xref: /kvm-unit-tests/x86/vmx_tests.c (revision b546af303081c4ddce726cac87041147aa7e2e63)
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 "vm.h"
13 #include "pci.h"
14 #include "fwcfg.h"
15 #include "isr.h"
16 #include "desc.h"
17 #include "apic.h"
18 #include "types.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 #define VPID_CAP_INVVPID_TYPES_SHIFT 40
27 
28 u64 ia32_pat;
29 u64 ia32_efer;
30 void *io_bitmap_a, *io_bitmap_b;
31 u16 ioport;
32 
33 unsigned long *pml4;
34 u64 eptp;
35 void *data_page1, *data_page2;
36 
37 phys_addr_t pci_physaddr;
38 
39 void *pml_log;
40 #define PML_INDEX 512
41 
42 static inline unsigned ffs(unsigned x)
43 {
44 	int pos = -1;
45 
46 	__asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0"
47 			     : "+r"(pos) : "rm"(x) : "eax");
48 	return pos + 1;
49 }
50 
51 static inline void vmcall(void)
52 {
53 	asm volatile("vmcall");
54 }
55 
56 static void basic_guest_main(void)
57 {
58 	report_pass("Basic VMX test");
59 }
60 
61 static int basic_exit_handler(union exit_reason exit_reason)
62 {
63 	report_fail("Basic VMX test");
64 	print_vmexit_info(exit_reason);
65 	return VMX_TEST_EXIT;
66 }
67 
68 static void vmenter_main(void)
69 {
70 	u64 rax;
71 	u64 rsp, resume_rsp;
72 
73 	report_pass("test vmlaunch");
74 
75 	asm volatile(
76 		"mov %%rsp, %0\n\t"
77 		"mov %3, %%rax\n\t"
78 		"vmcall\n\t"
79 		"mov %%rax, %1\n\t"
80 		"mov %%rsp, %2\n\t"
81 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
82 		: "g"(0xABCD));
83 	report((rax == 0xFFFF) && (rsp == resume_rsp), "test vmresume");
84 }
85 
86 static int vmenter_exit_handler(union exit_reason exit_reason)
87 {
88 	u64 guest_rip = vmcs_read(GUEST_RIP);
89 
90 	switch (exit_reason.basic) {
91 	case VMX_VMCALL:
92 		if (regs.rax != 0xABCD) {
93 			report_fail("test vmresume");
94 			return VMX_TEST_VMEXIT;
95 		}
96 		regs.rax = 0xFFFF;
97 		vmcs_write(GUEST_RIP, guest_rip + 3);
98 		return VMX_TEST_RESUME;
99 	default:
100 		report_fail("test vmresume");
101 		print_vmexit_info(exit_reason);
102 	}
103 	return VMX_TEST_VMEXIT;
104 }
105 
106 u32 preempt_scale;
107 volatile unsigned long long tsc_val;
108 volatile u32 preempt_val;
109 u64 saved_rip;
110 
111 static int preemption_timer_init(struct vmcs *vmcs)
112 {
113 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
114 		printf("\tPreemption timer is not supported\n");
115 		return VMX_TEST_EXIT;
116 	}
117 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
118 	preempt_val = 10000000;
119 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
120 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
121 
122 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
123 		printf("\tSave preemption value is not supported\n");
124 
125 	return VMX_TEST_START;
126 }
127 
128 static void preemption_timer_main(void)
129 {
130 	tsc_val = rdtsc();
131 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
132 		vmx_set_test_stage(0);
133 		vmcall();
134 		if (vmx_get_test_stage() == 1)
135 			vmcall();
136 	}
137 	vmx_set_test_stage(1);
138 	while (vmx_get_test_stage() == 1) {
139 		if (((rdtsc() - tsc_val) >> preempt_scale)
140 				> 10 * preempt_val) {
141 			vmx_set_test_stage(2);
142 			vmcall();
143 		}
144 	}
145 	tsc_val = rdtsc();
146 	asm volatile ("hlt");
147 	vmcall();
148 	vmx_set_test_stage(5);
149 	vmcall();
150 }
151 
152 static int preemption_timer_exit_handler(union exit_reason exit_reason)
153 {
154 	bool guest_halted;
155 	u64 guest_rip;
156 	u32 insn_len;
157 	u32 ctrl_exit;
158 
159 	guest_rip = vmcs_read(GUEST_RIP);
160 	insn_len = vmcs_read(EXI_INST_LEN);
161 	switch (exit_reason.basic) {
162 	case VMX_PREEMPT:
163 		switch (vmx_get_test_stage()) {
164 		case 1:
165 		case 2:
166 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val,
167 			       "busy-wait for preemption timer");
168 			vmx_set_test_stage(3);
169 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
170 			return VMX_TEST_RESUME;
171 		case 3:
172 			guest_halted =
173 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
174 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val
175 			        && guest_halted,
176 			       "preemption timer during hlt");
177 			vmx_set_test_stage(4);
178 			vmcs_write(PIN_CONTROLS,
179 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
180 			vmcs_write(EXI_CONTROLS,
181 				   vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT);
182 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
183 			return VMX_TEST_RESUME;
184 		case 4:
185 			report(saved_rip == guest_rip,
186 			       "preemption timer with 0 value");
187 			break;
188 		default:
189 			report_fail("Invalid stage.");
190 			print_vmexit_info(exit_reason);
191 			break;
192 		}
193 		break;
194 	case VMX_VMCALL:
195 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
196 		switch (vmx_get_test_stage()) {
197 		case 0:
198 			report(vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val,
199 			       "Keep preemption value");
200 			vmx_set_test_stage(1);
201 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
202 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
203 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
204 			vmcs_write(EXI_CONTROLS, ctrl_exit);
205 			return VMX_TEST_RESUME;
206 		case 1:
207 			report(vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val,
208 			       "Save preemption value");
209 			return VMX_TEST_RESUME;
210 		case 2:
211 			report_fail("busy-wait for preemption timer");
212 			vmx_set_test_stage(3);
213 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
214 			return VMX_TEST_RESUME;
215 		case 3:
216 			report_fail("preemption timer during hlt");
217 			vmx_set_test_stage(4);
218 			/* fall through */
219 		case 4:
220 			vmcs_write(PIN_CONTROLS,
221 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
222 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
223 			saved_rip = guest_rip + insn_len;
224 			return VMX_TEST_RESUME;
225 		case 5:
226 			report_fail("preemption timer with 0 value (vmcall stage 5)");
227 			break;
228 		default:
229 			// Should not reach here
230 			report_fail("unexpected stage, %d",
231 				    vmx_get_test_stage());
232 			print_vmexit_info(exit_reason);
233 			return VMX_TEST_VMEXIT;
234 		}
235 		break;
236 	default:
237 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
238 		print_vmexit_info(exit_reason);
239 	}
240 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
241 	return VMX_TEST_VMEXIT;
242 }
243 
244 static void msr_bmp_init(void)
245 {
246 	void *msr_bitmap;
247 	u32 ctrl_cpu0;
248 
249 	msr_bitmap = alloc_page();
250 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
251 	ctrl_cpu0 |= CPU_MSR_BITMAP;
252 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
253 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
254 }
255 
256 static void *get_msr_bitmap(void)
257 {
258 	void *msr_bitmap;
259 
260 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
261 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
262 	} else {
263 		msr_bitmap = alloc_page();
264 		memset(msr_bitmap, 0xff, PAGE_SIZE);
265 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
266 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
267 	}
268 
269 	return msr_bitmap;
270 }
271 
272 static void disable_intercept_for_x2apic_msrs(void)
273 {
274 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
275 	u32 msr;
276 
277 	for (msr = APIC_BASE_MSR;
278 		 msr < (APIC_BASE_MSR+0xff);
279 		 msr += BITS_PER_LONG) {
280 		unsigned int word = msr / BITS_PER_LONG;
281 
282 		msr_bitmap[word] = 0;
283 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
284 	}
285 }
286 
287 static int test_ctrl_pat_init(struct vmcs *vmcs)
288 {
289 	u64 ctrl_ent;
290 	u64 ctrl_exi;
291 
292 	msr_bmp_init();
293 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
294 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
295 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
296 		printf("\tSave/load PAT is not supported\n");
297 		return 1;
298 	}
299 
300 	ctrl_ent = vmcs_read(ENT_CONTROLS);
301 	ctrl_exi = vmcs_read(EXI_CONTROLS);
302 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
303 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
304 	vmcs_write(ENT_CONTROLS, ctrl_ent);
305 	vmcs_write(EXI_CONTROLS, ctrl_exi);
306 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
307 	vmcs_write(GUEST_PAT, 0x0);
308 	vmcs_write(HOST_PAT, ia32_pat);
309 	return VMX_TEST_START;
310 }
311 
312 static void test_ctrl_pat_main(void)
313 {
314 	u64 guest_ia32_pat;
315 
316 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
317 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
318 		printf("\tENT_LOAD_PAT is not supported.\n");
319 	else {
320 		if (guest_ia32_pat != 0) {
321 			report_fail("Entry load PAT");
322 			return;
323 		}
324 	}
325 	wrmsr(MSR_IA32_CR_PAT, 0x6);
326 	vmcall();
327 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
328 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
329 		report(guest_ia32_pat == ia32_pat, "Entry load PAT");
330 }
331 
332 static int test_ctrl_pat_exit_handler(union exit_reason exit_reason)
333 {
334 	u64 guest_rip;
335 	u64 guest_pat;
336 
337 	guest_rip = vmcs_read(GUEST_RIP);
338 	switch (exit_reason.basic) {
339 	case VMX_VMCALL:
340 		guest_pat = vmcs_read(GUEST_PAT);
341 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
342 			printf("\tEXI_SAVE_PAT is not supported\n");
343 			vmcs_write(GUEST_PAT, 0x6);
344 		} else {
345 			report(guest_pat == 0x6, "Exit save PAT");
346 		}
347 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
348 			printf("\tEXI_LOAD_PAT is not supported\n");
349 		else
350 			report(rdmsr(MSR_IA32_CR_PAT) == ia32_pat,
351 			       "Exit load PAT");
352 		vmcs_write(GUEST_PAT, ia32_pat);
353 		vmcs_write(GUEST_RIP, guest_rip + 3);
354 		return VMX_TEST_RESUME;
355 	default:
356 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
357 		break;
358 	}
359 	return VMX_TEST_VMEXIT;
360 }
361 
362 static int test_ctrl_efer_init(struct vmcs *vmcs)
363 {
364 	u64 ctrl_ent;
365 	u64 ctrl_exi;
366 
367 	msr_bmp_init();
368 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
369 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
370 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
371 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
372 	ia32_efer = rdmsr(MSR_EFER);
373 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
374 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
375 	return VMX_TEST_START;
376 }
377 
378 static void test_ctrl_efer_main(void)
379 {
380 	u64 guest_ia32_efer;
381 
382 	guest_ia32_efer = rdmsr(MSR_EFER);
383 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
384 		printf("\tENT_LOAD_EFER is not supported.\n");
385 	else {
386 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
387 			report_fail("Entry load EFER");
388 			return;
389 		}
390 	}
391 	wrmsr(MSR_EFER, ia32_efer);
392 	vmcall();
393 	guest_ia32_efer = rdmsr(MSR_EFER);
394 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
395 		report(guest_ia32_efer == ia32_efer, "Entry load EFER");
396 }
397 
398 static int test_ctrl_efer_exit_handler(union exit_reason exit_reason)
399 {
400 	u64 guest_rip;
401 	u64 guest_efer;
402 
403 	guest_rip = vmcs_read(GUEST_RIP);
404 	switch (exit_reason.basic) {
405 	case VMX_VMCALL:
406 		guest_efer = vmcs_read(GUEST_EFER);
407 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
408 			printf("\tEXI_SAVE_EFER is not supported\n");
409 			vmcs_write(GUEST_EFER, ia32_efer);
410 		} else {
411 			report(guest_efer == ia32_efer, "Exit save EFER");
412 		}
413 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
414 			printf("\tEXI_LOAD_EFER is not supported\n");
415 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
416 		} else {
417 			report(rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX),
418 			       "Exit load EFER");
419 		}
420 		vmcs_write(GUEST_PAT, ia32_efer);
421 		vmcs_write(GUEST_RIP, guest_rip + 3);
422 		return VMX_TEST_RESUME;
423 	default:
424 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
425 		break;
426 	}
427 	return VMX_TEST_VMEXIT;
428 }
429 
430 u32 guest_cr0, guest_cr4;
431 
432 static void cr_shadowing_main(void)
433 {
434 	u32 cr0, cr4, tmp;
435 
436 	// Test read through
437 	vmx_set_test_stage(0);
438 	guest_cr0 = read_cr0();
439 	if (vmx_get_test_stage() == 1)
440 		report_fail("Read through CR0");
441 	else
442 		vmcall();
443 	vmx_set_test_stage(1);
444 	guest_cr4 = read_cr4();
445 	if (vmx_get_test_stage() == 2)
446 		report_fail("Read through CR4");
447 	else
448 		vmcall();
449 	// Test write through
450 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
451 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
452 	vmx_set_test_stage(2);
453 	write_cr0(guest_cr0);
454 	if (vmx_get_test_stage() == 3)
455 		report_fail("Write throuth CR0");
456 	else
457 		vmcall();
458 	vmx_set_test_stage(3);
459 	write_cr4(guest_cr4);
460 	if (vmx_get_test_stage() == 4)
461 		report_fail("Write through CR4");
462 	else
463 		vmcall();
464 	// Test read shadow
465 	vmx_set_test_stage(4);
466 	vmcall();
467 	cr0 = read_cr0();
468 	if (vmx_get_test_stage() != 5)
469 		report(cr0 == guest_cr0, "Read shadowing CR0");
470 	vmx_set_test_stage(5);
471 	cr4 = read_cr4();
472 	if (vmx_get_test_stage() != 6)
473 		report(cr4 == guest_cr4, "Read shadowing CR4");
474 	// Test write shadow (same value with shadow)
475 	vmx_set_test_stage(6);
476 	write_cr0(guest_cr0);
477 	if (vmx_get_test_stage() == 7)
478 		report_fail("Write shadowing CR0 (same value with shadow)");
479 	else
480 		vmcall();
481 	vmx_set_test_stage(7);
482 	write_cr4(guest_cr4);
483 	if (vmx_get_test_stage() == 8)
484 		report_fail("Write shadowing CR4 (same value with shadow)");
485 	else
486 		vmcall();
487 	// Test write shadow (different value)
488 	vmx_set_test_stage(8);
489 	tmp = guest_cr0 ^ X86_CR0_TS;
490 	asm volatile("mov %0, %%rsi\n\t"
491 		"mov %%rsi, %%cr0\n\t"
492 		::"m"(tmp)
493 		:"rsi", "memory", "cc");
494 	report(vmx_get_test_stage() == 9,
495 	       "Write shadowing different X86_CR0_TS");
496 	vmx_set_test_stage(9);
497 	tmp = guest_cr0 ^ X86_CR0_MP;
498 	asm volatile("mov %0, %%rsi\n\t"
499 		"mov %%rsi, %%cr0\n\t"
500 		::"m"(tmp)
501 		:"rsi", "memory", "cc");
502 	report(vmx_get_test_stage() == 10,
503 	       "Write shadowing different X86_CR0_MP");
504 	vmx_set_test_stage(10);
505 	tmp = guest_cr4 ^ X86_CR4_TSD;
506 	asm volatile("mov %0, %%rsi\n\t"
507 		"mov %%rsi, %%cr4\n\t"
508 		::"m"(tmp)
509 		:"rsi", "memory", "cc");
510 	report(vmx_get_test_stage() == 11,
511 	       "Write shadowing different X86_CR4_TSD");
512 	vmx_set_test_stage(11);
513 	tmp = guest_cr4 ^ X86_CR4_DE;
514 	asm volatile("mov %0, %%rsi\n\t"
515 		"mov %%rsi, %%cr4\n\t"
516 		::"m"(tmp)
517 		:"rsi", "memory", "cc");
518 	report(vmx_get_test_stage() == 12,
519 	       "Write shadowing different X86_CR4_DE");
520 }
521 
522 static int cr_shadowing_exit_handler(union exit_reason exit_reason)
523 {
524 	u64 guest_rip;
525 	u32 insn_len;
526 	u32 exit_qual;
527 
528 	guest_rip = vmcs_read(GUEST_RIP);
529 	insn_len = vmcs_read(EXI_INST_LEN);
530 	exit_qual = vmcs_read(EXI_QUALIFICATION);
531 	switch (exit_reason.basic) {
532 	case VMX_VMCALL:
533 		switch (vmx_get_test_stage()) {
534 		case 0:
535 			report(guest_cr0 == vmcs_read(GUEST_CR0),
536 			       "Read through CR0");
537 			break;
538 		case 1:
539 			report(guest_cr4 == vmcs_read(GUEST_CR4),
540 			       "Read through CR4");
541 			break;
542 		case 2:
543 			report(guest_cr0 == vmcs_read(GUEST_CR0),
544 			       "Write through CR0");
545 			break;
546 		case 3:
547 			report(guest_cr4 == vmcs_read(GUEST_CR4),
548 			       "Write through CR4");
549 			break;
550 		case 4:
551 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
552 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
553 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
554 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
555 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
556 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
557 			break;
558 		case 6:
559 			report(guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)),
560 			       "Write shadowing CR0 (same value)");
561 			break;
562 		case 7:
563 			report(guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)),
564 			       "Write shadowing CR4 (same value)");
565 			break;
566 		default:
567 			// Should not reach here
568 			report_fail("unexpected stage, %d",
569 				    vmx_get_test_stage());
570 			print_vmexit_info(exit_reason);
571 			return VMX_TEST_VMEXIT;
572 		}
573 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
574 		return VMX_TEST_RESUME;
575 	case VMX_CR:
576 		switch (vmx_get_test_stage()) {
577 		case 4:
578 			report_fail("Read shadowing CR0");
579 			vmx_inc_test_stage();
580 			break;
581 		case 5:
582 			report_fail("Read shadowing CR4");
583 			vmx_inc_test_stage();
584 			break;
585 		case 6:
586 			report_fail("Write shadowing CR0 (same value)");
587 			vmx_inc_test_stage();
588 			break;
589 		case 7:
590 			report_fail("Write shadowing CR4 (same value)");
591 			vmx_inc_test_stage();
592 			break;
593 		case 8:
594 		case 9:
595 			// 0x600 encodes "mov %esi, %cr0"
596 			if (exit_qual == 0x600)
597 				vmx_inc_test_stage();
598 			break;
599 		case 10:
600 		case 11:
601 			// 0x604 encodes "mov %esi, %cr4"
602 			if (exit_qual == 0x604)
603 				vmx_inc_test_stage();
604 			break;
605 		default:
606 			// Should not reach here
607 			report_fail("unexpected stage, %d",
608 				    vmx_get_test_stage());
609 			print_vmexit_info(exit_reason);
610 			return VMX_TEST_VMEXIT;
611 		}
612 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
613 		return VMX_TEST_RESUME;
614 	default:
615 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
616 		print_vmexit_info(exit_reason);
617 	}
618 	return VMX_TEST_VMEXIT;
619 }
620 
621 static int iobmp_init(struct vmcs *vmcs)
622 {
623 	u32 ctrl_cpu0;
624 
625 	io_bitmap_a = alloc_page();
626 	io_bitmap_b = alloc_page();
627 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
628 	ctrl_cpu0 |= CPU_IO_BITMAP;
629 	ctrl_cpu0 &= (~CPU_IO);
630 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
631 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
632 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
633 	return VMX_TEST_START;
634 }
635 
636 static void iobmp_main(void)
637 {
638 	// stage 0, test IO pass
639 	vmx_set_test_stage(0);
640 	inb(0x5000);
641 	outb(0x0, 0x5000);
642 	report(vmx_get_test_stage() == 0, "I/O bitmap - I/O pass");
643 	// test IO width, in/out
644 	((u8 *)io_bitmap_a)[0] = 0xFF;
645 	vmx_set_test_stage(2);
646 	inb(0x0);
647 	report(vmx_get_test_stage() == 3, "I/O bitmap - trap in");
648 	vmx_set_test_stage(3);
649 	outw(0x0, 0x0);
650 	report(vmx_get_test_stage() == 4, "I/O bitmap - trap out");
651 	vmx_set_test_stage(4);
652 	inl(0x0);
653 	report(vmx_get_test_stage() == 5, "I/O bitmap - I/O width, long");
654 	// test low/high IO port
655 	vmx_set_test_stage(5);
656 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
657 	inb(0x5000);
658 	report(vmx_get_test_stage() == 6, "I/O bitmap - I/O port, low part");
659 	vmx_set_test_stage(6);
660 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
661 	inb(0x9000);
662 	report(vmx_get_test_stage() == 7, "I/O bitmap - I/O port, high part");
663 	// test partial pass
664 	vmx_set_test_stage(7);
665 	inl(0x4FFF);
666 	report(vmx_get_test_stage() == 8, "I/O bitmap - partial pass");
667 	// test overrun
668 	vmx_set_test_stage(8);
669 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
670 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
671 	inl(0xFFFF);
672 	report(vmx_get_test_stage() == 9, "I/O bitmap - overrun");
673 	vmx_set_test_stage(9);
674 	vmcall();
675 	outb(0x0, 0x0);
676 	report(vmx_get_test_stage() == 9,
677 	       "I/O bitmap - ignore unconditional exiting");
678 	vmx_set_test_stage(10);
679 	vmcall();
680 	outb(0x0, 0x0);
681 	report(vmx_get_test_stage() == 11,
682 	       "I/O bitmap - unconditional exiting");
683 }
684 
685 static int iobmp_exit_handler(union exit_reason exit_reason)
686 {
687 	u64 guest_rip;
688 	ulong exit_qual;
689 	u32 insn_len, ctrl_cpu0;
690 
691 	guest_rip = vmcs_read(GUEST_RIP);
692 	exit_qual = vmcs_read(EXI_QUALIFICATION);
693 	insn_len = vmcs_read(EXI_INST_LEN);
694 	switch (exit_reason.basic) {
695 	case VMX_IO:
696 		switch (vmx_get_test_stage()) {
697 		case 0:
698 		case 1:
699 			vmx_inc_test_stage();
700 			break;
701 		case 2:
702 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE,
703 			       "I/O bitmap - I/O width, byte");
704 			report(exit_qual & VMX_IO_IN,
705 			       "I/O bitmap - I/O direction, in");
706 			vmx_inc_test_stage();
707 			break;
708 		case 3:
709 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD,
710 			       "I/O bitmap - I/O width, word");
711 			report(!(exit_qual & VMX_IO_IN),
712 			       "I/O bitmap - I/O direction, out");
713 			vmx_inc_test_stage();
714 			break;
715 		case 4:
716 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG,
717 			       "I/O bitmap - I/O width, long");
718 			vmx_inc_test_stage();
719 			break;
720 		case 5:
721 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
722 				vmx_inc_test_stage();
723 			break;
724 		case 6:
725 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
726 				vmx_inc_test_stage();
727 			break;
728 		case 7:
729 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
730 				vmx_inc_test_stage();
731 			break;
732 		case 8:
733 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
734 				vmx_inc_test_stage();
735 			break;
736 		case 9:
737 		case 10:
738 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
739 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
740 			vmx_inc_test_stage();
741 			break;
742 		default:
743 			// Should not reach here
744 			report_fail("unexpected stage, %d",
745 				    vmx_get_test_stage());
746 			print_vmexit_info(exit_reason);
747 			return VMX_TEST_VMEXIT;
748 		}
749 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
750 		return VMX_TEST_RESUME;
751 	case VMX_VMCALL:
752 		switch (vmx_get_test_stage()) {
753 		case 9:
754 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
755 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
756 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
757 			break;
758 		case 10:
759 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
760 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
761 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
762 			break;
763 		default:
764 			// Should not reach here
765 			report_fail("unexpected stage, %d",
766 				    vmx_get_test_stage());
767 			print_vmexit_info(exit_reason);
768 			return VMX_TEST_VMEXIT;
769 		}
770 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
771 		return VMX_TEST_RESUME;
772 	default:
773 		printf("guest_rip = %#lx\n", guest_rip);
774 		printf("\tERROR : Unknown exit reason, 0x%x\n", exit_reason.full);
775 		break;
776 	}
777 	return VMX_TEST_VMEXIT;
778 }
779 
780 #define INSN_CPU0		0
781 #define INSN_CPU1		1
782 #define INSN_ALWAYS_TRAP	2
783 
784 #define FIELD_EXIT_QUAL		(1 << 0)
785 #define FIELD_INSN_INFO		(1 << 1)
786 
787 asm(
788 	"insn_hlt: hlt;ret\n\t"
789 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
790 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
791 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
792 	"insn_rdtsc: rdtsc;ret\n\t"
793 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
794 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
795 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
796 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
797 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
798 	"insn_pause: pause;ret\n\t"
799 	"insn_wbinvd: wbinvd;ret\n\t"
800 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
801 	"insn_invd: invd;ret\n\t"
802 	"insn_sgdt: sgdt gdt_descr;ret\n\t"
803 	"insn_lgdt: lgdt gdt_descr;ret\n\t"
804 	"insn_sidt: sidt idt_descr;ret\n\t"
805 	"insn_lidt: lidt idt_descr;ret\n\t"
806 	"insn_sldt: sldt %ax;ret\n\t"
807 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
808 	"insn_str: str %ax;ret\n\t"
809 	"insn_rdrand: rdrand %rax;ret\n\t"
810 	"insn_rdseed: rdseed %rax;ret\n\t"
811 );
812 extern void insn_hlt(void);
813 extern void insn_invlpg(void);
814 extern void insn_mwait(void);
815 extern void insn_rdpmc(void);
816 extern void insn_rdtsc(void);
817 extern void insn_cr3_load(void);
818 extern void insn_cr3_store(void);
819 extern void insn_cr8_load(void);
820 extern void insn_cr8_store(void);
821 extern void insn_monitor(void);
822 extern void insn_pause(void);
823 extern void insn_wbinvd(void);
824 extern void insn_sgdt(void);
825 extern void insn_lgdt(void);
826 extern void insn_sidt(void);
827 extern void insn_lidt(void);
828 extern void insn_sldt(void);
829 extern void insn_lldt(void);
830 extern void insn_str(void);
831 extern void insn_cpuid(void);
832 extern void insn_invd(void);
833 extern void insn_rdrand(void);
834 extern void insn_rdseed(void);
835 
836 u32 cur_insn;
837 u64 cr3;
838 
839 #define X86_FEATURE_MONITOR	(1 << 3)
840 
841 typedef bool (*supported_fn)(void);
842 
843 static bool monitor_supported(void)
844 {
845 	return this_cpu_has(X86_FEATURE_MWAIT);
846 }
847 
848 struct insn_table {
849 	const char *name;
850 	u32 flag;
851 	void (*insn_func)(void);
852 	u32 type;
853 	u32 reason;
854 	ulong exit_qual;
855 	u32 insn_info;
856 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
857 	// which field need to be tested, reason is always tested
858 	u32 test_field;
859 	const supported_fn supported_fn;
860 	u8 disabled;
861 };
862 
863 /*
864  * Add more test cases of instruction intercept here. Elements in this
865  * table is:
866  *	name/control flag/insn function/type/exit reason/exit qulification/
867  *	instruction info/field to test
868  * The last field defines which fields (exit_qual and insn_info) need to be
869  * tested in exit handler. If set to 0, only "reason" is checked.
870  */
871 static struct insn_table insn_table[] = {
872 	// Flags for Primary Processor-Based VM-Execution Controls
873 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
874 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
875 		0x12345678, 0, FIELD_EXIT_QUAL},
876 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0, &monitor_supported},
877 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0},
878 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
879 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
880 		FIELD_EXIT_QUAL},
881 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
882 		FIELD_EXIT_QUAL},
883 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
884 		FIELD_EXIT_QUAL},
885 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
886 		FIELD_EXIT_QUAL},
887 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0, &monitor_supported},
888 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
889 	// Flags for Secondary Processor-Based VM-Execution Controls
890 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
891 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
892 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
893 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
894 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
895 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
896 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
897 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
898 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
899 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
900 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
901 	// Instructions always trap
902 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
903 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
904 	// Instructions never trap
905 	{NULL},
906 };
907 
908 static int insn_intercept_init(struct vmcs *vmcs)
909 {
910 	u32 ctrl_cpu, cur_insn;
911 
912 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
913 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
914 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
915 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
916 	cr3 = read_cr3();
917 
918 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
919 		if (insn_table[cur_insn].supported_fn == NULL)
920 			continue;
921 		insn_table[cur_insn].disabled = !insn_table[cur_insn].supported_fn();
922 	}
923 	return VMX_TEST_START;
924 }
925 
926 static void insn_intercept_main(void)
927 {
928 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
929 		vmx_set_test_stage(cur_insn * 2);
930 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
931 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
932 		    (insn_table[cur_insn].type == INSN_CPU1 &&
933 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
934 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
935 			       insn_table[cur_insn].type - INSN_CPU0,
936 			       insn_table[cur_insn].name);
937 			continue;
938 		}
939 
940 		if (insn_table[cur_insn].disabled) {
941 			printf("\tFeature required for %s is not supported.\n",
942 			       insn_table[cur_insn].name);
943 			continue;
944 		}
945 
946 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
947 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
948 		    (insn_table[cur_insn].type == INSN_CPU1 &&
949 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
950 			/* skip hlt, it stalls the guest and is tested below */
951 			if (insn_table[cur_insn].insn_func != insn_hlt)
952 				insn_table[cur_insn].insn_func();
953 			report(vmx_get_test_stage() == cur_insn * 2,
954 					"execute %s",
955 					insn_table[cur_insn].name);
956 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
957 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
958 			       insn_table[cur_insn].type - INSN_CPU0,
959 			       insn_table[cur_insn].name);
960 
961 		vmcall();
962 
963 		insn_table[cur_insn].insn_func();
964 		report(vmx_get_test_stage() == cur_insn * 2 + 1,
965 				"intercept %s",
966 				insn_table[cur_insn].name);
967 
968 		vmx_set_test_stage(cur_insn * 2 + 1);
969 		vmcall();
970 	}
971 }
972 
973 static int insn_intercept_exit_handler(union exit_reason exit_reason)
974 {
975 	u64 guest_rip;
976 	ulong exit_qual;
977 	u32 insn_len;
978 	u32 insn_info;
979 	bool pass;
980 
981 	guest_rip = vmcs_read(GUEST_RIP);
982 	exit_qual = vmcs_read(EXI_QUALIFICATION);
983 	insn_len = vmcs_read(EXI_INST_LEN);
984 	insn_info = vmcs_read(EXI_INST_INFO);
985 
986 	if (exit_reason.basic == VMX_VMCALL) {
987 		u32 val = 0;
988 
989 		if (insn_table[cur_insn].type == INSN_CPU0)
990 			val = vmcs_read(CPU_EXEC_CTRL0);
991 		else if (insn_table[cur_insn].type == INSN_CPU1)
992 			val = vmcs_read(CPU_EXEC_CTRL1);
993 
994 		if (vmx_get_test_stage() & 1)
995 			val &= ~insn_table[cur_insn].flag;
996 		else
997 			val |= insn_table[cur_insn].flag;
998 
999 		if (insn_table[cur_insn].type == INSN_CPU0)
1000 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
1001 		else if (insn_table[cur_insn].type == INSN_CPU1)
1002 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
1003 	} else {
1004 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
1005 			insn_table[cur_insn].reason == exit_reason.full;
1006 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
1007 		    insn_table[cur_insn].exit_qual != exit_qual)
1008 			pass = false;
1009 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
1010 		    insn_table[cur_insn].insn_info != insn_info)
1011 			pass = false;
1012 		if (pass)
1013 			vmx_inc_test_stage();
1014 	}
1015 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1016 	return VMX_TEST_RESUME;
1017 }
1018 
1019 /**
1020  * __setup_ept - Setup the VMCS fields to enable Extended Page Tables (EPT)
1021  * @hpa:	Host physical address of the top-level, a.k.a. root, EPT table
1022  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1023  *
1024  * Returns 0 on success, 1 on failure.
1025  *
1026  * Note that @hpa doesn't need to point at actual memory if VM-Launch is
1027  * expected to fail, e.g. setup_dummy_ept() arbitrarily passes '0' to satisfy
1028  * the various EPTP consistency checks, but doesn't ensure backing for HPA '0'.
1029  */
1030 static int __setup_ept(u64 hpa, bool enable_ad)
1031 {
1032 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1033 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1034 		printf("\tEPT is not supported\n");
1035 		return 1;
1036 	}
1037 	if (!(ept_vpid.val & EPT_CAP_WB)) {
1038 		printf("\tWB memtype for EPT walks not supported\n");
1039 		return 1;
1040 	}
1041 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
1042 		printf("\tPWL4 is not supported\n");
1043 		return 1;
1044 	}
1045 
1046 	eptp = EPT_MEM_TYPE_WB;
1047 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1048 	eptp |= hpa;
1049 	if (enable_ad)
1050 		eptp |= EPTP_AD_FLAG;
1051 
1052 	vmcs_write(EPTP, eptp);
1053 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY);
1054 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT);
1055 
1056 	return 0;
1057 }
1058 
1059 /**
1060  * setup_ept - Enable Extended Page Tables (EPT) and setup an identity map
1061  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1062  *
1063  * Returns 0 on success, 1 on failure.
1064  *
1065  * This is the "real" function for setting up EPT tables, i.e. use this for
1066  * tests that need to run code in the guest with EPT enabled.
1067  */
1068 static int setup_ept(bool enable_ad)
1069 {
1070 	unsigned long end_of_memory;
1071 
1072 	pml4 = alloc_page();
1073 
1074 	if (__setup_ept(virt_to_phys(pml4), enable_ad))
1075 		return 1;
1076 
1077 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1078 	if (end_of_memory < (1ul << 32))
1079 		end_of_memory = (1ul << 32);
1080 	/* Cannot use large EPT pages if we need to track EPT
1081 	 * accessed/dirty bits at 4K granularity.
1082 	 */
1083 	setup_ept_range(pml4, 0, end_of_memory, 0,
1084 			!enable_ad && ept_2m_supported(),
1085 			EPT_WA | EPT_RA | EPT_EA);
1086 	return 0;
1087 }
1088 
1089 /**
1090  * setup_dummy_ept - Enable Extended Page Tables (EPT) with a dummy root HPA
1091  *
1092  * Setup EPT using a semi-arbitrary dummy root HPA.  This function is intended
1093  * for use by tests that need EPT enabled to verify dependent VMCS controls
1094  * but never expect to fully enter the guest, i.e. don't need setup the actual
1095  * EPT tables.
1096  */
1097 static void setup_dummy_ept(void)
1098 {
1099 	if (__setup_ept(0, false))
1100 		report_abort("EPT setup unexpectedly failed");
1101 }
1102 
1103 static int enable_unrestricted_guest(bool need_valid_ept)
1104 {
1105 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1106 	    !(ctrl_cpu_rev[1].clr & CPU_URG) ||
1107 	    !(ctrl_cpu_rev[1].clr & CPU_EPT))
1108 		return 1;
1109 
1110 	if (need_valid_ept)
1111 		setup_ept(false);
1112 	else
1113 		setup_dummy_ept();
1114 
1115 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
1116 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
1117 
1118 	return 0;
1119 }
1120 
1121 static void ept_enable_ad_bits(void)
1122 {
1123 	eptp |= EPTP_AD_FLAG;
1124 	vmcs_write(EPTP, eptp);
1125 }
1126 
1127 static void ept_disable_ad_bits(void)
1128 {
1129 	eptp &= ~EPTP_AD_FLAG;
1130 	vmcs_write(EPTP, eptp);
1131 }
1132 
1133 static int ept_ad_enabled(void)
1134 {
1135 	return eptp & EPTP_AD_FLAG;
1136 }
1137 
1138 static void ept_enable_ad_bits_or_skip_test(void)
1139 {
1140 	if (!ept_ad_bits_supported())
1141 		test_skip("EPT AD bits not supported.");
1142 	ept_enable_ad_bits();
1143 }
1144 
1145 static int apic_version;
1146 
1147 static int ept_init_common(bool have_ad)
1148 {
1149 	int ret;
1150 	struct pci_dev pcidev;
1151 
1152 	/* INVEPT is required by the EPT violation handler. */
1153 	if (!is_invept_type_supported(INVEPT_SINGLE))
1154 		return VMX_TEST_EXIT;
1155 
1156 	if (setup_ept(have_ad))
1157 		return VMX_TEST_EXIT;
1158 
1159 	data_page1 = alloc_page();
1160 	data_page2 = alloc_page();
1161 	*((u32 *)data_page1) = MAGIC_VAL_1;
1162 	*((u32 *)data_page2) = MAGIC_VAL_2;
1163 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1164 			EPT_RA | EPT_WA | EPT_EA);
1165 
1166 	apic_version = apic_read(APIC_LVR);
1167 
1168 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1169 	if (ret != PCIDEVADDR_INVALID) {
1170 		pci_dev_init(&pcidev, ret);
1171 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1172 	}
1173 
1174 	return VMX_TEST_START;
1175 }
1176 
1177 static int ept_init(struct vmcs *vmcs)
1178 {
1179 	return ept_init_common(false);
1180 }
1181 
1182 static void ept_common(void)
1183 {
1184 	vmx_set_test_stage(0);
1185 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1186 			*((u32 *)data_page1) != MAGIC_VAL_1)
1187 		report_fail("EPT basic framework - read");
1188 	else {
1189 		*((u32 *)data_page2) = MAGIC_VAL_3;
1190 		vmcall();
1191 		if (vmx_get_test_stage() == 1) {
1192 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1193 					*((u32 *)data_page2) == MAGIC_VAL_2)
1194 				report_pass("EPT basic framework");
1195 			else
1196 				report_pass("EPT basic framework - remap");
1197 		}
1198 	}
1199 	// Test EPT Misconfigurations
1200 	vmx_set_test_stage(1);
1201 	vmcall();
1202 	*((u32 *)data_page1) = MAGIC_VAL_1;
1203 	if (vmx_get_test_stage() != 2) {
1204 		report_fail("EPT misconfigurations");
1205 		goto t1;
1206 	}
1207 	vmx_set_test_stage(2);
1208 	vmcall();
1209 	*((u32 *)data_page1) = MAGIC_VAL_1;
1210 	report(vmx_get_test_stage() == 3, "EPT misconfigurations");
1211 t1:
1212 	// Test EPT violation
1213 	vmx_set_test_stage(3);
1214 	vmcall();
1215 	*((u32 *)data_page1) = MAGIC_VAL_1;
1216 	report(vmx_get_test_stage() == 4, "EPT violation - page permission");
1217 	// Violation caused by EPT paging structure
1218 	vmx_set_test_stage(4);
1219 	vmcall();
1220 	*((u32 *)data_page1) = MAGIC_VAL_2;
1221 	report(vmx_get_test_stage() == 5, "EPT violation - paging structure");
1222 
1223 	// MMIO Read/Write
1224 	vmx_set_test_stage(5);
1225 	vmcall();
1226 
1227 	*(u32 volatile *)pci_physaddr;
1228 	report(vmx_get_test_stage() == 6, "MMIO EPT violation - read");
1229 
1230 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1231 	report(vmx_get_test_stage() == 7, "MMIO EPT violation - write");
1232 }
1233 
1234 static void ept_main(void)
1235 {
1236 	ept_common();
1237 
1238 	// Test EPT access to L1 MMIO
1239 	vmx_set_test_stage(7);
1240 	report(*((u32 *)0xfee00030UL) == apic_version, "EPT - MMIO access");
1241 
1242 	// Test invalid operand for INVEPT
1243 	vmcall();
1244 	report(vmx_get_test_stage() == 8, "EPT - unsupported INVEPT");
1245 }
1246 
1247 static bool invept_test(int type, u64 eptp)
1248 {
1249 	bool ret, supported;
1250 
1251 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1252 	ret = __invept(type, eptp);
1253 
1254 	if (ret == !supported)
1255 		return false;
1256 
1257 	if (!supported)
1258 		printf("WARNING: unsupported invept passed!\n");
1259 	else
1260 		printf("WARNING: invept failed!\n");
1261 
1262 	return true;
1263 }
1264 
1265 static int pml_exit_handler(union exit_reason exit_reason)
1266 {
1267 	u16 index, count;
1268 	u64 *pmlbuf = pml_log;
1269 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1270 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1271 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1272 
1273 	switch (exit_reason.basic) {
1274 	case VMX_VMCALL:
1275 		switch (vmx_get_test_stage()) {
1276 		case 0:
1277 			index = vmcs_read(GUEST_PML_INDEX);
1278 			for (count = index + 1; count < PML_INDEX; count++) {
1279 				if (pmlbuf[count] == (u64)data_page2) {
1280 					vmx_inc_test_stage();
1281 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1282 					break;
1283 				}
1284 			}
1285 			break;
1286 		case 1:
1287 			index = vmcs_read(GUEST_PML_INDEX);
1288 			/* Keep clearing the dirty bit till a overflow */
1289 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1290 			break;
1291 		default:
1292 			report_fail("unexpected stage, %d.",
1293 			       vmx_get_test_stage());
1294 			print_vmexit_info(exit_reason);
1295 			return VMX_TEST_VMEXIT;
1296 		}
1297 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1298 		return VMX_TEST_RESUME;
1299 	case VMX_PML_FULL:
1300 		vmx_inc_test_stage();
1301 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1302 		return VMX_TEST_RESUME;
1303 	default:
1304 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1305 		print_vmexit_info(exit_reason);
1306 	}
1307 	return VMX_TEST_VMEXIT;
1308 }
1309 
1310 static int ept_exit_handler_common(union exit_reason exit_reason, bool have_ad)
1311 {
1312 	u64 guest_rip;
1313 	u64 guest_cr3;
1314 	u32 insn_len;
1315 	u32 exit_qual;
1316 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte,
1317 			     guest_pte_addr;
1318 
1319 	guest_rip = vmcs_read(GUEST_RIP);
1320 	guest_cr3 = vmcs_read(GUEST_CR3);
1321 	insn_len = vmcs_read(EXI_INST_LEN);
1322 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1323 	pteval_t *ptep;
1324 	switch (exit_reason.basic) {
1325 	case VMX_VMCALL:
1326 		switch (vmx_get_test_stage()) {
1327 		case 0:
1328 			check_ept_ad(pml4, guest_cr3,
1329 				     (unsigned long)data_page1,
1330 				     have_ad ? EPT_ACCESS_FLAG : 0,
1331 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1332 			check_ept_ad(pml4, guest_cr3,
1333 				     (unsigned long)data_page2,
1334 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1335 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1336 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1337 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1338 			if (have_ad)
1339 				invept(INVEPT_SINGLE, eptp);
1340 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1341 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1342 				vmx_inc_test_stage();
1343 				install_ept(pml4, (unsigned long)data_page2,
1344 						(unsigned long)data_page2,
1345 						EPT_RA | EPT_WA | EPT_EA);
1346 			} else
1347 				report_fail("EPT basic framework - write");
1348 			break;
1349 		case 1:
1350 			install_ept(pml4, (unsigned long)data_page1,
1351  				(unsigned long)data_page1, EPT_WA);
1352 			invept(INVEPT_SINGLE, eptp);
1353 			break;
1354 		case 2:
1355 			install_ept(pml4, (unsigned long)data_page1,
1356  				(unsigned long)data_page1,
1357  				EPT_RA | EPT_WA | EPT_EA |
1358  				(2 << EPT_MEM_TYPE_SHIFT));
1359 			invept(INVEPT_SINGLE, eptp);
1360 			break;
1361 		case 3:
1362 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1363 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1364 						1, &data_page1_pte));
1365 			set_ept_pte(pml4, (unsigned long)data_page1,
1366 				1, data_page1_pte & ~EPT_PRESENT);
1367 			invept(INVEPT_SINGLE, eptp);
1368 			break;
1369 		case 4:
1370 			ptep = get_pte_level((pgd_t *)guest_cr3, data_page1, /*level=*/2);
1371 			guest_pte_addr = virt_to_phys(ptep) & PAGE_MASK;
1372 
1373 			TEST_ASSERT(get_ept_pte(pml4, guest_pte_addr, 2, &data_page1_pte_pte));
1374 			set_ept_pte(pml4, guest_pte_addr, 2,
1375 				data_page1_pte_pte & ~EPT_PRESENT);
1376 			invept(INVEPT_SINGLE, eptp);
1377 			break;
1378 		case 5:
1379 			install_ept(pml4, (unsigned long)pci_physaddr,
1380 				(unsigned long)pci_physaddr, 0);
1381 			invept(INVEPT_SINGLE, eptp);
1382 			break;
1383 		case 7:
1384 			if (!invept_test(0, eptp))
1385 				vmx_inc_test_stage();
1386 			break;
1387 		// Should not reach here
1388 		default:
1389 			report_fail("ERROR - unexpected stage, %d.",
1390 			       vmx_get_test_stage());
1391 			print_vmexit_info(exit_reason);
1392 			return VMX_TEST_VMEXIT;
1393 		}
1394 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1395 		return VMX_TEST_RESUME;
1396 	case VMX_EPT_MISCONFIG:
1397 		switch (vmx_get_test_stage()) {
1398 		case 1:
1399 		case 2:
1400 			vmx_inc_test_stage();
1401 			install_ept(pml4, (unsigned long)data_page1,
1402  				(unsigned long)data_page1,
1403  				EPT_RA | EPT_WA | EPT_EA);
1404 			invept(INVEPT_SINGLE, eptp);
1405 			break;
1406 		// Should not reach here
1407 		default:
1408 			report_fail("ERROR - unexpected stage, %d.",
1409 			       vmx_get_test_stage());
1410 			print_vmexit_info(exit_reason);
1411 			return VMX_TEST_VMEXIT;
1412 		}
1413 		return VMX_TEST_RESUME;
1414 	case VMX_EPT_VIOLATION:
1415 		/*
1416 		 * Exit-qualifications are masked not to account for advanced
1417 		 * VM-exit information. Once KVM supports this feature, this
1418 		 * masking should be removed.
1419 		 */
1420 		exit_qual &= ~EPT_VLT_GUEST_MASK;
1421 
1422 		switch(vmx_get_test_stage()) {
1423 		case 3:
1424 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1425 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1426 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1427 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1428 					EPT_VLT_PADDR))
1429 				vmx_inc_test_stage();
1430 			set_ept_pte(pml4, (unsigned long)data_page1,
1431 				1, data_page1_pte | (EPT_PRESENT));
1432 			invept(INVEPT_SINGLE, eptp);
1433 			break;
1434 		case 4:
1435 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1436 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1437 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1438 			if (exit_qual == (EPT_VLT_RD |
1439 					  (have_ad ? EPT_VLT_WR : 0) |
1440 					  EPT_VLT_LADDR_VLD))
1441 				vmx_inc_test_stage();
1442 			set_ept_pte(pml4, guest_pte_addr, 2,
1443 				data_page1_pte_pte | (EPT_PRESENT));
1444 			invept(INVEPT_SINGLE, eptp);
1445 			break;
1446 		case 5:
1447 			if (exit_qual & EPT_VLT_RD)
1448 				vmx_inc_test_stage();
1449 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1450 						1, &memaddr_pte));
1451 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1452 			invept(INVEPT_SINGLE, eptp);
1453 			break;
1454 		case 6:
1455 			if (exit_qual & EPT_VLT_WR)
1456 				vmx_inc_test_stage();
1457 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1458 						1, &memaddr_pte));
1459 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1460 			invept(INVEPT_SINGLE, eptp);
1461 			break;
1462 		default:
1463 			// Should not reach here
1464 			report_fail("ERROR : unexpected stage, %d",
1465 			       vmx_get_test_stage());
1466 			print_vmexit_info(exit_reason);
1467 			return VMX_TEST_VMEXIT;
1468 		}
1469 		return VMX_TEST_RESUME;
1470 	default:
1471 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1472 		print_vmexit_info(exit_reason);
1473 	}
1474 	return VMX_TEST_VMEXIT;
1475 }
1476 
1477 static int ept_exit_handler(union exit_reason exit_reason)
1478 {
1479 	return ept_exit_handler_common(exit_reason, false);
1480 }
1481 
1482 static int eptad_init(struct vmcs *vmcs)
1483 {
1484 	int r = ept_init_common(true);
1485 
1486 	if (r == VMX_TEST_EXIT)
1487 		return r;
1488 
1489 	if (!ept_ad_bits_supported()) {
1490 		printf("\tEPT A/D bits are not supported");
1491 		return VMX_TEST_EXIT;
1492 	}
1493 
1494 	return r;
1495 }
1496 
1497 static int pml_init(struct vmcs *vmcs)
1498 {
1499 	u32 ctrl_cpu;
1500 	int r = eptad_init(vmcs);
1501 
1502 	if (r == VMX_TEST_EXIT)
1503 		return r;
1504 
1505 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1506 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1507 		printf("\tPML is not supported");
1508 		return VMX_TEST_EXIT;
1509 	}
1510 
1511 	pml_log = alloc_page();
1512 	vmcs_write(PMLADDR, (u64)pml_log);
1513 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1514 
1515 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1516 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1517 
1518 	return VMX_TEST_START;
1519 }
1520 
1521 static void pml_main(void)
1522 {
1523 	int count = 0;
1524 
1525 	vmx_set_test_stage(0);
1526 	*((u32 *)data_page2) = 0x1;
1527 	vmcall();
1528 	report(vmx_get_test_stage() == 1, "PML - Dirty GPA Logging");
1529 
1530 	while (vmx_get_test_stage() == 1) {
1531 		vmcall();
1532 		*((u32 *)data_page2) = 0x1;
1533 		if (count++ > PML_INDEX)
1534 			break;
1535 	}
1536 	report(vmx_get_test_stage() == 2, "PML Full Event");
1537 }
1538 
1539 static void eptad_main(void)
1540 {
1541 	ept_common();
1542 }
1543 
1544 static int eptad_exit_handler(union exit_reason exit_reason)
1545 {
1546 	return ept_exit_handler_common(exit_reason, true);
1547 }
1548 
1549 #define TIMER_VECTOR	222
1550 
1551 static volatile bool timer_fired;
1552 
1553 static void timer_isr(isr_regs_t *regs)
1554 {
1555 	timer_fired = true;
1556 	apic_write(APIC_EOI, 0);
1557 }
1558 
1559 static int interrupt_init(struct vmcs *vmcs)
1560 {
1561 	msr_bmp_init();
1562 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1563 	handle_irq(TIMER_VECTOR, timer_isr);
1564 	return VMX_TEST_START;
1565 }
1566 
1567 static void interrupt_main(void)
1568 {
1569 	long long start, loops;
1570 
1571 	vmx_set_test_stage(0);
1572 
1573 	apic_write(APIC_LVTT, TIMER_VECTOR);
1574 	irq_enable();
1575 
1576 	apic_write(APIC_TMICT, 1);
1577 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1578 		asm volatile ("nop");
1579 	report(timer_fired, "direct interrupt while running guest");
1580 
1581 	apic_write(APIC_TMICT, 0);
1582 	irq_disable();
1583 	vmcall();
1584 	timer_fired = false;
1585 	apic_write(APIC_TMICT, 1);
1586 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1587 		asm volatile ("nop");
1588 	report(timer_fired, "intercepted interrupt while running guest");
1589 
1590 	irq_enable();
1591 	apic_write(APIC_TMICT, 0);
1592 	irq_disable();
1593 	vmcall();
1594 	timer_fired = false;
1595 	start = rdtsc();
1596 	apic_write(APIC_TMICT, 1000000);
1597 
1598 	safe_halt();
1599 
1600 	report(rdtsc() - start > 1000000 && timer_fired,
1601 	       "direct interrupt + hlt");
1602 
1603 	apic_write(APIC_TMICT, 0);
1604 	irq_disable();
1605 	vmcall();
1606 	timer_fired = false;
1607 	start = rdtsc();
1608 	apic_write(APIC_TMICT, 1000000);
1609 
1610 	safe_halt();
1611 
1612 	report(rdtsc() - start > 10000 && timer_fired,
1613 	       "intercepted interrupt + hlt");
1614 
1615 	apic_write(APIC_TMICT, 0);
1616 	irq_disable();
1617 	vmcall();
1618 	timer_fired = false;
1619 	start = rdtsc();
1620 	apic_write(APIC_TMICT, 1000000);
1621 
1622 	irq_enable();
1623 	asm volatile ("nop");
1624 	vmcall();
1625 
1626 	report(rdtsc() - start > 10000 && timer_fired,
1627 	       "direct interrupt + activity state hlt");
1628 
1629 	apic_write(APIC_TMICT, 0);
1630 	irq_disable();
1631 	vmcall();
1632 	timer_fired = false;
1633 	start = rdtsc();
1634 	apic_write(APIC_TMICT, 1000000);
1635 
1636 	irq_enable();
1637 	asm volatile ("nop");
1638 	vmcall();
1639 
1640 	report(rdtsc() - start > 10000 && timer_fired,
1641 	       "intercepted interrupt + activity state hlt");
1642 
1643 	apic_write(APIC_TMICT, 0);
1644 	irq_disable();
1645 	vmx_set_test_stage(7);
1646 	vmcall();
1647 	timer_fired = false;
1648 	apic_write(APIC_TMICT, 1);
1649 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1650 		asm volatile ("nop");
1651 	report(timer_fired,
1652 	       "running a guest with interrupt acknowledgement set");
1653 
1654 	apic_write(APIC_TMICT, 0);
1655 	irq_enable();
1656 	timer_fired = false;
1657 	vmcall();
1658 	report(timer_fired, "Inject an event to a halted guest");
1659 }
1660 
1661 static int interrupt_exit_handler(union exit_reason exit_reason)
1662 {
1663 	u64 guest_rip = vmcs_read(GUEST_RIP);
1664 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1665 
1666 	switch (exit_reason.basic) {
1667 	case VMX_VMCALL:
1668 		switch (vmx_get_test_stage()) {
1669 		case 0:
1670 		case 2:
1671 		case 5:
1672 			vmcs_write(PIN_CONTROLS,
1673 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1674 			break;
1675 		case 7:
1676 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1677 			vmcs_write(PIN_CONTROLS,
1678 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1679 			break;
1680 		case 1:
1681 		case 3:
1682 			vmcs_write(PIN_CONTROLS,
1683 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1684 			break;
1685 		case 4:
1686 		case 6:
1687 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1688 			break;
1689 
1690 		case 8:
1691 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1692 			vmcs_write(ENT_INTR_INFO,
1693 				   TIMER_VECTOR |
1694 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1695 				   INTR_INFO_VALID_MASK);
1696 			break;
1697 		}
1698 		vmx_inc_test_stage();
1699 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1700 		return VMX_TEST_RESUME;
1701 	case VMX_EXTINT:
1702 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1703 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1704 			handle_external_interrupt(vector);
1705 		} else {
1706 			irq_enable();
1707 			asm volatile ("nop");
1708 			irq_disable();
1709 		}
1710 		if (vmx_get_test_stage() >= 2)
1711 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1712 		return VMX_TEST_RESUME;
1713 	default:
1714 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1715 		print_vmexit_info(exit_reason);
1716 	}
1717 
1718 	return VMX_TEST_VMEXIT;
1719 }
1720 
1721 
1722 static volatile int nmi_fired;
1723 
1724 #define NMI_DELAY 100000000ULL
1725 
1726 static void nmi_isr(isr_regs_t *regs)
1727 {
1728 	nmi_fired = true;
1729 }
1730 
1731 static int nmi_hlt_init(struct vmcs *vmcs)
1732 {
1733 	msr_bmp_init();
1734 	handle_irq(NMI_VECTOR, nmi_isr);
1735 	vmcs_write(PIN_CONTROLS,
1736 		   vmcs_read(PIN_CONTROLS) & ~PIN_NMI);
1737 	vmcs_write(PIN_CONTROLS,
1738 		   vmcs_read(PIN_CONTROLS) & ~PIN_VIRT_NMI);
1739 	return VMX_TEST_START;
1740 }
1741 
1742 static void nmi_message_thread(void *data)
1743 {
1744     while (vmx_get_test_stage() != 1)
1745         pause();
1746 
1747     delay(NMI_DELAY);
1748     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1749 
1750     while (vmx_get_test_stage() != 2)
1751         pause();
1752 
1753     delay(NMI_DELAY);
1754     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1755 }
1756 
1757 static void nmi_hlt_main(void)
1758 {
1759     long long start;
1760 
1761     if (cpu_count() < 2) {
1762         report_skip(__func__);
1763         vmx_set_test_stage(-1);
1764         return;
1765     }
1766 
1767     vmx_set_test_stage(0);
1768     on_cpu_async(1, nmi_message_thread, NULL);
1769     start = rdtsc();
1770     vmx_set_test_stage(1);
1771     asm volatile ("hlt");
1772     report((rdtsc() - start > NMI_DELAY) && nmi_fired,
1773             "direct NMI + hlt");
1774     if (!nmi_fired)
1775         vmx_set_test_stage(-1);
1776     nmi_fired = false;
1777 
1778     vmcall();
1779 
1780     start = rdtsc();
1781     vmx_set_test_stage(2);
1782     asm volatile ("hlt");
1783     report((rdtsc() - start > NMI_DELAY) && !nmi_fired,
1784             "intercepted NMI + hlt");
1785     if (nmi_fired) {
1786         report(!nmi_fired, "intercepted NMI was dispatched");
1787         vmx_set_test_stage(-1);
1788         return;
1789     }
1790     vmx_set_test_stage(3);
1791 }
1792 
1793 static int nmi_hlt_exit_handler(union exit_reason exit_reason)
1794 {
1795     u64 guest_rip = vmcs_read(GUEST_RIP);
1796     u32 insn_len = vmcs_read(EXI_INST_LEN);
1797 
1798     switch (vmx_get_test_stage()) {
1799     case 1:
1800         if (exit_reason.basic != VMX_VMCALL) {
1801             report_fail("VMEXIT not due to vmcall. Exit reason 0x%x",
1802                         exit_reason.full);
1803             print_vmexit_info(exit_reason);
1804             return VMX_TEST_VMEXIT;
1805         }
1806 
1807         vmcs_write(PIN_CONTROLS,
1808                vmcs_read(PIN_CONTROLS) | PIN_NMI);
1809         vmcs_write(PIN_CONTROLS,
1810                vmcs_read(PIN_CONTROLS) | PIN_VIRT_NMI);
1811         vmcs_write(GUEST_RIP, guest_rip + insn_len);
1812         break;
1813 
1814     case 2:
1815         if (exit_reason.basic != VMX_EXC_NMI) {
1816             report_fail("VMEXIT not due to NMI intercept. Exit reason 0x%x",
1817                         exit_reason.full);
1818             print_vmexit_info(exit_reason);
1819             return VMX_TEST_VMEXIT;
1820         }
1821         report_pass("NMI intercept while running guest");
1822         vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1823         break;
1824 
1825     case 3:
1826         break;
1827 
1828     default:
1829         return VMX_TEST_VMEXIT;
1830     }
1831 
1832     if (vmx_get_test_stage() == 3)
1833         return VMX_TEST_VMEXIT;
1834 
1835     return VMX_TEST_RESUME;
1836 }
1837 
1838 
1839 static int dbgctls_init(struct vmcs *vmcs)
1840 {
1841 	u64 dr7 = 0x402;
1842 	u64 zero = 0;
1843 
1844 	msr_bmp_init();
1845 	asm volatile(
1846 		"mov %0,%%dr0\n\t"
1847 		"mov %0,%%dr1\n\t"
1848 		"mov %0,%%dr2\n\t"
1849 		"mov %1,%%dr7\n\t"
1850 		: : "r" (zero), "r" (dr7));
1851 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1852 	vmcs_write(GUEST_DR7, 0x404);
1853 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1854 
1855 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1856 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1857 
1858 	return VMX_TEST_START;
1859 }
1860 
1861 static void dbgctls_main(void)
1862 {
1863 	u64 dr7, debugctl;
1864 
1865 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1866 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1867 	/* Commented out: KVM does not support DEBUGCTL so far */
1868 	(void)debugctl;
1869 	report(dr7 == 0x404, "Load debug controls" /* && debugctl == 0x2 */);
1870 
1871 	dr7 = 0x408;
1872 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1873 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1874 
1875 	vmx_set_test_stage(0);
1876 	vmcall();
1877 	report(vmx_get_test_stage() == 1, "Save debug controls");
1878 
1879 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1880 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1881 		printf("\tDebug controls are always loaded/saved\n");
1882 		return;
1883 	}
1884 	vmx_set_test_stage(2);
1885 	vmcall();
1886 
1887 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1888 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1889 	/* Commented out: KVM does not support DEBUGCTL so far */
1890 	(void)debugctl;
1891 	report(dr7 == 0x402,
1892 	       "Guest=host debug controls" /* && debugctl == 0x1 */);
1893 
1894 	dr7 = 0x408;
1895 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1896 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1897 
1898 	vmx_set_test_stage(3);
1899 	vmcall();
1900 	report(vmx_get_test_stage() == 4, "Don't save debug controls");
1901 }
1902 
1903 static int dbgctls_exit_handler(union exit_reason exit_reason)
1904 {
1905 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1906 	u64 guest_rip = vmcs_read(GUEST_RIP);
1907 	u64 dr7, debugctl;
1908 
1909 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1910 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1911 
1912 	switch (exit_reason.basic) {
1913 	case VMX_VMCALL:
1914 		switch (vmx_get_test_stage()) {
1915 		case 0:
1916 			if (dr7 == 0x400 && debugctl == 0 &&
1917 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1918 			    Commented out: KVM does not support DEBUGCTL so far
1919 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1920 				vmx_inc_test_stage();
1921 			break;
1922 		case 2:
1923 			dr7 = 0x402;
1924 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1925 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1926 			vmcs_write(GUEST_DR7, 0x404);
1927 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1928 
1929 			vmcs_write(ENT_CONTROLS,
1930 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1931 			vmcs_write(EXI_CONTROLS,
1932 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1933 			break;
1934 		case 3:
1935 			if (dr7 == 0x400 && debugctl == 0 &&
1936 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1937 			    Commented out: KVM does not support DEBUGCTL so far
1938 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1939 				vmx_inc_test_stage();
1940 			break;
1941 		}
1942 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1943 		return VMX_TEST_RESUME;
1944 	default:
1945 		report_fail("Unknown exit reason, %d", exit_reason.full);
1946 		print_vmexit_info(exit_reason);
1947 	}
1948 	return VMX_TEST_VMEXIT;
1949 }
1950 
1951 struct vmx_msr_entry {
1952 	u32 index;
1953 	u32 reserved;
1954 	u64 value;
1955 } __attribute__((packed));
1956 
1957 #define MSR_MAGIC 0x31415926
1958 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1959 
1960 static int msr_switch_init(struct vmcs *vmcs)
1961 {
1962 	msr_bmp_init();
1963 	exit_msr_store = alloc_page();
1964 	exit_msr_load = alloc_page();
1965 	entry_msr_load = alloc_page();
1966 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1967 	entry_msr_load[0].value = MSR_MAGIC;
1968 
1969 	vmx_set_test_stage(1);
1970 	vmcs_write(ENT_MSR_LD_CNT, 1);
1971 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1972 	vmcs_write(EXI_MSR_ST_CNT, 1);
1973 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1974 	vmcs_write(EXI_MSR_LD_CNT, 1);
1975 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1976 	return VMX_TEST_START;
1977 }
1978 
1979 static void msr_switch_main(void)
1980 {
1981 	if (vmx_get_test_stage() == 1) {
1982 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC,
1983 		       "VM entry MSR load");
1984 		vmx_set_test_stage(2);
1985 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
1986 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
1987 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
1988 		exit_msr_load[0].value = MSR_MAGIC + 2;
1989 	}
1990 	vmcall();
1991 }
1992 
1993 static int msr_switch_exit_handler(union exit_reason exit_reason)
1994 {
1995 	if (exit_reason.basic == VMX_VMCALL && vmx_get_test_stage() == 2) {
1996 		report(exit_msr_store[0].value == MSR_MAGIC + 1,
1997 		       "VM exit MSR store");
1998 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2,
1999 		       "VM exit MSR load");
2000 		vmx_set_test_stage(3);
2001 		entry_msr_load[0].index = MSR_FS_BASE;
2002 		return VMX_TEST_RESUME;
2003 	}
2004 	printf("ERROR %s: unexpected stage=%u or reason=0x%x\n",
2005 		__func__, vmx_get_test_stage(), exit_reason.full);
2006 	return VMX_TEST_EXIT;
2007 }
2008 
2009 static int msr_switch_entry_failure(struct vmentry_result *result)
2010 {
2011 	if (result->vm_fail) {
2012 		printf("ERROR %s: VM-Fail on %s\n", __func__, result->instr);
2013 		return VMX_TEST_EXIT;
2014 	}
2015 
2016 	if (result->exit_reason.failed_vmentry &&
2017 	    result->exit_reason.basic == VMX_FAIL_MSR &&
2018 	    vmx_get_test_stage() == 3) {
2019 		report(vmcs_read(EXI_QUALIFICATION) == 1,
2020 		       "VM entry MSR load: try to load FS_BASE");
2021 		return VMX_TEST_VMEXIT;
2022 	}
2023 	printf("ERROR %s: unexpected stage=%u or reason=%x\n",
2024 		__func__, vmx_get_test_stage(), result->exit_reason.full);
2025 	return VMX_TEST_EXIT;
2026 }
2027 
2028 static int vmmcall_init(struct vmcs *vmcs)
2029 {
2030 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
2031 	return VMX_TEST_START;
2032 }
2033 
2034 static void vmmcall_main(void)
2035 {
2036 	asm volatile(
2037 		"mov $0xABCD, %%rax\n\t"
2038 		"vmmcall\n\t"
2039 		::: "rax");
2040 
2041 	report_fail("VMMCALL");
2042 }
2043 
2044 static int vmmcall_exit_handler(union exit_reason exit_reason)
2045 {
2046 	switch (exit_reason.basic) {
2047 	case VMX_VMCALL:
2048 		printf("here\n");
2049 		report_fail("VMMCALL triggers #UD");
2050 		break;
2051 	case VMX_EXC_NMI:
2052 		report((vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR,
2053 		       "VMMCALL triggers #UD");
2054 		break;
2055 	default:
2056 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2057 		print_vmexit_info(exit_reason);
2058 	}
2059 
2060 	return VMX_TEST_VMEXIT;
2061 }
2062 
2063 static int disable_rdtscp_init(struct vmcs *vmcs)
2064 {
2065 	u32 ctrl_cpu1;
2066 
2067 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
2068 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
2069 		ctrl_cpu1 &= ~CPU_RDTSCP;
2070 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
2071 	}
2072 
2073 	return VMX_TEST_START;
2074 }
2075 
2076 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
2077 {
2078 	switch (vmx_get_test_stage()) {
2079 	case 0:
2080 		report_pass("RDTSCP triggers #UD");
2081 		vmx_inc_test_stage();
2082 		regs->rip += 3;
2083 		break;
2084 	case 2:
2085 		report_pass("RDPID triggers #UD");
2086 		vmx_inc_test_stage();
2087 		regs->rip += 4;
2088 		break;
2089 	}
2090 	return;
2091 
2092 }
2093 
2094 static void disable_rdtscp_main(void)
2095 {
2096 	/* Test that #UD is properly injected in L2.  */
2097 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2098 
2099 	vmx_set_test_stage(0);
2100 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2101 	vmcall();
2102 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2103 
2104 	handle_exception(UD_VECTOR, 0);
2105 	vmcall();
2106 }
2107 
2108 static int disable_rdtscp_exit_handler(union exit_reason exit_reason)
2109 {
2110 	switch (exit_reason.basic) {
2111 	case VMX_VMCALL:
2112 		switch (vmx_get_test_stage()) {
2113 		case 0:
2114 			report_fail("RDTSCP triggers #UD");
2115 			vmx_inc_test_stage();
2116 			/* fallthrough */
2117 		case 1:
2118 			vmx_inc_test_stage();
2119 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2120 			return VMX_TEST_RESUME;
2121 		case 2:
2122 			report_fail("RDPID triggers #UD");
2123 			break;
2124 		}
2125 		break;
2126 
2127 	default:
2128 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2129 		print_vmexit_info(exit_reason);
2130 	}
2131 	return VMX_TEST_VMEXIT;
2132 }
2133 
2134 static int int3_init(struct vmcs *vmcs)
2135 {
2136 	vmcs_write(EXC_BITMAP, ~0u);
2137 	return VMX_TEST_START;
2138 }
2139 
2140 static void int3_guest_main(void)
2141 {
2142 	asm volatile ("int3");
2143 }
2144 
2145 static int int3_exit_handler(union exit_reason exit_reason)
2146 {
2147 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2148 
2149 	report(exit_reason.basic == VMX_EXC_NMI &&
2150 	       (intr_info & INTR_INFO_VALID_MASK) &&
2151 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2152 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2153 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2154 	       "L1 intercepts #BP");
2155 
2156 	return VMX_TEST_VMEXIT;
2157 }
2158 
2159 static int into_init(struct vmcs *vmcs)
2160 {
2161 	vmcs_write(EXC_BITMAP, ~0u);
2162 	return VMX_TEST_START;
2163 }
2164 
2165 static void into_guest_main(void)
2166 {
2167 	struct far_pointer32 fp = {
2168 		.offset = (uintptr_t)&&into,
2169 		.selector = KERNEL_CS32,
2170 	};
2171 	uintptr_t rsp;
2172 
2173 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
2174 
2175 	if (fp.offset != (uintptr_t)&&into) {
2176 		printf("Code address too high.\n");
2177 		return;
2178 	}
2179 	if ((u32)rsp != rsp) {
2180 		printf("Stack address too high.\n");
2181 		return;
2182 	}
2183 
2184 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2185 	return;
2186 into:
2187 	asm volatile (".code32;"
2188 		      "movl $0x7fffffff, %eax;"
2189 		      "addl %eax, %eax;"
2190 		      "into;"
2191 		      "lret;"
2192 		      ".code64");
2193 	__builtin_unreachable();
2194 }
2195 
2196 static int into_exit_handler(union exit_reason exit_reason)
2197 {
2198 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2199 
2200 	report(exit_reason.basic == VMX_EXC_NMI &&
2201 	       (intr_info & INTR_INFO_VALID_MASK) &&
2202 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2203 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2204 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2205 	       "L1 intercepts #OF");
2206 
2207 	return VMX_TEST_VMEXIT;
2208 }
2209 
2210 static void exit_monitor_from_l2_main(void)
2211 {
2212 	printf("Calling exit(0) from l2...\n");
2213 	exit(0);
2214 }
2215 
2216 static int exit_monitor_from_l2_handler(union exit_reason exit_reason)
2217 {
2218 	report_fail("The guest should have killed the VMM");
2219 	return VMX_TEST_EXIT;
2220 }
2221 
2222 static void assert_exit_reason(u64 expected)
2223 {
2224 	u64 actual = vmcs_read(EXI_REASON);
2225 
2226 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2227 			   exit_reason_description(expected),
2228 			   exit_reason_description(actual));
2229 }
2230 
2231 static void skip_exit_insn(void)
2232 {
2233 	u64 guest_rip = vmcs_read(GUEST_RIP);
2234 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2235 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2236 }
2237 
2238 static void skip_exit_vmcall(void)
2239 {
2240 	assert_exit_reason(VMX_VMCALL);
2241 	skip_exit_insn();
2242 }
2243 
2244 static void v2_null_test_guest(void)
2245 {
2246 }
2247 
2248 static void v2_null_test(void)
2249 {
2250 	test_set_guest(v2_null_test_guest);
2251 	enter_guest();
2252 	report_pass(__func__);
2253 }
2254 
2255 static void v2_multiple_entries_test_guest(void)
2256 {
2257 	vmx_set_test_stage(1);
2258 	vmcall();
2259 	vmx_set_test_stage(2);
2260 }
2261 
2262 static void v2_multiple_entries_test(void)
2263 {
2264 	test_set_guest(v2_multiple_entries_test_guest);
2265 	enter_guest();
2266 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2267 	skip_exit_vmcall();
2268 	enter_guest();
2269 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2270 	report_pass(__func__);
2271 }
2272 
2273 static int fixture_test_data = 1;
2274 
2275 static void fixture_test_teardown(void *data)
2276 {
2277 	*((int *) data) = 1;
2278 }
2279 
2280 static void fixture_test_guest(void)
2281 {
2282 	fixture_test_data++;
2283 }
2284 
2285 
2286 static void fixture_test_setup(void)
2287 {
2288 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2289 			   "fixture_test_teardown didn't run?!");
2290 	fixture_test_data = 2;
2291 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2292 	test_set_guest(fixture_test_guest);
2293 }
2294 
2295 static void fixture_test_case1(void)
2296 {
2297 	fixture_test_setup();
2298 	TEST_ASSERT_EQ(2, fixture_test_data);
2299 	enter_guest();
2300 	TEST_ASSERT_EQ(3, fixture_test_data);
2301 	report_pass(__func__);
2302 }
2303 
2304 static void fixture_test_case2(void)
2305 {
2306 	fixture_test_setup();
2307 	TEST_ASSERT_EQ(2, fixture_test_data);
2308 	enter_guest();
2309 	TEST_ASSERT_EQ(3, fixture_test_data);
2310 	report_pass(__func__);
2311 }
2312 
2313 enum ept_access_op {
2314 	OP_READ,
2315 	OP_WRITE,
2316 	OP_EXEC,
2317 	OP_FLUSH_TLB,
2318 	OP_EXIT,
2319 };
2320 
2321 static struct ept_access_test_data {
2322 	unsigned long gpa;
2323 	unsigned long *gva;
2324 	unsigned long hpa;
2325 	unsigned long *hva;
2326 	enum ept_access_op op;
2327 } ept_access_test_data;
2328 
2329 extern unsigned char ret42_start;
2330 extern unsigned char ret42_end;
2331 
2332 /* Returns 42. */
2333 asm(
2334 	".align 64\n"
2335 	"ret42_start:\n"
2336 	"mov $42, %eax\n"
2337 	"ret\n"
2338 	"ret42_end:\n"
2339 );
2340 
2341 static void
2342 diagnose_ept_violation_qual(u64 expected, u64 actual)
2343 {
2344 
2345 #define DIAGNOSE(flag)							\
2346 do {									\
2347 	if ((expected & flag) != (actual & flag))			\
2348 		printf(#flag " %sexpected\n",				\
2349 		       (expected & flag) ? "" : "un");			\
2350 } while (0)
2351 
2352 	DIAGNOSE(EPT_VLT_RD);
2353 	DIAGNOSE(EPT_VLT_WR);
2354 	DIAGNOSE(EPT_VLT_FETCH);
2355 	DIAGNOSE(EPT_VLT_PERM_RD);
2356 	DIAGNOSE(EPT_VLT_PERM_WR);
2357 	DIAGNOSE(EPT_VLT_PERM_EX);
2358 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2359 	DIAGNOSE(EPT_VLT_PADDR);
2360 
2361 #undef DIAGNOSE
2362 }
2363 
2364 static void do_ept_access_op(enum ept_access_op op)
2365 {
2366 	ept_access_test_data.op = op;
2367 	enter_guest();
2368 }
2369 
2370 /*
2371  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2372  * needed by tests that modify guest PTEs.
2373  */
2374 static void ept_access_test_guest_flush_tlb(void)
2375 {
2376 	do_ept_access_op(OP_FLUSH_TLB);
2377 	skip_exit_vmcall();
2378 }
2379 
2380 /*
2381  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2382  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2383  * a huge page.
2384  */
2385 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2386 				 unsigned long clear, unsigned long set)
2387 {
2388 	struct ept_access_test_data *data = &ept_access_test_data;
2389 	unsigned long orig_pte;
2390 	unsigned long pte;
2391 
2392 	/* Screw with the mapping at the requested level. */
2393 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2394 	pte = orig_pte;
2395 	if (mkhuge)
2396 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2397 	else
2398 		pte = orig_pte;
2399 	pte = (pte & ~clear) | set;
2400 	set_ept_pte(pml4, gpa, level, pte);
2401 	invept(INVEPT_SINGLE, eptp);
2402 
2403 	return orig_pte;
2404 }
2405 
2406 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2407 {
2408 	set_ept_pte(pml4, gpa, level, orig_pte);
2409 	invept(INVEPT_SINGLE, eptp);
2410 }
2411 
2412 static void do_ept_violation(bool leaf, enum ept_access_op op,
2413 			     u64 expected_qual, u64 expected_paddr)
2414 {
2415 	u64 qual;
2416 
2417 	/* Try the access and observe the violation. */
2418 	do_ept_access_op(op);
2419 
2420 	assert_exit_reason(VMX_EPT_VIOLATION);
2421 
2422 	qual = vmcs_read(EXI_QUALIFICATION);
2423 
2424 	/* Mask undefined bits (which may later be defined in certain cases). */
2425 	qual &= ~(EPT_VLT_GUEST_USER | EPT_VLT_GUEST_RW | EPT_VLT_GUEST_EX |
2426 		 EPT_VLT_PERM_USER_EX);
2427 
2428 	diagnose_ept_violation_qual(expected_qual, qual);
2429 	TEST_EXPECT_EQ(expected_qual, qual);
2430 
2431 	#if 0
2432 	/* Disable for now otherwise every test will fail */
2433 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2434 		       (unsigned long) (
2435 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2436 	#endif
2437 	/*
2438 	 * TODO: tests that probe expected_paddr in pages other than the one at
2439 	 * the beginning of the 1g region.
2440 	 */
2441 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2442 }
2443 
2444 static void
2445 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2446 			      unsigned long set, enum ept_access_op op,
2447 			      u64 expected_qual)
2448 {
2449 	struct ept_access_test_data *data = &ept_access_test_data;
2450 	unsigned long orig_pte;
2451 
2452 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2453 
2454 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2455 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2456 					 data->gpa);
2457 
2458 	/* Fix the violation and resume the op loop. */
2459 	ept_untwiddle(data->gpa, level, orig_pte);
2460 	enter_guest();
2461 	skip_exit_vmcall();
2462 }
2463 
2464 static void
2465 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2466 		       enum ept_access_op op, u64 expected_qual)
2467 {
2468 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2469 				      expected_qual);
2470 	if (ept_huge_pages_supported(level))
2471 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2472 					      expected_qual);
2473 }
2474 
2475 static void ept_violation(unsigned long clear, unsigned long set,
2476 			  enum ept_access_op op, u64 expected_qual)
2477 {
2478 	ept_violation_at_level(1, clear, set, op, expected_qual);
2479 	ept_violation_at_level(2, clear, set, op, expected_qual);
2480 	ept_violation_at_level(3, clear, set, op, expected_qual);
2481 	ept_violation_at_level(4, clear, set, op, expected_qual);
2482 }
2483 
2484 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2485 				       u64 expected_qual)
2486 {
2487 	ept_violation(EPT_PRESENT, access, op,
2488 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2489 }
2490 
2491 /*
2492  * For translations that don't involve a GVA, that is physical address (paddr)
2493  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2494  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2495  * translations don't involve GVAs, such as when the hardware does the guest
2496  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2497  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2498  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2499  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2500  *
2501  * Note that paddr violations can also be triggered by loading PAE page tables
2502  * with wonky addresses. We don't test that yet.
2503  *
2504  * This function modifies the EPT entry that maps the GPA that the guest page
2505  * table entry mapping ept_access_test_data.gva resides on.
2506  *
2507  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2508  *
2509  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2510  *
2511  *	@op		Guest operation to perform with
2512  *			ept_access_test_data.gva.
2513  *
2514  *	@expect_violation
2515  *			Is a violation expected during the paddr access?
2516  *
2517  *	@expected_qual	Expected qualification for the EPT violation.
2518  *			EPT_VLT_PADDR should be clear.
2519  */
2520 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2521 			     enum ept_access_op op, bool expect_violation,
2522 			     u64 expected_qual)
2523 {
2524 	struct ept_access_test_data *data = &ept_access_test_data;
2525 	unsigned long *ptep;
2526 	unsigned long gpa;
2527 	unsigned long orig_epte;
2528 	unsigned long epte;
2529 	int i;
2530 
2531 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2532 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2533 	TEST_ASSERT(ptep);
2534 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2535 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2536 	ept_access_test_guest_flush_tlb();
2537 
2538 	/*
2539 	 * Now modify the access bits on the EPT entry for the GPA that the
2540 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2541 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2542 	 * constructed our test such that those other 511 PTEs aren't used by
2543 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2544 	 * PTE we're modifying is at the beginning of a 4K page and the
2545 	 * following 511 entires are also under our control (and not touched by
2546 	 * the guest).
2547 	 */
2548 	gpa = virt_to_phys(ptep);
2549 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2550 	/*
2551 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2552 	 * otherwise our level=1 twiddling below will fail. We use the
2553 	 * identity map (gpa = gpa) since page tables are shared with the host.
2554 	 */
2555 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2556 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2557 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2558 
2559 	if (expect_violation) {
2560 		do_ept_violation(/*leaf=*/true, op,
2561 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2562 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2563 		do_ept_access_op(op);
2564 	} else {
2565 		do_ept_access_op(op);
2566 		if (ept_ad_enabled()) {
2567 			for (i = EPT_PAGE_LEVEL; i > 0; i--) {
2568 				TEST_ASSERT(get_ept_pte(pml4, gpa, i, &epte));
2569 				TEST_ASSERT(epte & EPT_ACCESS_FLAG);
2570 				if (i == 1)
2571 					TEST_ASSERT(epte & EPT_DIRTY_FLAG);
2572 				else
2573 					TEST_ASSERT_EQ(epte & EPT_DIRTY_FLAG, 0);
2574 			}
2575 		}
2576 
2577 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2578 	}
2579 
2580 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2581 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2582 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2583 
2584 	skip_exit_vmcall();
2585 }
2586 
2587 static void ept_access_allowed_paddr(unsigned long ept_access,
2588 				     unsigned long pte_ad,
2589 				     enum ept_access_op op)
2590 {
2591 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2592 			 /*expected_qual=*/-1);
2593 }
2594 
2595 static void ept_access_violation_paddr(unsigned long ept_access,
2596 				       unsigned long pte_ad,
2597 				       enum ept_access_op op,
2598 				       u64 expected_qual)
2599 {
2600 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2601 			 expected_qual);
2602 }
2603 
2604 
2605 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2606 					unsigned long clear,
2607 					unsigned long set,
2608 					enum ept_access_op op)
2609 {
2610 	struct ept_access_test_data *data = &ept_access_test_data;
2611 	unsigned long orig_pte;
2612 
2613 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2614 
2615 	/* No violation. Should proceed to vmcall. */
2616 	do_ept_access_op(op);
2617 	skip_exit_vmcall();
2618 
2619 	ept_untwiddle(data->gpa, level, orig_pte);
2620 }
2621 
2622 static void ept_allowed_at_level(int level, unsigned long clear,
2623 				 unsigned long set, enum ept_access_op op)
2624 {
2625 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2626 	if (ept_huge_pages_supported(level))
2627 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2628 }
2629 
2630 static void ept_allowed(unsigned long clear, unsigned long set,
2631 			enum ept_access_op op)
2632 {
2633 	ept_allowed_at_level(1, clear, set, op);
2634 	ept_allowed_at_level(2, clear, set, op);
2635 	ept_allowed_at_level(3, clear, set, op);
2636 	ept_allowed_at_level(4, clear, set, op);
2637 }
2638 
2639 static void ept_ignored_bit(int bit)
2640 {
2641 	/* Set the bit. */
2642 	ept_allowed(0, 1ul << bit, OP_READ);
2643 	ept_allowed(0, 1ul << bit, OP_WRITE);
2644 	ept_allowed(0, 1ul << bit, OP_EXEC);
2645 
2646 	/* Clear the bit. */
2647 	ept_allowed(1ul << bit, 0, OP_READ);
2648 	ept_allowed(1ul << bit, 0, OP_WRITE);
2649 	ept_allowed(1ul << bit, 0, OP_EXEC);
2650 }
2651 
2652 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2653 {
2654 	ept_allowed(EPT_PRESENT, access, op);
2655 }
2656 
2657 
2658 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2659 					     unsigned long clear,
2660 					     unsigned long set,
2661 					     enum ept_access_op op)
2662 {
2663 	struct ept_access_test_data *data = &ept_access_test_data;
2664 	unsigned long orig_pte;
2665 
2666 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2667 
2668 	do_ept_access_op(op);
2669 	assert_exit_reason(VMX_EPT_MISCONFIG);
2670 
2671 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2672 	#if 0
2673 	/* broken: */
2674 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2675 	#endif
2676 	#if 0
2677 	/*
2678 	 * broken:
2679 	 * According to description of exit qual for EPT violation,
2680 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2681 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2682 	 * to be set for msiconfig.
2683 	 */
2684 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2685 		       (unsigned long) (
2686 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2687 	#endif
2688 
2689 	/* Fix the violation and resume the op loop. */
2690 	ept_untwiddle(data->gpa, level, orig_pte);
2691 	enter_guest();
2692 	skip_exit_vmcall();
2693 }
2694 
2695 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2696 					  unsigned long clear,
2697 					  unsigned long set)
2698 {
2699 	/* The op shouldn't matter (read, write, exec), so try them all! */
2700 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2701 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2702 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2703 }
2704 
2705 static void ept_misconfig_at_level(int level, unsigned long clear,
2706 				   unsigned long set)
2707 {
2708 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2709 	if (ept_huge_pages_supported(level))
2710 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2711 }
2712 
2713 static void ept_misconfig(unsigned long clear, unsigned long set)
2714 {
2715 	ept_misconfig_at_level(1, clear, set);
2716 	ept_misconfig_at_level(2, clear, set);
2717 	ept_misconfig_at_level(3, clear, set);
2718 	ept_misconfig_at_level(4, clear, set);
2719 }
2720 
2721 static void ept_access_misconfig(unsigned long access)
2722 {
2723 	ept_misconfig(EPT_PRESENT, access);
2724 }
2725 
2726 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2727 {
2728 	/* Setting the bit causes a misconfig. */
2729 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2730 
2731 	/* Making the entry non-present turns reserved bits into ignored. */
2732 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2733 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2734 }
2735 
2736 static void ept_reserved_bit_at_level_huge(int level, int bit)
2737 {
2738 	/* Setting the bit causes a misconfig. */
2739 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2740 
2741 	/* Making the entry non-present turns reserved bits into ignored. */
2742 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2743 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2744 }
2745 
2746 static void ept_reserved_bit_at_level(int level, int bit)
2747 {
2748 	/* Setting the bit causes a misconfig. */
2749 	ept_misconfig_at_level(level, 0, 1ul << bit);
2750 
2751 	/* Making the entry non-present turns reserved bits into ignored. */
2752 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2753 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2754 }
2755 
2756 static void ept_reserved_bit(int bit)
2757 {
2758 	ept_reserved_bit_at_level(1, bit);
2759 	ept_reserved_bit_at_level(2, bit);
2760 	ept_reserved_bit_at_level(3, bit);
2761 	ept_reserved_bit_at_level(4, bit);
2762 }
2763 
2764 #define PAGE_2M_ORDER 9
2765 #define PAGE_1G_ORDER 18
2766 
2767 static void *get_1g_page(void)
2768 {
2769 	static void *alloc;
2770 
2771 	if (!alloc)
2772 		alloc = alloc_pages(PAGE_1G_ORDER);
2773 	return alloc;
2774 }
2775 
2776 static void ept_access_test_teardown(void *unused)
2777 {
2778 	/* Exit the guest cleanly. */
2779 	do_ept_access_op(OP_EXIT);
2780 }
2781 
2782 static void ept_access_test_guest(void)
2783 {
2784 	struct ept_access_test_data *data = &ept_access_test_data;
2785 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2786 
2787 	while (true) {
2788 		switch (data->op) {
2789 		case OP_READ:
2790 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2791 			break;
2792 		case OP_WRITE:
2793 			*data->gva = MAGIC_VAL_2;
2794 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2795 			*data->gva = MAGIC_VAL_1;
2796 			break;
2797 		case OP_EXEC:
2798 			TEST_ASSERT_EQ(42, code());
2799 			break;
2800 		case OP_FLUSH_TLB:
2801 			write_cr3(read_cr3());
2802 			break;
2803 		case OP_EXIT:
2804 			return;
2805 		default:
2806 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2807 		}
2808 		vmcall();
2809 	}
2810 }
2811 
2812 static void ept_access_test_setup(void)
2813 {
2814 	struct ept_access_test_data *data = &ept_access_test_data;
2815 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2816 	unsigned long size = npages * PAGE_SIZE;
2817 	unsigned long *page_table = current_page_table();
2818 	unsigned long pte;
2819 
2820 	if (setup_ept(false))
2821 		test_skip("EPT not supported");
2822 
2823 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2824 	if (cpuid_maxphyaddr() < 40)
2825 		test_skip("Test needs MAXPHYADDR >= 40");
2826 
2827 	test_set_guest(ept_access_test_guest);
2828 	test_add_teardown(ept_access_test_teardown, NULL);
2829 
2830 	data->hva = get_1g_page();
2831 	TEST_ASSERT(data->hva);
2832 	data->hpa = virt_to_phys(data->hva);
2833 
2834 	data->gpa = 1ul << 39;
2835 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2836 				   size);
2837 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2838 	install_pages(page_table, data->gpa, size, data->gva);
2839 
2840 	/*
2841 	 * Make sure nothing's mapped here so the tests that screw with the
2842 	 * pml4 entry don't inadvertently break something.
2843 	 */
2844 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2845 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2846 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2847 
2848 	data->hva[0] = MAGIC_VAL_1;
2849 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2850 }
2851 
2852 static void ept_access_test_not_present(void)
2853 {
2854 	ept_access_test_setup();
2855 	/* --- */
2856 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2857 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2858 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2859 }
2860 
2861 static void ept_access_test_read_only(void)
2862 {
2863 	ept_access_test_setup();
2864 
2865 	/* r-- */
2866 	ept_access_allowed(EPT_RA, OP_READ);
2867 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2868 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2869 }
2870 
2871 static void ept_access_test_write_only(void)
2872 {
2873 	ept_access_test_setup();
2874 	/* -w- */
2875 	ept_access_misconfig(EPT_WA);
2876 }
2877 
2878 static void ept_access_test_read_write(void)
2879 {
2880 	ept_access_test_setup();
2881 	/* rw- */
2882 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2883 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2884 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2885 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2886 }
2887 
2888 
2889 static void ept_access_test_execute_only(void)
2890 {
2891 	ept_access_test_setup();
2892 	/* --x */
2893 	if (ept_execute_only_supported()) {
2894 		ept_access_violation(EPT_EA, OP_READ,
2895 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2896 		ept_access_violation(EPT_EA, OP_WRITE,
2897 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2898 		ept_access_allowed(EPT_EA, OP_EXEC);
2899 	} else {
2900 		ept_access_misconfig(EPT_EA);
2901 	}
2902 }
2903 
2904 static void ept_access_test_read_execute(void)
2905 {
2906 	ept_access_test_setup();
2907 	/* r-x */
2908 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2909 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2910 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2911 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2912 }
2913 
2914 static void ept_access_test_write_execute(void)
2915 {
2916 	ept_access_test_setup();
2917 	/* -wx */
2918 	ept_access_misconfig(EPT_WA | EPT_EA);
2919 }
2920 
2921 static void ept_access_test_read_write_execute(void)
2922 {
2923 	ept_access_test_setup();
2924 	/* rwx */
2925 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2926 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2927 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2928 }
2929 
2930 static void ept_access_test_reserved_bits(void)
2931 {
2932 	int i;
2933 	int maxphyaddr;
2934 
2935 	ept_access_test_setup();
2936 
2937 	/* Reserved bits above maxphyaddr. */
2938 	maxphyaddr = cpuid_maxphyaddr();
2939 	for (i = maxphyaddr; i <= 51; i++) {
2940 		report_prefix_pushf("reserved_bit=%d", i);
2941 		ept_reserved_bit(i);
2942 		report_prefix_pop();
2943 	}
2944 
2945 	/* Level-specific reserved bits. */
2946 	ept_reserved_bit_at_level_nohuge(2, 3);
2947 	ept_reserved_bit_at_level_nohuge(2, 4);
2948 	ept_reserved_bit_at_level_nohuge(2, 5);
2949 	ept_reserved_bit_at_level_nohuge(2, 6);
2950 	/* 2M alignment. */
2951 	for (i = 12; i < 20; i++) {
2952 		report_prefix_pushf("reserved_bit=%d", i);
2953 		ept_reserved_bit_at_level_huge(2, i);
2954 		report_prefix_pop();
2955 	}
2956 	ept_reserved_bit_at_level_nohuge(3, 3);
2957 	ept_reserved_bit_at_level_nohuge(3, 4);
2958 	ept_reserved_bit_at_level_nohuge(3, 5);
2959 	ept_reserved_bit_at_level_nohuge(3, 6);
2960 	/* 1G alignment. */
2961 	for (i = 12; i < 29; i++) {
2962 		report_prefix_pushf("reserved_bit=%d", i);
2963 		ept_reserved_bit_at_level_huge(3, i);
2964 		report_prefix_pop();
2965 	}
2966 	ept_reserved_bit_at_level(4, 3);
2967 	ept_reserved_bit_at_level(4, 4);
2968 	ept_reserved_bit_at_level(4, 5);
2969 	ept_reserved_bit_at_level(4, 6);
2970 	ept_reserved_bit_at_level(4, 7);
2971 }
2972 
2973 static void ept_access_test_ignored_bits(void)
2974 {
2975 	ept_access_test_setup();
2976 	/*
2977 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2978 	 * far as translation is concerned even if AD bits are enabled in the
2979 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2980 	 * control is 0.
2981 	 */
2982 	ept_ignored_bit(8);
2983 	ept_ignored_bit(9);
2984 	ept_ignored_bit(10);
2985 	ept_ignored_bit(11);
2986 	ept_ignored_bit(52);
2987 	ept_ignored_bit(53);
2988 	ept_ignored_bit(54);
2989 	ept_ignored_bit(55);
2990 	ept_ignored_bit(56);
2991 	ept_ignored_bit(57);
2992 	ept_ignored_bit(58);
2993 	ept_ignored_bit(59);
2994 	ept_ignored_bit(60);
2995 	ept_ignored_bit(61);
2996 	ept_ignored_bit(62);
2997 	ept_ignored_bit(63);
2998 }
2999 
3000 static void ept_access_test_paddr_not_present_ad_disabled(void)
3001 {
3002 	ept_access_test_setup();
3003 	ept_disable_ad_bits();
3004 
3005 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
3006 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
3007 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
3008 }
3009 
3010 static void ept_access_test_paddr_not_present_ad_enabled(void)
3011 {
3012 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
3013 
3014 	ept_access_test_setup();
3015 	ept_enable_ad_bits_or_skip_test();
3016 
3017 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
3018 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
3019 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
3020 }
3021 
3022 static void ept_access_test_paddr_read_only_ad_disabled(void)
3023 {
3024 	/*
3025 	 * When EPT AD bits are disabled, all accesses to guest paging
3026 	 * structures are reported separately as a read and (after
3027 	 * translation of the GPA to host physical address) a read+write
3028 	 * if the A/D bits have to be set.
3029 	 */
3030 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3031 
3032 	ept_access_test_setup();
3033 	ept_disable_ad_bits();
3034 
3035 	/* Can't update A bit, so all accesses fail. */
3036 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3037 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3038 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3039 	/* AD bits disabled, so only writes try to update the D bit. */
3040 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
3041 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3042 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
3043 	/* Both A and D already set, so read-only is OK. */
3044 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
3045 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
3046 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
3047 }
3048 
3049 static void ept_access_test_paddr_read_only_ad_enabled(void)
3050 {
3051 	/*
3052 	 * When EPT AD bits are enabled, all accesses to guest paging
3053 	 * structures are considered writes as far as EPT translation
3054 	 * is concerned.
3055 	 */
3056 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3057 
3058 	ept_access_test_setup();
3059 	ept_enable_ad_bits_or_skip_test();
3060 
3061 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3062 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3063 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3064 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
3065 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3066 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
3067 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
3068 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
3069 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
3070 }
3071 
3072 static void ept_access_test_paddr_read_write(void)
3073 {
3074 	ept_access_test_setup();
3075 	/* Read-write access to paging structure. */
3076 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
3077 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
3078 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
3079 }
3080 
3081 static void ept_access_test_paddr_read_write_execute(void)
3082 {
3083 	ept_access_test_setup();
3084 	/* RWX access to paging structure. */
3085 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
3086 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
3087 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
3088 }
3089 
3090 static void ept_access_test_paddr_read_execute_ad_disabled(void)
3091 {
3092   	/*
3093 	 * When EPT AD bits are disabled, all accesses to guest paging
3094 	 * structures are reported separately as a read and (after
3095 	 * translation of the GPA to host physical address) a read+write
3096 	 * if the A/D bits have to be set.
3097 	 */
3098 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3099 
3100 	ept_access_test_setup();
3101 	ept_disable_ad_bits();
3102 
3103 	/* Can't update A bit, so all accesses fail. */
3104 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3105 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3106 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3107 	/* AD bits disabled, so only writes try to update the D bit. */
3108 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
3109 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3110 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3111 	/* Both A and D already set, so read-only is OK. */
3112 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3113 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3114 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3115 }
3116 
3117 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3118 {
3119 	/*
3120 	 * When EPT AD bits are enabled, all accesses to guest paging
3121 	 * structures are considered writes as far as EPT translation
3122 	 * is concerned.
3123 	 */
3124 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3125 
3126 	ept_access_test_setup();
3127 	ept_enable_ad_bits_or_skip_test();
3128 
3129 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3130 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3131 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3132 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3133 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3134 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3135 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3136 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3137 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3138 }
3139 
3140 static void ept_access_test_paddr_not_present_page_fault(void)
3141 {
3142 	ept_access_test_setup();
3143 	/*
3144 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3145 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3146 	 * Thus guest page fault before host takes EPT violation for trying to
3147 	 * update A bit.
3148 	 */
3149 }
3150 
3151 static void ept_access_test_force_2m_page(void)
3152 {
3153 	ept_access_test_setup();
3154 
3155 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3156 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3157 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3158 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3159 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3160 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3161 }
3162 
3163 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3164 {
3165 	if (!is_invvpid_type_supported(type))
3166 		return false;
3167 
3168 	if (vpid >> 16)
3169 		return false;
3170 
3171 	if (type != INVVPID_ALL && !vpid)
3172 		return false;
3173 
3174 	if (type == INVVPID_ADDR && !is_canonical(gla))
3175 		return false;
3176 
3177 	return true;
3178 }
3179 
3180 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3181 {
3182 	int rc;
3183 	bool valid = invvpid_valid(type, vpid, gla);
3184 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3185 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3186 	/*
3187 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3188 	 * that we can tell if it is updated by INVVPID.
3189 	 */
3190 	vmcs_read(~0);
3191 	rc = __invvpid(type, vpid, gla);
3192 	report(!rc == valid, "INVVPID type %ld VPID %lx GLA %lx %s", type,
3193 	       vpid, gla,
3194 	       valid ? "passes" : "fails");
3195 	report(vmcs_read(VMX_INST_ERROR) == expected,
3196 	       "After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3197 	       rc ? "failed" : "successful",
3198 	       expected, vmcs_read(VMX_INST_ERROR));
3199 }
3200 
3201 static inline unsigned long get_first_supported_invvpid_type(void)
3202 {
3203 	u64 type = ffs(ept_vpid.val >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3204 
3205 	__TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3206 	return type;
3207 }
3208 
3209 static void ds_invvpid(void *data)
3210 {
3211 	asm volatile("invvpid %0, %1"
3212 		     :
3213 		     : "m"(*(struct invvpid_operand *)data),
3214 		       "r"(get_first_supported_invvpid_type()));
3215 }
3216 
3217 /*
3218  * The SS override is ignored in 64-bit mode, so we use an addressing
3219  * mode with %rsp as the base register to generate an implicit SS
3220  * reference.
3221  */
3222 static void ss_invvpid(void *data)
3223 {
3224 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3225 		     : "+r"(data)
3226 		     : "r"(get_first_supported_invvpid_type()));
3227 }
3228 
3229 static void invvpid_test_gp(void)
3230 {
3231 	bool fault;
3232 
3233 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3234 				   (void *)NONCANONICAL);
3235 	report(fault, "INVVPID with non-canonical DS operand raises #GP");
3236 }
3237 
3238 static void invvpid_test_ss(void)
3239 {
3240 	bool fault;
3241 
3242 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3243 				   (void *)NONCANONICAL);
3244 	report(fault, "INVVPID with non-canonical SS operand raises #SS");
3245 }
3246 
3247 static void invvpid_test_pf(void)
3248 {
3249 	void *vpage = alloc_vpage();
3250 	bool fault;
3251 
3252 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3253 	report(fault, "INVVPID with unmapped operand raises #PF");
3254 }
3255 
3256 static void try_compat_invvpid(void *unused)
3257 {
3258 	struct far_pointer32 fp = {
3259 		.offset = (uintptr_t)&&invvpid,
3260 		.selector = KERNEL_CS32,
3261 	};
3262 	uintptr_t rsp;
3263 
3264 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
3265 
3266 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3267 			"Code address too high.");
3268 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3269 
3270 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3271 	return;
3272 invvpid:
3273 	asm volatile (".code32;"
3274 		      "invvpid (%eax), %eax;"
3275 		      "lret;"
3276 		      ".code64");
3277 	__builtin_unreachable();
3278 }
3279 
3280 static void invvpid_test_compatibility_mode(void)
3281 {
3282 	bool fault;
3283 
3284 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3285 	report(fault, "Compatibility mode INVVPID raises #UD");
3286 }
3287 
3288 static void invvpid_test_not_in_vmx_operation(void)
3289 {
3290 	bool fault;
3291 
3292 	TEST_ASSERT(!vmx_off());
3293 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3294 	report(fault, "INVVPID outside of VMX operation raises #UD");
3295 	TEST_ASSERT(!vmx_on());
3296 }
3297 
3298 /*
3299  * This does not test real-address mode, virtual-8086 mode, protected mode,
3300  * or CPL > 0.
3301  */
3302 static void invvpid_test(void)
3303 {
3304 	int i;
3305 	unsigned types = 0;
3306 	unsigned type;
3307 
3308 	if (!is_vpid_supported())
3309 		test_skip("VPID not supported");
3310 
3311 	if (!is_invvpid_supported())
3312 		test_skip("INVVPID not supported.\n");
3313 
3314 	if (is_invvpid_type_supported(INVVPID_ADDR))
3315 		types |= 1u << INVVPID_ADDR;
3316 	if (is_invvpid_type_supported(INVVPID_CONTEXT_GLOBAL))
3317 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3318 	if (is_invvpid_type_supported(INVVPID_ALL))
3319 		types |= 1u << INVVPID_ALL;
3320 	if (is_invvpid_type_supported(INVVPID_CONTEXT_LOCAL))
3321 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3322 
3323 	if (!types)
3324 		test_skip("No INVVPID types supported.\n");
3325 
3326 	for (i = -127; i < 128; i++)
3327 		try_invvpid(i, 0xffff, 0);
3328 
3329 	/*
3330 	 * VPID must not be more than 16 bits.
3331 	 */
3332 	for (i = 0; i < 64; i++)
3333 		for (type = 0; type < 4; type++)
3334 			if (types & (1u << type))
3335 				try_invvpid(type, 1ul << i, 0);
3336 
3337 	/*
3338 	 * VPID must not be zero, except for "all contexts."
3339 	 */
3340 	for (type = 0; type < 4; type++)
3341 		if (types & (1u << type))
3342 			try_invvpid(type, 0, 0);
3343 
3344 	/*
3345 	 * The gla operand is only validated for single-address INVVPID.
3346 	 */
3347 	if (types & (1u << INVVPID_ADDR))
3348 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3349 
3350 	invvpid_test_gp();
3351 	invvpid_test_ss();
3352 	invvpid_test_pf();
3353 	invvpid_test_compatibility_mode();
3354 	invvpid_test_not_in_vmx_operation();
3355 }
3356 
3357 /*
3358  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3359  * at least as far as the guest-state checks. Returns false if the
3360  * VMLAUNCH fails early and execution falls through to the next
3361  * instruction.
3362  */
3363 static bool vmlaunch_succeeds(void)
3364 {
3365 	u32 exit_reason;
3366 
3367 	/*
3368 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3369 	 * unsupported VMCS component"). The caller can then check
3370 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3371 	 */
3372 	vmcs_write(~0u, 0);
3373 
3374 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3375 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3376 				   :
3377 				   : "r" ((u64)HOST_RSP)
3378 				   : "cc", "memory"
3379 				   : success);
3380 	return false;
3381 success:
3382 	exit_reason = vmcs_read(EXI_REASON);
3383 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3384 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3385 	return true;
3386 }
3387 
3388 /*
3389  * Try to launch the current VMCS.
3390  */
3391 static void test_vmx_vmlaunch(u32 xerror)
3392 {
3393 	bool success = vmlaunch_succeeds();
3394 	u32 vmx_inst_err;
3395 
3396 	report(success == !xerror, "vmlaunch %s",
3397 	       !xerror ? "succeeds" : "fails");
3398 	if (!success && xerror) {
3399 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3400 		report(vmx_inst_err == xerror,
3401 		       "VMX inst error is %d (actual %d)", xerror,
3402 		       vmx_inst_err);
3403 	}
3404 }
3405 
3406 /*
3407  * Try to launch the current VMCS, and expect one of two possible
3408  * errors (or success) codes.
3409  */
3410 static void test_vmx_vmlaunch2(u32 xerror1, u32 xerror2)
3411 {
3412 	bool success = vmlaunch_succeeds();
3413 	u32 vmx_inst_err;
3414 
3415 	if (!xerror1 == !xerror2)
3416 		report(success == !xerror1, "vmlaunch %s",
3417 		       !xerror1 ? "succeeds" : "fails");
3418 
3419 	if (!success && (xerror1 || xerror2)) {
3420 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3421 		report(vmx_inst_err == xerror1 || vmx_inst_err == xerror2,
3422 		       "VMX inst error is %d or %d (actual %d)", xerror1,
3423 		       xerror2, vmx_inst_err);
3424 	}
3425 }
3426 
3427 static void test_vmx_invalid_controls(void)
3428 {
3429 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3430 }
3431 
3432 static void test_vmx_valid_controls(void)
3433 {
3434 	test_vmx_vmlaunch(0);
3435 }
3436 
3437 /*
3438  * Test a particular value of a VM-execution control bit, if the value
3439  * is required or if the value is zero.
3440  */
3441 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3442 				    enum Encoding encoding, unsigned bit,
3443 				    unsigned val)
3444 {
3445 	u32 mask = 1u << bit;
3446 	bool expected;
3447 	u32 controls;
3448 
3449 	if (msr.set & mask)
3450 		TEST_ASSERT(msr.clr & mask);
3451 
3452 	/*
3453 	 * We can't arbitrarily turn on a control bit, because it may
3454 	 * introduce dependencies on other VMCS fields. So, we only
3455 	 * test turning on bits that have a required setting.
3456 	 */
3457 	if (val && (msr.clr & mask) && !(msr.set & mask))
3458 		return;
3459 
3460 	report_prefix_pushf("%s %s bit %d",
3461 			    val ? "Set" : "Clear", name, bit);
3462 
3463 	controls = vmcs_read(encoding);
3464 	if (val) {
3465 		vmcs_write(encoding, msr.set | mask);
3466 		expected = (msr.clr & mask);
3467 	} else {
3468 		vmcs_write(encoding, msr.set & ~mask);
3469 		expected = !(msr.set & mask);
3470 	}
3471 	if (expected)
3472 		test_vmx_valid_controls();
3473 	else
3474 		test_vmx_invalid_controls();
3475 	vmcs_write(encoding, controls);
3476 	report_prefix_pop();
3477 }
3478 
3479 /*
3480  * Test reserved values of a VM-execution control bit, based on the
3481  * allowed bit settings from the corresponding VMX capability MSR.
3482  */
3483 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3484 			      enum Encoding encoding, unsigned bit)
3485 {
3486 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3487 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3488 }
3489 
3490 /*
3491  * Reserved bits in the pin-based VM-execution controls must be set
3492  * properly. Software may consult the VMX capability MSRs to determine
3493  * the proper settings.
3494  * [Intel SDM]
3495  */
3496 static void test_pin_based_ctls(void)
3497 {
3498 	unsigned bit;
3499 
3500 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3501 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3502 	for (bit = 0; bit < 32; bit++)
3503 		test_rsvd_ctl_bit("pin-based controls",
3504 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3505 }
3506 
3507 /*
3508  * Reserved bits in the primary processor-based VM-execution controls
3509  * must be set properly. Software may consult the VMX capability MSRs
3510  * to determine the proper settings.
3511  * [Intel SDM]
3512  */
3513 static void test_primary_processor_based_ctls(void)
3514 {
3515 	unsigned bit;
3516 
3517 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3518 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3519 	for (bit = 0; bit < 32; bit++)
3520 		test_rsvd_ctl_bit("primary processor-based controls",
3521 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3522 }
3523 
3524 /*
3525  * If the "activate secondary controls" primary processor-based
3526  * VM-execution control is 1, reserved bits in the secondary
3527  * processor-based VM-execution controls must be cleared. Software may
3528  * consult the VMX capability MSRs to determine which bits are
3529  * reserved.
3530  * If the "activate secondary controls" primary processor-based
3531  * VM-execution control is 0 (or if the processor does not support the
3532  * 1-setting of that control), no checks are performed on the
3533  * secondary processor-based VM-execution controls.
3534  * [Intel SDM]
3535  */
3536 static void test_secondary_processor_based_ctls(void)
3537 {
3538 	u32 primary;
3539 	u32 secondary;
3540 	unsigned bit;
3541 
3542 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3543 		return;
3544 
3545 	primary = vmcs_read(CPU_EXEC_CTRL0);
3546 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3547 
3548 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3549 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3550 	for (bit = 0; bit < 32; bit++)
3551 		test_rsvd_ctl_bit("secondary processor-based controls",
3552 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3553 
3554 	/*
3555 	 * When the "activate secondary controls" VM-execution control
3556 	 * is clear, there are no checks on the secondary controls.
3557 	 */
3558 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3559 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3560 	report(vmlaunch_succeeds(),
3561 	       "Secondary processor-based controls ignored");
3562 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3563 	vmcs_write(CPU_EXEC_CTRL0, primary);
3564 }
3565 
3566 static void try_cr3_target_count(unsigned i, unsigned max)
3567 {
3568 	report_prefix_pushf("CR3 target count 0x%x", i);
3569 	vmcs_write(CR3_TARGET_COUNT, i);
3570 	if (i <= max)
3571 		test_vmx_valid_controls();
3572 	else
3573 		test_vmx_invalid_controls();
3574 	report_prefix_pop();
3575 }
3576 
3577 /*
3578  * The CR3-target count must not be greater than 4. Future processors
3579  * may support a different number of CR3-target values. Software
3580  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3581  * number of values supported.
3582  * [Intel SDM]
3583  */
3584 static void test_cr3_targets(void)
3585 {
3586 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3587 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3588 	unsigned i;
3589 
3590 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3591 	TEST_ASSERT(supported_targets <= 256);
3592 
3593 	try_cr3_target_count(-1u, supported_targets);
3594 	try_cr3_target_count(0x80000000, supported_targets);
3595 	try_cr3_target_count(0x7fffffff, supported_targets);
3596 	for (i = 0; i <= supported_targets + 1; i++)
3597 		try_cr3_target_count(i, supported_targets);
3598 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3599 
3600 	/* VMWRITE to nonexistent target fields should fail. */
3601 	for (i = supported_targets; i < 256; i++)
3602 		TEST_ASSERT(vmcs_write(CR3_TARGET_0 + i*2, 0));
3603 }
3604 
3605 /*
3606  * Test a particular address setting in the VMCS
3607  */
3608 static void test_vmcs_addr(const char *name,
3609 			   enum Encoding encoding,
3610 			   u64 align,
3611 			   bool ignored,
3612 			   bool skip_beyond_mapped_ram,
3613 			   u64 addr)
3614 {
3615 	report_prefix_pushf("%s = %lx", name, addr);
3616 	vmcs_write(encoding, addr);
3617 	if (skip_beyond_mapped_ram &&
3618 	    addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3619 	    addr < (1ul << cpuid_maxphyaddr()))
3620 		printf("Skipping physical address beyond mapped RAM\n");
3621 	else if (ignored || (IS_ALIGNED(addr, align) &&
3622 	    addr < (1ul << cpuid_maxphyaddr())))
3623 		test_vmx_valid_controls();
3624 	else
3625 		test_vmx_invalid_controls();
3626 	report_prefix_pop();
3627 }
3628 
3629 /*
3630  * Test interesting values for a VMCS address
3631  */
3632 static void test_vmcs_addr_values(const char *name,
3633 				  enum Encoding encoding,
3634 				  u64 align,
3635 				  bool ignored,
3636 				  bool skip_beyond_mapped_ram,
3637 				  u32 bit_start, u32 bit_end)
3638 {
3639 	unsigned i;
3640 	u64 orig_val = vmcs_read(encoding);
3641 
3642 	for (i = bit_start; i <= bit_end; i++)
3643 		test_vmcs_addr(name, encoding, align, ignored,
3644 			       skip_beyond_mapped_ram, 1ul << i);
3645 
3646 	test_vmcs_addr(name, encoding, align, ignored,
3647 		       skip_beyond_mapped_ram, PAGE_SIZE - 1);
3648 	test_vmcs_addr(name, encoding, align, ignored,
3649 		       skip_beyond_mapped_ram, PAGE_SIZE);
3650 	test_vmcs_addr(name, encoding, align, ignored,
3651 		       skip_beyond_mapped_ram,
3652 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3653 	test_vmcs_addr(name, encoding, align, ignored,
3654 		       skip_beyond_mapped_ram, -1ul);
3655 
3656 	vmcs_write(encoding, orig_val);
3657 }
3658 
3659 /*
3660  * Test a physical address reference in the VMCS, when the corresponding
3661  * feature is enabled and when the corresponding feature is disabled.
3662  */
3663 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3664 				     const char *field_name,
3665 				     const char *control_name, u64 align,
3666 				     bool skip_beyond_mapped_ram,
3667 				     bool control_primary)
3668 {
3669 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3670 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3671 	u64 page_addr;
3672 
3673 	if (control_primary) {
3674 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3675 			return;
3676 	} else {
3677 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3678 			return;
3679 	}
3680 
3681 	page_addr = vmcs_read(field);
3682 
3683 	report_prefix_pushf("%s enabled", control_name);
3684 	if (control_primary) {
3685 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3686 	} else {
3687 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3688 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3689 	}
3690 
3691 	test_vmcs_addr_values(field_name, field, align, false,
3692 			      skip_beyond_mapped_ram, 0, 63);
3693 	report_prefix_pop();
3694 
3695 	report_prefix_pushf("%s disabled", control_name);
3696 	if (control_primary) {
3697 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3698 	} else {
3699 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3700 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3701 	}
3702 
3703 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3704 	report_prefix_pop();
3705 
3706 	vmcs_write(field, page_addr);
3707 	vmcs_write(CPU_EXEC_CTRL0, primary);
3708 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3709 }
3710 
3711 /*
3712  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3713  * each I/O-bitmap address must be 0. Neither address should set any
3714  * bits beyond the processor's physical-address width.
3715  * [Intel SDM]
3716  */
3717 static void test_io_bitmaps(void)
3718 {
3719 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3720 				 "I/O bitmap A", "Use I/O bitmaps",
3721 				 PAGE_SIZE, false, true);
3722 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3723 				 "I/O bitmap B", "Use I/O bitmaps",
3724 				 PAGE_SIZE, false, true);
3725 }
3726 
3727 /*
3728  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3729  * the MSR-bitmap address must be 0. The address should not set any
3730  * bits beyond the processor's physical-address width.
3731  * [Intel SDM]
3732  */
3733 static void test_msr_bitmap(void)
3734 {
3735 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3736 				 "MSR bitmap", "Use MSR bitmaps",
3737 				 PAGE_SIZE, false, true);
3738 }
3739 
3740 /*
3741  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3742  * address must satisfy the following checks:
3743  * - Bits 11:0 of the address must be 0.
3744  * - The address should not set any bits beyond the processor's
3745  *   physical-address width.
3746  * [Intel SDM]
3747  */
3748 static void test_apic_virt_addr(void)
3749 {
3750 	/*
3751 	 * Ensure the processor will never use the virtual-APIC page, since
3752 	 * we will point it to invalid RAM.  Otherwise KVM is puzzled about
3753 	 * what we're trying to achieve and fails vmentry.
3754 	 */
3755 	u32 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
3756 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0 | CPU_CR8_LOAD | CPU_CR8_STORE);
3757 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3758 				 "virtual-APIC address", "Use TPR shadow",
3759 				 PAGE_SIZE, false, true);
3760 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
3761 }
3762 
3763 /*
3764  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3765  * APIC-access address must satisfy the following checks:
3766  *  - Bits 11:0 of the address must be 0.
3767  *  - The address should not set any bits beyond the processor's
3768  *    physical-address width.
3769  * [Intel SDM]
3770  */
3771 static void test_apic_access_addr(void)
3772 {
3773 	void *apic_access_page = alloc_page();
3774 
3775 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3776 
3777 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3778 				 "APIC-access address",
3779 				 "virtualize APIC-accesses", PAGE_SIZE,
3780 				 true, false);
3781 }
3782 
3783 static bool set_bit_pattern(u8 mask, u32 *secondary)
3784 {
3785 	u8 i;
3786 	bool flag = false;
3787 	u32 test_bits[3] = {
3788 		CPU_VIRT_X2APIC,
3789 		CPU_APIC_REG_VIRT,
3790 		CPU_VINTD
3791 	};
3792 
3793         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3794 		if ((mask & (1u << i)) &&
3795 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3796 			*secondary |= test_bits[i];
3797 			flag = true;
3798 		}
3799 	}
3800 
3801 	return (flag);
3802 }
3803 
3804 /*
3805  * If the "use TPR shadow" VM-execution control is 0, the following
3806  * VM-execution controls must also be 0:
3807  * 	- virtualize x2APIC mode
3808  *	- APIC-register virtualization
3809  *	- virtual-interrupt delivery
3810  *    [Intel SDM]
3811  *
3812  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3813  *    "virtualize APIC accesses" VM-execution control must be 0.
3814  *    [Intel SDM]
3815  */
3816 static void test_apic_virtual_ctls(void)
3817 {
3818 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3819 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3820 	u32 primary = saved_primary;
3821 	u32 secondary = saved_secondary;
3822 	bool is_ctrl_valid = false;
3823 	char str[10] = "disabled";
3824 	u8 i = 0, j;
3825 
3826 	/*
3827 	 * First test
3828 	 */
3829 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3830 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3831 		return;
3832 
3833 	primary |= CPU_SECONDARY;
3834 	primary &= ~CPU_TPR_SHADOW;
3835 	vmcs_write(CPU_EXEC_CTRL0, primary);
3836 
3837 	while (1) {
3838 		for (j = 1; j < 8; j++) {
3839 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3840 			if (primary & CPU_TPR_SHADOW) {
3841 				is_ctrl_valid = true;
3842 			} else {
3843 				if (! set_bit_pattern(j, &secondary))
3844 					is_ctrl_valid = true;
3845 				else
3846 					is_ctrl_valid = false;
3847 			}
3848 
3849 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3850 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3851 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3852 			if (is_ctrl_valid)
3853 				test_vmx_valid_controls();
3854 			else
3855 				test_vmx_invalid_controls();
3856 			report_prefix_pop();
3857 		}
3858 
3859 		if (i == 1)
3860 			break;
3861 		i++;
3862 
3863 		primary |= CPU_TPR_SHADOW;
3864 		vmcs_write(CPU_EXEC_CTRL0, primary);
3865 		strcpy(str, "enabled");
3866 	}
3867 
3868 	/*
3869 	 * Second test
3870 	 */
3871 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3872 
3873 	primary = saved_primary;
3874 	secondary = saved_secondary;
3875 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3876 		return;
3877 
3878 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3879 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3880 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3881 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3882 	test_vmx_valid_controls();
3883 	report_prefix_pop();
3884 
3885 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3886 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3887 	test_vmx_valid_controls();
3888 	report_prefix_pop();
3889 
3890 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3891 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3892 	test_vmx_invalid_controls();
3893 	report_prefix_pop();
3894 
3895 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3896 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3897 	test_vmx_valid_controls();
3898 	report_prefix_pop();
3899 
3900 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3901 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3902 }
3903 
3904 /*
3905  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3906  * "external-interrupt exiting" VM-execution control must be 1.
3907  * [Intel SDM]
3908  */
3909 static void test_virtual_intr_ctls(void)
3910 {
3911 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3912 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3913 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3914 	u32 primary = saved_primary;
3915 	u32 secondary = saved_secondary;
3916 	u32 pin = saved_pin;
3917 
3918 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3919 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
3920 		return;
3921 
3922 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3923 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
3924 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3925 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
3926 	test_vmx_valid_controls();
3927 	report_prefix_pop();
3928 
3929 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
3930 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3931 	test_vmx_invalid_controls();
3932 	report_prefix_pop();
3933 
3934 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
3935 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
3936 	test_vmx_valid_controls();
3937 	report_prefix_pop();
3938 
3939 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3940 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3941 	test_vmx_invalid_controls();
3942 	report_prefix_pop();
3943 
3944 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3945 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3946 	vmcs_write(PIN_CONTROLS, saved_pin);
3947 }
3948 
3949 static void test_pi_desc_addr(u64 addr, bool is_ctrl_valid)
3950 {
3951 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
3952 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
3953 	if (is_ctrl_valid)
3954 		test_vmx_valid_controls();
3955 	else
3956 		test_vmx_invalid_controls();
3957 	report_prefix_pop();
3958 }
3959 
3960 /*
3961  * If the "process posted interrupts" VM-execution control is 1, the
3962  * following must be true:
3963  *
3964  *	- The "virtual-interrupt delivery" VM-execution control is 1.
3965  *	- The "acknowledge interrupt on exit" VM-exit control is 1.
3966  *	- The posted-interrupt notification vector has a value in the
3967  *	- range 0 - 255 (bits 15:8 are all 0).
3968  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
3969  *	- The posted-interrupt descriptor address does not set any bits
3970  *	  beyond the processor's physical-address width.
3971  * [Intel SDM]
3972  */
3973 static void test_posted_intr(void)
3974 {
3975 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3976 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3977 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3978 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
3979 	u32 primary = saved_primary;
3980 	u32 secondary = saved_secondary;
3981 	u32 pin = saved_pin;
3982 	u32 exit_ctl = exit_ctl_saved;
3983 	u16 vec;
3984 	int i;
3985 
3986 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
3987 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3988 	    (ctrl_exit_rev.clr & EXI_INTA)))
3989 		return;
3990 
3991 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3992 
3993 	/*
3994 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
3995 	 */
3996 	pin |= PIN_POST_INTR;
3997 	vmcs_write(PIN_CONTROLS, pin);
3998 	secondary &= ~CPU_VINTD;
3999 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4000 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
4001 	test_vmx_invalid_controls();
4002 	report_prefix_pop();
4003 
4004 	secondary |= CPU_VINTD;
4005 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4006 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
4007 	test_vmx_invalid_controls();
4008 	report_prefix_pop();
4009 
4010 	exit_ctl &= ~EXI_INTA;
4011 	vmcs_write(EXI_CONTROLS, exit_ctl);
4012 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
4013 	test_vmx_invalid_controls();
4014 	report_prefix_pop();
4015 
4016 	exit_ctl |= EXI_INTA;
4017 	vmcs_write(EXI_CONTROLS, exit_ctl);
4018 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4019 	test_vmx_valid_controls();
4020 	report_prefix_pop();
4021 
4022 	secondary &= ~CPU_VINTD;
4023 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4024 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
4025 	test_vmx_invalid_controls();
4026 	report_prefix_pop();
4027 
4028 	secondary |= CPU_VINTD;
4029 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4030 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4031 	test_vmx_valid_controls();
4032 	report_prefix_pop();
4033 
4034 	/*
4035 	 * Test posted-interrupt notification vector
4036 	 */
4037 	for (i = 0; i < 8; i++) {
4038 		vec = (1ul << i);
4039 		vmcs_write(PINV, vec);
4040 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4041 		test_vmx_valid_controls();
4042 		report_prefix_pop();
4043 	}
4044 	for (i = 8; i < 16; i++) {
4045 		vec = (1ul << i);
4046 		vmcs_write(PINV, vec);
4047 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4048 		test_vmx_invalid_controls();
4049 		report_prefix_pop();
4050 	}
4051 
4052 	vec &= ~(0xff << 8);
4053 	vmcs_write(PINV, vec);
4054 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4055 	test_vmx_valid_controls();
4056 	report_prefix_pop();
4057 
4058 	/*
4059 	 * Test posted-interrupt descriptor addresss
4060 	 */
4061 	for (i = 0; i < 6; i++) {
4062 		test_pi_desc_addr(1ul << i, false);
4063 	}
4064 
4065 	test_pi_desc_addr(0xf0, false);
4066 	test_pi_desc_addr(0xff, false);
4067 	test_pi_desc_addr(0x0f, false);
4068 	test_pi_desc_addr(0x8000, true);
4069 	test_pi_desc_addr(0x00, true);
4070 	test_pi_desc_addr(0xc000, true);
4071 
4072 	test_vmcs_addr_values("process-posted interrupts",
4073 			       POSTED_INTR_DESC_ADDR, 64,
4074 			       false, false, 0, 63);
4075 
4076 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4077 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4078 	vmcs_write(PIN_CONTROLS, saved_pin);
4079 }
4080 
4081 static void test_apic_ctls(void)
4082 {
4083 	test_apic_virt_addr();
4084 	test_apic_access_addr();
4085 	test_apic_virtual_ctls();
4086 	test_virtual_intr_ctls();
4087 	test_posted_intr();
4088 }
4089 
4090 /*
4091  * If the "enable VPID" VM-execution control is 1, the value of the
4092  * of the VPID VM-execution control field must not be 0000H.
4093  * [Intel SDM]
4094  */
4095 static void test_vpid(void)
4096 {
4097 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4098 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4099 	u16 vpid = 0x0000;
4100 	int i;
4101 
4102 	if (!is_vpid_supported()) {
4103 		printf("Secondary controls and/or VPID not supported\n");
4104 		return;
4105 	}
4106 
4107 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
4108 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
4109 	vmcs_write(VPID, vpid);
4110 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
4111 	test_vmx_valid_controls();
4112 	report_prefix_pop();
4113 
4114 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
4115 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4116 	test_vmx_invalid_controls();
4117 	report_prefix_pop();
4118 
4119 	for (i = 0; i < 16; i++) {
4120 		vpid = (short)1 << i;;
4121 		vmcs_write(VPID, vpid);
4122 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4123 		test_vmx_valid_controls();
4124 		report_prefix_pop();
4125 	}
4126 
4127 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4128 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4129 }
4130 
4131 static void set_vtpr(unsigned vtpr)
4132 {
4133 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
4134 }
4135 
4136 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
4137 {
4138 	bool valid = true;
4139 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4140 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4141 
4142 	if ((primary & CPU_TPR_SHADOW) &&
4143 	    (!(primary & CPU_SECONDARY) ||
4144 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
4145 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
4146 
4147 	set_vtpr(vtpr);
4148 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
4149 	    threshold, (vtpr >> 4) & 0xf);
4150 	if (valid)
4151 		test_vmx_valid_controls();
4152 	else
4153 		test_vmx_invalid_controls();
4154 	report_prefix_pop();
4155 }
4156 
4157 static void test_invalid_event_injection(void)
4158 {
4159 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
4160 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
4161 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
4162 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
4163 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
4164 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4165 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4166 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4167 	u32 cnt;
4168 
4169 	/* Setup */
4170 	report_prefix_push("invalid event injection");
4171 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4172 	vmcs_write(ENT_INST_LEN, 0x00000001);
4173 
4174 	/* The field's interruption type is not set to a reserved value. */
4175 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4176 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4177 			    "RESERVED interruption type invalid [-]",
4178 			    ent_intr_info);
4179 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4180 	test_vmx_invalid_controls();
4181 	report_prefix_pop();
4182 
4183 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4184 			DE_VECTOR;
4185 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4186 			    "RESERVED interruption type invalid [+]",
4187 			    ent_intr_info);
4188 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4189 	test_vmx_valid_controls();
4190 	report_prefix_pop();
4191 
4192 	/* If the interruption type is other event, the vector is 0. */
4193 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4194 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4195 			    "(OTHER EVENT && vector != 0) invalid [-]",
4196 			    ent_intr_info);
4197 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4198 	test_vmx_invalid_controls();
4199 	report_prefix_pop();
4200 
4201 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4202 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4203 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4204 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4205 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4206 	test_vmx_invalid_controls();
4207 	report_prefix_pop();
4208 
4209 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4210 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4211 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4212 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4213 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4214 	test_vmx_valid_controls();
4215 	report_prefix_pop();
4216 
4217 	/*
4218 	 * If the interruption type
4219 	 * is HW exception, the vector is at most 31.
4220 	 */
4221 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4222 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4223 			    "(HW exception && vector > 31) invalid [-]",
4224 			    ent_intr_info);
4225 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4226 	test_vmx_invalid_controls();
4227 	report_prefix_pop();
4228 
4229 	/*
4230 	 * deliver-error-code is 1 iff either
4231 	 * (a) the "unrestricted guest" VM-execution control is 0
4232 	 * (b) CR0.PE is set.
4233 	 */
4234 
4235 	/* Assert that unrestricted guest is disabled or unsupported */
4236 	assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
4237 	       !(secondary_save & CPU_URG));
4238 
4239 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4240 			GP_VECTOR;
4241 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4242 			    "error code <-> (!URG || prot_mode) [-]",
4243 			    ent_intr_info);
4244 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4245 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4246 	test_vmx_invalid_controls();
4247 	report_prefix_pop();
4248 
4249 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4250 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4251 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4252 			    "error code <-> (!URG || prot_mode) [+]",
4253 			    ent_intr_info);
4254 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4255 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4256 	test_vmx_valid_controls();
4257 	report_prefix_pop();
4258 
4259 	if (enable_unrestricted_guest(false))
4260 		goto skip_unrestricted_guest;
4261 
4262 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4263 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4264 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4265 			    "error code <-> (!URG || prot_mode) [-]",
4266 			    ent_intr_info);
4267 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4268 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4269 	test_vmx_invalid_controls();
4270 	report_prefix_pop();
4271 
4272 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4273 			GP_VECTOR;
4274 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4275 			    "error code <-> (!URG || prot_mode) [-]",
4276 			    ent_intr_info);
4277 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4278 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4279 	test_vmx_invalid_controls();
4280 	report_prefix_pop();
4281 
4282 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4283 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4284 
4285 skip_unrestricted_guest:
4286 	vmcs_write(GUEST_CR0, guest_cr0_save);
4287 
4288 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4289 	report_prefix_push("error code <-> HW exception");
4290 	for (cnt = 0; cnt < 8; cnt++) {
4291 		u32 exception_type_mask = cnt << 8;
4292 		u32 deliver_error_code_mask =
4293 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4294 			INTR_INFO_DELIVER_CODE_MASK : 0;
4295 
4296 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4297 				exception_type_mask | GP_VECTOR;
4298 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4299 				    ent_intr_info);
4300 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4301 		test_vmx_invalid_controls();
4302 		report_prefix_pop();
4303 	}
4304 	report_prefix_pop();
4305 
4306 	/*
4307 	 * deliver-error-code is 1 iff the the vector
4308 	 * indicates an exception that would normally deliver an error code
4309 	 */
4310 	report_prefix_push("error code <-> vector delivers error code");
4311 	for (cnt = 0; cnt < 32; cnt++) {
4312 		bool has_error_code = false;
4313 		u32 deliver_error_code_mask;
4314 
4315 		switch (cnt) {
4316 		case DF_VECTOR:
4317 		case TS_VECTOR:
4318 		case NP_VECTOR:
4319 		case SS_VECTOR:
4320 		case GP_VECTOR:
4321 		case PF_VECTOR:
4322 		case AC_VECTOR:
4323 			has_error_code = true;
4324 		case CP_VECTOR:
4325 			/* Some CPUs have error code and some do not, skip */
4326 			continue;
4327 		}
4328 
4329 		/* Negative case */
4330 		deliver_error_code_mask = has_error_code ?
4331 						0 :
4332 						INTR_INFO_DELIVER_CODE_MASK;
4333 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4334 				INTR_TYPE_HARD_EXCEPTION | cnt;
4335 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4336 				    ent_intr_info);
4337 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4338 		test_vmx_invalid_controls();
4339 		report_prefix_pop();
4340 
4341 		/* Positive case */
4342 		deliver_error_code_mask = has_error_code ?
4343 						INTR_INFO_DELIVER_CODE_MASK :
4344 						0;
4345 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4346 				INTR_TYPE_HARD_EXCEPTION | cnt;
4347 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4348 				    ent_intr_info);
4349 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4350 		test_vmx_valid_controls();
4351 		report_prefix_pop();
4352 	}
4353 	report_prefix_pop();
4354 
4355 	/* Reserved bits in the field (30:12) are 0. */
4356 	report_prefix_push("reserved bits clear");
4357 	for (cnt = 12; cnt <= 30; cnt++) {
4358 		ent_intr_info = ent_intr_info_base |
4359 				INTR_INFO_DELIVER_CODE_MASK |
4360 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4361 				(1U << cnt);
4362 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4363 				    ent_intr_info);
4364 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4365 		test_vmx_invalid_controls();
4366 		report_prefix_pop();
4367 	}
4368 	report_prefix_pop();
4369 
4370 	/*
4371 	 * If deliver-error-code is 1
4372 	 * bits 31:16 of the VM-entry exception error-code field are 0.
4373 	 */
4374 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4375 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4376 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4377 			    "VM-entry exception error code[31:16] clear",
4378 			    ent_intr_info);
4379 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4380 	for (cnt = 16; cnt <= 31; cnt++) {
4381 		ent_intr_err = 1U << cnt;
4382 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4383 				    ent_intr_err);
4384 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4385 		test_vmx_invalid_controls();
4386 		report_prefix_pop();
4387 	}
4388 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4389 	report_prefix_pop();
4390 
4391 	/*
4392 	 * If the interruption type is software interrupt, software exception,
4393 	 * or privileged software exception, the VM-entry instruction-length
4394 	 * field is in the range 0 - 15.
4395 	 */
4396 
4397 	for (cnt = 0; cnt < 3; cnt++) {
4398 		switch (cnt) {
4399 		case 0:
4400 			ent_intr_info = ent_intr_info_base |
4401 					INTR_TYPE_SOFT_INTR;
4402 			break;
4403 		case 1:
4404 			ent_intr_info = ent_intr_info_base |
4405 					INTR_TYPE_SOFT_EXCEPTION;
4406 			break;
4407 		case 2:
4408 			ent_intr_info = ent_intr_info_base |
4409 					INTR_TYPE_PRIV_SW_EXCEPTION;
4410 			break;
4411 		}
4412 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4413 				    "VM-entry instruction-length check",
4414 				    ent_intr_info);
4415 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4416 
4417 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4418 		ent_intr_len = -1;
4419 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4420 				    ent_intr_len);
4421 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4422 		test_vmx_invalid_controls();
4423 		report_prefix_pop();
4424 
4425 		/* Instruction length set to 16 should fail */
4426 		ent_intr_len = 0x00000010;
4427 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4428 				    ent_intr_len);
4429 		vmcs_write(ENT_INST_LEN, 0x00000010);
4430 		test_vmx_invalid_controls();
4431 		report_prefix_pop();
4432 
4433 		report_prefix_pop();
4434 	}
4435 
4436 	/* Cleanup */
4437 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4438 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4439 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4440 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4441 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4442 	vmcs_write(GUEST_CR0, guest_cr0_save);
4443 	report_prefix_pop();
4444 }
4445 
4446 /*
4447  * Test interesting vTPR values for a given TPR threshold.
4448  */
4449 static void test_vtpr_values(unsigned threshold)
4450 {
4451 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4452 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4453 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4454 }
4455 
4456 static void try_tpr_threshold(unsigned threshold)
4457 {
4458 	bool valid = true;
4459 
4460 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4461 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4462 
4463 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4464 	    (secondary & CPU_VINTD)))
4465 		valid = !(threshold >> 4);
4466 
4467 	set_vtpr(-1);
4468 	vmcs_write(TPR_THRESHOLD, threshold);
4469 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4470 	if (valid)
4471 		test_vmx_valid_controls();
4472 	else
4473 		test_vmx_invalid_controls();
4474 	report_prefix_pop();
4475 
4476 	if (valid)
4477 		test_vtpr_values(threshold);
4478 }
4479 
4480 /*
4481  * Test interesting TPR threshold values.
4482  */
4483 static void test_tpr_threshold_values(void)
4484 {
4485 	unsigned i;
4486 
4487 	for (i = 0; i < 0x10; i++)
4488 		try_tpr_threshold(i);
4489 	for (i = 4; i < 32; i++)
4490 		try_tpr_threshold(1u << i);
4491 	try_tpr_threshold(-1u);
4492 	try_tpr_threshold(0x7fffffff);
4493 }
4494 
4495 /*
4496  * This test covers the following two VM entry checks:
4497  *
4498  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4499  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4500  *         31:4 of the TPR threshold VM-execution control field must
4501 	   be 0.
4502  *         [Intel SDM]
4503  *
4504  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4505  *          "virtual-interrupt delivery" VM-execution control is 0
4506  *          and the "virtualize APIC accesses" VM-execution control
4507  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4508  *          control field must not be greater than the value of bits
4509  *          7:4 of VTPR.
4510  *          [Intel SDM]
4511  */
4512 static void test_tpr_threshold(void)
4513 {
4514 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4515 	u64 apic_virt_addr = vmcs_read(APIC_VIRT_ADDR);
4516 	u64 threshold = vmcs_read(TPR_THRESHOLD);
4517 	void *virtual_apic_page;
4518 
4519 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4520 		return;
4521 
4522 	virtual_apic_page = alloc_page();
4523 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4524 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4525 
4526 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4527 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4528 	test_tpr_threshold_values();
4529 	report_prefix_pop();
4530 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4531 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4532 	test_tpr_threshold_values();
4533 	report_prefix_pop();
4534 
4535 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4536 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES))))
4537 		goto out;
4538 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4539 
4540 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4541 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4542 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4543 		test_tpr_threshold_values();
4544 		report_prefix_pop();
4545 
4546 		vmcs_write(CPU_EXEC_CTRL0,
4547 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4548 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4549 		test_tpr_threshold_values();
4550 		report_prefix_pop();
4551 	}
4552 
4553 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4554 		vmcs_write(CPU_EXEC_CTRL0,
4555 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4556 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4557 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4558 		test_tpr_threshold_values();
4559 		report_prefix_pop();
4560 
4561 		vmcs_write(CPU_EXEC_CTRL0,
4562 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4563 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4564 		test_tpr_threshold_values();
4565 		report_prefix_pop();
4566 	}
4567 
4568 	if ((ctrl_cpu_rev[1].clr &
4569 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4570 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4571 		vmcs_write(CPU_EXEC_CTRL0,
4572 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4573 		vmcs_write(CPU_EXEC_CTRL1,
4574 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4575 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4576 		test_tpr_threshold_values();
4577 		report_prefix_pop();
4578 
4579 		vmcs_write(CPU_EXEC_CTRL0,
4580 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4581 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4582 		test_tpr_threshold_values();
4583 		report_prefix_pop();
4584 	}
4585 
4586 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4587 out:
4588 	vmcs_write(TPR_THRESHOLD, threshold);
4589 	vmcs_write(APIC_VIRT_ADDR, apic_virt_addr);
4590 	vmcs_write(CPU_EXEC_CTRL0, primary);
4591 }
4592 
4593 /*
4594  * This test verifies the following two vmentry checks:
4595  *
4596  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4597  *  VM-execution control must be 0.
4598  *  [Intel SDM]
4599  *
4600  *  If the "virtual NMIs" VM-execution control is 0, the "NMI-window
4601  *  exiting" VM-execution control must be 0.
4602  *  [Intel SDM]
4603  */
4604 static void test_nmi_ctrls(void)
4605 {
4606 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4607 
4608 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4609 	    (PIN_NMI | PIN_VIRT_NMI)) {
4610 		printf("NMI exiting and Virtual NMIs are not supported !\n");
4611 		return;
4612 	}
4613 
4614 	/* Save the controls so that we can restore them after our tests */
4615 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4616 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4617 
4618 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4619 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4620 
4621 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4622 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4623 	test_vmx_valid_controls();
4624 	report_prefix_pop();
4625 
4626 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4627 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4628 	test_vmx_invalid_controls();
4629 	report_prefix_pop();
4630 
4631 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4632 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4633 	test_vmx_valid_controls();
4634 	report_prefix_pop();
4635 
4636 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4637 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4638 	test_vmx_valid_controls();
4639 	report_prefix_pop();
4640 
4641 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4642 		report_info("NMI-window exiting is not supported, skipping...");
4643 		goto done;
4644 	}
4645 
4646 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4647 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4648 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4649 	test_vmx_invalid_controls();
4650 	report_prefix_pop();
4651 
4652 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4653 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4654 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4655 	test_vmx_valid_controls();
4656 	report_prefix_pop();
4657 
4658 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4659 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4660 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4661 	test_vmx_valid_controls();
4662 	report_prefix_pop();
4663 
4664 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4665 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4666 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4667 	test_vmx_valid_controls();
4668 	report_prefix_pop();
4669 
4670 	/* Restore the controls to their original values */
4671 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4672 done:
4673 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4674 }
4675 
4676 static void test_eptp_ad_bit(u64 eptp, bool is_ctrl_valid)
4677 {
4678 	vmcs_write(EPTP, eptp);
4679 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4680 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4681 	if (is_ctrl_valid)
4682 		test_vmx_valid_controls();
4683 	else
4684 		test_vmx_invalid_controls();
4685 	report_prefix_pop();
4686 
4687 }
4688 
4689 /*
4690  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4691  *    control field must satisfy the following checks:
4692  *
4693  *     - The EPT memory type (bits 2:0) must be a value supported by the
4694  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4695  *     - Bits 5:3 (1 less than the EPT page-walk length) must indicate a
4696  *	 supported EPT page-walk length.
4697  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4698  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4699  *	 indicating that the processor does not support accessed and dirty
4700  *	 dirty flags for EPT.
4701  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4702  *	 physical-address width) must all be 0.
4703  *
4704  * 2. If the "unrestricted guest" VM-execution control is 1, the
4705  *    "enable EPT" VM-execution control must also be 1.
4706  */
4707 static void test_ept_eptp(void)
4708 {
4709 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4710 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4711 	u64 eptp_saved = vmcs_read(EPTP);
4712 	u32 primary = primary_saved;
4713 	u32 secondary = secondary_saved;
4714 	u64 eptp = eptp_saved;
4715 	u32 i, maxphysaddr;
4716 	u64 j, resv_bits_mask = 0;
4717 
4718 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4719 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4720 		printf("\"CPU secondary\" and/or \"enable EPT\" execution controls are not supported !\n");
4721 		return;
4722 	}
4723 
4724 	/* Support for 4-level EPT is mandatory. */
4725 	report(is_4_level_ept_supported(), "4-level EPT support check");
4726 
4727 	primary |= CPU_SECONDARY;
4728 	vmcs_write(CPU_EXEC_CTRL0, primary);
4729 	secondary |= CPU_EPT;
4730 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4731 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4732 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4733 	vmcs_write(EPTP, eptp);
4734 
4735 	for (i = 0; i < 8; i++) {
4736 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4737 		vmcs_write(EPTP, eptp);
4738 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4739 		    eptp & EPT_MEM_TYPE_MASK);
4740 		if (is_ept_memtype_supported(i))
4741 			test_vmx_valid_controls();
4742 		else
4743 			test_vmx_invalid_controls();
4744 		report_prefix_pop();
4745 	}
4746 
4747 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4748 
4749 	/*
4750 	 * Page walk length (bits 5:3).  Note, the value in VMCS.EPTP "is 1
4751 	 * less than the EPT page-walk length".
4752 	 */
4753 	for (i = 0; i < 8; i++) {
4754 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4755 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4756 
4757 		vmcs_write(EPTP, eptp);
4758 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4759 		    eptp & EPTP_PG_WALK_LEN_MASK);
4760 		if (i == 3 || (i == 4 && is_5_level_ept_supported()))
4761 			test_vmx_valid_controls();
4762 		else
4763 			test_vmx_invalid_controls();
4764 		report_prefix_pop();
4765 	}
4766 
4767 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4768 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4769 
4770 	/*
4771 	 * Accessed and dirty flag (bit 6)
4772 	 */
4773 	if (ept_ad_bits_supported()) {
4774 		report_info("Processor supports accessed and dirty flag");
4775 		eptp &= ~EPTP_AD_FLAG;
4776 		test_eptp_ad_bit(eptp, true);
4777 
4778 		eptp |= EPTP_AD_FLAG;
4779 		test_eptp_ad_bit(eptp, true);
4780 	} else {
4781 		report_info("Processor does not supports accessed and dirty flag");
4782 		eptp &= ~EPTP_AD_FLAG;
4783 		test_eptp_ad_bit(eptp, true);
4784 
4785 		eptp |= EPTP_AD_FLAG;
4786 		test_eptp_ad_bit(eptp, false);
4787 	}
4788 
4789 	/*
4790 	 * Reserved bits [11:7] and [63:N]
4791 	 */
4792 	for (i = 0; i < 32; i++) {
4793 		eptp = (eptp &
4794 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4795 		    (i << EPTP_RESERV_BITS_SHIFT);
4796 		vmcs_write(EPTP, eptp);
4797 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4798 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4799 		    EPTP_RESERV_BITS_MASK);
4800 		if (i == 0)
4801 			test_vmx_valid_controls();
4802 		else
4803 			test_vmx_invalid_controls();
4804 		report_prefix_pop();
4805 	}
4806 
4807 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4808 
4809 	maxphysaddr = cpuid_maxphyaddr();
4810 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4811 		resv_bits_mask |= 1ul << i;
4812 	}
4813 
4814 	for (j = maxphysaddr - 1; j <= 63; j++) {
4815 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4816 		    (j < maxphysaddr ? 0 : 1ul << j);
4817 		vmcs_write(EPTP, eptp);
4818 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4819 		    (eptp >> maxphysaddr) & resv_bits_mask);
4820 		if (j < maxphysaddr)
4821 			test_vmx_valid_controls();
4822 		else
4823 			test_vmx_invalid_controls();
4824 		report_prefix_pop();
4825 	}
4826 
4827 	secondary &= ~(CPU_EPT | CPU_URG);
4828 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4829 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4830 	test_vmx_valid_controls();
4831 	report_prefix_pop();
4832 
4833 	if (!(ctrl_cpu_rev[1].clr & CPU_URG))
4834 		goto skip_unrestricted_guest;
4835 
4836 	secondary |= CPU_URG;
4837 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4838 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4839 	test_vmx_invalid_controls();
4840 	report_prefix_pop();
4841 
4842 	secondary |= CPU_EPT;
4843 	setup_dummy_ept();
4844 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4845 	test_vmx_valid_controls();
4846 	report_prefix_pop();
4847 
4848 skip_unrestricted_guest:
4849 	secondary &= ~CPU_URG;
4850 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4851 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4852 	test_vmx_valid_controls();
4853 	report_prefix_pop();
4854 
4855 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4856 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4857 	vmcs_write(EPTP, eptp_saved);
4858 }
4859 
4860 /*
4861  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4862  * VM-execution control must also be 1. In addition, the PML address
4863  * must satisfy the following checks:
4864  *
4865  *    * Bits 11:0 of the address must be 0.
4866  *    * The address should not set any bits beyond the processor's
4867  *	physical-address width.
4868  *
4869  *  [Intel SDM]
4870  */
4871 static void test_pml(void)
4872 {
4873 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4874 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4875 	u32 primary = primary_saved;
4876 	u32 secondary = secondary_saved;
4877 
4878 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4879 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
4880 		printf("\"Secondary execution\" control or \"enable EPT\" control or \"enable PML\" control is not supported !\n");
4881 		return;
4882 	}
4883 
4884 	primary |= CPU_SECONDARY;
4885 	vmcs_write(CPU_EXEC_CTRL0, primary);
4886 	secondary &= ~(CPU_PML | CPU_EPT);
4887 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4888 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
4889 	test_vmx_valid_controls();
4890 	report_prefix_pop();
4891 
4892 	secondary |= CPU_PML;
4893 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4894 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
4895 	test_vmx_invalid_controls();
4896 	report_prefix_pop();
4897 
4898 	secondary |= CPU_EPT;
4899 	setup_dummy_ept();
4900 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
4901 	test_vmx_valid_controls();
4902 	report_prefix_pop();
4903 
4904 	secondary &= ~CPU_PML;
4905 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4906 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
4907 	test_vmx_valid_controls();
4908 	report_prefix_pop();
4909 
4910 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
4911 				 PAGE_SIZE, false, false);
4912 
4913 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4914 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4915 }
4916 
4917  /*
4918  * If the "activate VMX-preemption timer" VM-execution control is 0, the
4919  * the "save VMX-preemption timer value" VM-exit control must also be 0.
4920  *
4921  *  [Intel SDM]
4922  */
4923 static void test_vmx_preemption_timer(void)
4924 {
4925 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4926 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
4927 	u32 pin = saved_pin;
4928 	u32 exit = saved_exit;
4929 
4930 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
4931 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
4932 		printf("\"Save-VMX-preemption-timer\" control and/or \"Enable-VMX-preemption-timer\" control is not supported\n");
4933 		return;
4934 	}
4935 
4936 	pin |= PIN_PREEMPT;
4937 	vmcs_write(PIN_CONTROLS, pin);
4938 	exit &= ~EXI_SAVE_PREEMPT;
4939 	vmcs_write(EXI_CONTROLS, exit);
4940 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
4941 	test_vmx_valid_controls();
4942 	report_prefix_pop();
4943 
4944 	exit |= EXI_SAVE_PREEMPT;
4945 	vmcs_write(EXI_CONTROLS, exit);
4946 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
4947 	test_vmx_valid_controls();
4948 	report_prefix_pop();
4949 
4950 	pin &= ~PIN_PREEMPT;
4951 	vmcs_write(PIN_CONTROLS, pin);
4952 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
4953 	test_vmx_invalid_controls();
4954 	report_prefix_pop();
4955 
4956 	exit &= ~EXI_SAVE_PREEMPT;
4957 	vmcs_write(EXI_CONTROLS, exit);
4958 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
4959 	test_vmx_valid_controls();
4960 	report_prefix_pop();
4961 
4962 	vmcs_write(PIN_CONTROLS, saved_pin);
4963 	vmcs_write(EXI_CONTROLS, saved_exit);
4964 }
4965 
4966 extern unsigned char test_mtf1;
4967 extern unsigned char test_mtf2;
4968 extern unsigned char test_mtf3;
4969 extern unsigned char test_mtf4;
4970 
4971 static void test_mtf_guest(void)
4972 {
4973 	asm ("vmcall;\n\t"
4974 	     "out %al, $0x80;\n\t"
4975 	     "test_mtf1:\n\t"
4976 	     "vmcall;\n\t"
4977 	     "out %al, $0x80;\n\t"
4978 	     "test_mtf2:\n\t"
4979 	     /*
4980 	      * Prepare for the 'MOV CR3' test. Attempt to induce a
4981 	      * general-protection fault by moving a non-canonical address into
4982 	      * CR3. The 'MOV CR3' instruction does not take an imm64 operand,
4983 	      * so we must MOV the desired value into a register first.
4984 	      *
4985 	      * MOV RAX is done before the VMCALL such that MTF is only enabled
4986 	      * for the instruction under test.
4987 	      */
4988 	     "mov $0xaaaaaaaaaaaaaaaa, %rax;\n\t"
4989 	     "vmcall;\n\t"
4990 	     "mov %rax, %cr3;\n\t"
4991 	     "test_mtf3:\n\t"
4992 	     "vmcall;\n\t"
4993 	     /*
4994 	      * ICEBP/INT1 instruction. Though the instruction is now
4995 	      * documented, don't rely on assemblers enumerating the
4996 	      * instruction. Resort to hand assembly.
4997 	      */
4998 	     ".byte 0xf1;\n\t"
4999 	     "vmcall;\n\t"
5000 	     "test_mtf4:\n\t"
5001 	     "mov $0, %eax;\n\t");
5002 }
5003 
5004 static void test_mtf_gp_handler(struct ex_regs *regs)
5005 {
5006 	regs->rip = (unsigned long) &test_mtf3;
5007 }
5008 
5009 static void test_mtf_db_handler(struct ex_regs *regs)
5010 {
5011 }
5012 
5013 static void enable_mtf(void)
5014 {
5015 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5016 
5017 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 | CPU_MTF);
5018 }
5019 
5020 static void disable_mtf(void)
5021 {
5022 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5023 
5024 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 & ~CPU_MTF);
5025 }
5026 
5027 static void enable_tf(void)
5028 {
5029 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5030 
5031 	vmcs_write(GUEST_RFLAGS, rflags | X86_EFLAGS_TF);
5032 }
5033 
5034 static void disable_tf(void)
5035 {
5036 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5037 
5038 	vmcs_write(GUEST_RFLAGS, rflags & ~X86_EFLAGS_TF);
5039 }
5040 
5041 static void report_mtf(const char *insn_name, unsigned long exp_rip)
5042 {
5043 	unsigned long rip = vmcs_read(GUEST_RIP);
5044 
5045 	assert_exit_reason(VMX_MTF);
5046 	report(rip == exp_rip, "MTF VM-exit after %s. RIP: 0x%lx (expected 0x%lx)",
5047 	       insn_name, rip, exp_rip);
5048 }
5049 
5050 static void vmx_mtf_test(void)
5051 {
5052 	unsigned long pending_dbg;
5053 	handler old_gp, old_db;
5054 
5055 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5056 		printf("CPU does not support the 'monitor trap flag' processor-based VM-execution control.\n");
5057 		return;
5058 	}
5059 
5060 	test_set_guest(test_mtf_guest);
5061 
5062 	/* Expect an MTF VM-exit after OUT instruction */
5063 	enter_guest();
5064 	skip_exit_vmcall();
5065 
5066 	enable_mtf();
5067 	enter_guest();
5068 	report_mtf("OUT", (unsigned long) &test_mtf1);
5069 	disable_mtf();
5070 
5071 	/*
5072 	 * Concurrent #DB trap and MTF on instruction boundary. Expect MTF
5073 	 * VM-exit with populated 'pending debug exceptions' VMCS field.
5074 	 */
5075 	enter_guest();
5076 	skip_exit_vmcall();
5077 
5078 	enable_mtf();
5079 	enable_tf();
5080 
5081 	enter_guest();
5082 	report_mtf("OUT", (unsigned long) &test_mtf2);
5083 	pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
5084 	report(pending_dbg & DR_STEP,
5085 	       "'pending debug exceptions' field after MTF VM-exit: 0x%lx (expected 0x%lx)",
5086 	       pending_dbg, (unsigned long) DR_STEP);
5087 
5088 	disable_mtf();
5089 	disable_tf();
5090 	vmcs_write(GUEST_PENDING_DEBUG, 0);
5091 
5092 	/*
5093 	 * #GP exception takes priority over MTF. Expect MTF VM-exit with RIP
5094 	 * advanced to first instruction of #GP handler.
5095 	 */
5096 	enter_guest();
5097 	skip_exit_vmcall();
5098 
5099 	old_gp = handle_exception(GP_VECTOR, test_mtf_gp_handler);
5100 
5101 	enable_mtf();
5102 	enter_guest();
5103 	report_mtf("MOV CR3", (unsigned long) get_idt_addr(&boot_idt[GP_VECTOR]));
5104 	disable_mtf();
5105 
5106 	/*
5107 	 * Concurrent MTF and privileged software exception (i.e. ICEBP/INT1).
5108 	 * MTF should follow the delivery of #DB trap, though the SDM doesn't
5109 	 * provide clear indication of the relative priority.
5110 	 */
5111 	enter_guest();
5112 	skip_exit_vmcall();
5113 
5114 	handle_exception(GP_VECTOR, old_gp);
5115 	old_db = handle_exception(DB_VECTOR, test_mtf_db_handler);
5116 
5117 	enable_mtf();
5118 	enter_guest();
5119 	report_mtf("INT1", (unsigned long) get_idt_addr(&boot_idt[DB_VECTOR]));
5120 	disable_mtf();
5121 
5122 	enter_guest();
5123 	skip_exit_vmcall();
5124 	handle_exception(DB_VECTOR, old_db);
5125 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK | INTR_TYPE_OTHER_EVENT);
5126 	enter_guest();
5127 	report_mtf("injected MTF", (unsigned long) &test_mtf4);
5128 	enter_guest();
5129 }
5130 
5131 extern char vmx_mtf_pdpte_guest_begin;
5132 extern char vmx_mtf_pdpte_guest_end;
5133 
5134 asm("vmx_mtf_pdpte_guest_begin:\n\t"
5135     "mov %cr0, %rax\n\t"    /* save CR0 with PG=1                 */
5136     "vmcall\n\t"            /* on return from this CR0.PG=0       */
5137     "mov %rax, %cr0\n\t"    /* restore CR0.PG=1 to enter PAE mode */
5138     "vmcall\n\t"
5139     "retq\n\t"
5140     "vmx_mtf_pdpte_guest_end:");
5141 
5142 static void vmx_mtf_pdpte_test(void)
5143 {
5144 	void *test_mtf_pdpte_guest;
5145 	pteval_t *pdpt;
5146 	u32 guest_ar_cs;
5147 	u64 guest_efer;
5148 	pteval_t *pte;
5149 	u64 guest_cr0;
5150 	u64 guest_cr3;
5151 	u64 guest_cr4;
5152 	u64 ent_ctls;
5153 	int i;
5154 
5155 	if (setup_ept(false))
5156 		return;
5157 
5158 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5159 		printf("CPU does not support 'monitor trap flag.'\n");
5160 		return;
5161 	}
5162 
5163 	if (!(ctrl_cpu_rev[1].clr & CPU_URG)) {
5164 		printf("CPU does not support 'unrestricted guest.'\n");
5165 		return;
5166 	}
5167 
5168 	vmcs_write(EXC_BITMAP, ~0);
5169 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
5170 
5171 	/*
5172 	 * Copy the guest code to an identity-mapped page.
5173 	 */
5174 	test_mtf_pdpte_guest = alloc_page();
5175 	memcpy(test_mtf_pdpte_guest, &vmx_mtf_pdpte_guest_begin,
5176 	       &vmx_mtf_pdpte_guest_end - &vmx_mtf_pdpte_guest_begin);
5177 
5178 	test_set_guest(test_mtf_pdpte_guest);
5179 
5180 	enter_guest();
5181 	skip_exit_vmcall();
5182 
5183 	/*
5184 	 * Put the guest in non-paged 32-bit protected mode, ready to enter
5185 	 * PAE mode when CR0.PG is set. CR4.PAE will already have been set
5186 	 * when the guest started out in long mode.
5187 	 */
5188 	ent_ctls = vmcs_read(ENT_CONTROLS);
5189 	vmcs_write(ENT_CONTROLS, ent_ctls & ~ENT_GUEST_64);
5190 
5191 	guest_efer = vmcs_read(GUEST_EFER);
5192 	vmcs_write(GUEST_EFER, guest_efer & ~(EFER_LMA | EFER_LME));
5193 
5194 	/*
5195 	 * Set CS access rights bits for 32-bit protected mode:
5196 	 * 3:0    B execute/read/accessed
5197 	 * 4      1 code or data
5198 	 * 6:5    0 descriptor privilege level
5199 	 * 7      1 present
5200 	 * 11:8   0 reserved
5201 	 * 12     0 available for use by system software
5202 	 * 13     0 64 bit mode not active
5203 	 * 14     1 default operation size 32-bit segment
5204 	 * 15     1 page granularity: segment limit in 4K units
5205 	 * 16     0 segment usable
5206 	 * 31:17  0 reserved
5207 	 */
5208 	guest_ar_cs = vmcs_read(GUEST_AR_CS);
5209 	vmcs_write(GUEST_AR_CS, 0xc09b);
5210 
5211 	guest_cr0 = vmcs_read(GUEST_CR0);
5212 	vmcs_write(GUEST_CR0, guest_cr0 & ~X86_CR0_PG);
5213 
5214 	guest_cr4 = vmcs_read(GUEST_CR4);
5215 	vmcs_write(GUEST_CR4, guest_cr4 & ~X86_CR4_PCIDE);
5216 
5217 	guest_cr3 = vmcs_read(GUEST_CR3);
5218 
5219 	/*
5220 	 * Turn the 4-level page table into a PAE page table by following the 0th
5221 	 * PML4 entry to a PDPT page, and grab the first four PDPTEs from that
5222 	 * page.
5223 	 *
5224 	 * Why does this work?
5225 	 *
5226 	 * PAE uses 32-bit addressing which implies:
5227 	 * Bits 11:0   page offset
5228 	 * Bits 20:12  entry into 512-entry page table
5229 	 * Bits 29:21  entry into a 512-entry directory table
5230 	 * Bits 31:30  entry into the page directory pointer table.
5231 	 * Bits 63:32  zero
5232 	 *
5233 	 * As only 2 bits are needed to select the PDPTEs for the entire
5234 	 * 32-bit address space, take the first 4 PDPTEs in the level 3 page
5235 	 * directory pointer table. It doesn't matter which of these PDPTEs
5236 	 * are present because they must cover the guest code given that it
5237 	 * has already run successfully.
5238 	 *
5239 	 * Get a pointer to PTE for GVA=0 in the page directory pointer table
5240 	 */
5241 	pte = get_pte_level(
5242             (pgd_t *)phys_to_virt(guest_cr3 & ~X86_CR3_PCID_MASK), 0,
5243             PDPT_LEVEL);
5244 
5245 	/*
5246 	 * Need some memory for the 4-entry PAE page directory pointer
5247 	 * table. Use the end of the identity-mapped page where the guest code
5248 	 * is stored. There is definitely space as the guest code is only a
5249 	 * few bytes.
5250 	 */
5251 	pdpt = test_mtf_pdpte_guest + PAGE_SIZE - 4 * sizeof(pteval_t);
5252 
5253 	/*
5254 	 * Copy the first four PDPTEs into the PAE page table with reserved
5255 	 * bits cleared. Note that permission bits from the PML4E and PDPTE
5256 	 * are not propagated.
5257 	 */
5258 	for (i = 0; i < 4; i++) {
5259 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_RSVD_MASK),
5260 				   "PDPTE has invalid reserved bits");
5261 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_PAGE_SIZE_MASK),
5262 				   "Cannot use 1GB super pages for PAE");
5263 		pdpt[i] = pte[i] & ~(PAE_PDPTE_RSVD_MASK);
5264 	}
5265 	vmcs_write(GUEST_CR3, virt_to_phys(pdpt));
5266 
5267 	enable_mtf();
5268 	enter_guest();
5269 	assert_exit_reason(VMX_MTF);
5270 	disable_mtf();
5271 
5272 	/*
5273 	 * The four PDPTEs should have been loaded into the VMCS when
5274 	 * the guest set CR0.PG to enter PAE mode.
5275 	 */
5276 	for (i = 0; i < 4; i++) {
5277 		u64 pdpte = vmcs_read(GUEST_PDPTE + 2 * i);
5278 
5279 		report(pdpte == pdpt[i], "PDPTE%d is 0x%lx (expected 0x%lx)",
5280 		       i, pdpte, pdpt[i]);
5281 	}
5282 
5283 	/*
5284 	 * Now, try to enter the guest in PAE mode. If the PDPTEs in the
5285 	 * vmcs are wrong, this will fail.
5286 	 */
5287 	enter_guest();
5288 	skip_exit_vmcall();
5289 
5290 	/*
5291 	 * Return guest to 64-bit mode and wrap up.
5292 	 */
5293 	vmcs_write(ENT_CONTROLS, ent_ctls);
5294 	vmcs_write(GUEST_EFER, guest_efer);
5295 	vmcs_write(GUEST_AR_CS, guest_ar_cs);
5296 	vmcs_write(GUEST_CR0, guest_cr0);
5297 	vmcs_write(GUEST_CR4, guest_cr4);
5298 	vmcs_write(GUEST_CR3, guest_cr3);
5299 
5300 	enter_guest();
5301 }
5302 
5303 /*
5304  * Tests for VM-execution control fields
5305  */
5306 static void test_vm_execution_ctls(void)
5307 {
5308 	test_pin_based_ctls();
5309 	test_primary_processor_based_ctls();
5310 	test_secondary_processor_based_ctls();
5311 	test_cr3_targets();
5312 	test_io_bitmaps();
5313 	test_msr_bitmap();
5314 	test_apic_ctls();
5315 	test_tpr_threshold();
5316 	test_nmi_ctrls();
5317 	test_pml();
5318 	test_vpid();
5319 	test_ept_eptp();
5320 	test_vmx_preemption_timer();
5321 }
5322 
5323  /*
5324   * The following checks are performed for the VM-entry MSR-load address if
5325   * the VM-entry MSR-load count field is non-zero:
5326   *
5327   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
5328   *      The address should not set any bits beyond the processor's
5329   *      physical-address width.
5330   *
5331   *    - The address of the last byte in the VM-entry MSR-load area
5332   *      should not set any bits beyond the processor's physical-address
5333   *      width. The address of this last byte is VM-entry MSR-load address
5334   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5335   *      uses more bits than the processor's physical-address width.)
5336   *
5337   *
5338   *  [Intel SDM]
5339   */
5340 static void test_entry_msr_load(void)
5341 {
5342 	entry_msr_load = alloc_page();
5343 	u64 tmp;
5344 	u32 entry_msr_ld_cnt = 1;
5345 	int i;
5346 	u32 addr_len = 64;
5347 
5348 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
5349 
5350 	/* Check first 4 bits of VM-entry MSR-load address */
5351 	for (i = 0; i < 4; i++) {
5352 		tmp = (u64)entry_msr_load | 1ull << i;
5353 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
5354 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
5355 				    tmp & 0xf);
5356 		test_vmx_invalid_controls();
5357 		report_prefix_pop();
5358 	}
5359 
5360 	if (basic.val & (1ul << 48))
5361 		addr_len = 32;
5362 
5363 	test_vmcs_addr_values("VM-entry-MSR-load address",
5364 				ENTER_MSR_LD_ADDR, 16, false, false,
5365 				4, addr_len - 1);
5366 
5367 	/*
5368 	 * Check last byte of VM-entry MSR-load address
5369 	 */
5370 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
5371 
5372 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5373 							i < 64; i++) {
5374 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
5375 			1ul << i;
5376 		vmcs_write(ENTER_MSR_LD_ADDR,
5377 			   tmp - (entry_msr_ld_cnt * 16 - 1));
5378 		test_vmx_invalid_controls();
5379 	}
5380 
5381 	vmcs_write(ENT_MSR_LD_CNT, 2);
5382 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5383 	test_vmx_invalid_controls();
5384 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5385 	test_vmx_valid_controls();
5386 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5387 	test_vmx_valid_controls();
5388 }
5389 
5390 static struct vmx_state_area_test_data {
5391 	u32 msr;
5392 	u64 exp;
5393 	bool enabled;
5394 } vmx_state_area_test_data;
5395 
5396 static void guest_state_test_main(void)
5397 {
5398 	u64 obs;
5399 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
5400 
5401 	while (1) {
5402 		if (vmx_get_test_stage() == 2)
5403 			break;
5404 
5405 		if (data->enabled) {
5406 			obs = rdmsr(data->msr);
5407 			report(data->exp == obs,
5408 			       "Guest state is 0x%lx (expected 0x%lx)",
5409 			       obs, data->exp);
5410 		}
5411 
5412 		vmcall();
5413 	}
5414 
5415 	asm volatile("fnop");
5416 }
5417 
5418 static void test_guest_state(const char *test, bool xfail, u64 field,
5419 			     const char * field_name)
5420 {
5421 	struct vmentry_result result;
5422 	u8 abort_flags;
5423 
5424 	abort_flags = ABORT_ON_EARLY_VMENTRY_FAIL;
5425 	if (!xfail)
5426 		abort_flags = ABORT_ON_INVALID_GUEST_STATE;
5427 
5428 	__enter_guest(abort_flags, &result);
5429 
5430 	report(result.exit_reason.failed_vmentry == xfail &&
5431 	       ((xfail && result.exit_reason.basic == VMX_FAIL_STATE) ||
5432 	        (!xfail && result.exit_reason.basic == VMX_VMCALL)) &&
5433 		(!xfail || vmcs_read(EXI_QUALIFICATION) == ENTRY_FAIL_DEFAULT),
5434 	        "%s, %s = %lx", test, field_name, field);
5435 
5436 	if (!result.exit_reason.failed_vmentry)
5437 		skip_exit_insn();
5438 }
5439 
5440 /*
5441  * Tests for VM-entry control fields
5442  */
5443 static void test_vm_entry_ctls(void)
5444 {
5445 	test_invalid_event_injection();
5446 	test_entry_msr_load();
5447 }
5448 
5449 /*
5450  * The following checks are performed for the VM-exit MSR-store address if
5451  * the VM-exit MSR-store count field is non-zero:
5452  *
5453  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
5454  *      The address should not set any bits beyond the processor's
5455  *      physical-address width.
5456  *
5457  *    - The address of the last byte in the VM-exit MSR-store area
5458  *      should not set any bits beyond the processor's physical-address
5459  *      width. The address of this last byte is VM-exit MSR-store address
5460  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5461  *      uses more bits than the processor's physical-address width.)
5462  *
5463  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
5464  * in the range 63:32.
5465  *
5466  *  [Intel SDM]
5467  */
5468 static void test_exit_msr_store(void)
5469 {
5470 	exit_msr_store = alloc_page();
5471 	u64 tmp;
5472 	u32 exit_msr_st_cnt = 1;
5473 	int i;
5474 	u32 addr_len = 64;
5475 
5476 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
5477 
5478 	/* Check first 4 bits of VM-exit MSR-store address */
5479 	for (i = 0; i < 4; i++) {
5480 		tmp = (u64)exit_msr_store | 1ull << i;
5481 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
5482 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
5483 				    tmp & 0xf);
5484 		test_vmx_invalid_controls();
5485 		report_prefix_pop();
5486 	}
5487 
5488 	if (basic.val & (1ul << 48))
5489 		addr_len = 32;
5490 
5491 	test_vmcs_addr_values("VM-exit-MSR-store address",
5492 				EXIT_MSR_ST_ADDR, 16, false, false,
5493 				4, addr_len - 1);
5494 
5495 	/*
5496 	 * Check last byte of VM-exit MSR-store address
5497 	 */
5498 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
5499 
5500 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5501 							i < 64; i++) {
5502 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
5503 			1ul << i;
5504 		vmcs_write(EXIT_MSR_ST_ADDR,
5505 			   tmp - (exit_msr_st_cnt * 16 - 1));
5506 		test_vmx_invalid_controls();
5507 	}
5508 
5509 	vmcs_write(EXI_MSR_ST_CNT, 2);
5510 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5511 	test_vmx_invalid_controls();
5512 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5513 	test_vmx_valid_controls();
5514 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5515 	test_vmx_valid_controls();
5516 }
5517 
5518 /*
5519  * Tests for VM-exit controls
5520  */
5521 static void test_vm_exit_ctls(void)
5522 {
5523 	test_exit_msr_store();
5524 }
5525 
5526 /*
5527  * Check that the virtual CPU checks all of the VMX controls as
5528  * documented in the Intel SDM.
5529  */
5530 static void vmx_controls_test(void)
5531 {
5532 	/*
5533 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
5534 	 * fail due to invalid guest state, should we make it that
5535 	 * far.
5536 	 */
5537 	vmcs_write(GUEST_RFLAGS, 0);
5538 
5539 	test_vm_execution_ctls();
5540 	test_vm_exit_ctls();
5541 	test_vm_entry_ctls();
5542 }
5543 
5544 struct apic_reg_virt_config {
5545 	bool apic_register_virtualization;
5546 	bool use_tpr_shadow;
5547 	bool virtualize_apic_accesses;
5548 	bool virtualize_x2apic_mode;
5549 	bool activate_secondary_controls;
5550 };
5551 
5552 struct apic_reg_test {
5553 	const char *name;
5554 	struct apic_reg_virt_config apic_reg_virt_config;
5555 };
5556 
5557 struct apic_reg_virt_expectation {
5558 	enum Reason rd_exit_reason;
5559 	enum Reason wr_exit_reason;
5560 	u32 val;
5561 	u32 (*virt_fn)(u32);
5562 
5563 	/*
5564 	 * If false, accessing the APIC access address from L2 is treated as a
5565 	 * normal memory operation, rather than triggering virtualization.
5566 	 */
5567 	bool virtualize_apic_accesses;
5568 };
5569 
5570 static u32 apic_virt_identity(u32 val)
5571 {
5572 	return val;
5573 }
5574 
5575 static u32 apic_virt_nibble1(u32 val)
5576 {
5577 	return val & 0xf0;
5578 }
5579 
5580 static u32 apic_virt_byte3(u32 val)
5581 {
5582 	return val & (0xff << 24);
5583 }
5584 
5585 static bool apic_reg_virt_exit_expectation(
5586 	u32 reg, struct apic_reg_virt_config *config,
5587 	struct apic_reg_virt_expectation *expectation)
5588 {
5589 	/* Good configs, where some L2 APIC accesses are virtualized. */
5590 	bool virtualize_apic_accesses_only =
5591 		config->virtualize_apic_accesses &&
5592 		!config->use_tpr_shadow &&
5593 		!config->apic_register_virtualization &&
5594 		!config->virtualize_x2apic_mode &&
5595 		config->activate_secondary_controls;
5596 	bool virtualize_apic_accesses_and_use_tpr_shadow =
5597 		config->virtualize_apic_accesses &&
5598 		config->use_tpr_shadow &&
5599 		!config->apic_register_virtualization &&
5600 		!config->virtualize_x2apic_mode &&
5601 		config->activate_secondary_controls;
5602 	bool apic_register_virtualization =
5603 		config->virtualize_apic_accesses &&
5604 		config->use_tpr_shadow &&
5605 		config->apic_register_virtualization &&
5606 		!config->virtualize_x2apic_mode &&
5607 		config->activate_secondary_controls;
5608 
5609 	expectation->val = MAGIC_VAL_1;
5610 	expectation->virt_fn = apic_virt_identity;
5611 	expectation->virtualize_apic_accesses =
5612 		config->virtualize_apic_accesses &&
5613 		config->activate_secondary_controls;
5614 	if (virtualize_apic_accesses_only) {
5615 		expectation->rd_exit_reason = VMX_APIC_ACCESS;
5616 		expectation->wr_exit_reason = VMX_APIC_ACCESS;
5617 	} else if (virtualize_apic_accesses_and_use_tpr_shadow) {
5618 		switch (reg) {
5619 		case APIC_TASKPRI:
5620 			expectation->rd_exit_reason = VMX_VMCALL;
5621 			expectation->wr_exit_reason = VMX_VMCALL;
5622 			expectation->virt_fn = apic_virt_nibble1;
5623 			break;
5624 		default:
5625 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5626 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5627 		}
5628 	} else if (apic_register_virtualization) {
5629 		expectation->rd_exit_reason = VMX_VMCALL;
5630 
5631 		switch (reg) {
5632 		case APIC_ID:
5633 		case APIC_EOI:
5634 		case APIC_LDR:
5635 		case APIC_DFR:
5636 		case APIC_SPIV:
5637 		case APIC_ESR:
5638 		case APIC_ICR:
5639 		case APIC_LVTT:
5640 		case APIC_LVTTHMR:
5641 		case APIC_LVTPC:
5642 		case APIC_LVT0:
5643 		case APIC_LVT1:
5644 		case APIC_LVTERR:
5645 		case APIC_TMICT:
5646 		case APIC_TDCR:
5647 			expectation->wr_exit_reason = VMX_APIC_WRITE;
5648 			break;
5649 		case APIC_LVR:
5650 		case APIC_ISR ... APIC_ISR + 0x70:
5651 		case APIC_TMR ... APIC_TMR + 0x70:
5652 		case APIC_IRR ... APIC_IRR + 0x70:
5653 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5654 			break;
5655 		case APIC_TASKPRI:
5656 			expectation->wr_exit_reason = VMX_VMCALL;
5657 			expectation->virt_fn = apic_virt_nibble1;
5658 			break;
5659 		case APIC_ICR2:
5660 			expectation->wr_exit_reason = VMX_VMCALL;
5661 			expectation->virt_fn = apic_virt_byte3;
5662 			break;
5663 		default:
5664 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5665 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5666 		}
5667 	} else if (!expectation->virtualize_apic_accesses) {
5668 		/*
5669 		 * No APIC registers are directly virtualized. This includes
5670 		 * VTPR, which can be virtualized through MOV to/from CR8 via
5671 		 * the use TPR shadow control, but not through directly
5672 		 * accessing VTPR.
5673 		 */
5674 		expectation->rd_exit_reason = VMX_VMCALL;
5675 		expectation->wr_exit_reason = VMX_VMCALL;
5676 	} else {
5677 		printf("Cannot parse APIC register virtualization config:\n"
5678 		       "\tvirtualize_apic_accesses: %d\n"
5679 		       "\tuse_tpr_shadow: %d\n"
5680 		       "\tapic_register_virtualization: %d\n"
5681 		       "\tvirtualize_x2apic_mode: %d\n"
5682 		       "\tactivate_secondary_controls: %d\n",
5683 		       config->virtualize_apic_accesses,
5684 		       config->use_tpr_shadow,
5685 		       config->apic_register_virtualization,
5686 		       config->virtualize_x2apic_mode,
5687 		       config->activate_secondary_controls);
5688 
5689 		return false;
5690 	}
5691 
5692 	return true;
5693 }
5694 
5695 struct apic_reg_test apic_reg_tests[] = {
5696 	/* Good configs, where some L2 APIC accesses are virtualized. */
5697 	{
5698 		.name = "Virtualize APIC accesses",
5699 		.apic_reg_virt_config = {
5700 			.virtualize_apic_accesses = true,
5701 			.use_tpr_shadow = false,
5702 			.apic_register_virtualization = false,
5703 			.virtualize_x2apic_mode = false,
5704 			.activate_secondary_controls = true,
5705 		},
5706 	},
5707 	{
5708 		.name = "Virtualize APIC accesses + Use TPR shadow",
5709 		.apic_reg_virt_config = {
5710 			.virtualize_apic_accesses = true,
5711 			.use_tpr_shadow = true,
5712 			.apic_register_virtualization = false,
5713 			.virtualize_x2apic_mode = false,
5714 			.activate_secondary_controls = true,
5715 		},
5716 	},
5717 	{
5718 		.name = "APIC-register virtualization",
5719 		.apic_reg_virt_config = {
5720 			.virtualize_apic_accesses = true,
5721 			.use_tpr_shadow = true,
5722 			.apic_register_virtualization = true,
5723 			.virtualize_x2apic_mode = false,
5724 			.activate_secondary_controls = true,
5725 		},
5726 	},
5727 
5728 	/*
5729 	 * Test that the secondary processor-based VM-execution controls are
5730 	 * correctly ignored when "activate secondary controls" is disabled.
5731 	 */
5732 	{
5733 		.name = "Activate secondary controls off",
5734 		.apic_reg_virt_config = {
5735 			.virtualize_apic_accesses = true,
5736 			.use_tpr_shadow = false,
5737 			.apic_register_virtualization = true,
5738 			.virtualize_x2apic_mode = true,
5739 			.activate_secondary_controls = false,
5740 		},
5741 	},
5742 	{
5743 		.name = "Activate secondary controls off + Use TPR shadow",
5744 		.apic_reg_virt_config = {
5745 			.virtualize_apic_accesses = true,
5746 			.use_tpr_shadow = true,
5747 			.apic_register_virtualization = true,
5748 			.virtualize_x2apic_mode = true,
5749 			.activate_secondary_controls = false,
5750 		},
5751 	},
5752 
5753 	/*
5754 	 * Test that the APIC access address is treated like an arbitrary memory
5755 	 * address when "virtualize APIC accesses" is disabled.
5756 	 */
5757 	{
5758 		.name = "Virtualize APIC accesses off + Use TPR shadow",
5759 		.apic_reg_virt_config = {
5760 			.virtualize_apic_accesses = false,
5761 			.use_tpr_shadow = true,
5762 			.apic_register_virtualization = true,
5763 			.virtualize_x2apic_mode = true,
5764 			.activate_secondary_controls = true,
5765 		},
5766 	},
5767 
5768 	/*
5769 	 * Test that VM entry fails due to invalid controls when
5770 	 * "APIC-register virtualization" is enabled while "use TPR shadow" is
5771 	 * disabled.
5772 	 */
5773 	{
5774 		.name = "APIC-register virtualization + Use TPR shadow off",
5775 		.apic_reg_virt_config = {
5776 			.virtualize_apic_accesses = true,
5777 			.use_tpr_shadow = false,
5778 			.apic_register_virtualization = true,
5779 			.virtualize_x2apic_mode = false,
5780 			.activate_secondary_controls = true,
5781 		},
5782 	},
5783 
5784 	/*
5785 	 * Test that VM entry fails due to invalid controls when
5786 	 * "Virtualize x2APIC mode" is enabled while "use TPR shadow" is
5787 	 * disabled.
5788 	 */
5789 	{
5790 		.name = "Virtualize x2APIC mode + Use TPR shadow off",
5791 		.apic_reg_virt_config = {
5792 			.virtualize_apic_accesses = false,
5793 			.use_tpr_shadow = false,
5794 			.apic_register_virtualization = false,
5795 			.virtualize_x2apic_mode = true,
5796 			.activate_secondary_controls = true,
5797 		},
5798 	},
5799 	{
5800 		.name = "Virtualize x2APIC mode + Use TPR shadow off v2",
5801 		.apic_reg_virt_config = {
5802 			.virtualize_apic_accesses = false,
5803 			.use_tpr_shadow = false,
5804 			.apic_register_virtualization = true,
5805 			.virtualize_x2apic_mode = true,
5806 			.activate_secondary_controls = true,
5807 		},
5808 	},
5809 
5810 	/*
5811 	 * Test that VM entry fails due to invalid controls when
5812 	 * "virtualize x2APIC mode" is enabled while "virtualize APIC accesses"
5813 	 * is enabled.
5814 	 */
5815 	{
5816 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses",
5817 		.apic_reg_virt_config = {
5818 			.virtualize_apic_accesses = true,
5819 			.use_tpr_shadow = true,
5820 			.apic_register_virtualization = false,
5821 			.virtualize_x2apic_mode = true,
5822 			.activate_secondary_controls = true,
5823 		},
5824 	},
5825 	{
5826 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses v2",
5827 		.apic_reg_virt_config = {
5828 			.virtualize_apic_accesses = true,
5829 			.use_tpr_shadow = true,
5830 			.apic_register_virtualization = true,
5831 			.virtualize_x2apic_mode = true,
5832 			.activate_secondary_controls = true,
5833 		},
5834 	},
5835 };
5836 
5837 enum Apic_op {
5838 	APIC_OP_XAPIC_RD,
5839 	APIC_OP_XAPIC_WR,
5840 	TERMINATE,
5841 };
5842 
5843 static u32 vmx_xapic_read(u32 *apic_access_address, u32 reg)
5844 {
5845 	return *(volatile u32 *)((uintptr_t)apic_access_address + reg);
5846 }
5847 
5848 static void vmx_xapic_write(u32 *apic_access_address, u32 reg, u32 val)
5849 {
5850 	*(volatile u32 *)((uintptr_t)apic_access_address + reg) = val;
5851 }
5852 
5853 struct apic_reg_virt_guest_args {
5854 	enum Apic_op op;
5855 	u32 *apic_access_address;
5856 	u32 reg;
5857 	u32 val;
5858 	bool check_rd;
5859 	u32 (*virt_fn)(u32);
5860 } apic_reg_virt_guest_args;
5861 
5862 static void apic_reg_virt_guest(void)
5863 {
5864 	volatile struct apic_reg_virt_guest_args *args =
5865 		&apic_reg_virt_guest_args;
5866 
5867 	for (;;) {
5868 		enum Apic_op op = args->op;
5869 		u32 *apic_access_address = args->apic_access_address;
5870 		u32 reg = args->reg;
5871 		u32 val = args->val;
5872 		bool check_rd = args->check_rd;
5873 		u32 (*virt_fn)(u32) = args->virt_fn;
5874 
5875 		if (op == TERMINATE)
5876 			break;
5877 
5878 		if (op == APIC_OP_XAPIC_RD) {
5879 			u32 ret = vmx_xapic_read(apic_access_address, reg);
5880 
5881 			if (check_rd) {
5882 				u32 want = virt_fn(val);
5883 				u32 got = virt_fn(ret);
5884 
5885 				report(got == want,
5886 				       "read 0x%x, expected 0x%x.", got, want);
5887 			}
5888 		} else if (op == APIC_OP_XAPIC_WR) {
5889 			vmx_xapic_write(apic_access_address, reg, val);
5890 		}
5891 
5892 		/*
5893 		 * The L1 should always execute a vmcall after it's done testing
5894 		 * an individual APIC operation. This helps to validate that the
5895 		 * L1 and L2 are in sync with each other, as expected.
5896 		 */
5897 		vmcall();
5898 	}
5899 }
5900 
5901 static void test_xapic_rd(
5902 	u32 reg, struct apic_reg_virt_expectation *expectation,
5903 	u32 *apic_access_address, u32 *virtual_apic_page)
5904 {
5905 	u32 val = expectation->val;
5906 	u32 exit_reason_want = expectation->rd_exit_reason;
5907 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5908 
5909 	report_prefix_pushf("xapic - reading 0x%03x", reg);
5910 
5911 	/* Configure guest to do an xapic read */
5912 	args->op = APIC_OP_XAPIC_RD;
5913 	args->apic_access_address = apic_access_address;
5914 	args->reg = reg;
5915 	args->val = val;
5916 	args->check_rd = exit_reason_want == VMX_VMCALL;
5917 	args->virt_fn = expectation->virt_fn;
5918 
5919 	/* Setup virtual APIC page */
5920 	if (!expectation->virtualize_apic_accesses) {
5921 		apic_access_address[apic_reg_index(reg)] = val;
5922 		virtual_apic_page[apic_reg_index(reg)] = 0;
5923 	} else if (exit_reason_want == VMX_VMCALL) {
5924 		apic_access_address[apic_reg_index(reg)] = 0;
5925 		virtual_apic_page[apic_reg_index(reg)] = val;
5926 	}
5927 
5928 	/* Enter guest */
5929 	enter_guest();
5930 
5931 	/*
5932 	 * Validate the behavior and
5933 	 * pass a magic value back to the guest.
5934 	 */
5935 	if (exit_reason_want == VMX_APIC_ACCESS) {
5936 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5937 
5938 		assert_exit_reason(exit_reason_want);
5939 		report(apic_page_offset == reg,
5940 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
5941 		       apic_page_offset, reg);
5942 		skip_exit_insn();
5943 
5944 		/* Reenter guest so it can consume/check rcx and exit again. */
5945 		enter_guest();
5946 	} else if (exit_reason_want != VMX_VMCALL) {
5947 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
5948 	}
5949 
5950 	skip_exit_vmcall();
5951 	report_prefix_pop();
5952 }
5953 
5954 static void test_xapic_wr(
5955 	u32 reg, struct apic_reg_virt_expectation *expectation,
5956 	u32 *apic_access_address, u32 *virtual_apic_page)
5957 {
5958 	u32 val = expectation->val;
5959 	u32 exit_reason_want = expectation->wr_exit_reason;
5960 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5961 	bool virtualized =
5962 		expectation->virtualize_apic_accesses &&
5963 		(exit_reason_want == VMX_APIC_WRITE ||
5964 		 exit_reason_want == VMX_VMCALL);
5965 	bool checked = false;
5966 
5967 	report_prefix_pushf("xapic - writing 0x%x to 0x%03x", val, reg);
5968 
5969 	/* Configure guest to do an xapic read */
5970 	args->op = APIC_OP_XAPIC_WR;
5971 	args->apic_access_address = apic_access_address;
5972 	args->reg = reg;
5973 	args->val = val;
5974 
5975 	/* Setup virtual APIC page */
5976 	if (virtualized || !expectation->virtualize_apic_accesses) {
5977 		apic_access_address[apic_reg_index(reg)] = 0;
5978 		virtual_apic_page[apic_reg_index(reg)] = 0;
5979 	}
5980 
5981 	/* Enter guest */
5982 	enter_guest();
5983 
5984 	/*
5985 	 * Validate the behavior and
5986 	 * pass a magic value back to the guest.
5987 	 */
5988 	if (exit_reason_want == VMX_APIC_ACCESS) {
5989 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5990 
5991 		assert_exit_reason(exit_reason_want);
5992 		report(apic_page_offset == reg,
5993 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
5994 		       apic_page_offset, reg);
5995 		skip_exit_insn();
5996 
5997 		/* Reenter guest so it can consume/check rcx and exit again. */
5998 		enter_guest();
5999 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6000 		assert_exit_reason(exit_reason_want);
6001 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6002 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%x",
6003 		       apic_reg_index(reg),
6004 		       virtual_apic_page[apic_reg_index(reg)], val);
6005 		checked = true;
6006 
6007 		/* Reenter guest so it can consume/check rcx and exit again. */
6008 		enter_guest();
6009 	} else if (exit_reason_want != VMX_VMCALL) {
6010 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6011 	}
6012 
6013 	assert_exit_reason(VMX_VMCALL);
6014 	if (virtualized && !checked) {
6015 		u32 want = expectation->virt_fn(val);
6016 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6017 		got = expectation->virt_fn(got);
6018 
6019 		report(got == want, "exitless write; val is 0x%x, want 0x%x",
6020 		       got, want);
6021 	} else if (!expectation->virtualize_apic_accesses && !checked) {
6022 		u32 got = apic_access_address[apic_reg_index(reg)];
6023 
6024 		report(got == val,
6025 		       "non-virtualized write; val is 0x%x, want 0x%x", got,
6026 		       val);
6027 	} else if (!expectation->virtualize_apic_accesses && checked) {
6028 		report_fail("Non-virtualized write was prematurely checked!");
6029 	}
6030 
6031 	skip_exit_vmcall();
6032 	report_prefix_pop();
6033 }
6034 
6035 enum Config_type {
6036 	CONFIG_TYPE_GOOD,
6037 	CONFIG_TYPE_UNSUPPORTED,
6038 	CONFIG_TYPE_VMENTRY_FAILS_EARLY,
6039 };
6040 
6041 static enum Config_type configure_apic_reg_virt_test(
6042 	struct apic_reg_virt_config *apic_reg_virt_config)
6043 {
6044 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6045 	u32 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6046 	/* Configs where L2 entry fails early, due to invalid controls. */
6047 	bool use_tpr_shadow_incorrectly_off =
6048 		!apic_reg_virt_config->use_tpr_shadow &&
6049 		(apic_reg_virt_config->apic_register_virtualization ||
6050 		 apic_reg_virt_config->virtualize_x2apic_mode) &&
6051 		apic_reg_virt_config->activate_secondary_controls;
6052 	bool virtualize_apic_accesses_incorrectly_on =
6053 		apic_reg_virt_config->virtualize_apic_accesses &&
6054 		apic_reg_virt_config->virtualize_x2apic_mode &&
6055 		apic_reg_virt_config->activate_secondary_controls;
6056 	bool vmentry_fails_early =
6057 		use_tpr_shadow_incorrectly_off ||
6058 		virtualize_apic_accesses_incorrectly_on;
6059 
6060 	if (apic_reg_virt_config->activate_secondary_controls) {
6061 		if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
6062 			printf("VM-execution control \"activate secondary controls\" NOT supported.\n");
6063 			return CONFIG_TYPE_UNSUPPORTED;
6064 		}
6065 		cpu_exec_ctrl0 |= CPU_SECONDARY;
6066 	} else {
6067 		cpu_exec_ctrl0 &= ~CPU_SECONDARY;
6068 	}
6069 
6070 	if (apic_reg_virt_config->virtualize_apic_accesses) {
6071 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES)) {
6072 			printf("VM-execution control \"virtualize APIC accesses\" NOT supported.\n");
6073 			return CONFIG_TYPE_UNSUPPORTED;
6074 		}
6075 		cpu_exec_ctrl1 |= CPU_VIRT_APIC_ACCESSES;
6076 	} else {
6077 		cpu_exec_ctrl1 &= ~CPU_VIRT_APIC_ACCESSES;
6078 	}
6079 
6080 	if (apic_reg_virt_config->use_tpr_shadow) {
6081 		if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6082 			printf("VM-execution control \"use TPR shadow\" NOT supported.\n");
6083 			return CONFIG_TYPE_UNSUPPORTED;
6084 		}
6085 		cpu_exec_ctrl0 |= CPU_TPR_SHADOW;
6086 	} else {
6087 		cpu_exec_ctrl0 &= ~CPU_TPR_SHADOW;
6088 	}
6089 
6090 	if (apic_reg_virt_config->apic_register_virtualization) {
6091 		if (!(ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT)) {
6092 			printf("VM-execution control \"APIC-register virtualization\" NOT supported.\n");
6093 			return CONFIG_TYPE_UNSUPPORTED;
6094 		}
6095 		cpu_exec_ctrl1 |= CPU_APIC_REG_VIRT;
6096 	} else {
6097 		cpu_exec_ctrl1 &= ~CPU_APIC_REG_VIRT;
6098 	}
6099 
6100 	if (apic_reg_virt_config->virtualize_x2apic_mode) {
6101 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_X2APIC)) {
6102 			printf("VM-execution control \"virtualize x2APIC mode\" NOT supported.\n");
6103 			return CONFIG_TYPE_UNSUPPORTED;
6104 		}
6105 		cpu_exec_ctrl1 |= CPU_VIRT_X2APIC;
6106 	} else {
6107 		cpu_exec_ctrl1 &= ~CPU_VIRT_X2APIC;
6108 	}
6109 
6110 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6111 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6112 
6113 	if (vmentry_fails_early)
6114 		return CONFIG_TYPE_VMENTRY_FAILS_EARLY;
6115 
6116 	return CONFIG_TYPE_GOOD;
6117 }
6118 
6119 static bool cpu_has_apicv(void)
6120 {
6121 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
6122 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
6123 		(ctrl_pin_rev.clr & PIN_POST_INTR));
6124 }
6125 
6126 /* Validates APIC register access across valid virtualization configurations. */
6127 static void apic_reg_virt_test(void)
6128 {
6129 	u32 *apic_access_address;
6130 	u32 *virtual_apic_page;
6131 	u64 control;
6132 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6133 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6134 	int i;
6135 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
6136 
6137 	if (!cpu_has_apicv()) {
6138 		report_skip(__func__);
6139 		return;
6140 	}
6141 
6142 	control = cpu_exec_ctrl1;
6143 	control &= ~CPU_VINTD;
6144 	vmcs_write(CPU_EXEC_CTRL1, control);
6145 
6146 	test_set_guest(apic_reg_virt_guest);
6147 
6148 	/*
6149 	 * From the SDM: The 1-setting of the "virtualize APIC accesses"
6150 	 * VM-execution is guaranteed to apply only if translations to the
6151 	 * APIC-access address use a 4-KByte page.
6152 	 */
6153 	apic_access_address = alloc_page();
6154 	force_4k_page(apic_access_address);
6155 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_address));
6156 
6157 	virtual_apic_page = alloc_page();
6158 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6159 
6160 	for (i = 0; i < ARRAY_SIZE(apic_reg_tests); i++) {
6161 		struct apic_reg_test *apic_reg_test = &apic_reg_tests[i];
6162 		struct apic_reg_virt_config *apic_reg_virt_config =
6163 				&apic_reg_test->apic_reg_virt_config;
6164 		enum Config_type config_type;
6165 		u32 reg;
6166 
6167 		printf("--- %s test ---\n", apic_reg_test->name);
6168 		config_type =
6169 			configure_apic_reg_virt_test(apic_reg_virt_config);
6170 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6171 			printf("Skip because of missing features.\n");
6172 			continue;
6173 		}
6174 
6175 		if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6176 			enter_guest_with_bad_controls();
6177 			continue;
6178 		}
6179 
6180 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6181 			struct apic_reg_virt_expectation expectation = {};
6182 			bool ok;
6183 
6184 			ok = apic_reg_virt_exit_expectation(
6185 				reg, apic_reg_virt_config, &expectation);
6186 			if (!ok) {
6187 				report_fail("Malformed test.");
6188 				break;
6189 			}
6190 
6191 			test_xapic_rd(reg, &expectation, apic_access_address,
6192 				      virtual_apic_page);
6193 			test_xapic_wr(reg, &expectation, apic_access_address,
6194 				      virtual_apic_page);
6195 		}
6196 	}
6197 
6198 	/* Terminate the guest */
6199 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6200 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6201 	args->op = TERMINATE;
6202 	enter_guest();
6203 	assert_exit_reason(VMX_VMCALL);
6204 }
6205 
6206 struct virt_x2apic_mode_config {
6207 	struct apic_reg_virt_config apic_reg_virt_config;
6208 	bool virtual_interrupt_delivery;
6209 	bool use_msr_bitmaps;
6210 	bool disable_x2apic_msr_intercepts;
6211 	bool disable_x2apic;
6212 };
6213 
6214 struct virt_x2apic_mode_test_case {
6215 	const char *name;
6216 	struct virt_x2apic_mode_config virt_x2apic_mode_config;
6217 };
6218 
6219 enum Virt_x2apic_mode_behavior_type {
6220 	X2APIC_ACCESS_VIRTUALIZED,
6221 	X2APIC_ACCESS_PASSED_THROUGH,
6222 	X2APIC_ACCESS_TRIGGERS_GP,
6223 };
6224 
6225 struct virt_x2apic_mode_expectation {
6226 	enum Reason rd_exit_reason;
6227 	enum Reason wr_exit_reason;
6228 
6229 	/*
6230 	 * RDMSR and WRMSR handle 64-bit values. However, except for ICR, all of
6231 	 * the x2APIC registers are 32 bits. Notice:
6232 	 *   1. vmx_x2apic_read() clears the upper 32 bits for 32-bit registers.
6233 	 *   2. vmx_x2apic_write() expects the val arg to be well-formed.
6234 	 */
6235 	u64 rd_val;
6236 	u64 wr_val;
6237 
6238 	/*
6239 	 * Compares input to virtualized output;
6240 	 * 1st arg is pointer to return expected virtualization output.
6241 	 */
6242 	u64 (*virt_fn)(u64);
6243 
6244 	enum Virt_x2apic_mode_behavior_type rd_behavior;
6245 	enum Virt_x2apic_mode_behavior_type wr_behavior;
6246 	bool wr_only;
6247 };
6248 
6249 static u64 virt_x2apic_mode_identity(u64 val)
6250 {
6251 	return val;
6252 }
6253 
6254 static u64 virt_x2apic_mode_nibble1(u64 val)
6255 {
6256 	return val & 0xf0;
6257 }
6258 
6259 static void virt_x2apic_mode_rd_expectation(
6260 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6261 	bool apic_register_virtualization, bool virtual_interrupt_delivery,
6262 	struct virt_x2apic_mode_expectation *expectation)
6263 {
6264 	bool readable =
6265 		!x2apic_reg_reserved(reg) &&
6266 		reg != APIC_EOI;
6267 
6268 	expectation->rd_exit_reason = VMX_VMCALL;
6269 	expectation->virt_fn = virt_x2apic_mode_identity;
6270 	if (virt_x2apic_mode_on && apic_register_virtualization) {
6271 		expectation->rd_val = MAGIC_VAL_1;
6272 		if (reg == APIC_PROCPRI && virtual_interrupt_delivery)
6273 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6274 		else if (reg == APIC_TASKPRI)
6275 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6276 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6277 	} else if (virt_x2apic_mode_on && !apic_register_virtualization &&
6278 		   reg == APIC_TASKPRI) {
6279 		expectation->rd_val = MAGIC_VAL_1;
6280 		expectation->virt_fn = virt_x2apic_mode_nibble1;
6281 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6282 	} else if (!disable_x2apic && readable) {
6283 		expectation->rd_val = apic_read(reg);
6284 		expectation->rd_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6285 	} else {
6286 		expectation->rd_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6287 	}
6288 }
6289 
6290 /*
6291  * get_x2apic_wr_val() creates an innocuous write value for an x2APIC register.
6292  *
6293  * For writable registers, get_x2apic_wr_val() deposits the write value into the
6294  * val pointer arg and returns true. For non-writable registers, val is not
6295  * modified and get_x2apic_wr_val() returns false.
6296  */
6297 static bool get_x2apic_wr_val(u32 reg, u64 *val)
6298 {
6299 	switch (reg) {
6300 	case APIC_TASKPRI:
6301 		/* Bits 31:8 are reserved. */
6302 		*val &= 0xff;
6303 		break;
6304 	case APIC_EOI:
6305 	case APIC_ESR:
6306 	case APIC_TMICT:
6307 		/*
6308 		 * EOI, ESR: WRMSR of a non-zero value causes #GP(0).
6309 		 * TMICT: A write of 0 to the initial-count register effectively
6310 		 *        stops the local APIC timer, in both one-shot and
6311 		 *        periodic mode.
6312 		 */
6313 		*val = 0;
6314 		break;
6315 	case APIC_SPIV:
6316 	case APIC_LVTT:
6317 	case APIC_LVTTHMR:
6318 	case APIC_LVTPC:
6319 	case APIC_LVT0:
6320 	case APIC_LVT1:
6321 	case APIC_LVTERR:
6322 	case APIC_TDCR:
6323 		/*
6324 		 * To avoid writing a 1 to a reserved bit or causing some other
6325 		 * unintended side effect, read the current value and use it as
6326 		 * the write value.
6327 		 */
6328 		*val = apic_read(reg);
6329 		break;
6330 	case APIC_CMCI:
6331 		if (!apic_lvt_entry_supported(6))
6332 			return false;
6333 		*val = apic_read(reg);
6334 		break;
6335 	case APIC_ICR:
6336 		*val = 0x40000 | 0xf1;
6337 		break;
6338 	case APIC_SELF_IPI:
6339 		/*
6340 		 * With special processing (i.e., virtualize x2APIC mode +
6341 		 * virtual interrupt delivery), writing zero causes an
6342 		 * APIC-write VM exit. We plan to add a test for enabling
6343 		 * "virtual-interrupt delivery" in VMCS12, and that's where we
6344 		 * will test a self IPI with special processing.
6345 		 */
6346 		*val = 0x0;
6347 		break;
6348 	default:
6349 		return false;
6350 	}
6351 
6352 	return true;
6353 }
6354 
6355 static bool special_processing_applies(u32 reg, u64 *val,
6356 				       bool virt_int_delivery)
6357 {
6358 	bool special_processing =
6359 		(reg == APIC_TASKPRI) ||
6360 		(virt_int_delivery &&
6361 		 (reg == APIC_EOI || reg == APIC_SELF_IPI));
6362 
6363 	if (special_processing) {
6364 		TEST_ASSERT(get_x2apic_wr_val(reg, val));
6365 		return true;
6366 	}
6367 
6368 	return false;
6369 }
6370 
6371 static void virt_x2apic_mode_wr_expectation(
6372 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6373 	bool virt_int_delivery,
6374 	struct virt_x2apic_mode_expectation *expectation)
6375 {
6376 	expectation->wr_exit_reason = VMX_VMCALL;
6377 	expectation->wr_val = MAGIC_VAL_1;
6378 	expectation->wr_only = false;
6379 
6380 	if (virt_x2apic_mode_on &&
6381 	    special_processing_applies(reg, &expectation->wr_val,
6382 				       virt_int_delivery)) {
6383 		expectation->wr_behavior = X2APIC_ACCESS_VIRTUALIZED;
6384 		if (reg == APIC_SELF_IPI)
6385 			expectation->wr_exit_reason = VMX_APIC_WRITE;
6386 	} else if (!disable_x2apic &&
6387 		   get_x2apic_wr_val(reg, &expectation->wr_val)) {
6388 		expectation->wr_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6389 		if (reg == APIC_EOI || reg == APIC_SELF_IPI)
6390 			expectation->wr_only = true;
6391 		if (reg == APIC_ICR)
6392 			expectation->wr_exit_reason = VMX_EXTINT;
6393 	} else {
6394 		expectation->wr_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6395 		/*
6396 		 * Writing 1 to a reserved bit triggers a #GP.
6397 		 * Thus, set the write value to 0, which seems
6398 		 * the most likely to detect a missed #GP.
6399 		 */
6400 		expectation->wr_val = 0;
6401 	}
6402 }
6403 
6404 static void virt_x2apic_mode_exit_expectation(
6405 	u32 reg, struct virt_x2apic_mode_config *config,
6406 	struct virt_x2apic_mode_expectation *expectation)
6407 {
6408 	struct apic_reg_virt_config *base_config =
6409 		&config->apic_reg_virt_config;
6410 	bool virt_x2apic_mode_on =
6411 		base_config->virtualize_x2apic_mode &&
6412 		config->use_msr_bitmaps &&
6413 		config->disable_x2apic_msr_intercepts &&
6414 		base_config->activate_secondary_controls;
6415 
6416 	virt_x2apic_mode_wr_expectation(
6417 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6418 		config->virtual_interrupt_delivery, expectation);
6419 	virt_x2apic_mode_rd_expectation(
6420 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6421 		base_config->apic_register_virtualization,
6422 		config->virtual_interrupt_delivery, expectation);
6423 }
6424 
6425 struct virt_x2apic_mode_test_case virt_x2apic_mode_tests[] = {
6426 	/*
6427 	 * Baseline "virtualize x2APIC mode" configuration:
6428 	 *   - virtualize x2APIC mode
6429 	 *   - virtual-interrupt delivery
6430 	 *   - APIC-register virtualization
6431 	 *   - x2APIC MSR intercepts disabled
6432 	 *
6433 	 * Reads come from virtual APIC page, special processing applies to
6434 	 * VTPR, EOI, and SELF IPI, and all other writes pass through to L1
6435 	 * APIC.
6436 	 */
6437 	{
6438 		.name = "Baseline",
6439 		.virt_x2apic_mode_config = {
6440 			.virtual_interrupt_delivery = true,
6441 			.use_msr_bitmaps = true,
6442 			.disable_x2apic_msr_intercepts = true,
6443 			.disable_x2apic = false,
6444 			.apic_reg_virt_config = {
6445 				.apic_register_virtualization = true,
6446 				.use_tpr_shadow = true,
6447 				.virtualize_apic_accesses = false,
6448 				.virtualize_x2apic_mode = true,
6449 				.activate_secondary_controls = true,
6450 			},
6451 		},
6452 	},
6453 	{
6454 		.name = "Baseline w/ x2apic disabled",
6455 		.virt_x2apic_mode_config = {
6456 			.virtual_interrupt_delivery = true,
6457 			.use_msr_bitmaps = true,
6458 			.disable_x2apic_msr_intercepts = true,
6459 			.disable_x2apic = true,
6460 			.apic_reg_virt_config = {
6461 				.apic_register_virtualization = true,
6462 				.use_tpr_shadow = true,
6463 				.virtualize_apic_accesses = false,
6464 				.virtualize_x2apic_mode = true,
6465 				.activate_secondary_controls = true,
6466 			},
6467 		},
6468 	},
6469 
6470 	/*
6471 	 * Baseline, minus virtual-interrupt delivery. Reads come from virtual
6472 	 * APIC page, special processing applies to VTPR, and all other writes
6473 	 * pass through to L1 APIC.
6474 	 */
6475 	{
6476 		.name = "Baseline - virtual interrupt delivery",
6477 		.virt_x2apic_mode_config = {
6478 			.virtual_interrupt_delivery = false,
6479 			.use_msr_bitmaps = true,
6480 			.disable_x2apic_msr_intercepts = true,
6481 			.disable_x2apic = false,
6482 			.apic_reg_virt_config = {
6483 				.apic_register_virtualization = true,
6484 				.use_tpr_shadow = true,
6485 				.virtualize_apic_accesses = false,
6486 				.virtualize_x2apic_mode = true,
6487 				.activate_secondary_controls = true,
6488 			},
6489 		},
6490 	},
6491 
6492 	/*
6493 	 * Baseline, minus APIC-register virtualization. x2APIC reads pass
6494 	 * through to L1's APIC, unless reading VTPR
6495 	 */
6496 	{
6497 		.name = "Virtualize x2APIC mode, no APIC reg virt",
6498 		.virt_x2apic_mode_config = {
6499 			.virtual_interrupt_delivery = true,
6500 			.use_msr_bitmaps = true,
6501 			.disable_x2apic_msr_intercepts = true,
6502 			.disable_x2apic = false,
6503 			.apic_reg_virt_config = {
6504 				.apic_register_virtualization = false,
6505 				.use_tpr_shadow = true,
6506 				.virtualize_apic_accesses = false,
6507 				.virtualize_x2apic_mode = true,
6508 				.activate_secondary_controls = true,
6509 			},
6510 		},
6511 	},
6512 	{
6513 		.name = "Virtualize x2APIC mode, no APIC reg virt, x2APIC off",
6514 		.virt_x2apic_mode_config = {
6515 			.virtual_interrupt_delivery = true,
6516 			.use_msr_bitmaps = true,
6517 			.disable_x2apic_msr_intercepts = true,
6518 			.disable_x2apic = true,
6519 			.apic_reg_virt_config = {
6520 				.apic_register_virtualization = false,
6521 				.use_tpr_shadow = true,
6522 				.virtualize_apic_accesses = false,
6523 				.virtualize_x2apic_mode = true,
6524 				.activate_secondary_controls = true,
6525 			},
6526 		},
6527 	},
6528 
6529 	/*
6530 	 * Enable "virtualize x2APIC mode" and "APIC-register virtualization",
6531 	 * and disable intercepts for the x2APIC MSRs, but fail to enable
6532 	 * "activate secondary controls" (i.e. L2 gets access to L1's x2APIC
6533 	 * MSRs).
6534 	 */
6535 	{
6536 		.name = "Fail to enable activate secondary controls",
6537 		.virt_x2apic_mode_config = {
6538 			.virtual_interrupt_delivery = true,
6539 			.use_msr_bitmaps = true,
6540 			.disable_x2apic_msr_intercepts = true,
6541 			.disable_x2apic = false,
6542 			.apic_reg_virt_config = {
6543 				.apic_register_virtualization = true,
6544 				.use_tpr_shadow = true,
6545 				.virtualize_apic_accesses = false,
6546 				.virtualize_x2apic_mode = true,
6547 				.activate_secondary_controls = false,
6548 			},
6549 		},
6550 	},
6551 
6552 	/*
6553 	 * Enable "APIC-register virtualization" and enable "activate secondary
6554 	 * controls" and disable intercepts for the x2APIC MSRs, but do not
6555 	 * enable the "virtualize x2APIC mode" VM-execution control (i.e. L2
6556 	 * gets access to L1's x2APIC MSRs).
6557 	 */
6558 	{
6559 		.name = "Fail to enable virtualize x2APIC mode",
6560 		.virt_x2apic_mode_config = {
6561 			.virtual_interrupt_delivery = true,
6562 			.use_msr_bitmaps = true,
6563 			.disable_x2apic_msr_intercepts = true,
6564 			.disable_x2apic = false,
6565 			.apic_reg_virt_config = {
6566 				.apic_register_virtualization = true,
6567 				.use_tpr_shadow = true,
6568 				.virtualize_apic_accesses = false,
6569 				.virtualize_x2apic_mode = false,
6570 				.activate_secondary_controls = true,
6571 			},
6572 		},
6573 	},
6574 
6575 	/*
6576 	 * Disable "Virtualize x2APIC mode", disable x2APIC MSR intercepts, and
6577 	 * enable "APIC-register virtualization" --> L2 gets L1's x2APIC MSRs.
6578 	 */
6579 	{
6580 		.name = "Baseline",
6581 		.virt_x2apic_mode_config = {
6582 			.virtual_interrupt_delivery = true,
6583 			.use_msr_bitmaps = true,
6584 			.disable_x2apic_msr_intercepts = true,
6585 			.disable_x2apic = false,
6586 			.apic_reg_virt_config = {
6587 				.apic_register_virtualization = true,
6588 				.use_tpr_shadow = true,
6589 				.virtualize_apic_accesses = false,
6590 				.virtualize_x2apic_mode = false,
6591 				.activate_secondary_controls = true,
6592 			},
6593 		},
6594 	},
6595 };
6596 
6597 enum X2apic_op {
6598 	X2APIC_OP_RD,
6599 	X2APIC_OP_WR,
6600 	X2APIC_TERMINATE,
6601 };
6602 
6603 static u64 vmx_x2apic_read(u32 reg)
6604 {
6605 	u32 msr_addr = x2apic_msr(reg);
6606 	u64 val;
6607 
6608 	val = rdmsr(msr_addr);
6609 
6610 	return val;
6611 }
6612 
6613 static void vmx_x2apic_write(u32 reg, u64 val)
6614 {
6615 	u32 msr_addr = x2apic_msr(reg);
6616 
6617 	wrmsr(msr_addr, val);
6618 }
6619 
6620 struct virt_x2apic_mode_guest_args {
6621 	enum X2apic_op op;
6622 	u32 reg;
6623 	u64 val;
6624 	bool should_gp;
6625 	u64 (*virt_fn)(u64);
6626 } virt_x2apic_mode_guest_args;
6627 
6628 static volatile bool handle_x2apic_gp_ran;
6629 static volatile u32 handle_x2apic_gp_insn_len;
6630 static void handle_x2apic_gp(struct ex_regs *regs)
6631 {
6632 	handle_x2apic_gp_ran = true;
6633 	regs->rip += handle_x2apic_gp_insn_len;
6634 }
6635 
6636 static handler setup_x2apic_gp_handler(void)
6637 {
6638 	handler old_handler;
6639 
6640 	old_handler = handle_exception(GP_VECTOR, handle_x2apic_gp);
6641 	/* RDMSR and WRMSR are both 2 bytes, assuming no prefixes. */
6642 	handle_x2apic_gp_insn_len = 2;
6643 
6644 	return old_handler;
6645 }
6646 
6647 static void teardown_x2apic_gp_handler(handler old_handler)
6648 {
6649 	handle_exception(GP_VECTOR, old_handler);
6650 
6651 	/*
6652 	 * Defensively reset instruction length, so that if the handler is
6653 	 * incorrectly used, it will loop infinitely, rather than run off into
6654 	 * la la land.
6655 	 */
6656 	handle_x2apic_gp_insn_len = 0;
6657 	handle_x2apic_gp_ran = false;
6658 }
6659 
6660 static void virt_x2apic_mode_guest(void)
6661 {
6662 	volatile struct virt_x2apic_mode_guest_args *args =
6663 		&virt_x2apic_mode_guest_args;
6664 
6665 	for (;;) {
6666 		enum X2apic_op op = args->op;
6667 		u32 reg = args->reg;
6668 		u64 val = args->val;
6669 		bool should_gp = args->should_gp;
6670 		u64 (*virt_fn)(u64) = args->virt_fn;
6671 		handler old_handler;
6672 
6673 		if (op == X2APIC_TERMINATE)
6674 			break;
6675 
6676 		if (should_gp) {
6677 			TEST_ASSERT(!handle_x2apic_gp_ran);
6678 			old_handler = setup_x2apic_gp_handler();
6679 		}
6680 
6681 		if (op == X2APIC_OP_RD) {
6682 			u64 ret = vmx_x2apic_read(reg);
6683 
6684 			if (!should_gp) {
6685 				u64 want = virt_fn(val);
6686 				u64 got = virt_fn(ret);
6687 
6688 				report(got == want,
6689 				       "APIC read; got 0x%lx, want 0x%lx.",
6690 				       got, want);
6691 			}
6692 		} else if (op == X2APIC_OP_WR) {
6693 			vmx_x2apic_write(reg, val);
6694 		}
6695 
6696 		if (should_gp) {
6697 			report(handle_x2apic_gp_ran,
6698 			       "x2APIC op triggered GP.");
6699 			teardown_x2apic_gp_handler(old_handler);
6700 		}
6701 
6702 		/*
6703 		 * The L1 should always execute a vmcall after it's done testing
6704 		 * an individual APIC operation. This helps to validate that the
6705 		 * L1 and L2 are in sync with each other, as expected.
6706 		 */
6707 		vmcall();
6708 	}
6709 }
6710 
6711 static void test_x2apic_rd(
6712 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6713 	u32 *virtual_apic_page)
6714 {
6715 	u64 val = expectation->rd_val;
6716 	u32 exit_reason_want = expectation->rd_exit_reason;
6717 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6718 
6719 	report_prefix_pushf("x2apic - reading 0x%03x", reg);
6720 
6721 	/* Configure guest to do an x2apic read */
6722 	args->op = X2APIC_OP_RD;
6723 	args->reg = reg;
6724 	args->val = val;
6725 	args->should_gp = expectation->rd_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6726 	args->virt_fn = expectation->virt_fn;
6727 
6728 	/* Setup virtual APIC page */
6729 	if (expectation->rd_behavior == X2APIC_ACCESS_VIRTUALIZED)
6730 		virtual_apic_page[apic_reg_index(reg)] = (u32)val;
6731 
6732 	/* Enter guest */
6733 	enter_guest();
6734 
6735 	if (exit_reason_want != VMX_VMCALL) {
6736 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6737 	}
6738 
6739 	skip_exit_vmcall();
6740 	report_prefix_pop();
6741 }
6742 
6743 static volatile bool handle_x2apic_ipi_ran;
6744 static void handle_x2apic_ipi(isr_regs_t *regs)
6745 {
6746 	handle_x2apic_ipi_ran = true;
6747 	eoi();
6748 }
6749 
6750 static void test_x2apic_wr(
6751 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6752 	u32 *virtual_apic_page)
6753 {
6754 	u64 val = expectation->wr_val;
6755 	u32 exit_reason_want = expectation->wr_exit_reason;
6756 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6757 	int ipi_vector = 0xf1;
6758 	u32 restore_val = 0;
6759 
6760 	report_prefix_pushf("x2apic - writing 0x%lx to 0x%03x", val, reg);
6761 
6762 	/* Configure guest to do an x2apic read */
6763 	args->op = X2APIC_OP_WR;
6764 	args->reg = reg;
6765 	args->val = val;
6766 	args->should_gp = expectation->wr_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6767 
6768 	/* Setup virtual APIC page */
6769 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED)
6770 		virtual_apic_page[apic_reg_index(reg)] = 0;
6771 	if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH && !expectation->wr_only)
6772 		restore_val = apic_read(reg);
6773 
6774 	/* Setup IPI handler */
6775 	handle_x2apic_ipi_ran = false;
6776 	handle_irq(ipi_vector, handle_x2apic_ipi);
6777 
6778 	/* Enter guest */
6779 	enter_guest();
6780 
6781 	/*
6782 	 * Validate the behavior and
6783 	 * pass a magic value back to the guest.
6784 	 */
6785 	if (exit_reason_want == VMX_EXTINT) {
6786 		assert_exit_reason(exit_reason_want);
6787 
6788 		/* Clear the external interrupt. */
6789 		irq_enable();
6790 		asm volatile ("nop");
6791 		irq_disable();
6792 		report(handle_x2apic_ipi_ran,
6793 		       "Got pending interrupt after IRQ enabled.");
6794 
6795 		enter_guest();
6796 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6797 		assert_exit_reason(exit_reason_want);
6798 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6799 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%lx",
6800 		       apic_reg_index(reg),
6801 		       virtual_apic_page[apic_reg_index(reg)], val);
6802 
6803 		/* Reenter guest so it can consume/check rcx and exit again. */
6804 		enter_guest();
6805 	} else if (exit_reason_want != VMX_VMCALL) {
6806 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6807 	}
6808 
6809 	assert_exit_reason(VMX_VMCALL);
6810 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) {
6811 		u64 want = val;
6812 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6813 
6814 		report(got == want, "x2APIC write; got 0x%x, want 0x%lx", got,
6815 		       want);
6816 	} else if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH) {
6817 		if (!expectation->wr_only) {
6818 			u32 got = apic_read(reg);
6819 			bool ok;
6820 
6821 			/*
6822 			 * When L1's TPR is passed through to L2, the lower
6823 			 * nibble can be lost. For example, if L2 executes
6824 			 * WRMSR(0x808, 0x78), then, L1 might read 0x70.
6825 			 *
6826 			 * Here's how the lower nibble can get lost:
6827 			 *   1. L2 executes WRMSR(0x808, 0x78).
6828 			 *   2. L2 exits to L0 with a WRMSR exit.
6829 			 *   3. L0 emulates WRMSR, by writing L1's TPR.
6830 			 *   4. L0 re-enters L2.
6831 			 *   5. L2 exits to L0 (reason doesn't matter).
6832 			 *   6. L0 reflects L2's exit to L1.
6833 			 *   7. Before entering L1, L0 exits to user-space
6834 			 *      (e.g., to satisfy TPR access reporting).
6835 			 *   8. User-space executes KVM_SET_REGS ioctl, which
6836 			 *      clears the lower nibble of L1's TPR.
6837 			 */
6838 			if (reg == APIC_TASKPRI) {
6839 				got = apic_virt_nibble1(got);
6840 				val = apic_virt_nibble1(val);
6841 			}
6842 
6843 			ok = got == val;
6844 			report(ok,
6845 			       "non-virtualized write; val is 0x%x, want 0x%lx",
6846 			       got, val);
6847 			apic_write(reg, restore_val);
6848 		} else {
6849 			report_pass("non-virtualized and write-only OK");
6850 		}
6851 	}
6852 	skip_exit_insn();
6853 
6854 	report_prefix_pop();
6855 }
6856 
6857 static enum Config_type configure_virt_x2apic_mode_test(
6858 	struct virt_x2apic_mode_config *virt_x2apic_mode_config,
6859 	u8 *msr_bitmap_page)
6860 {
6861 	int msr;
6862 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6863 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6864 
6865 	/* x2apic-specific VMCS config */
6866 	if (virt_x2apic_mode_config->use_msr_bitmaps) {
6867 		/* virt_x2apic_mode_test() checks for MSR bitmaps support */
6868 		cpu_exec_ctrl0 |= CPU_MSR_BITMAP;
6869 	} else {
6870 		cpu_exec_ctrl0 &= ~CPU_MSR_BITMAP;
6871 	}
6872 
6873 	if (virt_x2apic_mode_config->virtual_interrupt_delivery) {
6874 		if (!(ctrl_cpu_rev[1].clr & CPU_VINTD)) {
6875 			report_skip("VM-execution control \"virtual-interrupt delivery\" NOT supported.\n");
6876 			return CONFIG_TYPE_UNSUPPORTED;
6877 		}
6878 		cpu_exec_ctrl1 |= CPU_VINTD;
6879 	} else {
6880 		cpu_exec_ctrl1 &= ~CPU_VINTD;
6881 	}
6882 
6883 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6884 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6885 
6886 	/* x2APIC MSR intercepts are usually off for "Virtualize x2APIC mode" */
6887 	for (msr = 0x800; msr <= 0x8ff; msr++) {
6888 		if (virt_x2apic_mode_config->disable_x2apic_msr_intercepts) {
6889 			clear_bit(msr, msr_bitmap_page + 0x000);
6890 			clear_bit(msr, msr_bitmap_page + 0x800);
6891 		} else {
6892 			set_bit(msr, msr_bitmap_page + 0x000);
6893 			set_bit(msr, msr_bitmap_page + 0x800);
6894 		}
6895 	}
6896 
6897 	/* x2APIC mode can impact virtualization */
6898 	reset_apic();
6899 	if (!virt_x2apic_mode_config->disable_x2apic)
6900 		enable_x2apic();
6901 
6902 	return configure_apic_reg_virt_test(
6903 		&virt_x2apic_mode_config->apic_reg_virt_config);
6904 }
6905 
6906 static void virt_x2apic_mode_test(void)
6907 {
6908 	u32 *virtual_apic_page;
6909 	u8 *msr_bitmap_page;
6910 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6911 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6912 	int i;
6913 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6914 
6915 	if (!cpu_has_apicv()) {
6916 		report_skip(__func__);
6917 		return;
6918 	}
6919 
6920 	/*
6921 	 * This is to exercise an issue in KVM's logic to merge L0's and L1's
6922 	 * MSR bitmaps. Previously, an L1 could get at L0's x2APIC MSRs by
6923 	 * writing the IA32_SPEC_CTRL MSR or the IA32_PRED_CMD MSRs. KVM would
6924 	 * then proceed to manipulate the MSR bitmaps, as if VMCS12 had the
6925 	 * "Virtualize x2APIC mod" control set, even when it didn't.
6926 	 */
6927 	if (has_spec_ctrl())
6928 		wrmsr(MSR_IA32_SPEC_CTRL, 1);
6929 
6930 	/*
6931 	 * Check that VMCS12 supports:
6932 	 *   - "Virtual-APIC address", indicated by "use TPR shadow"
6933 	 *   - "MSR-bitmap address", indicated by "use MSR bitmaps"
6934 	 */
6935 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6936 		report_skip("VM-execution control \"use TPR shadow\" NOT supported.\n");
6937 		return;
6938 	} else if (!(ctrl_cpu_rev[0].clr & CPU_MSR_BITMAP)) {
6939 		report_skip("VM-execution control \"use MSR bitmaps\" NOT supported.\n");
6940 		return;
6941 	}
6942 
6943 	test_set_guest(virt_x2apic_mode_guest);
6944 
6945 	virtual_apic_page = alloc_page();
6946 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6947 
6948 	msr_bitmap_page = alloc_page();
6949 	memset(msr_bitmap_page, 0xff, PAGE_SIZE);
6950 	vmcs_write(MSR_BITMAP, virt_to_phys(msr_bitmap_page));
6951 
6952 	for (i = 0; i < ARRAY_SIZE(virt_x2apic_mode_tests); i++) {
6953 		struct virt_x2apic_mode_test_case *virt_x2apic_mode_test_case =
6954 			&virt_x2apic_mode_tests[i];
6955 		struct virt_x2apic_mode_config *virt_x2apic_mode_config =
6956 			&virt_x2apic_mode_test_case->virt_x2apic_mode_config;
6957 		enum Config_type config_type;
6958 		u32 reg;
6959 
6960 		printf("--- %s test ---\n", virt_x2apic_mode_test_case->name);
6961 		config_type =
6962 			configure_virt_x2apic_mode_test(virt_x2apic_mode_config,
6963 							msr_bitmap_page);
6964 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6965 			report_skip("Skip because of missing features.\n");
6966 			continue;
6967 		} else if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6968 			enter_guest_with_bad_controls();
6969 			continue;
6970 		}
6971 
6972 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6973 			struct virt_x2apic_mode_expectation expectation;
6974 
6975 			virt_x2apic_mode_exit_expectation(
6976 				reg, virt_x2apic_mode_config, &expectation);
6977 
6978 			test_x2apic_rd(reg, &expectation, virtual_apic_page);
6979 			test_x2apic_wr(reg, &expectation, virtual_apic_page);
6980 		}
6981 	}
6982 
6983 
6984 	/* Terminate the guest */
6985 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6986 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6987 	args->op = X2APIC_TERMINATE;
6988 	enter_guest();
6989 	assert_exit_reason(VMX_VMCALL);
6990 }
6991 
6992 static void test_ctl_reg(const char *cr_name, u64 cr, u64 fixed0, u64 fixed1)
6993 {
6994 	u64 val;
6995 	u64 cr_saved = vmcs_read(cr);
6996 	int i;
6997 
6998 	val = fixed0 & fixed1;
6999 	if (cr == HOST_CR4)
7000 		vmcs_write(cr, val | X86_CR4_PAE);
7001 	else
7002 		vmcs_write(cr, val);
7003 	report_prefix_pushf("%s %lx", cr_name, val);
7004 	if (val == fixed0)
7005 		test_vmx_vmlaunch(0);
7006 	else
7007 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7008 	report_prefix_pop();
7009 
7010 	for (i = 0; i < 64; i++) {
7011 
7012 		/* Set a bit when the corresponding bit in fixed1 is 0 */
7013 		if ((fixed1 & (1ull << i)) == 0) {
7014 			if (cr == HOST_CR4 && ((1ull << i) & X86_CR4_SMEP ||
7015 					       (1ull << i) & X86_CR4_SMAP))
7016 				continue;
7017 
7018 			vmcs_write(cr, cr_saved | (1ull << i));
7019 			report_prefix_pushf("%s %llx", cr_name,
7020 						cr_saved | (1ull << i));
7021 			test_vmx_vmlaunch(
7022 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7023 			report_prefix_pop();
7024 		}
7025 
7026 		/* Unset a bit when the corresponding bit in fixed0 is 1 */
7027 		if (fixed0 & (1ull << i)) {
7028 			vmcs_write(cr, cr_saved & ~(1ull << i));
7029 			report_prefix_pushf("%s %llx", cr_name,
7030 						cr_saved & ~(1ull << i));
7031 			test_vmx_vmlaunch(
7032 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7033 			report_prefix_pop();
7034 		}
7035 	}
7036 
7037 	vmcs_write(cr, cr_saved);
7038 }
7039 
7040 /*
7041  * 1. The CR0 field must not set any bit to a value not supported in VMX
7042  *    operation.
7043  * 2. The CR4 field must not set any bit to a value not supported in VMX
7044  *    operation.
7045  * 3. On processors that support Intel 64 architecture, the CR3 field must
7046  *    be such that bits 63:52 and bits in the range 51:32 beyond the
7047  *    processor's physical-address width must be 0.
7048  *
7049  *  [Intel SDM]
7050  */
7051 static void test_host_ctl_regs(void)
7052 {
7053 	u64 fixed0, fixed1, cr3, cr3_saved;
7054 	int i;
7055 
7056 	/* Test CR0 */
7057 	fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0);
7058 	fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1);
7059 	test_ctl_reg("HOST_CR0", HOST_CR0, fixed0, fixed1);
7060 
7061 	/* Test CR4 */
7062 	fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0);
7063 	fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1) &
7064 		 ~(X86_CR4_SMEP | X86_CR4_SMAP);
7065 	test_ctl_reg("HOST_CR4", HOST_CR4, fixed0, fixed1);
7066 
7067 	/* Test CR3 */
7068 	cr3_saved = vmcs_read(HOST_CR3);
7069 	for (i = cpuid_maxphyaddr(); i < 64; i++) {
7070 		cr3 = cr3_saved | (1ul << i);
7071 		vmcs_write(HOST_CR3, cr3);
7072 		report_prefix_pushf("HOST_CR3 %lx", cr3);
7073 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7074 		report_prefix_pop();
7075 	}
7076 
7077 	vmcs_write(HOST_CR3, cr3_saved);
7078 }
7079 
7080 static void test_efer_vmlaunch(u32 fld, bool ok)
7081 {
7082 	if (fld == HOST_EFER) {
7083 		if (ok)
7084 			test_vmx_vmlaunch(0);
7085 		else
7086 			test_vmx_vmlaunch2(VMXERR_ENTRY_INVALID_CONTROL_FIELD,
7087 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7088 	} else {
7089 		test_guest_state("EFER test", !ok, GUEST_EFER, "GUEST_EFER");
7090 	}
7091 }
7092 
7093 static void test_efer_one(u32 fld, const char * fld_name, u64 efer,
7094 			  u32 ctrl_fld, u64 ctrl,
7095 			  int i, const char *efer_bit_name)
7096 {
7097 	bool ok;
7098 
7099 	ok = true;
7100 	if (ctrl_fld == EXI_CONTROLS && (ctrl & EXI_LOAD_EFER)) {
7101 		if (!!(efer & EFER_LMA) != !!(ctrl & EXI_HOST_64))
7102 			ok = false;
7103 		if (!!(efer & EFER_LME) != !!(ctrl & EXI_HOST_64))
7104 			ok = false;
7105 	}
7106 	if (ctrl_fld == ENT_CONTROLS && (ctrl & ENT_LOAD_EFER)) {
7107 		/* Check LMA too since CR0.PG is set.  */
7108 		if (!!(efer & EFER_LMA) != !!(ctrl & ENT_GUEST_64))
7109 			ok = false;
7110 		if (!!(efer & EFER_LME) != !!(ctrl & ENT_GUEST_64))
7111 			ok = false;
7112 	}
7113 
7114 	/*
7115 	 * Skip the test if it would enter the guest in 32-bit mode.
7116 	 * Perhaps write the test in assembly and make sure it
7117 	 * can be run in either mode?
7118 	 */
7119 	if (fld == GUEST_EFER && ok && !(ctrl & ENT_GUEST_64))
7120 		return;
7121 
7122 	vmcs_write(ctrl_fld, ctrl);
7123 	vmcs_write(fld, efer);
7124 	report_prefix_pushf("%s %s bit turned %s, controls %s",
7125 			    fld_name, efer_bit_name,
7126 			    (i & 1) ? "on" : "off",
7127 			    (i & 2) ? "on" : "off");
7128 
7129 	test_efer_vmlaunch(fld, ok);
7130 	report_prefix_pop();
7131 }
7132 
7133 static void test_efer_bit(u32 fld, const char * fld_name,
7134 			  u32 ctrl_fld, u64 ctrl_bit, u64 efer_bit,
7135 			  const char *efer_bit_name)
7136 {
7137 	u64 efer_saved = vmcs_read(fld);
7138 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7139 	int i;
7140 
7141 	for (i = 0; i < 4; i++) {
7142 		u64 efer = efer_saved & ~efer_bit;
7143 		u64 ctrl = ctrl_saved & ~ctrl_bit;
7144 
7145 		if (i & 1)
7146 			efer |= efer_bit;
7147 		if (i & 2)
7148 			ctrl |= ctrl_bit;
7149 
7150 		test_efer_one(fld, fld_name, efer, ctrl_fld, ctrl,
7151 			      i, efer_bit_name);
7152 	}
7153 
7154 	vmcs_write(ctrl_fld, ctrl_saved);
7155 	vmcs_write(fld, efer_saved);
7156 }
7157 
7158 static void test_efer(u32 fld, const char * fld_name, u32 ctrl_fld,
7159 		      u64 ctrl_bit1, u64 ctrl_bit2)
7160 {
7161 	u64 efer_saved = vmcs_read(fld);
7162 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7163 	u64 efer_reserved_bits =  ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
7164 	u64 i;
7165 	u64 efer;
7166 
7167 	if (cpu_has_efer_nx())
7168 		efer_reserved_bits &= ~EFER_NX;
7169 
7170 	if (!ctrl_bit1) {
7171 		printf("\"Load-IA32-EFER\" exit control not supported\n");
7172 		goto test_entry_exit_mode;
7173 	}
7174 
7175 	report_prefix_pushf("%s %lx", fld_name, efer_saved);
7176 	test_efer_vmlaunch(fld, true);
7177 	report_prefix_pop();
7178 
7179 	/*
7180 	 * Check reserved bits
7181 	 */
7182 	vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit1);
7183 	for (i = 0; i < 64; i++) {
7184 		if ((1ull << i) & efer_reserved_bits) {
7185 			efer = efer_saved | (1ull << i);
7186 			vmcs_write(fld, efer);
7187 			report_prefix_pushf("%s %lx", fld_name, efer);
7188 			test_efer_vmlaunch(fld, true);
7189 			report_prefix_pop();
7190 		}
7191 	}
7192 
7193 	vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit1);
7194 	for (i = 0; i < 64; i++) {
7195 		if ((1ull << i) & efer_reserved_bits) {
7196 			efer = efer_saved | (1ull << i);
7197 			vmcs_write(fld, efer);
7198 			report_prefix_pushf("%s %lx", fld_name, efer);
7199 			test_efer_vmlaunch(fld, false);
7200 			report_prefix_pop();
7201 		}
7202 	}
7203 
7204 	vmcs_write(ctrl_fld, ctrl_saved);
7205 	vmcs_write(fld, efer_saved);
7206 
7207 	/*
7208 	 * Check LMA and LME bits
7209 	 */
7210 	test_efer_bit(fld, fld_name,
7211 		      ctrl_fld, ctrl_bit1,
7212 		      EFER_LMA,
7213 		      "EFER_LMA");
7214 	test_efer_bit(fld, fld_name,
7215 		      ctrl_fld, ctrl_bit1,
7216 		      EFER_LME,
7217 		      "EFER_LME");
7218 
7219 test_entry_exit_mode:
7220 	test_efer_bit(fld, fld_name,
7221 		      ctrl_fld, ctrl_bit2,
7222 		      EFER_LMA,
7223 		      "EFER_LMA");
7224 	test_efer_bit(fld, fld_name,
7225 		      ctrl_fld, ctrl_bit2,
7226 		      EFER_LME,
7227 		      "EFER_LME");
7228 }
7229 
7230 /*
7231  * If the 'load IA32_EFER' VM-exit control is 1, bits reserved in the
7232  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7233  * the values of the LMA and LME bits in the field must each be that of
7234  * the 'host address-space size' VM-exit control.
7235  *
7236  *  [Intel SDM]
7237  */
7238 static void test_host_efer(void)
7239 {
7240 	test_efer(HOST_EFER, "HOST_EFER", EXI_CONTROLS,
7241 		  ctrl_exit_rev.clr & EXI_LOAD_EFER,
7242 		  EXI_HOST_64);
7243 }
7244 
7245 /*
7246  * If the 'load IA32_EFER' VM-enter control is 1, bits reserved in the
7247  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7248  * the values of the LMA and LME bits in the field must each be that of
7249  * the 'IA32e-mode guest' VM-exit control.
7250  */
7251 static void test_guest_efer(void)
7252 {
7253 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER)) {
7254 		printf("\"Load-IA32-EFER\" entry control not supported\n");
7255 		return;
7256 	}
7257 
7258 	vmcs_write(GUEST_EFER, rdmsr(MSR_EFER));
7259 	test_efer(GUEST_EFER, "GUEST_EFER", ENT_CONTROLS,
7260 		  ctrl_enter_rev.clr & ENT_LOAD_EFER,
7261 		  ENT_GUEST_64);
7262 }
7263 
7264 /*
7265  * PAT values higher than 8 are uninteresting since they're likely lumped
7266  * in with "8". We only test values above 8 one bit at a time,
7267  * in order to reduce the number of VM-Entries and keep the runtime reasonable.
7268  */
7269 #define	PAT_VAL_LIMIT	8
7270 
7271 static void test_pat(u32 field, const char * field_name, u32 ctrl_field,
7272 		     u64 ctrl_bit)
7273 {
7274 	u32 ctrl_saved = vmcs_read(ctrl_field);
7275 	u64 pat_saved = vmcs_read(field);
7276 	u64 i, val;
7277 	u32 j;
7278 	int error;
7279 
7280 	vmcs_clear_bits(ctrl_field, ctrl_bit);
7281 
7282 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7283 		/* Test PAT0..PAT7 fields */
7284 		for (j = 0; j < (i ? 8 : 1); j++) {
7285 			val = i << j * 8;
7286 			vmcs_write(field, val);
7287 			if (field == HOST_PAT) {
7288 				report_prefix_pushf("%s %lx", field_name, val);
7289 				test_vmx_vmlaunch(0);
7290 				report_prefix_pop();
7291 
7292 			} else {	// GUEST_PAT
7293 				test_guest_state("ENT_LOAD_PAT enabled", false,
7294 						 val, "GUEST_PAT");
7295 			}
7296 		}
7297 	}
7298 
7299 	vmcs_set_bits(ctrl_field, ctrl_bit);
7300 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7301 		/* Test PAT0..PAT7 fields */
7302 		for (j = 0; j < (i ? 8 : 1); j++) {
7303 			val = i << j * 8;
7304 			vmcs_write(field, val);
7305 
7306 			if (field == HOST_PAT) {
7307 				report_prefix_pushf("%s %lx", field_name, val);
7308 				if (i == 0x2 || i == 0x3 || i >= 0x8)
7309 					error =
7310 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
7311 				else
7312 					error = 0;
7313 
7314 				test_vmx_vmlaunch(error);
7315 				report_prefix_pop();
7316 
7317 			} else {	// GUEST_PAT
7318 				error = (i == 0x2 || i == 0x3 || i >= 0x8);
7319 				test_guest_state("ENT_LOAD_PAT enabled", !!error,
7320 						 val, "GUEST_PAT");
7321 			}
7322 
7323 		}
7324 	}
7325 
7326 	vmcs_write(ctrl_field, ctrl_saved);
7327 	vmcs_write(field, pat_saved);
7328 }
7329 
7330 /*
7331  *  If the "load IA32_PAT" VM-exit control is 1, the value of the field
7332  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7333  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7334  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7335  *  6 (WB), or 7 (UC-).
7336  *
7337  *  [Intel SDM]
7338  */
7339 static void test_load_host_pat(void)
7340 {
7341 	/*
7342 	 * "load IA32_PAT" VM-exit control
7343 	 */
7344 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) {
7345 		printf("\"Load-IA32-PAT\" exit control not supported\n");
7346 		return;
7347 	}
7348 
7349 	test_pat(HOST_PAT, "HOST_PAT", EXI_CONTROLS, EXI_LOAD_PAT);
7350 }
7351 
7352 union cpuidA_eax {
7353 	struct {
7354 		unsigned int version_id:8;
7355 		unsigned int num_counters_gp:8;
7356 		unsigned int bit_width:8;
7357 		unsigned int mask_length:8;
7358 	} split;
7359 	unsigned int full;
7360 };
7361 
7362 union cpuidA_edx {
7363 	struct {
7364 		unsigned int num_counters_fixed:5;
7365 		unsigned int bit_width_fixed:8;
7366 		unsigned int reserved:9;
7367 	} split;
7368 	unsigned int full;
7369 };
7370 
7371 static bool valid_pgc(u64 val)
7372 {
7373 	struct cpuid id;
7374 	union cpuidA_eax eax;
7375 	union cpuidA_edx edx;
7376 	u64 mask;
7377 
7378 	id = cpuid(0xA);
7379 	eax.full = id.a;
7380 	edx.full = id.d;
7381 	mask = ~(((1ull << eax.split.num_counters_gp) - 1) |
7382 		 (((1ull << edx.split.num_counters_fixed) - 1) << 32));
7383 
7384 	return !(val & mask);
7385 }
7386 
7387 static void test_pgc_vmlaunch(u32 xerror, u32 xreason, bool xfail, bool host)
7388 {
7389 	u32 inst_err;
7390 	u64 obs;
7391 	bool success;
7392 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7393 
7394 	if (host) {
7395 		success = vmlaunch_succeeds();
7396 		obs = rdmsr(data->msr);
7397 		if (!success) {
7398 			inst_err = vmcs_read(VMX_INST_ERROR);
7399 			report(xerror == inst_err, "vmlaunch failed, "
7400 			       "VMX Inst Error is %d (expected %d)",
7401 			       inst_err, xerror);
7402 		} else {
7403 			report(!data->enabled || data->exp == obs,
7404 			       "Host state is 0x%lx (expected 0x%lx)",
7405 			       obs, data->exp);
7406 			report(success != xfail, "vmlaunch succeeded");
7407 		}
7408 	} else {
7409 		test_guest_state("load GUEST_PERF_GLOBAL_CTRL", xfail,
7410 				 GUEST_PERF_GLOBAL_CTRL,
7411 				 "GUEST_PERF_GLOBAL_CTRL");
7412 	}
7413 }
7414 
7415 /*
7416  * test_load_perf_global_ctrl is a generic function for testing the
7417  * "load IA32_PERF_GLOBAL_CTRL" VM-{Entry,Exit} controls. This test function
7418  * tests the provided ctrl_val when disabled and enabled.
7419  *
7420  * @nr: VMCS field number corresponding to the host/guest state field
7421  * @name: Name of the above VMCS field for printing in test report
7422  * @ctrl_nr: VMCS field number corresponding to the VM-{Entry,Exit} control
7423  * @ctrl_val: Bit to set on the ctrl_field
7424  */
7425 static void test_perf_global_ctrl(u32 nr, const char *name, u32 ctrl_nr,
7426 				  const char *ctrl_name, u64 ctrl_val)
7427 {
7428 	u64 ctrl_saved = vmcs_read(ctrl_nr);
7429 	u64 pgc_saved = vmcs_read(nr);
7430 	u64 i, val;
7431 	bool host = nr == HOST_PERF_GLOBAL_CTRL;
7432 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7433 
7434 	data->msr = MSR_CORE_PERF_GLOBAL_CTRL;
7435 	msr_bmp_init();
7436 	vmcs_write(ctrl_nr, ctrl_saved & ~ctrl_val);
7437 	data->enabled = false;
7438 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=0 on %s",
7439 			    ctrl_name);
7440 
7441 	for (i = 0; i < 64; i++) {
7442 		val = 1ull << i;
7443 		vmcs_write(nr, val);
7444 		report_prefix_pushf("%s = 0x%lx", name, val);
7445 		test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7446 		report_prefix_pop();
7447 	}
7448 	report_prefix_pop();
7449 
7450 	vmcs_write(ctrl_nr, ctrl_saved | ctrl_val);
7451 	data->enabled = true;
7452 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=1 on %s",
7453 			    ctrl_name);
7454 	for (i = 0; i < 64; i++) {
7455 		val = 1ull << i;
7456 		data->exp = val;
7457 		vmcs_write(nr, val);
7458 		report_prefix_pushf("%s = 0x%lx", name, val);
7459 		if (valid_pgc(val)) {
7460 			test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7461 		} else {
7462 			if (host)
7463 				test_pgc_vmlaunch(
7464 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
7465 					0,
7466 					true,
7467 					host);
7468 			else
7469 				test_pgc_vmlaunch(
7470 					0,
7471 					VMX_ENTRY_FAILURE | VMX_FAIL_STATE,
7472 					true,
7473 					host);
7474 		}
7475 		report_prefix_pop();
7476 	}
7477 
7478 	data->enabled = false;
7479 	report_prefix_pop();
7480 	vmcs_write(ctrl_nr, ctrl_saved);
7481 	vmcs_write(nr, pgc_saved);
7482 }
7483 
7484 static void test_load_host_perf_global_ctrl(void)
7485 {
7486 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PERF)) {
7487 		printf("\"load IA32_PERF_GLOBAL_CTRL\" exit control not supported\n");
7488 		return;
7489 	}
7490 
7491 	test_perf_global_ctrl(HOST_PERF_GLOBAL_CTRL, "HOST_PERF_GLOBAL_CTRL",
7492 				   EXI_CONTROLS, "EXI_CONTROLS", EXI_LOAD_PERF);
7493 }
7494 
7495 
7496 static void test_load_guest_perf_global_ctrl(void)
7497 {
7498 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PERF)) {
7499 		printf("\"load IA32_PERF_GLOBAL_CTRL\" entry control not supported\n");
7500 		return;
7501 	}
7502 
7503 	test_perf_global_ctrl(GUEST_PERF_GLOBAL_CTRL, "GUEST_PERF_GLOBAL_CTRL",
7504 				   ENT_CONTROLS, "ENT_CONTROLS", ENT_LOAD_PERF);
7505 }
7506 
7507 
7508 /*
7509  * test_vmcs_field - test a value for the given VMCS field
7510  * @field: VMCS field
7511  * @field_name: string name of VMCS field
7512  * @bit_start: starting bit
7513  * @bit_end: ending bit
7514  * @val: value that the bit range must or must not contain
7515  * @valid_val: whether value given in 'val' must be valid or not
7516  * @error: expected VMCS error when vmentry fails for an invalid value
7517  */
7518 static void test_vmcs_field(u64 field, const char *field_name, u32 bit_start,
7519 			    u32 bit_end, u64 val, bool valid_val, u32 error)
7520 {
7521 	u64 field_saved = vmcs_read(field);
7522 	u32 i;
7523 	u64 tmp;
7524 	u32 bit_on;
7525 	u64 mask = ~0ull;
7526 
7527 	mask = (mask >> bit_end) << bit_end;
7528 	mask = mask | ((1 << bit_start) - 1);
7529 	tmp = (field_saved & mask) | (val << bit_start);
7530 
7531 	vmcs_write(field, tmp);
7532 	report_prefix_pushf("%s %lx", field_name, tmp);
7533 	if (valid_val)
7534 		test_vmx_vmlaunch(0);
7535 	else
7536 		test_vmx_vmlaunch(error);
7537 	report_prefix_pop();
7538 
7539 	for (i = bit_start; i <= bit_end; i = i + 2) {
7540 		bit_on = ((1ull < i) & (val << bit_start)) ? 0 : 1;
7541 		if (bit_on)
7542 			tmp = field_saved | (1ull << i);
7543 		else
7544 			tmp = field_saved & ~(1ull << i);
7545 		vmcs_write(field, tmp);
7546 		report_prefix_pushf("%s %lx", field_name, tmp);
7547 		if (valid_val)
7548 			test_vmx_vmlaunch(error);
7549 		else
7550 			test_vmx_vmlaunch(0);
7551 		report_prefix_pop();
7552 	}
7553 
7554 	vmcs_write(field, field_saved);
7555 }
7556 
7557 static void test_canonical(u64 field, const char * field_name, bool host)
7558 {
7559 	u64 addr_saved = vmcs_read(field);
7560 
7561 	/*
7562 	 * Use the existing value if possible.  Writing a random canonical
7563 	 * value is not an option as doing so would corrupt the field being
7564 	 * tested and likely hose the test.
7565 	 */
7566 	if (is_canonical(addr_saved)) {
7567 		if (host) {
7568 			report_prefix_pushf("%s %lx", field_name, addr_saved);
7569 			test_vmx_vmlaunch(0);
7570 			report_prefix_pop();
7571 		} else {
7572 			test_guest_state("Test canonical address", false,
7573 					 addr_saved, field_name);
7574 		}
7575 	}
7576 
7577 	vmcs_write(field, NONCANONICAL);
7578 
7579 	if (host) {
7580 		report_prefix_pushf("%s %llx", field_name, NONCANONICAL);
7581 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7582 		report_prefix_pop();
7583 	} else {
7584 		test_guest_state("Test non-canonical address", true,
7585 				 NONCANONICAL, field_name);
7586 	}
7587 
7588 	vmcs_write(field, addr_saved);
7589 }
7590 
7591 #define TEST_RPL_TI_FLAGS(reg, name)				\
7592 	test_vmcs_field(reg, name, 0, 2, 0x0, true,		\
7593 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7594 
7595 #define TEST_CS_TR_FLAGS(reg, name)				\
7596 	test_vmcs_field(reg, name, 3, 15, 0x0000, false,	\
7597 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7598 
7599 /*
7600  * 1. In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the
7601  *    RPL (bits 1:0) and the TI flag (bit 2) must be 0.
7602  * 2. The selector fields for CS and TR cannot be 0000H.
7603  * 3. The selector field for SS cannot be 0000H if the "host address-space
7604  *    size" VM-exit control is 0.
7605  * 4. On processors that support Intel 64 architecture, the base-address
7606  *    fields for FS, GS and TR must contain canonical addresses.
7607  */
7608 static void test_host_segment_regs(void)
7609 {
7610 	u16 selector_saved;
7611 
7612 	/*
7613 	 * Test RPL and TI flags
7614 	 */
7615 	TEST_RPL_TI_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7616 	TEST_RPL_TI_FLAGS(HOST_SEL_SS, "HOST_SEL_SS");
7617 	TEST_RPL_TI_FLAGS(HOST_SEL_DS, "HOST_SEL_DS");
7618 	TEST_RPL_TI_FLAGS(HOST_SEL_ES, "HOST_SEL_ES");
7619 	TEST_RPL_TI_FLAGS(HOST_SEL_FS, "HOST_SEL_FS");
7620 	TEST_RPL_TI_FLAGS(HOST_SEL_GS, "HOST_SEL_GS");
7621 	TEST_RPL_TI_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7622 
7623 	/*
7624 	 * Test that CS and TR fields can not be 0x0000
7625 	 */
7626 	TEST_CS_TR_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7627 	TEST_CS_TR_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7628 
7629 	/*
7630 	 * SS field can not be 0x0000 if "host address-space size" VM-exit
7631 	 * control is 0
7632 	 */
7633 	selector_saved = vmcs_read(HOST_SEL_SS);
7634 	vmcs_write(HOST_SEL_SS, 0);
7635 	report_prefix_pushf("HOST_SEL_SS 0");
7636 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7637 		test_vmx_vmlaunch(0);
7638 	} else {
7639 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7640 	}
7641 	report_prefix_pop();
7642 
7643 	vmcs_write(HOST_SEL_SS, selector_saved);
7644 
7645 	/*
7646 	 * Base address for FS, GS and TR must be canonical
7647 	 */
7648 	test_canonical(HOST_BASE_FS, "HOST_BASE_FS", true);
7649 	test_canonical(HOST_BASE_GS, "HOST_BASE_GS", true);
7650 	test_canonical(HOST_BASE_TR, "HOST_BASE_TR", true);
7651 }
7652 
7653 /*
7654  *  On processors that support Intel 64 architecture, the base-address
7655  *  fields for GDTR and IDTR must contain canonical addresses.
7656  */
7657 static void test_host_desc_tables(void)
7658 {
7659 	test_canonical(HOST_BASE_GDTR, "HOST_BASE_GDTR", true);
7660 	test_canonical(HOST_BASE_IDTR, "HOST_BASE_IDTR", true);
7661 }
7662 
7663 /*
7664  * If the "host address-space size" VM-exit control is 0, the following must
7665  * hold:
7666  *    - The "IA-32e mode guest" VM-entry control is 0.
7667  *    - Bit 17 of the CR4 field (corresponding to CR4.PCIDE) is 0.
7668  *    - Bits 63:32 in the RIP field are 0.
7669  *
7670  * If the "host address-space size" VM-exit control is 1, the following must
7671  * hold:
7672  *    - Bit 5 of the CR4 field (corresponding to CR4.PAE) is 1.
7673  *    - The RIP field contains a canonical address.
7674  *
7675  */
7676 static void test_host_addr_size(void)
7677 {
7678 	u64 cr4_saved = vmcs_read(HOST_CR4);
7679 	u64 rip_saved = vmcs_read(HOST_RIP);
7680 	u64 entry_ctrl_saved = vmcs_read(ENT_CONTROLS);
7681 	int i;
7682 	u64 tmp;
7683 
7684 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7685 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7686 		report_prefix_pushf("\"IA-32e mode guest\" enabled");
7687 		test_vmx_vmlaunch(0);
7688 		report_prefix_pop();
7689 
7690 		vmcs_write(HOST_CR4, cr4_saved | X86_CR4_PCIDE);
7691 		report_prefix_pushf("\"CR4.PCIDE\" set");
7692 		test_vmx_vmlaunch(0);
7693 		report_prefix_pop();
7694 
7695 		for (i = 32; i <= 63; i = i + 4) {
7696 			tmp = rip_saved | 1ull << i;
7697 			vmcs_write(HOST_RIP, tmp);
7698 			report_prefix_pushf("HOST_RIP %lx", tmp);
7699 			test_vmx_vmlaunch(0);
7700 			report_prefix_pop();
7701 		}
7702 
7703 		if (cr4_saved & X86_CR4_PAE) {
7704 			vmcs_write(HOST_CR4, cr4_saved  & ~X86_CR4_PAE);
7705 			report_prefix_pushf("\"CR4.PAE\" unset");
7706 			test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7707 		} else {
7708 			report_prefix_pushf("\"CR4.PAE\" set");
7709 			test_vmx_vmlaunch(0);
7710 		}
7711 		report_prefix_pop();
7712 
7713 		vmcs_write(HOST_RIP, NONCANONICAL);
7714 		report_prefix_pushf("HOST_RIP %llx", NONCANONICAL);
7715 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7716 		report_prefix_pop();
7717 
7718 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7719 		vmcs_write(HOST_RIP, rip_saved);
7720 		vmcs_write(HOST_CR4, cr4_saved);
7721 
7722 		/* Restore host's active RIP and CR4 values. */
7723 		report_prefix_pushf("restore host state");
7724 		test_vmx_vmlaunch(0);
7725 		report_prefix_pop();
7726 	}
7727 }
7728 
7729 /*
7730  * Check that the virtual CPU checks the VMX Host State Area as
7731  * documented in the Intel SDM.
7732  */
7733 static void vmx_host_state_area_test(void)
7734 {
7735 	/*
7736 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
7737 	 * fail due to invalid guest state, should we make it that
7738 	 * far.
7739 	 */
7740 	vmcs_write(GUEST_RFLAGS, 0);
7741 
7742 	test_host_ctl_regs();
7743 
7744 	test_canonical(HOST_SYSENTER_ESP, "HOST_SYSENTER_ESP", true);
7745 	test_canonical(HOST_SYSENTER_EIP, "HOST_SYSENTER_EIP", true);
7746 
7747 	test_host_efer();
7748 	test_load_host_pat();
7749 	test_host_segment_regs();
7750 	test_host_desc_tables();
7751 	test_host_addr_size();
7752 	test_load_host_perf_global_ctrl();
7753 }
7754 
7755 /*
7756  * If the "load debug controls" VM-entry control is 1, bits 63:32 in
7757  * the DR7 field must be 0.
7758  *
7759  * [Intel SDM]
7760  */
7761 static void test_guest_dr7(void)
7762 {
7763 	u32 ent_saved = vmcs_read(ENT_CONTROLS);
7764 	u64 dr7_saved = vmcs_read(GUEST_DR7);
7765 	u64 val;
7766 	int i;
7767 
7768 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS) {
7769 		vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7770 		for (i = 0; i < 64; i++) {
7771 			val = 1ull << i;
7772 			vmcs_write(GUEST_DR7, val);
7773 			test_guest_state("ENT_LOAD_DBGCTLS disabled", false,
7774 					 val, "GUEST_DR7");
7775 		}
7776 	}
7777 	if (ctrl_enter_rev.clr & ENT_LOAD_DBGCTLS) {
7778 		vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7779 		for (i = 0; i < 64; i++) {
7780 			val = 1ull << i;
7781 			vmcs_write(GUEST_DR7, val);
7782 			test_guest_state("ENT_LOAD_DBGCTLS enabled", i >= 32,
7783 					 val, "GUEST_DR7");
7784 		}
7785 	}
7786 	vmcs_write(GUEST_DR7, dr7_saved);
7787 	vmcs_write(ENT_CONTROLS, ent_saved);
7788 }
7789 
7790 /*
7791  *  If the "load IA32_PAT" VM-entry control is 1, the value of the field
7792  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7793  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7794  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7795  *  6 (WB), or 7 (UC-).
7796  *
7797  *  [Intel SDM]
7798  */
7799 static void test_load_guest_pat(void)
7800 {
7801 	/*
7802 	 * "load IA32_PAT" VM-entry control
7803 	 */
7804 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
7805 		printf("\"Load-IA32-PAT\" entry control not supported\n");
7806 		return;
7807 	}
7808 
7809 	test_pat(GUEST_PAT, "GUEST_PAT", ENT_CONTROLS, ENT_LOAD_PAT);
7810 }
7811 
7812 #define MSR_IA32_BNDCFGS_RSVD_MASK	0x00000ffc
7813 
7814 /*
7815  * If the "load IA32_BNDCFGS" VM-entry control is 1, the following
7816  * checks are performed on the field for the IA32_BNDCFGS MSR:
7817  *
7818  *   - Bits reserved in the IA32_BNDCFGS MSR must be 0.
7819  *   - The linear address in bits 63:12 must be canonical.
7820  *
7821  *  [Intel SDM]
7822  */
7823 static void test_load_guest_bndcfgs(void)
7824 {
7825 	u64 bndcfgs_saved = vmcs_read(GUEST_BNDCFGS);
7826 	u64 bndcfgs;
7827 
7828 	if (!(ctrl_enter_rev.clr & ENT_LOAD_BNDCFGS)) {
7829 		printf("\"Load-IA32-BNDCFGS\" entry control not supported\n");
7830 		return;
7831 	}
7832 
7833 	vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7834 
7835 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7836 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7837 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7838 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7839 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7840 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7841 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7842 
7843 	vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7844 
7845 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7846 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7847 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7848 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7849 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7850 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7851 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7852 
7853 	vmcs_write(GUEST_BNDCFGS, bndcfgs_saved);
7854 }
7855 
7856 #define	GUEST_SEG_UNUSABLE_MASK	(1u << 16)
7857 #define	GUEST_SEG_SEL_TI_MASK	(1u << 2)
7858 
7859 
7860 #define	TEST_SEGMENT_SEL(test, xfail, sel, val)				\
7861 do {									\
7862 	vmcs_write(sel, val);						\
7863 	test_guest_state(test " segment", xfail, val, xstr(sel));	\
7864 } while (0)
7865 
7866 #define	TEST_INVALID_SEG_SEL(sel, val) \
7867 	TEST_SEGMENT_SEL("Invalid: " xstr(val), true, sel, val);
7868 
7869 #define	TEST_VALID_SEG_SEL(sel, val) \
7870 	TEST_SEGMENT_SEL("Valid: " xstr(val), false, sel, val);
7871 
7872 /*
7873  * The following checks are done on the Selector field of the Guest Segment
7874  * Registers:
7875  *    - TR. The TI flag (bit 2) must be 0.
7876  *    - LDTR. If LDTR is usable, the TI flag (bit 2) must be 0.
7877  *    - SS. If the guest will not be virtual-8086 and the "unrestricted
7878  *	guest" VM-execution control is 0, the RPL (bits 1:0) must equal
7879  *	the RPL of the selector field for CS.
7880  *
7881  *  [Intel SDM]
7882  */
7883 static void test_guest_segment_sel_fields(void)
7884 {
7885 	u16 sel_saved;
7886 	u32 ar_saved;
7887 	u32 cpu_ctrl0_saved;
7888 	u32 cpu_ctrl1_saved;
7889 	u16 cs_rpl_bits;
7890 
7891 	/*
7892 	 * Test for GUEST_SEL_TR
7893 	 */
7894 	sel_saved = vmcs_read(GUEST_SEL_TR);
7895 	TEST_INVALID_SEG_SEL(GUEST_SEL_TR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7896 	vmcs_write(GUEST_SEL_TR, sel_saved);
7897 
7898 	/*
7899 	 * Test for GUEST_SEL_LDTR
7900 	 */
7901 	sel_saved = vmcs_read(GUEST_SEL_LDTR);
7902 	ar_saved = vmcs_read(GUEST_AR_LDTR);
7903 	/* LDTR is set unusable */
7904 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7905 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7906 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7907 	/* LDTR is set usable */
7908 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7909 	TEST_INVALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7910 
7911 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7912 
7913 	vmcs_write(GUEST_AR_LDTR, ar_saved);
7914 	vmcs_write(GUEST_SEL_LDTR, sel_saved);
7915 
7916 	/*
7917 	 * Test for GUEST_SEL_SS
7918 	 */
7919 	cpu_ctrl0_saved = vmcs_read(CPU_EXEC_CTRL0);
7920 	cpu_ctrl1_saved = vmcs_read(CPU_EXEC_CTRL1);
7921 	ar_saved = vmcs_read(GUEST_AR_SS);
7922 	/* Turn off "unrestricted guest" vm-execution control */
7923 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved & ~CPU_URG);
7924 	cs_rpl_bits = vmcs_read(GUEST_SEL_CS) & 0x3;
7925 	sel_saved = vmcs_read(GUEST_SEL_SS);
7926 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7927 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7928 	/* Make SS usable if it's unusable or vice-versa */
7929 	if (ar_saved & GUEST_SEG_UNUSABLE_MASK)
7930 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7931 	else
7932 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7933 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7934 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7935 
7936 	/* Need a valid EPTP as the passing case fully enters the guest. */
7937 	if (enable_unrestricted_guest(true))
7938 		goto skip_ss_tests;
7939 
7940 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7941 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7942 
7943 	/* Make SS usable if it's unusable or vice-versa */
7944 	if (vmcs_read(GUEST_AR_SS) & GUEST_SEG_UNUSABLE_MASK)
7945 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7946 	else
7947 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7948 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7949 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7950 skip_ss_tests:
7951 
7952 	vmcs_write(GUEST_AR_SS, ar_saved);
7953 	vmcs_write(GUEST_SEL_SS, sel_saved);
7954 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrl0_saved);
7955 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved);
7956 }
7957 
7958 #define	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(xfail, seg_base)			\
7959 do {										\
7960 	addr_saved = vmcs_read(seg_base);					\
7961 	for (i = 32; i < 63; i = i + 4) {					\
7962 		addr = addr_saved | 1ull << i;					\
7963 		vmcs_write(seg_base, addr);					\
7964 		test_guest_state("seg.BASE[63:32] != 0, usable = " xstr(xfail),	\
7965 				 xfail, addr, xstr(seg_base));			\
7966 	}									\
7967 	vmcs_write(seg_base, addr_saved);					\
7968 } while (0)
7969 
7970 #define	TEST_SEGMENT_BASE_ADDR_CANONICAL(xfail, seg_base)		  \
7971 do {									  \
7972 	addr_saved = vmcs_read(seg_base);				  \
7973 	vmcs_write(seg_base, NONCANONICAL);				  \
7974 	test_guest_state("seg.BASE non-canonical, usable = " xstr(xfail), \
7975 			 xfail, NONCANONICAL, xstr(seg_base));		  \
7976 	vmcs_write(seg_base, addr_saved);				  \
7977 } while (0)
7978 
7979 /*
7980  * The following checks are done on the Base Address field of the Guest
7981  * Segment Registers on processors that support Intel 64 architecture:
7982  *    - TR, FS, GS : The address must be canonical.
7983  *    - LDTR : If LDTR is usable, the address must be canonical.
7984  *    - CS : Bits 63:32 of the address must be zero.
7985  *    - SS, DS, ES : If the register is usable, bits 63:32 of the address
7986  *	must be zero.
7987  *
7988  *  [Intel SDM]
7989  */
7990 static void test_guest_segment_base_addr_fields(void)
7991 {
7992 	u64 addr_saved;
7993 	u64 addr;
7994 	u32 ar_saved;
7995 	int i;
7996 
7997 	/*
7998 	 * The address of TR, FS, GS and LDTR must be canonical.
7999 	 */
8000 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_TR);
8001 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_FS);
8002 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_GS);
8003 	ar_saved = vmcs_read(GUEST_AR_LDTR);
8004 	/* Make LDTR unusable */
8005 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8006 	TEST_SEGMENT_BASE_ADDR_CANONICAL(false, GUEST_BASE_LDTR);
8007 	/* Make LDTR usable */
8008 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8009 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_LDTR);
8010 
8011 	vmcs_write(GUEST_AR_LDTR, ar_saved);
8012 
8013 	/*
8014 	 * Bits 63:32 in CS, SS, DS and ES base address must be zero
8015 	 */
8016 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_CS);
8017 	ar_saved = vmcs_read(GUEST_AR_SS);
8018 	/* Make SS unusable */
8019 	vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8020 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_SS);
8021 	/* Make SS usable */
8022 	vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8023 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_SS);
8024 	vmcs_write(GUEST_AR_SS, ar_saved);
8025 
8026 	ar_saved = vmcs_read(GUEST_AR_DS);
8027 	/* Make DS unusable */
8028 	vmcs_write(GUEST_AR_DS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8029 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_DS);
8030 	/* Make DS usable */
8031 	vmcs_write(GUEST_AR_DS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8032 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_DS);
8033 	vmcs_write(GUEST_AR_DS, ar_saved);
8034 
8035 	ar_saved = vmcs_read(GUEST_AR_ES);
8036 	/* Make ES unusable */
8037 	vmcs_write(GUEST_AR_ES, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8038 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_ES);
8039 	/* Make ES usable */
8040 	vmcs_write(GUEST_AR_ES, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8041 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_ES);
8042 	vmcs_write(GUEST_AR_ES, ar_saved);
8043 }
8044 
8045 /*
8046  * Check that the virtual CPU checks the VMX Guest State Area as
8047  * documented in the Intel SDM.
8048  */
8049 static void vmx_guest_state_area_test(void)
8050 {
8051 	vmx_set_test_stage(1);
8052 	test_set_guest(guest_state_test_main);
8053 
8054 	/*
8055 	 * The IA32_SYSENTER_ESP field and the IA32_SYSENTER_EIP field
8056 	 * must each contain a canonical address.
8057 	 */
8058 	test_canonical(GUEST_SYSENTER_ESP, "GUEST_SYSENTER_ESP", false);
8059 	test_canonical(GUEST_SYSENTER_EIP, "GUEST_SYSENTER_EIP", false);
8060 
8061 	test_guest_dr7();
8062 	test_load_guest_pat();
8063 	test_guest_efer();
8064 	test_load_guest_perf_global_ctrl();
8065 	test_load_guest_bndcfgs();
8066 
8067 	test_guest_segment_sel_fields();
8068 	test_guest_segment_base_addr_fields();
8069 
8070 	test_canonical(GUEST_BASE_GDTR, "GUEST_BASE_GDTR", false);
8071 	test_canonical(GUEST_BASE_IDTR, "GUEST_BASE_IDTR", false);
8072 
8073 	u32 guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_GDTR);
8074 	int i;
8075 	for (i = 16; i <= 31; i++) {
8076 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8077 		vmcs_write(GUEST_LIMIT_GDTR, tmp);
8078 		test_guest_state("GDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_GDTR");
8079 	}
8080 	vmcs_write(GUEST_LIMIT_GDTR, guest_desc_limit_saved);
8081 
8082 	guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_IDTR);
8083 	for (i = 16; i <= 31; i++) {
8084 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8085 		vmcs_write(GUEST_LIMIT_IDTR, tmp);
8086 		test_guest_state("IDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_IDTR");
8087 	}
8088 	vmcs_write(GUEST_LIMIT_IDTR, guest_desc_limit_saved);
8089 
8090 	/*
8091 	 * Let the guest finish execution
8092 	 */
8093 	vmx_set_test_stage(2);
8094 	enter_guest();
8095 }
8096 
8097 extern void unrestricted_guest_main(void);
8098 asm (".code32\n"
8099 	"unrestricted_guest_main:\n"
8100 	"vmcall\n"
8101 	"nop\n"
8102 	"mov $1, %edi\n"
8103 	"call hypercall\n"
8104 	".code64\n");
8105 
8106 static void setup_unrestricted_guest(void)
8107 {
8108 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) & ~(X86_CR0_PG));
8109 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) & ~ENT_GUEST_64);
8110 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) & ~EFER_LMA);
8111 	vmcs_write(GUEST_RIP, virt_to_phys(unrestricted_guest_main));
8112 }
8113 
8114 static void unsetup_unrestricted_guest(void)
8115 {
8116 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) | X86_CR0_PG);
8117 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_GUEST_64);
8118 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) | EFER_LMA);
8119 	vmcs_write(GUEST_RIP, (u64) phys_to_virt(vmcs_read(GUEST_RIP)));
8120 	vmcs_write(GUEST_RSP, (u64) phys_to_virt(vmcs_read(GUEST_RSP)));
8121 }
8122 
8123 /*
8124  * If "unrestricted guest" secondary VM-execution control is set, guests
8125  * can run in unpaged protected mode.
8126  */
8127 static void vmentry_unrestricted_guest_test(void)
8128 {
8129 	if (enable_unrestricted_guest(true)) {
8130 		report_skip("Unrestricted guest not supported");
8131 		return;
8132 	}
8133 
8134 	test_set_guest(unrestricted_guest_main);
8135 	setup_unrestricted_guest();
8136 	test_guest_state("Unrestricted guest test", false, CPU_URG, "CPU_URG");
8137 
8138 	/*
8139 	 * Let the guest finish execution as a regular guest
8140 	 */
8141 	unsetup_unrestricted_guest();
8142 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) & ~CPU_URG);
8143 	enter_guest();
8144 }
8145 
8146 static bool valid_vmcs_for_vmentry(void)
8147 {
8148 	struct vmcs *current_vmcs = NULL;
8149 
8150 	if (vmcs_save(&current_vmcs))
8151 		return false;
8152 
8153 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
8154 }
8155 
8156 static void try_vmentry_in_movss_shadow(void)
8157 {
8158 	u32 vm_inst_err;
8159 	u32 flags;
8160 	bool early_failure = false;
8161 	u32 expected_flags = X86_EFLAGS_FIXED;
8162 	bool valid_vmcs = valid_vmcs_for_vmentry();
8163 
8164 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
8165 
8166 	/*
8167 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
8168 	 * unsupported VMCS component").
8169 	 */
8170 	vmcs_write(~0u, 0);
8171 
8172 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
8173 			      "vmwrite %%rsp, %%rdx;"
8174 			      "mov 0f, %%rax;"
8175 			      "mov %[host_rip], %%edx;"
8176 			      "vmwrite %%rax, %%rdx;"
8177 			      "mov $-1, %%ah;"
8178 			      "sahf;"
8179 			      "mov %%ss, %%ax;"
8180 			      "mov %%ax, %%ss;"
8181 			      "vmlaunch;"
8182 			      "mov $1, %[early_failure];"
8183 			      "0: lahf;"
8184 			      "movzbl %%ah, %[flags]"
8185 			      : [early_failure] "+r" (early_failure),
8186 				[flags] "=&a" (flags)
8187 			      : [host_rsp] "i" (HOST_RSP),
8188 				[host_rip] "i" (HOST_RIP)
8189 			      : "rdx", "cc", "memory");
8190 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
8191 
8192 	report(early_failure, "Early VM-entry failure");
8193 	report(flags == expected_flags, "RFLAGS[8:0] is %x (actual %x)",
8194 	       expected_flags, flags);
8195 	if (valid_vmcs)
8196 		report(vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
8197 		       "VM-instruction error is %d (actual %d)",
8198 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
8199 }
8200 
8201 static void vmentry_movss_shadow_test(void)
8202 {
8203 	struct vmcs *orig_vmcs;
8204 
8205 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
8206 
8207 	/*
8208 	 * Set the launched flag on the current VMCS to verify the correct
8209 	 * error priority, below.
8210 	 */
8211 	test_set_guest(v2_null_test_guest);
8212 	enter_guest();
8213 
8214 	/*
8215 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
8216 	 * fail due to invalid guest state (if we make it that far).
8217 	 */
8218 	vmcs_write(GUEST_RFLAGS, 0);
8219 
8220 	/*
8221 	 * "VM entry with events blocked by MOV SS" takes precedence over
8222 	 * "VMLAUNCH with non-clear VMCS."
8223 	 */
8224 	report_prefix_push("valid current-VMCS");
8225 	try_vmentry_in_movss_shadow();
8226 	report_prefix_pop();
8227 
8228 	/*
8229 	 * VMfailInvalid takes precedence over "VM entry with events
8230 	 * blocked by MOV SS."
8231 	 */
8232 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
8233 	report_prefix_push("no current-VMCS");
8234 	try_vmentry_in_movss_shadow();
8235 	report_prefix_pop();
8236 
8237 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
8238 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8239 }
8240 
8241 static void vmx_ldtr_test_guest(void)
8242 {
8243 	u16 ldtr = sldt();
8244 
8245 	report(ldtr == NP_SEL, "Expected %x for L2 LDTR selector (got %x)",
8246 	       NP_SEL, ldtr);
8247 }
8248 
8249 /*
8250  * Ensure that the L1 LDTR is set to 0 on VM-exit.
8251  */
8252 static void vmx_ldtr_test(void)
8253 {
8254 	const u8 ldt_ar = 0x82; /* Present LDT */
8255 	u16 sel = FIRST_SPARE_SEL;
8256 
8257 	/* Set up a non-zero L1 LDTR prior to VM-entry. */
8258 	set_gdt_entry(sel, 0, 0, ldt_ar, 0);
8259 	lldt(sel);
8260 
8261 	test_set_guest(vmx_ldtr_test_guest);
8262 	/*
8263 	 * Set up a different LDTR for L2. The actual GDT contents are
8264 	 * irrelevant, since we stuff the hidden descriptor state
8265 	 * straight into the VMCS rather than reading it from the GDT.
8266 	 */
8267 	vmcs_write(GUEST_SEL_LDTR, NP_SEL);
8268 	vmcs_write(GUEST_AR_LDTR, ldt_ar);
8269 	enter_guest();
8270 
8271 	/*
8272 	 * VM-exit should clear LDTR (and make it unusable, but we
8273 	 * won't verify that here).
8274 	 */
8275 	sel = sldt();
8276 	report(!sel, "Expected 0 for L1 LDTR selector (got %x)", sel);
8277 }
8278 
8279 static void vmx_single_vmcall_guest(void)
8280 {
8281 	vmcall();
8282 }
8283 
8284 static void vmx_cr_load_test(void)
8285 {
8286 	unsigned long cr3, cr4, orig_cr3, orig_cr4;
8287 	u32 ctrls[2] = {0};
8288 	pgd_t *pml5;
8289 
8290 	orig_cr4 = read_cr4();
8291 	orig_cr3 = read_cr3();
8292 
8293 	if (!this_cpu_has(X86_FEATURE_PCID)) {
8294 		report_skip("PCID not detected");
8295 		return;
8296 	}
8297 	if (!this_cpu_has(X86_FEATURE_MCE)) {
8298 		report_skip("MCE not detected");
8299 		return;
8300 	}
8301 
8302 	TEST_ASSERT(!(orig_cr3 & X86_CR3_PCID_MASK));
8303 
8304 	/* Enable PCID for L1. */
8305 	cr4 = orig_cr4 | X86_CR4_PCIDE;
8306 	cr3 = orig_cr3 | 0x1;
8307 	TEST_ASSERT(!write_cr4_checking(cr4));
8308 	write_cr3(cr3);
8309 
8310 	test_set_guest(vmx_single_vmcall_guest);
8311 	vmcs_write(HOST_CR4, cr4);
8312 	vmcs_write(HOST_CR3, cr3);
8313 	enter_guest();
8314 
8315 	/*
8316 	 * No exception is expected.
8317 	 *
8318 	 * NB. KVM loads the last guest write to CR4 into CR4 read
8319 	 *     shadow. In order to trigger an exit to KVM, we can toggle a
8320 	 *     bit that is owned by KVM. We use CR4.MCE, which shall
8321 	 *     have no side effect because normally no guest MCE (e.g., as the
8322 	 *     result of bad memory) would happen during this test.
8323 	 */
8324 	TEST_ASSERT(!write_cr4_checking(cr4 ^ X86_CR4_MCE));
8325 
8326 	/* Cleanup L1 state. */
8327 	write_cr3(orig_cr3);
8328 	TEST_ASSERT(!write_cr4_checking(orig_cr4));
8329 
8330 	if (!this_cpu_has(X86_FEATURE_LA57))
8331 		goto done;
8332 
8333 	/*
8334 	 * Allocate a full page for PML5 to guarantee alignment, though only
8335 	 * the first entry needs to be filled (the test's virtual addresses
8336 	 * most definitely do not have any of bits 56:48 set).
8337 	 */
8338 	pml5 = alloc_page();
8339 	*pml5 = orig_cr3 | PT_PRESENT_MASK | PT_WRITABLE_MASK;
8340 
8341 	/*
8342 	 * Transition to/from 5-level paging in the host via VM-Exit.  CR4.LA57
8343 	 * can't be toggled while long is active via MOV CR4, but there are no
8344 	 * such restrictions on VM-Exit.
8345 	 */
8346 lol_5level:
8347 	vmcs_write(HOST_CR4, orig_cr4 | X86_CR4_LA57);
8348 	vmcs_write(HOST_CR3, virt_to_phys(pml5));
8349 	enter_guest();
8350 
8351 	/*
8352 	 * VMREAD with a memory operand to verify KVM detects the LA57 change,
8353 	 * e.g. uses the correct guest root level in gva_to_gpa().
8354 	 */
8355 	TEST_ASSERT(vmcs_readm(HOST_CR3) == virt_to_phys(pml5));
8356 	TEST_ASSERT(vmcs_readm(HOST_CR4) == (orig_cr4 | X86_CR4_LA57));
8357 
8358 	vmcs_write(HOST_CR4, orig_cr4);
8359 	vmcs_write(HOST_CR3, orig_cr3);
8360 	enter_guest();
8361 
8362 	TEST_ASSERT(vmcs_readm(HOST_CR3) == orig_cr3);
8363 	TEST_ASSERT(vmcs_readm(HOST_CR4) == orig_cr4);
8364 
8365 	/*
8366 	 * And now do the same LA57 shenanigans with EPT enabled.  KVM uses
8367 	 * two separate MMUs when L1 uses TDP, whereas the above shadow paging
8368 	 * version shares an MMU between L1 and L2.
8369 	 *
8370 	 * If the saved execution controls are non-zero then the EPT version
8371 	 * has already run.  In that case, restore the old controls.  If EPT
8372 	 * setup fails, e.g. EPT isn't supported, fall through and finish up.
8373 	 */
8374 	if (ctrls[0]) {
8375 		vmcs_write(CPU_EXEC_CTRL0, ctrls[0]);
8376 		vmcs_write(CPU_EXEC_CTRL1, ctrls[1]);
8377 	} else if (!setup_ept(false)) {
8378 		ctrls[0] = vmcs_read(CPU_EXEC_CTRL0);
8379 		ctrls[1]  = vmcs_read(CPU_EXEC_CTRL1);
8380 		goto lol_5level;
8381 	}
8382 
8383 	free_page(pml5);
8384 
8385 done:
8386 	skip_exit_vmcall();
8387 	enter_guest();
8388 }
8389 
8390 static void vmx_cr4_osxsave_test_guest(void)
8391 {
8392 	write_cr4(read_cr4() & ~X86_CR4_OSXSAVE);
8393 }
8394 
8395 /*
8396  * Ensure that kvm recalculates the L1 guest's CPUID.01H:ECX.OSXSAVE
8397  * after VM-exit from an L2 guest that sets CR4.OSXSAVE to a different
8398  * value than in L1.
8399  */
8400 static void vmx_cr4_osxsave_test(void)
8401 {
8402 	if (!this_cpu_has(X86_FEATURE_XSAVE)) {
8403 		report_skip("XSAVE not detected");
8404 		return;
8405 	}
8406 
8407 	if (!(read_cr4() & X86_CR4_OSXSAVE)) {
8408 		unsigned long cr4 = read_cr4() | X86_CR4_OSXSAVE;
8409 
8410 		write_cr4(cr4);
8411 		vmcs_write(GUEST_CR4, cr4);
8412 		vmcs_write(HOST_CR4, cr4);
8413 	}
8414 
8415 	TEST_ASSERT(cpuid_osxsave());
8416 
8417 	test_set_guest(vmx_cr4_osxsave_test_guest);
8418 	enter_guest();
8419 
8420 	TEST_ASSERT(cpuid_osxsave());
8421 }
8422 
8423 static void vmx_nm_test_guest(void)
8424 {
8425 	write_cr0(read_cr0() | X86_CR0_TS);
8426 	asm volatile("fnop");
8427 }
8428 
8429 static void check_nm_exit(const char *test)
8430 {
8431 	u32 reason = vmcs_read(EXI_REASON);
8432 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
8433 	const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8434 		NM_VECTOR;
8435 
8436 	report(reason == VMX_EXC_NMI && intr_info == expected, "%s", test);
8437 }
8438 
8439 /*
8440  * This test checks that:
8441  *
8442  * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then
8443  *     a subsequent #NM VM-exit is reflected to L1.
8444  *
8445  * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a
8446  *     subsequent #NM VM-exit is reflected to L1.
8447  */
8448 static void vmx_nm_test(void)
8449 {
8450 	unsigned long cr0 = read_cr0();
8451 
8452 	test_set_guest(vmx_nm_test_guest);
8453 
8454 	/*
8455 	 * L1 wants to intercept #NM exceptions encountered in L2.
8456 	 */
8457 	vmcs_write(EXC_BITMAP, 1 << NM_VECTOR);
8458 
8459 	/*
8460 	 * Launch L2 with CR0.TS clear, but don't claim host ownership of
8461 	 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop,
8462 	 * which will raise #NM. L0 should reflect the #NM VM-exit to L1.
8463 	 */
8464 	vmcs_write(CR0_MASK, 0);
8465 	vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS);
8466 	enter_guest();
8467 	check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1");
8468 
8469 	/*
8470 	 * Re-enter L2 at the fnop instruction, with CR0.TS clear but
8471 	 * CR0.EM set. The fnop will still raise #NM, and L0 should
8472 	 * reflect the #NM VM-exit to L1.
8473 	 */
8474 	vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM);
8475 	enter_guest();
8476 	check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1");
8477 
8478 	/*
8479 	 * Re-enter L2 at the fnop instruction, with both CR0.TS and
8480 	 * CR0.EM clear. There will be no #NM, and the L2 guest should
8481 	 * exit normally.
8482 	 */
8483 	vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM));
8484 	enter_guest();
8485 }
8486 
8487 bool vmx_pending_event_ipi_fired;
8488 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
8489 {
8490 	vmx_pending_event_ipi_fired = true;
8491 	eoi();
8492 }
8493 
8494 bool vmx_pending_event_guest_run;
8495 static void vmx_pending_event_guest(void)
8496 {
8497 	vmcall();
8498 	vmx_pending_event_guest_run = true;
8499 }
8500 
8501 static void vmx_pending_event_test_core(bool guest_hlt)
8502 {
8503 	int ipi_vector = 0xf1;
8504 
8505 	vmx_pending_event_ipi_fired = false;
8506 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
8507 
8508 	vmx_pending_event_guest_run = false;
8509 	test_set_guest(vmx_pending_event_guest);
8510 
8511 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
8512 
8513 	enter_guest();
8514 	skip_exit_vmcall();
8515 
8516 	if (guest_hlt)
8517 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8518 
8519 	irq_disable();
8520 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
8521 				   APIC_DM_FIXED | ipi_vector,
8522 				   0);
8523 
8524 	enter_guest();
8525 
8526 	assert_exit_reason(VMX_EXTINT);
8527 	report(!vmx_pending_event_guest_run,
8528 	       "Guest did not run before host received IPI");
8529 
8530 	irq_enable();
8531 	asm volatile ("nop");
8532 	irq_disable();
8533 	report(vmx_pending_event_ipi_fired,
8534 	       "Got pending interrupt after IRQ enabled");
8535 
8536 	if (guest_hlt)
8537 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8538 
8539 	enter_guest();
8540 	report(vmx_pending_event_guest_run,
8541 	       "Guest finished running when no interrupt");
8542 }
8543 
8544 static void vmx_pending_event_test(void)
8545 {
8546 	vmx_pending_event_test_core(false);
8547 }
8548 
8549 static void vmx_pending_event_hlt_test(void)
8550 {
8551 	vmx_pending_event_test_core(true);
8552 }
8553 
8554 static int vmx_window_test_db_count;
8555 
8556 static void vmx_window_test_db_handler(struct ex_regs *regs)
8557 {
8558 	vmx_window_test_db_count++;
8559 }
8560 
8561 static void vmx_nmi_window_test_guest(void)
8562 {
8563 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8564 
8565 	asm volatile("vmcall\n\t"
8566 		     "nop\n\t");
8567 
8568 	handle_exception(DB_VECTOR, NULL);
8569 }
8570 
8571 static void verify_nmi_window_exit(u64 rip)
8572 {
8573 	u32 exit_reason = vmcs_read(EXI_REASON);
8574 
8575 	report(exit_reason == VMX_NMI_WINDOW,
8576 	       "Exit reason (%d) is 'NMI window'", exit_reason);
8577 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8578 	       vmcs_read(GUEST_RIP), rip);
8579 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8580 }
8581 
8582 static void vmx_nmi_window_test(void)
8583 {
8584 	u64 nop_addr;
8585 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8586 
8587 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
8588 		report_skip("CPU does not support the \"Virtual NMIs\" VM-execution control.");
8589 		return;
8590 	}
8591 
8592 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
8593 		report_skip("CPU does not support the \"NMI-window exiting\" VM-execution control.");
8594 		return;
8595 	}
8596 
8597 	vmx_window_test_db_count = 0;
8598 
8599 	report_prefix_push("NMI-window");
8600 	test_set_guest(vmx_nmi_window_test_guest);
8601 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
8602 	enter_guest();
8603 	skip_exit_vmcall();
8604 	nop_addr = vmcs_read(GUEST_RIP);
8605 
8606 	/*
8607 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
8608 	 * RIP will not advance.
8609 	 */
8610 	report_prefix_push("active, no blocking");
8611 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8612 	enter_guest();
8613 	verify_nmi_window_exit(nop_addr);
8614 	report_prefix_pop();
8615 
8616 	/*
8617 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
8618 	 * a VM-exit on the next instruction after the nop. (The nop
8619 	 * is one byte.)
8620 	 */
8621 	report_prefix_push("active, blocking by MOV-SS");
8622 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8623 	enter_guest();
8624 	verify_nmi_window_exit(nop_addr + 1);
8625 	report_prefix_pop();
8626 
8627 	/*
8628 	 * Ask for "NMI-window exiting" (with event injection), and
8629 	 * expect a VM-exit after the event is injected. (RIP should
8630 	 * be at the address specified in the IDT entry for #DB.)
8631 	 */
8632 	report_prefix_push("active, no blocking, injecting #DB");
8633 	vmcs_write(ENT_INTR_INFO,
8634 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8635 	enter_guest();
8636 	verify_nmi_window_exit((u64)db_fault_addr);
8637 	report_prefix_pop();
8638 
8639 	/*
8640 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
8641 	 * a VM-exit after the next IRET (i.e. after the #DB handler
8642 	 * returns). So, RIP should be back at one byte past the nop.
8643 	 */
8644 	report_prefix_push("active, blocking by NMI");
8645 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
8646 	enter_guest();
8647 	verify_nmi_window_exit(nop_addr + 1);
8648 	report(vmx_window_test_db_count == 1,
8649 	       "#DB handler executed once (actual %d times)",
8650 	       vmx_window_test_db_count);
8651 	report_prefix_pop();
8652 
8653 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8654 		report_skip("CPU does not support activity state HLT.");
8655 	} else {
8656 		/*
8657 		 * Ask for "NMI-window exiting" when entering activity
8658 		 * state HLT, and expect an immediate VM-exit. RIP is
8659 		 * still one byte past the nop.
8660 		 */
8661 		report_prefix_push("halted, no blocking");
8662 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8663 		enter_guest();
8664 		verify_nmi_window_exit(nop_addr + 1);
8665 		report_prefix_pop();
8666 
8667 		/*
8668 		 * Ask for "NMI-window exiting" when entering activity
8669 		 * state HLT (with event injection), and expect a
8670 		 * VM-exit after the event is injected. (RIP should be
8671 		 * at the address specified in the IDT entry for #DB.)
8672 		 */
8673 		report_prefix_push("halted, no blocking, injecting #DB");
8674 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8675 		vmcs_write(ENT_INTR_INFO,
8676 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8677 			   DB_VECTOR);
8678 		enter_guest();
8679 		verify_nmi_window_exit((u64)db_fault_addr);
8680 		report_prefix_pop();
8681 	}
8682 
8683 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8684 	enter_guest();
8685 	report_prefix_pop();
8686 }
8687 
8688 static void vmx_intr_window_test_guest(void)
8689 {
8690 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8691 
8692 	/*
8693 	 * The two consecutive STIs are to ensure that only the first
8694 	 * one has a shadow. Note that NOP and STI are one byte
8695 	 * instructions.
8696 	 */
8697 	asm volatile("vmcall\n\t"
8698 		     "nop\n\t"
8699 		     "sti\n\t"
8700 		     "sti\n\t");
8701 
8702 	handle_exception(DB_VECTOR, NULL);
8703 }
8704 
8705 static void verify_intr_window_exit(u64 rip)
8706 {
8707 	u32 exit_reason = vmcs_read(EXI_REASON);
8708 
8709 	report(exit_reason == VMX_INTR_WINDOW,
8710 	       "Exit reason (%d) is 'interrupt window'", exit_reason);
8711 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8712 	       vmcs_read(GUEST_RIP), rip);
8713 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8714 }
8715 
8716 static void vmx_intr_window_test(void)
8717 {
8718 	u64 vmcall_addr;
8719 	u64 nop_addr;
8720 	unsigned int orig_db_gate_type;
8721 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8722 
8723 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
8724 		report_skip("CPU does not support the \"interrupt-window exiting\" VM-execution control.");
8725 		return;
8726 	}
8727 
8728 	/*
8729 	 * Change the IDT entry for #DB from interrupt gate to trap gate,
8730 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
8731 	 * be disabled after vectoring a #DB.
8732 	 */
8733 	orig_db_gate_type = boot_idt[DB_VECTOR].type;
8734 	boot_idt[DB_VECTOR].type = 15;
8735 
8736 	report_prefix_push("interrupt-window");
8737 	test_set_guest(vmx_intr_window_test_guest);
8738 	enter_guest();
8739 	assert_exit_reason(VMX_VMCALL);
8740 	vmcall_addr = vmcs_read(GUEST_RIP);
8741 
8742 	/*
8743 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
8744 	 * no blocking; expect an immediate VM-exit. Note that we have
8745 	 * not advanced past the vmcall instruction yet, so RIP should
8746 	 * point to the vmcall instruction.
8747 	 */
8748 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
8749 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8750 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
8751 	enter_guest();
8752 	verify_intr_window_exit(vmcall_addr);
8753 	report_prefix_pop();
8754 
8755 	/*
8756 	 * Ask for "interrupt-window exiting" (with event injection)
8757 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
8758 	 * the event is injected. That is, RIP should should be at the
8759 	 * address specified in the IDT entry for #DB.
8760 	 */
8761 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #DB");
8762 	vmcs_write(ENT_INTR_INFO,
8763 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8764 	vmcall_addr = vmcs_read(GUEST_RIP);
8765 	enter_guest();
8766 	verify_intr_window_exit((u64)db_fault_addr);
8767 	report_prefix_pop();
8768 
8769 	/*
8770 	 * Let the L2 guest run through the IRET, back to the VMCALL.
8771 	 * We have to clear the "interrupt-window exiting"
8772 	 * VM-execution control, or it would just keep causing
8773 	 * VM-exits. Then, advance past the VMCALL and set the
8774 	 * "interrupt-window exiting" VM-execution control again.
8775 	 */
8776 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8777 	enter_guest();
8778 	skip_exit_vmcall();
8779 	nop_addr = vmcs_read(GUEST_RIP);
8780 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8781 
8782 	/*
8783 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
8784 	 * RFLAGS.IF set, and expect a VM-exit on the next
8785 	 * instruction. (NOP is one byte.)
8786 	 */
8787 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
8788 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8789 	enter_guest();
8790 	verify_intr_window_exit(nop_addr + 1);
8791 	report_prefix_pop();
8792 
8793 	/*
8794 	 * Back up to the NOP and ask for "interrupt-window exiting"
8795 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
8796 	 * on the next instruction. (NOP is one byte.)
8797 	 */
8798 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
8799 	vmcs_write(GUEST_RIP, nop_addr);
8800 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
8801 	enter_guest();
8802 	verify_intr_window_exit(nop_addr + 1);
8803 	report_prefix_pop();
8804 
8805 	/*
8806 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
8807 	 * and expect a VM-exit on the instruction following the STI
8808 	 * shadow. Only the first STI (which is one byte past the NOP)
8809 	 * should have a shadow. The second STI (which is two bytes
8810 	 * past the NOP) has no shadow. Therefore, the interrupt
8811 	 * window opens at three bytes past the NOP.
8812 	 */
8813 	report_prefix_push("active, RFLAGS.IF = 0");
8814 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8815 	enter_guest();
8816 	verify_intr_window_exit(nop_addr + 3);
8817 	report_prefix_pop();
8818 
8819 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8820 		report_skip("CPU does not support activity state HLT.");
8821 	} else {
8822 		/*
8823 		 * Ask for "interrupt-window exiting" when entering
8824 		 * activity state HLT, and expect an immediate
8825 		 * VM-exit. RIP is still three bytes past the nop.
8826 		 */
8827 		report_prefix_push("halted, no blocking");
8828 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8829 		enter_guest();
8830 		verify_intr_window_exit(nop_addr + 3);
8831 		report_prefix_pop();
8832 
8833 		/*
8834 		 * Ask for "interrupt-window exiting" when entering
8835 		 * activity state HLT (with event injection), and
8836 		 * expect a VM-exit after the event is injected. That
8837 		 * is, RIP should should be at the address specified
8838 		 * in the IDT entry for #DB.
8839 		 */
8840 		report_prefix_push("halted, no blocking, injecting #DB");
8841 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8842 		vmcs_write(ENT_INTR_INFO,
8843 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8844 			   DB_VECTOR);
8845 		enter_guest();
8846 		verify_intr_window_exit((u64)db_fault_addr);
8847 		report_prefix_pop();
8848 	}
8849 
8850 	boot_idt[DB_VECTOR].type = orig_db_gate_type;
8851 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8852 	enter_guest();
8853 	report_prefix_pop();
8854 }
8855 
8856 #define GUEST_TSC_OFFSET (1u << 30)
8857 
8858 static u64 guest_tsc;
8859 
8860 static void vmx_store_tsc_test_guest(void)
8861 {
8862 	guest_tsc = rdtsc();
8863 }
8864 
8865 /*
8866  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
8867  * list, the value saved is not subject to the TSC offset that is
8868  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
8869  */
8870 static void vmx_store_tsc_test(void)
8871 {
8872 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
8873 	u64 low, high;
8874 
8875 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
8876 		report_skip("'Use TSC offsetting' not supported");
8877 		return;
8878 	}
8879 
8880 	test_set_guest(vmx_store_tsc_test_guest);
8881 
8882 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
8883 	vmcs_write(EXI_MSR_ST_CNT, 1);
8884 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
8885 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
8886 
8887 	low = rdtsc();
8888 	enter_guest();
8889 	high = rdtsc();
8890 
8891 	report(low + GUEST_TSC_OFFSET <= guest_tsc &&
8892 	       guest_tsc <= high + GUEST_TSC_OFFSET,
8893 	       "RDTSC value in the guest (%lu) is in range [%lu, %lu]",
8894 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
8895 	report(low <= msr_entry.value && msr_entry.value <= high,
8896 	       "IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
8897 	       msr_entry.value, low, high);
8898 }
8899 
8900 static void vmx_preemption_timer_zero_test_db_handler(struct ex_regs *regs)
8901 {
8902 }
8903 
8904 static void vmx_preemption_timer_zero_test_guest(void)
8905 {
8906 	while (vmx_get_test_stage() < 3)
8907 		vmcall();
8908 }
8909 
8910 static void vmx_preemption_timer_zero_activate_preemption_timer(void)
8911 {
8912 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
8913 	vmcs_write(PREEMPT_TIMER_VALUE, 0);
8914 }
8915 
8916 static void vmx_preemption_timer_zero_advance_past_vmcall(void)
8917 {
8918 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
8919 	enter_guest();
8920 	skip_exit_vmcall();
8921 }
8922 
8923 static void vmx_preemption_timer_zero_inject_db(bool intercept_db)
8924 {
8925 	vmx_preemption_timer_zero_activate_preemption_timer();
8926 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK |
8927 		   INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8928 	vmcs_write(EXC_BITMAP, intercept_db ? 1 << DB_VECTOR : 0);
8929 	enter_guest();
8930 }
8931 
8932 static void vmx_preemption_timer_zero_set_pending_dbg(u32 exception_bitmap)
8933 {
8934 	vmx_preemption_timer_zero_activate_preemption_timer();
8935 	vmcs_write(GUEST_PENDING_DEBUG, BIT(12) | DR_TRAP1);
8936 	vmcs_write(EXC_BITMAP, exception_bitmap);
8937 	enter_guest();
8938 }
8939 
8940 static void vmx_preemption_timer_zero_expect_preempt_at_rip(u64 expected_rip)
8941 {
8942 	u32 reason = (u32)vmcs_read(EXI_REASON);
8943 	u64 guest_rip = vmcs_read(GUEST_RIP);
8944 
8945 	report(reason == VMX_PREEMPT && guest_rip == expected_rip,
8946 	       "Exit reason is 0x%x (expected 0x%x) and guest RIP is %lx (0x%lx expected).",
8947 	       reason, VMX_PREEMPT, guest_rip, expected_rip);
8948 }
8949 
8950 /*
8951  * This test ensures that when the VMX preemption timer is zero at
8952  * VM-entry, a VM-exit occurs after any event injection and after any
8953  * pending debug exceptions are raised, but before execution of any
8954  * guest instructions.
8955  */
8956 static void vmx_preemption_timer_zero_test(void)
8957 {
8958 	u64 db_fault_address = (u64)get_idt_addr(&boot_idt[DB_VECTOR]);
8959 	handler old_db;
8960 	u32 reason;
8961 
8962 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
8963 		report_skip("'Activate VMX-preemption timer' not supported");
8964 		return;
8965 	}
8966 
8967 	/*
8968 	 * Install a custom #DB handler that doesn't abort.
8969 	 */
8970 	old_db = handle_exception(DB_VECTOR,
8971 				  vmx_preemption_timer_zero_test_db_handler);
8972 
8973 	test_set_guest(vmx_preemption_timer_zero_test_guest);
8974 
8975 	/*
8976 	 * VMX-preemption timer should fire after event injection.
8977 	 */
8978 	vmx_set_test_stage(0);
8979 	vmx_preemption_timer_zero_inject_db(0);
8980 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8981 	vmx_preemption_timer_zero_advance_past_vmcall();
8982 
8983 	/*
8984 	 * VMX-preemption timer should fire after event injection.
8985 	 * Exception bitmap is irrelevant, since you can't intercept
8986 	 * an event that you injected.
8987 	 */
8988 	vmx_set_test_stage(1);
8989 	vmx_preemption_timer_zero_inject_db(true);
8990 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8991 	vmx_preemption_timer_zero_advance_past_vmcall();
8992 
8993 	/*
8994 	 * VMX-preemption timer should fire after pending debug exceptions
8995 	 * have delivered a #DB trap.
8996 	 */
8997 	vmx_set_test_stage(2);
8998 	vmx_preemption_timer_zero_set_pending_dbg(0);
8999 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
9000 	vmx_preemption_timer_zero_advance_past_vmcall();
9001 
9002 	/*
9003 	 * VMX-preemption timer would fire after pending debug exceptions
9004 	 * have delivered a #DB trap, but in this case, the #DB trap is
9005 	 * intercepted.
9006 	 */
9007 	vmx_set_test_stage(3);
9008 	vmx_preemption_timer_zero_set_pending_dbg(1 << DB_VECTOR);
9009 	reason = (u32)vmcs_read(EXI_REASON);
9010 	report(reason == VMX_EXC_NMI, "Exit reason is 0x%x (expected 0x%x)",
9011 	       reason, VMX_EXC_NMI);
9012 
9013 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9014 	enter_guest();
9015 
9016 	handle_exception(DB_VECTOR, old_db);
9017 }
9018 
9019 static u64 vmx_preemption_timer_tf_test_prev_rip;
9020 
9021 static void vmx_preemption_timer_tf_test_db_handler(struct ex_regs *regs)
9022 {
9023 	extern char vmx_preemption_timer_tf_test_endloop;
9024 
9025 	if (vmx_get_test_stage() == 2) {
9026 		/*
9027 		 * Stage 2 means that we're done, one way or another.
9028 		 * Arrange for the iret to drop us out of the wbinvd
9029 		 * loop and stop single-stepping.
9030 		 */
9031 		regs->rip = (u64)&vmx_preemption_timer_tf_test_endloop;
9032 		regs->rflags &= ~X86_EFLAGS_TF;
9033 	} else if (regs->rip == vmx_preemption_timer_tf_test_prev_rip) {
9034 		/*
9035 		 * The RIP should alternate between the wbinvd and the
9036 		 * jmp instruction in the code below. If we ever see
9037 		 * the same instruction twice in a row, that means a
9038 		 * single-step trap has been dropped. Let the
9039 		 * hypervisor know about the failure by executing a
9040 		 * VMCALL.
9041 		 */
9042 		vmcall();
9043 	}
9044 	vmx_preemption_timer_tf_test_prev_rip = regs->rip;
9045 }
9046 
9047 static void vmx_preemption_timer_tf_test_guest(void)
9048 {
9049 	/*
9050 	 * The hypervisor doesn't intercept WBINVD, so the loop below
9051 	 * shouldn't be a problem--it's just two instructions
9052 	 * executing in VMX non-root mode. However, when the
9053 	 * hypervisor is running in a virtual environment, the parent
9054 	 * hypervisor might intercept WBINVD and emulate it. If the
9055 	 * parent hypervisor is broken, the single-step trap after the
9056 	 * WBINVD might be lost.
9057 	 */
9058 	asm volatile("vmcall\n\t"
9059 		     "0: wbinvd\n\t"
9060 		     "1: jmp 0b\n\t"
9061 		     "vmx_preemption_timer_tf_test_endloop:");
9062 }
9063 
9064 /*
9065  * Ensure that the delivery of a "VMX-preemption timer expired"
9066  * VM-exit doesn't disrupt single-stepping in the guest. Note that
9067  * passing this test doesn't ensure correctness, because the test will
9068  * only fail if the VMX-preemtion timer fires at the right time (or
9069  * the wrong time, as it were).
9070  */
9071 static void vmx_preemption_timer_tf_test(void)
9072 {
9073 	handler old_db;
9074 	u32 reason;
9075 	int i;
9076 
9077 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9078 		report_skip("'Activate VMX-preemption timer' not supported");
9079 		return;
9080 	}
9081 
9082 	old_db = handle_exception(DB_VECTOR,
9083 				  vmx_preemption_timer_tf_test_db_handler);
9084 
9085 	test_set_guest(vmx_preemption_timer_tf_test_guest);
9086 
9087 	enter_guest();
9088 	skip_exit_vmcall();
9089 
9090 	vmx_set_test_stage(1);
9091 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9092 	vmcs_write(PREEMPT_TIMER_VALUE, 50000);
9093 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9094 
9095 	/*
9096 	 * The only exit we should see is "VMX-preemption timer
9097 	 * expired."  If we get a VMCALL exit, that means the #DB
9098 	 * handler has detected a missing single-step trap. It doesn't
9099 	 * matter where the guest RIP is when the VMX-preemption timer
9100 	 * expires (whether it's in the WBINVD loop or in the #DB
9101 	 * handler)--a single-step trap should never be discarded.
9102 	 */
9103 	for (i = 0; i < 10000; i++) {
9104 		enter_guest();
9105 		reason = (u32)vmcs_read(EXI_REASON);
9106 		if (reason == VMX_PREEMPT)
9107 			continue;
9108 		TEST_ASSERT(reason == VMX_VMCALL);
9109 		skip_exit_insn();
9110 		break;
9111 	}
9112 
9113 	report(reason == VMX_PREEMPT, "No single-step traps skipped");
9114 
9115 	vmx_set_test_stage(2);
9116 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9117 	enter_guest();
9118 
9119 	handle_exception(DB_VECTOR, old_db);
9120 }
9121 
9122 #define VMX_PREEMPTION_TIMER_EXPIRY_CYCLES 1000000
9123 
9124 static u64 vmx_preemption_timer_expiry_start;
9125 static u64 vmx_preemption_timer_expiry_finish;
9126 
9127 static void vmx_preemption_timer_expiry_test_guest(void)
9128 {
9129 	vmcall();
9130 	vmx_preemption_timer_expiry_start = fenced_rdtsc();
9131 
9132 	while (vmx_get_test_stage() == 0)
9133 		vmx_preemption_timer_expiry_finish = fenced_rdtsc();
9134 }
9135 
9136 /*
9137  * Test that the VMX-preemption timer is not excessively delayed.
9138  *
9139  * Per the SDM, volume 3, VM-entry starts the VMX-preemption timer
9140  * with the unsigned value in the VMX-preemption timer-value field,
9141  * and the VMX-preemption timer counts down by 1 every time bit X in
9142  * the TSC changes due to a TSC increment (where X is
9143  * IA32_VMX_MISC[4:0]). If the timer counts down to zero in any state
9144  * other than the wait-for-SIPI state, the logical processor
9145  * transitions to the C0 C-state and causes a VM-exit.
9146  *
9147  * The guest code above reads the starting TSC after VM-entry. At this
9148  * point, the VMX-preemption timer has already been activated. Next,
9149  * the guest code reads the current TSC in a loop, storing the value
9150  * read to memory.
9151  *
9152  * If the RDTSC in the loop reads a value past the VMX-preemption
9153  * timer deadline, then the VMX-preemption timer VM-exit must be
9154  * delivered before the next instruction retires. Even if a higher
9155  * priority SMI is delivered first, the VMX-preemption timer VM-exit
9156  * must be delivered before the next instruction retires. Hence, a TSC
9157  * value past the VMX-preemption timer deadline might be read, but it
9158  * cannot be stored. If a TSC value past the deadline *is* stored,
9159  * then the architectural specification has been violated.
9160  */
9161 static void vmx_preemption_timer_expiry_test(void)
9162 {
9163 	u32 preemption_timer_value;
9164 	union vmx_misc misc;
9165 	u64 tsc_deadline;
9166 	u32 reason;
9167 
9168 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9169 		report_skip("'Activate VMX-preemption timer' not supported");
9170 		return;
9171 	}
9172 
9173 	test_set_guest(vmx_preemption_timer_expiry_test_guest);
9174 
9175 	enter_guest();
9176 	skip_exit_vmcall();
9177 
9178 	misc.val = rdmsr(MSR_IA32_VMX_MISC);
9179 	preemption_timer_value =
9180 		VMX_PREEMPTION_TIMER_EXPIRY_CYCLES >> misc.pt_bit;
9181 
9182 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9183 	vmcs_write(PREEMPT_TIMER_VALUE, preemption_timer_value);
9184 	vmx_set_test_stage(0);
9185 
9186 	enter_guest();
9187 	reason = (u32)vmcs_read(EXI_REASON);
9188 	TEST_ASSERT(reason == VMX_PREEMPT);
9189 
9190 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9191 	vmx_set_test_stage(1);
9192 	enter_guest();
9193 
9194 	tsc_deadline = ((vmx_preemption_timer_expiry_start >> misc.pt_bit) <<
9195 			misc.pt_bit) + (preemption_timer_value << misc.pt_bit);
9196 
9197 	report(vmx_preemption_timer_expiry_finish < tsc_deadline,
9198 	       "Last stored guest TSC (%lu) < TSC deadline (%lu)",
9199 	       vmx_preemption_timer_expiry_finish, tsc_deadline);
9200 }
9201 
9202 static void vmx_db_test_guest(void)
9203 {
9204 	/*
9205 	 * For a hardware generated single-step #DB.
9206 	 */
9207 	asm volatile("vmcall;"
9208 		     "nop;"
9209 		     ".Lpost_nop:");
9210 	/*
9211 	 * ...in a MOVSS shadow, with pending debug exceptions.
9212 	 */
9213 	asm volatile("vmcall;"
9214 		     "nop;"
9215 		     ".Lpost_movss_nop:");
9216 	/*
9217 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
9218 	 * emulates it in software.)
9219 	 */
9220 	asm volatile("vmcall;"
9221 		     "wbinvd;"
9222 		     ".Lpost_wbinvd:");
9223 	/*
9224 	 * ...in a MOVSS shadow, with pending debug exceptions.
9225 	 */
9226 	asm volatile("vmcall;"
9227 		     "wbinvd;"
9228 		     ".Lpost_movss_wbinvd:");
9229 	/*
9230 	 * For a hardware generated single-step #DB in a transactional region.
9231 	 */
9232 	asm volatile("vmcall;"
9233 		     ".Lxbegin: xbegin .Lskip_rtm;"
9234 		     "xend;"
9235 		     ".Lskip_rtm:");
9236 }
9237 
9238 /*
9239  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
9240  * L2. No #DB is delivered and L2 continues to the next point of
9241  * interest.
9242  */
9243 static void dismiss_db(void)
9244 {
9245 	vmcs_write(GUEST_PENDING_DEBUG, 0);
9246 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
9247 	enter_guest();
9248 }
9249 
9250 /*
9251  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
9252  * Then throw away the #DB exception and resume L2.
9253  */
9254 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
9255 			  void *expected_rip, u64 expected_exit_qual,
9256 			  u64 expected_dr6)
9257 {
9258 	u32 reason = vmcs_read(EXI_REASON);
9259 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
9260 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
9261 	u64 guest_rip = vmcs_read(GUEST_RIP);
9262 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
9263 	u64 dr6 = read_dr6();
9264 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
9265 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
9266 
9267 	report(reason == VMX_EXC_NMI && intr_info == expected_intr_info,
9268 	       "Expected #DB VM-exit");
9269 	report((u64)expected_rip == guest_rip, "Expected RIP %p (actual %lx)",
9270 	       expected_rip, guest_rip);
9271 	report_xfail(xfail_pdbg, 0 == guest_pending_dbg,
9272 		     "Expected pending debug exceptions 0 (actual %lx)",
9273 		     guest_pending_dbg);
9274 	report_xfail(xfail_qual, expected_exit_qual == exit_qual,
9275 		     "Expected exit qualification %lx (actual %lx)",
9276 		     expected_exit_qual, exit_qual);
9277 	report_xfail(xfail_dr6, expected_dr6 == dr6,
9278 		     "Expected DR6 %lx (actual %lx)", expected_dr6, dr6);
9279 	dismiss_db();
9280 }
9281 
9282 /*
9283  * Assuming the guest has just exited on a VMCALL instruction, skip
9284  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
9285  * pending debug exceptions are non-zero, set the VMCS up as if the
9286  * previous instruction was a MOVSS that generated the indicated
9287  * pending debug exceptions. Then enter L2.
9288  */
9289 static void single_step_guest(const char *test_name, u64 starting_dr6,
9290 			      u64 pending_debug_exceptions)
9291 {
9292 	printf("\n%s\n", test_name);
9293 	skip_exit_vmcall();
9294 	write_dr6(starting_dr6);
9295 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9296 	if (pending_debug_exceptions) {
9297 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
9298 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
9299 	}
9300 	enter_guest();
9301 }
9302 
9303 /*
9304  * When L1 intercepts #DB, verify that a single-step trap clears
9305  * pending debug exceptions, populates the exit qualification field
9306  * properly, and that DR6 is not prematurely clobbered. In a
9307  * (simulated) MOVSS shadow, make sure that the pending debug
9308  * exception bits are properly accumulated into the exit qualification
9309  * field.
9310  */
9311 static void vmx_db_test(void)
9312 {
9313 	/*
9314 	 * We are going to set a few arbitrary bits in DR6 to verify that
9315 	 * (a) DR6 is not modified by an intercepted #DB, and
9316 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
9317          *     the exit qualification field for a subsequent #DB exception.
9318 	 */
9319 	const u64 starting_dr6 = DR6_RESERVED | BIT(13) | DR_TRAP3 | DR_TRAP1;
9320 	extern char post_nop asm(".Lpost_nop");
9321 	extern char post_movss_nop asm(".Lpost_movss_nop");
9322 	extern char post_wbinvd asm(".Lpost_wbinvd");
9323 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
9324 	extern char xbegin asm(".Lxbegin");
9325 	extern char skip_rtm asm(".Lskip_rtm");
9326 
9327 	/*
9328 	 * L1 wants to intercept #DB exceptions encountered in L2.
9329 	 */
9330 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
9331 
9332 	/*
9333 	 * Start L2 and run it up to the first point of interest.
9334 	 */
9335 	test_set_guest(vmx_db_test_guest);
9336 	enter_guest();
9337 
9338 	/*
9339 	 * Hardware-delivered #DB trap for single-step sets the
9340 	 * standard that L0 has to follow for emulated instructions.
9341 	 */
9342 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
9343 	check_db_exit(false, false, false, &post_nop, DR_STEP, starting_dr6);
9344 
9345 	/*
9346 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
9347 	 * also sets the standard that L0 has to follow for emulated
9348 	 * instructions. Here, we establish the VMCS pending debug
9349 	 * exceptions to indicate that the simulated MOVSS triggered a
9350 	 * data breakpoint as well as the single-step trap.
9351 	 */
9352 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
9353 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0 );
9354 	check_db_exit(false, false, false, &post_movss_nop, DR_STEP | DR_TRAP0,
9355 		      starting_dr6);
9356 
9357 	/*
9358 	 * L0 synthesized #DB trap for single-step is buggy, because
9359 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
9360 	 * reconstitute the exit qualification from the prematurely
9361 	 * modified DR6, but fails miserably.
9362 	 */
9363 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
9364 	check_db_exit(false, false, false, &post_wbinvd, DR_STEP, starting_dr6);
9365 
9366 	/*
9367 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
9368 	 * even worse, because L0 also leaves the pending debug
9369 	 * exceptions in the VMCS instead of accumulating them into
9370 	 * the exit qualification field for the #DB exception.
9371 	 */
9372 	single_step_guest("Software synthesized single-step in MOVSS shadow",
9373 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0);
9374 	check_db_exit(true, false, true, &post_movss_wbinvd, DR_STEP | DR_TRAP0,
9375 		      starting_dr6);
9376 
9377 	/*
9378 	 * Optional RTM test for hardware that supports RTM, to
9379 	 * demonstrate that the current volume 3 of the SDM
9380 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
9381 	 * qualification for debug exceptions is not reserved. It is
9382 	 * set to 1 if a debug exception (#DB) or a breakpoint
9383 	 * exception (#BP) occurs inside an RTM region while advanced
9384 	 * debugging of RTM transactional regions is enabled.
9385 	 */
9386 	if (this_cpu_has(X86_FEATURE_RTM)) {
9387 		vmcs_write(ENT_CONTROLS,
9388 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
9389 		/*
9390 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
9391 		 * in the guest to enable advanced debugging of RTM
9392 		 * transactional regions.
9393 		 */
9394 		vmcs_write(GUEST_DR7, BIT(11));
9395 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
9396 		single_step_guest("Hardware delivered single-step in "
9397 				  "transactional region", starting_dr6, 0);
9398 		check_db_exit(false, false, false, &xbegin, BIT(16),
9399 			      starting_dr6);
9400 	} else {
9401 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
9402 		enter_guest();
9403 	}
9404 }
9405 
9406 static void enable_vid(void)
9407 {
9408 	void *virtual_apic_page;
9409 
9410 	assert(cpu_has_apicv());
9411 
9412 	disable_intercept_for_x2apic_msrs();
9413 
9414 	virtual_apic_page = alloc_page();
9415 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
9416 
9417 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
9418 
9419 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
9420 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
9421 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
9422 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
9423 
9424 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
9425 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
9426 }
9427 
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 
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 
9453 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
9454 {
9455 	handle_irq(0x79, irq_79_handler_guest);
9456 	irq_enable();
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 
9474 static void vmx_eoi_bitmap_ioapic_scan_test(void)
9475 {
9476 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
9477 		report_skip(__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 asserion 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;
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 
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 	irq_enable();
9514 	asm volatile ("nop");
9515 
9516 	vmcall();
9517 }
9518 
9519 static void vmx_hlt_with_rvi_test(void)
9520 {
9521 	if (!cpu_has_apicv()) {
9522 		report_skip(__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 
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;
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 
9562 static void vmx_apic_passthrough_guest(void)
9563 {
9564 	handle_irq(0x78, irq_78_handler_guest);
9565 	irq_enable();
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 
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(__func__);
9581 		return;
9582 	}
9583 
9584 	/* Test device is required for generating IRQs */
9585 	if (!test_device_enabled()) {
9586 		report_skip(__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 
9628 static void vmx_apic_passthrough_test(void)
9629 {
9630 	vmx_apic_passthrough(false);
9631 }
9632 
9633 static void vmx_apic_passthrough_thread_test(void)
9634 {
9635 	vmx_apic_passthrough(true);
9636 }
9637 
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;
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 
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();
9674 	asm volatile ("nop");
9675 	report(vmx_apic_passthrough_tpr_threshold_ipi_isr_fired, "self-IPI fired");
9676 
9677 	report_pass(__func__);
9678 }
9679 
9680 static u64 init_signal_test_exit_reason;
9681 static bool init_signal_test_thread_continued;
9682 
9683 static void init_signal_test_thread(void *data)
9684 {
9685 	struct vmcs *test_vmcs = data;
9686 
9687 	/* Enter VMX operation (i.e. exec VMXON) */
9688 	u64 *ap_vmxon_region = alloc_page();
9689 	enable_vmx();
9690 	init_vmx(ap_vmxon_region);
9691 	_vmx_on(ap_vmxon_region);
9692 
9693 	/* Signal CPU have entered VMX operation */
9694 	vmx_set_test_stage(1);
9695 
9696 	/* Wait for BSP CPU to send INIT signal */
9697 	while (vmx_get_test_stage() != 2)
9698 		;
9699 
9700 	/*
9701 	 * Signal that we continue as usual as INIT signal
9702 	 * should be blocked while CPU is in VMX operation
9703 	 */
9704 	vmx_set_test_stage(3);
9705 
9706 	/* Wait for signal to enter VMX non-root mode */
9707 	while (vmx_get_test_stage() != 4)
9708 		;
9709 
9710 	/* Enter VMX non-root mode */
9711 	test_set_guest(v2_null_test_guest);
9712 	make_vmcs_current(test_vmcs);
9713 	enter_guest();
9714 	/* Save exit reason for BSP CPU to compare to expected result */
9715 	init_signal_test_exit_reason = vmcs_read(EXI_REASON);
9716 	/* VMCLEAR test-vmcs so it could be loaded by BSP CPU */
9717 	vmcs_clear(test_vmcs);
9718 	launched = false;
9719 	/* Signal that CPU exited to VMX root mode */
9720 	vmx_set_test_stage(5);
9721 
9722 	/* Wait for BSP CPU to signal to exit VMX operation */
9723 	while (vmx_get_test_stage() != 6)
9724 		;
9725 
9726 	/* Exit VMX operation (i.e. exec VMXOFF) */
9727 	vmx_off();
9728 
9729 	/*
9730 	 * Signal to BSP CPU that we continue as usual as INIT signal
9731 	 * should have been consumed by VMX_INIT exit from guest
9732 	 */
9733 	vmx_set_test_stage(7);
9734 
9735 	/* Wait for BSP CPU to signal to enter VMX operation */
9736 	while (vmx_get_test_stage() != 8)
9737 		;
9738 	/* Enter VMX operation (i.e. exec VMXON) */
9739 	_vmx_on(ap_vmxon_region);
9740 	/* Signal to BSP we are in VMX operation */
9741 	vmx_set_test_stage(9);
9742 
9743 	/* Wait for BSP CPU to send INIT signal */
9744 	while (vmx_get_test_stage() != 10)
9745 		;
9746 
9747 	/* Exit VMX operation (i.e. exec VMXOFF) */
9748 	vmx_off();
9749 
9750 	/*
9751 	 * Exiting VMX operation should result in latched
9752 	 * INIT signal being processed. Therefore, we should
9753 	 * never reach the below code. Thus, signal to BSP
9754 	 * CPU if we have reached here so it is able to
9755 	 * report an issue if it happens.
9756 	 */
9757 	init_signal_test_thread_continued = true;
9758 }
9759 
9760 #define INIT_SIGNAL_TEST_DELAY	100000000ULL
9761 
9762 static void vmx_init_signal_test(void)
9763 {
9764 	struct vmcs *test_vmcs;
9765 
9766 	if (cpu_count() < 2) {
9767 		report_skip(__func__);
9768 		return;
9769 	}
9770 
9771 	/* VMCLEAR test-vmcs so it could be loaded by other CPU */
9772 	vmcs_save(&test_vmcs);
9773 	vmcs_clear(test_vmcs);
9774 
9775 	vmx_set_test_stage(0);
9776 	on_cpu_async(1, init_signal_test_thread, test_vmcs);
9777 
9778 	/* Wait for other CPU to enter VMX operation */
9779 	while (vmx_get_test_stage() != 1)
9780 		;
9781 
9782 	/* Send INIT signal to other CPU */
9783 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9784 				   id_map[1]);
9785 	/* Signal other CPU we have sent INIT signal */
9786 	vmx_set_test_stage(2);
9787 
9788 	/*
9789 	 * Wait reasonable amount of time for INIT signal to
9790 	 * be received on other CPU and verify that other CPU
9791 	 * have proceed as usual to next test stage as INIT
9792 	 * signal should be blocked while other CPU in
9793 	 * VMX operation
9794 	 */
9795 	delay(INIT_SIGNAL_TEST_DELAY);
9796 	report(vmx_get_test_stage() == 3,
9797 	       "INIT signal blocked when CPU in VMX operation");
9798 	/* No point to continue if we failed at this point */
9799 	if (vmx_get_test_stage() != 3)
9800 		return;
9801 
9802 	/* Signal other CPU to enter VMX non-root mode */
9803 	init_signal_test_exit_reason = -1ull;
9804 	vmx_set_test_stage(4);
9805 	/*
9806 	 * Wait reasonable amont of time for other CPU
9807 	 * to exit to VMX root mode
9808 	 */
9809 	delay(INIT_SIGNAL_TEST_DELAY);
9810 	if (vmx_get_test_stage() != 5) {
9811 		report_fail("Pending INIT signal didn't result in VMX exit");
9812 		return;
9813 	}
9814 	report(init_signal_test_exit_reason == VMX_INIT,
9815 			"INIT signal during VMX non-root mode result in exit-reason %s (%lu)",
9816 			exit_reason_description(init_signal_test_exit_reason),
9817 			init_signal_test_exit_reason);
9818 
9819 	/* Run guest to completion */
9820 	make_vmcs_current(test_vmcs);
9821 	enter_guest();
9822 
9823 	/* Signal other CPU to exit VMX operation */
9824 	init_signal_test_thread_continued = false;
9825 	vmx_set_test_stage(6);
9826 
9827 	/* Wait reasonable amount of time for other CPU to exit VMX operation */
9828 	delay(INIT_SIGNAL_TEST_DELAY);
9829 	report(vmx_get_test_stage() == 7,
9830 	       "INIT signal consumed on VMX_INIT exit");
9831 	/* No point to continue if we failed at this point */
9832 	if (vmx_get_test_stage() != 7)
9833 		return;
9834 
9835 	/* Signal other CPU to enter VMX operation */
9836 	vmx_set_test_stage(8);
9837 	/* Wait for other CPU to enter VMX operation */
9838 	while (vmx_get_test_stage() != 9)
9839 		;
9840 
9841 	/* Send INIT signal to other CPU */
9842 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9843 				   id_map[1]);
9844 	/* Signal other CPU we have sent INIT signal */
9845 	vmx_set_test_stage(10);
9846 
9847 	/*
9848 	 * Wait reasonable amount of time for other CPU
9849 	 * to exit VMX operation and process INIT signal
9850 	 */
9851 	delay(INIT_SIGNAL_TEST_DELAY);
9852 	report(!init_signal_test_thread_continued,
9853 	       "INIT signal processed after exit VMX operation");
9854 
9855 	/*
9856 	 * TODO: Send SIPI to other CPU to sipi_entry (See x86/cstart64.S)
9857 	 * to re-init it to kvm-unit-tests standard environment.
9858 	 * Somehow (?) verify that SIPI was indeed received.
9859 	 */
9860 }
9861 
9862 #define SIPI_SIGNAL_TEST_DELAY	100000000ULL
9863 
9864 static void vmx_sipi_test_guest(void)
9865 {
9866 	if (apic_id() == 0) {
9867 		/* wait AP enter guest with activity=WAIT_SIPI */
9868 		while (vmx_get_test_stage() != 1)
9869 			;
9870 		delay(SIPI_SIGNAL_TEST_DELAY);
9871 
9872 		/* First SIPI signal */
9873 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9874 		report_pass("BSP(L2): Send first SIPI to cpu[%d]", id_map[1]);
9875 
9876 		/* wait AP enter guest */
9877 		while (vmx_get_test_stage() != 2)
9878 			;
9879 		delay(SIPI_SIGNAL_TEST_DELAY);
9880 
9881 		/* Second SIPI signal should be ignored since AP is not in WAIT_SIPI state */
9882 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9883 		report_pass("BSP(L2): Send second SIPI to cpu[%d]", id_map[1]);
9884 
9885 		/* Delay a while to check whether second SIPI would cause VMExit */
9886 		delay(SIPI_SIGNAL_TEST_DELAY);
9887 
9888 		/* Test is done, notify AP to exit test */
9889 		vmx_set_test_stage(3);
9890 
9891 		/* wait AP exit non-root mode */
9892 		while (vmx_get_test_stage() != 5)
9893 			;
9894 	} else {
9895 		/* wait BSP notify test is done */
9896 		while (vmx_get_test_stage() != 3)
9897 			;
9898 
9899 		/* AP exit guest */
9900 		vmx_set_test_stage(4);
9901 	}
9902 }
9903 
9904 static void sipi_test_ap_thread(void *data)
9905 {
9906 	struct vmcs *ap_vmcs;
9907 	u64 *ap_vmxon_region;
9908 	void *ap_stack, *ap_syscall_stack;
9909 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9910 	u64 cpu_ctrl_1 = 0;
9911 
9912 	/* Enter VMX operation (i.e. exec VMXON) */
9913 	ap_vmxon_region = alloc_page();
9914 	enable_vmx();
9915 	init_vmx(ap_vmxon_region);
9916 	_vmx_on(ap_vmxon_region);
9917 	init_vmcs(&ap_vmcs);
9918 	make_vmcs_current(ap_vmcs);
9919 
9920 	/* Set stack for AP */
9921 	ap_stack = alloc_page();
9922 	ap_syscall_stack = alloc_page();
9923 	vmcs_write(GUEST_RSP, (u64)(ap_stack + PAGE_SIZE - 1));
9924 	vmcs_write(GUEST_SYSENTER_ESP, (u64)(ap_syscall_stack + PAGE_SIZE - 1));
9925 
9926 	/* passthrough lapic to L2 */
9927 	disable_intercept_for_x2apic_msrs();
9928 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9929 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9930 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9931 
9932 	/* Set guest activity state to wait-for-SIPI state */
9933 	vmcs_write(GUEST_ACTV_STATE, ACTV_WAIT_SIPI);
9934 
9935 	vmx_set_test_stage(1);
9936 
9937 	/* AP enter guest */
9938 	enter_guest();
9939 
9940 	if (vmcs_read(EXI_REASON) == VMX_SIPI) {
9941 		report_pass("AP: Handle SIPI VMExit");
9942 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
9943 		vmx_set_test_stage(2);
9944 	} else {
9945 		report_fail("AP: Unexpected VMExit, reason=%ld", vmcs_read(EXI_REASON));
9946 		vmx_off();
9947 		return;
9948 	}
9949 
9950 	/* AP enter guest */
9951 	enter_guest();
9952 
9953 	report(vmcs_read(EXI_REASON) != VMX_SIPI,
9954 		"AP: should no SIPI VMExit since activity is not in WAIT_SIPI state");
9955 
9956 	/* notify BSP that AP is already exit from non-root mode */
9957 	vmx_set_test_stage(5);
9958 
9959 	/* Leave VMX operation */
9960 	vmx_off();
9961 }
9962 
9963 static void vmx_sipi_signal_test(void)
9964 {
9965 	if (!(rdmsr(MSR_IA32_VMX_MISC) & MSR_IA32_VMX_MISC_ACTIVITY_WAIT_SIPI)) {
9966 		printf("\tACTIVITY_WAIT_SIPI state is not supported.\n");
9967 		return;
9968 	}
9969 
9970 	if (cpu_count() < 2) {
9971 		report_skip(__func__);
9972 		return;
9973 	}
9974 
9975 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9976 	u64 cpu_ctrl_1 = 0;
9977 
9978 	/* passthrough lapic to L2 */
9979 	disable_intercept_for_x2apic_msrs();
9980 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9981 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9982 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9983 
9984 	test_set_guest(vmx_sipi_test_guest);
9985 
9986 	/* update CR3 on AP */
9987 	on_cpu(1, update_cr3, (void *)read_cr3());
9988 
9989 	/* start AP */
9990 	on_cpu_async(1, sipi_test_ap_thread, NULL);
9991 
9992 	vmx_set_test_stage(0);
9993 
9994 	/* BSP enter guest */
9995 	enter_guest();
9996 }
9997 
9998 
9999 enum vmcs_access {
10000 	ACCESS_VMREAD,
10001 	ACCESS_VMWRITE,
10002 	ACCESS_NONE,
10003 };
10004 
10005 struct vmcs_shadow_test_common {
10006 	enum vmcs_access op;
10007 	enum Reason reason;
10008 	u64 field;
10009 	u64 value;
10010 	u64 flags;
10011 	u64 time;
10012 } l1_l2_common;
10013 
10014 static inline u64 vmread_flags(u64 field, u64 *val)
10015 {
10016 	u64 flags;
10017 
10018 	asm volatile ("vmread %2, %1; pushf; pop %0"
10019 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
10020 	return flags & X86_EFLAGS_ALU;
10021 }
10022 
10023 static inline u64 vmwrite_flags(u64 field, u64 val)
10024 {
10025 	u64 flags;
10026 
10027 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
10028 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
10029 	return flags & X86_EFLAGS_ALU;
10030 }
10031 
10032 static void vmx_vmcs_shadow_test_guest(void)
10033 {
10034 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10035 	u64 start;
10036 
10037 	while (c->op != ACCESS_NONE) {
10038 		start = rdtsc();
10039 		switch (c->op) {
10040 		default:
10041 			c->flags = -1ull;
10042 			break;
10043 		case ACCESS_VMREAD:
10044 			c->flags = vmread_flags(c->field, &c->value);
10045 			break;
10046 		case ACCESS_VMWRITE:
10047 			c->flags = vmwrite_flags(c->field, 0);
10048 			break;
10049 		}
10050 		c->time = rdtsc() - start;
10051 		vmcall();
10052 	}
10053 }
10054 
10055 static u64 vmread_from_shadow(u64 field)
10056 {
10057 	struct vmcs *primary;
10058 	struct vmcs *shadow;
10059 	u64 value;
10060 
10061 	TEST_ASSERT(!vmcs_save(&primary));
10062 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10063 	TEST_ASSERT(!make_vmcs_current(shadow));
10064 	value = vmcs_read(field);
10065 	TEST_ASSERT(!make_vmcs_current(primary));
10066 	return value;
10067 }
10068 
10069 static u64 vmwrite_to_shadow(u64 field, u64 value)
10070 {
10071 	struct vmcs *primary;
10072 	struct vmcs *shadow;
10073 
10074 	TEST_ASSERT(!vmcs_save(&primary));
10075 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10076 	TEST_ASSERT(!make_vmcs_current(shadow));
10077 	vmcs_write(field, value);
10078 	value = vmcs_read(field);
10079 	TEST_ASSERT(!make_vmcs_current(primary));
10080 	return value;
10081 }
10082 
10083 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
10084 {
10085 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10086 
10087 	c->op = access;
10088 	vmcs_write(VMX_INST_ERROR, 0);
10089 	enter_guest();
10090 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
10091 	if (c->reason != VMX_VMCALL) {
10092 		skip_exit_insn();
10093 		enter_guest();
10094 	}
10095 	skip_exit_vmcall();
10096 }
10097 
10098 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
10099 {
10100 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10101 	struct vmcs *shadow;
10102 	u64 value;
10103 	uintptr_t flags[2];
10104 	bool good_shadow;
10105 	u32 vmx_inst_error;
10106 
10107 	report_prefix_pushf("field %lx", field);
10108 	c->field = field;
10109 
10110 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10111 	if (shadow != (struct vmcs *)-1ull) {
10112 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
10113 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
10114 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
10115 	} else {
10116 		/*
10117 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
10118 		 * shadowed-fields should fail with setting RFLAGS.CF.
10119 		 */
10120 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
10121 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
10122 		good_shadow = false;
10123 	}
10124 
10125 	/* Intercept both VMREAD and VMWRITE. */
10126 	report_prefix_push("no VMREAD/VMWRITE permission");
10127 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
10128 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
10129 		set_bit(field, bitmap[ACCESS_VMREAD]);
10130 		set_bit(field, bitmap[ACCESS_VMWRITE]);
10131 	}
10132 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10133 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10134 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10135 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10136 	report_prefix_pop();
10137 
10138 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
10139 		goto out;
10140 
10141 	/* Permit shadowed VMREAD. */
10142 	report_prefix_push("VMREAD permission only");
10143 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10144 	set_bit(field, bitmap[ACCESS_VMWRITE]);
10145 	if (good_shadow)
10146 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10147 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10148 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10149 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10150 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10151 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10152 	       c->time);
10153 	report(c->flags == flags[ACCESS_VMREAD],
10154 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10155 	       c->flags, flags[ACCESS_VMREAD]);
10156 	if (good_shadow)
10157 		report(c->value == value,
10158 		       "value read from shadow (%lx) is as expected (%lx)",
10159 		       c->value, value);
10160 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10161 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10162 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10163 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10164 	report_prefix_pop();
10165 
10166 	/* Permit shadowed VMWRITE. */
10167 	report_prefix_push("VMWRITE permission only");
10168 	set_bit(field, bitmap[ACCESS_VMREAD]);
10169 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10170 	if (good_shadow)
10171 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10172 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10173 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10174 	report(c->reason == VMX_VMCALL,
10175 		"shadowed for VMWRITE (in %ld cycles)",
10176 		c->time);
10177 	report(c->flags == flags[ACCESS_VMREAD],
10178 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10179 	       c->flags, flags[ACCESS_VMREAD]);
10180 	if (good_shadow) {
10181 		value = vmread_from_shadow(field);
10182 		report(value == 0,
10183 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10184 		       0ul);
10185 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10186 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10187 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10188 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10189 	}
10190 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10191 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10192 	report_prefix_pop();
10193 
10194 	/* Permit shadowed VMREAD and VMWRITE. */
10195 	report_prefix_push("VMREAD and VMWRITE permission");
10196 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10197 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10198 	if (good_shadow)
10199 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10200 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10201 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10202 	report(c->reason == VMX_VMCALL,
10203 		"shadowed for VMWRITE (in %ld cycles)",
10204 		c->time);
10205 	report(c->flags == flags[ACCESS_VMREAD],
10206 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10207 	       c->flags, flags[ACCESS_VMREAD]);
10208 	if (good_shadow) {
10209 		value = vmread_from_shadow(field);
10210 		report(value == 0,
10211 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10212 		       0ul);
10213 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10214 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10215 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10216 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10217 	}
10218 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10219 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10220 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10221 	       c->time);
10222 	report(c->flags == flags[ACCESS_VMREAD],
10223 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10224 	       c->flags, flags[ACCESS_VMREAD]);
10225 	if (good_shadow)
10226 		report(c->value == 0,
10227 		       "value read from shadow (%lx) is as expected (%lx)",
10228 		       c->value, 0ul);
10229 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10230 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10231 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10232 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10233 	report_prefix_pop();
10234 
10235 out:
10236 	report_prefix_pop();
10237 }
10238 
10239 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
10240 {
10241 	unsigned base;
10242 	unsigned index;
10243 	unsigned bit;
10244 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
10245 
10246 	/* Run test on all possible valid VMCS fields */
10247 	for (base = 0;
10248 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
10249 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
10250 		for (index = 0; index <= highest_index; index++)
10251 			vmcs_shadow_test_field(bitmap, base + index);
10252 
10253 	/*
10254 	 * Run tests on some invalid VMCS fields
10255 	 * (Have reserved bit set).
10256 	 */
10257 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
10258 		vmcs_shadow_test_field(bitmap, (1ull << bit));
10259 }
10260 
10261 static void vmx_vmcs_shadow_test(void)
10262 {
10263 	u8 *bitmap[2];
10264 	struct vmcs *shadow;
10265 
10266 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
10267 		printf("\t'Activate secondary controls' not supported.\n");
10268 		return;
10269 	}
10270 
10271 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
10272 		printf("\t'VMCS shadowing' not supported.\n");
10273 		return;
10274 	}
10275 
10276 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
10277 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
10278 		printf("\tVMWRITE can't modify VM-exit information fields.\n");
10279 		return;
10280 	}
10281 
10282 	test_set_guest(vmx_vmcs_shadow_test_guest);
10283 
10284 	bitmap[ACCESS_VMREAD] = alloc_page();
10285 	bitmap[ACCESS_VMWRITE] = alloc_page();
10286 
10287 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
10288 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
10289 
10290 	shadow = alloc_page();
10291 	shadow->hdr.revision_id = basic.revision;
10292 	shadow->hdr.shadow_vmcs = 1;
10293 	TEST_ASSERT(!vmcs_clear(shadow));
10294 
10295 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
10296 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
10297 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
10298 
10299 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
10300 	report_prefix_push("valid link pointer");
10301 	vmx_vmcs_shadow_test_body(bitmap);
10302 	report_prefix_pop();
10303 
10304 	vmcs_write(VMCS_LINK_PTR, -1ull);
10305 	report_prefix_push("invalid link pointer");
10306 	vmx_vmcs_shadow_test_body(bitmap);
10307 	report_prefix_pop();
10308 
10309 	l1_l2_common.op = ACCESS_NONE;
10310 	enter_guest();
10311 }
10312 
10313 /*
10314  * This test monitors the difference between a guest RDTSC instruction
10315  * and the IA32_TIME_STAMP_COUNTER MSR value stored in the VMCS12
10316  * VM-exit MSR-store list when taking a VM-exit on the instruction
10317  * following RDTSC.
10318  */
10319 #define RDTSC_DIFF_ITERS 100000
10320 #define RDTSC_DIFF_FAILS 100
10321 #define HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD 750
10322 
10323 /*
10324  * Set 'use TSC offsetting' and set the guest offset to the
10325  * inverse of the host's current TSC value, so that the guest starts running
10326  * with an effective TSC value of 0.
10327  */
10328 static void reset_guest_tsc_to_zero(void)
10329 {
10330 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
10331 	vmcs_write(TSC_OFFSET, -rdtsc());
10332 }
10333 
10334 static void rdtsc_vmexit_diff_test_guest(void)
10335 {
10336 	int i;
10337 
10338 	for (i = 0; i < RDTSC_DIFF_ITERS; i++)
10339 		/* Ensure rdtsc is the last instruction before the vmcall. */
10340 		asm volatile("rdtsc; vmcall" : : : "eax", "edx");
10341 }
10342 
10343 /*
10344  * This function only considers the "use TSC offsetting" VM-execution
10345  * control.  It does not handle "use TSC scaling" (because the latter
10346  * isn't available to the host today.)
10347  */
10348 static unsigned long long host_time_to_guest_time(unsigned long long t)
10349 {
10350 	TEST_ASSERT(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
10351 		    !(vmcs_read(CPU_EXEC_CTRL1) & CPU_USE_TSC_SCALING));
10352 
10353 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_USE_TSC_OFFSET)
10354 		t += vmcs_read(TSC_OFFSET);
10355 
10356 	return t;
10357 }
10358 
10359 static unsigned long long rdtsc_vmexit_diff_test_iteration(void)
10360 {
10361 	unsigned long long guest_tsc, host_to_guest_tsc;
10362 
10363 	enter_guest();
10364 	skip_exit_vmcall();
10365 	guest_tsc = (u32) regs.rax + (regs.rdx << 32);
10366 	host_to_guest_tsc = host_time_to_guest_time(exit_msr_store[0].value);
10367 
10368 	return host_to_guest_tsc - guest_tsc;
10369 }
10370 
10371 static void rdtsc_vmexit_diff_test(void)
10372 {
10373 	unsigned long long delta;
10374 	int fail = 0;
10375 	int i;
10376 
10377 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET))
10378 		test_skip("CPU doesn't support the 'use TSC offsetting' processor-based VM-execution control.\n");
10379 
10380 	test_set_guest(rdtsc_vmexit_diff_test_guest);
10381 
10382 	reset_guest_tsc_to_zero();
10383 
10384 	/*
10385 	 * Set up the VMCS12 VM-exit MSR-store list to store just one
10386 	 * MSR: IA32_TIME_STAMP_COUNTER. Note that the value stored is
10387 	 * in the host time domain (i.e., it is not adjusted according
10388 	 * to the TSC multiplier and TSC offset fields in the VMCS12,
10389 	 * as a guest RDTSC would be.)
10390 	 */
10391 	exit_msr_store = alloc_page();
10392 	exit_msr_store[0].index = MSR_IA32_TSC;
10393 	vmcs_write(EXI_MSR_ST_CNT, 1);
10394 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(exit_msr_store));
10395 
10396 	for (i = 0; i < RDTSC_DIFF_ITERS && fail < RDTSC_DIFF_FAILS; i++) {
10397 		delta = rdtsc_vmexit_diff_test_iteration();
10398 		if (delta >= HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD)
10399 			fail++;
10400 	}
10401 
10402 	enter_guest();
10403 
10404 	report(fail < RDTSC_DIFF_FAILS,
10405 	       "RDTSC to VM-exit delta too high in %d of %d iterations, last = %llu",
10406 	       fail, i, delta);
10407 }
10408 
10409 static int invalid_msr_init(struct vmcs *vmcs)
10410 {
10411 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
10412 		printf("\tPreemption timer is not supported\n");
10413 		return VMX_TEST_EXIT;
10414 	}
10415 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
10416 	preempt_val = 10000000;
10417 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
10418 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
10419 
10420 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
10421 		printf("\tSave preemption value is not supported\n");
10422 
10423 	vmcs_write(ENT_MSR_LD_CNT, 1);
10424 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
10425 
10426 	return VMX_TEST_START;
10427 }
10428 
10429 
10430 static void invalid_msr_main(void)
10431 {
10432 	report_fail("Invalid MSR load");
10433 }
10434 
10435 static int invalid_msr_exit_handler(union exit_reason exit_reason)
10436 {
10437 	report_fail("Invalid MSR load");
10438 	print_vmexit_info(exit_reason);
10439 	return VMX_TEST_EXIT;
10440 }
10441 
10442 static int invalid_msr_entry_failure(struct vmentry_result *result)
10443 {
10444 	report(result->exit_reason.failed_vmentry &&
10445 	       result->exit_reason.basic == VMX_FAIL_MSR, "Invalid MSR load");
10446 	return VMX_TEST_VMEXIT;
10447 }
10448 
10449 /*
10450  * The max number of MSRs in an atomic switch MSR list is:
10451  * (111B + 1) * 512 = 4096
10452  *
10453  * Each list entry consumes:
10454  * 4-byte MSR index + 4 bytes reserved + 8-byte data = 16 bytes
10455  *
10456  * Allocate 128 kB to cover max_msr_list_size (i.e., 64 kB) and then some.
10457  */
10458 static const u32 msr_list_page_order = 5;
10459 
10460 static void atomic_switch_msr_limit_test_guest(void)
10461 {
10462 	vmcall();
10463 }
10464 
10465 static void populate_msr_list(struct vmx_msr_entry *msr_list,
10466 			      size_t byte_capacity, int count)
10467 {
10468 	int i;
10469 
10470 	for (i = 0; i < count; i++) {
10471 		msr_list[i].index = MSR_IA32_TSC;
10472 		msr_list[i].reserved = 0;
10473 		msr_list[i].value = 0x1234567890abcdef;
10474 	}
10475 
10476 	memset(msr_list + count, 0xff,
10477 	       byte_capacity - count * sizeof(*msr_list));
10478 }
10479 
10480 static int max_msr_list_size(void)
10481 {
10482 	u32 vmx_misc = rdmsr(MSR_IA32_VMX_MISC);
10483 	u32 factor = ((vmx_misc & GENMASK(27, 25)) >> 25) + 1;
10484 
10485 	return factor * 512;
10486 }
10487 
10488 static void atomic_switch_msrs_test(int count)
10489 {
10490 	struct vmx_msr_entry *vm_enter_load;
10491         struct vmx_msr_entry *vm_exit_load;
10492         struct vmx_msr_entry *vm_exit_store;
10493 	int max_allowed = max_msr_list_size();
10494 	int byte_capacity = 1ul << (msr_list_page_order + PAGE_SHIFT);
10495 	/* Exceeding the max MSR list size at exit trigers KVM to abort. */
10496 	int exit_count = count > max_allowed ? max_allowed : count;
10497 	int cleanup_count = count > max_allowed ? 2 : 1;
10498 	int i;
10499 
10500 	/*
10501 	 * Check for the IA32_TSC MSR,
10502 	 * available with the "TSC flag" and used to populate the MSR lists.
10503 	 */
10504 	if (!(cpuid(1).d & (1 << 4))) {
10505 		report_skip(__func__);
10506 		return;
10507 	}
10508 
10509 	/* Set L2 guest. */
10510 	test_set_guest(atomic_switch_msr_limit_test_guest);
10511 
10512 	/* Setup atomic MSR switch lists. */
10513 	vm_enter_load = alloc_pages(msr_list_page_order);
10514 	vm_exit_load = alloc_pages(msr_list_page_order);
10515 	vm_exit_store = alloc_pages(msr_list_page_order);
10516 
10517 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)vm_enter_load);
10518 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)vm_exit_load);
10519 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)vm_exit_store);
10520 
10521 	/*
10522 	 * VM-Enter should succeed up to the max number of MSRs per list, and
10523 	 * should not consume junk beyond the last entry.
10524 	 */
10525 	populate_msr_list(vm_enter_load, byte_capacity, count);
10526 	populate_msr_list(vm_exit_load, byte_capacity, exit_count);
10527 	populate_msr_list(vm_exit_store, byte_capacity, exit_count);
10528 
10529 	vmcs_write(ENT_MSR_LD_CNT, count);
10530 	vmcs_write(EXI_MSR_LD_CNT, exit_count);
10531 	vmcs_write(EXI_MSR_ST_CNT, exit_count);
10532 
10533 	if (count <= max_allowed) {
10534 		enter_guest();
10535 		assert_exit_reason(VMX_VMCALL);
10536 		skip_exit_vmcall();
10537 	} else {
10538 		u32 exit_qual;
10539 
10540 		test_guest_state("Invalid MSR Load Count", true, count,
10541 				 "ENT_MSR_LD_CNT");
10542 
10543 		exit_qual = vmcs_read(EXI_QUALIFICATION);
10544 		report(exit_qual == max_allowed + 1, "exit_qual, %u, is %u.",
10545 		       exit_qual, max_allowed + 1);
10546 	}
10547 
10548 	/* Cleanup. */
10549 	vmcs_write(ENT_MSR_LD_CNT, 0);
10550 	vmcs_write(EXI_MSR_LD_CNT, 0);
10551 	vmcs_write(EXI_MSR_ST_CNT, 0);
10552 	for (i = 0; i < cleanup_count; i++) {
10553 		enter_guest();
10554 		skip_exit_vmcall();
10555 	}
10556 	free_pages_by_order(vm_enter_load, msr_list_page_order);
10557 	free_pages_by_order(vm_exit_load, msr_list_page_order);
10558 	free_pages_by_order(vm_exit_store, msr_list_page_order);
10559 }
10560 
10561 static void atomic_switch_max_msrs_test(void)
10562 {
10563 	atomic_switch_msrs_test(max_msr_list_size());
10564 }
10565 
10566 static void atomic_switch_overflow_msrs_test(void)
10567 {
10568 	if (test_device_enabled())
10569 		atomic_switch_msrs_test(max_msr_list_size() + 1);
10570 	else
10571 		test_skip("Test is only supported on KVM");
10572 }
10573 
10574 static void vmx_pf_exception_test_guest(void)
10575 {
10576 	ac_test_run(PT_LEVEL_PML4);
10577 }
10578 
10579 typedef void (*invalidate_tlb_t)(void *data);
10580 
10581 static void __vmx_pf_exception_test(invalidate_tlb_t inv_fn, void *data)
10582 {
10583 	u64 efer;
10584 	struct cpuid cpuid;
10585 
10586 	test_set_guest(vmx_pf_exception_test_guest);
10587 
10588 	/* Intercept INVLPG when to perform TLB invalidation from L1 (this). */
10589 	if (inv_fn)
10590 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INVLPG);
10591 	else
10592 		vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INVLPG);
10593 
10594 	enter_guest();
10595 
10596 	while (vmcs_read(EXI_REASON) != VMX_VMCALL) {
10597 		switch (vmcs_read(EXI_REASON)) {
10598 		case VMX_RDMSR:
10599 			assert(regs.rcx == MSR_EFER);
10600 			efer = vmcs_read(GUEST_EFER);
10601 			regs.rdx = efer >> 32;
10602 			regs.rax = efer & 0xffffffff;
10603 			break;
10604 		case VMX_WRMSR:
10605 			assert(regs.rcx == MSR_EFER);
10606 			efer = regs.rdx << 32 | (regs.rax & 0xffffffff);
10607 			vmcs_write(GUEST_EFER, efer);
10608 			break;
10609 		case VMX_CPUID:
10610 			cpuid = (struct cpuid) {0, 0, 0, 0};
10611 			cpuid = raw_cpuid(regs.rax, regs.rcx);
10612 			regs.rax = cpuid.a;
10613 			regs.rbx = cpuid.b;
10614 			regs.rcx = cpuid.c;
10615 			regs.rdx = cpuid.d;
10616 			break;
10617 		case VMX_INVLPG:
10618 			inv_fn(data);
10619 			break;
10620 		default:
10621 			assert_msg(false,
10622 				"Unexpected exit to L1, exit_reason: %s (0x%lx)",
10623 				exit_reason_description(vmcs_read(EXI_REASON)),
10624 				vmcs_read(EXI_REASON));
10625 		}
10626 		skip_exit_insn();
10627 		enter_guest();
10628 	}
10629 
10630 	assert_exit_reason(VMX_VMCALL);
10631 }
10632 
10633 static void vmx_pf_exception_test(void)
10634 {
10635 	__vmx_pf_exception_test(NULL, NULL);
10636 }
10637 
10638 static void invalidate_tlb_no_vpid(void *data)
10639 {
10640 	/* If VPID is disabled, the TLB is flushed on VM-Enter and VM-Exit. */
10641 }
10642 
10643 static void vmx_pf_no_vpid_test(void)
10644 {
10645 	if (is_vpid_supported())
10646 		vmcs_clear_bits(CPU_EXEC_CTRL1, CPU_VPID);
10647 
10648 	__vmx_pf_exception_test(invalidate_tlb_no_vpid, NULL);
10649 }
10650 
10651 static void invalidate_tlb_invvpid_addr(void *data)
10652 {
10653 	invvpid(INVVPID_ALL, *(u16 *)data, vmcs_read(EXI_QUALIFICATION));
10654 }
10655 
10656 static void invalidate_tlb_new_vpid(void *data)
10657 {
10658 	u16 *vpid = data;
10659 
10660 	/*
10661 	 * Bump VPID to effectively flush L2's TLB from L0's perspective.
10662 	 * Invalidate all VPIDs when the VPID wraps to zero as hardware/KVM is
10663 	 * architecturally allowed to keep TLB entries indefinitely.
10664 	 */
10665 	++(*vpid);
10666 	if (*vpid == 0) {
10667 		++(*vpid);
10668 		invvpid(INVVPID_ALL, 0, 0);
10669 	}
10670 	vmcs_write(VPID, *vpid);
10671 }
10672 
10673 static void __vmx_pf_vpid_test(invalidate_tlb_t inv_fn, u16 vpid)
10674 {
10675 	if (!is_vpid_supported())
10676 		test_skip("VPID unsupported");
10677 
10678 	if (!is_invvpid_supported())
10679 		test_skip("INVVPID unsupported");
10680 
10681 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
10682 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VPID);
10683 	vmcs_write(VPID, vpid);
10684 
10685 	__vmx_pf_exception_test(inv_fn, &vpid);
10686 }
10687 
10688 static void vmx_pf_invvpid_test(void)
10689 {
10690 	if (!is_invvpid_type_supported(INVVPID_ADDR))
10691 		test_skip("INVVPID ADDR unsupported");
10692 
10693 	__vmx_pf_vpid_test(invalidate_tlb_invvpid_addr, 0xaaaa);
10694 }
10695 
10696 static void vmx_pf_vpid_test(void)
10697 {
10698 	/* Need INVVPID(ALL) to flush VPIDs upon wrap/reuse. */
10699 	if (!is_invvpid_type_supported(INVVPID_ALL))
10700 		test_skip("INVVPID ALL unsupported");
10701 
10702 	__vmx_pf_vpid_test(invalidate_tlb_new_vpid, 1);
10703 }
10704 
10705 static void vmx_l2_gp_test(void)
10706 {
10707 	*(volatile u64 *)NONCANONICAL = 0;
10708 }
10709 
10710 static void vmx_l2_ud_test(void)
10711 {
10712 	asm volatile ("ud2");
10713 }
10714 
10715 static void vmx_l2_de_test(void)
10716 {
10717 	asm volatile (
10718 		"xor %%eax, %%eax\n\t"
10719 		"xor %%ebx, %%ebx\n\t"
10720 		"xor %%edx, %%edx\n\t"
10721 		"idiv %%ebx\n\t"
10722 		::: "eax", "ebx", "edx");
10723 }
10724 
10725 static void vmx_l2_bp_test(void)
10726 {
10727 	asm volatile ("int3");
10728 }
10729 
10730 static void vmx_l2_db_test(void)
10731 {
10732 	write_rflags(read_rflags() | X86_EFLAGS_TF);
10733 }
10734 
10735 static uint64_t usermode_callback(void)
10736 {
10737 	/*
10738 	 * Trigger an #AC by writing 8 bytes to a 4-byte aligned address.
10739 	 * Disclaimer: It is assumed that the stack pointer is aligned
10740 	 * on a 16-byte boundary as x86_64 stacks should be.
10741 	 */
10742 	asm volatile("movq $0, -0x4(%rsp)");
10743 
10744 	return 0;
10745 }
10746 
10747 static void vmx_l2_ac_test(void)
10748 {
10749 	bool hit_ac = false;
10750 
10751 	write_cr0(read_cr0() | X86_CR0_AM);
10752 	write_rflags(read_rflags() | X86_EFLAGS_AC);
10753 
10754 	run_in_user(usermode_callback, AC_VECTOR, 0, 0, 0, 0, &hit_ac);
10755 	report(hit_ac, "Usermode #AC handled in L2");
10756 	vmcall();
10757 }
10758 
10759 struct vmx_exception_test {
10760 	u8 vector;
10761 	void (*guest_code)(void);
10762 };
10763 
10764 struct vmx_exception_test vmx_exception_tests[] = {
10765 	{ GP_VECTOR, vmx_l2_gp_test },
10766 	{ UD_VECTOR, vmx_l2_ud_test },
10767 	{ DE_VECTOR, vmx_l2_de_test },
10768 	{ DB_VECTOR, vmx_l2_db_test },
10769 	{ BP_VECTOR, vmx_l2_bp_test },
10770 	{ AC_VECTOR, vmx_l2_ac_test },
10771 };
10772 
10773 static u8 vmx_exception_test_vector;
10774 
10775 static void vmx_exception_handler(struct ex_regs *regs)
10776 {
10777 	report(regs->vector == vmx_exception_test_vector,
10778 	       "Handling %s in L2's exception handler",
10779 	       exception_mnemonic(vmx_exception_test_vector));
10780 	vmcall();
10781 }
10782 
10783 static void handle_exception_in_l2(u8 vector)
10784 {
10785 	handler old_handler = handle_exception(vector, vmx_exception_handler);
10786 
10787 	vmx_exception_test_vector = vector;
10788 
10789 	enter_guest();
10790 	report(vmcs_read(EXI_REASON) == VMX_VMCALL,
10791 	       "%s handled by L2", exception_mnemonic(vector));
10792 
10793 	handle_exception(vector, old_handler);
10794 }
10795 
10796 static void handle_exception_in_l1(u32 vector)
10797 {
10798 	u32 old_eb = vmcs_read(EXC_BITMAP);
10799 
10800 	vmcs_write(EXC_BITMAP, old_eb | (1u << vector));
10801 
10802 	enter_guest();
10803 
10804 	report((vmcs_read(EXI_REASON) == VMX_EXC_NMI) &&
10805 	       ((vmcs_read(EXI_INTR_INFO) & 0xff) == vector),
10806 	       "%s handled by L1", exception_mnemonic(vector));
10807 
10808 	vmcs_write(EXC_BITMAP, old_eb);
10809 }
10810 
10811 static void vmx_exception_test(void)
10812 {
10813 	struct vmx_exception_test *t;
10814 	int i;
10815 
10816 	for (i = 0; i < ARRAY_SIZE(vmx_exception_tests); i++) {
10817 		t = &vmx_exception_tests[i];
10818 
10819 		/*
10820 		 * Override the guest code before each run even though it's the
10821 		 * same code, the VMCS guest state needs to be reinitialized.
10822 		 */
10823 		test_override_guest(t->guest_code);
10824 		handle_exception_in_l2(t->vector);
10825 
10826 		test_override_guest(t->guest_code);
10827 		handle_exception_in_l1(t->vector);
10828 	}
10829 
10830 	test_set_guest_finished();
10831 }
10832 
10833 #define TEST(name) { #name, .v2 = name }
10834 
10835 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
10836 struct vmx_test vmx_tests[] = {
10837 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
10838 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
10839 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
10840 		preemption_timer_exit_handler, NULL, {0} },
10841 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
10842 		test_ctrl_pat_exit_handler, NULL, {0} },
10843 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
10844 		test_ctrl_efer_exit_handler, NULL, {0} },
10845 	{ "CR shadowing", NULL, cr_shadowing_main,
10846 		cr_shadowing_exit_handler, NULL, {0} },
10847 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
10848 		NULL, {0} },
10849 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
10850 		insn_intercept_exit_handler, NULL, {0} },
10851 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
10852 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
10853 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
10854 	{ "interrupt", interrupt_init, interrupt_main,
10855 		interrupt_exit_handler, NULL, {0} },
10856 	{ "nmi_hlt", nmi_hlt_init, nmi_hlt_main,
10857 		nmi_hlt_exit_handler, NULL, {0} },
10858 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
10859 		NULL, {0} },
10860 	{ "MSR switch", msr_switch_init, msr_switch_main,
10861 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
10862 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
10863 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
10864 		disable_rdtscp_exit_handler, NULL, {0} },
10865 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
10866 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
10867 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
10868 		exit_monitor_from_l2_handler, NULL, {0} },
10869 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
10870 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
10871 	/* Basic V2 tests. */
10872 	TEST(v2_null_test),
10873 	TEST(v2_multiple_entries_test),
10874 	TEST(fixture_test_case1),
10875 	TEST(fixture_test_case2),
10876 	/* Opcode tests. */
10877 	TEST(invvpid_test),
10878 	/* VM-entry tests */
10879 	TEST(vmx_controls_test),
10880 	TEST(vmx_host_state_area_test),
10881 	TEST(vmx_guest_state_area_test),
10882 	TEST(vmentry_movss_shadow_test),
10883 	TEST(vmentry_unrestricted_guest_test),
10884 	/* APICv tests */
10885 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
10886 	TEST(vmx_hlt_with_rvi_test),
10887 	TEST(apic_reg_virt_test),
10888 	TEST(virt_x2apic_mode_test),
10889 	/* APIC pass-through tests */
10890 	TEST(vmx_apic_passthrough_test),
10891 	TEST(vmx_apic_passthrough_thread_test),
10892 	TEST(vmx_apic_passthrough_tpr_threshold_test),
10893 	TEST(vmx_init_signal_test),
10894 	TEST(vmx_sipi_signal_test),
10895 	/* VMCS Shadowing tests */
10896 	TEST(vmx_vmcs_shadow_test),
10897 	/* Regression tests */
10898 	TEST(vmx_ldtr_test),
10899 	TEST(vmx_cr_load_test),
10900 	TEST(vmx_cr4_osxsave_test),
10901 	TEST(vmx_nm_test),
10902 	TEST(vmx_db_test),
10903 	TEST(vmx_nmi_window_test),
10904 	TEST(vmx_intr_window_test),
10905 	TEST(vmx_pending_event_test),
10906 	TEST(vmx_pending_event_hlt_test),
10907 	TEST(vmx_store_tsc_test),
10908 	TEST(vmx_preemption_timer_zero_test),
10909 	TEST(vmx_preemption_timer_tf_test),
10910 	TEST(vmx_preemption_timer_expiry_test),
10911 	/* EPT access tests. */
10912 	TEST(ept_access_test_not_present),
10913 	TEST(ept_access_test_read_only),
10914 	TEST(ept_access_test_write_only),
10915 	TEST(ept_access_test_read_write),
10916 	TEST(ept_access_test_execute_only),
10917 	TEST(ept_access_test_read_execute),
10918 	TEST(ept_access_test_write_execute),
10919 	TEST(ept_access_test_read_write_execute),
10920 	TEST(ept_access_test_reserved_bits),
10921 	TEST(ept_access_test_ignored_bits),
10922 	TEST(ept_access_test_paddr_not_present_ad_disabled),
10923 	TEST(ept_access_test_paddr_not_present_ad_enabled),
10924 	TEST(ept_access_test_paddr_read_only_ad_disabled),
10925 	TEST(ept_access_test_paddr_read_only_ad_enabled),
10926 	TEST(ept_access_test_paddr_read_write),
10927 	TEST(ept_access_test_paddr_read_write_execute),
10928 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
10929 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
10930 	TEST(ept_access_test_paddr_not_present_page_fault),
10931 	TEST(ept_access_test_force_2m_page),
10932 	/* Atomic MSR switch tests. */
10933 	TEST(atomic_switch_max_msrs_test),
10934 	TEST(atomic_switch_overflow_msrs_test),
10935 	TEST(rdtsc_vmexit_diff_test),
10936 	TEST(vmx_mtf_test),
10937 	TEST(vmx_mtf_pdpte_test),
10938 	TEST(vmx_pf_exception_test),
10939 	TEST(vmx_pf_no_vpid_test),
10940 	TEST(vmx_pf_invvpid_test),
10941 	TEST(vmx_pf_vpid_test),
10942 	TEST(vmx_exception_test),
10943 	{ NULL, NULL, NULL, NULL, NULL, {0} },
10944 };
10945