal to exit VMX operation */
9722 	while (vmx_get_test_stage() != 6)
9723 		;
9724 
9725 	/* Exit VMX operation (i.e. exec VMXOFF) */
9726 	vmx_off();
9727 
9728 	/*
9729 	 * Signal to BSP CPU that we continue as usual as INIT signal
9730 	 * should have been consumed by VMX_INIT exit from guest
9731 	 */
9732 	vmx_set_test_stage(7);
9733 
9734 	/* Wait for BSP CPU to signal to enter VMX operation */
9735 	while (vmx_get_test_stage() != 8)
9736 		;
9737 	/* Enter VMX operation (i.e. exec VMXON) */
9738 	TEST_ASSERT(!__vmxon_safe(ap_vmxon_region));
9739 	/* Signal to BSP we are in VMX operation */
9740 	vmx_set_test_stage(9);
9741 
9742 	/* Wait for BSP CPU to send INIT signal */
9743 	while (vmx_get_test_stage() != 10)
9744 		;
9745 
9746 	/* Exit VMX operation (i.e. exec VMXOFF) */
9747 	vmx_off();
9748 
9749 	/*
9750 	 * Exiting VMX operation should result in latched
9751 	 * INIT signal being processed. Therefore, we should
9752 	 * never reach the below code. Thus, signal to BSP
9753 	 * CPU if we have reached here so it is able to
9754 	 * report an issue if it happens.
9755 	 */
9756 	init_signal_test_thread_continued = true;
9757 }
9758 
9759 #define INIT_SIGNAL_TEST_DELAY	100000000ULL
9760 
vmx_init_signal_test(void)9761 static void vmx_init_signal_test(void)
9762 {
9763 	struct vmcs *test_vmcs;
9764 
9765 	if (cpu_count() < 2) {
9766 		report_skip("%s : CPU count < 2", __func__);
9767 		return;
9768 	}
9769 
9770 	/* VMCLEAR test-vmcs so it could be loaded by other CPU */
9771 	vmcs_save(&test_vmcs);
9772 	vmcs_clear(test_vmcs);
9773 
9774 	vmx_set_test_stage(0);
9775 	on_cpu_async(1, init_signal_test_thread, test_vmcs);
9776 
9777 	/* Wait for other CPU to enter VMX operation */
9778 	while (vmx_get_test_stage() != 1)
9779 		;
9780 
9781 	/* Send INIT signal to other CPU */
9782 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9783 				   id_map[1]);
9784 	/* Signal other CPU we have sent INIT signal */
9785 	vmx_set_test_stage(2);
9786 
9787 	/*
9788 	 * Wait reasonable amount of time for INIT signal to
9789 	 * be received on other CPU and verify that other CPU
9790 	 * have proceed as usual to next test stage as INIT
9791 	 * signal should be blocked while other CPU in
9792 	 * VMX operation
9793 	 */
9794 	delay(INIT_SIGNAL_TEST_DELAY);
9795 	report(vmx_get_test_stage() == 3,
9796 	       "INIT signal blocked when CPU in VMX operation");
9797 	/* No point to continue if we failed at this point */
9798 	if (vmx_get_test_stage() != 3)
9799 		return;
9800 
9801 	/* Signal other CPU to enter VMX non-root mode */
9802 	init_signal_test_exit_reason = -1ull;
9803 	vmx_set_test_stage(4);
9804 	/*
9805 	 * Wait reasonable amount of time for other CPU
9806 	 * to exit to VMX root mode
9807 	 */
9808 	delay(INIT_SIGNAL_TEST_DELAY);
9809 	if (vmx_get_test_stage() != 5) {
9810 		report_fail("Pending INIT signal didn't result in VMX exit");
9811 		return;
9812 	}
9813 	report(init_signal_test_exit_reason == VMX_INIT,
9814 			"INIT signal during VMX non-root mode result in exit-reason %s (%lu)",
9815 			exit_reason_description(init_signal_test_exit_reason),
9816 			init_signal_test_exit_reason);
9817 
9818 	/* Run guest to completion */
9819 	make_vmcs_current(test_vmcs);
9820 	enter_guest();
9821 
9822 	/* Signal other CPU to exit VMX operation */
9823 	init_signal_test_thread_continued = false;
9824 	vmx_set_test_stage(6);
9825 
9826 	/* Wait reasonable amount of time for other CPU to exit VMX operation */
9827 	delay(INIT_SIGNAL_TEST_DELAY);
9828 	report(vmx_get_test_stage() == 7,
9829 	       "INIT signal consumed on VMX_INIT exit");
9830 	/* No point to continue if we failed at this point */
9831 	if (vmx_get_test_stage() != 7)
9832 		return;
9833 
9834 	/* Signal other CPU to enter VMX operation */
9835 	vmx_set_test_stage(8);
9836 	/* Wait for other CPU to enter VMX operation */
9837 	while (vmx_get_test_stage() != 9)
9838 		;
9839 
9840 	/* Send INIT signal to other CPU */
9841 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9842 				   id_map[1]);
9843 	/* Signal other CPU we have sent INIT signal */
9844 	vmx_set_test_stage(10);
9845 
9846 	/*
9847 	 * Wait reasonable amount of time for other CPU
9848 	 * to exit VMX operation and process INIT signal
9849 	 */
9850 	delay(INIT_SIGNAL_TEST_DELAY);
9851 	report(!init_signal_test_thread_continued,
9852 	       "INIT signal processed after exit VMX operation");
9853 
9854 	/*
9855 	 * TODO: Send SIPI to other CPU to sipi_entry (See x86/cstart64.S)
9856 	 * to re-init it to kvm-unit-tests standard environment.
9857 	 * Somehow (?) verify that SIPI was indeed received.
9858 	 */
9859 }
9860 
9861 #define SIPI_SIGNAL_TEST_DELAY	100000000ULL
9862 
vmx_sipi_test_guest(void)9863 static void vmx_sipi_test_guest(void)
9864 {
9865 	if (apic_id() == 0) {
9866 		/* wait AP enter guest with activity=WAIT_SIPI */
9867 		while (vmx_get_test_stage() != 1)
9868 			;
9869 		delay(SIPI_SIGNAL_TEST_DELAY);
9870 
9871 		/* First SIPI signal */
9872 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9873 		report_pass("BSP(L2): Send first SIPI to cpu[%d]", id_map[1]);
9874 
9875 		/* wait AP enter guest */
9876 		while (vmx_get_test_stage() != 2)
9877 			;
9878 		delay(SIPI_SIGNAL_TEST_DELAY);
9879 
9880 		/* Second SIPI signal should be ignored since AP is not in WAIT_SIPI state */
9881 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9882 		report_pass("BSP(L2): Send second SIPI to cpu[%d]", id_map[1]);
9883 
9884 		/* Delay a while to check whether second SIPI would cause VMExit */
9885 		delay(SIPI_SIGNAL_TEST_DELAY);
9886 
9887 		/* Test is done, notify AP to exit test */
9888 		vmx_set_test_stage(3);
9889 
9890 		/* wait AP exit non-root mode */
9891 		while (vmx_get_test_stage() != 5)
9892 			;
9893 	} else {
9894 		/* wait BSP notify test is done */
9895 		while (vmx_get_test_stage() != 3)
9896 			;
9897 
9898 		/* AP exit guest */
9899 		vmx_set_test_stage(4);
9900 	}
9901 }
9902 
sipi_test_ap_thread(void * data)9903 static void sipi_test_ap_thread(void *data)
9904 {
9905 	struct vmcs *ap_vmcs;
9906 	u64 *ap_vmxon_region;
9907 	void *ap_stack, *ap_syscall_stack;
9908 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9909 	u64 cpu_ctrl_1 = 0;
9910 
9911 	/* Enter VMX operation (i.e. exec VMXON) */
9912 	ap_vmxon_region = alloc_page();
9913 	enable_vmx();
9914 	init_vmx(ap_vmxon_region);
9915 	TEST_ASSERT(!__vmxon_safe(ap_vmxon_region));
9916 	init_vmcs(&ap_vmcs);
9917 	make_vmcs_current(ap_vmcs);
9918 
9919 	/* Set stack for AP */
9920 	ap_stack = alloc_page();
9921 	ap_syscall_stack = alloc_page();
9922 	vmcs_write(GUEST_RSP, (u64)(ap_stack + PAGE_SIZE - 1));
9923 	vmcs_write(GUEST_SYSENTER_ESP, (u64)(ap_syscall_stack + PAGE_SIZE - 1));
9924 
9925 	/* passthrough lapic to L2 */
9926 	disable_intercept_for_x2apic_msrs();
9927 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9928 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9929 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9930 
9931 	/* Set guest activity state to wait-for-SIPI state */
9932 	vmcs_write(GUEST_ACTV_STATE, ACTV_WAIT_SIPI);
9933 
9934 	vmx_set_test_stage(1);
9935 
9936 	/* AP enter guest */
9937 	enter_guest();
9938 
9939 	if (vmcs_read(EXI_REASON) == VMX_SIPI) {
9940 		report_pass("AP: Handle SIPI VMExit");
9941 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
9942 		vmx_set_test_stage(2);
9943 	} else {
9944 		report_fail("AP: Unexpected VMExit, reason=%ld", vmcs_read(EXI_REASON));
9945 		vmx_off();
9946 		return;
9947 	}
9948 
9949 	/* AP enter guest */
9950 	enter_guest();
9951 
9952 	report(vmcs_read(EXI_REASON) != VMX_SIPI,
9953 		"AP: should no SIPI VMExit since activity is not in WAIT_SIPI state");
9954 
9955 	/* notify BSP that AP is already exit from non-root mode */
9956 	vmx_set_test_stage(5);
9957 
9958 	/* Leave VMX operation */
9959 	vmx_off();
9960 }
9961 
vmx_sipi_signal_test(void)9962 static void vmx_sipi_signal_test(void)
9963 {
9964 	if (!(rdmsr(MSR_IA32_VMX_MISC) & MSR_IA32_VMX_MISC_ACTIVITY_WAIT_SIPI)) {
9965 		report_skip("%s : \"ACTIVITY_WAIT_SIPI state\" not supported", __func__);
9966 		return;
9967 	}
9968 
9969 	if (cpu_count() < 2) {
9970 		report_skip("%s : CPU count < 2", __func__);
9971 		return;
9972 	}
9973 
9974 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9975 	u64 cpu_ctrl_1 = 0;
9976 
9977 	/* passthrough lapic to L2 */
9978 	disable_intercept_for_x2apic_msrs();
9979 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9980 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9981 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9982 
9983 	test_set_guest(vmx_sipi_test_guest);
9984 
9985 	/* update CR3 on AP */
9986 	on_cpu(1, update_cr3, (void *)read_cr3());
9987 
9988 	/* start AP */
9989 	on_cpu_async(1, sipi_test_ap_thread, NULL);
9990 
9991 	vmx_set_test_stage(0);
9992 
9993 	/* BSP enter guest */
9994 	enter_guest();
9995 }
9996 
9997 
9998 enum vmcs_access {
9999 	ACCESS_VMREAD,
10000 	ACCESS_VMWRITE,
10001 	ACCESS_NONE,
10002 };
10003 
10004 struct vmcs_shadow_test_common {
10005 	enum vmcs_access op;
10006 	enum Reason reason;
10007 	u64 field;
10008 	u64 value;
10009 	u64 flags;
10010 	u64 time;
10011 } l1_l2_common;
10012 
vmread_flags(u64 field,u64 * val)10013 static inline u64 vmread_flags(u64 field, u64 *val)
10014 {
10015 	u64 flags;
10016 
10017 	asm volatile ("vmread %2, %1; pushf; pop %0"
10018 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
10019 	return flags & X86_EFLAGS_ALU;
10020 }
10021 
vmwrite_flags(u64 field,u64 val)10022 static inline u64 vmwrite_flags(u64 field, u64 val)
10023 {
10024 	u64 flags;
10025 
10026 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
10027 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
10028 	return flags & X86_EFLAGS_ALU;
10029 }
10030 
vmx_vmcs_shadow_test_guest(void)10031 static void vmx_vmcs_shadow_test_guest(void)
10032 {
10033 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10034 	u64 start;
10035 
10036 	while (c->op != ACCESS_NONE) {
10037 		start = rdtsc();
10038 		switch (c->op) {
10039 		default:
10040 			c->flags = -1ull;
10041 			break;
10042 		case ACCESS_VMREAD:
10043 			c->flags = vmread_flags(c->field, &c->value);
10044 			break;
10045 		case ACCESS_VMWRITE:
10046 			c->flags = vmwrite_flags(c->field, 0);
10047 			break;
10048 		}
10049 		c->time = rdtsc() - start;
10050 		vmcall();
10051 	}
10052 }
10053 
vmread_from_shadow(u64 field)10054 static u64 vmread_from_shadow(u64 field)
10055 {
10056 	struct vmcs *primary;
10057 	struct vmcs *shadow;
10058 	u64 value;
10059 
10060 	TEST_ASSERT(!vmcs_save(&primary));
10061 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10062 	TEST_ASSERT(!make_vmcs_current(shadow));
10063 	value = vmcs_read(field);
10064 	TEST_ASSERT(!make_vmcs_current(primary));
10065 	return value;
10066 }
10067 
vmwrite_to_shadow(u64 field,u64 value)10068 static u64 vmwrite_to_shadow(u64 field, u64 value)
10069 {
10070 	struct vmcs *primary;
10071 	struct vmcs *shadow;
10072 
10073 	TEST_ASSERT(!vmcs_save(&primary));
10074 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10075 	TEST_ASSERT(!make_vmcs_current(shadow));
10076 	vmcs_write(field, value);
10077 	value = vmcs_read(field);
10078 	TEST_ASSERT(!make_vmcs_current(primary));
10079 	return value;
10080 }
10081 
vmcs_shadow_test_access(u8 * bitmap[2],enum vmcs_access access)10082 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
10083 {
10084 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10085 
10086 	c->op = access;
10087 	vmcs_write(VMX_INST_ERROR, 0);
10088 	enter_guest();
10089 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
10090 	if (c->reason != VMX_VMCALL) {
10091 		skip_exit_insn();
10092 		enter_guest();
10093 	}
10094 	skip_exit_vmcall();
10095 }
10096 
vmcs_shadow_test_field(u8 * bitmap[2],u64 field)10097 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
10098 {
10099 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10100 	struct vmcs *shadow;
10101 	u64 value;
10102 	uintptr_t flags[2];
10103 	bool good_shadow;
10104 	u32 vmx_inst_error;
10105 
10106 	report_prefix_pushf("field %lx", field);
10107 	c->field = field;
10108 
10109 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10110 	if (shadow != (struct vmcs *)-1ull) {
10111 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
10112 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
10113 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
10114 	} else {
10115 		/*
10116 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
10117 		 * shadowed-fields should fail with setting RFLAGS.CF.
10118 		 */
10119 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
10120 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
10121 		good_shadow = false;
10122 	}
10123 
10124 	/* Intercept both VMREAD and VMWRITE. */
10125 	report_prefix_push("no VMREAD/VMWRITE permission");
10126 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
10127 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
10128 		set_bit(field, bitmap[ACCESS_VMREAD]);
10129 		set_bit(field, bitmap[ACCESS_VMWRITE]);
10130 	}
10131 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10132 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10133 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10134 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10135 	report_prefix_pop();
10136 
10137 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
10138 		goto out;
10139 
10140 	/* Permit shadowed VMREAD. */
10141 	report_prefix_push("VMREAD permission only");
10142 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10143 	set_bit(field, bitmap[ACCESS_VMWRITE]);
10144 	if (good_shadow)
10145 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10146 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10147 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10148 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10149 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10150 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10151 	       c->time);
10152 	report(c->flags == flags[ACCESS_VMREAD],
10153 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10154 	       c->flags, flags[ACCESS_VMREAD]);
10155 	if (good_shadow)
10156 		report(c->value == value,
10157 		       "value read from shadow (%lx) is as expected (%lx)",
10158 		       c->value, value);
10159 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10160 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10161 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10162 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10163 	report_prefix_pop();
10164 
10165 	/* Permit shadowed VMWRITE. */
10166 	report_prefix_push("VMWRITE permission only");
10167 	set_bit(field, bitmap[ACCESS_VMREAD]);
10168 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10169 	if (good_shadow)
10170 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10171 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10172 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10173 	report(c->reason == VMX_VMCALL,
10174 		"shadowed for VMWRITE (in %ld cycles)",
10175 		c->time);
10176 	report(c->flags == flags[ACCESS_VMREAD],
10177 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10178 	       c->flags, flags[ACCESS_VMREAD]);
10179 	if (good_shadow) {
10180 		value = vmread_from_shadow(field);
10181 		report(value == 0,
10182 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10183 		       0ul);
10184 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10185 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10186 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10187 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10188 	}
10189 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10190 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10191 	report_prefix_pop();
10192 
10193 	/* Permit shadowed VMREAD and VMWRITE. */
10194 	report_prefix_push("VMREAD and VMWRITE permission");
10195 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10196 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10197 	if (good_shadow)
10198 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10199 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10200 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10201 	report(c->reason == VMX_VMCALL,
10202 		"shadowed for VMWRITE (in %ld cycles)",
10203 		c->time);
10204 	report(c->flags == flags[ACCESS_VMREAD],
10205 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10206 	       c->flags, flags[ACCESS_VMREAD]);
10207 	if (good_shadow) {
10208 		value = vmread_from_shadow(field);
10209 		report(value == 0,
10210 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10211 		       0ul);
10212 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10213 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10214 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10215 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10216 	}
10217 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10218 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10219 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10220 	       c->time);
10221 	report(c->flags == flags[ACCESS_VMREAD],
10222 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10223 	       c->flags, flags[ACCESS_VMREAD]);
10224 	if (good_shadow)
10225 		report(c->value == 0,
10226 		       "value read from shadow (%lx) is as expected (%lx)",
10227 		       c->value, 0ul);
10228 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10229 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10230 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10231 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10232 	report_prefix_pop();
10233 
10234 out:
10235 	report_prefix_pop();
10236 }
10237 
vmx_vmcs_shadow_test_body(u8 * bitmap[2])10238 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
10239 {
10240 	unsigned base;
10241 	unsigned index;
10242 	unsigned bit;
10243 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
10244 
10245 	/* Run test on all possible valid VMCS fields */
10246 	for (base = 0;
10247 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
10248 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
10249 		for (index = 0; index <= highest_index; index++)
10250 			vmcs_shadow_test_field(bitmap, base + index);
10251 
10252 	/*
10253 	 * Run tests on some invalid VMCS fields
10254 	 * (Have reserved bit set).
10255 	 */
10256 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
10257 		vmcs_shadow_test_field(bitmap, (1ull << bit));
10258 }
10259 
vmx_vmcs_shadow_test(void)10260 static void vmx_vmcs_shadow_test(void)
10261 {
10262 	u8 *bitmap[2];
10263 	struct vmcs *shadow;
10264 
10265 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
10266 		report_skip("%s : \"Activate secondary controls\" not supported", __func__);
10267 		return;
10268 	}
10269 
10270 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
10271 		report_skip("%s : \"VMCS shadowing\" not supported", __func__);
10272 		return;
10273 	}
10274 
10275 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
10276 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
10277 		report_skip("%s : VMWRITE can't modify VM-exit information fields.", __func__);
10278 		return;
10279 	}
10280 
10281 	test_set_guest(vmx_vmcs_shadow_test_guest);
10282 
10283 	bitmap[ACCESS_VMREAD] = alloc_page();
10284 	bitmap[ACCESS_VMWRITE] = alloc_page();
10285 
10286 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
10287 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
10288 
10289 	shadow = alloc_page();
10290 	shadow->hdr.revision_id = basic_msr.revision;
10291 	shadow->hdr.shadow_vmcs = 1;
10292 	TEST_ASSERT(!vmcs_clear(shadow));
10293 
10294 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
10295 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
10296 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
10297 
10298 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
10299 	report_prefix_push("valid link pointer");
10300 	vmx_vmcs_shadow_test_body(bitmap);
10301 	report_prefix_pop();
10302 
10303 	vmcs_write(VMCS_LINK_PTR, -1ull);
10304 	report_prefix_push("invalid link pointer");
10305 	vmx_vmcs_shadow_test_body(bitmap);
10306 	report_prefix_pop();
10307 
10308 	l1_l2_common.op = ACCESS_NONE;
10309 	enter_guest();
10310 }
10311 
10312 /*
10313  * This test monitors the difference between a guest RDTSC instruction
10314  * and the IA32_TIME_STAMP_COUNTER MSR value stored in the VMCS12
10315  * VM-exit MSR-store list when taking a VM-exit on the instruction
10316  * following RDTSC.
10317  */
10318 #define RDTSC_DIFF_ITERS 100000
10319 #define RDTSC_DIFF_FAILS 100
10320 #define HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD 750
10321 
10322 /*
10323  * Set 'use TSC offsetting' and set the guest offset to the
10324  * inverse of the host's current TSC value, so that the guest starts running
10325  * with an effective TSC value of 0.
10326  */
reset_guest_tsc_to_zero(void)10327 static void reset_guest_tsc_to_zero(void)
10328 {
10329 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
10330 	vmcs_write(TSC_OFFSET, -rdtsc());
10331 }
10332 
rdtsc_vmexit_diff_test_guest(void)10333 static void rdtsc_vmexit_diff_test_guest(void)
10334 {
10335 	int i;
10336 
10337 	for (i = 0; i < RDTSC_DIFF_ITERS; i++)
10338 		/* Ensure rdtsc is the last instruction before the vmcall. */
10339 		asm volatile("rdtsc; vmcall" : : : "eax", "edx");
10340 }
10341 
10342 /*
10343  * This function only considers the "use TSC offsetting" VM-execution
10344  * control.  It does not handle "use TSC scaling" (because the latter
10345  * isn't available to the host today.)
10346  */
host_time_to_guest_time(unsigned long long t)10347 static unsigned long long host_time_to_guest_time(unsigned long long t)
10348 {
10349 	TEST_ASSERT(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
10350 		    !(vmcs_read(CPU_EXEC_CTRL1) & CPU_USE_TSC_SCALING));
10351 
10352 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_USE_TSC_OFFSET)
10353 		t += vmcs_read(TSC_OFFSET);
10354 
10355 	return t;
10356 }
10357 
rdtsc_vmexit_diff_test_iteration(void)10358 static unsigned long long rdtsc_vmexit_diff_test_iteration(void)
10359 {
10360 	unsigned long long guest_tsc, host_to_guest_tsc;
10361 
10362 	enter_guest();
10363 	skip_exit_vmcall();
10364 	guest_tsc = (u32) regs.rax + (regs.rdx << 32);
10365 	host_to_guest_tsc = host_time_to_guest_time(exit_msr_store[0].value);
10366 
10367 	return host_to_guest_tsc - guest_tsc;
10368 }
10369 
rdtsc_vmexit_diff_test(void)10370 static void rdtsc_vmexit_diff_test(void)
10371 {
10372 	unsigned long long delta;
10373 	int fail = 0;
10374 	int i;
10375 
10376 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET))
10377 		test_skip("CPU doesn't support the 'use TSC offsetting' processor-based VM-execution control.\n");
10378 
10379 	test_set_guest(rdtsc_vmexit_diff_test_guest);
10380 
10381 	reset_guest_tsc_to_zero();
10382 
10383 	/*
10384 	 * Set up the VMCS12 VM-exit MSR-store list to store just one
10385 	 * MSR: IA32_TIME_STAMP_COUNTER. Note that the value stored is
10386 	 * in the host time domain (i.e., it is not adjusted according
10387 	 * to the TSC multiplier and TSC offset fields in the VMCS12,
10388 	 * as a guest RDTSC would be.)
10389 	 */
10390 	exit_msr_store = alloc_page();
10391 	exit_msr_store[0].index = MSR_IA32_TSC;
10392 	vmcs_write(EXI_MSR_ST_CNT, 1);
10393 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(exit_msr_store));
10394 
10395 	for (i = 0; i < RDTSC_DIFF_ITERS && fail < RDTSC_DIFF_FAILS; i++) {
10396 		delta = rdtsc_vmexit_diff_test_iteration();
10397 		if (delta >= HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD)
10398 			fail++;
10399 	}
10400 
10401 	enter_guest();
10402 
10403 	report(fail < RDTSC_DIFF_FAILS,
10404 	       "RDTSC to VM-exit delta too high in %d of %d iterations, last = %llu",
10405 	       fail, i, delta);
10406 }
10407 
invalid_msr_init(struct vmcs * vmcs)10408 static int invalid_msr_init(struct vmcs *vmcs)
10409 {
10410 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
10411 		printf("\tPreemption timer is not supported\n");
10412 		return VMX_TEST_EXIT;
10413 	}
10414 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
10415 	preempt_val = 10000000;
10416 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
10417 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
10418 
10419 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
10420 		printf("\tSave preemption value is not supported\n");
10421 
10422 	vmcs_write(ENT_MSR_LD_CNT, 1);
10423 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
10424 
10425 	return VMX_TEST_START;
10426 }
10427 
10428 
invalid_msr_main(void)10429 static void invalid_msr_main(void)
10430 {
10431 	report_fail("Invalid MSR load");
10432 }
10433 
invalid_msr_exit_handler(union exit_reason exit_reason)10434 static int invalid_msr_exit_handler(union exit_reason exit_reason)
10435 {
10436 	report_fail("Invalid MSR load");
10437 	print_vmexit_info(exit_reason);
10438 	return VMX_TEST_EXIT;
10439 }
10440 
invalid_msr_entry_failure(struct vmentry_result * result)10441 static int invalid_msr_entry_failure(struct vmentry_result *result)
10442 {
10443 	report(result->exit_reason.failed_vmentry &&
10444 	       result->exit_reason.basic == VMX_FAIL_MSR, "Invalid MSR load");
10445 	return VMX_TEST_VMEXIT;
10446 }
10447 
10448 /*
10449  * The max number of MSRs in an atomic switch MSR list is:
10450  * (111B + 1) * 512 = 4096
10451  *
10452  * Each list entry consumes:
10453  * 4-byte MSR index + 4 bytes reserved + 8-byte data = 16 bytes
10454  *
10455  * Allocate 128 kB to cover max_msr_list_size (i.e., 64 kB) and then some.
10456  */
10457 static const u32 msr_list_page_order = 5;
10458 
atomic_switch_msr_limit_test_guest(void)10459 static void atomic_switch_msr_limit_test_guest(void)
10460 {
10461 	vmcall();
10462 }
10463 
populate_msr_list(struct vmx_msr_entry * msr_list,size_t byte_capacity,int count)10464 static void populate_msr_list(struct vmx_msr_entry *msr_list,
10465 			      size_t byte_capacity, int count)
10466 {
10467 	int i;
10468 
10469 	for (i = 0; i < count; i++) {
10470 		msr_list[i].index = MSR_IA32_TSC;
10471 		msr_list[i].reserved = 0;
10472 		msr_list[i].value = 0x1234567890abcdef;
10473 	}
10474 
10475 	memset(msr_list + count, 0xff,
10476 	       byte_capacity - count * sizeof(*msr_list));
10477 }
10478 
max_msr_list_size(void)10479 static int max_msr_list_size(void)
10480 {
10481 	u32 vmx_misc = rdmsr(MSR_IA32_VMX_MISC);
10482 	u32 factor = ((vmx_misc & GENMASK(27, 25)) >> 25) + 1;
10483 
10484 	return factor * 512;
10485 }
10486 
atomic_switch_msrs_test(int count)10487 static void atomic_switch_msrs_test(int count)
10488 {
10489 	struct vmx_msr_entry *vm_enter_load;
10490         struct vmx_msr_entry *vm_exit_load;
10491         struct vmx_msr_entry *vm_exit_store;
10492 	int max_allowed = max_msr_list_size();
10493 	int byte_capacity = 1ul << (msr_list_page_order + PAGE_SHIFT);
10494 	/* Exceeding the max MSR list size at exit triggers KVM to abort. */
10495 	int exit_count = count > max_allowed ? max_allowed : count;
10496 	int cleanup_count = count > max_allowed ? 2 : 1;
10497 	int i;
10498 
10499 	/*
10500 	 * Check for the IA32_TSC MSR,
10501 	 * available with the "TSC flag" and used to populate the MSR lists.
10502 	 */
10503 	if (!(cpuid(1).d & (1 << 4))) {
10504 		report_skip("%s : \"Time Stamp Counter\" not supported", __func__);
10505 		return;
10506 	}
10507 
10508 	/* Set L2 guest. */
10509 	test_set_guest(atomic_switch_msr_limit_test_guest);
10510 
10511 	/* Setup atomic MSR switch lists. */
10512 	vm_enter_load = alloc_pages(msr_list_page_order);
10513 	vm_exit_load = alloc_pages(msr_list_page_order);
10514 	vm_exit_store = alloc_pages(msr_list_page_order);
10515 
10516 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)vm_enter_load);
10517 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)vm_exit_load);
10518 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)vm_exit_store);
10519 
10520 	/*
10521 	 * VM-Enter should succeed up to the max number of MSRs per list, and
10522 	 * should not consume junk beyond the last entry.
10523 	 */
10524 	populate_msr_list(vm_enter_load, byte_capacity, count);
10525 	populate_msr_list(vm_exit_load, byte_capacity, exit_count);
10526 	populate_msr_list(vm_exit_store, byte_capacity, exit_count);
10527 
10528 	vmcs_write(ENT_MSR_LD_CNT, count);
10529 	vmcs_write(EXI_MSR_LD_CNT, exit_count);
10530 	vmcs_write(EXI_MSR_ST_CNT, exit_count);
10531 
10532 	if (count <= max_allowed) {
10533 		enter_guest();
10534 		assert_exit_reason(VMX_VMCALL);
10535 		skip_exit_vmcall();
10536 	} else {
10537 		u32 exit_qual;
10538 
10539 		test_guest_state("Invalid MSR Load Count", true, count,
10540 				 "ENT_MSR_LD_CNT");
10541 
10542 		exit_qual = vmcs_read(EXI_QUALIFICATION);
10543 		report(exit_qual == max_allowed + 1, "exit_qual, %u, is %u.",
10544 		       exit_qual, max_allowed + 1);
10545 	}
10546 
10547 	/* Cleanup. */
10548 	vmcs_write(ENT_MSR_LD_CNT, 0);
10549 	vmcs_write(EXI_MSR_LD_CNT, 0);
10550 	vmcs_write(EXI_MSR_ST_CNT, 0);
10551 	for (i = 0; i < cleanup_count; i++) {
10552 		enter_guest();
10553 		skip_exit_vmcall();
10554 	}
10555 	free_pages_by_order(vm_enter_load, msr_list_page_order);
10556 	free_pages_by_order(vm_exit_load, msr_list_page_order);
10557 	free_pages_by_order(vm_exit_store, msr_list_page_order);
10558 }
10559 
atomic_switch_max_msrs_test(void)10560 static void atomic_switch_max_msrs_test(void)
10561 {
10562 	atomic_switch_msrs_test(max_msr_list_size());
10563 }
10564 
atomic_switch_overflow_msrs_test(void)10565 static void atomic_switch_overflow_msrs_test(void)
10566 {
10567 	if (test_device_enabled())
10568 		atomic_switch_msrs_test(max_msr_list_size() + 1);
10569 	else
10570 		test_skip("Test is only supported on KVM");
10571 }
10572 
vmx_pf_exception_test_guest(void)10573 static void vmx_pf_exception_test_guest(void)
10574 {
10575 	ac_test_run(PT_LEVEL_PML4, false);
10576 }
10577 
vmx_pf_exception_forced_emulation_test_guest(void)10578 static void vmx_pf_exception_forced_emulation_test_guest(void)
10579 {
10580 	ac_test_run(PT_LEVEL_PML4, true);
10581 }
10582 
10583 typedef void (*invalidate_tlb_t)(void *data);
10584 typedef void (*pf_exception_test_guest_t)(void);
10585 
10586 
__vmx_pf_exception_test(invalidate_tlb_t inv_fn,void * data,pf_exception_test_guest_t guest_fn)10587 static void __vmx_pf_exception_test(invalidate_tlb_t inv_fn, void *data,
10588 				    pf_exception_test_guest_t guest_fn)
10589 {
10590 	u64 efer;
10591 	struct cpuid cpuid;
10592 
10593 	test_set_guest(guest_fn);
10594 
10595 	/* Intercept INVLPG when to perform TLB invalidation from L1 (this). */
10596 	if (inv_fn)
10597 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INVLPG);
10598 	else
10599 		vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INVLPG);
10600 
10601 	enter_guest();
10602 
10603 	while (vmcs_read(EXI_REASON) != VMX_VMCALL) {
10604 		switch (vmcs_read(EXI_REASON)) {
10605 		case VMX_RDMSR:
10606 			assert(regs.rcx == MSR_EFER);
10607 			efer = vmcs_read(GUEST_EFER);
10608 			regs.rdx = efer >> 32;
10609 			regs.rax = efer & 0xffffffff;
10610 			break;
10611 		case VMX_WRMSR:
10612 			assert(regs.rcx == MSR_EFER);
10613 			efer = regs.rdx << 32 | (regs.rax & 0xffffffff);
10614 			vmcs_write(GUEST_EFER, efer);
10615 			break;
10616 		case VMX_CPUID:
10617 			cpuid = (struct cpuid) {0, 0, 0, 0};
10618 			cpuid = raw_cpuid(regs.rax, regs.rcx);
10619 			regs.rax = cpuid.a;
10620 			regs.rbx = cpuid.b;
10621 			regs.rcx = cpuid.c;
10622 			regs.rdx = cpuid.d;
10623 			break;
10624 		case VMX_INVLPG:
10625 			inv_fn(data);
10626 			break;
10627 		default:
10628 			assert_msg(false,
10629 				"Unexpected exit to L1, exit_reason: %s (0x%lx)",
10630 				exit_reason_description(vmcs_read(EXI_REASON)),
10631 				vmcs_read(EXI_REASON));
10632 		}
10633 		skip_exit_insn();
10634 		enter_guest();
10635 	}
10636 
10637 	assert_exit_reason(VMX_VMCALL);
10638 }
10639 
vmx_pf_exception_test(void)10640 static void vmx_pf_exception_test(void)
10641 {
10642 	__vmx_pf_exception_test(NULL, NULL, vmx_pf_exception_test_guest);
10643 }
10644 
vmx_pf_exception_forced_emulation_test(void)10645 static void vmx_pf_exception_forced_emulation_test(void)
10646 {
10647 	__vmx_pf_exception_test(NULL, NULL, vmx_pf_exception_forced_emulation_test_guest);
10648 }
10649 
invalidate_tlb_no_vpid(void * data)10650 static void invalidate_tlb_no_vpid(void *data)
10651 {
10652 	/* If VPID is disabled, the TLB is flushed on VM-Enter and VM-Exit. */
10653 }
10654 
vmx_pf_no_vpid_test(void)10655 static void vmx_pf_no_vpid_test(void)
10656 {
10657 	if (is_vpid_supported())
10658 		vmcs_clear_bits(CPU_EXEC_CTRL1, CPU_VPID);
10659 
10660 	__vmx_pf_exception_test(invalidate_tlb_no_vpid, NULL,
10661 				vmx_pf_exception_test_guest);
10662 }
10663 
invalidate_tlb_invvpid_addr(void * data)10664 static void invalidate_tlb_invvpid_addr(void *data)
10665 {
10666 	invvpid(INVVPID_ALL, *(u16 *)data, vmcs_read(EXI_QUALIFICATION));
10667 }
10668 
invalidate_tlb_new_vpid(void * data)10669 static void invalidate_tlb_new_vpid(void *data)
10670 {
10671 	u16 *vpid = data;
10672 
10673 	/*
10674 	 * Bump VPID to effectively flush L2's TLB from L0's perspective.
10675 	 * Invalidate all VPIDs when the VPID wraps to zero as hardware/KVM is
10676 	 * architecturally allowed to keep TLB entries indefinitely.
10677 	 */
10678 	++(*vpid);
10679 	if (*vpid == 0) {
10680 		++(*vpid);
10681 		invvpid(INVVPID_ALL, 0, 0);
10682 	}
10683 	vmcs_write(VPID, *vpid);
10684 }
10685 
__vmx_pf_vpid_test(invalidate_tlb_t inv_fn,u16 vpid)10686 static void __vmx_pf_vpid_test(invalidate_tlb_t inv_fn, u16 vpid)
10687 {
10688 	if (!is_vpid_supported())
10689 		test_skip("VPID unsupported");
10690 
10691 	if (!is_invvpid_supported())
10692 		test_skip("INVVPID unsupported");
10693 
10694 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
10695 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VPID);
10696 	vmcs_write(VPID, vpid);
10697 
10698 	__vmx_pf_exception_test(inv_fn, &vpid, vmx_pf_exception_test_guest);
10699 }
10700 
vmx_pf_invvpid_test(void)10701 static void vmx_pf_invvpid_test(void)
10702 {
10703 	if (!is_invvpid_type_supported(INVVPID_ADDR))
10704 		test_skip("INVVPID ADDR unsupported");
10705 
10706 	__vmx_pf_vpid_test(invalidate_tlb_invvpid_addr, 0xaaaa);
10707 }
10708 
vmx_pf_vpid_test(void)10709 static void vmx_pf_vpid_test(void)
10710 {
10711 	/* Need INVVPID(ALL) to flush VPIDs upon wrap/reuse. */
10712 	if (!is_invvpid_type_supported(INVVPID_ALL))
10713 		test_skip("INVVPID ALL unsupported");
10714 
10715 	__vmx_pf_vpid_test(invalidate_tlb_new_vpid, 1);
10716 }
10717 
vmx_l2_ac_test(void)10718 static void vmx_l2_ac_test(void)
10719 {
10720 	bool hit_ac = false;
10721 
10722 	write_cr0(read_cr0() | X86_CR0_AM);
10723 	write_rflags(read_rflags() | X86_EFLAGS_AC);
10724 
10725 	run_in_user(generate_usermode_ac, AC_VECTOR, 0, 0, 0, 0, &hit_ac);
10726 	report(hit_ac, "Usermode #AC handled in L2");
10727 	vmcall();
10728 }
10729 
10730 struct vmx_exception_test {
10731 	u8 vector;
10732 	void (*guest_code)(void);
10733 };
10734 
10735 struct vmx_exception_test vmx_exception_tests[] = {
10736 	{ GP_VECTOR, generate_non_canonical_gp },
10737 	{ UD_VECTOR, generate_ud },
10738 	{ DE_VECTOR, generate_de },
10739 	{ DB_VECTOR, generate_single_step_db },
10740 	{ BP_VECTOR, generate_bp },
10741 	{ AC_VECTOR, vmx_l2_ac_test },
10742 	{ OF_VECTOR, generate_of },
10743 	{ NM_VECTOR, generate_cr0_ts_nm },
10744 	{ NM_VECTOR, generate_cr0_em_nm },
10745 };
10746 
10747 static u8 vmx_exception_test_vector;
10748 
vmx_exception_handler(struct ex_regs * regs)10749 static void vmx_exception_handler(struct ex_regs *regs)
10750 {
10751 	report(regs->vector == vmx_exception_test_vector,
10752 	       "Handling %s in L2's exception handler",
10753 	       exception_mnemonic(vmx_exception_test_vector));
10754 	vmcall();
10755 }
10756 
handle_exception_in_l2(u8 vector)10757 static void handle_exception_in_l2(u8 vector)
10758 {
10759 	handler old_handler = handle_exception(vector, vmx_exception_handler);
10760 
10761 	vmx_exception_test_vector = vector;
10762 
10763 	enter_guest();
10764 	report(vmcs_read(EXI_REASON) == VMX_VMCALL,
10765 	       "%s handled by L2", exception_mnemonic(vector));
10766 
10767 	handle_exception(vector, old_handler);
10768 }
10769 
handle_exception_in_l1(u32 vector)10770 static void handle_exception_in_l1(u32 vector)
10771 {
10772 	u32 old_eb = vmcs_read(EXC_BITMAP);
10773 	u32 intr_type;
10774 	u32 intr_info;
10775 
10776 	vmcs_write(EXC_BITMAP, old_eb | (1u << vector));
10777 
10778 	enter_guest();
10779 
10780 	if (vector == BP_VECTOR || vector == OF_VECTOR)
10781 		intr_type = VMX_INTR_TYPE_SOFT_EXCEPTION;
10782 	else
10783 		intr_type = VMX_INTR_TYPE_HARD_EXCEPTION;
10784 
10785 	intr_info = vmcs_read(EXI_INTR_INFO);
10786 	report((vmcs_read(EXI_REASON) == VMX_EXC_NMI) &&
10787 	       (intr_info & INTR_INFO_VALID_MASK) &&
10788 	       (intr_info & INTR_INFO_VECTOR_MASK) == vector &&
10789 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >> INTR_INFO_INTR_TYPE_SHIFT) == intr_type,
10790 	       "%s correctly routed to L1", exception_mnemonic(vector));
10791 
10792 	vmcs_write(EXC_BITMAP, old_eb);
10793 }
10794 
vmx_exception_test(void)10795 static void vmx_exception_test(void)
10796 {
10797 	struct vmx_exception_test *t;
10798 	int i;
10799 
10800 	for (i = 0; i < ARRAY_SIZE(vmx_exception_tests); i++) {
10801 		t = &vmx_exception_tests[i];
10802 
10803 		/*
10804 		 * Override the guest code before each run even though it's the
10805 		 * same code, the VMCS guest state needs to be reinitialized.
10806 		 */
10807 		test_override_guest(t->guest_code);
10808 		handle_exception_in_l2(t->vector);
10809 
10810 		test_override_guest(t->guest_code);
10811 		handle_exception_in_l1(t->vector);
10812 	}
10813 
10814 	test_set_guest_finished();
10815 }
10816 
10817 /*
10818  * Arbitrary canonical values for validating direct writes in the host, e.g. to
10819  * an MSR, and for indirect writes via loads from VMCS fields on VM-Exit.
10820  */
10821 #define TEST_DIRECT_VALUE 	0xff45454545000000
10822 #define TEST_VMCS_VALUE		0xff55555555000000
10823 
vmx_canonical_test_guest(void)10824 static void vmx_canonical_test_guest(void)
10825 {
10826 	while (true)
10827 		vmcall();
10828 }
10829 
get_host_value(u64 vmcs_field,u64 * value)10830 static int get_host_value(u64 vmcs_field, u64 *value)
10831 {
10832 	struct descriptor_table_ptr dt_ptr;
10833 
10834 	switch (vmcs_field) {
10835 	case HOST_SYSENTER_ESP:
10836 		*value = rdmsr(MSR_IA32_SYSENTER_ESP);
10837 		break;
10838 	case HOST_SYSENTER_EIP:
10839 		*value =  rdmsr(MSR_IA32_SYSENTER_EIP);
10840 		break;
10841 	case HOST_BASE_FS:
10842 		*value =  rdmsr(MSR_FS_BASE);
10843 		break;
10844 	case HOST_BASE_GS:
10845 		*value =  rdmsr(MSR_GS_BASE);
10846 		break;
10847 	case HOST_BASE_GDTR:
10848 		sgdt(&dt_ptr);
10849 		*value =  dt_ptr.base;
10850 		break;
10851 	case HOST_BASE_IDTR:
10852 		sidt(&dt_ptr);
10853 		*value =  dt_ptr.base;
10854 		break;
10855 	case HOST_BASE_TR:
10856 		*value = get_gdt_entry_base(get_tss_descr());
10857 		/* value might not reflect the actual base if changed by VMX */
10858 		return 1;
10859 	default:
10860 		assert(0);
10861 		return 1;
10862 	}
10863 	return 0;
10864 }
10865 
set_host_value(u64 vmcs_field,u64 value)10866 static int set_host_value(u64 vmcs_field, u64 value)
10867 {
10868 	struct descriptor_table_ptr dt_ptr;
10869 
10870 	switch (vmcs_field) {
10871 	case HOST_SYSENTER_ESP:
10872 		return wrmsr_safe(MSR_IA32_SYSENTER_ESP, value);
10873 	case HOST_SYSENTER_EIP:
10874 		return wrmsr_safe(MSR_IA32_SYSENTER_EIP, value);
10875 	case HOST_BASE_FS:
10876 		return wrmsr_safe(MSR_FS_BASE, value);
10877 	case HOST_BASE_GS:
10878 		/* TODO: _safe variants assume per-cpu gs base*/
10879 		wrmsr(MSR_GS_BASE, value);
10880 		return 0;
10881 	case HOST_BASE_GDTR:
10882 		sgdt(&dt_ptr);
10883 		dt_ptr.base = value;
10884 		lgdt(&dt_ptr);
10885 		return lgdt_fep_safe(&dt_ptr);
10886 	case HOST_BASE_IDTR:
10887 		sidt(&dt_ptr);
10888 		dt_ptr.base = value;
10889 		return lidt_fep_safe(&dt_ptr);
10890 	case HOST_BASE_TR:
10891 		/* Set the base and clear the busy bit */
10892 		set_gdt_entry(FIRST_SPARE_SEL, value, 0x200, 0x89, 0);
10893 		return ltr_safe(FIRST_SPARE_SEL);
10894 	default:
10895 		assert(0);
10896 	}
10897 }
10898 
test_host_value_direct(const char * field_name,u64 vmcs_field)10899 static void test_host_value_direct(const char *field_name, u64 vmcs_field)
10900 {
10901 	u64 value = 0;
10902 	int vector;
10903 
10904 	/*
10905 	 * Set the directly register via host ISA (e.g lgdt) and check that we
10906 	 * got no exception.
10907 	 */
10908 	vector = set_host_value(vmcs_field, TEST_DIRECT_VALUE);
10909 	if (vector) {
10910 		report_fail("Exception %d when setting %s to 0x%lx via direct write",
10911 			    vector, field_name, TEST_DIRECT_VALUE);
10912 		return;
10913 	}
10914 
10915 	/*
10916 	 * Now check that the host value matches what we expect for fields
10917 	 * that can be read back (these that we can't we assume that are correct)
10918 	 */
10919 	report(get_host_value(vmcs_field, &value) || value == TEST_DIRECT_VALUE,
10920 	       "%s: HOST value set to 0x%lx (wanted 0x%lx) via direct write",
10921 	       field_name, value, TEST_DIRECT_VALUE);
10922 }
10923 
test_host_value_vmcs(const char * field_name,u64 vmcs_field)10924 static void test_host_value_vmcs(const char *field_name, u64 vmcs_field)
10925 {
10926 	u64 value = 0;
10927 
10928 	/* Set host state field in the vmcs and do the VM entry
10929 	 * Success of VM entry already shows that L0 accepted the value
10930 	 */
10931 	vmcs_write(vmcs_field, TEST_VMCS_VALUE);
10932 	enter_guest();
10933 	skip_exit_vmcall();
10934 
10935 	/*
10936 	 * Now check that the host value matches what we expect for fields
10937 	 * that can be read back (these that we can't we assume that are correct)
10938 	 */
10939 	report(get_host_value(vmcs_field, &value) || value == TEST_VMCS_VALUE,
10940 	       "%s: HOST value set to 0x%lx (wanted 0x%lx) via VMLAUNCH/VMRESUME",
10941 	       field_name, value, TEST_VMCS_VALUE);
10942 }
10943 
do_vmx_canonical_test_one_field(const char * field_name,u64 field)10944 static void do_vmx_canonical_test_one_field(const char *field_name, u64 field)
10945 {
10946 	u64 host_org_value, field_org_value;
10947 
10948 	/* Backup the current host value and vmcs field value values */
10949 	get_host_value(field, &host_org_value);
10950 	field_org_value = vmcs_read(field);
10951 
10952 	test_host_value_direct(field_name, field);
10953 	test_host_value_vmcs(field_name, field);
10954 
10955 	/* Restore original values */
10956 	vmcs_write(field, field_org_value);
10957 	set_host_value(field, host_org_value);
10958 }
10959 
10960 #define vmx_canonical_test_one_field(field) \
10961 	do_vmx_canonical_test_one_field(#field, field)
10962 
vmx_canonical_test(void)10963 static void vmx_canonical_test(void)
10964 {
10965 	report(!(read_cr4() & X86_CR4_LA57), "4 level paging");
10966 
10967 	if (!this_cpu_has(X86_FEATURE_LA57))
10968 		test_skip("5 level paging not supported");
10969 
10970 	test_set_guest(vmx_canonical_test_guest);
10971 
10972 	vmx_canonical_test_one_field(HOST_SYSENTER_ESP);
10973 	vmx_canonical_test_one_field(HOST_SYSENTER_EIP);
10974 	vmx_canonical_test_one_field(HOST_BASE_FS);
10975 	vmx_canonical_test_one_field(HOST_BASE_GS);
10976 	vmx_canonical_test_one_field(HOST_BASE_GDTR);
10977 	vmx_canonical_test_one_field(HOST_BASE_IDTR);
10978 	vmx_canonical_test_one_field(HOST_BASE_TR);
10979 
10980 	test_set_guest_finished();
10981 }
10982 
10983 enum Vid_op {
10984 	VID_OP_SET_ISR,
10985 	VID_OP_NOP,
10986 	VID_OP_SET_CR8,
10987 	VID_OP_SELF_IPI,
10988 	VID_OP_TERMINATE,
10989 	VID_OP_SPIN,
10990 	VID_OP_SPIN_IRR,
10991 	VID_OP_HLT,
10992 };
10993 
10994 struct vmx_basic_vid_test_guest_args {
10995 	enum Vid_op op;
10996 	u8 nr;
10997 	u32 isr_exec_cnt;
10998 	u32 *virtual_apic_page;
10999 	u64 *pi_desc;
11000 	u32 dest;
11001 	bool in_guest;
11002 } vmx_basic_vid_test_guest_args;
11003 
11004 /*
11005  * From the SDM, Bit x of the VIRR is
11006  *     at bit position (x & 1FH)
11007  *     at offset (200H | ((x & E0H) >> 1)).
11008  */
set_virr_bit(volatile u32 * virtual_apic_page,u8 nr)11009 static void set_virr_bit(volatile u32 *virtual_apic_page, u8 nr)
11010 {
11011 	u32 page_offset = (0x200 | ((nr & 0xE0) >> 1)) / sizeof(u32);
11012 	u32 mask = 1 << (nr & 0x1f);
11013 
11014 	virtual_apic_page[page_offset] |= mask;
11015 }
11016 
clear_virr_bit(volatile u32 * virtual_apic_page,u8 nr)11017 static void clear_virr_bit(volatile u32 *virtual_apic_page, u8 nr)
11018 {
11019 	u32 page_offset = (0x200 | ((nr & 0xE0) >> 1)) / sizeof(u32);
11020 	u32 mask = 1 << (nr & 0x1f);
11021 
11022 	virtual_apic_page[page_offset] &= ~mask;
11023 }
11024 
get_virr_bit(volatile u32 * virtual_apic_page,u8 nr)11025 static bool get_virr_bit(volatile u32 *virtual_apic_page, u8 nr)
11026 {
11027 	u32 page_offset = (0x200 | ((nr & 0xE0) >> 1)) / sizeof(u32);
11028 	u32 mask = 1 << (nr & 0x1f);
11029 
11030 	return virtual_apic_page[page_offset] & mask;
11031 }
11032 
vmx_vid_test_isr(isr_regs_t * regs)11033 static void vmx_vid_test_isr(isr_regs_t *regs)
11034 {
11035 	volatile struct vmx_basic_vid_test_guest_args *args =
11036 		&vmx_basic_vid_test_guest_args;
11037 
11038 	args->isr_exec_cnt++;
11039 	barrier();
11040 	eoi();
11041 }
11042 
vmx_basic_vid_test_guest(void)11043 static void vmx_basic_vid_test_guest(void)
11044 {
11045 	volatile struct vmx_basic_vid_test_guest_args *args =
11046 		&vmx_basic_vid_test_guest_args;
11047 
11048 	sti_nop();
11049 	for (;;) {
11050 		enum Vid_op op = args->op;
11051 		u8 nr = args->nr;
11052 
11053 		switch (op) {
11054 		case VID_OP_TERMINATE:
11055 			return;
11056 		case VID_OP_SET_ISR:
11057 			handle_irq(nr, vmx_vid_test_isr);
11058 			break;
11059 		case VID_OP_SET_CR8:
11060 			write_cr8(nr);
11061 			break;
11062 		case VID_OP_SELF_IPI:
11063 			vmx_x2apic_write(APIC_SELF_IPI, nr);
11064 			break;
11065 		case VID_OP_HLT:
11066 			cli();
11067 			barrier();
11068 			args->in_guest = true;
11069 			barrier();
11070 			safe_halt();
11071 			break;
11072 		case VID_OP_SPIN:
11073 			args->in_guest = true;
11074 			while (!args->isr_exec_cnt)
11075 				pause();
11076 			break;
11077 		case VID_OP_SPIN_IRR: {
11078 			u32 *virtual_apic_page = args->virtual_apic_page;
11079 			u8 nr = args->nr;
11080 
11081 			args->in_guest = true;
11082 			while (!get_virr_bit(virtual_apic_page, nr))
11083 				pause();
11084 			clear_virr_bit(virtual_apic_page, nr);
11085 			break;
11086 		}
11087 		default:
11088 			break;
11089 		}
11090 
11091 		vmcall();
11092 	}
11093 }
11094 
set_isrs_for_vmx_basic_vid_test(void)11095 static void set_isrs_for_vmx_basic_vid_test(void)
11096 {
11097 	volatile struct vmx_basic_vid_test_guest_args *args =
11098 		&vmx_basic_vid_test_guest_args;
11099 	u16 nr;
11100 
11101 	/*
11102 	 * kvm-unit-tests uses vector 32 for IPIs, so don't install a test ISR
11103 	 * for that vector.
11104 	 */
11105 	for (nr = 0x21; nr < 0x100; nr++) {
11106 		vmcs_write(GUEST_INT_STATUS, 0);
11107 		args->virtual_apic_page = get_vapic_page();
11108 		args->op = VID_OP_SET_ISR;
11109 		args->nr = nr;
11110 		args->isr_exec_cnt = 0;
11111 		enter_guest();
11112 		skip_exit_vmcall();
11113 	}
11114 	report(true, "Set ISR for vectors 33-255.");
11115 }
11116 
vmx_posted_interrupts_test_worker(void * data)11117 static void vmx_posted_interrupts_test_worker(void *data)
11118 {
11119 	volatile struct vmx_basic_vid_test_guest_args *args =
11120 		&vmx_basic_vid_test_guest_args;
11121 
11122 	while (!args->in_guest)
11123 		pause();
11124 
11125 	test_and_set_bit(args->nr, args->pi_desc);
11126 	test_and_set_bit(256, args->pi_desc);
11127 	apic_icr_write(PI_VECTOR, args->dest);
11128 }
11129 
11130 /*
11131  * Test virtual interrupt delivery (VID) at VM-entry or TPR virtualization
11132  *
11133  * Args:
11134  *   nr: vector under test
11135  *   tpr: task priority under test
11136  *   tpr_virt: If true, then test VID during TPR virtualization. Otherwise,
11137  *       test VID during VM-entry.
11138  */
test_basic_vid(u8 nr,u8 tpr,enum Vid_op op,u32 isr_exec_cnt_want,bool eoi_exit_induced)11139 static void test_basic_vid(u8 nr, u8 tpr, enum Vid_op op, u32 isr_exec_cnt_want,
11140 			   bool eoi_exit_induced)
11141 {
11142 	volatile struct vmx_basic_vid_test_guest_args *args =
11143 		&vmx_basic_vid_test_guest_args;
11144 	u16 rvi_want = isr_exec_cnt_want ? 0 : nr;
11145 	u16 int_status;
11146 
11147 	/*
11148 	 * From the SDM:
11149 	 *     IF "interrupt-window exiting" is 0 AND
11150 	 *     RVI[7:4] > VPPR[7:4] (see Section 29.1.1 for definition of VPPR)
11151 	 *             THEN recognize a pending virtual interrupt;
11152 	 *         ELSE
11153 	 *             do not recognize a pending virtual interrupt;
11154 	 *     FI;
11155 	 *
11156 	 * Thus, VPPR dictates whether a virtual interrupt is recognized.
11157 	 * However, PPR virtualization, which occurs before virtual interrupt
11158 	 * delivery, sets VPPR to VTPR, when SVI is 0.
11159 	 */
11160 	args->isr_exec_cnt = 0;
11161 	args->virtual_apic_page = get_vapic_page();
11162 	args->op = op;
11163 	args->in_guest = false;
11164 	switch (op) {
11165 	case VID_OP_SELF_IPI:
11166 		vmcs_write(GUEST_INT_STATUS, 0);
11167 		args->nr = nr;
11168 		set_vtpr(0);
11169 		break;
11170 	case VID_OP_SET_CR8:
11171 		vmcs_write(GUEST_INT_STATUS, nr);
11172 		args->nr = task_priority_class(tpr);
11173 		set_vtpr(0xff);
11174 		break;
11175 	case VID_OP_SPIN:
11176 	case VID_OP_SPIN_IRR:
11177 	case VID_OP_HLT:
11178 		vmcs_write(GUEST_INT_STATUS, 0);
11179 		args->nr = nr;
11180 		set_vtpr(tpr);
11181 		barrier();
11182 		on_cpu_async(1, vmx_posted_interrupts_test_worker, NULL);
11183 		break;
11184 	default:
11185 		vmcs_write(GUEST_INT_STATUS, nr);
11186 		set_vtpr(tpr);
11187 		break;
11188 	}
11189 
11190 	enter_guest();
11191 	if (eoi_exit_induced) {
11192 		u32 exit_cnt;
11193 
11194 		assert_exit_reason(VMX_EOI_INDUCED);
11195 		for (exit_cnt = 1; exit_cnt < isr_exec_cnt_want; exit_cnt++) {
11196 			enter_guest();
11197 			assert_exit_reason(VMX_EOI_INDUCED);
11198 		}
11199 		enter_guest();
11200 	}
11201 	skip_exit_vmcall();
11202 	TEST_ASSERT_EQ(args->isr_exec_cnt, isr_exec_cnt_want);
11203 	int_status = vmcs_read(GUEST_INT_STATUS);
11204 	TEST_ASSERT_EQ(int_status, rvi_want);
11205 }
11206 
11207 /*
11208  * Test recognizing and delivering virtual interrupts via "Virtual-interrupt
11209  * delivery" for two scenarios:
11210  *   1. When there is a pending interrupt at VM-entry.
11211  *   2. When there is a pending interrupt during TPR virtualization.
11212  */
vmx_basic_vid_test(void)11213 static void vmx_basic_vid_test(void)
11214 {
11215 	volatile struct vmx_basic_vid_test_guest_args *args =
11216 		&vmx_basic_vid_test_guest_args;
11217 	u8 nr_class;
11218 
11219 	if (!cpu_has_apicv()) {
11220 		report_skip("%s : Not all required APICv bits supported", __func__);
11221 		return;
11222 	}
11223 
11224 	enable_vid();
11225 	test_set_guest(vmx_basic_vid_test_guest);
11226 	set_isrs_for_vmx_basic_vid_test();
11227 
11228 	for (nr_class = 2; nr_class < 16; nr_class++) {
11229 		u16 nr;
11230 		u8 nr_sub_class;
11231 
11232 		for (nr_sub_class = 0; nr_sub_class < 16; nr_sub_class++) {
11233 			u16 tpr;
11234 
11235 			nr = (nr_class << 4) | nr_sub_class;
11236 
11237 			/*
11238 			 * Don't test the reserved IPI vector, as the test ISR
11239 			 * was not installed.
11240 			 */
11241 			if (nr == 0x20)
11242 				continue;
11243 
11244 			test_basic_vid(nr, /*tpr=*/0, VID_OP_SELF_IPI,
11245 				       /*isr_exec_cnt_want=*/1,
11246 				       /*eoi_exit_induced=*/false);
11247 			for (tpr = 0; tpr < 256; tpr++) {
11248 				u32 isr_exec_cnt_want =
11249 					task_priority_class(nr) >
11250 					task_priority_class(tpr) ? 1 : 0;
11251 
11252 				test_basic_vid(nr, tpr, VID_OP_NOP,
11253 					       isr_exec_cnt_want,
11254 					       /*eoi_exit_induced=*/false);
11255 				test_basic_vid(nr, tpr, VID_OP_SET_CR8,
11256 					       isr_exec_cnt_want,
11257 					       /*eoi_exit_induced=*/false);
11258 			}
11259 			report(true, "TPR 0-255 for vector 0x%x.", nr);
11260 		}
11261 	}
11262 
11263 	/* Terminate the guest */
11264 	args->op = VID_OP_TERMINATE;
11265 	enter_guest();
11266 	assert_exit_reason(VMX_VMCALL);
11267 }
11268 
test_eoi_virt(u8 nr,u8 lo_pri_nr,bool eoi_exit_induced)11269 static void test_eoi_virt(u8 nr, u8 lo_pri_nr, bool eoi_exit_induced)
11270 {
11271 	u32 *virtual_apic_page = get_vapic_page();
11272 
11273 	set_virr_bit(virtual_apic_page, lo_pri_nr);
11274 	test_basic_vid(nr, /*tpr=*/0, VID_OP_NOP, /*isr_exec_cnt_want=*/2,
11275 		       eoi_exit_induced);
11276 	TEST_ASSERT(!get_virr_bit(virtual_apic_page, lo_pri_nr));
11277 	TEST_ASSERT(!get_virr_bit(virtual_apic_page, nr));
11278 }
11279 
vmx_eoi_virt_test(void)11280 static void vmx_eoi_virt_test(void)
11281 {
11282 	volatile struct vmx_basic_vid_test_guest_args *args =
11283 		&vmx_basic_vid_test_guest_args;
11284 	u16 nr;
11285 	u16 lo_pri_nr;
11286 
11287 	if (!cpu_has_apicv()) {
11288 		report_skip("%s : Not all required APICv bits supported", __func__);
11289 		return;
11290 	}
11291 
11292 	enable_vid();  /* Note, enable_vid sets APIC_VIRT_ADDR field in VMCS. */
11293 	test_set_guest(vmx_basic_vid_test_guest);
11294 	set_isrs_for_vmx_basic_vid_test();
11295 
11296 	/* Now test EOI virtualization without induced EOI exits. */
11297 	for (nr = 0x22; nr < 0x100; nr++) {
11298 		for (lo_pri_nr = 0x21; lo_pri_nr < nr; lo_pri_nr++)
11299 			test_eoi_virt(nr, lo_pri_nr,
11300 				      /*eoi_exit_induced=*/false);
11301 
11302 		report(true, "Low priority nrs 0x21-0x%x for nr 0x%x.",
11303 		       nr - 1, nr);
11304 	}
11305 
11306 	/* Finally, test EOI virtualization with induced EOI exits. */
11307 	vmcs_write(EOI_EXIT_BITMAP0, GENMASK_ULL(63, 0));
11308 	vmcs_write(EOI_EXIT_BITMAP1, GENMASK_ULL(63, 0));
11309 	vmcs_write(EOI_EXIT_BITMAP2, GENMASK_ULL(63, 0));
11310 	vmcs_write(EOI_EXIT_BITMAP3, GENMASK_ULL(63, 0));
11311 	for (nr = 0x22; nr < 0x100; nr++) {
11312 		for (lo_pri_nr = 0x21; lo_pri_nr < nr; lo_pri_nr++)
11313 			test_eoi_virt(nr, lo_pri_nr,
11314 				      /*eoi_exit_induced=*/true);
11315 
11316 		report(true,
11317 		       "Low priority nrs 0x21-0x%x for nr 0x%x, with induced EOI exits.",
11318 		       nr - 1, nr);
11319 	}
11320 
11321 	/* Terminate the guest */
11322 	args->op = VID_OP_TERMINATE;
11323 	enter_guest();
11324 	assert_exit_reason(VMX_VMCALL);
11325 }
11326 
vmx_posted_interrupts_test(void)11327 static void vmx_posted_interrupts_test(void)
11328 {
11329 	volatile struct vmx_basic_vid_test_guest_args *args =
11330 		&vmx_basic_vid_test_guest_args;
11331 	u16 vector;
11332 	u8 class;
11333 
11334 	if (!cpu_has_apicv()) {
11335 		report_skip("%s : Not all required APICv bits supported", __func__);
11336 		return;
11337 	}
11338 
11339 	if (cpu_count() < 2) {
11340 		report_skip("%s : CPU count < 2", __func__);
11341 		return;
11342 	}
11343 
11344 	enable_vid();
11345 	enable_posted_interrupts();
11346 	args->pi_desc = get_pi_desc();
11347 	args->dest = apic_id();
11348 
11349 	test_set_guest(vmx_basic_vid_test_guest);
11350 	set_isrs_for_vmx_basic_vid_test();
11351 
11352 	for (class = 0; class < 16; class++) {
11353 		for (vector = 33; vector < 256; vector++) {
11354 			/*
11355 			 * If the vector isn't above TPR, then the vector should
11356 			 * be moved from PIR to the IRR, but never serviced.
11357 			 *
11358 			 * Only test posted interrupts to a halted vCPU if the
11359 			 * interrupt is expected to be serviced.  Otherwise, the
11360 			 * vCPU will HLT indefinitely.
11361 			 */
11362 			if (task_priority_class(vector) <= class) {
11363 				test_basic_vid(vector, class << 4,
11364 					       VID_OP_SPIN_IRR, 0, false);
11365 				continue;
11366 			}
11367 
11368 			test_basic_vid(vector, class << 4, VID_OP_SPIN, 1, false);
11369 			test_basic_vid(vector, class << 4, VID_OP_HLT, 1, false);
11370 		}
11371 	}
11372 	report(true, "Posted vectors 33-25 cross TPR classes 0-0xf, running and sometimes halted\n");
11373 
11374 	/* Terminate the guest */
11375 	args->op = VID_OP_TERMINATE;
11376 	enter_guest();
11377 }
11378 
11379 #define TEST(name) { #name, .v2 = name }
11380 
11381 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
11382 struct vmx_test vmx_tests[] = {
11383 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
11384 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
11385 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
11386 		preemption_timer_exit_handler, NULL, {0} },
11387 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
11388 		test_ctrl_pat_exit_handler, NULL, {0} },
11389 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
11390 		test_ctrl_efer_exit_handler, NULL, {0} },
11391 	{ "CR shadowing", NULL, cr_shadowing_main,
11392 		cr_shadowing_exit_handler, NULL, {0} },
11393 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
11394 		NULL, {0} },
11395 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
11396 		insn_intercept_exit_handler, NULL, {0} },
11397 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
11398 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
11399 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
11400 	{ "interrupt", interrupt_init, interrupt_main,
11401 		interrupt_exit_handler, NULL, {0} },
11402 	{ "nmi_hlt", nmi_hlt_init, nmi_hlt_main,
11403 		nmi_hlt_exit_handler, NULL, {0} },
11404 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
11405 		NULL, {0} },
11406 	{ "MSR switch", msr_switch_init, msr_switch_main,
11407 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
11408 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
11409 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
11410 		disable_rdtscp_exit_handler, NULL, {0} },
11411 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
11412 		exit_monitor_from_l2_handler, NULL, {0} },
11413 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
11414 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
11415 	/* Basic V2 tests. */
11416 	TEST(v2_null_test),
11417 	TEST(v2_multiple_entries_test),
11418 	TEST(fixture_test_case1),
11419 	TEST(fixture_test_case2),
11420 	/* Opcode tests. */
11421 	TEST(invvpid_test),
11422 	/* VM-entry tests */
11423 	TEST(vmx_controls_test),
11424 	TEST(vmx_host_state_area_test),
11425 	TEST(vmx_guest_state_area_test),
11426 	TEST(vmentry_movss_shadow_test),
11427 	TEST(vmentry_unrestricted_guest_test),
11428 	/* APICv tests */
11429 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
11430 	TEST(vmx_hlt_with_rvi_test),
11431 	TEST(apic_reg_virt_test),
11432 	TEST(virt_x2apic_mode_test),
11433 	TEST(vmx_basic_vid_test),
11434 	TEST(vmx_eoi_virt_test),
11435 	TEST(vmx_posted_interrupts_test),
11436 	/* APIC pass-through tests */
11437 	TEST(vmx_apic_passthrough_test),
11438 	TEST(vmx_apic_passthrough_thread_test),
11439 	TEST(vmx_apic_passthrough_tpr_threshold_test),
11440 	TEST(vmx_init_signal_test),
11441 	TEST(vmx_sipi_signal_test),
11442 	/* VMCS Shadowing tests */
11443 	TEST(vmx_vmcs_shadow_test),
11444 	/* Regression tests */
11445 	TEST(vmx_ldtr_test),
11446 	TEST(vmx_cr_load_test),
11447 	TEST(vmx_cr4_osxsave_test),
11448 	TEST(vmx_no_nm_test),
11449 	TEST(vmx_db_test),
11450 	TEST(vmx_nmi_window_test),
11451 	TEST(vmx_intr_window_test),
11452 	TEST(vmx_pending_event_test),
11453 	TEST(vmx_pending_event_hlt_test),
11454 	TEST(vmx_store_tsc_test),
11455 	TEST(vmx_preemption_timer_zero_test),
11456 	TEST(vmx_preemption_timer_tf_test),
11457 	TEST(vmx_preemption_timer_expiry_test),
11458 	/* EPT access tests. */
11459 	TEST(ept_access_test_not_present),
11460 	TEST(ept_access_test_read_only),
11461 	TEST(ept_access_test_write_only),
11462 	TEST(ept_access_test_read_write),
11463 	TEST(ept_access_test_execute_only),
11464 	TEST(ept_access_test_read_execute),
11465 	TEST(ept_access_test_write_execute),
11466 	TEST(ept_access_test_read_write_execute),
11467 	TEST(ept_access_test_reserved_bits),
11468 	TEST(ept_access_test_ignored_bits),
11469 	TEST(ept_access_test_paddr_not_present_ad_disabled),
11470 	TEST(ept_access_test_paddr_not_present_ad_enabled),
11471 	TEST(ept_access_test_paddr_read_only_ad_disabled),
11472 	TEST(ept_access_test_paddr_read_only_ad_enabled),
11473 	TEST(ept_access_test_paddr_read_write),
11474 	TEST(ept_access_test_paddr_read_write_execute),
11475 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
11476 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
11477 	TEST(ept_access_test_paddr_not_present_page_fault),
11478 	TEST(ept_access_test_force_2m_page),
11479 	/* Atomic MSR switch tests. */
11480 	TEST(atomic_switch_max_msrs_test),
11481 	TEST(atomic_switch_overflow_msrs_test),
11482 	TEST(rdtsc_vmexit_diff_test),
11483 	TEST(vmx_mtf_test),
11484 	TEST(vmx_mtf_pdpte_test),
11485 	TEST(vmx_pf_exception_test),
11486 	TEST(vmx_pf_exception_forced_emulation_test),
11487 	TEST(vmx_pf_no_vpid_test),
11488 	TEST(vmx_pf_invvpid_test),
11489 	TEST(vmx_pf_vpid_test),
11490 	TEST(vmx_exception_test),
11491 	TEST(vmx_canonical_test),
11492 	{ NULL, NULL, NULL, NULL, NULL, {0} },
11493 };
11494