xref: /kvm-unit-tests/x86/vmx_tests.c (revision 222532747637a03d234151b75399483c2eb9189b)
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 
24 #define NONCANONICAL            0xaaaaaaaaaaaaaaaaull
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("Basic VMX test", 1);
59 }
60 
61 static int basic_exit_handler(void)
62 {
63 	report("Basic VMX test", 0);
64 	print_vmexit_info();
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("test vmlaunch", 1);
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("test vmresume", (rax == 0xFFFF) && (rsp == resume_rsp));
84 }
85 
86 static int vmenter_exit_handler(void)
87 {
88 	u64 guest_rip;
89 	ulong reason;
90 
91 	guest_rip = vmcs_read(GUEST_RIP);
92 	reason = vmcs_read(EXI_REASON) & 0xff;
93 	switch (reason) {
94 	case VMX_VMCALL:
95 		if (regs.rax != 0xABCD) {
96 			report("test vmresume", 0);
97 			return VMX_TEST_VMEXIT;
98 		}
99 		regs.rax = 0xFFFF;
100 		vmcs_write(GUEST_RIP, guest_rip + 3);
101 		return VMX_TEST_RESUME;
102 	default:
103 		report("test vmresume", 0);
104 		print_vmexit_info();
105 	}
106 	return VMX_TEST_VMEXIT;
107 }
108 
109 u32 preempt_scale;
110 volatile unsigned long long tsc_val;
111 volatile u32 preempt_val;
112 u64 saved_rip;
113 
114 static int preemption_timer_init(struct vmcs *vmcs)
115 {
116 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
117 		printf("\tPreemption timer is not supported\n");
118 		return VMX_TEST_EXIT;
119 	}
120 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
121 	preempt_val = 10000000;
122 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
123 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
124 
125 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
126 		printf("\tSave preemption value is not supported\n");
127 
128 	return VMX_TEST_START;
129 }
130 
131 static void preemption_timer_main(void)
132 {
133 	tsc_val = rdtsc();
134 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
135 		vmx_set_test_stage(0);
136 		vmcall();
137 		if (vmx_get_test_stage() == 1)
138 			vmcall();
139 	}
140 	vmx_set_test_stage(1);
141 	while (vmx_get_test_stage() == 1) {
142 		if (((rdtsc() - tsc_val) >> preempt_scale)
143 				> 10 * preempt_val) {
144 			vmx_set_test_stage(2);
145 			vmcall();
146 		}
147 	}
148 	tsc_val = rdtsc();
149 	asm volatile ("hlt");
150 	vmcall();
151 	vmx_set_test_stage(5);
152 	vmcall();
153 }
154 
155 static int preemption_timer_exit_handler(void)
156 {
157 	bool guest_halted;
158 	u64 guest_rip;
159 	ulong reason;
160 	u32 insn_len;
161 	u32 ctrl_exit;
162 
163 	guest_rip = vmcs_read(GUEST_RIP);
164 	reason = vmcs_read(EXI_REASON) & 0xff;
165 	insn_len = vmcs_read(EXI_INST_LEN);
166 	switch (reason) {
167 	case VMX_PREEMPT:
168 		switch (vmx_get_test_stage()) {
169 		case 1:
170 		case 2:
171 			report("busy-wait for preemption timer",
172 			       ((rdtsc() - tsc_val) >> preempt_scale) >=
173 			       preempt_val);
174 			vmx_set_test_stage(3);
175 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
176 			return VMX_TEST_RESUME;
177 		case 3:
178 			guest_halted =
179 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
180 			report("preemption timer during hlt",
181 			       ((rdtsc() - tsc_val) >> preempt_scale) >=
182 			       preempt_val && guest_halted);
183 			vmx_set_test_stage(4);
184 			vmcs_write(PIN_CONTROLS,
185 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
186 			vmcs_write(EXI_CONTROLS,
187 				   vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT);
188 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
189 			return VMX_TEST_RESUME;
190 		case 4:
191 			report("preemption timer with 0 value",
192 			       saved_rip == guest_rip);
193 			break;
194 		default:
195 			report("Invalid stage.", false);
196 			print_vmexit_info();
197 			break;
198 		}
199 		break;
200 	case VMX_VMCALL:
201 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
202 		switch (vmx_get_test_stage()) {
203 		case 0:
204 			report("Keep preemption value",
205 			       vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val);
206 			vmx_set_test_stage(1);
207 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
208 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
209 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
210 			vmcs_write(EXI_CONTROLS, ctrl_exit);
211 			return VMX_TEST_RESUME;
212 		case 1:
213 			report("Save preemption value",
214 			       vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val);
215 			return VMX_TEST_RESUME;
216 		case 2:
217 			report("busy-wait for preemption timer", 0);
218 			vmx_set_test_stage(3);
219 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
220 			return VMX_TEST_RESUME;
221 		case 3:
222 			report("preemption timer during hlt", 0);
223 			vmx_set_test_stage(4);
224 			/* fall through */
225 		case 4:
226 			vmcs_write(PIN_CONTROLS,
227 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
228 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
229 			saved_rip = guest_rip + insn_len;
230 			return VMX_TEST_RESUME;
231 		case 5:
232 			report("preemption timer with 0 value (vmcall stage 5)", 0);
233 			break;
234 		default:
235 			// Should not reach here
236 			report("unexpected stage, %d", false,
237 			       vmx_get_test_stage());
238 			print_vmexit_info();
239 			return VMX_TEST_VMEXIT;
240 		}
241 		break;
242 	default:
243 		report("Unknown exit reason, %ld", false, reason);
244 		print_vmexit_info();
245 	}
246 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
247 	return VMX_TEST_VMEXIT;
248 }
249 
250 static void msr_bmp_init(void)
251 {
252 	void *msr_bitmap;
253 	u32 ctrl_cpu0;
254 
255 	msr_bitmap = alloc_page();
256 	memset(msr_bitmap, 0x0, PAGE_SIZE);
257 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
258 	ctrl_cpu0 |= CPU_MSR_BITMAP;
259 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
260 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
261 }
262 
263 static void *get_msr_bitmap(void)
264 {
265 	void *msr_bitmap;
266 
267 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
268 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
269 	} else {
270 		msr_bitmap = alloc_page();
271 		memset(msr_bitmap, 0xff, PAGE_SIZE);
272 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
273 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
274 	}
275 
276 	return msr_bitmap;
277 }
278 
279 static void disable_intercept_for_x2apic_msrs(void)
280 {
281 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
282 	u32 msr;
283 
284 	for (msr = APIC_BASE_MSR;
285 		 msr < (APIC_BASE_MSR+0xff);
286 		 msr += BITS_PER_LONG) {
287 		unsigned int word = msr / BITS_PER_LONG;
288 
289 		msr_bitmap[word] = 0;
290 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
291 	}
292 }
293 
294 static int test_ctrl_pat_init(struct vmcs *vmcs)
295 {
296 	u64 ctrl_ent;
297 	u64 ctrl_exi;
298 
299 	msr_bmp_init();
300 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
301 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
302 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
303 		printf("\tSave/load PAT is not supported\n");
304 		return 1;
305 	}
306 
307 	ctrl_ent = vmcs_read(ENT_CONTROLS);
308 	ctrl_exi = vmcs_read(EXI_CONTROLS);
309 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
310 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
311 	vmcs_write(ENT_CONTROLS, ctrl_ent);
312 	vmcs_write(EXI_CONTROLS, ctrl_exi);
313 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
314 	vmcs_write(GUEST_PAT, 0x0);
315 	vmcs_write(HOST_PAT, ia32_pat);
316 	return VMX_TEST_START;
317 }
318 
319 static void test_ctrl_pat_main(void)
320 {
321 	u64 guest_ia32_pat;
322 
323 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
324 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
325 		printf("\tENT_LOAD_PAT is not supported.\n");
326 	else {
327 		if (guest_ia32_pat != 0) {
328 			report("Entry load PAT", 0);
329 			return;
330 		}
331 	}
332 	wrmsr(MSR_IA32_CR_PAT, 0x6);
333 	vmcall();
334 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
335 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
336 		report("Entry load PAT", guest_ia32_pat == ia32_pat);
337 }
338 
339 static int test_ctrl_pat_exit_handler(void)
340 {
341 	u64 guest_rip;
342 	ulong reason;
343 	u64 guest_pat;
344 
345 	guest_rip = vmcs_read(GUEST_RIP);
346 	reason = vmcs_read(EXI_REASON) & 0xff;
347 	switch (reason) {
348 	case VMX_VMCALL:
349 		guest_pat = vmcs_read(GUEST_PAT);
350 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
351 			printf("\tEXI_SAVE_PAT is not supported\n");
352 			vmcs_write(GUEST_PAT, 0x6);
353 		} else {
354 			report("Exit save PAT", guest_pat == 0x6);
355 		}
356 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
357 			printf("\tEXI_LOAD_PAT is not supported\n");
358 		else
359 			report("Exit load PAT", rdmsr(MSR_IA32_CR_PAT) == ia32_pat);
360 		vmcs_write(GUEST_PAT, ia32_pat);
361 		vmcs_write(GUEST_RIP, guest_rip + 3);
362 		return VMX_TEST_RESUME;
363 	default:
364 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
365 		break;
366 	}
367 	return VMX_TEST_VMEXIT;
368 }
369 
370 static int test_ctrl_efer_init(struct vmcs *vmcs)
371 {
372 	u64 ctrl_ent;
373 	u64 ctrl_exi;
374 
375 	msr_bmp_init();
376 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
377 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
378 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
379 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
380 	ia32_efer = rdmsr(MSR_EFER);
381 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
382 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
383 	return VMX_TEST_START;
384 }
385 
386 static void test_ctrl_efer_main(void)
387 {
388 	u64 guest_ia32_efer;
389 
390 	guest_ia32_efer = rdmsr(MSR_EFER);
391 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
392 		printf("\tENT_LOAD_EFER is not supported.\n");
393 	else {
394 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
395 			report("Entry load EFER", 0);
396 			return;
397 		}
398 	}
399 	wrmsr(MSR_EFER, ia32_efer);
400 	vmcall();
401 	guest_ia32_efer = rdmsr(MSR_EFER);
402 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
403 		report("Entry load EFER", guest_ia32_efer == ia32_efer);
404 }
405 
406 static int test_ctrl_efer_exit_handler(void)
407 {
408 	u64 guest_rip;
409 	ulong reason;
410 	u64 guest_efer;
411 
412 	guest_rip = vmcs_read(GUEST_RIP);
413 	reason = vmcs_read(EXI_REASON) & 0xff;
414 	switch (reason) {
415 	case VMX_VMCALL:
416 		guest_efer = vmcs_read(GUEST_EFER);
417 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
418 			printf("\tEXI_SAVE_EFER is not supported\n");
419 			vmcs_write(GUEST_EFER, ia32_efer);
420 		} else {
421 			report("Exit save EFER", guest_efer == ia32_efer);
422 		}
423 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
424 			printf("\tEXI_LOAD_EFER is not supported\n");
425 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
426 		} else {
427 			report("Exit load EFER", rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX));
428 		}
429 		vmcs_write(GUEST_PAT, ia32_efer);
430 		vmcs_write(GUEST_RIP, guest_rip + 3);
431 		return VMX_TEST_RESUME;
432 	default:
433 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
434 		break;
435 	}
436 	return VMX_TEST_VMEXIT;
437 }
438 
439 u32 guest_cr0, guest_cr4;
440 
441 static void cr_shadowing_main(void)
442 {
443 	u32 cr0, cr4, tmp;
444 
445 	// Test read through
446 	vmx_set_test_stage(0);
447 	guest_cr0 = read_cr0();
448 	if (vmx_get_test_stage() == 1)
449 		report("Read through CR0", 0);
450 	else
451 		vmcall();
452 	vmx_set_test_stage(1);
453 	guest_cr4 = read_cr4();
454 	if (vmx_get_test_stage() == 2)
455 		report("Read through CR4", 0);
456 	else
457 		vmcall();
458 	// Test write through
459 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
460 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
461 	vmx_set_test_stage(2);
462 	write_cr0(guest_cr0);
463 	if (vmx_get_test_stage() == 3)
464 		report("Write throuth CR0", 0);
465 	else
466 		vmcall();
467 	vmx_set_test_stage(3);
468 	write_cr4(guest_cr4);
469 	if (vmx_get_test_stage() == 4)
470 		report("Write through CR4", 0);
471 	else
472 		vmcall();
473 	// Test read shadow
474 	vmx_set_test_stage(4);
475 	vmcall();
476 	cr0 = read_cr0();
477 	if (vmx_get_test_stage() != 5)
478 		report("Read shadowing CR0", cr0 == guest_cr0);
479 	vmx_set_test_stage(5);
480 	cr4 = read_cr4();
481 	if (vmx_get_test_stage() != 6)
482 		report("Read shadowing CR4", cr4 == guest_cr4);
483 	// Test write shadow (same value with shadow)
484 	vmx_set_test_stage(6);
485 	write_cr0(guest_cr0);
486 	if (vmx_get_test_stage() == 7)
487 		report("Write shadowing CR0 (same value with shadow)", 0);
488 	else
489 		vmcall();
490 	vmx_set_test_stage(7);
491 	write_cr4(guest_cr4);
492 	if (vmx_get_test_stage() == 8)
493 		report("Write shadowing CR4 (same value with shadow)", 0);
494 	else
495 		vmcall();
496 	// Test write shadow (different value)
497 	vmx_set_test_stage(8);
498 	tmp = guest_cr0 ^ X86_CR0_TS;
499 	asm volatile("mov %0, %%rsi\n\t"
500 		"mov %%rsi, %%cr0\n\t"
501 		::"m"(tmp)
502 		:"rsi", "memory", "cc");
503 	report("Write shadowing different X86_CR0_TS", vmx_get_test_stage() == 9);
504 	vmx_set_test_stage(9);
505 	tmp = guest_cr0 ^ X86_CR0_MP;
506 	asm volatile("mov %0, %%rsi\n\t"
507 		"mov %%rsi, %%cr0\n\t"
508 		::"m"(tmp)
509 		:"rsi", "memory", "cc");
510 	report("Write shadowing different X86_CR0_MP", vmx_get_test_stage() == 10);
511 	vmx_set_test_stage(10);
512 	tmp = guest_cr4 ^ X86_CR4_TSD;
513 	asm volatile("mov %0, %%rsi\n\t"
514 		"mov %%rsi, %%cr4\n\t"
515 		::"m"(tmp)
516 		:"rsi", "memory", "cc");
517 	report("Write shadowing different X86_CR4_TSD", vmx_get_test_stage() == 11);
518 	vmx_set_test_stage(11);
519 	tmp = guest_cr4 ^ X86_CR4_DE;
520 	asm volatile("mov %0, %%rsi\n\t"
521 		"mov %%rsi, %%cr4\n\t"
522 		::"m"(tmp)
523 		:"rsi", "memory", "cc");
524 	report("Write shadowing different X86_CR4_DE", vmx_get_test_stage() == 12);
525 }
526 
527 static int cr_shadowing_exit_handler(void)
528 {
529 	u64 guest_rip;
530 	ulong reason;
531 	u32 insn_len;
532 	u32 exit_qual;
533 
534 	guest_rip = vmcs_read(GUEST_RIP);
535 	reason = vmcs_read(EXI_REASON) & 0xff;
536 	insn_len = vmcs_read(EXI_INST_LEN);
537 	exit_qual = vmcs_read(EXI_QUALIFICATION);
538 	switch (reason) {
539 	case VMX_VMCALL:
540 		switch (vmx_get_test_stage()) {
541 		case 0:
542 			report("Read through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
543 			break;
544 		case 1:
545 			report("Read through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
546 			break;
547 		case 2:
548 			report("Write through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
549 			break;
550 		case 3:
551 			report("Write through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
552 			break;
553 		case 4:
554 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
555 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
556 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
557 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
558 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
559 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
560 			break;
561 		case 6:
562 			report("Write shadowing CR0 (same value)",
563 					guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)));
564 			break;
565 		case 7:
566 			report("Write shadowing CR4 (same value)",
567 					guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)));
568 			break;
569 		default:
570 			// Should not reach here
571 			report("unexpected stage, %d", false,
572 			       vmx_get_test_stage());
573 			print_vmexit_info();
574 			return VMX_TEST_VMEXIT;
575 		}
576 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
577 		return VMX_TEST_RESUME;
578 	case VMX_CR:
579 		switch (vmx_get_test_stage()) {
580 		case 4:
581 			report("Read shadowing CR0", 0);
582 			vmx_inc_test_stage();
583 			break;
584 		case 5:
585 			report("Read shadowing CR4", 0);
586 			vmx_inc_test_stage();
587 			break;
588 		case 6:
589 			report("Write shadowing CR0 (same value)", 0);
590 			vmx_inc_test_stage();
591 			break;
592 		case 7:
593 			report("Write shadowing CR4 (same value)", 0);
594 			vmx_inc_test_stage();
595 			break;
596 		case 8:
597 		case 9:
598 			// 0x600 encodes "mov %esi, %cr0"
599 			if (exit_qual == 0x600)
600 				vmx_inc_test_stage();
601 			break;
602 		case 10:
603 		case 11:
604 			// 0x604 encodes "mov %esi, %cr4"
605 			if (exit_qual == 0x604)
606 				vmx_inc_test_stage();
607 			break;
608 		default:
609 			// Should not reach here
610 			report("unexpected stage, %d", false,
611 			       vmx_get_test_stage());
612 			print_vmexit_info();
613 			return VMX_TEST_VMEXIT;
614 		}
615 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
616 		return VMX_TEST_RESUME;
617 	default:
618 		report("Unknown exit reason, %ld", false, reason);
619 		print_vmexit_info();
620 	}
621 	return VMX_TEST_VMEXIT;
622 }
623 
624 static int iobmp_init(struct vmcs *vmcs)
625 {
626 	u32 ctrl_cpu0;
627 
628 	io_bitmap_a = alloc_page();
629 	io_bitmap_b = alloc_page();
630 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
631 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
632 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
633 	ctrl_cpu0 |= CPU_IO_BITMAP;
634 	ctrl_cpu0 &= (~CPU_IO);
635 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
636 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
637 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
638 	return VMX_TEST_START;
639 }
640 
641 static void iobmp_main(void)
642 {
643 	// stage 0, test IO pass
644 	vmx_set_test_stage(0);
645 	inb(0x5000);
646 	outb(0x0, 0x5000);
647 	report("I/O bitmap - I/O pass", vmx_get_test_stage() == 0);
648 	// test IO width, in/out
649 	((u8 *)io_bitmap_a)[0] = 0xFF;
650 	vmx_set_test_stage(2);
651 	inb(0x0);
652 	report("I/O bitmap - trap in", vmx_get_test_stage() == 3);
653 	vmx_set_test_stage(3);
654 	outw(0x0, 0x0);
655 	report("I/O bitmap - trap out", vmx_get_test_stage() == 4);
656 	vmx_set_test_stage(4);
657 	inl(0x0);
658 	report("I/O bitmap - I/O width, long", vmx_get_test_stage() == 5);
659 	// test low/high IO port
660 	vmx_set_test_stage(5);
661 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
662 	inb(0x5000);
663 	report("I/O bitmap - I/O port, low part", vmx_get_test_stage() == 6);
664 	vmx_set_test_stage(6);
665 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
666 	inb(0x9000);
667 	report("I/O bitmap - I/O port, high part", vmx_get_test_stage() == 7);
668 	// test partial pass
669 	vmx_set_test_stage(7);
670 	inl(0x4FFF);
671 	report("I/O bitmap - partial pass", vmx_get_test_stage() == 8);
672 	// test overrun
673 	vmx_set_test_stage(8);
674 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
675 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
676 	inl(0xFFFF);
677 	report("I/O bitmap - overrun", vmx_get_test_stage() == 9);
678 	vmx_set_test_stage(9);
679 	vmcall();
680 	outb(0x0, 0x0);
681 	report("I/O bitmap - ignore unconditional exiting",
682 	       vmx_get_test_stage() == 9);
683 	vmx_set_test_stage(10);
684 	vmcall();
685 	outb(0x0, 0x0);
686 	report("I/O bitmap - unconditional exiting",
687 	       vmx_get_test_stage() == 11);
688 }
689 
690 static int iobmp_exit_handler(void)
691 {
692 	u64 guest_rip;
693 	ulong reason, exit_qual;
694 	u32 insn_len, ctrl_cpu0;
695 
696 	guest_rip = vmcs_read(GUEST_RIP);
697 	reason = vmcs_read(EXI_REASON) & 0xff;
698 	exit_qual = vmcs_read(EXI_QUALIFICATION);
699 	insn_len = vmcs_read(EXI_INST_LEN);
700 	switch (reason) {
701 	case VMX_IO:
702 		switch (vmx_get_test_stage()) {
703 		case 0:
704 		case 1:
705 			vmx_inc_test_stage();
706 			break;
707 		case 2:
708 			report("I/O bitmap - I/O width, byte",
709 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE);
710 			report("I/O bitmap - I/O direction, in", exit_qual & VMX_IO_IN);
711 			vmx_inc_test_stage();
712 			break;
713 		case 3:
714 			report("I/O bitmap - I/O width, word",
715 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD);
716 			report("I/O bitmap - I/O direction, out",
717 					!(exit_qual & VMX_IO_IN));
718 			vmx_inc_test_stage();
719 			break;
720 		case 4:
721 			report("I/O bitmap - I/O width, long",
722 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG);
723 			vmx_inc_test_stage();
724 			break;
725 		case 5:
726 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
727 				vmx_inc_test_stage();
728 			break;
729 		case 6:
730 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
731 				vmx_inc_test_stage();
732 			break;
733 		case 7:
734 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
735 				vmx_inc_test_stage();
736 			break;
737 		case 8:
738 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
739 				vmx_inc_test_stage();
740 			break;
741 		case 9:
742 		case 10:
743 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
744 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
745 			vmx_inc_test_stage();
746 			break;
747 		default:
748 			// Should not reach here
749 			report("unexpected stage, %d", false,
750 			       vmx_get_test_stage());
751 			print_vmexit_info();
752 			return VMX_TEST_VMEXIT;
753 		}
754 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
755 		return VMX_TEST_RESUME;
756 	case VMX_VMCALL:
757 		switch (vmx_get_test_stage()) {
758 		case 9:
759 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
760 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
761 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
762 			break;
763 		case 10:
764 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
765 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
766 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
767 			break;
768 		default:
769 			// Should not reach here
770 			report("unexpected stage, %d", false,
771 			       vmx_get_test_stage());
772 			print_vmexit_info();
773 			return VMX_TEST_VMEXIT;
774 		}
775 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
776 		return VMX_TEST_RESUME;
777 	default:
778 		printf("guest_rip = %#lx\n", guest_rip);
779 		printf("\tERROR : Undefined exit reason, reason = %ld.\n", reason);
780 		break;
781 	}
782 	return VMX_TEST_VMEXIT;
783 }
784 
785 #define INSN_CPU0		0
786 #define INSN_CPU1		1
787 #define INSN_ALWAYS_TRAP	2
788 
789 #define FIELD_EXIT_QUAL		(1 << 0)
790 #define FIELD_INSN_INFO		(1 << 1)
791 
792 asm(
793 	"insn_hlt: hlt;ret\n\t"
794 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
795 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
796 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
797 	"insn_rdtsc: rdtsc;ret\n\t"
798 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
799 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
800 #ifdef __x86_64__
801 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
802 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
803 #endif
804 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
805 	"insn_pause: pause;ret\n\t"
806 	"insn_wbinvd: wbinvd;ret\n\t"
807 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
808 	"insn_invd: invd;ret\n\t"
809 	"insn_sgdt: sgdt gdt64_desc;ret\n\t"
810 	"insn_lgdt: lgdt gdt64_desc;ret\n\t"
811 	"insn_sidt: sidt idt_descr;ret\n\t"
812 	"insn_lidt: lidt idt_descr;ret\n\t"
813 	"insn_sldt: sldt %ax;ret\n\t"
814 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
815 	"insn_str: str %ax;ret\n\t"
816 	"insn_rdrand: rdrand %rax;ret\n\t"
817 	"insn_rdseed: rdseed %rax;ret\n\t"
818 );
819 extern void insn_hlt(void);
820 extern void insn_invlpg(void);
821 extern void insn_mwait(void);
822 extern void insn_rdpmc(void);
823 extern void insn_rdtsc(void);
824 extern void insn_cr3_load(void);
825 extern void insn_cr3_store(void);
826 #ifdef __x86_64__
827 extern void insn_cr8_load(void);
828 extern void insn_cr8_store(void);
829 #endif
830 extern void insn_monitor(void);
831 extern void insn_pause(void);
832 extern void insn_wbinvd(void);
833 extern void insn_sgdt(void);
834 extern void insn_lgdt(void);
835 extern void insn_sidt(void);
836 extern void insn_lidt(void);
837 extern void insn_sldt(void);
838 extern void insn_lldt(void);
839 extern void insn_str(void);
840 extern void insn_cpuid(void);
841 extern void insn_invd(void);
842 extern void insn_rdrand(void);
843 extern void insn_rdseed(void);
844 
845 u32 cur_insn;
846 u64 cr3;
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 };
860 
861 /*
862  * Add more test cases of instruction intercept here. Elements in this
863  * table is:
864  *	name/control flag/insn function/type/exit reason/exit qulification/
865  *	instruction info/field to test
866  * The last field defines which fields (exit_qual and insn_info) need to be
867  * tested in exit handler. If set to 0, only "reason" is checked.
868  */
869 static struct insn_table insn_table[] = {
870 	// Flags for Primary Processor-Based VM-Execution Controls
871 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
872 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
873 		0x12345678, 0, FIELD_EXIT_QUAL},
874 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0},
875 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0},
876 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
877 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
878 		FIELD_EXIT_QUAL},
879 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
880 		FIELD_EXIT_QUAL},
881 #ifdef __x86_64__
882 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
883 		FIELD_EXIT_QUAL},
884 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
885 		FIELD_EXIT_QUAL},
886 #endif
887 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0},
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;
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 	return VMX_TEST_START;
918 }
919 
920 static void insn_intercept_main(void)
921 {
922 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
923 		vmx_set_test_stage(cur_insn * 2);
924 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
925 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
926 		    (insn_table[cur_insn].type == INSN_CPU1 &&
927 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
928 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
929 			       insn_table[cur_insn].type - INSN_CPU0,
930 			       insn_table[cur_insn].name);
931 			continue;
932 		}
933 
934 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
935 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
936 		    (insn_table[cur_insn].type == INSN_CPU1 &&
937 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
938 			/* skip hlt, it stalls the guest and is tested below */
939 			if (insn_table[cur_insn].insn_func != insn_hlt)
940 				insn_table[cur_insn].insn_func();
941 			report("execute %s", vmx_get_test_stage() == cur_insn * 2,
942 					insn_table[cur_insn].name);
943 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
944 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
945 			       insn_table[cur_insn].type - INSN_CPU0,
946 			       insn_table[cur_insn].name);
947 
948 		vmcall();
949 
950 		insn_table[cur_insn].insn_func();
951 		report("intercept %s", vmx_get_test_stage() == cur_insn * 2 + 1,
952 				insn_table[cur_insn].name);
953 
954 		vmx_set_test_stage(cur_insn * 2 + 1);
955 		vmcall();
956 	}
957 }
958 
959 static int insn_intercept_exit_handler(void)
960 {
961 	u64 guest_rip;
962 	u32 reason;
963 	ulong exit_qual;
964 	u32 insn_len;
965 	u32 insn_info;
966 	bool pass;
967 
968 	guest_rip = vmcs_read(GUEST_RIP);
969 	reason = vmcs_read(EXI_REASON) & 0xff;
970 	exit_qual = vmcs_read(EXI_QUALIFICATION);
971 	insn_len = vmcs_read(EXI_INST_LEN);
972 	insn_info = vmcs_read(EXI_INST_INFO);
973 
974 	if (reason == VMX_VMCALL) {
975 		u32 val = 0;
976 
977 		if (insn_table[cur_insn].type == INSN_CPU0)
978 			val = vmcs_read(CPU_EXEC_CTRL0);
979 		else if (insn_table[cur_insn].type == INSN_CPU1)
980 			val = vmcs_read(CPU_EXEC_CTRL1);
981 
982 		if (vmx_get_test_stage() & 1)
983 			val &= ~insn_table[cur_insn].flag;
984 		else
985 			val |= insn_table[cur_insn].flag;
986 
987 		if (insn_table[cur_insn].type == INSN_CPU0)
988 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
989 		else if (insn_table[cur_insn].type == INSN_CPU1)
990 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
991 	} else {
992 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
993 			insn_table[cur_insn].reason == reason;
994 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
995 		    insn_table[cur_insn].exit_qual != exit_qual)
996 			pass = false;
997 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
998 		    insn_table[cur_insn].insn_info != insn_info)
999 			pass = false;
1000 		if (pass)
1001 			vmx_inc_test_stage();
1002 	}
1003 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1004 	return VMX_TEST_RESUME;
1005 }
1006 
1007 
1008 /* Enables EPT and sets up the identity map. */
1009 static int setup_ept(bool enable_ad)
1010 {
1011 	unsigned long end_of_memory;
1012 
1013 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1014 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1015 		printf("\tEPT is not supported");
1016 		return 1;
1017 	}
1018 
1019 
1020 	if (!(ept_vpid.val & EPT_CAP_UC) &&
1021 			!(ept_vpid.val & EPT_CAP_WB)) {
1022 		printf("\tEPT paging-structure memory type "
1023 				"UC&WB are not supported\n");
1024 		return 1;
1025 	}
1026 	if (ept_vpid.val & EPT_CAP_UC)
1027 		eptp = EPT_MEM_TYPE_UC;
1028 	else
1029 		eptp = EPT_MEM_TYPE_WB;
1030 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
1031 		printf("\tPWL4 is not supported\n");
1032 		return 1;
1033 	}
1034 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY);
1035 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT);
1036 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1037 	pml4 = alloc_page();
1038 	memset(pml4, 0, PAGE_SIZE);
1039 	eptp |= virt_to_phys(pml4);
1040 	if (enable_ad)
1041 		eptp |= EPTP_AD_FLAG;
1042 	vmcs_write(EPTP, eptp);
1043 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1044 	if (end_of_memory < (1ul << 32))
1045 		end_of_memory = (1ul << 32);
1046 	/* Cannot use large EPT pages if we need to track EPT
1047 	 * accessed/dirty bits at 4K granularity.
1048 	 */
1049 	setup_ept_range(pml4, 0, end_of_memory, 0,
1050 			!enable_ad && ept_2m_supported(),
1051 			EPT_WA | EPT_RA | EPT_EA);
1052 	return 0;
1053 }
1054 
1055 static void ept_enable_ad_bits(void)
1056 {
1057 	eptp |= EPTP_AD_FLAG;
1058 	vmcs_write(EPTP, eptp);
1059 }
1060 
1061 static void ept_disable_ad_bits(void)
1062 {
1063 	eptp &= ~EPTP_AD_FLAG;
1064 	vmcs_write(EPTP, eptp);
1065 }
1066 
1067 static void ept_enable_ad_bits_or_skip_test(void)
1068 {
1069 	if (!ept_ad_bits_supported())
1070 		test_skip("EPT AD bits not supported.");
1071 	ept_enable_ad_bits();
1072 }
1073 
1074 static int apic_version;
1075 
1076 static int ept_init_common(bool have_ad)
1077 {
1078 	int ret;
1079 	struct pci_dev pcidev;
1080 
1081 	if (setup_ept(have_ad))
1082 		return VMX_TEST_EXIT;
1083 	data_page1 = alloc_page();
1084 	data_page2 = alloc_page();
1085 	memset(data_page1, 0x0, PAGE_SIZE);
1086 	memset(data_page2, 0x0, PAGE_SIZE);
1087 	*((u32 *)data_page1) = MAGIC_VAL_1;
1088 	*((u32 *)data_page2) = MAGIC_VAL_2;
1089 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1090 			EPT_RA | EPT_WA | EPT_EA);
1091 
1092 	apic_version = apic_read(APIC_LVR);
1093 
1094 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1095 	if (ret != PCIDEVADDR_INVALID) {
1096 		pci_dev_init(&pcidev, ret);
1097 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1098 	}
1099 
1100 	return VMX_TEST_START;
1101 }
1102 
1103 static int ept_init(struct vmcs *vmcs)
1104 {
1105 	return ept_init_common(false);
1106 }
1107 
1108 static void ept_common(void)
1109 {
1110 	vmx_set_test_stage(0);
1111 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1112 			*((u32 *)data_page1) != MAGIC_VAL_1)
1113 		report("EPT basic framework - read", 0);
1114 	else {
1115 		*((u32 *)data_page2) = MAGIC_VAL_3;
1116 		vmcall();
1117 		if (vmx_get_test_stage() == 1) {
1118 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1119 					*((u32 *)data_page2) == MAGIC_VAL_2)
1120 				report("EPT basic framework", 1);
1121 			else
1122 				report("EPT basic framework - remap", 1);
1123 		}
1124 	}
1125 	// Test EPT Misconfigurations
1126 	vmx_set_test_stage(1);
1127 	vmcall();
1128 	*((u32 *)data_page1) = MAGIC_VAL_1;
1129 	if (vmx_get_test_stage() != 2) {
1130 		report("EPT misconfigurations", 0);
1131 		goto t1;
1132 	}
1133 	vmx_set_test_stage(2);
1134 	vmcall();
1135 	*((u32 *)data_page1) = MAGIC_VAL_1;
1136 	report("EPT misconfigurations", vmx_get_test_stage() == 3);
1137 t1:
1138 	// Test EPT violation
1139 	vmx_set_test_stage(3);
1140 	vmcall();
1141 	*((u32 *)data_page1) = MAGIC_VAL_1;
1142 	report("EPT violation - page permission", vmx_get_test_stage() == 4);
1143 	// Violation caused by EPT paging structure
1144 	vmx_set_test_stage(4);
1145 	vmcall();
1146 	*((u32 *)data_page1) = MAGIC_VAL_2;
1147 	report("EPT violation - paging structure", vmx_get_test_stage() == 5);
1148 
1149 	// MMIO Read/Write
1150 	vmx_set_test_stage(5);
1151 	vmcall();
1152 
1153 	*(u32 volatile *)pci_physaddr;
1154 	report("MMIO EPT violation - read", vmx_get_test_stage() == 6);
1155 
1156 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1157 	report("MMIO EPT violation - write", vmx_get_test_stage() == 7);
1158 }
1159 
1160 static void ept_main(void)
1161 {
1162 	ept_common();
1163 
1164 	// Test EPT access to L1 MMIO
1165 	vmx_set_test_stage(7);
1166 	report("EPT - MMIO access", *((u32 *)0xfee00030UL) == apic_version);
1167 
1168 	// Test invalid operand for INVEPT
1169 	vmcall();
1170 	report("EPT - unsupported INVEPT", vmx_get_test_stage() == 8);
1171 }
1172 
1173 static bool invept_test(int type, u64 eptp)
1174 {
1175 	bool ret, supported;
1176 
1177 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1178 	ret = invept(type, eptp);
1179 
1180 	if (ret == !supported)
1181 		return false;
1182 
1183 	if (!supported)
1184 		printf("WARNING: unsupported invept passed!\n");
1185 	else
1186 		printf("WARNING: invept failed!\n");
1187 
1188 	return true;
1189 }
1190 
1191 static int pml_exit_handler(void)
1192 {
1193 	u16 index, count;
1194 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1195 	u64 *pmlbuf = pml_log;
1196 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1197 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1198 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1199 
1200 	switch (reason) {
1201 	case VMX_VMCALL:
1202 		switch (vmx_get_test_stage()) {
1203 		case 0:
1204 			index = vmcs_read(GUEST_PML_INDEX);
1205 			for (count = index + 1; count < PML_INDEX; count++) {
1206 				if (pmlbuf[count] == (u64)data_page2) {
1207 					vmx_inc_test_stage();
1208 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1209 					break;
1210 				}
1211 			}
1212 			break;
1213 		case 1:
1214 			index = vmcs_read(GUEST_PML_INDEX);
1215 			/* Keep clearing the dirty bit till a overflow */
1216 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1217 			break;
1218 		default:
1219 			report("unexpected stage, %d.", false,
1220 			       vmx_get_test_stage());
1221 			print_vmexit_info();
1222 			return VMX_TEST_VMEXIT;
1223 		}
1224 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1225 		return VMX_TEST_RESUME;
1226 	case VMX_PML_FULL:
1227 		vmx_inc_test_stage();
1228 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1229 		return VMX_TEST_RESUME;
1230 	default:
1231 		report("Unknown exit reason, %ld", false, reason);
1232 		print_vmexit_info();
1233 	}
1234 	return VMX_TEST_VMEXIT;
1235 }
1236 
1237 static int ept_exit_handler_common(bool have_ad)
1238 {
1239 	u64 guest_rip;
1240 	u64 guest_cr3;
1241 	ulong reason;
1242 	u32 insn_len;
1243 	u32 exit_qual;
1244 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte;
1245 
1246 	guest_rip = vmcs_read(GUEST_RIP);
1247 	guest_cr3 = vmcs_read(GUEST_CR3);
1248 	reason = vmcs_read(EXI_REASON) & 0xff;
1249 	insn_len = vmcs_read(EXI_INST_LEN);
1250 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1251 	switch (reason) {
1252 	case VMX_VMCALL:
1253 		switch (vmx_get_test_stage()) {
1254 		case 0:
1255 			check_ept_ad(pml4, guest_cr3,
1256 				     (unsigned long)data_page1,
1257 				     have_ad ? EPT_ACCESS_FLAG : 0,
1258 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1259 			check_ept_ad(pml4, guest_cr3,
1260 				     (unsigned long)data_page2,
1261 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1262 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1263 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1264 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1265 			if (have_ad)
1266 				ept_sync(INVEPT_SINGLE, eptp);;
1267 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1268 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1269 				vmx_inc_test_stage();
1270 				install_ept(pml4, (unsigned long)data_page2,
1271 						(unsigned long)data_page2,
1272 						EPT_RA | EPT_WA | EPT_EA);
1273 			} else
1274 				report("EPT basic framework - write", 0);
1275 			break;
1276 		case 1:
1277 			install_ept(pml4, (unsigned long)data_page1,
1278  				(unsigned long)data_page1, EPT_WA);
1279 			ept_sync(INVEPT_SINGLE, eptp);
1280 			break;
1281 		case 2:
1282 			install_ept(pml4, (unsigned long)data_page1,
1283  				(unsigned long)data_page1,
1284  				EPT_RA | EPT_WA | EPT_EA |
1285  				(2 << EPT_MEM_TYPE_SHIFT));
1286 			ept_sync(INVEPT_SINGLE, eptp);
1287 			break;
1288 		case 3:
1289 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1290 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1291 						1, &data_page1_pte));
1292 			set_ept_pte(pml4, (unsigned long)data_page1,
1293 				1, data_page1_pte & ~EPT_PRESENT);
1294 			ept_sync(INVEPT_SINGLE, eptp);
1295 			break;
1296 		case 4:
1297 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1298 						2, &data_page1_pte));
1299 			data_page1_pte &= PAGE_MASK;
1300 			TEST_ASSERT(get_ept_pte(pml4, data_page1_pte,
1301 						2, &data_page1_pte_pte));
1302 			set_ept_pte(pml4, data_page1_pte, 2,
1303 				data_page1_pte_pte & ~EPT_PRESENT);
1304 			ept_sync(INVEPT_SINGLE, eptp);
1305 			break;
1306 		case 5:
1307 			install_ept(pml4, (unsigned long)pci_physaddr,
1308 				(unsigned long)pci_physaddr, 0);
1309 			ept_sync(INVEPT_SINGLE, eptp);
1310 			break;
1311 		case 7:
1312 			if (!invept_test(0, eptp))
1313 				vmx_inc_test_stage();
1314 			break;
1315 		// Should not reach here
1316 		default:
1317 			report("ERROR - unexpected stage, %d.", false,
1318 			       vmx_get_test_stage());
1319 			print_vmexit_info();
1320 			return VMX_TEST_VMEXIT;
1321 		}
1322 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1323 		return VMX_TEST_RESUME;
1324 	case VMX_EPT_MISCONFIG:
1325 		switch (vmx_get_test_stage()) {
1326 		case 1:
1327 		case 2:
1328 			vmx_inc_test_stage();
1329 			install_ept(pml4, (unsigned long)data_page1,
1330  				(unsigned long)data_page1,
1331  				EPT_RA | EPT_WA | EPT_EA);
1332 			ept_sync(INVEPT_SINGLE, eptp);
1333 			break;
1334 		// Should not reach here
1335 		default:
1336 			report("ERROR - unexpected stage, %d.", false,
1337 			       vmx_get_test_stage());
1338 			print_vmexit_info();
1339 			return VMX_TEST_VMEXIT;
1340 		}
1341 		return VMX_TEST_RESUME;
1342 	case VMX_EPT_VIOLATION:
1343 		switch(vmx_get_test_stage()) {
1344 		case 3:
1345 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1346 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1347 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1348 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1349 					EPT_VLT_PADDR))
1350 				vmx_inc_test_stage();
1351 			set_ept_pte(pml4, (unsigned long)data_page1,
1352 				1, data_page1_pte | (EPT_PRESENT));
1353 			ept_sync(INVEPT_SINGLE, eptp);
1354 			break;
1355 		case 4:
1356 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1357 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1358 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1359 			if (exit_qual == (EPT_VLT_RD |
1360 					  (have_ad ? EPT_VLT_WR : 0) |
1361 					  EPT_VLT_LADDR_VLD))
1362 				vmx_inc_test_stage();
1363 			set_ept_pte(pml4, data_page1_pte, 2,
1364 				data_page1_pte_pte | (EPT_PRESENT));
1365 			ept_sync(INVEPT_SINGLE, eptp);
1366 			break;
1367 		case 5:
1368 			if (exit_qual & EPT_VLT_RD)
1369 				vmx_inc_test_stage();
1370 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1371 						1, &memaddr_pte));
1372 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1373 			ept_sync(INVEPT_SINGLE, eptp);
1374 			break;
1375 		case 6:
1376 			if (exit_qual & EPT_VLT_WR)
1377 				vmx_inc_test_stage();
1378 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1379 						1, &memaddr_pte));
1380 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1381 			ept_sync(INVEPT_SINGLE, eptp);
1382 			break;
1383 		default:
1384 			// Should not reach here
1385 			report("ERROR : unexpected stage, %d", false,
1386 			       vmx_get_test_stage());
1387 			print_vmexit_info();
1388 			return VMX_TEST_VMEXIT;
1389 		}
1390 		return VMX_TEST_RESUME;
1391 	default:
1392 		report("Unknown exit reason, %ld", false, reason);
1393 		print_vmexit_info();
1394 	}
1395 	return VMX_TEST_VMEXIT;
1396 }
1397 
1398 static int ept_exit_handler(void)
1399 {
1400 	return ept_exit_handler_common(false);
1401 }
1402 
1403 static int eptad_init(struct vmcs *vmcs)
1404 {
1405 	int r = ept_init_common(true);
1406 
1407 	if (r == VMX_TEST_EXIT)
1408 		return r;
1409 
1410 	if ((rdmsr(MSR_IA32_VMX_EPT_VPID_CAP) & EPT_CAP_AD_FLAG) == 0) {
1411 		printf("\tEPT A/D bits are not supported");
1412 		return VMX_TEST_EXIT;
1413 	}
1414 
1415 	return r;
1416 }
1417 
1418 static int pml_init(struct vmcs *vmcs)
1419 {
1420 	u32 ctrl_cpu;
1421 	int r = eptad_init(vmcs);
1422 
1423 	if (r == VMX_TEST_EXIT)
1424 		return r;
1425 
1426 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1427 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1428 		printf("\tPML is not supported");
1429 		return VMX_TEST_EXIT;
1430 	}
1431 
1432 	pml_log = alloc_page();
1433 	memset(pml_log, 0x0, PAGE_SIZE);
1434 	vmcs_write(PMLADDR, (u64)pml_log);
1435 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1436 
1437 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1438 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1439 
1440 	return VMX_TEST_START;
1441 }
1442 
1443 static void pml_main(void)
1444 {
1445 	int count = 0;
1446 
1447 	vmx_set_test_stage(0);
1448 	*((u32 *)data_page2) = 0x1;
1449 	vmcall();
1450 	report("PML - Dirty GPA Logging", vmx_get_test_stage() == 1);
1451 
1452 	while (vmx_get_test_stage() == 1) {
1453 		vmcall();
1454 		*((u32 *)data_page2) = 0x1;
1455 		if (count++ > PML_INDEX)
1456 			break;
1457 	}
1458 	report("PML Full Event", vmx_get_test_stage() == 2);
1459 }
1460 
1461 static void eptad_main(void)
1462 {
1463 	ept_common();
1464 }
1465 
1466 static int eptad_exit_handler(void)
1467 {
1468 	return ept_exit_handler_common(true);
1469 }
1470 
1471 static bool invvpid_test(int type, u16 vpid)
1472 {
1473 	bool ret, supported;
1474 
1475 	supported = ept_vpid.val &
1476 		(VPID_CAP_INVVPID_ADDR >> INVVPID_ADDR << type);
1477 	ret = invvpid(type, vpid, 0);
1478 
1479 	if (ret == !supported)
1480 		return false;
1481 
1482 	if (!supported)
1483 		printf("WARNING: unsupported invvpid passed!\n");
1484 	else
1485 		printf("WARNING: invvpid failed!\n");
1486 
1487 	return true;
1488 }
1489 
1490 static int vpid_init(struct vmcs *vmcs)
1491 {
1492 	u32 ctrl_cpu1;
1493 
1494 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1495 		!(ctrl_cpu_rev[1].clr & CPU_VPID)) {
1496 		printf("\tVPID is not supported");
1497 		return VMX_TEST_EXIT;
1498 	}
1499 
1500 	ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
1501 	ctrl_cpu1 |= CPU_VPID;
1502 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
1503 	return VMX_TEST_START;
1504 }
1505 
1506 static void vpid_main(void)
1507 {
1508 	vmx_set_test_stage(0);
1509 	vmcall();
1510 	report("INVVPID SINGLE ADDRESS", vmx_get_test_stage() == 1);
1511 	vmx_set_test_stage(2);
1512 	vmcall();
1513 	report("INVVPID SINGLE", vmx_get_test_stage() == 3);
1514 	vmx_set_test_stage(4);
1515 	vmcall();
1516 	report("INVVPID ALL", vmx_get_test_stage() == 5);
1517 }
1518 
1519 static int vpid_exit_handler(void)
1520 {
1521 	u64 guest_rip;
1522 	ulong reason;
1523 	u32 insn_len;
1524 
1525 	guest_rip = vmcs_read(GUEST_RIP);
1526 	reason = vmcs_read(EXI_REASON) & 0xff;
1527 	insn_len = vmcs_read(EXI_INST_LEN);
1528 
1529 	switch (reason) {
1530 	case VMX_VMCALL:
1531 		switch(vmx_get_test_stage()) {
1532 		case 0:
1533 			if (!invvpid_test(INVVPID_ADDR, 1))
1534 				vmx_inc_test_stage();
1535 			break;
1536 		case 2:
1537 			if (!invvpid_test(INVVPID_CONTEXT_GLOBAL, 1))
1538 				vmx_inc_test_stage();
1539 			break;
1540 		case 4:
1541 			if (!invvpid_test(INVVPID_ALL, 1))
1542 				vmx_inc_test_stage();
1543 			break;
1544 		default:
1545 			report("ERROR: unexpected stage, %d", false,
1546 					vmx_get_test_stage());
1547 			print_vmexit_info();
1548 			return VMX_TEST_VMEXIT;
1549 		}
1550 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1551 		return VMX_TEST_RESUME;
1552 	default:
1553 		report("Unknown exit reason, %ld", false, reason);
1554 		print_vmexit_info();
1555 	}
1556 	return VMX_TEST_VMEXIT;
1557 }
1558 
1559 #define TIMER_VECTOR	222
1560 
1561 static volatile bool timer_fired;
1562 
1563 static void timer_isr(isr_regs_t *regs)
1564 {
1565 	timer_fired = true;
1566 	apic_write(APIC_EOI, 0);
1567 }
1568 
1569 static int interrupt_init(struct vmcs *vmcs)
1570 {
1571 	msr_bmp_init();
1572 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1573 	handle_irq(TIMER_VECTOR, timer_isr);
1574 	return VMX_TEST_START;
1575 }
1576 
1577 static void interrupt_main(void)
1578 {
1579 	long long start, loops;
1580 
1581 	vmx_set_test_stage(0);
1582 
1583 	apic_write(APIC_LVTT, TIMER_VECTOR);
1584 	irq_enable();
1585 
1586 	apic_write(APIC_TMICT, 1);
1587 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1588 		asm volatile ("nop");
1589 	report("direct interrupt while running guest", timer_fired);
1590 
1591 	apic_write(APIC_TMICT, 0);
1592 	irq_disable();
1593 	vmcall();
1594 	timer_fired = false;
1595 	apic_write(APIC_TMICT, 1);
1596 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1597 		asm volatile ("nop");
1598 	report("intercepted interrupt while running guest", timer_fired);
1599 
1600 	irq_enable();
1601 	apic_write(APIC_TMICT, 0);
1602 	irq_disable();
1603 	vmcall();
1604 	timer_fired = false;
1605 	start = rdtsc();
1606 	apic_write(APIC_TMICT, 1000000);
1607 
1608 	asm volatile ("sti; hlt");
1609 
1610 	report("direct interrupt + hlt",
1611 	       rdtsc() - start > 1000000 && timer_fired);
1612 
1613 	apic_write(APIC_TMICT, 0);
1614 	irq_disable();
1615 	vmcall();
1616 	timer_fired = false;
1617 	start = rdtsc();
1618 	apic_write(APIC_TMICT, 1000000);
1619 
1620 	asm volatile ("sti; hlt");
1621 
1622 	report("intercepted interrupt + hlt",
1623 	       rdtsc() - start > 10000 && timer_fired);
1624 
1625 	apic_write(APIC_TMICT, 0);
1626 	irq_disable();
1627 	vmcall();
1628 	timer_fired = false;
1629 	start = rdtsc();
1630 	apic_write(APIC_TMICT, 1000000);
1631 
1632 	irq_enable();
1633 	asm volatile ("nop");
1634 	vmcall();
1635 
1636 	report("direct interrupt + activity state hlt",
1637 	       rdtsc() - start > 10000 && timer_fired);
1638 
1639 	apic_write(APIC_TMICT, 0);
1640 	irq_disable();
1641 	vmcall();
1642 	timer_fired = false;
1643 	start = rdtsc();
1644 	apic_write(APIC_TMICT, 1000000);
1645 
1646 	irq_enable();
1647 	asm volatile ("nop");
1648 	vmcall();
1649 
1650 	report("intercepted interrupt + activity state hlt",
1651 	       rdtsc() - start > 10000 && timer_fired);
1652 
1653 	apic_write(APIC_TMICT, 0);
1654 	irq_disable();
1655 	vmx_set_test_stage(7);
1656 	vmcall();
1657 	timer_fired = false;
1658 	apic_write(APIC_TMICT, 1);
1659 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1660 		asm volatile ("nop");
1661 	report("running a guest with interrupt acknowledgement set", timer_fired);
1662 
1663 	apic_write(APIC_TMICT, 0);
1664 	irq_enable();
1665 	timer_fired = false;
1666 	vmcall();
1667 	report("Inject an event to a halted guest", timer_fired);
1668 }
1669 
1670 static int interrupt_exit_handler(void)
1671 {
1672 	u64 guest_rip = vmcs_read(GUEST_RIP);
1673 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1674 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1675 
1676 	switch (reason) {
1677 	case VMX_VMCALL:
1678 		switch (vmx_get_test_stage()) {
1679 		case 0:
1680 		case 2:
1681 		case 5:
1682 			vmcs_write(PIN_CONTROLS,
1683 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1684 			break;
1685 		case 7:
1686 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1687 			vmcs_write(PIN_CONTROLS,
1688 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1689 			break;
1690 		case 1:
1691 		case 3:
1692 			vmcs_write(PIN_CONTROLS,
1693 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1694 			break;
1695 		case 4:
1696 		case 6:
1697 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1698 			break;
1699 
1700 		case 8:
1701 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1702 			vmcs_write(ENT_INTR_INFO,
1703 				   TIMER_VECTOR |
1704 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1705 				   INTR_INFO_VALID_MASK);
1706 			break;
1707 		}
1708 		vmx_inc_test_stage();
1709 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1710 		return VMX_TEST_RESUME;
1711 	case VMX_EXTINT:
1712 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1713 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1714 			handle_external_interrupt(vector);
1715 		} else {
1716 			irq_enable();
1717 			asm volatile ("nop");
1718 			irq_disable();
1719 		}
1720 		if (vmx_get_test_stage() >= 2)
1721 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1722 		return VMX_TEST_RESUME;
1723 	default:
1724 		report("Unknown exit reason, %ld", false, reason);
1725 		print_vmexit_info();
1726 	}
1727 
1728 	return VMX_TEST_VMEXIT;
1729 }
1730 
1731 static int dbgctls_init(struct vmcs *vmcs)
1732 {
1733 	u64 dr7 = 0x402;
1734 	u64 zero = 0;
1735 
1736 	msr_bmp_init();
1737 	asm volatile(
1738 		"mov %0,%%dr0\n\t"
1739 		"mov %0,%%dr1\n\t"
1740 		"mov %0,%%dr2\n\t"
1741 		"mov %1,%%dr7\n\t"
1742 		: : "r" (zero), "r" (dr7));
1743 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1744 	vmcs_write(GUEST_DR7, 0x404);
1745 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1746 
1747 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1748 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1749 
1750 	return VMX_TEST_START;
1751 }
1752 
1753 static void dbgctls_main(void)
1754 {
1755 	u64 dr7, debugctl;
1756 
1757 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1758 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1759 	/* Commented out: KVM does not support DEBUGCTL so far */
1760 	(void)debugctl;
1761 	report("Load debug controls", dr7 == 0x404 /* && debugctl == 0x2 */);
1762 
1763 	dr7 = 0x408;
1764 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1765 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1766 
1767 	vmx_set_test_stage(0);
1768 	vmcall();
1769 	report("Save debug controls", vmx_get_test_stage() == 1);
1770 
1771 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1772 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1773 		printf("\tDebug controls are always loaded/saved\n");
1774 		return;
1775 	}
1776 	vmx_set_test_stage(2);
1777 	vmcall();
1778 
1779 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1780 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1781 	/* Commented out: KVM does not support DEBUGCTL so far */
1782 	(void)debugctl;
1783 	report("Guest=host debug controls", dr7 == 0x402 /* && debugctl == 0x1 */);
1784 
1785 	dr7 = 0x408;
1786 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1787 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1788 
1789 	vmx_set_test_stage(3);
1790 	vmcall();
1791 	report("Don't save debug controls", vmx_get_test_stage() == 4);
1792 }
1793 
1794 static int dbgctls_exit_handler(void)
1795 {
1796 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
1797 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1798 	u64 guest_rip = vmcs_read(GUEST_RIP);
1799 	u64 dr7, debugctl;
1800 
1801 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1802 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1803 
1804 	switch (reason) {
1805 	case VMX_VMCALL:
1806 		switch (vmx_get_test_stage()) {
1807 		case 0:
1808 			if (dr7 == 0x400 && debugctl == 0 &&
1809 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1810 			    Commented out: KVM does not support DEBUGCTL so far
1811 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1812 				vmx_inc_test_stage();
1813 			break;
1814 		case 2:
1815 			dr7 = 0x402;
1816 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1817 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1818 			vmcs_write(GUEST_DR7, 0x404);
1819 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1820 
1821 			vmcs_write(ENT_CONTROLS,
1822 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1823 			vmcs_write(EXI_CONTROLS,
1824 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1825 			break;
1826 		case 3:
1827 			if (dr7 == 0x400 && debugctl == 0 &&
1828 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1829 			    Commented out: KVM does not support DEBUGCTL so far
1830 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1831 				vmx_inc_test_stage();
1832 			break;
1833 		}
1834 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1835 		return VMX_TEST_RESUME;
1836 	default:
1837 		report("Unknown exit reason, %d", false, reason);
1838 		print_vmexit_info();
1839 	}
1840 	return VMX_TEST_VMEXIT;
1841 }
1842 
1843 struct vmx_msr_entry {
1844 	u32 index;
1845 	u32 reserved;
1846 	u64 value;
1847 } __attribute__((packed));
1848 
1849 #define MSR_MAGIC 0x31415926
1850 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1851 
1852 static int msr_switch_init(struct vmcs *vmcs)
1853 {
1854 	msr_bmp_init();
1855 	exit_msr_store = alloc_page();
1856 	exit_msr_load = alloc_page();
1857 	entry_msr_load = alloc_page();
1858 	memset(exit_msr_store, 0, PAGE_SIZE);
1859 	memset(exit_msr_load, 0, PAGE_SIZE);
1860 	memset(entry_msr_load, 0, PAGE_SIZE);
1861 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1862 	entry_msr_load[0].value = MSR_MAGIC;
1863 
1864 	vmx_set_test_stage(1);
1865 	vmcs_write(ENT_MSR_LD_CNT, 1);
1866 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1867 	vmcs_write(EXI_MSR_ST_CNT, 1);
1868 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1869 	vmcs_write(EXI_MSR_LD_CNT, 1);
1870 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1871 	return VMX_TEST_START;
1872 }
1873 
1874 static void msr_switch_main(void)
1875 {
1876 	if (vmx_get_test_stage() == 1) {
1877 		report("VM entry MSR load",
1878 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC);
1879 		vmx_set_test_stage(2);
1880 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
1881 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
1882 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
1883 		exit_msr_load[0].value = MSR_MAGIC + 2;
1884 	}
1885 	vmcall();
1886 }
1887 
1888 static int msr_switch_exit_handler(void)
1889 {
1890 	ulong reason;
1891 
1892 	reason = vmcs_read(EXI_REASON);
1893 	if (reason == VMX_VMCALL && vmx_get_test_stage() == 2) {
1894 		report("VM exit MSR store",
1895 			exit_msr_store[0].value == MSR_MAGIC + 1);
1896 		report("VM exit MSR load",
1897 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2);
1898 		vmx_set_test_stage(3);
1899 		entry_msr_load[0].index = MSR_FS_BASE;
1900 		return VMX_TEST_RESUME;
1901 	}
1902 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1903 		__func__, vmx_get_test_stage(), reason);
1904 	return VMX_TEST_EXIT;
1905 }
1906 
1907 static int msr_switch_entry_failure(struct vmentry_failure *failure)
1908 {
1909 	ulong reason;
1910 
1911 	if (failure->early) {
1912 		printf("ERROR %s: early exit\n", __func__);
1913 		return VMX_TEST_EXIT;
1914 	}
1915 
1916 	reason = vmcs_read(EXI_REASON);
1917 	if (reason == (VMX_ENTRY_FAILURE | VMX_FAIL_MSR) &&
1918 	    vmx_get_test_stage() == 3) {
1919 		report("VM entry MSR load: try to load FS_BASE",
1920 			vmcs_read(EXI_QUALIFICATION) == 1);
1921 		return VMX_TEST_VMEXIT;
1922 	}
1923 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1924 		__func__, vmx_get_test_stage(), reason);
1925 	return VMX_TEST_EXIT;
1926 }
1927 
1928 static int vmmcall_init(struct vmcs *vmcs)
1929 {
1930 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
1931 	return VMX_TEST_START;
1932 }
1933 
1934 static void vmmcall_main(void)
1935 {
1936 	asm volatile(
1937 		"mov $0xABCD, %%rax\n\t"
1938 		"vmmcall\n\t"
1939 		::: "rax");
1940 
1941 	report("VMMCALL", 0);
1942 }
1943 
1944 static int vmmcall_exit_handler(void)
1945 {
1946 	ulong reason;
1947 
1948 	reason = vmcs_read(EXI_REASON);
1949 	switch (reason) {
1950 	case VMX_VMCALL:
1951 		printf("here\n");
1952 		report("VMMCALL triggers #UD", 0);
1953 		break;
1954 	case VMX_EXC_NMI:
1955 		report("VMMCALL triggers #UD",
1956 		       (vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR);
1957 		break;
1958 	default:
1959 		report("Unknown exit reason, %ld", false, reason);
1960 		print_vmexit_info();
1961 	}
1962 
1963 	return VMX_TEST_VMEXIT;
1964 }
1965 
1966 static int disable_rdtscp_init(struct vmcs *vmcs)
1967 {
1968 	u32 ctrl_cpu1;
1969 
1970 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
1971 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
1972 		ctrl_cpu1 &= ~CPU_RDTSCP;
1973 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
1974 	}
1975 
1976 	return VMX_TEST_START;
1977 }
1978 
1979 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
1980 {
1981 	switch (vmx_get_test_stage()) {
1982 	case 0:
1983 		report("RDTSCP triggers #UD", true);
1984 		vmx_inc_test_stage();
1985 		regs->rip += 3;
1986 		break;
1987 	case 2:
1988 		report("RDPID triggers #UD", true);
1989 		vmx_inc_test_stage();
1990 		regs->rip += 4;
1991 		break;
1992 	}
1993 	return;
1994 
1995 }
1996 
1997 static void disable_rdtscp_main(void)
1998 {
1999 	/* Test that #UD is properly injected in L2.  */
2000 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2001 
2002 	vmx_set_test_stage(0);
2003 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2004 	vmcall();
2005 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2006 
2007 	handle_exception(UD_VECTOR, 0);
2008 	vmcall();
2009 }
2010 
2011 static int disable_rdtscp_exit_handler(void)
2012 {
2013 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
2014 
2015 	switch (reason) {
2016 	case VMX_VMCALL:
2017 		switch (vmx_get_test_stage()) {
2018 		case 0:
2019 			report("RDTSCP triggers #UD", false);
2020 			vmx_inc_test_stage();
2021 			/* fallthrough */
2022 		case 1:
2023 			vmx_inc_test_stage();
2024 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2025 			return VMX_TEST_RESUME;
2026 		case 2:
2027 			report("RDPID triggers #UD", false);
2028 			break;
2029 		}
2030 		break;
2031 
2032 	default:
2033 		report("Unknown exit reason, %d", false, reason);
2034 		print_vmexit_info();
2035 	}
2036 	return VMX_TEST_VMEXIT;
2037 }
2038 
2039 static int int3_init(struct vmcs *vmcs)
2040 {
2041 	vmcs_write(EXC_BITMAP, ~0u);
2042 	return VMX_TEST_START;
2043 }
2044 
2045 static void int3_guest_main(void)
2046 {
2047 	asm volatile ("int3");
2048 }
2049 
2050 static int int3_exit_handler(void)
2051 {
2052 	u32 reason = vmcs_read(EXI_REASON);
2053 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2054 
2055 	report("L1 intercepts #BP", reason == VMX_EXC_NMI &&
2056 	       (intr_info & INTR_INFO_VALID_MASK) &&
2057 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2058 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2059 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2060 
2061 	return VMX_TEST_VMEXIT;
2062 }
2063 
2064 static int into_init(struct vmcs *vmcs)
2065 {
2066 	vmcs_write(EXC_BITMAP, ~0u);
2067 	return VMX_TEST_START;
2068 }
2069 
2070 static void into_guest_main(void)
2071 {
2072 	struct far_pointer32 fp = {
2073 		.offset = (uintptr_t)&&into,
2074 		.selector = KERNEL_CS32,
2075 	};
2076 	register uintptr_t rsp asm("rsp");
2077 
2078 	if (fp.offset != (uintptr_t)&&into) {
2079 		printf("Code address too high.\n");
2080 		return;
2081 	}
2082 	if ((u32)rsp != rsp) {
2083 		printf("Stack address too high.\n");
2084 		return;
2085 	}
2086 
2087 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2088 	return;
2089 into:
2090 	asm volatile (".code32;"
2091 		      "movl $0x7fffffff, %eax;"
2092 		      "addl %eax, %eax;"
2093 		      "into;"
2094 		      "lret;"
2095 		      ".code64");
2096 	__builtin_unreachable();
2097 }
2098 
2099 static int into_exit_handler(void)
2100 {
2101 	u32 reason = vmcs_read(EXI_REASON);
2102 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2103 
2104 	report("L1 intercepts #OF", reason == VMX_EXC_NMI &&
2105 	       (intr_info & INTR_INFO_VALID_MASK) &&
2106 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2107 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2108 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2109 
2110 	return VMX_TEST_VMEXIT;
2111 }
2112 
2113 static void exit_monitor_from_l2_main(void)
2114 {
2115 	printf("Calling exit(0) from l2...\n");
2116 	exit(0);
2117 }
2118 
2119 static int exit_monitor_from_l2_handler(void)
2120 {
2121 	report("The guest should have killed the VMM", false);
2122 	return VMX_TEST_EXIT;
2123 }
2124 
2125 static void assert_exit_reason(u64 expected)
2126 {
2127 	u64 actual = vmcs_read(EXI_REASON);
2128 
2129 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2130 			   exit_reason_description(expected),
2131 			   exit_reason_description(actual));
2132 }
2133 
2134 static void skip_exit_insn(void)
2135 {
2136 	u64 guest_rip = vmcs_read(GUEST_RIP);
2137 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2138 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2139 }
2140 
2141 static void skip_exit_vmcall(void)
2142 {
2143 	assert_exit_reason(VMX_VMCALL);
2144 	skip_exit_insn();
2145 }
2146 
2147 static void v2_null_test_guest(void)
2148 {
2149 }
2150 
2151 static void v2_null_test(void)
2152 {
2153 	test_set_guest(v2_null_test_guest);
2154 	enter_guest();
2155 	report(__func__, 1);
2156 }
2157 
2158 static void v2_multiple_entries_test_guest(void)
2159 {
2160 	vmx_set_test_stage(1);
2161 	vmcall();
2162 	vmx_set_test_stage(2);
2163 }
2164 
2165 static void v2_multiple_entries_test(void)
2166 {
2167 	test_set_guest(v2_multiple_entries_test_guest);
2168 	enter_guest();
2169 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2170 	skip_exit_vmcall();
2171 	enter_guest();
2172 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2173 	report(__func__, 1);
2174 }
2175 
2176 static int fixture_test_data = 1;
2177 
2178 static void fixture_test_teardown(void *data)
2179 {
2180 	*((int *) data) = 1;
2181 }
2182 
2183 static void fixture_test_guest(void)
2184 {
2185 	fixture_test_data++;
2186 }
2187 
2188 
2189 static void fixture_test_setup(void)
2190 {
2191 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2192 			   "fixture_test_teardown didn't run?!");
2193 	fixture_test_data = 2;
2194 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2195 	test_set_guest(fixture_test_guest);
2196 }
2197 
2198 static void fixture_test_case1(void)
2199 {
2200 	fixture_test_setup();
2201 	TEST_ASSERT_EQ(2, fixture_test_data);
2202 	enter_guest();
2203 	TEST_ASSERT_EQ(3, fixture_test_data);
2204 	report(__func__, 1);
2205 }
2206 
2207 static void fixture_test_case2(void)
2208 {
2209 	fixture_test_setup();
2210 	TEST_ASSERT_EQ(2, fixture_test_data);
2211 	enter_guest();
2212 	TEST_ASSERT_EQ(3, fixture_test_data);
2213 	report(__func__, 1);
2214 }
2215 
2216 enum ept_access_op {
2217 	OP_READ,
2218 	OP_WRITE,
2219 	OP_EXEC,
2220 	OP_FLUSH_TLB,
2221 	OP_EXIT,
2222 };
2223 
2224 static struct ept_access_test_data {
2225 	unsigned long gpa;
2226 	unsigned long *gva;
2227 	unsigned long hpa;
2228 	unsigned long *hva;
2229 	enum ept_access_op op;
2230 } ept_access_test_data;
2231 
2232 extern unsigned char ret42_start;
2233 extern unsigned char ret42_end;
2234 
2235 /* Returns 42. */
2236 asm(
2237 	".align 64\n"
2238 	"ret42_start:\n"
2239 	"mov $42, %eax\n"
2240 	"ret\n"
2241 	"ret42_end:\n"
2242 );
2243 
2244 static void
2245 diagnose_ept_violation_qual(u64 expected, u64 actual)
2246 {
2247 
2248 #define DIAGNOSE(flag)							\
2249 do {									\
2250 	if ((expected & flag) != (actual & flag))			\
2251 		printf(#flag " %sexpected\n",				\
2252 		       (expected & flag) ? "" : "un");			\
2253 } while (0)
2254 
2255 	DIAGNOSE(EPT_VLT_RD);
2256 	DIAGNOSE(EPT_VLT_WR);
2257 	DIAGNOSE(EPT_VLT_FETCH);
2258 	DIAGNOSE(EPT_VLT_PERM_RD);
2259 	DIAGNOSE(EPT_VLT_PERM_WR);
2260 	DIAGNOSE(EPT_VLT_PERM_EX);
2261 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2262 	DIAGNOSE(EPT_VLT_PADDR);
2263 
2264 #undef DIAGNOSE
2265 }
2266 
2267 static void do_ept_access_op(enum ept_access_op op)
2268 {
2269 	ept_access_test_data.op = op;
2270 	enter_guest();
2271 }
2272 
2273 /*
2274  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2275  * needed by tests that modify guest PTEs.
2276  */
2277 static void ept_access_test_guest_flush_tlb(void)
2278 {
2279 	do_ept_access_op(OP_FLUSH_TLB);
2280 	skip_exit_vmcall();
2281 }
2282 
2283 /*
2284  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2285  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2286  * a huge page.
2287  */
2288 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2289 				 unsigned long clear, unsigned long set)
2290 {
2291 	struct ept_access_test_data *data = &ept_access_test_data;
2292 	unsigned long orig_pte;
2293 	unsigned long pte;
2294 
2295 	/* Screw with the mapping at the requested level. */
2296 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2297 	pte = orig_pte;
2298 	if (mkhuge)
2299 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2300 	else
2301 		pte = orig_pte;
2302 	pte = (pte & ~clear) | set;
2303 	set_ept_pte(pml4, gpa, level, pte);
2304 	ept_sync(INVEPT_SINGLE, eptp);
2305 
2306 	return orig_pte;
2307 }
2308 
2309 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2310 {
2311 	set_ept_pte(pml4, gpa, level, orig_pte);
2312 }
2313 
2314 static void do_ept_violation(bool leaf, enum ept_access_op op,
2315 			     u64 expected_qual, u64 expected_paddr)
2316 {
2317 	u64 qual;
2318 
2319 	/* Try the access and observe the violation. */
2320 	do_ept_access_op(op);
2321 
2322 	assert_exit_reason(VMX_EPT_VIOLATION);
2323 
2324 	qual = vmcs_read(EXI_QUALIFICATION);
2325 
2326 	diagnose_ept_violation_qual(expected_qual, qual);
2327 	TEST_EXPECT_EQ(expected_qual, qual);
2328 
2329 	#if 0
2330 	/* Disable for now otherwise every test will fail */
2331 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2332 		       (unsigned long) (
2333 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2334 	#endif
2335 	/*
2336 	 * TODO: tests that probe expected_paddr in pages other than the one at
2337 	 * the beginning of the 1g region.
2338 	 */
2339 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2340 }
2341 
2342 static void
2343 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2344 			      unsigned long set, enum ept_access_op op,
2345 			      u64 expected_qual)
2346 {
2347 	struct ept_access_test_data *data = &ept_access_test_data;
2348 	unsigned long orig_pte;
2349 
2350 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2351 
2352 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2353 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2354 					 data->gpa);
2355 
2356 	/* Fix the violation and resume the op loop. */
2357 	ept_untwiddle(data->gpa, level, orig_pte);
2358 	enter_guest();
2359 	skip_exit_vmcall();
2360 }
2361 
2362 static void
2363 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2364 		       enum ept_access_op op, u64 expected_qual)
2365 {
2366 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2367 				      expected_qual);
2368 	if (ept_huge_pages_supported(level))
2369 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2370 					      expected_qual);
2371 }
2372 
2373 static void ept_violation(unsigned long clear, unsigned long set,
2374 			  enum ept_access_op op, u64 expected_qual)
2375 {
2376 	ept_violation_at_level(1, clear, set, op, expected_qual);
2377 	ept_violation_at_level(2, clear, set, op, expected_qual);
2378 	ept_violation_at_level(3, clear, set, op, expected_qual);
2379 	ept_violation_at_level(4, clear, set, op, expected_qual);
2380 }
2381 
2382 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2383 				       u64 expected_qual)
2384 {
2385 	ept_violation(EPT_PRESENT, access, op,
2386 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2387 }
2388 
2389 /*
2390  * For translations that don't involve a GVA, that is physical address (paddr)
2391  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2392  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2393  * translations don't involve GVAs, such as when the hardware does the guest
2394  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2395  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2396  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2397  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2398  *
2399  * Note that paddr violations can also be triggered by loading PAE page tables
2400  * with wonky addresses. We don't test that yet.
2401  *
2402  * This function modifies the EPT entry that maps the GPA that the guest page
2403  * table entry mapping ept_access_data.gva resides on.
2404  *
2405  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2406  *
2407  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2408  *
2409  *	@op		Guest operation to perform with ept_access_data.gva.
2410  *
2411  *	@expect_violation
2412  *			Is a violation expected during the paddr access?
2413  *
2414  *	@expected_qual	Expected qualification for the EPT violation.
2415  *			EPT_VLT_PADDR should be clear.
2416  */
2417 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2418 			     enum ept_access_op op, bool expect_violation,
2419 			     u64 expected_qual)
2420 {
2421 	struct ept_access_test_data *data = &ept_access_test_data;
2422 	unsigned long *ptep;
2423 	unsigned long gpa;
2424 	unsigned long orig_epte;
2425 
2426 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2427 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2428 	TEST_ASSERT(ptep);
2429 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2430 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2431 	ept_access_test_guest_flush_tlb();
2432 
2433 	/*
2434 	 * Now modify the access bits on the EPT entry for the GPA that the
2435 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2436 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2437 	 * constructed our test such that those other 511 PTEs aren't used by
2438 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2439 	 * PTE we're modifying is at the beginning of a 4K page and the
2440 	 * following 511 entires are also under our control (and not touched by
2441 	 * the guest).
2442 	 */
2443 	gpa = virt_to_phys(ptep);
2444 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2445 	/*
2446 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2447 	 * otherwise our level=1 twiddling below will fail. We use the
2448 	 * identity map (gpa = gpa) since page tables are shared with the host.
2449 	 */
2450 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2451 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2452 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2453 
2454 	if (expect_violation) {
2455 		do_ept_violation(/*leaf=*/true, op,
2456 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2457 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2458 		do_ept_access_op(op);
2459 	} else {
2460 		do_ept_access_op(op);
2461 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2462 	}
2463 
2464 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2465 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2466 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2467 
2468 	skip_exit_vmcall();
2469 }
2470 
2471 static void ept_access_allowed_paddr(unsigned long ept_access,
2472 				     unsigned long pte_ad,
2473 				     enum ept_access_op op)
2474 {
2475 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2476 			 /*expected_qual=*/-1);
2477 }
2478 
2479 static void ept_access_violation_paddr(unsigned long ept_access,
2480 				       unsigned long pte_ad,
2481 				       enum ept_access_op op,
2482 				       u64 expected_qual)
2483 {
2484 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2485 			 expected_qual);
2486 }
2487 
2488 
2489 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2490 					unsigned long clear,
2491 					unsigned long set,
2492 					enum ept_access_op op)
2493 {
2494 	struct ept_access_test_data *data = &ept_access_test_data;
2495 	unsigned long orig_pte;
2496 
2497 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2498 
2499 	/* No violation. Should proceed to vmcall. */
2500 	do_ept_access_op(op);
2501 	skip_exit_vmcall();
2502 
2503 	ept_untwiddle(data->gpa, level, orig_pte);
2504 }
2505 
2506 static void ept_allowed_at_level(int level, unsigned long clear,
2507 				 unsigned long set, enum ept_access_op op)
2508 {
2509 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2510 	if (ept_huge_pages_supported(level))
2511 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2512 }
2513 
2514 static void ept_allowed(unsigned long clear, unsigned long set,
2515 			enum ept_access_op op)
2516 {
2517 	ept_allowed_at_level(1, clear, set, op);
2518 	ept_allowed_at_level(2, clear, set, op);
2519 	ept_allowed_at_level(3, clear, set, op);
2520 	ept_allowed_at_level(4, clear, set, op);
2521 }
2522 
2523 static void ept_ignored_bit(int bit)
2524 {
2525 	/* Set the bit. */
2526 	ept_allowed(0, 1ul << bit, OP_READ);
2527 	ept_allowed(0, 1ul << bit, OP_WRITE);
2528 	ept_allowed(0, 1ul << bit, OP_EXEC);
2529 
2530 	/* Clear the bit. */
2531 	ept_allowed(1ul << bit, 0, OP_READ);
2532 	ept_allowed(1ul << bit, 0, OP_WRITE);
2533 	ept_allowed(1ul << bit, 0, OP_EXEC);
2534 }
2535 
2536 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2537 {
2538 	ept_allowed(EPT_PRESENT, access, op);
2539 }
2540 
2541 
2542 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2543 					     unsigned long clear,
2544 					     unsigned long set,
2545 					     enum ept_access_op op)
2546 {
2547 	struct ept_access_test_data *data = &ept_access_test_data;
2548 	unsigned long orig_pte;
2549 
2550 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2551 
2552 	do_ept_access_op(op);
2553 	assert_exit_reason(VMX_EPT_MISCONFIG);
2554 
2555 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2556 	#if 0
2557 	/* broken: */
2558 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2559 	#endif
2560 	#if 0
2561 	/*
2562 	 * broken:
2563 	 * According to description of exit qual for EPT violation,
2564 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2565 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2566 	 * to be set for msiconfig.
2567 	 */
2568 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2569 		       (unsigned long) (
2570 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2571 	#endif
2572 
2573 	/* Fix the violation and resume the op loop. */
2574 	ept_untwiddle(data->gpa, level, orig_pte);
2575 	enter_guest();
2576 	skip_exit_vmcall();
2577 }
2578 
2579 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2580 					  unsigned long clear,
2581 					  unsigned long set)
2582 {
2583 	/* The op shouldn't matter (read, write, exec), so try them all! */
2584 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2585 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2586 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2587 }
2588 
2589 static void ept_misconfig_at_level(int level, unsigned long clear,
2590 				   unsigned long set)
2591 {
2592 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2593 	if (ept_huge_pages_supported(level))
2594 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2595 }
2596 
2597 static void ept_misconfig(unsigned long clear, unsigned long set)
2598 {
2599 	ept_misconfig_at_level(1, clear, set);
2600 	ept_misconfig_at_level(2, clear, set);
2601 	ept_misconfig_at_level(3, clear, set);
2602 	ept_misconfig_at_level(4, clear, set);
2603 }
2604 
2605 static void ept_access_misconfig(unsigned long access)
2606 {
2607 	ept_misconfig(EPT_PRESENT, access);
2608 }
2609 
2610 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2611 {
2612 	/* Setting the bit causes a misconfig. */
2613 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2614 
2615 	/* Making the entry non-present turns reserved bits into ignored. */
2616 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2617 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2618 }
2619 
2620 static void ept_reserved_bit_at_level_huge(int level, int bit)
2621 {
2622 	/* Setting the bit causes a misconfig. */
2623 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2624 
2625 	/* Making the entry non-present turns reserved bits into ignored. */
2626 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2627 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2628 }
2629 
2630 static void ept_reserved_bit_at_level(int level, int bit)
2631 {
2632 	/* Setting the bit causes a misconfig. */
2633 	ept_misconfig_at_level(level, 0, 1ul << bit);
2634 
2635 	/* Making the entry non-present turns reserved bits into ignored. */
2636 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2637 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2638 }
2639 
2640 static void ept_reserved_bit(int bit)
2641 {
2642 	ept_reserved_bit_at_level(1, bit);
2643 	ept_reserved_bit_at_level(2, bit);
2644 	ept_reserved_bit_at_level(3, bit);
2645 	ept_reserved_bit_at_level(4, bit);
2646 }
2647 
2648 #define PAGE_2M_ORDER 9
2649 #define PAGE_1G_ORDER 18
2650 
2651 static void *get_1g_page(void)
2652 {
2653 	static void *alloc;
2654 
2655 	if (!alloc)
2656 		alloc = alloc_pages(PAGE_1G_ORDER);
2657 	return alloc;
2658 }
2659 
2660 static void ept_access_test_teardown(void *unused)
2661 {
2662 	/* Exit the guest cleanly. */
2663 	do_ept_access_op(OP_EXIT);
2664 }
2665 
2666 static void ept_access_test_guest(void)
2667 {
2668 	struct ept_access_test_data *data = &ept_access_test_data;
2669 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2670 
2671 	while (true) {
2672 		switch (data->op) {
2673 		case OP_READ:
2674 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2675 			break;
2676 		case OP_WRITE:
2677 			*data->gva = MAGIC_VAL_2;
2678 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2679 			*data->gva = MAGIC_VAL_1;
2680 			break;
2681 		case OP_EXEC:
2682 			TEST_ASSERT_EQ(42, code());
2683 			break;
2684 		case OP_FLUSH_TLB:
2685 			write_cr3(read_cr3());
2686 			break;
2687 		case OP_EXIT:
2688 			return;
2689 		default:
2690 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2691 		}
2692 		vmcall();
2693 	}
2694 }
2695 
2696 static void ept_access_test_setup(void)
2697 {
2698 	struct ept_access_test_data *data = &ept_access_test_data;
2699 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2700 	unsigned long size = npages * PAGE_SIZE;
2701 	unsigned long *page_table = current_page_table();
2702 	unsigned long pte;
2703 
2704 	if (setup_ept(false))
2705 		test_skip("EPT not supported");
2706 
2707 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2708 	if (cpuid_maxphyaddr() < 40)
2709 		test_skip("Test needs MAXPHYADDR >= 40");
2710 
2711 	test_set_guest(ept_access_test_guest);
2712 	test_add_teardown(ept_access_test_teardown, NULL);
2713 
2714 	data->hva = get_1g_page();
2715 	TEST_ASSERT(data->hva);
2716 	data->hpa = virt_to_phys(data->hva);
2717 
2718 	data->gpa = 1ul << 39;
2719 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2720 				   size);
2721 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2722 	install_pages(page_table, data->gpa, size, data->gva);
2723 
2724 	/*
2725 	 * Make sure nothing's mapped here so the tests that screw with the
2726 	 * pml4 entry don't inadvertently break something.
2727 	 */
2728 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2729 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2730 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2731 
2732 	data->hva[0] = MAGIC_VAL_1;
2733 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2734 }
2735 
2736 static void ept_access_test_not_present(void)
2737 {
2738 	ept_access_test_setup();
2739 	/* --- */
2740 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2741 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2742 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2743 }
2744 
2745 static void ept_access_test_read_only(void)
2746 {
2747 	ept_access_test_setup();
2748 
2749 	/* r-- */
2750 	ept_access_allowed(EPT_RA, OP_READ);
2751 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2752 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2753 }
2754 
2755 static void ept_access_test_write_only(void)
2756 {
2757 	ept_access_test_setup();
2758 	/* -w- */
2759 	ept_access_misconfig(EPT_WA);
2760 }
2761 
2762 static void ept_access_test_read_write(void)
2763 {
2764 	ept_access_test_setup();
2765 	/* rw- */
2766 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2767 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2768 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2769 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2770 }
2771 
2772 
2773 static void ept_access_test_execute_only(void)
2774 {
2775 	ept_access_test_setup();
2776 	/* --x */
2777 	if (ept_execute_only_supported()) {
2778 		ept_access_violation(EPT_EA, OP_READ,
2779 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2780 		ept_access_violation(EPT_EA, OP_WRITE,
2781 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2782 		ept_access_allowed(EPT_EA, OP_EXEC);
2783 	} else {
2784 		ept_access_misconfig(EPT_EA);
2785 	}
2786 }
2787 
2788 static void ept_access_test_read_execute(void)
2789 {
2790 	ept_access_test_setup();
2791 	/* r-x */
2792 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2793 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2794 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2795 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2796 }
2797 
2798 static void ept_access_test_write_execute(void)
2799 {
2800 	ept_access_test_setup();
2801 	/* -wx */
2802 	ept_access_misconfig(EPT_WA | EPT_EA);
2803 }
2804 
2805 static void ept_access_test_read_write_execute(void)
2806 {
2807 	ept_access_test_setup();
2808 	/* rwx */
2809 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2810 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2811 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2812 }
2813 
2814 static void ept_access_test_reserved_bits(void)
2815 {
2816 	int i;
2817 	int maxphyaddr;
2818 
2819 	ept_access_test_setup();
2820 
2821 	/* Reserved bits above maxphyaddr. */
2822 	maxphyaddr = cpuid_maxphyaddr();
2823 	for (i = maxphyaddr; i <= 51; i++) {
2824 		report_prefix_pushf("reserved_bit=%d", i);
2825 		ept_reserved_bit(i);
2826 		report_prefix_pop();
2827 	}
2828 
2829 	/* Level-specific reserved bits. */
2830 	ept_reserved_bit_at_level_nohuge(2, 3);
2831 	ept_reserved_bit_at_level_nohuge(2, 4);
2832 	ept_reserved_bit_at_level_nohuge(2, 5);
2833 	ept_reserved_bit_at_level_nohuge(2, 6);
2834 	/* 2M alignment. */
2835 	for (i = 12; i < 20; i++) {
2836 		report_prefix_pushf("reserved_bit=%d", i);
2837 		ept_reserved_bit_at_level_huge(2, i);
2838 		report_prefix_pop();
2839 	}
2840 	ept_reserved_bit_at_level_nohuge(3, 3);
2841 	ept_reserved_bit_at_level_nohuge(3, 4);
2842 	ept_reserved_bit_at_level_nohuge(3, 5);
2843 	ept_reserved_bit_at_level_nohuge(3, 6);
2844 	/* 1G alignment. */
2845 	for (i = 12; i < 29; i++) {
2846 		report_prefix_pushf("reserved_bit=%d", i);
2847 		ept_reserved_bit_at_level_huge(3, i);
2848 		report_prefix_pop();
2849 	}
2850 	ept_reserved_bit_at_level(4, 3);
2851 	ept_reserved_bit_at_level(4, 4);
2852 	ept_reserved_bit_at_level(4, 5);
2853 	ept_reserved_bit_at_level(4, 6);
2854 	ept_reserved_bit_at_level(4, 7);
2855 }
2856 
2857 static void ept_access_test_ignored_bits(void)
2858 {
2859 	ept_access_test_setup();
2860 	/*
2861 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2862 	 * far as translation is concerned even if AD bits are enabled in the
2863 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2864 	 * control is 0.
2865 	 */
2866 	ept_ignored_bit(8);
2867 	ept_ignored_bit(9);
2868 	ept_ignored_bit(10);
2869 	ept_ignored_bit(11);
2870 	ept_ignored_bit(52);
2871 	ept_ignored_bit(53);
2872 	ept_ignored_bit(54);
2873 	ept_ignored_bit(55);
2874 	ept_ignored_bit(56);
2875 	ept_ignored_bit(57);
2876 	ept_ignored_bit(58);
2877 	ept_ignored_bit(59);
2878 	ept_ignored_bit(60);
2879 	ept_ignored_bit(61);
2880 	ept_ignored_bit(62);
2881 	ept_ignored_bit(63);
2882 }
2883 
2884 static void ept_access_test_paddr_not_present_ad_disabled(void)
2885 {
2886 	ept_access_test_setup();
2887 	ept_disable_ad_bits();
2888 
2889 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
2890 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
2891 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
2892 }
2893 
2894 static void ept_access_test_paddr_not_present_ad_enabled(void)
2895 {
2896 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
2897 
2898 	ept_access_test_setup();
2899 	ept_enable_ad_bits_or_skip_test();
2900 
2901 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
2902 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
2903 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
2904 }
2905 
2906 static void ept_access_test_paddr_read_only_ad_disabled(void)
2907 {
2908 	/*
2909 	 * When EPT AD bits are disabled, all accesses to guest paging
2910 	 * structures are reported separately as a read and (after
2911 	 * translation of the GPA to host physical address) a read+write
2912 	 * if the A/D bits have to be set.
2913 	 */
2914 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2915 
2916 	ept_access_test_setup();
2917 	ept_disable_ad_bits();
2918 
2919 	/* Can't update A bit, so all accesses fail. */
2920 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2921 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
2922 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
2923 	/* AD bits disabled, so only writes try to update the D bit. */
2924 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
2925 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
2926 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
2927 	/* Both A and D already set, so read-only is OK. */
2928 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
2929 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
2930 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
2931 }
2932 
2933 static void ept_access_test_paddr_read_only_ad_enabled(void)
2934 {
2935 	/*
2936 	 * When EPT AD bits are enabled, all accesses to guest paging
2937 	 * structures are considered writes as far as EPT translation
2938 	 * is concerned.
2939 	 */
2940 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2941 
2942 	ept_access_test_setup();
2943 	ept_enable_ad_bits_or_skip_test();
2944 
2945 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2946 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
2947 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
2948 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
2949 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
2950 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
2951 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
2952 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
2953 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
2954 }
2955 
2956 static void ept_access_test_paddr_read_write(void)
2957 {
2958 	ept_access_test_setup();
2959 	/* Read-write access to paging structure. */
2960 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
2961 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
2962 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
2963 }
2964 
2965 static void ept_access_test_paddr_read_write_execute(void)
2966 {
2967 	ept_access_test_setup();
2968 	/* RWX access to paging structure. */
2969 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
2970 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
2971 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
2972 }
2973 
2974 static void ept_access_test_paddr_read_execute_ad_disabled(void)
2975 {
2976   	/*
2977 	 * When EPT AD bits are disabled, all accesses to guest paging
2978 	 * structures are reported separately as a read and (after
2979 	 * translation of the GPA to host physical address) a read+write
2980 	 * if the A/D bits have to be set.
2981 	 */
2982 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
2983 
2984 	ept_access_test_setup();
2985 	ept_disable_ad_bits();
2986 
2987 	/* Can't update A bit, so all accesses fail. */
2988 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
2989 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
2990 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
2991 	/* AD bits disabled, so only writes try to update the D bit. */
2992 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
2993 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
2994 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
2995 	/* Both A and D already set, so read-only is OK. */
2996 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
2997 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
2998 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
2999 }
3000 
3001 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3002 {
3003 	/*
3004 	 * When EPT AD bits are enabled, all accesses to guest paging
3005 	 * structures are considered writes as far as EPT translation
3006 	 * is concerned.
3007 	 */
3008 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3009 
3010 	ept_access_test_setup();
3011 	ept_enable_ad_bits_or_skip_test();
3012 
3013 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3014 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3015 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3016 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3017 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3018 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3019 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3020 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3021 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3022 }
3023 
3024 static void ept_access_test_paddr_not_present_page_fault(void)
3025 {
3026 	ept_access_test_setup();
3027 	/*
3028 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3029 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3030 	 * Thus guest page fault before host takes EPT violation for trying to
3031 	 * update A bit.
3032 	 */
3033 }
3034 
3035 static void ept_access_test_force_2m_page(void)
3036 {
3037 	ept_access_test_setup();
3038 
3039 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3040 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3041 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3042 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3043 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3044 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3045 }
3046 
3047 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3048 {
3049 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3050 
3051 	TEST_ASSERT(msr & VPID_CAP_INVVPID);
3052 
3053 	if (type < INVVPID_ADDR || type > INVVPID_CONTEXT_LOCAL)
3054 		return false;
3055 
3056 	if (!(msr & (1ull << (type + VPID_CAP_INVVPID_TYPES_SHIFT))))
3057 		return false;
3058 
3059 	if (vpid >> 16)
3060 		return false;
3061 
3062 	if (type != INVVPID_ALL && !vpid)
3063 		return false;
3064 
3065 	if (type == INVVPID_ADDR && !is_canonical(gla))
3066 		return false;
3067 
3068 	return true;
3069 }
3070 
3071 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3072 {
3073 	int rc;
3074 	bool valid = invvpid_valid(type, vpid, gla);
3075 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3076 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3077 	/*
3078 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3079 	 * that we can tell if it is updated by INVVPID.
3080 	 */
3081 	vmcs_read(~0);
3082 	rc = invvpid(type, vpid, gla);
3083 	report("INVVPID type %ld VPID %lx GLA %lx %s",
3084 	       !rc == valid, type, vpid, gla,
3085 	       valid ? "passes" : "fails");
3086 	report("After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3087 	       vmcs_read(VMX_INST_ERROR) == expected,
3088 	       rc ? "failed" : "successful",
3089 	       expected, vmcs_read(VMX_INST_ERROR));
3090 }
3091 
3092 static void ds_invvpid(void *data)
3093 {
3094 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3095 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3096 
3097 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3098 	asm volatile("invvpid %0, %1"
3099 		     :
3100 		     : "m"(*(struct invvpid_operand *)data),
3101 		       "r"(type));
3102 }
3103 
3104 /*
3105  * The SS override is ignored in 64-bit mode, so we use an addressing
3106  * mode with %rsp as the base register to generate an implicit SS
3107  * reference.
3108  */
3109 static void ss_invvpid(void *data)
3110 {
3111 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3112 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3113 
3114 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3115 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3116 		     : "+r"(data)
3117 		     : "r"(type));
3118 }
3119 
3120 static void invvpid_test_gp(void)
3121 {
3122 	bool fault;
3123 
3124 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3125 				   (void *)NONCANONICAL);
3126 	report("INVVPID with non-canonical DS operand raises #GP", fault);
3127 }
3128 
3129 static void invvpid_test_ss(void)
3130 {
3131 	bool fault;
3132 
3133 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3134 				   (void *)NONCANONICAL);
3135 	report("INVVPID with non-canonical SS operand raises #SS", fault);
3136 }
3137 
3138 static void invvpid_test_pf(void)
3139 {
3140 	void *vpage = alloc_vpage();
3141 	bool fault;
3142 
3143 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3144 	report("INVVPID with unmapped operand raises #PF", fault);
3145 }
3146 
3147 static void try_compat_invvpid(void *unused)
3148 {
3149 	struct far_pointer32 fp = {
3150 		.offset = (uintptr_t)&&invvpid,
3151 		.selector = KERNEL_CS32,
3152 	};
3153 	register uintptr_t rsp asm("rsp");
3154 
3155 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3156 			"Code address too high.");
3157 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3158 
3159 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3160 	return;
3161 invvpid:
3162 	asm volatile (".code32;"
3163 		      "invvpid (%eax), %eax;"
3164 		      "lret;"
3165 		      ".code64");
3166 	__builtin_unreachable();
3167 }
3168 
3169 static void invvpid_test_compatibility_mode(void)
3170 {
3171 	bool fault;
3172 
3173 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3174 	report("Compatibility mode INVVPID raises #UD", fault);
3175 }
3176 
3177 static void invvpid_test_not_in_vmx_operation(void)
3178 {
3179 	bool fault;
3180 
3181 	TEST_ASSERT(!vmx_off());
3182 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3183 	report("INVVPID outside of VMX operation raises #UD", fault);
3184 	TEST_ASSERT(!vmx_on());
3185 }
3186 
3187 /*
3188  * This does not test real-address mode, virtual-8086 mode, protected mode,
3189  * or CPL > 0.
3190  */
3191 static void invvpid_test_v2(void)
3192 {
3193 	u64 msr;
3194 	int i;
3195 	unsigned types = 0;
3196 	unsigned type;
3197 
3198 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
3199 	    !(ctrl_cpu_rev[1].clr & CPU_VPID))
3200 		test_skip("VPID not supported");
3201 
3202 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3203 
3204 	if (!(msr & VPID_CAP_INVVPID))
3205 		test_skip("INVVPID not supported.\n");
3206 
3207 	if (msr & VPID_CAP_INVVPID_ADDR)
3208 		types |= 1u << INVVPID_ADDR;
3209 	if (msr & VPID_CAP_INVVPID_CXTGLB)
3210 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3211 	if (msr & VPID_CAP_INVVPID_ALL)
3212 		types |= 1u << INVVPID_ALL;
3213 	if (msr & VPID_CAP_INVVPID_CXTLOC)
3214 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3215 
3216 	if (!types)
3217 		test_skip("No INVVPID types supported.\n");
3218 
3219 	for (i = -127; i < 128; i++)
3220 		try_invvpid(i, 0xffff, 0);
3221 
3222 	/*
3223 	 * VPID must not be more than 16 bits.
3224 	 */
3225 	for (i = 0; i < 64; i++)
3226 		for (type = 0; type < 4; type++)
3227 			if (types & (1u << type))
3228 				try_invvpid(type, 1ul << i, 0);
3229 
3230 	/*
3231 	 * VPID must not be zero, except for "all contexts."
3232 	 */
3233 	for (type = 0; type < 4; type++)
3234 		if (types & (1u << type))
3235 			try_invvpid(type, 0, 0);
3236 
3237 	/*
3238 	 * The gla operand is only validated for single-address INVVPID.
3239 	 */
3240 	if (types & (1u << INVVPID_ADDR))
3241 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3242 
3243 	invvpid_test_gp();
3244 	invvpid_test_ss();
3245 	invvpid_test_pf();
3246 	invvpid_test_compatibility_mode();
3247 	invvpid_test_not_in_vmx_operation();
3248 }
3249 
3250 /*
3251  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3252  * at least as far as the guest-state checks. Returns false if the
3253  * VMLAUNCH fails early and execution falls through to the next
3254  * instruction.
3255  */
3256 static bool vmlaunch_succeeds(void)
3257 {
3258 	u32 exit_reason;
3259 
3260 	/*
3261 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3262 	 * unsupported VMCS component"). The caller can then check
3263 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3264 	 */
3265 	vmcs_write(~0u, 0);
3266 
3267 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3268 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3269 				   :
3270 				   : "r" ((u64)HOST_RSP)
3271 				   : "cc", "memory"
3272 				   : success);
3273 	return false;
3274 success:
3275 	exit_reason = vmcs_read(EXI_REASON);
3276 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3277 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3278 	return true;
3279 }
3280 
3281 /*
3282  * Try to launch the current VMCS.
3283  */
3284 static void test_vmx_controls(bool controls_valid, bool xfail)
3285 {
3286 	bool success = vmlaunch_succeeds();
3287 	u32 vmx_inst_err;
3288 
3289 	report_xfail("vmlaunch %s", xfail, success == controls_valid,
3290 		     controls_valid ? "succeeds" : "fails");
3291 	if (!success) {
3292 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3293 		report("VMX inst error is %d (actual %d)",
3294 		       vmx_inst_err == VMXERR_ENTRY_INVALID_CONTROL_FIELD,
3295 		       VMXERR_ENTRY_INVALID_CONTROL_FIELD, vmx_inst_err);
3296 	}
3297 }
3298 
3299 /*
3300  * Test a particular value of a VM-execution control bit, if the value
3301  * is required or if the value is zero.
3302  */
3303 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3304 				    enum Encoding encoding, unsigned bit,
3305 				    unsigned val)
3306 {
3307 	u32 mask = 1u << bit;
3308 	bool expected;
3309 	u32 controls;
3310 
3311 	if (msr.set & mask)
3312 		TEST_ASSERT(msr.clr & mask);
3313 
3314 	/*
3315 	 * We can't arbitrarily turn on a control bit, because it may
3316 	 * introduce dependencies on other VMCS fields. So, we only
3317 	 * test turning on bits that have a required setting.
3318 	 */
3319 	if (val && (msr.clr & mask) && !(msr.set & mask))
3320 		return;
3321 
3322 	report_prefix_pushf("%s %s bit %d",
3323 			    val ? "Set" : "Clear", name, bit);
3324 
3325 	controls = vmcs_read(encoding);
3326 	if (val) {
3327 		vmcs_write(encoding, msr.set | mask);
3328 		expected = (msr.clr & mask);
3329 	} else {
3330 		vmcs_write(encoding, msr.set & ~mask);
3331 		expected = !(msr.set & mask);
3332 	}
3333 	test_vmx_controls(expected, false);
3334 	vmcs_write(encoding, controls);
3335 	report_prefix_pop();
3336 }
3337 
3338 /*
3339  * Test reserved values of a VM-execution control bit, based on the
3340  * allowed bit settings from the corresponding VMX capability MSR.
3341  */
3342 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3343 			      enum Encoding encoding, unsigned bit)
3344 {
3345 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3346 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3347 }
3348 
3349 /*
3350  * Reserved bits in the pin-based VM-execution controls must be set
3351  * properly. Software may consult the VMX capability MSRs to determine
3352  * the proper settings.
3353  * [Intel SDM]
3354  */
3355 static void test_pin_based_ctls(void)
3356 {
3357 	unsigned bit;
3358 
3359 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3360 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3361 	for (bit = 0; bit < 32; bit++)
3362 		test_rsvd_ctl_bit("pin-based controls",
3363 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3364 }
3365 
3366 /*
3367  * Reserved bits in the primary processor-based VM-execution controls
3368  * must be set properly. Software may consult the VMX capability MSRs
3369  * to determine the proper settings.
3370  * [Intel SDM]
3371  */
3372 static void test_primary_processor_based_ctls(void)
3373 {
3374 	unsigned bit;
3375 
3376 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3377 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3378 	for (bit = 0; bit < 32; bit++)
3379 		test_rsvd_ctl_bit("primary processor-based controls",
3380 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3381 }
3382 
3383 /*
3384  * If the "activate secondary controls" primary processor-based
3385  * VM-execution control is 1, reserved bits in the secondary
3386  * processor-based VM-execution controls must be cleared. Software may
3387  * consult the VMX capability MSRs to determine which bits are
3388  * reserved.
3389  * If the "activate secondary controls" primary processor-based
3390  * VM-execution control is 0 (or if the processor does not support the
3391  * 1-setting of that control), no checks are performed on the
3392  * secondary processor-based VM-execution controls.
3393  * [Intel SDM]
3394  */
3395 static void test_secondary_processor_based_ctls(void)
3396 {
3397 	u32 primary;
3398 	u32 secondary;
3399 	unsigned bit;
3400 
3401 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3402 		return;
3403 
3404 	primary = vmcs_read(CPU_EXEC_CTRL0);
3405 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3406 
3407 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3408 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3409 	for (bit = 0; bit < 32; bit++)
3410 		test_rsvd_ctl_bit("secondary processor-based controls",
3411 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3412 
3413 	/*
3414 	 * When the "activate secondary controls" VM-execution control
3415 	 * is clear, there are no checks on the secondary controls.
3416 	 */
3417 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3418 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3419 	report("Secondary processor-based controls ignored",
3420 	       vmlaunch_succeeds());
3421 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3422 	vmcs_write(CPU_EXEC_CTRL0, primary);
3423 }
3424 
3425 static void try_cr3_target_count(unsigned i, unsigned max)
3426 {
3427 	report_prefix_pushf("CR3 target count 0x%x", i);
3428 	vmcs_write(CR3_TARGET_COUNT, i);
3429 	test_vmx_controls(i <= max, false);
3430 	report_prefix_pop();
3431 }
3432 
3433 /*
3434  * The CR3-target count must not be greater than 4. Future processors
3435  * may support a different number of CR3-target values. Software
3436  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3437  * number of values supported.
3438  * [Intel SDM]
3439  */
3440 static void test_cr3_targets(void)
3441 {
3442 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3443 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3444 	unsigned i;
3445 
3446 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3447 	TEST_ASSERT(supported_targets <= 256);
3448 
3449 	try_cr3_target_count(-1u, supported_targets);
3450 	try_cr3_target_count(0x80000000, supported_targets);
3451 	try_cr3_target_count(0x7fffffff, supported_targets);
3452 	for (i = 0; i <= supported_targets + 1; i++)
3453 		try_cr3_target_count(i, supported_targets);
3454 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3455 }
3456 
3457 /*
3458  * Test a particular address setting in the VMCS
3459  */
3460 static void test_vmcs_addr(const char *name,
3461 			   enum Encoding encoding,
3462 			   u64 align,
3463 			   bool ignored,
3464 			   bool xfail_beyond_mapped_ram,
3465 			   u64 addr)
3466 {
3467 	bool xfail =
3468 		(xfail_beyond_mapped_ram &&
3469 		 addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3470 		 addr < (1ul << cpuid_maxphyaddr()));
3471 
3472 	report_prefix_pushf("%s = %lx", name, addr);
3473 	vmcs_write(encoding, addr);
3474 	test_vmx_controls(ignored || (IS_ALIGNED(addr, align) &&
3475 				  addr < (1ul << cpuid_maxphyaddr())),
3476 			  xfail);
3477 	report_prefix_pop();
3478 	xfail = false;
3479 }
3480 
3481 /*
3482  * Test interesting values for a VMCS address
3483  */
3484 static void test_vmcs_addr_values(const char *name,
3485 				  enum Encoding encoding,
3486 				  u64 align,
3487 				  bool ignored,
3488 				  bool xfail_beyond_mapped_ram,
3489 				  u32 bit_start, u32 bit_end)
3490 {
3491 	unsigned i;
3492 	u64 orig_val = vmcs_read(encoding);
3493 
3494 	for (i = bit_start; i <= bit_end; i++)
3495 		test_vmcs_addr(name, encoding, align, ignored,
3496 			       xfail_beyond_mapped_ram, 1ul << i);
3497 
3498 	test_vmcs_addr(name, encoding, align, ignored,
3499 		       xfail_beyond_mapped_ram, PAGE_SIZE - 1);
3500 	test_vmcs_addr(name, encoding, align, ignored,
3501 		       xfail_beyond_mapped_ram, PAGE_SIZE);
3502 	test_vmcs_addr(name, encoding, align, ignored,
3503 		       xfail_beyond_mapped_ram,
3504 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3505 	test_vmcs_addr(name, encoding, align, ignored,
3506 		       xfail_beyond_mapped_ram, -1ul);
3507 
3508 	vmcs_write(encoding, orig_val);
3509 }
3510 
3511 /*
3512  * Test a physical address reference in the VMCS, when the corresponding
3513  * feature is enabled and when the corresponding feature is disabled.
3514  */
3515 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3516 				     const char *field_name,
3517 				     const char *control_name, u64 align,
3518 				     bool xfail_beyond_mapped_ram,
3519 				     bool control_primary)
3520 {
3521 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3522 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3523 	u64 page_addr;
3524 
3525 	if (control_primary) {
3526 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3527 			return;
3528 	} else {
3529 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3530 			return;
3531 	}
3532 
3533 	page_addr = vmcs_read(field);
3534 
3535 	report_prefix_pushf("%s enabled", control_name);
3536 	if (control_primary) {
3537 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3538 	} else {
3539 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3540 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3541 	}
3542 
3543 	test_vmcs_addr_values(field_name, field, align, false,
3544 			      xfail_beyond_mapped_ram, 0, 63);
3545 	report_prefix_pop();
3546 
3547 	report_prefix_pushf("%s disabled", control_name);
3548 	if (control_primary) {
3549 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3550 	} else {
3551 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3552 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3553 	}
3554 
3555 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3556 	report_prefix_pop();
3557 
3558 	vmcs_write(field, page_addr);
3559 	vmcs_write(CPU_EXEC_CTRL0, primary);
3560 }
3561 
3562 /*
3563  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3564  * each I/O-bitmap address must be 0. Neither address should set any
3565  * bits beyond the processor's physical-address width.
3566  * [Intel SDM]
3567  */
3568 static void test_io_bitmaps(void)
3569 {
3570 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3571 				 "I/O bitmap A", "Use I/O bitmaps",
3572 				 PAGE_SIZE, false, true);
3573 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3574 				 "I/O bitmap B", "Use I/O bitmaps",
3575 				 PAGE_SIZE, false, true);
3576 }
3577 
3578 /*
3579  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3580  * the MSR-bitmap address must be 0. The address should not set any
3581  * bits beyond the processor's physical-address width.
3582  * [Intel SDM]
3583  */
3584 static void test_msr_bitmap(void)
3585 {
3586 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3587 				 "MSR bitmap", "Use MSR bitmaps",
3588 				 PAGE_SIZE, false, true);
3589 }
3590 
3591 /*
3592  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3593  * address must satisfy the following checks:
3594  * - Bits 11:0 of the address must be 0.
3595  * - The address should not set any bits beyond the processor's
3596  *   physical-address width.
3597  * [Intel SDM]
3598  */
3599 static void test_apic_virt_addr(void)
3600 {
3601 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3602 				 "virtual-APIC address", "Use TPR shadow",
3603 				 PAGE_SIZE, true, true);
3604 }
3605 
3606 /*
3607  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3608  * APIC-access address must satisfy the following checks:
3609  *  - Bits 11:0 of the address must be 0.
3610  *  - The address should not set any bits beyond the processor's
3611  *    physical-address width.
3612  * [Intel SDM]
3613  */
3614 static void test_apic_access_addr(void)
3615 {
3616 	void *apic_access_page = alloc_page();
3617 
3618 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3619 
3620 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3621 				 "APIC-access address",
3622 				 "virtualize APIC-accesses", PAGE_SIZE,
3623 				 false, false);
3624 }
3625 
3626 static bool set_bit_pattern(u8 mask, u32 *secondary)
3627 {
3628 	u8 i;
3629 	bool flag = false;
3630 	u32 test_bits[3] = {
3631 		CPU_VIRT_X2APIC,
3632 		CPU_APIC_REG_VIRT,
3633 		CPU_VINTD
3634 	};
3635 
3636         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3637 		if ((mask & (1u << i)) &&
3638 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3639 			*secondary |= test_bits[i];
3640 			flag = true;
3641 		}
3642 	}
3643 
3644 	return (flag);
3645 }
3646 
3647 /*
3648  * If the "use TPR shadow" VM-execution control is 0, the following
3649  * VM-execution controls must also be 0:
3650  * 	- virtualize x2APIC mode
3651  *	- APIC-register virtualization
3652  *	- virtual-interrupt delivery
3653  *    [Intel SDM]
3654  *
3655  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3656  *    "virtualize APIC accesses" VM-execution control must be 0.
3657  *    [Intel SDM]
3658  */
3659 static void test_apic_virtual_ctls(void)
3660 {
3661 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3662 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3663 	u32 primary = saved_primary;
3664 	u32 secondary = saved_secondary;
3665 	bool ctrl = false;
3666 	char str[10] = "disabled";
3667 	u8 i = 0, j;
3668 
3669 	/*
3670 	 * First test
3671 	 */
3672 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3673 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3674 		return;
3675 
3676 	primary |= CPU_SECONDARY;
3677 	primary &= ~CPU_TPR_SHADOW;
3678 	vmcs_write(CPU_EXEC_CTRL0, primary);
3679 
3680 	while (1) {
3681 		for (j = 1; j < 8; j++) {
3682 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3683 			if (primary & CPU_TPR_SHADOW) {
3684 				ctrl = true;
3685 			} else {
3686 				if (! set_bit_pattern(j, &secondary))
3687 					ctrl = true;
3688 				else
3689 					ctrl = false;
3690 			}
3691 
3692 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3693 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3694 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3695 			test_vmx_controls(ctrl, false);
3696 			report_prefix_pop();
3697 		}
3698 
3699 		if (i == 1)
3700 			break;
3701 		i++;
3702 
3703 		primary |= CPU_TPR_SHADOW;
3704 		vmcs_write(CPU_EXEC_CTRL0, primary);
3705 		strcpy(str, "enabled");
3706 	}
3707 
3708 	/*
3709 	 * Second test
3710 	 */
3711 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3712 
3713 	primary = saved_primary;
3714 	secondary = saved_secondary;
3715 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3716 		return;
3717 
3718 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3719 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3720 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3721 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3722 	test_vmx_controls(true, false);
3723 	report_prefix_pop();
3724 
3725 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3726 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3727 	test_vmx_controls(true, false);
3728 	report_prefix_pop();
3729 
3730 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3731 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3732 	test_vmx_controls(false, false);
3733 	report_prefix_pop();
3734 
3735 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3736 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3737 	test_vmx_controls(true, false);
3738 	report_prefix_pop();
3739 
3740 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3741 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3742 }
3743 
3744 /*
3745  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3746  * "external-interrupt exiting" VM-execution control must be 1.
3747  * [Intel SDM]
3748  */
3749 static void test_virtual_intr_ctls(void)
3750 {
3751 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3752 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3753 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3754 	u32 primary = saved_primary;
3755 	u32 secondary = saved_secondary;
3756 	u32 pin = saved_pin;
3757 
3758 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3759 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
3760 		return;
3761 
3762 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3763 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
3764 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3765 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
3766 	test_vmx_controls(true, false);
3767 	report_prefix_pop();
3768 
3769 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
3770 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3771 	test_vmx_controls(false, false);
3772 	report_prefix_pop();
3773 
3774 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
3775 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
3776 	test_vmx_controls(true, false);
3777 	report_prefix_pop();
3778 
3779 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3780 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3781 	test_vmx_controls(false, false);
3782 	report_prefix_pop();
3783 
3784 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3785 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3786 	vmcs_write(PIN_CONTROLS, saved_pin);
3787 }
3788 
3789 static void test_pi_desc_addr(u64 addr, bool ctrl)
3790 {
3791 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
3792 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
3793 	test_vmx_controls(ctrl, false);
3794 	report_prefix_pop();
3795 }
3796 
3797 /*
3798  * If the “process posted interrupts†VM-execution control is 1, the
3799  * following must be true:
3800  *
3801  *	- The “virtual-interrupt delivery†VM-execution control is 1.
3802  *	- The “acknowledge interrupt on exit†VM-exit control is 1.
3803  *	- The posted-interrupt notification vector has a value in the
3804  *	- range 0–255 (bits 15:8 are all 0).
3805  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
3806  *	- The posted-interrupt descriptor address does not set any bits
3807  *	  beyond the processor's physical-address width.
3808  * [Intel SDM]
3809  */
3810 static void test_posted_intr(void)
3811 {
3812 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3813 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3814 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3815 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
3816 	u32 primary = saved_primary;
3817 	u32 secondary = saved_secondary;
3818 	u32 pin = saved_pin;
3819 	u32 exit_ctl = exit_ctl_saved;
3820 	u16 vec;
3821 	int i;
3822 
3823 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
3824 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3825 	    (ctrl_exit_rev.clr & EXI_INTA)))
3826 		return;
3827 
3828 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3829 
3830 	/*
3831 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
3832 	 */
3833 	pin |= PIN_POST_INTR;
3834 	vmcs_write(PIN_CONTROLS, pin);
3835 	secondary &= ~CPU_VINTD;
3836 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3837 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
3838 	test_vmx_controls(false, false);
3839 	report_prefix_pop();
3840 
3841 	secondary |= CPU_VINTD;
3842 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3843 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
3844 	test_vmx_controls(false, false);
3845 	report_prefix_pop();
3846 
3847 	exit_ctl &= ~EXI_INTA;
3848 	vmcs_write(EXI_CONTROLS, exit_ctl);
3849 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
3850 	test_vmx_controls(false, false);
3851 	report_prefix_pop();
3852 
3853 	exit_ctl |= EXI_INTA;
3854 	vmcs_write(EXI_CONTROLS, exit_ctl);
3855 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3856 	test_vmx_controls(true, false);
3857 	report_prefix_pop();
3858 
3859 	secondary &= ~CPU_VINTD;
3860 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3861 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
3862 	test_vmx_controls(false, false);
3863 	report_prefix_pop();
3864 
3865 	secondary |= CPU_VINTD;
3866 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3867 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3868 	test_vmx_controls(true, false);
3869 	report_prefix_pop();
3870 
3871 	/*
3872 	 * Test posted-interrupt notification vector
3873 	 */
3874 	for (i = 0; i < 8; i++) {
3875 		vec = (1ul << i);
3876 		vmcs_write(PINV, vec);
3877 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3878 		test_vmx_controls(true, false);
3879 		report_prefix_pop();
3880 	}
3881 	for (i = 8; i < 16; i++) {
3882 		vec = (1ul << i);
3883 		vmcs_write(PINV, vec);
3884 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3885 		test_vmx_controls(false, false);
3886 		report_prefix_pop();
3887 	}
3888 
3889 	vec &= ~(0xff << 8);
3890 	vmcs_write(PINV, vec);
3891 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3892 	test_vmx_controls(true, false);
3893 	report_prefix_pop();
3894 
3895 	/*
3896 	 * Test posted-interrupt descriptor addresss
3897 	 */
3898 	for (i = 0; i < 6; i++) {
3899 		test_pi_desc_addr(1ul << i, false);
3900 	}
3901 
3902 	test_pi_desc_addr(0xf0, false);
3903 	test_pi_desc_addr(0xff, false);
3904 	test_pi_desc_addr(0x0f, false);
3905 	test_pi_desc_addr(0x8000, true);
3906 	test_pi_desc_addr(0x00, true);
3907 	test_pi_desc_addr(0xc000, true);
3908 
3909 	test_vmcs_addr_values("process-posted interrupts",
3910 			       POSTED_INTR_DESC_ADDR, 64,
3911 			       false, false, 0, 63);
3912 
3913 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3914 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3915 	vmcs_write(PIN_CONTROLS, saved_pin);
3916 }
3917 
3918 static void test_apic_ctls(void)
3919 {
3920 	test_apic_virt_addr();
3921 	test_apic_access_addr();
3922 	test_apic_virtual_ctls();
3923 	test_virtual_intr_ctls();
3924 	test_posted_intr();
3925 }
3926 
3927 /*
3928  * If the “enable VPID†VM-execution control is 1, the value of the
3929  * of the VPID VM-execution control field must not be 0000H.
3930  * [Intel SDM]
3931  */
3932 static void test_vpid(void)
3933 {
3934 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3935 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3936 	u16 vpid = 0x0000;
3937 	int i;
3938 
3939 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
3940 	    (ctrl_cpu_rev[1].clr & CPU_VPID))) {
3941 		test_skip("Secondary controls and/or VPID not supported");
3942 		return;
3943 	}
3944 
3945 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
3946 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
3947 	vmcs_write(VPID, vpid);
3948 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
3949 	test_vmx_controls(true, false);
3950 	report_prefix_pop();
3951 
3952 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
3953 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
3954 	test_vmx_controls(false, false);
3955 	report_prefix_pop();
3956 
3957 	for (i = 0; i < 16; i++) {
3958 		vpid = (short)1 << i;;
3959 		vmcs_write(VPID, vpid);
3960 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
3961 		test_vmx_controls(true, false);
3962 		report_prefix_pop();
3963 	}
3964 
3965 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3966 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3967 }
3968 
3969 static void set_vtpr(unsigned vtpr)
3970 {
3971 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
3972 }
3973 
3974 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
3975 {
3976 	bool valid = true;
3977 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3978 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3979 
3980 	if ((primary & CPU_TPR_SHADOW) &&
3981 	    (!(primary & CPU_SECONDARY) ||
3982 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
3983 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
3984 
3985 	set_vtpr(vtpr);
3986 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
3987 	    threshold, (vtpr >> 4) & 0xf);
3988 	test_vmx_controls(valid, false);
3989 	report_prefix_pop();
3990 }
3991 
3992 static void test_invalid_event_injection(void)
3993 {
3994 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
3995 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
3996 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
3997 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
3998 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
3999 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4000 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4001 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4002 	u32 cnt;
4003 
4004 	/* Setup */
4005 	report_prefix_push("invalid event injection");
4006 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4007 	vmcs_write(ENT_INST_LEN, 0x00000001);
4008 
4009 	/* The field’s interruption type is not set to a reserved value. */
4010 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4011 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4012 			    "RESERVED interruption type invalid [-]",
4013 			    ent_intr_info);
4014 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4015 	test_vmx_controls(false, false);
4016 	report_prefix_pop();
4017 
4018 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4019 			DE_VECTOR;
4020 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4021 			    "RESERVED interruption type invalid [+]",
4022 			    ent_intr_info);
4023 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4024 	test_vmx_controls(true, false);
4025 	report_prefix_pop();
4026 
4027 	/* If the interruption type is other event, the vector is 0. */
4028 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4029 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4030 			    "(OTHER EVENT && vector != 0) invalid [-]",
4031 			    ent_intr_info);
4032 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4033 	test_vmx_controls(false, false);
4034 	report_prefix_pop();
4035 
4036 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4037 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4038 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4039 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4040 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4041 	test_vmx_controls(false, false);
4042 	report_prefix_pop();
4043 
4044 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4045 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4046 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4047 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4048 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4049 	test_vmx_controls(true, false);
4050 	report_prefix_pop();
4051 
4052 	/*
4053 	 * If the interruption type
4054 	 * is HW exception, the vector is at most 31.
4055 	 */
4056 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4057 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4058 			    "(HW exception && vector > 31) invalid [-]",
4059 			    ent_intr_info);
4060 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4061 	test_vmx_controls(false, false);
4062 	report_prefix_pop();
4063 
4064 	/*
4065 	 * deliver-error-code is 1 iff either
4066 	 * (a) the "unrestricted guest" VM-execution control is 0
4067 	 * (b) CR0.PE is set.
4068 	 */
4069 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4070 			GP_VECTOR;
4071 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4072 			    "error code <-> (!URG || prot_mode) [-]",
4073 			    ent_intr_info);
4074 	disable_unrestricted_guest();
4075 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4076 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4077 	test_vmx_controls(false, false);
4078 	report_prefix_pop();
4079 
4080 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4081 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4082 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4083 			    "error code <-> (!URG || prot_mode) [+]",
4084 			    ent_intr_info);
4085 	disable_unrestricted_guest();
4086 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4087 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4088 	test_vmx_controls(true, false);
4089 	report_prefix_pop();
4090 
4091 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4092 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4093 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4094 			    "error code <-> (!URG || prot_mode) [-]",
4095 			    ent_intr_info);
4096 	enable_unrestricted_guest();
4097 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4098 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4099 	test_vmx_controls(false, false);
4100 	report_prefix_pop();
4101 
4102 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4103 			GP_VECTOR;
4104 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4105 			    "error code <-> (!URG || prot_mode) [-]",
4106 			    ent_intr_info);
4107 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4108 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4109 	test_vmx_controls(false, false);
4110 	report_prefix_pop();
4111 
4112 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4113 	report_prefix_push("error code <-> HW exception");
4114 	for (cnt = 0; cnt < 8; cnt++) {
4115 		u32 exception_type_mask = cnt << 8;
4116 		u32 deliver_error_code_mask =
4117 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4118 			INTR_INFO_DELIVER_CODE_MASK : 0;
4119 
4120 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4121 				exception_type_mask | GP_VECTOR;
4122 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4123 				    ent_intr_info);
4124 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4125 		test_vmx_controls(false, false);
4126 		report_prefix_pop();
4127 	}
4128 	report_prefix_pop();
4129 
4130 	/*
4131 	 * deliver-error-code is 1 iff the the vector
4132 	 * indicates an exception that would normally deliver an error code
4133 	 */
4134 	report_prefix_push("error code <-> vector delivers error code");
4135 	for (cnt = 0; cnt < 32; cnt++) {
4136 		bool has_error_code = false;
4137 		u32 deliver_error_code_mask;
4138 
4139 		switch (cnt) {
4140 		case DF_VECTOR:
4141 		case TS_VECTOR:
4142 		case NP_VECTOR:
4143 		case SS_VECTOR:
4144 		case GP_VECTOR:
4145 		case PF_VECTOR:
4146 		case AC_VECTOR:
4147 			has_error_code = true;
4148 		}
4149 
4150 		/* Negative case */
4151 		deliver_error_code_mask = has_error_code ?
4152 						0 :
4153 						INTR_INFO_DELIVER_CODE_MASK;
4154 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4155 				INTR_TYPE_HARD_EXCEPTION | cnt;
4156 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4157 				    ent_intr_info);
4158 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4159 		test_vmx_controls(false, false);
4160 		report_prefix_pop();
4161 
4162 		/* Positive case */
4163 		deliver_error_code_mask = has_error_code ?
4164 						INTR_INFO_DELIVER_CODE_MASK :
4165 						0;
4166 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4167 				INTR_TYPE_HARD_EXCEPTION | cnt;
4168 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4169 				    ent_intr_info);
4170 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4171 		test_vmx_controls(true, false);
4172 		report_prefix_pop();
4173 	}
4174 	report_prefix_pop();
4175 
4176 	/* Reserved bits in the field (30:12) are 0. */
4177 	report_prefix_push("reserved bits clear");
4178 	for (cnt = 12; cnt <= 30; cnt++) {
4179 		ent_intr_info = ent_intr_info_base |
4180 				INTR_INFO_DELIVER_CODE_MASK |
4181 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4182 				(1U << cnt);
4183 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4184 				    ent_intr_info);
4185 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4186 		test_vmx_controls(false, false);
4187 		report_prefix_pop();
4188 	}
4189 	report_prefix_pop();
4190 
4191 	/*
4192 	 * If deliver-error-code is 1
4193 	 * bits 31:15 of the VM-entry exception error-code field are 0.
4194 	 */
4195 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4196 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4197 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4198 			    "VM-entry exception error code[31:15] clear",
4199 			    ent_intr_info);
4200 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4201 	for (cnt = 15; cnt <= 31; cnt++) {
4202 		ent_intr_err = 1U << cnt;
4203 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4204 				    ent_intr_err);
4205 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4206 		test_vmx_controls(false, false);
4207 		report_prefix_pop();
4208 	}
4209 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4210 	report_prefix_pop();
4211 
4212 	/*
4213 	 * If the interruption type is software interrupt, software exception,
4214 	 * or privileged software exception, the VM-entry instruction-length
4215 	 * field is in the range 0–15.
4216 	 */
4217 
4218 	for (cnt = 0; cnt < 3; cnt++) {
4219 		switch (cnt) {
4220 		case 0:
4221 			ent_intr_info = ent_intr_info_base |
4222 					INTR_TYPE_SOFT_INTR;
4223 			break;
4224 		case 1:
4225 			ent_intr_info = ent_intr_info_base |
4226 					INTR_TYPE_SOFT_EXCEPTION;
4227 			break;
4228 		case 2:
4229 			ent_intr_info = ent_intr_info_base |
4230 					INTR_TYPE_PRIV_SW_EXCEPTION;
4231 			break;
4232 		}
4233 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4234 				    "VM-entry instruction-length check",
4235 				    ent_intr_info);
4236 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4237 
4238 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4239 		ent_intr_len = -1;
4240 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4241 				    ent_intr_len);
4242 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4243 		test_vmx_controls(false, false);
4244 		report_prefix_pop();
4245 
4246 		/* Instruction length set to 16 should fail */
4247 		ent_intr_len = 0x00000010;
4248 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4249 				    ent_intr_len);
4250 		vmcs_write(ENT_INST_LEN, 0x00000010);
4251 		test_vmx_controls(false, false);
4252 		report_prefix_pop();
4253 
4254 		report_prefix_pop();
4255 	}
4256 
4257 	/* Cleanup */
4258 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4259 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4260 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4261 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4262 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4263 	vmcs_write(GUEST_CR0, guest_cr0_save);
4264 	report_prefix_pop();
4265 }
4266 
4267 /*
4268  * Test interesting vTPR values for a given TPR threshold.
4269  */
4270 static void test_vtpr_values(unsigned threshold)
4271 {
4272 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4273 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4274 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4275 }
4276 
4277 static void try_tpr_threshold(unsigned threshold)
4278 {
4279 	bool valid = true;
4280 
4281 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4282 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4283 
4284 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4285 	    (secondary & CPU_VINTD)))
4286 		valid = !(threshold >> 4);
4287 
4288 	set_vtpr(-1);
4289 	vmcs_write(TPR_THRESHOLD, threshold);
4290 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4291 	test_vmx_controls(valid, false);
4292 	report_prefix_pop();
4293 
4294 	if (valid)
4295 		test_vtpr_values(threshold);
4296 }
4297 
4298 /*
4299  * Test interesting TPR threshold values.
4300  */
4301 static void test_tpr_threshold_values(void)
4302 {
4303 	unsigned i;
4304 
4305 	for (i = 0; i < 0x10; i++)
4306 		try_tpr_threshold(i);
4307 	for (i = 4; i < 32; i++)
4308 		try_tpr_threshold(1u << i);
4309 	try_tpr_threshold(-1u);
4310 	try_tpr_threshold(0x7fffffff);
4311 }
4312 
4313 /*
4314  * This test covers the following two VM entry checks:
4315  *
4316  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4317  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4318  *         31:4 of the TPR threshold VM-execution control field must
4319 	   be 0.
4320  *         [Intel SDM]
4321  *
4322  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4323  *          "virtual-interrupt delivery" VM-execution control is 0
4324  *          and the "virtualize APIC accesses" VM-execution control
4325  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4326  *          control field must not be greater than the value of bits
4327  *          7:4 of VTPR.
4328  *          [Intel SDM]
4329  */
4330 static void test_tpr_threshold(void)
4331 {
4332 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4333 	void *virtual_apic_page;
4334 
4335 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4336 		return;
4337 
4338 	virtual_apic_page = alloc_page();
4339 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4340 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4341 
4342 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4343 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4344 	test_tpr_threshold_values();
4345 	report_prefix_pop();
4346 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4347 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4348 	test_tpr_threshold_values();
4349 	report_prefix_pop();
4350 
4351 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4352 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES)))) {
4353 		vmcs_write(CPU_EXEC_CTRL0, primary);
4354 		return;
4355 	}
4356 
4357 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4358 
4359 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4360 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4361 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4362 		test_tpr_threshold_values();
4363 		report_prefix_pop();
4364 
4365 		vmcs_write(CPU_EXEC_CTRL0,
4366 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4367 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4368 		test_tpr_threshold_values();
4369 		report_prefix_pop();
4370 	}
4371 
4372 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4373 		vmcs_write(CPU_EXEC_CTRL0,
4374 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4375 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4376 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4377 		test_tpr_threshold_values();
4378 		report_prefix_pop();
4379 
4380 		vmcs_write(CPU_EXEC_CTRL0,
4381 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4382 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4383 		test_tpr_threshold_values();
4384 		report_prefix_pop();
4385 	}
4386 
4387 	if ((ctrl_cpu_rev[1].clr &
4388 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4389 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4390 		vmcs_write(CPU_EXEC_CTRL0,
4391 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4392 		vmcs_write(CPU_EXEC_CTRL1,
4393 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4394 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4395 		test_tpr_threshold_values();
4396 		report_prefix_pop();
4397 
4398 		vmcs_write(CPU_EXEC_CTRL0,
4399 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4400 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4401 		test_tpr_threshold_values();
4402 		report_prefix_pop();
4403 	}
4404 
4405 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4406 	vmcs_write(CPU_EXEC_CTRL0, primary);
4407 }
4408 
4409 /*
4410  * This test verifies the following two vmentry checks:
4411  *
4412  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4413  *  VM-execution control must be 0.
4414  *  [Intel SDM]
4415  *
4416  *  If the “virtual NMIs” VM-execution control is 0, the “NMI-window
4417  *  exiting” VM-execution control must be 0.
4418  *  [Intel SDM]
4419  */
4420 static void test_nmi_ctrls(void)
4421 {
4422 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4423 
4424 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4425 	    (PIN_NMI | PIN_VIRT_NMI)) {
4426 		test_skip("NMI exiting and Virtual NMIs are not supported !");
4427 		return;
4428 	}
4429 
4430 	/* Save the controls so that we can restore them after our tests */
4431 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4432 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4433 
4434 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4435 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4436 
4437 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4438 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4439 	test_vmx_controls(true, false);
4440 	report_prefix_pop();
4441 
4442 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4443 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4444 	test_vmx_controls(false, false);
4445 	report_prefix_pop();
4446 
4447 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4448 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4449 	test_vmx_controls(true, false);
4450 	report_prefix_pop();
4451 
4452 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4453 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4454 	test_vmx_controls(true, false);
4455 	report_prefix_pop();
4456 
4457 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4458 		report_info("NMI-window exiting is not supported, skipping...");
4459 		goto done;
4460 	}
4461 
4462 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4463 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4464 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4465 	test_vmx_controls(false, false);
4466 	report_prefix_pop();
4467 
4468 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4469 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4470 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4471 	test_vmx_controls(true, false);
4472 	report_prefix_pop();
4473 
4474 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4475 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4476 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4477 	test_vmx_controls(true, false);
4478 	report_prefix_pop();
4479 
4480 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4481 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4482 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4483 	test_vmx_controls(true, false);
4484 	report_prefix_pop();
4485 
4486 	/* Restore the controls to their original values */
4487 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4488 done:
4489 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4490 }
4491 
4492 static void test_eptp_ad_bit(u64 eptp, bool ctrl)
4493 {
4494 	vmcs_write(EPTP, eptp);
4495 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4496 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4497 	test_vmx_controls(ctrl, false);
4498 	report_prefix_pop();
4499 
4500 }
4501 
4502 /*
4503  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4504  *    control field must satisfy the following checks:
4505  *
4506  *     - The EPT memory type (bits 2:0) must be a value supported by the
4507  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4508  *     - Bits 5:3 (1 less than the EPT page-walk length) must be 3,
4509  *	 indicating an EPT page-walk length of 4.
4510  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4511  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4512  *	 indicating that the processor does not support accessed and dirty
4513  *	 dirty flags for EPT.
4514  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4515  *	 physical-address width) must all be 0.
4516  *
4517  * 2. If the "unrestricted guest" VM-execution control is 1, the
4518  *    "enable EPT" VM-execution control must also be 1.
4519  */
4520 static void test_ept_eptp(void)
4521 {
4522 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4523 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4524 	u64 eptp_saved = vmcs_read(EPTP);
4525 	u32 primary = primary_saved;
4526 	u32 secondary = secondary_saved;
4527 	u64 msr, eptp = eptp_saved;
4528 	bool un_cache = false;
4529 	bool wr_bk = false;
4530 	bool ctrl;
4531 	u32 i, maxphysaddr;
4532 	u64 j, resv_bits_mask = 0;
4533 
4534 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4535 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4536 		test_skip("\"CPU secondary\" and/or \"enable EPT\" execution controls are not supported !");
4537 		return;
4538 	}
4539 
4540 	/*
4541 	 * Memory type (bits 2:0)
4542 	 */
4543 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
4544 	if (msr & EPT_CAP_UC)
4545 		un_cache = true;
4546 	if (msr & EPT_CAP_WB)
4547 		wr_bk = true;
4548 
4549 	primary |= CPU_SECONDARY;
4550 	vmcs_write(CPU_EXEC_CTRL0, primary);
4551 	secondary |= CPU_EPT;
4552 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4553 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4554 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4555 	vmcs_write(EPTP, eptp);
4556 
4557 	for (i = 0; i < 8; i++) {
4558 		if (i == 0) {
4559 			if (un_cache) {
4560 				report_info("EPT paging structure memory-type is Un-cacheable\n");
4561 				ctrl = true;
4562 			} else {
4563 				ctrl = false;
4564 			}
4565 		} else if (i == 6) {
4566 			if (wr_bk) {
4567 				report_info("EPT paging structure memory-type is Write-back\n");
4568 				ctrl = true;
4569 			} else {
4570 				ctrl = false;
4571 			}
4572 		} else {
4573 			ctrl = false;
4574 		}
4575 
4576 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4577 		vmcs_write(EPTP, eptp);
4578 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4579 		    eptp & EPT_MEM_TYPE_MASK);
4580 		test_vmx_controls(ctrl, false);
4581 		report_prefix_pop();
4582 	}
4583 
4584 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4585 
4586 	/*
4587 	 * Page walk length (bits 5:3)
4588 	 */
4589 	for (i = 0; i < 8; i++) {
4590 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4591 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4592 		if (i == 3)
4593 			ctrl = true;
4594 		else
4595 			ctrl = false;
4596 
4597 		vmcs_write(EPTP, eptp);
4598 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4599 		    eptp & EPTP_PG_WALK_LEN_MASK);
4600 		test_vmx_controls(ctrl, false);
4601 		report_prefix_pop();
4602 	}
4603 
4604 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4605 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4606 
4607 	/*
4608 	 * Accessed and dirty flag (bit 6)
4609 	 */
4610 	if (msr & EPT_CAP_AD_FLAG) {
4611 		report_info("Processor supports accessed and dirty flag");
4612 		eptp &= ~EPTP_AD_FLAG;
4613 		test_eptp_ad_bit(eptp, true);
4614 
4615 		eptp |= EPTP_AD_FLAG;
4616 		test_eptp_ad_bit(eptp, true);
4617 	} else {
4618 		report_info("Processor does not supports accessed and dirty flag");
4619 		eptp &= ~EPTP_AD_FLAG;
4620 		test_eptp_ad_bit(eptp, true);
4621 
4622 		eptp |= EPTP_AD_FLAG;
4623 		test_eptp_ad_bit(eptp, false);
4624 	}
4625 
4626 	/*
4627 	 * Reserved bits [11:7] and [63:N]
4628 	 */
4629 	for (i = 0; i < 32; i++) {
4630 		if (i  == 0)
4631 			ctrl = true;
4632 		else
4633 			ctrl = false;
4634 
4635 		eptp = (eptp &
4636 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4637 		    (i << EPTP_RESERV_BITS_SHIFT);
4638 		vmcs_write(EPTP, eptp);
4639 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4640 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4641 		    EPTP_RESERV_BITS_MASK);
4642 		test_vmx_controls(ctrl, false);
4643 		report_prefix_pop();
4644 	}
4645 
4646 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4647 
4648 	maxphysaddr = cpuid_maxphyaddr();
4649 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4650 		resv_bits_mask |= 1ul << i;
4651 	}
4652 
4653 	for (j = 0; j < (63 - maxphysaddr + 1); j++) {
4654 		if (j  == 0)
4655 			ctrl = true;
4656 		else
4657 			ctrl = false;
4658 
4659 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4660 		    (j << maxphysaddr);
4661 		vmcs_write(EPTP, eptp);
4662 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4663 		    (eptp >> maxphysaddr) & resv_bits_mask);
4664 		test_vmx_controls(ctrl, false);
4665 		report_prefix_pop();
4666 	}
4667 
4668 	secondary &= ~(CPU_EPT | CPU_URG);
4669 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4670 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4671 	test_vmx_controls(true, false);
4672 	report_prefix_pop();
4673 
4674 	secondary |= CPU_URG;
4675 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4676 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4677 	test_vmx_controls(false, false);
4678 	report_prefix_pop();
4679 
4680 	secondary |= CPU_EPT;
4681 	setup_ept(false);
4682 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4683 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4684 	test_vmx_controls(true, false);
4685 	report_prefix_pop();
4686 
4687 	secondary &= ~CPU_URG;
4688 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4689 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4690 	test_vmx_controls(true, false);
4691 	report_prefix_pop();
4692 
4693 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4694 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4695 	vmcs_write(EPTP, eptp_saved);
4696 }
4697 
4698 /*
4699  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4700  * VM-execution control must also be 1. In addition, the PML address
4701  * must satisfy the following checks:
4702  *
4703  *    * Bits 11:0 of the address must be 0.
4704  *    * The address should not set any bits beyond the processor's
4705  *	physical-address width.
4706  *
4707  *  [Intel SDM]
4708  */
4709 static void test_pml(void)
4710 {
4711 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4712 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4713 	u32 primary = primary_saved;
4714 	u32 secondary = secondary_saved;
4715 
4716 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4717 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
4718 		test_skip("\"Secondary execution\" control or \"enable EPT\" control or \"enable PML\" control is not supported !");
4719 		return;
4720 	}
4721 
4722 	primary |= CPU_SECONDARY;
4723 	vmcs_write(CPU_EXEC_CTRL0, primary);
4724 	secondary &= ~(CPU_PML | CPU_EPT);
4725 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4726 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
4727 	test_vmx_controls(true, false);
4728 	report_prefix_pop();
4729 
4730 	secondary |= CPU_PML;
4731 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4732 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
4733 	test_vmx_controls(false, false);
4734 	report_prefix_pop();
4735 
4736 	secondary |= CPU_EPT;
4737 	setup_ept(false);
4738 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4739 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
4740 	test_vmx_controls(true, false);
4741 	report_prefix_pop();
4742 
4743 	secondary &= ~CPU_PML;
4744 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4745 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
4746 	test_vmx_controls(true, false);
4747 	report_prefix_pop();
4748 
4749 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
4750 				 PAGE_SIZE, false, false);
4751 
4752 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4753 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4754 }
4755 
4756  /*
4757  * If the "activate VMX-preemption timer" VM-execution control is 0, the
4758  * the "save VMX-preemption timer value" VM-exit control must also be 0.
4759  *
4760  *  [Intel SDM]
4761  */
4762 static void test_vmx_preemption_timer(void)
4763 {
4764 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4765 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
4766 	u32 pin = saved_pin;
4767 	u32 exit = saved_exit;
4768 
4769 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
4770 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
4771 		printf("\"Save-VMX-preemption-timer\" control and/or \"Enable-VMX-preemption-timer\" control is not supported\n");
4772 		return;
4773 	}
4774 
4775 	pin |= PIN_PREEMPT;
4776 	vmcs_write(PIN_CONTROLS, pin);
4777 	exit &= ~EXI_SAVE_PREEMPT;
4778 	vmcs_write(EXI_CONTROLS, exit);
4779 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
4780 	test_vmx_controls(true, false);
4781 	report_prefix_pop();
4782 
4783 	exit |= EXI_SAVE_PREEMPT;
4784 	vmcs_write(EXI_CONTROLS, exit);
4785 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
4786 	test_vmx_controls(true, false);
4787 	report_prefix_pop();
4788 
4789 	pin &= ~PIN_PREEMPT;
4790 	vmcs_write(PIN_CONTROLS, pin);
4791 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
4792 	test_vmx_controls(false, false);
4793 	report_prefix_pop();
4794 
4795 	exit &= ~EXI_SAVE_PREEMPT;
4796 	vmcs_write(EXI_CONTROLS, exit);
4797 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
4798 	test_vmx_controls(true, false);
4799 	report_prefix_pop();
4800 
4801 	vmcs_write(PIN_CONTROLS, saved_pin);
4802 	vmcs_write(EXI_CONTROLS, saved_exit);
4803 }
4804 
4805 /*
4806  * Tests for VM-execution control fields
4807  */
4808 static void test_vm_execution_ctls(void)
4809 {
4810 	test_pin_based_ctls();
4811 	test_primary_processor_based_ctls();
4812 	test_secondary_processor_based_ctls();
4813 	test_cr3_targets();
4814 	test_io_bitmaps();
4815 	test_msr_bitmap();
4816 	test_apic_ctls();
4817 	test_tpr_threshold();
4818 	test_nmi_ctrls();
4819 	test_pml();
4820 	test_vpid();
4821 	test_ept_eptp();
4822 	test_vmx_preemption_timer();
4823 }
4824 
4825  /*
4826   * The following checks are performed for the VM-entry MSR-load address if
4827   * the VM-entry MSR-load count field is non-zero:
4828   *
4829   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
4830   *      The address should not set any bits beyond the processor’s
4831   *      physical-address width.
4832   *
4833   *    - The address of the last byte in the VM-entry MSR-load area
4834   *      should not set any bits beyond the processor’s physical-address
4835   *      width. The address of this last byte is VM-entry MSR-load address
4836   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
4837   *      uses more bits than the processor’s physical-address width.)
4838   *
4839   *
4840   *  [Intel SDM]
4841   */
4842 static void test_entry_msr_load(void)
4843 {
4844 	entry_msr_load = alloc_page();
4845 	u64 tmp;
4846 	u32 entry_msr_ld_cnt = 1;
4847 	int i;
4848 	u32 addr_len = 64;
4849 
4850 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
4851 
4852 	/* Check first 4 bits of VM-entry MSR-load address */
4853 	for (i = 0; i < 4; i++) {
4854 		tmp = (u64)entry_msr_load | 1ull << i;
4855 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
4856 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
4857 				    tmp & 0xf);
4858 		test_vmx_controls(false, false);
4859 		report_prefix_pop();
4860 	}
4861 
4862 	if (basic.val & (1ul << 48))
4863 		addr_len = 32;
4864 
4865 	test_vmcs_addr_values("VM-entry-MSR-load address",
4866 				ENTER_MSR_LD_ADDR, 16, false, false,
4867 				4, addr_len - 1);
4868 
4869 	/*
4870 	 * Check last byte of VM-entry MSR-load address
4871 	 */
4872 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
4873 
4874 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
4875 							i < 64; i++) {
4876 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
4877 			1ul << i;
4878 		vmcs_write(ENTER_MSR_LD_ADDR,
4879 			   tmp - (entry_msr_ld_cnt * 16 - 1));
4880 		test_vmx_controls(false, false);
4881 	}
4882 
4883 	vmcs_write(ENT_MSR_LD_CNT, 2);
4884 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
4885 	test_vmx_controls(false, false);
4886 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
4887 	test_vmx_controls(true, false);
4888 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
4889 	test_vmx_controls(true, false);
4890 }
4891 
4892 /*
4893  * Tests for VM-entry control fields
4894  */
4895 static void test_vm_entry_ctls(void)
4896 {
4897 	test_invalid_event_injection();
4898 	test_entry_msr_load();
4899 }
4900 
4901 /*
4902  * The following checks are performed for the VM-exit MSR-store address if
4903  * the VM-exit MSR-store count field is non-zero:
4904  *
4905  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
4906  *      The address should not set any bits beyond the processor’s
4907  *      physical-address width.
4908  *
4909  *    - The address of the last byte in the VM-exit MSR-store area
4910  *      should not set any bits beyond the processor’s physical-address
4911  *      width. The address of this last byte is VM-exit MSR-store address
4912  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
4913  *      uses more bits than the processor’s physical-address width.)
4914  *
4915  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
4916  * in the range 63:32.
4917  *
4918  *  [Intel SDM]
4919  */
4920 static void test_exit_msr_store(void)
4921 {
4922 	exit_msr_store = alloc_page();
4923 	u64 tmp;
4924 	u32 exit_msr_st_cnt = 1;
4925 	int i;
4926 	u32 addr_len = 64;
4927 
4928 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
4929 
4930 	/* Check first 4 bits of VM-exit MSR-store address */
4931 	for (i = 0; i < 4; i++) {
4932 		tmp = (u64)exit_msr_store | 1ull << i;
4933 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
4934 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
4935 				    tmp & 0xf);
4936 		test_vmx_controls(false, false);
4937 		report_prefix_pop();
4938 	}
4939 
4940 	if (basic.val & (1ul << 48))
4941 		addr_len = 32;
4942 
4943 	test_vmcs_addr_values("VM-exit-MSR-store address",
4944 				EXIT_MSR_ST_ADDR, 16, false, false,
4945 				4, addr_len - 1);
4946 
4947 	/*
4948 	 * Check last byte of VM-exit MSR-store address
4949 	 */
4950 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
4951 
4952 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
4953 							i < 64; i++) {
4954 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
4955 			1ul << i;
4956 		vmcs_write(EXIT_MSR_ST_ADDR,
4957 			   tmp - (exit_msr_st_cnt * 16 - 1));
4958                 test_vmx_controls(false, false);
4959 	}
4960 
4961 	vmcs_write(EXI_MSR_ST_CNT, 2);
4962 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
4963 	test_vmx_controls(false, false);
4964 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
4965 	test_vmx_controls(true, false);
4966 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
4967 	test_vmx_controls(true, false);
4968 }
4969 
4970 /*
4971  * Tests for VM-exit controls
4972  */
4973 static void test_vm_exit_ctls(void)
4974 {
4975 	test_exit_msr_store();
4976 }
4977 
4978 /*
4979  * Check that the virtual CPU checks all of the VMX controls as
4980  * documented in the Intel SDM.
4981  */
4982 static void vmx_controls_test(void)
4983 {
4984 	/*
4985 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
4986 	 * fail due to invalid guest state, should we make it that
4987 	 * far.
4988 	 */
4989 	vmcs_write(GUEST_RFLAGS, 0);
4990 
4991 	test_vm_execution_ctls();
4992 	test_vm_entry_ctls();
4993 	test_vm_exit_ctls();
4994 }
4995 
4996 static bool valid_vmcs_for_vmentry(void)
4997 {
4998 	struct vmcs *current_vmcs = NULL;
4999 
5000 	if (vmcs_save(&current_vmcs))
5001 		return false;
5002 
5003 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
5004 }
5005 
5006 static void try_vmentry_in_movss_shadow(void)
5007 {
5008 	u32 vm_inst_err;
5009 	u32 flags;
5010 	bool early_failure = false;
5011 	u32 expected_flags = X86_EFLAGS_FIXED;
5012 	bool valid_vmcs = valid_vmcs_for_vmentry();
5013 
5014 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
5015 
5016 	/*
5017 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
5018 	 * unsupported VMCS component").
5019 	 */
5020 	vmcs_write(~0u, 0);
5021 
5022 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
5023 			      "vmwrite %%rsp, %%rdx;"
5024 			      "mov 0f, %%rax;"
5025 			      "mov %[host_rip], %%edx;"
5026 			      "vmwrite %%rax, %%rdx;"
5027 			      "mov $-1, %%ah;"
5028 			      "sahf;"
5029 			      "mov %%ss, %%ax;"
5030 			      "mov %%ax, %%ss;"
5031 			      "vmlaunch;"
5032 			      "mov $1, %[early_failure];"
5033 			      "0: lahf;"
5034 			      "movzbl %%ah, %[flags]"
5035 			      : [early_failure] "+r" (early_failure),
5036 				[flags] "=&a" (flags)
5037 			      : [host_rsp] "i" (HOST_RSP),
5038 				[host_rip] "i" (HOST_RIP)
5039 			      : "rdx", "cc", "memory");
5040 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
5041 
5042 	report("Early VM-entry failure", early_failure);
5043 	report("RFLAGS[8:0] is %x (actual %x)", flags == expected_flags,
5044 	       expected_flags, flags);
5045 	if (valid_vmcs)
5046 		report("VM-instruction error is %d (actual %d)",
5047 		       vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
5048 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
5049 }
5050 
5051 static void vmentry_movss_shadow_test(void)
5052 {
5053 	struct vmcs *orig_vmcs;
5054 
5055 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
5056 
5057 	/*
5058 	 * Set the launched flag on the current VMCS to verify the correct
5059 	 * error priority, below.
5060 	 */
5061 	test_set_guest(v2_null_test_guest);
5062 	enter_guest();
5063 
5064 	/*
5065 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
5066 	 * fail due to invalid guest state (if we make it that far).
5067 	 */
5068 	vmcs_write(GUEST_RFLAGS, 0);
5069 
5070 	/*
5071 	 * "VM entry with events blocked by MOV SS" takes precedence over
5072 	 * "VMLAUNCH with non-clear VMCS."
5073 	 */
5074 	report_prefix_push("valid current-VMCS");
5075 	try_vmentry_in_movss_shadow();
5076 	report_prefix_pop();
5077 
5078 	/*
5079 	 * VMfailInvalid takes precedence over "VM entry with events
5080 	 * blocked by MOV SS."
5081 	 */
5082 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
5083 	report_prefix_push("no current-VMCS");
5084 	try_vmentry_in_movss_shadow();
5085 	report_prefix_pop();
5086 
5087 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
5088 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
5089 }
5090 
5091 #define X86_FEATURE_PCID       (1 << 17)
5092 #define X86_FEATURE_MCE        (1 << 7)
5093 
5094 static int write_cr4_checking(unsigned long val)
5095 {
5096 	asm volatile(ASM_TRY("1f")
5097 		     "mov %0, %%cr4\n\t"
5098 		     "1:": : "r" (val));
5099 	return exception_vector();
5100 }
5101 
5102 static void vmx_cr_load_test(void)
5103 {
5104 	struct cpuid _cpuid = cpuid(1);
5105 	unsigned long cr4 = read_cr4(), cr3 = read_cr3();
5106 
5107 	if (!(_cpuid.c & X86_FEATURE_PCID)) {
5108 		report_skip("PCID not detected");
5109 		return;
5110 	}
5111 	if (!(_cpuid.d & X86_FEATURE_MCE)) {
5112 		report_skip("MCE not detected");
5113 		return;
5114 	}
5115 
5116 	TEST_ASSERT(!(cr4 & (X86_CR4_PCIDE | X86_CR4_MCE)));
5117 	TEST_ASSERT(!(cr3 & X86_CR3_PCID_MASK));
5118 
5119 	/* Enable PCID for L1. */
5120 	cr4 |= X86_CR4_PCIDE;
5121 	cr3 |= 0x1;
5122 	TEST_ASSERT(!write_cr4_checking(cr4));
5123 	write_cr3(cr3);
5124 
5125 	test_set_guest(v2_null_test_guest);
5126 	vmcs_write(HOST_CR4, cr4);
5127 	vmcs_write(HOST_CR3, cr3);
5128 	enter_guest();
5129 
5130 	/*
5131 	 * No exception is expected.
5132 	 *
5133 	 * NB. KVM loads the last guest write to CR4 into CR4 read
5134 	 *     shadow. In order to trigger an exit to KVM, we can set a
5135 	 *     bit that was zero in the above CR4 write and is owned by
5136 	 *     KVM. We choose to set CR4.MCE, which shall have no side
5137 	 *     effect because normally no guest MCE (e.g., as the result
5138 	 *     of bad memory) would happen during this test.
5139 	 */
5140 	TEST_ASSERT(!write_cr4_checking(cr4 | X86_CR4_MCE));
5141 
5142 	/* Cleanup L1 state: disable PCID. */
5143 	write_cr3(cr3 & ~X86_CR3_PCID_MASK);
5144 	TEST_ASSERT(!write_cr4_checking(cr4 & ~X86_CR4_PCIDE));
5145 }
5146 
5147 static void vmx_nm_test_guest(void)
5148 {
5149 	write_cr0(read_cr0() | X86_CR0_TS);
5150 	asm volatile("fnop");
5151 }
5152 
5153 static void check_nm_exit(const char *test)
5154 {
5155 	u32 reason = vmcs_read(EXI_REASON);
5156 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
5157 	const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
5158 		NM_VECTOR;
5159 
5160 	report("%s", reason == VMX_EXC_NMI && intr_info == expected, test);
5161 }
5162 
5163 /*
5164  * This test checks that:
5165  *
5166  * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then
5167  *     a subsequent #NM VM-exit is reflected to L1.
5168  *
5169  * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a
5170  *     subsequent #NM VM-exit is reflected to L1.
5171  */
5172 static void vmx_nm_test(void)
5173 {
5174 	unsigned long cr0 = read_cr0();
5175 
5176 	test_set_guest(vmx_nm_test_guest);
5177 
5178 	/*
5179 	 * L1 wants to intercept #NM exceptions encountered in L2.
5180 	 */
5181 	vmcs_write(EXC_BITMAP, 1 << NM_VECTOR);
5182 
5183 	/*
5184 	 * Launch L2 with CR0.TS clear, but don't claim host ownership of
5185 	 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop,
5186 	 * which will raise #NM. L0 should reflect the #NM VM-exit to L1.
5187 	 */
5188 	vmcs_write(CR0_MASK, 0);
5189 	vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS);
5190 	enter_guest();
5191 	check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1");
5192 
5193 	/*
5194 	 * Re-enter L2 at the fnop instruction, with CR0.TS clear but
5195 	 * CR0.EM set. The fnop will still raise #NM, and L0 should
5196 	 * reflect the #NM VM-exit to L1.
5197 	 */
5198 	vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM);
5199 	enter_guest();
5200 	check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1");
5201 
5202 	/*
5203 	 * Re-enter L2 at the fnop instruction, with both CR0.TS and
5204 	 * CR0.EM clear. There will be no #NM, and the L2 guest should
5205 	 * exit normally.
5206 	 */
5207 	vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM));
5208 	enter_guest();
5209 }
5210 
5211 bool vmx_pending_event_ipi_fired;
5212 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
5213 {
5214 	vmx_pending_event_ipi_fired = true;
5215 	eoi();
5216 }
5217 
5218 bool vmx_pending_event_guest_run;
5219 static void vmx_pending_event_guest(void)
5220 {
5221 	vmcall();
5222 	vmx_pending_event_guest_run = true;
5223 }
5224 
5225 static void vmx_pending_event_test_core(bool guest_hlt)
5226 {
5227 	int ipi_vector = 0xf1;
5228 
5229 	vmx_pending_event_ipi_fired = false;
5230 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
5231 
5232 	vmx_pending_event_guest_run = false;
5233 	test_set_guest(vmx_pending_event_guest);
5234 
5235 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
5236 
5237 	enter_guest();
5238 	skip_exit_vmcall();
5239 
5240 	if (guest_hlt)
5241 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
5242 
5243 	irq_disable();
5244 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
5245 				   APIC_DM_FIXED | ipi_vector,
5246 				   0);
5247 
5248 	enter_guest();
5249 
5250 	assert_exit_reason(VMX_EXTINT);
5251 	report("Guest did not run before host received IPI",
5252 		   !vmx_pending_event_guest_run);
5253 
5254 	irq_enable();
5255 	asm volatile ("nop");
5256 	irq_disable();
5257 	report("Got pending interrupt after IRQ enabled",
5258 		   vmx_pending_event_ipi_fired);
5259 
5260 	if (guest_hlt)
5261 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
5262 
5263 	enter_guest();
5264 	report("Guest finished running when no interrupt",
5265 		   vmx_pending_event_guest_run);
5266 }
5267 
5268 static void vmx_pending_event_test(void)
5269 {
5270 	vmx_pending_event_test_core(false);
5271 }
5272 
5273 static void vmx_pending_event_hlt_test(void)
5274 {
5275 	vmx_pending_event_test_core(true);
5276 }
5277 
5278 static int vmx_window_test_ud_count;
5279 
5280 static void vmx_window_test_ud_handler(struct ex_regs *regs)
5281 {
5282 	vmx_window_test_ud_count++;
5283 }
5284 
5285 static void vmx_nmi_window_test_guest(void)
5286 {
5287 	handle_exception(UD_VECTOR, vmx_window_test_ud_handler);
5288 
5289 	asm volatile("vmcall\n\t"
5290 		     "nop\n\t");
5291 
5292 	handle_exception(UD_VECTOR, NULL);
5293 }
5294 
5295 static void verify_nmi_window_exit(u64 rip)
5296 {
5297 	u32 exit_reason = vmcs_read(EXI_REASON);
5298 
5299 	report("Exit reason (%d) is 'NMI window'",
5300 	       exit_reason == VMX_NMI_WINDOW, exit_reason);
5301 	report("RIP (%#lx) is %#lx", vmcs_read(GUEST_RIP) == rip,
5302 	       vmcs_read(GUEST_RIP), rip);
5303 	report("Activity state (%ld) is 'ACTIVE'",
5304 	       vmcs_read(GUEST_ACTV_STATE) == ACTV_ACTIVE,
5305 	       vmcs_read(GUEST_ACTV_STATE));
5306 }
5307 
5308 static void vmx_nmi_window_test(void)
5309 {
5310 	u64 nop_addr;
5311 	void *ud_fault_addr = get_idt_addr(&boot_idt[UD_VECTOR]);
5312 
5313 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
5314 		report_skip("CPU does not support the \"Virtual NMIs\" VM-execution control.");
5315 		return;
5316 	}
5317 
5318 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
5319 		report_skip("CPU does not support the \"NMI-window exiting\" VM-execution control.");
5320 		return;
5321 	}
5322 
5323 	vmx_window_test_ud_count = 0;
5324 
5325 	report_prefix_push("NMI-window");
5326 	test_set_guest(vmx_nmi_window_test_guest);
5327 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
5328 	enter_guest();
5329 	skip_exit_vmcall();
5330 	nop_addr = vmcs_read(GUEST_RIP);
5331 
5332 	/*
5333 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
5334 	 * RIP will not advance.
5335 	 */
5336 	report_prefix_push("active, no blocking");
5337 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
5338 	enter_guest();
5339 	verify_nmi_window_exit(nop_addr);
5340 	report_prefix_pop();
5341 
5342 	/*
5343 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
5344 	 * a VM-exit on the next instruction after the nop. (The nop
5345 	 * is one byte.)
5346 	 */
5347 	report_prefix_push("active, blocking by MOV-SS");
5348 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
5349 	enter_guest();
5350 	verify_nmi_window_exit(nop_addr + 1);
5351 	report_prefix_pop();
5352 
5353 	/*
5354 	 * Ask for "NMI-window exiting" (with event injection), and
5355 	 * expect a VM-exit after the event is injected. (RIP should
5356 	 * be at the address specified in the IDT entry for #UD.)
5357 	 */
5358 	report_prefix_push("active, no blocking, injecting #UD");
5359 	vmcs_write(ENT_INTR_INFO,
5360 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | UD_VECTOR);
5361 	enter_guest();
5362 	verify_nmi_window_exit((u64)ud_fault_addr);
5363 	report_prefix_pop();
5364 
5365 	/*
5366 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
5367 	 * a VM-exit after the next IRET (i.e. after the #UD handler
5368 	 * returns). So, RIP should be back at one byte past the nop.
5369 	 */
5370 	report_prefix_push("active, blocking by NMI");
5371 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
5372 	enter_guest();
5373 	verify_nmi_window_exit(nop_addr + 1);
5374 	report("#UD handler executed once (actual %d times)",
5375 	       vmx_window_test_ud_count == 1,
5376 	       vmx_window_test_ud_count);
5377 	report_prefix_pop();
5378 
5379 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
5380 		report_skip("CPU does not support activity state HLT.");
5381 	} else {
5382 		/*
5383 		 * Ask for "NMI-window exiting" when entering activity
5384 		 * state HLT, and expect an immediate VM-exit. RIP is
5385 		 * still one byte past the nop.
5386 		 */
5387 		report_prefix_push("halted, no blocking");
5388 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
5389 		enter_guest();
5390 		verify_nmi_window_exit(nop_addr + 1);
5391 		report_prefix_pop();
5392 
5393 		/*
5394 		 * Ask for "NMI-window exiting" when entering activity
5395 		 * state HLT (with event injection), and expect a
5396 		 * VM-exit after the event is injected. (RIP should be
5397 		 * at the address specified in the IDT entry for #UD.)
5398 		 */
5399 		report_prefix_push("halted, no blocking, injecting #UD");
5400 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
5401 		vmcs_write(ENT_INTR_INFO,
5402 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
5403 			   UD_VECTOR);
5404 		enter_guest();
5405 		verify_nmi_window_exit((u64)ud_fault_addr);
5406 		report_prefix_pop();
5407 	}
5408 
5409 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
5410 	enter_guest();
5411 	report_prefix_pop();
5412 }
5413 
5414 static void vmx_intr_window_test_guest(void)
5415 {
5416 	handle_exception(UD_VECTOR, vmx_window_test_ud_handler);
5417 
5418 	/*
5419 	 * The two consecutive STIs are to ensure that only the first
5420 	 * one has a shadow. Note that NOP and STI are one byte
5421 	 * instructions.
5422 	 */
5423 	asm volatile("vmcall\n\t"
5424 		     "nop\n\t"
5425 		     "sti\n\t"
5426 		     "sti\n\t");
5427 
5428 	handle_exception(UD_VECTOR, NULL);
5429 }
5430 
5431 static void verify_intr_window_exit(u64 rip)
5432 {
5433 	u32 exit_reason = vmcs_read(EXI_REASON);
5434 
5435 	report("Exit reason (%d) is 'interrupt window'",
5436 	       exit_reason == VMX_INTR_WINDOW, exit_reason);
5437 	report("RIP (%#lx) is %#lx", vmcs_read(GUEST_RIP) == rip,
5438 	       vmcs_read(GUEST_RIP), rip);
5439 	report("Activity state (%ld) is 'ACTIVE'",
5440 	       vmcs_read(GUEST_ACTV_STATE) == ACTV_ACTIVE,
5441 	       vmcs_read(GUEST_ACTV_STATE));
5442 }
5443 
5444 static void vmx_intr_window_test(void)
5445 {
5446 	u64 vmcall_addr;
5447 	u64 nop_addr;
5448 	unsigned int orig_ud_gate_type;
5449 	void *ud_fault_addr = get_idt_addr(&boot_idt[UD_VECTOR]);
5450 
5451 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
5452 		report_skip("CPU does not support the \"interrupt-window exiting\" VM-execution control.");
5453 		return;
5454 	}
5455 
5456 	/*
5457 	 * Change the IDT entry for #UD from interrupt gate to trap gate,
5458 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
5459 	 * be disabled after vectoring a #UD.
5460 	 */
5461 	orig_ud_gate_type = boot_idt[UD_VECTOR].type;
5462 	boot_idt[UD_VECTOR].type = 15;
5463 
5464 	report_prefix_push("interrupt-window");
5465 	test_set_guest(vmx_intr_window_test_guest);
5466 	enter_guest();
5467 	assert_exit_reason(VMX_VMCALL);
5468 	vmcall_addr = vmcs_read(GUEST_RIP);
5469 
5470 	/*
5471 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
5472 	 * no blocking; expect an immediate VM-exit. Note that we have
5473 	 * not advanced past the vmcall instruction yet, so RIP should
5474 	 * point to the vmcall instruction.
5475 	 */
5476 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
5477 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
5478 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
5479 	enter_guest();
5480 	verify_intr_window_exit(vmcall_addr);
5481 	report_prefix_pop();
5482 
5483 	/*
5484 	 * Ask for "interrupt-window exiting" (with event injection)
5485 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
5486 	 * the event is injected. That is, RIP should should be at the
5487 	 * address specified in the IDT entry for #UD.
5488 	 */
5489 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #UD");
5490 	vmcs_write(ENT_INTR_INFO,
5491 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | UD_VECTOR);
5492 	vmcall_addr = vmcs_read(GUEST_RIP);
5493 	enter_guest();
5494 	verify_intr_window_exit((u64)ud_fault_addr);
5495 	report_prefix_pop();
5496 
5497 	/*
5498 	 * Let the L2 guest run through the IRET, back to the VMCALL.
5499 	 * We have to clear the "interrupt-window exiting"
5500 	 * VM-execution control, or it would just keep causing
5501 	 * VM-exits. Then, advance past the VMCALL and set the
5502 	 * "interrupt-window exiting" VM-execution control again.
5503 	 */
5504 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
5505 	enter_guest();
5506 	skip_exit_vmcall();
5507 	nop_addr = vmcs_read(GUEST_RIP);
5508 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
5509 
5510 	/*
5511 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
5512 	 * RFLAGS.IF set, and expect a VM-exit on the next
5513 	 * instruction. (NOP is one byte.)
5514 	 */
5515 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
5516 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
5517 	enter_guest();
5518 	verify_intr_window_exit(nop_addr + 1);
5519 	report_prefix_pop();
5520 
5521 	/*
5522 	 * Back up to the NOP and ask for "interrupt-window exiting"
5523 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
5524 	 * on the next instruction. (NOP is one byte.)
5525 	 */
5526 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
5527 	vmcs_write(GUEST_RIP, nop_addr);
5528 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
5529 	enter_guest();
5530 	verify_intr_window_exit(nop_addr + 1);
5531 	report_prefix_pop();
5532 
5533 	/*
5534 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
5535 	 * and expect a VM-exit on the instruction following the STI
5536 	 * shadow. Only the first STI (which is one byte past the NOP)
5537 	 * should have a shadow. The second STI (which is two bytes
5538 	 * past the NOP) has no shadow. Therefore, the interrupt
5539 	 * window opens at three bytes past the NOP.
5540 	 */
5541 	report_prefix_push("active, RFLAGS.IF = 0");
5542 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
5543 	enter_guest();
5544 	verify_intr_window_exit(nop_addr + 3);
5545 	report_prefix_pop();
5546 
5547 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
5548 		report_skip("CPU does not support activity state HLT.");
5549 	} else {
5550 		/*
5551 		 * Ask for "interrupt-window exiting" when entering
5552 		 * activity state HLT, and expect an immediate
5553 		 * VM-exit. RIP is still three bytes past the nop.
5554 		 */
5555 		report_prefix_push("halted, no blocking");
5556 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
5557 		enter_guest();
5558 		verify_intr_window_exit(nop_addr + 3);
5559 		report_prefix_pop();
5560 
5561 		/*
5562 		 * Ask for "interrupt-window exiting" when entering
5563 		 * activity state HLT (with event injection), and
5564 		 * expect a VM-exit after the event is injected. That
5565 		 * is, RIP should should be at the address specified
5566 		 * in the IDT entry for #UD.
5567 		 */
5568 		report_prefix_push("halted, no blocking, injecting #UD");
5569 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
5570 		vmcs_write(ENT_INTR_INFO,
5571 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
5572 			   UD_VECTOR);
5573 		enter_guest();
5574 		verify_intr_window_exit((u64)ud_fault_addr);
5575 		report_prefix_pop();
5576 	}
5577 
5578 	boot_idt[UD_VECTOR].type = orig_ud_gate_type;
5579 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
5580 	enter_guest();
5581 	report_prefix_pop();
5582 }
5583 
5584 #define GUEST_TSC_OFFSET (1u << 30)
5585 
5586 static u64 guest_tsc;
5587 
5588 static void vmx_store_tsc_test_guest(void)
5589 {
5590 	guest_tsc = rdtsc();
5591 }
5592 
5593 /*
5594  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
5595  * list, the value saved is not subject to the TSC offset that is
5596  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
5597  */
5598 static void vmx_store_tsc_test(void)
5599 {
5600 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
5601 	u64 low, high;
5602 
5603 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
5604 		report_skip("'Use TSC offsetting' not supported");
5605 		return;
5606 	}
5607 
5608 	test_set_guest(vmx_store_tsc_test_guest);
5609 
5610 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
5611 	vmcs_write(EXI_MSR_ST_CNT, 1);
5612 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
5613 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
5614 
5615 	low = rdtsc();
5616 	enter_guest();
5617 	high = rdtsc();
5618 
5619 	report("RDTSC value in the guest (%lu) is in range [%lu, %lu]",
5620 	       low + GUEST_TSC_OFFSET <= guest_tsc &&
5621 	       guest_tsc <= high + GUEST_TSC_OFFSET,
5622 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
5623 	report("IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
5624 	       low <= msr_entry.value && msr_entry.value <= high,
5625 	       msr_entry.value, low, high);
5626 }
5627 
5628 static void vmx_db_test_guest(void)
5629 {
5630 	/*
5631 	 * For a hardware generated single-step #DB.
5632 	 */
5633 	asm volatile("vmcall;"
5634 		     "nop;"
5635 		     ".Lpost_nop:");
5636 	/*
5637 	 * ...in a MOVSS shadow, with pending debug exceptions.
5638 	 */
5639 	asm volatile("vmcall;"
5640 		     "nop;"
5641 		     ".Lpost_movss_nop:");
5642 	/*
5643 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
5644 	 * emulates it in software.)
5645 	 */
5646 	asm volatile("vmcall;"
5647 		     "wbinvd;"
5648 		     ".Lpost_wbinvd:");
5649 	/*
5650 	 * ...in a MOVSS shadow, with pending debug exceptions.
5651 	 */
5652 	asm volatile("vmcall;"
5653 		     "wbinvd;"
5654 		     ".Lpost_movss_wbinvd:");
5655 	/*
5656 	 * For a hardware generated single-step #DB in a transactional region.
5657 	 */
5658 	asm volatile("vmcall;"
5659 		     ".Lxbegin: xbegin .Lskip_rtm;"
5660 		     "xend;"
5661 		     ".Lskip_rtm:");
5662 }
5663 
5664 /*
5665  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
5666  * L2. No #DB is delivered and L2 continues to the next point of
5667  * interest.
5668  */
5669 static void dismiss_db(void)
5670 {
5671 	vmcs_write(GUEST_PENDING_DEBUG, 0);
5672 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
5673 	enter_guest();
5674 }
5675 
5676 /*
5677  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
5678  * Then throw away the #DB exception and resume L2.
5679  */
5680 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
5681 			  void *expected_rip, u64 expected_exit_qual,
5682 			  u64 expected_dr6)
5683 {
5684 	u32 reason = vmcs_read(EXI_REASON);
5685 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
5686 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
5687 	u64 guest_rip = vmcs_read(GUEST_RIP);
5688 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
5689 	u64 dr6 = read_dr6();
5690 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
5691 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
5692 
5693 	report("Expected #DB VM-exit",
5694 	       reason == VMX_EXC_NMI && intr_info == expected_intr_info);
5695 	report("Expected RIP %p (actual %lx)", (u64)expected_rip == guest_rip,
5696 	       expected_rip, guest_rip);
5697 	report_xfail("Expected pending debug exceptions 0 (actual %lx)",
5698 		     xfail_pdbg, 0 == guest_pending_dbg, guest_pending_dbg);
5699 	report_xfail("Expected exit qualification %lx (actual %lx)", xfail_qual,
5700 		     expected_exit_qual == exit_qual,
5701 		     expected_exit_qual, exit_qual);
5702 	report_xfail("Expected DR6 %lx (actual %lx)", xfail_dr6,
5703 		     expected_dr6 == dr6, expected_dr6, dr6);
5704 	dismiss_db();
5705 }
5706 
5707 /*
5708  * Assuming the guest has just exited on a VMCALL instruction, skip
5709  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
5710  * pending debug exceptions are non-zero, set the VMCS up as if the
5711  * previous instruction was a MOVSS that generated the indicated
5712  * pending debug exceptions. Then enter L2.
5713  */
5714 static void single_step_guest(const char *test_name, u64 starting_dr6,
5715 			      u64 pending_debug_exceptions)
5716 {
5717 	printf("\n%s\n", test_name);
5718 	skip_exit_vmcall();
5719 	write_dr6(starting_dr6);
5720 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
5721 	if (pending_debug_exceptions) {
5722 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
5723 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
5724 	}
5725 	enter_guest();
5726 }
5727 
5728 /*
5729  * When L1 intercepts #DB, verify that a single-step trap clears
5730  * pending debug exceptions, populates the exit qualification field
5731  * properly, and that DR6 is not prematurely clobbered. In a
5732  * (simulated) MOVSS shadow, make sure that the pending debug
5733  * exception bits are properly accumulated into the exit qualification
5734  * field.
5735  */
5736 static void vmx_db_test(void)
5737 {
5738 	/*
5739 	 * We are going to set a few arbitrary bits in DR6 to verify that
5740 	 * (a) DR6 is not modified by an intercepted #DB, and
5741 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
5742          *     the exit qualification field for a subsequent #DB exception.
5743 	 */
5744 	const u64 starting_dr6 = DR6_RESERVED | BIT(13) | DR_TRAP3 | DR_TRAP1;
5745 	extern char post_nop asm(".Lpost_nop");
5746 	extern char post_movss_nop asm(".Lpost_movss_nop");
5747 	extern char post_wbinvd asm(".Lpost_wbinvd");
5748 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
5749 	extern char xbegin asm(".Lxbegin");
5750 	extern char skip_rtm asm(".Lskip_rtm");
5751 
5752 	/*
5753 	 * L1 wants to intercept #DB exceptions encountered in L2.
5754 	 */
5755 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
5756 
5757 	/*
5758 	 * Start L2 and run it up to the first point of interest.
5759 	 */
5760 	test_set_guest(vmx_db_test_guest);
5761 	enter_guest();
5762 
5763 	/*
5764 	 * Hardware-delivered #DB trap for single-step sets the
5765 	 * standard that L0 has to follow for emulated instructions.
5766 	 */
5767 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
5768 	check_db_exit(false, false, false, &post_nop, DR_STEP, starting_dr6);
5769 
5770 	/*
5771 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
5772 	 * also sets the standard that L0 has to follow for emulated
5773 	 * instructions. Here, we establish the VMCS pending debug
5774 	 * exceptions to indicate that the simulated MOVSS triggered a
5775 	 * data breakpoint as well as the single-step trap.
5776 	 */
5777 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
5778 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0 );
5779 	check_db_exit(false, false, false, &post_movss_nop, DR_STEP | DR_TRAP0,
5780 		      starting_dr6);
5781 
5782 	/*
5783 	 * L0 synthesized #DB trap for single-step is buggy, because
5784 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
5785 	 * reconstitute the exit qualification from the prematurely
5786 	 * modified DR6, but fails miserably.
5787 	 */
5788 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
5789 	check_db_exit(true, true, false, &post_wbinvd, DR_STEP, starting_dr6);
5790 
5791 	/*
5792 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
5793 	 * even worse, because L0 also leaves the pending debug
5794 	 * exceptions in the VMCS instead of accumulating them into
5795 	 * the exit qualification field for the #DB exception.
5796 	 */
5797 	single_step_guest("Software synthesized single-step in MOVSS shadow",
5798 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0);
5799 	check_db_exit(true, true, true, &post_movss_wbinvd, DR_STEP | DR_TRAP0,
5800 		      starting_dr6);
5801 
5802 	/*
5803 	 * Optional RTM test for hardware that supports RTM, to
5804 	 * demonstrate that the current volume 3 of the SDM
5805 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
5806 	 * qualification for debug exceptions is not reserved. It is
5807 	 * set to 1 if a debug exception (#DB) or a breakpoint
5808 	 * exception (#BP) occurs inside an RTM region while advanced
5809 	 * debugging of RTM transactional regions is enabled.
5810 	 */
5811 	if (cpuid(7).b & BIT(11)) {
5812 		vmcs_write(ENT_CONTROLS,
5813 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
5814 		/*
5815 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
5816 		 * in the guest to enable advanced debugging of RTM
5817 		 * transactional regions.
5818 		 */
5819 		vmcs_write(GUEST_DR7, BIT(11));
5820 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
5821 		single_step_guest("Hardware delivered single-step in "
5822 				  "transactional region", starting_dr6, 0);
5823 		check_db_exit(false, false, false, &xbegin, BIT(16),
5824 			      starting_dr6);
5825 	} else {
5826 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
5827 		enter_guest();
5828 	}
5829 }
5830 
5831 static bool cpu_has_apicv(void)
5832 {
5833 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
5834 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
5835 		(ctrl_pin_rev.clr & PIN_POST_INTR));
5836 }
5837 
5838 static void enable_vid(void)
5839 {
5840 	void *virtual_apic_page;
5841 
5842 	assert(cpu_has_apicv());
5843 
5844 	disable_intercept_for_x2apic_msrs();
5845 
5846 	virtual_apic_page = alloc_page();
5847 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
5848 
5849 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
5850 
5851 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
5852 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
5853 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
5854 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
5855 
5856 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
5857 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
5858 }
5859 
5860 static void trigger_ioapic_scan_thread(void *data)
5861 {
5862 	/* Wait until other CPU entered L2 */
5863 	while (vmx_get_test_stage() != 1)
5864 		;
5865 
5866 	/* Trigger ioapic scan */
5867 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
5868 	vmx_set_test_stage(2);
5869 }
5870 
5871 static void irq_79_handler_guest(isr_regs_t *regs)
5872 {
5873 	eoi();
5874 
5875 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
5876 	vmcall();
5877 }
5878 
5879 /*
5880  * Constant for num of busy-loop iterations after which
5881  * a timer interrupt should have happened in host
5882  */
5883 #define TIMER_INTERRUPT_DELAY 100000000
5884 
5885 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
5886 {
5887 	handle_irq(0x79, irq_79_handler_guest);
5888 	irq_enable();
5889 
5890 	/* Signal to L1 CPU to trigger ioapic scan */
5891 	vmx_set_test_stage(1);
5892 	/* Wait until L1 CPU to trigger ioapic scan */
5893 	while (vmx_get_test_stage() != 2)
5894 		;
5895 
5896 	/*
5897 	 * Wait for L0 timer interrupt to be raised while we run in L2
5898 	 * such that L0 will process the IOAPIC scan request before
5899 	 * resuming L2
5900 	 */
5901 	delay(TIMER_INTERRUPT_DELAY);
5902 
5903 	asm volatile ("int $0x79");
5904 }
5905 
5906 static void vmx_eoi_bitmap_ioapic_scan_test(void)
5907 {
5908 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
5909 		report_skip(__func__);
5910 		return;
5911 	}
5912 
5913 	enable_vid();
5914 
5915 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
5916 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
5917 
5918 	/*
5919 	 * Launch L2.
5920 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
5921 	 * In case the reason isn't VMX_VMCALL, the asserion inside
5922 	 * skip_exit_vmcall() will fail.
5923 	 */
5924 	enter_guest();
5925 	skip_exit_vmcall();
5926 
5927 	/* Let L2 finish */
5928 	enter_guest();
5929 	report(__func__, 1);
5930 }
5931 
5932 #define HLT_WITH_RVI_VECTOR		(0xf1)
5933 
5934 bool vmx_hlt_with_rvi_guest_isr_fired;
5935 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
5936 {
5937 	vmx_hlt_with_rvi_guest_isr_fired = true;
5938 	eoi();
5939 }
5940 
5941 static void vmx_hlt_with_rvi_guest(void)
5942 {
5943 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
5944 
5945 	irq_enable();
5946 	asm volatile ("nop");
5947 
5948 	vmcall();
5949 }
5950 
5951 static void vmx_hlt_with_rvi_test(void)
5952 {
5953 	if (!cpu_has_apicv()) {
5954 		report_skip(__func__);
5955 		return;
5956 	}
5957 
5958 	enable_vid();
5959 
5960 	vmx_hlt_with_rvi_guest_isr_fired = false;
5961 	test_set_guest(vmx_hlt_with_rvi_guest);
5962 
5963 	enter_guest();
5964 	skip_exit_vmcall();
5965 
5966 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
5967 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
5968 	enter_guest();
5969 
5970 	report("Interrupt raised in guest", vmx_hlt_with_rvi_guest_isr_fired);
5971 }
5972 
5973 static void set_irq_line_thread(void *data)
5974 {
5975 	/* Wait until other CPU entered L2 */
5976 	while (vmx_get_test_stage() != 1)
5977 		;
5978 
5979 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
5980 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
5981 	vmx_set_test_stage(2);
5982 }
5983 
5984 static bool irq_78_handler_vmcall_before_eoi;
5985 static void irq_78_handler_guest(isr_regs_t *regs)
5986 {
5987 	set_irq_line(0xf, 0);
5988 	if (irq_78_handler_vmcall_before_eoi)
5989 		vmcall();
5990 	eoi();
5991 	vmcall();
5992 }
5993 
5994 static void vmx_apic_passthrough_guest(void)
5995 {
5996 	handle_irq(0x78, irq_78_handler_guest);
5997 	irq_enable();
5998 
5999 	/* If requested, wait for other CPU to trigger ioapic scan */
6000 	if (vmx_get_test_stage() < 1) {
6001 		vmx_set_test_stage(1);
6002 		while (vmx_get_test_stage() != 2)
6003 			;
6004 	}
6005 
6006 	set_irq_line(0xf, 1);
6007 }
6008 
6009 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
6010 {
6011 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
6012 		report_skip(__func__);
6013 		return;
6014 	}
6015 
6016 	u64 cpu_ctrl_0 = CPU_SECONDARY;
6017 	u64 cpu_ctrl_1 = 0;
6018 
6019 	disable_intercept_for_x2apic_msrs();
6020 
6021 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
6022 
6023 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
6024 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
6025 
6026 	if (set_irq_line_from_thread) {
6027 		irq_78_handler_vmcall_before_eoi = false;
6028 		on_cpu_async(1, set_irq_line_thread, NULL);
6029 	} else {
6030 		irq_78_handler_vmcall_before_eoi = true;
6031 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
6032 		vmx_set_test_stage(2);
6033 	}
6034 	test_set_guest(vmx_apic_passthrough_guest);
6035 
6036 	if (irq_78_handler_vmcall_before_eoi) {
6037 		/* Before EOI remote_irr should still be set */
6038 		enter_guest();
6039 		skip_exit_vmcall();
6040 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
6041 			"IOAPIC pass-through: remote_irr=1 before EOI");
6042 	}
6043 
6044 	/* After EOI remote_irr should be cleared */
6045 	enter_guest();
6046 	skip_exit_vmcall();
6047 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
6048 		"IOAPIC pass-through: remote_irr=0 after EOI");
6049 
6050 	/* Let L2 finish */
6051 	enter_guest();
6052 	report(__func__, 1);
6053 }
6054 
6055 static void vmx_apic_passthrough_test(void)
6056 {
6057 	vmx_apic_passthrough(false);
6058 }
6059 
6060 static void vmx_apic_passthrough_thread_test(void)
6061 {
6062 	vmx_apic_passthrough(true);
6063 }
6064 
6065 enum vmcs_access {
6066 	ACCESS_VMREAD,
6067 	ACCESS_VMWRITE,
6068 	ACCESS_NONE,
6069 };
6070 
6071 struct vmcs_shadow_test_common {
6072 	enum vmcs_access op;
6073 	enum Reason reason;
6074 	u64 field;
6075 	u64 value;
6076 	u64 flags;
6077 	u64 time;
6078 } l1_l2_common;
6079 
6080 static inline u64 vmread_flags(u64 field, u64 *val)
6081 {
6082 	u64 flags;
6083 
6084 	asm volatile ("vmread %2, %1; pushf; pop %0"
6085 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
6086 	return flags & X86_EFLAGS_ALU;
6087 }
6088 
6089 static inline u64 vmwrite_flags(u64 field, u64 val)
6090 {
6091 	u64 flags;
6092 
6093 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
6094 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
6095 	return flags & X86_EFLAGS_ALU;
6096 }
6097 
6098 static void vmx_vmcs_shadow_test_guest(void)
6099 {
6100 	struct vmcs_shadow_test_common *c = &l1_l2_common;
6101 	u64 start;
6102 
6103 	while (c->op != ACCESS_NONE) {
6104 		start = rdtsc();
6105 		switch (c->op) {
6106 		default:
6107 			c->flags = -1ull;
6108 			break;
6109 		case ACCESS_VMREAD:
6110 			c->flags = vmread_flags(c->field, &c->value);
6111 			break;
6112 		case ACCESS_VMWRITE:
6113 			c->flags = vmwrite_flags(c->field, 0);
6114 			break;
6115 		}
6116 		c->time = rdtsc() - start;
6117 		vmcall();
6118 	}
6119 }
6120 
6121 static u64 vmread_from_shadow(u64 field)
6122 {
6123 	struct vmcs *primary;
6124 	struct vmcs *shadow;
6125 	u64 value;
6126 
6127 	TEST_ASSERT(!vmcs_save(&primary));
6128 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
6129 	TEST_ASSERT(!make_vmcs_current(shadow));
6130 	value = vmcs_read(field);
6131 	TEST_ASSERT(!make_vmcs_current(primary));
6132 	return value;
6133 }
6134 
6135 static u64 vmwrite_to_shadow(u64 field, u64 value)
6136 {
6137 	struct vmcs *primary;
6138 	struct vmcs *shadow;
6139 
6140 	TEST_ASSERT(!vmcs_save(&primary));
6141 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
6142 	TEST_ASSERT(!make_vmcs_current(shadow));
6143 	vmcs_write(field, value);
6144 	value = vmcs_read(field);
6145 	TEST_ASSERT(!make_vmcs_current(primary));
6146 	return value;
6147 }
6148 
6149 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
6150 {
6151 	struct vmcs_shadow_test_common *c = &l1_l2_common;
6152 
6153 	c->op = access;
6154 	vmcs_write(VMX_INST_ERROR, 0);
6155 	enter_guest();
6156 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
6157 	if (c->reason != VMX_VMCALL) {
6158 		skip_exit_insn();
6159 		enter_guest();
6160 	}
6161 	skip_exit_vmcall();
6162 }
6163 
6164 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
6165 {
6166 	struct vmcs_shadow_test_common *c = &l1_l2_common;
6167 	struct vmcs *shadow;
6168 	u64 value;
6169 	uintptr_t flags[2];
6170 	bool good_shadow;
6171 	u32 vmx_inst_error;
6172 
6173 	report_prefix_pushf("field %lx", field);
6174 	c->field = field;
6175 
6176 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
6177 	if (shadow != (struct vmcs *)-1ull) {
6178 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
6179 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
6180 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
6181 	} else {
6182 		/*
6183 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
6184 		 * shadowed-fields should fail with setting RFLAGS.CF.
6185 		 */
6186 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
6187 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
6188 		good_shadow = false;
6189 	}
6190 
6191 	/* Intercept both VMREAD and VMWRITE. */
6192 	report_prefix_push("no VMREAD/VMWRITE permission");
6193 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
6194 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
6195 		set_bit(field, bitmap[ACCESS_VMREAD]);
6196 		set_bit(field, bitmap[ACCESS_VMWRITE]);
6197 	}
6198 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
6199 	report("not shadowed for VMWRITE", c->reason == VMX_VMWRITE);
6200 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
6201 	report("not shadowed for VMREAD", c->reason == VMX_VMREAD);
6202 	report_prefix_pop();
6203 
6204 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
6205 		goto out;
6206 
6207 	/* Permit shadowed VMREAD. */
6208 	report_prefix_push("VMREAD permission only");
6209 	clear_bit(field, bitmap[ACCESS_VMREAD]);
6210 	set_bit(field, bitmap[ACCESS_VMWRITE]);
6211 	if (good_shadow)
6212 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
6213 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
6214 	report("not shadowed for VMWRITE", c->reason == VMX_VMWRITE);
6215 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
6216 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
6217 	report("shadowed for VMREAD (in %ld cycles)", c->reason == VMX_VMCALL,
6218 	       c->time);
6219 	report("ALU flags after VMREAD (%lx) are as expected (%lx)",
6220 	       c->flags == flags[ACCESS_VMREAD],
6221 	       c->flags, flags[ACCESS_VMREAD]);
6222 	if (good_shadow)
6223 		report("value read from shadow (%lx) is as expected (%lx)",
6224 		       c->value == value, c->value, value);
6225 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
6226 		report("VMX_INST_ERROR (%d) is as expected (%d)",
6227 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
6228 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6229 	report_prefix_pop();
6230 
6231 	/* Permit shadowed VMWRITE. */
6232 	report_prefix_push("VMWRITE permission only");
6233 	set_bit(field, bitmap[ACCESS_VMREAD]);
6234 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
6235 	if (good_shadow)
6236 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
6237 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
6238 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
6239 	report("shadowed for VMWRITE (in %ld cycles)", c->reason == VMX_VMCALL,
6240 		c->time);
6241 	report("ALU flags after VMWRITE (%lx) are as expected (%lx)",
6242 	       c->flags == flags[ACCESS_VMREAD],
6243 	       c->flags, flags[ACCESS_VMREAD]);
6244 	if (good_shadow) {
6245 		value = vmread_from_shadow(field);
6246 		report("shadow VMCS value (%lx) is as expected (%lx)",
6247 		       value == 0, value, 0ul);
6248 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
6249 		report("VMX_INST_ERROR (%d) is as expected (%d)",
6250 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
6251 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6252 	}
6253 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
6254 	report("not shadowed for VMREAD", c->reason == VMX_VMREAD);
6255 	report_prefix_pop();
6256 
6257 	/* Permit shadowed VMREAD and VMWRITE. */
6258 	report_prefix_push("VMREAD and VMWRITE permission");
6259 	clear_bit(field, bitmap[ACCESS_VMREAD]);
6260 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
6261 	if (good_shadow)
6262 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
6263 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
6264 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
6265 	report("shadowed for VMWRITE (in %ld cycles)", c->reason == VMX_VMCALL,
6266 		c->time);
6267 	report("ALU flags after VMWRITE (%lx) are as expected (%lx)",
6268 	       c->flags == flags[ACCESS_VMREAD],
6269 	       c->flags, flags[ACCESS_VMREAD]);
6270 	if (good_shadow) {
6271 		value = vmread_from_shadow(field);
6272 		report("shadow VMCS value (%lx) is as expected (%lx)",
6273 		       value == 0, value, 0ul);
6274 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
6275 		report("VMX_INST_ERROR (%d) is as expected (%d)",
6276 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
6277 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6278 	}
6279 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
6280 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
6281 	report("shadowed for VMREAD (in %ld cycles)", c->reason == VMX_VMCALL,
6282 	       c->time);
6283 	report("ALU flags after VMREAD (%lx) are as expected (%lx)",
6284 	       c->flags == flags[ACCESS_VMREAD],
6285 	       c->flags, flags[ACCESS_VMREAD]);
6286 	if (good_shadow)
6287 		report("value read from shadow (%lx) is as expected (%lx)",
6288 		       c->value == 0, c->value, 0ul);
6289 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
6290 		report("VMX_INST_ERROR (%d) is as expected (%d)",
6291 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
6292 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6293 	report_prefix_pop();
6294 
6295 out:
6296 	report_prefix_pop();
6297 }
6298 
6299 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
6300 {
6301 	unsigned base;
6302 	unsigned index;
6303 	unsigned bit;
6304 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
6305 
6306 	/* Run test on all possible valid VMCS fields */
6307 	for (base = 0;
6308 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
6309 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
6310 		for (index = 0; index <= highest_index; index++)
6311 			vmcs_shadow_test_field(bitmap, base + index);
6312 
6313 	/*
6314 	 * Run tests on some invalid VMCS fields
6315 	 * (Have reserved bit set).
6316 	 */
6317 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
6318 		vmcs_shadow_test_field(bitmap, (1ull << bit));
6319 }
6320 
6321 static void vmx_vmcs_shadow_test(void)
6322 {
6323 	u8 *bitmap[2];
6324 	struct vmcs *shadow;
6325 
6326 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
6327 		printf("\t'Activate secondary controls' not supported.\n");
6328 		return;
6329 	}
6330 
6331 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
6332 		printf("\t'VMCS shadowing' not supported.\n");
6333 		return;
6334 	}
6335 
6336 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
6337 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
6338 		printf("\tVMWRITE can't modify VM-exit information fields.\n");
6339 		return;
6340 	}
6341 
6342 	test_set_guest(vmx_vmcs_shadow_test_guest);
6343 
6344 	bitmap[ACCESS_VMREAD] = alloc_page();
6345 	bitmap[ACCESS_VMWRITE] = alloc_page();
6346 
6347 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
6348 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
6349 
6350 	shadow = alloc_page();
6351 	shadow->hdr.revision_id = basic.revision;
6352 	shadow->hdr.shadow_vmcs = 1;
6353 	TEST_ASSERT(!vmcs_clear(shadow));
6354 
6355 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
6356 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
6357 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
6358 
6359 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
6360 	report_prefix_push("valid link pointer");
6361 	vmx_vmcs_shadow_test_body(bitmap);
6362 	report_prefix_pop();
6363 
6364 	vmcs_write(VMCS_LINK_PTR, -1ull);
6365 	report_prefix_push("invalid link pointer");
6366 	vmx_vmcs_shadow_test_body(bitmap);
6367 	report_prefix_pop();
6368 
6369 	l1_l2_common.op = ACCESS_NONE;
6370 	enter_guest();
6371 }
6372 
6373 
6374 
6375 static int invalid_msr_init(struct vmcs *vmcs)
6376 {
6377 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
6378 		printf("\tPreemption timer is not supported\n");
6379 		return VMX_TEST_EXIT;
6380 	}
6381 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
6382 	preempt_val = 10000000;
6383 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
6384 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
6385 
6386 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
6387 		printf("\tSave preemption value is not supported\n");
6388 
6389 	vmcs_write(ENT_MSR_LD_CNT, 1);
6390 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
6391 
6392 	return VMX_TEST_START;
6393 }
6394 
6395 
6396 static void invalid_msr_main(void)
6397 {
6398 	report("Invalid MSR load", 0);
6399 }
6400 
6401 static int invalid_msr_exit_handler(void)
6402 {
6403 	report("Invalid MSR load", 0);
6404 	print_vmexit_info();
6405 	return VMX_TEST_EXIT;
6406 }
6407 
6408 static int invalid_msr_entry_failure(struct vmentry_failure *failure)
6409 {
6410 	ulong reason;
6411 
6412 	reason = vmcs_read(EXI_REASON);
6413 	report("Invalid MSR load", reason == (0x80000000u | VMX_FAIL_MSR));
6414 	return VMX_TEST_VMEXIT;
6415 }
6416 
6417 
6418 #define TEST(name) { #name, .v2 = name }
6419 
6420 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
6421 struct vmx_test vmx_tests[] = {
6422 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
6423 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
6424 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
6425 		preemption_timer_exit_handler, NULL, {0} },
6426 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
6427 		test_ctrl_pat_exit_handler, NULL, {0} },
6428 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
6429 		test_ctrl_efer_exit_handler, NULL, {0} },
6430 	{ "CR shadowing", NULL, cr_shadowing_main,
6431 		cr_shadowing_exit_handler, NULL, {0} },
6432 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
6433 		NULL, {0} },
6434 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
6435 		insn_intercept_exit_handler, NULL, {0} },
6436 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
6437 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
6438 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
6439 	{ "VPID", vpid_init, vpid_main, vpid_exit_handler, NULL, {0} },
6440 	{ "interrupt", interrupt_init, interrupt_main,
6441 		interrupt_exit_handler, NULL, {0} },
6442 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
6443 		NULL, {0} },
6444 	{ "MSR switch", msr_switch_init, msr_switch_main,
6445 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
6446 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
6447 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
6448 		disable_rdtscp_exit_handler, NULL, {0} },
6449 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
6450 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
6451 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
6452 		exit_monitor_from_l2_handler, NULL, {0} },
6453 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
6454 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
6455 	/* Basic V2 tests. */
6456 	TEST(v2_null_test),
6457 	TEST(v2_multiple_entries_test),
6458 	TEST(fixture_test_case1),
6459 	TEST(fixture_test_case2),
6460 	/* Opcode tests. */
6461 	TEST(invvpid_test_v2),
6462 	/* VM-entry tests */
6463 	TEST(vmx_controls_test),
6464 	TEST(vmentry_movss_shadow_test),
6465 	/* APICv tests */
6466 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
6467 	TEST(vmx_hlt_with_rvi_test),
6468 	/* APIC pass-through tests */
6469 	TEST(vmx_apic_passthrough_test),
6470 	TEST(vmx_apic_passthrough_thread_test),
6471 	/* VMCS Shadowing tests */
6472 	TEST(vmx_vmcs_shadow_test),
6473 	/* Regression tests */
6474 	TEST(vmx_cr_load_test),
6475 	TEST(vmx_nm_test),
6476 	TEST(vmx_db_test),
6477 	TEST(vmx_nmi_window_test),
6478 	TEST(vmx_intr_window_test),
6479 	TEST(vmx_pending_event_test),
6480 	TEST(vmx_pending_event_hlt_test),
6481 	TEST(vmx_store_tsc_test),
6482 	/* EPT access tests. */
6483 	TEST(ept_access_test_not_present),
6484 	TEST(ept_access_test_read_only),
6485 	TEST(ept_access_test_write_only),
6486 	TEST(ept_access_test_read_write),
6487 	TEST(ept_access_test_execute_only),
6488 	TEST(ept_access_test_read_execute),
6489 	TEST(ept_access_test_write_execute),
6490 	TEST(ept_access_test_read_write_execute),
6491 	TEST(ept_access_test_reserved_bits),
6492 	TEST(ept_access_test_ignored_bits),
6493 	TEST(ept_access_test_paddr_not_present_ad_disabled),
6494 	TEST(ept_access_test_paddr_not_present_ad_enabled),
6495 	TEST(ept_access_test_paddr_read_only_ad_disabled),
6496 	TEST(ept_access_test_paddr_read_only_ad_enabled),
6497 	TEST(ept_access_test_paddr_read_write),
6498 	TEST(ept_access_test_paddr_read_write_execute),
6499 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
6500 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
6501 	TEST(ept_access_test_paddr_not_present_page_fault),
6502 	TEST(ept_access_test_force_2m_page),
6503 	{ NULL, NULL, NULL, NULL, NULL, {0} },
6504 };
6505