xref: /kvm-unit-tests/x86/vmx_tests.c (revision 84a43d344a5c70df847521b2a4e1a736ca97d922)
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(1, "Basic VMX test");
59 }
60 
61 static int basic_exit_handler(union exit_reason exit_reason)
62 {
63 	report(0, "Basic VMX test");
64 	print_vmexit_info(exit_reason);
65 	return VMX_TEST_EXIT;
66 }
67 
68 static void vmenter_main(void)
69 {
70 	u64 rax;
71 	u64 rsp, resume_rsp;
72 
73 	report(1, "test vmlaunch");
74 
75 	asm volatile(
76 		"mov %%rsp, %0\n\t"
77 		"mov %3, %%rax\n\t"
78 		"vmcall\n\t"
79 		"mov %%rax, %1\n\t"
80 		"mov %%rsp, %2\n\t"
81 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
82 		: "g"(0xABCD));
83 	report((rax == 0xFFFF) && (rsp == resume_rsp), "test vmresume");
84 }
85 
86 static int vmenter_exit_handler(union exit_reason exit_reason)
87 {
88 	u64 guest_rip = vmcs_read(GUEST_RIP);
89 
90 	switch (exit_reason.basic) {
91 	case VMX_VMCALL:
92 		if (regs.rax != 0xABCD) {
93 			report(0, "test vmresume");
94 			return VMX_TEST_VMEXIT;
95 		}
96 		regs.rax = 0xFFFF;
97 		vmcs_write(GUEST_RIP, guest_rip + 3);
98 		return VMX_TEST_RESUME;
99 	default:
100 		report(0, "test vmresume");
101 		print_vmexit_info(exit_reason);
102 	}
103 	return VMX_TEST_VMEXIT;
104 }
105 
106 u32 preempt_scale;
107 volatile unsigned long long tsc_val;
108 volatile u32 preempt_val;
109 u64 saved_rip;
110 
111 static int preemption_timer_init(struct vmcs *vmcs)
112 {
113 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
114 		printf("\tPreemption timer is not supported\n");
115 		return VMX_TEST_EXIT;
116 	}
117 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
118 	preempt_val = 10000000;
119 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
120 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
121 
122 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
123 		printf("\tSave preemption value is not supported\n");
124 
125 	return VMX_TEST_START;
126 }
127 
128 static void preemption_timer_main(void)
129 {
130 	tsc_val = rdtsc();
131 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
132 		vmx_set_test_stage(0);
133 		vmcall();
134 		if (vmx_get_test_stage() == 1)
135 			vmcall();
136 	}
137 	vmx_set_test_stage(1);
138 	while (vmx_get_test_stage() == 1) {
139 		if (((rdtsc() - tsc_val) >> preempt_scale)
140 				> 10 * preempt_val) {
141 			vmx_set_test_stage(2);
142 			vmcall();
143 		}
144 	}
145 	tsc_val = rdtsc();
146 	asm volatile ("hlt");
147 	vmcall();
148 	vmx_set_test_stage(5);
149 	vmcall();
150 }
151 
152 static int preemption_timer_exit_handler(union exit_reason exit_reason)
153 {
154 	bool guest_halted;
155 	u64 guest_rip;
156 	u32 insn_len;
157 	u32 ctrl_exit;
158 
159 	guest_rip = vmcs_read(GUEST_RIP);
160 	insn_len = vmcs_read(EXI_INST_LEN);
161 	switch (exit_reason.basic) {
162 	case VMX_PREEMPT:
163 		switch (vmx_get_test_stage()) {
164 		case 1:
165 		case 2:
166 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val,
167 			       "busy-wait for preemption timer");
168 			vmx_set_test_stage(3);
169 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
170 			return VMX_TEST_RESUME;
171 		case 3:
172 			guest_halted =
173 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
174 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val
175 			        && guest_halted,
176 			       "preemption timer during hlt");
177 			vmx_set_test_stage(4);
178 			vmcs_write(PIN_CONTROLS,
179 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
180 			vmcs_write(EXI_CONTROLS,
181 				   vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT);
182 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
183 			return VMX_TEST_RESUME;
184 		case 4:
185 			report(saved_rip == guest_rip,
186 			       "preemption timer with 0 value");
187 			break;
188 		default:
189 			report(false, "Invalid stage.");
190 			print_vmexit_info(exit_reason);
191 			break;
192 		}
193 		break;
194 	case VMX_VMCALL:
195 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
196 		switch (vmx_get_test_stage()) {
197 		case 0:
198 			report(vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val,
199 			       "Keep preemption value");
200 			vmx_set_test_stage(1);
201 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
202 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
203 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
204 			vmcs_write(EXI_CONTROLS, ctrl_exit);
205 			return VMX_TEST_RESUME;
206 		case 1:
207 			report(vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val,
208 			       "Save preemption value");
209 			return VMX_TEST_RESUME;
210 		case 2:
211 			report(0, "busy-wait for preemption timer");
212 			vmx_set_test_stage(3);
213 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
214 			return VMX_TEST_RESUME;
215 		case 3:
216 			report(0, "preemption timer during hlt");
217 			vmx_set_test_stage(4);
218 			/* fall through */
219 		case 4:
220 			vmcs_write(PIN_CONTROLS,
221 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
222 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
223 			saved_rip = guest_rip + insn_len;
224 			return VMX_TEST_RESUME;
225 		case 5:
226 			report(0,
227 			       "preemption timer with 0 value (vmcall stage 5)");
228 			break;
229 		default:
230 			// Should not reach here
231 			report(false, "unexpected stage, %d",
232 			       vmx_get_test_stage());
233 			print_vmexit_info(exit_reason);
234 			return VMX_TEST_VMEXIT;
235 		}
236 		break;
237 	default:
238 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
239 		print_vmexit_info(exit_reason);
240 	}
241 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
242 	return VMX_TEST_VMEXIT;
243 }
244 
245 static void msr_bmp_init(void)
246 {
247 	void *msr_bitmap;
248 	u32 ctrl_cpu0;
249 
250 	msr_bitmap = alloc_page();
251 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
252 	ctrl_cpu0 |= CPU_MSR_BITMAP;
253 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
254 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
255 }
256 
257 static void *get_msr_bitmap(void)
258 {
259 	void *msr_bitmap;
260 
261 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
262 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
263 	} else {
264 		msr_bitmap = alloc_page();
265 		memset(msr_bitmap, 0xff, PAGE_SIZE);
266 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
267 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
268 	}
269 
270 	return msr_bitmap;
271 }
272 
273 static void disable_intercept_for_x2apic_msrs(void)
274 {
275 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
276 	u32 msr;
277 
278 	for (msr = APIC_BASE_MSR;
279 		 msr < (APIC_BASE_MSR+0xff);
280 		 msr += BITS_PER_LONG) {
281 		unsigned int word = msr / BITS_PER_LONG;
282 
283 		msr_bitmap[word] = 0;
284 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
285 	}
286 }
287 
288 static int test_ctrl_pat_init(struct vmcs *vmcs)
289 {
290 	u64 ctrl_ent;
291 	u64 ctrl_exi;
292 
293 	msr_bmp_init();
294 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
295 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
296 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
297 		printf("\tSave/load PAT is not supported\n");
298 		return 1;
299 	}
300 
301 	ctrl_ent = vmcs_read(ENT_CONTROLS);
302 	ctrl_exi = vmcs_read(EXI_CONTROLS);
303 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
304 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
305 	vmcs_write(ENT_CONTROLS, ctrl_ent);
306 	vmcs_write(EXI_CONTROLS, ctrl_exi);
307 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
308 	vmcs_write(GUEST_PAT, 0x0);
309 	vmcs_write(HOST_PAT, ia32_pat);
310 	return VMX_TEST_START;
311 }
312 
313 static void test_ctrl_pat_main(void)
314 {
315 	u64 guest_ia32_pat;
316 
317 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
318 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
319 		printf("\tENT_LOAD_PAT is not supported.\n");
320 	else {
321 		if (guest_ia32_pat != 0) {
322 			report(0, "Entry load PAT");
323 			return;
324 		}
325 	}
326 	wrmsr(MSR_IA32_CR_PAT, 0x6);
327 	vmcall();
328 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
329 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
330 		report(guest_ia32_pat == ia32_pat, "Entry load PAT");
331 }
332 
333 static int test_ctrl_pat_exit_handler(union exit_reason exit_reason)
334 {
335 	u64 guest_rip;
336 	u64 guest_pat;
337 
338 	guest_rip = vmcs_read(GUEST_RIP);
339 	switch (exit_reason.basic) {
340 	case VMX_VMCALL:
341 		guest_pat = vmcs_read(GUEST_PAT);
342 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
343 			printf("\tEXI_SAVE_PAT is not supported\n");
344 			vmcs_write(GUEST_PAT, 0x6);
345 		} else {
346 			report(guest_pat == 0x6, "Exit save PAT");
347 		}
348 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
349 			printf("\tEXI_LOAD_PAT is not supported\n");
350 		else
351 			report(rdmsr(MSR_IA32_CR_PAT) == ia32_pat,
352 			       "Exit load PAT");
353 		vmcs_write(GUEST_PAT, ia32_pat);
354 		vmcs_write(GUEST_RIP, guest_rip + 3);
355 		return VMX_TEST_RESUME;
356 	default:
357 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
358 		break;
359 	}
360 	return VMX_TEST_VMEXIT;
361 }
362 
363 static int test_ctrl_efer_init(struct vmcs *vmcs)
364 {
365 	u64 ctrl_ent;
366 	u64 ctrl_exi;
367 
368 	msr_bmp_init();
369 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
370 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
371 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
372 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
373 	ia32_efer = rdmsr(MSR_EFER);
374 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
375 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
376 	return VMX_TEST_START;
377 }
378 
379 static void test_ctrl_efer_main(void)
380 {
381 	u64 guest_ia32_efer;
382 
383 	guest_ia32_efer = rdmsr(MSR_EFER);
384 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
385 		printf("\tENT_LOAD_EFER is not supported.\n");
386 	else {
387 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
388 			report(0, "Entry load EFER");
389 			return;
390 		}
391 	}
392 	wrmsr(MSR_EFER, ia32_efer);
393 	vmcall();
394 	guest_ia32_efer = rdmsr(MSR_EFER);
395 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
396 		report(guest_ia32_efer == ia32_efer, "Entry load EFER");
397 }
398 
399 static int test_ctrl_efer_exit_handler(union exit_reason exit_reason)
400 {
401 	u64 guest_rip;
402 	u64 guest_efer;
403 
404 	guest_rip = vmcs_read(GUEST_RIP);
405 	switch (exit_reason.basic) {
406 	case VMX_VMCALL:
407 		guest_efer = vmcs_read(GUEST_EFER);
408 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
409 			printf("\tEXI_SAVE_EFER is not supported\n");
410 			vmcs_write(GUEST_EFER, ia32_efer);
411 		} else {
412 			report(guest_efer == ia32_efer, "Exit save EFER");
413 		}
414 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
415 			printf("\tEXI_LOAD_EFER is not supported\n");
416 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
417 		} else {
418 			report(rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX),
419 			       "Exit load EFER");
420 		}
421 		vmcs_write(GUEST_PAT, ia32_efer);
422 		vmcs_write(GUEST_RIP, guest_rip + 3);
423 		return VMX_TEST_RESUME;
424 	default:
425 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
426 		break;
427 	}
428 	return VMX_TEST_VMEXIT;
429 }
430 
431 u32 guest_cr0, guest_cr4;
432 
433 static void cr_shadowing_main(void)
434 {
435 	u32 cr0, cr4, tmp;
436 
437 	// Test read through
438 	vmx_set_test_stage(0);
439 	guest_cr0 = read_cr0();
440 	if (vmx_get_test_stage() == 1)
441 		report(0, "Read through CR0");
442 	else
443 		vmcall();
444 	vmx_set_test_stage(1);
445 	guest_cr4 = read_cr4();
446 	if (vmx_get_test_stage() == 2)
447 		report(0, "Read through CR4");
448 	else
449 		vmcall();
450 	// Test write through
451 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
452 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
453 	vmx_set_test_stage(2);
454 	write_cr0(guest_cr0);
455 	if (vmx_get_test_stage() == 3)
456 		report(0, "Write throuth CR0");
457 	else
458 		vmcall();
459 	vmx_set_test_stage(3);
460 	write_cr4(guest_cr4);
461 	if (vmx_get_test_stage() == 4)
462 		report(0, "Write through CR4");
463 	else
464 		vmcall();
465 	// Test read shadow
466 	vmx_set_test_stage(4);
467 	vmcall();
468 	cr0 = read_cr0();
469 	if (vmx_get_test_stage() != 5)
470 		report(cr0 == guest_cr0, "Read shadowing CR0");
471 	vmx_set_test_stage(5);
472 	cr4 = read_cr4();
473 	if (vmx_get_test_stage() != 6)
474 		report(cr4 == guest_cr4, "Read shadowing CR4");
475 	// Test write shadow (same value with shadow)
476 	vmx_set_test_stage(6);
477 	write_cr0(guest_cr0);
478 	if (vmx_get_test_stage() == 7)
479 		report(0, "Write shadowing CR0 (same value with shadow)");
480 	else
481 		vmcall();
482 	vmx_set_test_stage(7);
483 	write_cr4(guest_cr4);
484 	if (vmx_get_test_stage() == 8)
485 		report(0, "Write shadowing CR4 (same value with shadow)");
486 	else
487 		vmcall();
488 	// Test write shadow (different value)
489 	vmx_set_test_stage(8);
490 	tmp = guest_cr0 ^ X86_CR0_TS;
491 	asm volatile("mov %0, %%rsi\n\t"
492 		"mov %%rsi, %%cr0\n\t"
493 		::"m"(tmp)
494 		:"rsi", "memory", "cc");
495 	report(vmx_get_test_stage() == 9,
496 	       "Write shadowing different X86_CR0_TS");
497 	vmx_set_test_stage(9);
498 	tmp = guest_cr0 ^ X86_CR0_MP;
499 	asm volatile("mov %0, %%rsi\n\t"
500 		"mov %%rsi, %%cr0\n\t"
501 		::"m"(tmp)
502 		:"rsi", "memory", "cc");
503 	report(vmx_get_test_stage() == 10,
504 	       "Write shadowing different X86_CR0_MP");
505 	vmx_set_test_stage(10);
506 	tmp = guest_cr4 ^ X86_CR4_TSD;
507 	asm volatile("mov %0, %%rsi\n\t"
508 		"mov %%rsi, %%cr4\n\t"
509 		::"m"(tmp)
510 		:"rsi", "memory", "cc");
511 	report(vmx_get_test_stage() == 11,
512 	       "Write shadowing different X86_CR4_TSD");
513 	vmx_set_test_stage(11);
514 	tmp = guest_cr4 ^ X86_CR4_DE;
515 	asm volatile("mov %0, %%rsi\n\t"
516 		"mov %%rsi, %%cr4\n\t"
517 		::"m"(tmp)
518 		:"rsi", "memory", "cc");
519 	report(vmx_get_test_stage() == 12,
520 	       "Write shadowing different X86_CR4_DE");
521 }
522 
523 static int cr_shadowing_exit_handler(union exit_reason exit_reason)
524 {
525 	u64 guest_rip;
526 	u32 insn_len;
527 	u32 exit_qual;
528 
529 	guest_rip = vmcs_read(GUEST_RIP);
530 	insn_len = vmcs_read(EXI_INST_LEN);
531 	exit_qual = vmcs_read(EXI_QUALIFICATION);
532 	switch (exit_reason.basic) {
533 	case VMX_VMCALL:
534 		switch (vmx_get_test_stage()) {
535 		case 0:
536 			report(guest_cr0 == vmcs_read(GUEST_CR0),
537 			       "Read through CR0");
538 			break;
539 		case 1:
540 			report(guest_cr4 == vmcs_read(GUEST_CR4),
541 			       "Read through CR4");
542 			break;
543 		case 2:
544 			report(guest_cr0 == vmcs_read(GUEST_CR0),
545 			       "Write through CR0");
546 			break;
547 		case 3:
548 			report(guest_cr4 == vmcs_read(GUEST_CR4),
549 			       "Write through CR4");
550 			break;
551 		case 4:
552 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
553 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
554 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
555 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
556 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
557 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
558 			break;
559 		case 6:
560 			report(guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)),
561 			       "Write shadowing CR0 (same value)");
562 			break;
563 		case 7:
564 			report(guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)),
565 			       "Write shadowing CR4 (same value)");
566 			break;
567 		default:
568 			// Should not reach here
569 			report(false, "unexpected stage, %d",
570 			       vmx_get_test_stage());
571 			print_vmexit_info(exit_reason);
572 			return VMX_TEST_VMEXIT;
573 		}
574 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
575 		return VMX_TEST_RESUME;
576 	case VMX_CR:
577 		switch (vmx_get_test_stage()) {
578 		case 4:
579 			report(0, "Read shadowing CR0");
580 			vmx_inc_test_stage();
581 			break;
582 		case 5:
583 			report(0, "Read shadowing CR4");
584 			vmx_inc_test_stage();
585 			break;
586 		case 6:
587 			report(0, "Write shadowing CR0 (same value)");
588 			vmx_inc_test_stage();
589 			break;
590 		case 7:
591 			report(0, "Write shadowing CR4 (same value)");
592 			vmx_inc_test_stage();
593 			break;
594 		case 8:
595 		case 9:
596 			// 0x600 encodes "mov %esi, %cr0"
597 			if (exit_qual == 0x600)
598 				vmx_inc_test_stage();
599 			break;
600 		case 10:
601 		case 11:
602 			// 0x604 encodes "mov %esi, %cr4"
603 			if (exit_qual == 0x604)
604 				vmx_inc_test_stage();
605 			break;
606 		default:
607 			// Should not reach here
608 			report(false, "unexpected stage, %d",
609 			       vmx_get_test_stage());
610 			print_vmexit_info(exit_reason);
611 			return VMX_TEST_VMEXIT;
612 		}
613 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
614 		return VMX_TEST_RESUME;
615 	default:
616 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
617 		print_vmexit_info(exit_reason);
618 	}
619 	return VMX_TEST_VMEXIT;
620 }
621 
622 static int iobmp_init(struct vmcs *vmcs)
623 {
624 	u32 ctrl_cpu0;
625 
626 	io_bitmap_a = alloc_page();
627 	io_bitmap_b = alloc_page();
628 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
629 	ctrl_cpu0 |= CPU_IO_BITMAP;
630 	ctrl_cpu0 &= (~CPU_IO);
631 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
632 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
633 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
634 	return VMX_TEST_START;
635 }
636 
637 static void iobmp_main(void)
638 {
639 	// stage 0, test IO pass
640 	vmx_set_test_stage(0);
641 	inb(0x5000);
642 	outb(0x0, 0x5000);
643 	report(vmx_get_test_stage() == 0, "I/O bitmap - I/O pass");
644 	// test IO width, in/out
645 	((u8 *)io_bitmap_a)[0] = 0xFF;
646 	vmx_set_test_stage(2);
647 	inb(0x0);
648 	report(vmx_get_test_stage() == 3, "I/O bitmap - trap in");
649 	vmx_set_test_stage(3);
650 	outw(0x0, 0x0);
651 	report(vmx_get_test_stage() == 4, "I/O bitmap - trap out");
652 	vmx_set_test_stage(4);
653 	inl(0x0);
654 	report(vmx_get_test_stage() == 5, "I/O bitmap - I/O width, long");
655 	// test low/high IO port
656 	vmx_set_test_stage(5);
657 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
658 	inb(0x5000);
659 	report(vmx_get_test_stage() == 6, "I/O bitmap - I/O port, low part");
660 	vmx_set_test_stage(6);
661 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
662 	inb(0x9000);
663 	report(vmx_get_test_stage() == 7, "I/O bitmap - I/O port, high part");
664 	// test partial pass
665 	vmx_set_test_stage(7);
666 	inl(0x4FFF);
667 	report(vmx_get_test_stage() == 8, "I/O bitmap - partial pass");
668 	// test overrun
669 	vmx_set_test_stage(8);
670 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
671 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
672 	inl(0xFFFF);
673 	report(vmx_get_test_stage() == 9, "I/O bitmap - overrun");
674 	vmx_set_test_stage(9);
675 	vmcall();
676 	outb(0x0, 0x0);
677 	report(vmx_get_test_stage() == 9,
678 	       "I/O bitmap - ignore unconditional exiting");
679 	vmx_set_test_stage(10);
680 	vmcall();
681 	outb(0x0, 0x0);
682 	report(vmx_get_test_stage() == 11,
683 	       "I/O bitmap - unconditional exiting");
684 }
685 
686 static int iobmp_exit_handler(union exit_reason exit_reason)
687 {
688 	u64 guest_rip;
689 	ulong exit_qual;
690 	u32 insn_len, ctrl_cpu0;
691 
692 	guest_rip = vmcs_read(GUEST_RIP);
693 	exit_qual = vmcs_read(EXI_QUALIFICATION);
694 	insn_len = vmcs_read(EXI_INST_LEN);
695 	switch (exit_reason.basic) {
696 	case VMX_IO:
697 		switch (vmx_get_test_stage()) {
698 		case 0:
699 		case 1:
700 			vmx_inc_test_stage();
701 			break;
702 		case 2:
703 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE,
704 			       "I/O bitmap - I/O width, byte");
705 			report(exit_qual & VMX_IO_IN,
706 			       "I/O bitmap - I/O direction, in");
707 			vmx_inc_test_stage();
708 			break;
709 		case 3:
710 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD,
711 			       "I/O bitmap - I/O width, word");
712 			report(!(exit_qual & VMX_IO_IN),
713 			       "I/O bitmap - I/O direction, out");
714 			vmx_inc_test_stage();
715 			break;
716 		case 4:
717 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG,
718 			       "I/O bitmap - I/O width, long");
719 			vmx_inc_test_stage();
720 			break;
721 		case 5:
722 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
723 				vmx_inc_test_stage();
724 			break;
725 		case 6:
726 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
727 				vmx_inc_test_stage();
728 			break;
729 		case 7:
730 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
731 				vmx_inc_test_stage();
732 			break;
733 		case 8:
734 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
735 				vmx_inc_test_stage();
736 			break;
737 		case 9:
738 		case 10:
739 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
740 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
741 			vmx_inc_test_stage();
742 			break;
743 		default:
744 			// Should not reach here
745 			report(false, "unexpected stage, %d",
746 			       vmx_get_test_stage());
747 			print_vmexit_info(exit_reason);
748 			return VMX_TEST_VMEXIT;
749 		}
750 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
751 		return VMX_TEST_RESUME;
752 	case VMX_VMCALL:
753 		switch (vmx_get_test_stage()) {
754 		case 9:
755 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
756 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
757 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
758 			break;
759 		case 10:
760 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
761 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
762 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
763 			break;
764 		default:
765 			// Should not reach here
766 			report(false, "unexpected stage, %d",
767 			       vmx_get_test_stage());
768 			print_vmexit_info(exit_reason);
769 			return VMX_TEST_VMEXIT;
770 		}
771 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
772 		return VMX_TEST_RESUME;
773 	default:
774 		printf("guest_rip = %#lx\n", guest_rip);
775 		printf("\tERROR : Unknown exit reason, 0x%x\n", exit_reason.full);
776 		break;
777 	}
778 	return VMX_TEST_VMEXIT;
779 }
780 
781 #define INSN_CPU0		0
782 #define INSN_CPU1		1
783 #define INSN_ALWAYS_TRAP	2
784 
785 #define FIELD_EXIT_QUAL		(1 << 0)
786 #define FIELD_INSN_INFO		(1 << 1)
787 
788 asm(
789 	"insn_hlt: hlt;ret\n\t"
790 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
791 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
792 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
793 	"insn_rdtsc: rdtsc;ret\n\t"
794 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
795 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
796 #ifdef __x86_64__
797 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
798 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
799 #endif
800 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
801 	"insn_pause: pause;ret\n\t"
802 	"insn_wbinvd: wbinvd;ret\n\t"
803 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
804 	"insn_invd: invd;ret\n\t"
805 	"insn_sgdt: sgdt gdt64_desc;ret\n\t"
806 	"insn_lgdt: lgdt gdt64_desc;ret\n\t"
807 	"insn_sidt: sidt idt_descr;ret\n\t"
808 	"insn_lidt: lidt idt_descr;ret\n\t"
809 	"insn_sldt: sldt %ax;ret\n\t"
810 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
811 	"insn_str: str %ax;ret\n\t"
812 	"insn_rdrand: rdrand %rax;ret\n\t"
813 	"insn_rdseed: rdseed %rax;ret\n\t"
814 );
815 extern void insn_hlt(void);
816 extern void insn_invlpg(void);
817 extern void insn_mwait(void);
818 extern void insn_rdpmc(void);
819 extern void insn_rdtsc(void);
820 extern void insn_cr3_load(void);
821 extern void insn_cr3_store(void);
822 #ifdef __x86_64__
823 extern void insn_cr8_load(void);
824 extern void insn_cr8_store(void);
825 #endif
826 extern void insn_monitor(void);
827 extern void insn_pause(void);
828 extern void insn_wbinvd(void);
829 extern void insn_sgdt(void);
830 extern void insn_lgdt(void);
831 extern void insn_sidt(void);
832 extern void insn_lidt(void);
833 extern void insn_sldt(void);
834 extern void insn_lldt(void);
835 extern void insn_str(void);
836 extern void insn_cpuid(void);
837 extern void insn_invd(void);
838 extern void insn_rdrand(void);
839 extern void insn_rdseed(void);
840 
841 u32 cur_insn;
842 u64 cr3;
843 
844 #define X86_FEATURE_MONITOR	(1 << 3)
845 
846 typedef bool (*supported_fn)(void);
847 
848 static bool monitor_supported(void)
849 {
850 	return this_cpu_has(X86_FEATURE_MWAIT);
851 }
852 
853 struct insn_table {
854 	const char *name;
855 	u32 flag;
856 	void (*insn_func)(void);
857 	u32 type;
858 	u32 reason;
859 	ulong exit_qual;
860 	u32 insn_info;
861 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
862 	// which field need to be tested, reason is always tested
863 	u32 test_field;
864 	const supported_fn supported_fn;
865 	u8 disabled;
866 };
867 
868 /*
869  * Add more test cases of instruction intercept here. Elements in this
870  * table is:
871  *	name/control flag/insn function/type/exit reason/exit qulification/
872  *	instruction info/field to test
873  * The last field defines which fields (exit_qual and insn_info) need to be
874  * tested in exit handler. If set to 0, only "reason" is checked.
875  */
876 static struct insn_table insn_table[] = {
877 	// Flags for Primary Processor-Based VM-Execution Controls
878 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
879 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
880 		0x12345678, 0, FIELD_EXIT_QUAL},
881 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0, &monitor_supported},
882 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0},
883 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
884 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
885 		FIELD_EXIT_QUAL},
886 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
887 		FIELD_EXIT_QUAL},
888 #ifdef __x86_64__
889 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
890 		FIELD_EXIT_QUAL},
891 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
892 		FIELD_EXIT_QUAL},
893 #endif
894 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0, &monitor_supported},
895 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
896 	// Flags for Secondary Processor-Based VM-Execution Controls
897 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
898 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
899 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
900 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
901 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
902 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
903 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
904 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
905 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
906 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
907 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
908 	// Instructions always trap
909 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
910 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
911 	// Instructions never trap
912 	{NULL},
913 };
914 
915 static int insn_intercept_init(struct vmcs *vmcs)
916 {
917 	u32 ctrl_cpu, cur_insn;
918 
919 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
920 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
921 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
922 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
923 	cr3 = read_cr3();
924 
925 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
926 		if (insn_table[cur_insn].supported_fn == NULL)
927 			continue;
928 		insn_table[cur_insn].disabled = !insn_table[cur_insn].supported_fn();
929 	}
930 	return VMX_TEST_START;
931 }
932 
933 static void insn_intercept_main(void)
934 {
935 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
936 		vmx_set_test_stage(cur_insn * 2);
937 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
938 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
939 		    (insn_table[cur_insn].type == INSN_CPU1 &&
940 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
941 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
942 			       insn_table[cur_insn].type - INSN_CPU0,
943 			       insn_table[cur_insn].name);
944 			continue;
945 		}
946 
947 		if (insn_table[cur_insn].disabled) {
948 			printf("\tFeature required for %s is not supported.\n",
949 			       insn_table[cur_insn].name);
950 			continue;
951 		}
952 
953 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
954 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
955 		    (insn_table[cur_insn].type == INSN_CPU1 &&
956 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
957 			/* skip hlt, it stalls the guest and is tested below */
958 			if (insn_table[cur_insn].insn_func != insn_hlt)
959 				insn_table[cur_insn].insn_func();
960 			report(vmx_get_test_stage() == cur_insn * 2,
961 					"execute %s",
962 					insn_table[cur_insn].name);
963 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
964 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
965 			       insn_table[cur_insn].type - INSN_CPU0,
966 			       insn_table[cur_insn].name);
967 
968 		vmcall();
969 
970 		insn_table[cur_insn].insn_func();
971 		report(vmx_get_test_stage() == cur_insn * 2 + 1,
972 				"intercept %s",
973 				insn_table[cur_insn].name);
974 
975 		vmx_set_test_stage(cur_insn * 2 + 1);
976 		vmcall();
977 	}
978 }
979 
980 static int insn_intercept_exit_handler(union exit_reason exit_reason)
981 {
982 	u64 guest_rip;
983 	ulong exit_qual;
984 	u32 insn_len;
985 	u32 insn_info;
986 	bool pass;
987 
988 	guest_rip = vmcs_read(GUEST_RIP);
989 	exit_qual = vmcs_read(EXI_QUALIFICATION);
990 	insn_len = vmcs_read(EXI_INST_LEN);
991 	insn_info = vmcs_read(EXI_INST_INFO);
992 
993 	if (exit_reason.basic == VMX_VMCALL) {
994 		u32 val = 0;
995 
996 		if (insn_table[cur_insn].type == INSN_CPU0)
997 			val = vmcs_read(CPU_EXEC_CTRL0);
998 		else if (insn_table[cur_insn].type == INSN_CPU1)
999 			val = vmcs_read(CPU_EXEC_CTRL1);
1000 
1001 		if (vmx_get_test_stage() & 1)
1002 			val &= ~insn_table[cur_insn].flag;
1003 		else
1004 			val |= insn_table[cur_insn].flag;
1005 
1006 		if (insn_table[cur_insn].type == INSN_CPU0)
1007 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
1008 		else if (insn_table[cur_insn].type == INSN_CPU1)
1009 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
1010 	} else {
1011 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
1012 			insn_table[cur_insn].reason == exit_reason.full;
1013 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
1014 		    insn_table[cur_insn].exit_qual != exit_qual)
1015 			pass = false;
1016 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
1017 		    insn_table[cur_insn].insn_info != insn_info)
1018 			pass = false;
1019 		if (pass)
1020 			vmx_inc_test_stage();
1021 	}
1022 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1023 	return VMX_TEST_RESUME;
1024 }
1025 
1026 /**
1027  * __setup_ept - Setup the VMCS fields to enable Extended Page Tables (EPT)
1028  * @hpa:	Host physical address of the top-level, a.k.a. root, EPT table
1029  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1030  *
1031  * Returns 0 on success, 1 on failure.
1032  *
1033  * Note that @hpa doesn't need to point at actual memory if VM-Launch is
1034  * expected to fail, e.g. setup_dummy_ept() arbitrarily passes '0' to satisfy
1035  * the various EPTP consistency checks, but doesn't ensure backing for HPA '0'.
1036  */
1037 static int __setup_ept(u64 hpa, bool enable_ad)
1038 {
1039 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1040 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1041 		printf("\tEPT is not supported");
1042 		return 1;
1043 	}
1044 	if (!(ept_vpid.val & EPT_CAP_WB)) {
1045 		printf("WB memtype for EPT walks not supported\n");
1046 		return 1;
1047 	}
1048 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
1049 		printf("\tPWL4 is not supported\n");
1050 		return 1;
1051 	}
1052 
1053 	eptp = EPT_MEM_TYPE_WB;
1054 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1055 	eptp |= hpa;
1056 	if (enable_ad)
1057 		eptp |= EPTP_AD_FLAG;
1058 
1059 	vmcs_write(EPTP, eptp);
1060 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY);
1061 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT);
1062 
1063 	return 0;
1064 }
1065 
1066 /**
1067  * setup_ept - Enable Extended Page Tables (EPT) and setup an identity map
1068  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1069  *
1070  * Returns 0 on success, 1 on failure.
1071  *
1072  * This is the "real" function for setting up EPT tables, i.e. use this for
1073  * tests that need to run code in the guest with EPT enabled.
1074  */
1075 static int setup_ept(bool enable_ad)
1076 {
1077 	unsigned long end_of_memory;
1078 
1079 	pml4 = alloc_page();
1080 
1081 	if (__setup_ept(virt_to_phys(pml4), enable_ad))
1082 		return 1;
1083 
1084 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1085 	if (end_of_memory < (1ul << 32))
1086 		end_of_memory = (1ul << 32);
1087 	/* Cannot use large EPT pages if we need to track EPT
1088 	 * accessed/dirty bits at 4K granularity.
1089 	 */
1090 	setup_ept_range(pml4, 0, end_of_memory, 0,
1091 			!enable_ad && ept_2m_supported(),
1092 			EPT_WA | EPT_RA | EPT_EA);
1093 	return 0;
1094 }
1095 
1096 /**
1097  * setup_dummy_ept - Enable Extended Page Tables (EPT) with a dummy root HPA
1098  *
1099  * Setup EPT using a semi-arbitrary dummy root HPA.  This function is intended
1100  * for use by tests that need EPT enabled to verify dependent VMCS controls
1101  * but never expect to fully enter the guest, i.e. don't need setup the actual
1102  * EPT tables.
1103  */
1104 static void setup_dummy_ept(void)
1105 {
1106 	if (__setup_ept(0, false))
1107 		report_abort("EPT setup unexpectedly failed");
1108 }
1109 
1110 static int enable_unrestricted_guest(void)
1111 {
1112 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1113 	    !(ctrl_cpu_rev[1].clr & CPU_URG) ||
1114 	    !(ctrl_cpu_rev[1].clr & CPU_EPT))
1115 		return 1;
1116 
1117 	setup_dummy_ept();
1118 
1119 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
1120 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
1121 
1122 	return 0;
1123 }
1124 
1125 static void ept_enable_ad_bits(void)
1126 {
1127 	eptp |= EPTP_AD_FLAG;
1128 	vmcs_write(EPTP, eptp);
1129 }
1130 
1131 static void ept_disable_ad_bits(void)
1132 {
1133 	eptp &= ~EPTP_AD_FLAG;
1134 	vmcs_write(EPTP, eptp);
1135 }
1136 
1137 static int ept_ad_enabled(void)
1138 {
1139 	return eptp & EPTP_AD_FLAG;
1140 }
1141 
1142 static void ept_enable_ad_bits_or_skip_test(void)
1143 {
1144 	if (!ept_ad_bits_supported())
1145 		test_skip("EPT AD bits not supported.");
1146 	ept_enable_ad_bits();
1147 }
1148 
1149 static int apic_version;
1150 
1151 static int ept_init_common(bool have_ad)
1152 {
1153 	int ret;
1154 	struct pci_dev pcidev;
1155 
1156 	if (setup_ept(have_ad))
1157 		return VMX_TEST_EXIT;
1158 	data_page1 = alloc_page();
1159 	data_page2 = alloc_page();
1160 	*((u32 *)data_page1) = MAGIC_VAL_1;
1161 	*((u32 *)data_page2) = MAGIC_VAL_2;
1162 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1163 			EPT_RA | EPT_WA | EPT_EA);
1164 
1165 	apic_version = apic_read(APIC_LVR);
1166 
1167 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1168 	if (ret != PCIDEVADDR_INVALID) {
1169 		pci_dev_init(&pcidev, ret);
1170 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1171 	}
1172 
1173 	return VMX_TEST_START;
1174 }
1175 
1176 static int ept_init(struct vmcs *vmcs)
1177 {
1178 	return ept_init_common(false);
1179 }
1180 
1181 static void ept_common(void)
1182 {
1183 	vmx_set_test_stage(0);
1184 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1185 			*((u32 *)data_page1) != MAGIC_VAL_1)
1186 		report(0, "EPT basic framework - read");
1187 	else {
1188 		*((u32 *)data_page2) = MAGIC_VAL_3;
1189 		vmcall();
1190 		if (vmx_get_test_stage() == 1) {
1191 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1192 					*((u32 *)data_page2) == MAGIC_VAL_2)
1193 				report(1, "EPT basic framework");
1194 			else
1195 				report(1, "EPT basic framework - remap");
1196 		}
1197 	}
1198 	// Test EPT Misconfigurations
1199 	vmx_set_test_stage(1);
1200 	vmcall();
1201 	*((u32 *)data_page1) = MAGIC_VAL_1;
1202 	if (vmx_get_test_stage() != 2) {
1203 		report(0, "EPT misconfigurations");
1204 		goto t1;
1205 	}
1206 	vmx_set_test_stage(2);
1207 	vmcall();
1208 	*((u32 *)data_page1) = MAGIC_VAL_1;
1209 	report(vmx_get_test_stage() == 3, "EPT misconfigurations");
1210 t1:
1211 	// Test EPT violation
1212 	vmx_set_test_stage(3);
1213 	vmcall();
1214 	*((u32 *)data_page1) = MAGIC_VAL_1;
1215 	report(vmx_get_test_stage() == 4, "EPT violation - page permission");
1216 	// Violation caused by EPT paging structure
1217 	vmx_set_test_stage(4);
1218 	vmcall();
1219 	*((u32 *)data_page1) = MAGIC_VAL_2;
1220 	report(vmx_get_test_stage() == 5, "EPT violation - paging structure");
1221 
1222 	// MMIO Read/Write
1223 	vmx_set_test_stage(5);
1224 	vmcall();
1225 
1226 	*(u32 volatile *)pci_physaddr;
1227 	report(vmx_get_test_stage() == 6, "MMIO EPT violation - read");
1228 
1229 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1230 	report(vmx_get_test_stage() == 7, "MMIO EPT violation - write");
1231 }
1232 
1233 static void ept_main(void)
1234 {
1235 	ept_common();
1236 
1237 	// Test EPT access to L1 MMIO
1238 	vmx_set_test_stage(7);
1239 	report(*((u32 *)0xfee00030UL) == apic_version, "EPT - MMIO access");
1240 
1241 	// Test invalid operand for INVEPT
1242 	vmcall();
1243 	report(vmx_get_test_stage() == 8, "EPT - unsupported INVEPT");
1244 }
1245 
1246 static bool invept_test(int type, u64 eptp)
1247 {
1248 	bool ret, supported;
1249 
1250 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1251 	ret = invept(type, eptp);
1252 
1253 	if (ret == !supported)
1254 		return false;
1255 
1256 	if (!supported)
1257 		printf("WARNING: unsupported invept passed!\n");
1258 	else
1259 		printf("WARNING: invept failed!\n");
1260 
1261 	return true;
1262 }
1263 
1264 static int pml_exit_handler(union exit_reason exit_reason)
1265 {
1266 	u16 index, count;
1267 	u64 *pmlbuf = pml_log;
1268 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1269 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1270 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1271 
1272 	switch (exit_reason.basic) {
1273 	case VMX_VMCALL:
1274 		switch (vmx_get_test_stage()) {
1275 		case 0:
1276 			index = vmcs_read(GUEST_PML_INDEX);
1277 			for (count = index + 1; count < PML_INDEX; count++) {
1278 				if (pmlbuf[count] == (u64)data_page2) {
1279 					vmx_inc_test_stage();
1280 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1281 					break;
1282 				}
1283 			}
1284 			break;
1285 		case 1:
1286 			index = vmcs_read(GUEST_PML_INDEX);
1287 			/* Keep clearing the dirty bit till a overflow */
1288 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1289 			break;
1290 		default:
1291 			report(false, "unexpected stage, %d.",
1292 			       vmx_get_test_stage());
1293 			print_vmexit_info(exit_reason);
1294 			return VMX_TEST_VMEXIT;
1295 		}
1296 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1297 		return VMX_TEST_RESUME;
1298 	case VMX_PML_FULL:
1299 		vmx_inc_test_stage();
1300 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1301 		return VMX_TEST_RESUME;
1302 	default:
1303 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
1304 		print_vmexit_info(exit_reason);
1305 	}
1306 	return VMX_TEST_VMEXIT;
1307 }
1308 
1309 static int ept_exit_handler_common(union exit_reason exit_reason, bool have_ad)
1310 {
1311 	u64 guest_rip;
1312 	u64 guest_cr3;
1313 	u32 insn_len;
1314 	u32 exit_qual;
1315 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte,
1316 			     guest_pte_addr;
1317 
1318 	guest_rip = vmcs_read(GUEST_RIP);
1319 	guest_cr3 = vmcs_read(GUEST_CR3);
1320 	insn_len = vmcs_read(EXI_INST_LEN);
1321 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1322 	pteval_t *ptep;
1323 	switch (exit_reason.basic) {
1324 	case VMX_VMCALL:
1325 		switch (vmx_get_test_stage()) {
1326 		case 0:
1327 			check_ept_ad(pml4, guest_cr3,
1328 				     (unsigned long)data_page1,
1329 				     have_ad ? EPT_ACCESS_FLAG : 0,
1330 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1331 			check_ept_ad(pml4, guest_cr3,
1332 				     (unsigned long)data_page2,
1333 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1334 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1335 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1336 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1337 			if (have_ad)
1338 				ept_sync(INVEPT_SINGLE, eptp);;
1339 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1340 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1341 				vmx_inc_test_stage();
1342 				install_ept(pml4, (unsigned long)data_page2,
1343 						(unsigned long)data_page2,
1344 						EPT_RA | EPT_WA | EPT_EA);
1345 			} else
1346 				report(0, "EPT basic framework - write");
1347 			break;
1348 		case 1:
1349 			install_ept(pml4, (unsigned long)data_page1,
1350  				(unsigned long)data_page1, EPT_WA);
1351 			ept_sync(INVEPT_SINGLE, eptp);
1352 			break;
1353 		case 2:
1354 			install_ept(pml4, (unsigned long)data_page1,
1355  				(unsigned long)data_page1,
1356  				EPT_RA | EPT_WA | EPT_EA |
1357  				(2 << EPT_MEM_TYPE_SHIFT));
1358 			ept_sync(INVEPT_SINGLE, eptp);
1359 			break;
1360 		case 3:
1361 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1362 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1363 						1, &data_page1_pte));
1364 			set_ept_pte(pml4, (unsigned long)data_page1,
1365 				1, data_page1_pte & ~EPT_PRESENT);
1366 			ept_sync(INVEPT_SINGLE, eptp);
1367 			break;
1368 		case 4:
1369 			ptep = get_pte_level((pgd_t *)guest_cr3, data_page1, /*level=*/2);
1370 			guest_pte_addr = virt_to_phys(ptep) & PAGE_MASK;
1371 
1372 			TEST_ASSERT(get_ept_pte(pml4, guest_pte_addr, 2, &data_page1_pte_pte));
1373 			set_ept_pte(pml4, guest_pte_addr, 2,
1374 				data_page1_pte_pte & ~EPT_PRESENT);
1375 			ept_sync(INVEPT_SINGLE, eptp);
1376 			break;
1377 		case 5:
1378 			install_ept(pml4, (unsigned long)pci_physaddr,
1379 				(unsigned long)pci_physaddr, 0);
1380 			ept_sync(INVEPT_SINGLE, eptp);
1381 			break;
1382 		case 7:
1383 			if (!invept_test(0, eptp))
1384 				vmx_inc_test_stage();
1385 			break;
1386 		// Should not reach here
1387 		default:
1388 			report(false, "ERROR - unexpected stage, %d.",
1389 			       vmx_get_test_stage());
1390 			print_vmexit_info(exit_reason);
1391 			return VMX_TEST_VMEXIT;
1392 		}
1393 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1394 		return VMX_TEST_RESUME;
1395 	case VMX_EPT_MISCONFIG:
1396 		switch (vmx_get_test_stage()) {
1397 		case 1:
1398 		case 2:
1399 			vmx_inc_test_stage();
1400 			install_ept(pml4, (unsigned long)data_page1,
1401  				(unsigned long)data_page1,
1402  				EPT_RA | EPT_WA | EPT_EA);
1403 			ept_sync(INVEPT_SINGLE, eptp);
1404 			break;
1405 		// Should not reach here
1406 		default:
1407 			report(false, "ERROR - unexpected stage, %d.",
1408 			       vmx_get_test_stage());
1409 			print_vmexit_info(exit_reason);
1410 			return VMX_TEST_VMEXIT;
1411 		}
1412 		return VMX_TEST_RESUME;
1413 	case VMX_EPT_VIOLATION:
1414 		/*
1415 		 * Exit-qualifications are masked not to account for advanced
1416 		 * VM-exit information. Once KVM supports this feature, this
1417 		 * masking should be removed.
1418 		 */
1419 		exit_qual &= ~EPT_VLT_GUEST_MASK;
1420 
1421 		switch(vmx_get_test_stage()) {
1422 		case 3:
1423 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1424 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1425 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1426 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1427 					EPT_VLT_PADDR))
1428 				vmx_inc_test_stage();
1429 			set_ept_pte(pml4, (unsigned long)data_page1,
1430 				1, data_page1_pte | (EPT_PRESENT));
1431 			ept_sync(INVEPT_SINGLE, eptp);
1432 			break;
1433 		case 4:
1434 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1435 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1436 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1437 			if (exit_qual == (EPT_VLT_RD |
1438 					  (have_ad ? EPT_VLT_WR : 0) |
1439 					  EPT_VLT_LADDR_VLD))
1440 				vmx_inc_test_stage();
1441 			set_ept_pte(pml4, guest_pte_addr, 2,
1442 				data_page1_pte_pte | (EPT_PRESENT));
1443 			ept_sync(INVEPT_SINGLE, eptp);
1444 			break;
1445 		case 5:
1446 			if (exit_qual & EPT_VLT_RD)
1447 				vmx_inc_test_stage();
1448 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1449 						1, &memaddr_pte));
1450 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1451 			ept_sync(INVEPT_SINGLE, eptp);
1452 			break;
1453 		case 6:
1454 			if (exit_qual & EPT_VLT_WR)
1455 				vmx_inc_test_stage();
1456 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1457 						1, &memaddr_pte));
1458 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1459 			ept_sync(INVEPT_SINGLE, eptp);
1460 			break;
1461 		default:
1462 			// Should not reach here
1463 			report(false, "ERROR : unexpected stage, %d",
1464 			       vmx_get_test_stage());
1465 			print_vmexit_info(exit_reason);
1466 			return VMX_TEST_VMEXIT;
1467 		}
1468 		return VMX_TEST_RESUME;
1469 	default:
1470 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
1471 		print_vmexit_info(exit_reason);
1472 	}
1473 	return VMX_TEST_VMEXIT;
1474 }
1475 
1476 static int ept_exit_handler(union exit_reason exit_reason)
1477 {
1478 	return ept_exit_handler_common(exit_reason, false);
1479 }
1480 
1481 static int eptad_init(struct vmcs *vmcs)
1482 {
1483 	int r = ept_init_common(true);
1484 
1485 	if (r == VMX_TEST_EXIT)
1486 		return r;
1487 
1488 	if ((rdmsr(MSR_IA32_VMX_EPT_VPID_CAP) & EPT_CAP_AD_FLAG) == 0) {
1489 		printf("\tEPT A/D bits are not supported");
1490 		return VMX_TEST_EXIT;
1491 	}
1492 
1493 	return r;
1494 }
1495 
1496 static int pml_init(struct vmcs *vmcs)
1497 {
1498 	u32 ctrl_cpu;
1499 	int r = eptad_init(vmcs);
1500 
1501 	if (r == VMX_TEST_EXIT)
1502 		return r;
1503 
1504 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1505 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1506 		printf("\tPML is not supported");
1507 		return VMX_TEST_EXIT;
1508 	}
1509 
1510 	pml_log = alloc_page();
1511 	vmcs_write(PMLADDR, (u64)pml_log);
1512 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1513 
1514 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1515 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1516 
1517 	return VMX_TEST_START;
1518 }
1519 
1520 static void pml_main(void)
1521 {
1522 	int count = 0;
1523 
1524 	vmx_set_test_stage(0);
1525 	*((u32 *)data_page2) = 0x1;
1526 	vmcall();
1527 	report(vmx_get_test_stage() == 1, "PML - Dirty GPA Logging");
1528 
1529 	while (vmx_get_test_stage() == 1) {
1530 		vmcall();
1531 		*((u32 *)data_page2) = 0x1;
1532 		if (count++ > PML_INDEX)
1533 			break;
1534 	}
1535 	report(vmx_get_test_stage() == 2, "PML Full Event");
1536 }
1537 
1538 static void eptad_main(void)
1539 {
1540 	ept_common();
1541 }
1542 
1543 static int eptad_exit_handler(union exit_reason exit_reason)
1544 {
1545 	return ept_exit_handler_common(exit_reason, true);
1546 }
1547 
1548 static bool invvpid_test(int type, u16 vpid)
1549 {
1550 	bool ret, supported;
1551 
1552 	supported = ept_vpid.val &
1553 		(VPID_CAP_INVVPID_ADDR >> INVVPID_ADDR << type);
1554 	ret = invvpid(type, vpid, 0);
1555 
1556 	if (ret == !supported)
1557 		return false;
1558 
1559 	if (!supported)
1560 		printf("WARNING: unsupported invvpid passed!\n");
1561 	else
1562 		printf("WARNING: invvpid failed!\n");
1563 
1564 	return true;
1565 }
1566 
1567 static int vpid_init(struct vmcs *vmcs)
1568 {
1569 	u32 ctrl_cpu1;
1570 
1571 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1572 		!(ctrl_cpu_rev[1].clr & CPU_VPID)) {
1573 		printf("\tVPID is not supported");
1574 		return VMX_TEST_EXIT;
1575 	}
1576 
1577 	ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
1578 	ctrl_cpu1 |= CPU_VPID;
1579 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
1580 	return VMX_TEST_START;
1581 }
1582 
1583 static void vpid_main(void)
1584 {
1585 	vmx_set_test_stage(0);
1586 	vmcall();
1587 	report(vmx_get_test_stage() == 1, "INVVPID SINGLE ADDRESS");
1588 	vmx_set_test_stage(2);
1589 	vmcall();
1590 	report(vmx_get_test_stage() == 3, "INVVPID SINGLE");
1591 	vmx_set_test_stage(4);
1592 	vmcall();
1593 	report(vmx_get_test_stage() == 5, "INVVPID ALL");
1594 }
1595 
1596 static int vpid_exit_handler(union exit_reason exit_reason)
1597 {
1598 	u64 guest_rip;
1599 	u32 insn_len;
1600 
1601 	guest_rip = vmcs_read(GUEST_RIP);
1602 	insn_len = vmcs_read(EXI_INST_LEN);
1603 
1604 	switch (exit_reason.basic) {
1605 	case VMX_VMCALL:
1606 		switch(vmx_get_test_stage()) {
1607 		case 0:
1608 			if (!invvpid_test(INVVPID_ADDR, 1))
1609 				vmx_inc_test_stage();
1610 			break;
1611 		case 2:
1612 			if (!invvpid_test(INVVPID_CONTEXT_GLOBAL, 1))
1613 				vmx_inc_test_stage();
1614 			break;
1615 		case 4:
1616 			if (!invvpid_test(INVVPID_ALL, 1))
1617 				vmx_inc_test_stage();
1618 			break;
1619 		default:
1620 			report(false, "ERROR: unexpected stage, %d",
1621 					vmx_get_test_stage());
1622 			print_vmexit_info(exit_reason);
1623 			return VMX_TEST_VMEXIT;
1624 		}
1625 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1626 		return VMX_TEST_RESUME;
1627 	default:
1628 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
1629 		print_vmexit_info(exit_reason);
1630 	}
1631 	return VMX_TEST_VMEXIT;
1632 }
1633 
1634 #define TIMER_VECTOR	222
1635 
1636 static volatile bool timer_fired;
1637 
1638 static void timer_isr(isr_regs_t *regs)
1639 {
1640 	timer_fired = true;
1641 	apic_write(APIC_EOI, 0);
1642 }
1643 
1644 static int interrupt_init(struct vmcs *vmcs)
1645 {
1646 	msr_bmp_init();
1647 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1648 	handle_irq(TIMER_VECTOR, timer_isr);
1649 	return VMX_TEST_START;
1650 }
1651 
1652 static void interrupt_main(void)
1653 {
1654 	long long start, loops;
1655 
1656 	vmx_set_test_stage(0);
1657 
1658 	apic_write(APIC_LVTT, TIMER_VECTOR);
1659 	irq_enable();
1660 
1661 	apic_write(APIC_TMICT, 1);
1662 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1663 		asm volatile ("nop");
1664 	report(timer_fired, "direct interrupt while running guest");
1665 
1666 	apic_write(APIC_TMICT, 0);
1667 	irq_disable();
1668 	vmcall();
1669 	timer_fired = false;
1670 	apic_write(APIC_TMICT, 1);
1671 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1672 		asm volatile ("nop");
1673 	report(timer_fired, "intercepted interrupt while running guest");
1674 
1675 	irq_enable();
1676 	apic_write(APIC_TMICT, 0);
1677 	irq_disable();
1678 	vmcall();
1679 	timer_fired = false;
1680 	start = rdtsc();
1681 	apic_write(APIC_TMICT, 1000000);
1682 
1683 	asm volatile ("sti; hlt");
1684 
1685 	report(rdtsc() - start > 1000000 && timer_fired,
1686 	       "direct interrupt + hlt");
1687 
1688 	apic_write(APIC_TMICT, 0);
1689 	irq_disable();
1690 	vmcall();
1691 	timer_fired = false;
1692 	start = rdtsc();
1693 	apic_write(APIC_TMICT, 1000000);
1694 
1695 	asm volatile ("sti; hlt");
1696 
1697 	report(rdtsc() - start > 10000 && timer_fired,
1698 	       "intercepted interrupt + hlt");
1699 
1700 	apic_write(APIC_TMICT, 0);
1701 	irq_disable();
1702 	vmcall();
1703 	timer_fired = false;
1704 	start = rdtsc();
1705 	apic_write(APIC_TMICT, 1000000);
1706 
1707 	irq_enable();
1708 	asm volatile ("nop");
1709 	vmcall();
1710 
1711 	report(rdtsc() - start > 10000 && timer_fired,
1712 	       "direct interrupt + activity state hlt");
1713 
1714 	apic_write(APIC_TMICT, 0);
1715 	irq_disable();
1716 	vmcall();
1717 	timer_fired = false;
1718 	start = rdtsc();
1719 	apic_write(APIC_TMICT, 1000000);
1720 
1721 	irq_enable();
1722 	asm volatile ("nop");
1723 	vmcall();
1724 
1725 	report(rdtsc() - start > 10000 && timer_fired,
1726 	       "intercepted interrupt + activity state hlt");
1727 
1728 	apic_write(APIC_TMICT, 0);
1729 	irq_disable();
1730 	vmx_set_test_stage(7);
1731 	vmcall();
1732 	timer_fired = false;
1733 	apic_write(APIC_TMICT, 1);
1734 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1735 		asm volatile ("nop");
1736 	report(timer_fired,
1737 	       "running a guest with interrupt acknowledgement set");
1738 
1739 	apic_write(APIC_TMICT, 0);
1740 	irq_enable();
1741 	timer_fired = false;
1742 	vmcall();
1743 	report(timer_fired, "Inject an event to a halted guest");
1744 }
1745 
1746 static int interrupt_exit_handler(union exit_reason exit_reason)
1747 {
1748 	u64 guest_rip = vmcs_read(GUEST_RIP);
1749 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1750 
1751 	switch (exit_reason.basic) {
1752 	case VMX_VMCALL:
1753 		switch (vmx_get_test_stage()) {
1754 		case 0:
1755 		case 2:
1756 		case 5:
1757 			vmcs_write(PIN_CONTROLS,
1758 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1759 			break;
1760 		case 7:
1761 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1762 			vmcs_write(PIN_CONTROLS,
1763 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1764 			break;
1765 		case 1:
1766 		case 3:
1767 			vmcs_write(PIN_CONTROLS,
1768 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1769 			break;
1770 		case 4:
1771 		case 6:
1772 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1773 			break;
1774 
1775 		case 8:
1776 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1777 			vmcs_write(ENT_INTR_INFO,
1778 				   TIMER_VECTOR |
1779 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1780 				   INTR_INFO_VALID_MASK);
1781 			break;
1782 		}
1783 		vmx_inc_test_stage();
1784 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1785 		return VMX_TEST_RESUME;
1786 	case VMX_EXTINT:
1787 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1788 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1789 			handle_external_interrupt(vector);
1790 		} else {
1791 			irq_enable();
1792 			asm volatile ("nop");
1793 			irq_disable();
1794 		}
1795 		if (vmx_get_test_stage() >= 2)
1796 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1797 		return VMX_TEST_RESUME;
1798 	default:
1799 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
1800 		print_vmexit_info(exit_reason);
1801 	}
1802 
1803 	return VMX_TEST_VMEXIT;
1804 }
1805 
1806 
1807 static volatile int nmi_fired;
1808 
1809 #define NMI_DELAY 100000000ULL
1810 
1811 static void nmi_isr(isr_regs_t *regs)
1812 {
1813 	nmi_fired = true;
1814 }
1815 
1816 static int nmi_hlt_init(struct vmcs *vmcs)
1817 {
1818 	msr_bmp_init();
1819 	handle_irq(NMI_VECTOR, nmi_isr);
1820 	vmcs_write(PIN_CONTROLS,
1821 		   vmcs_read(PIN_CONTROLS) & ~PIN_NMI);
1822 	vmcs_write(PIN_CONTROLS,
1823 		   vmcs_read(PIN_CONTROLS) & ~PIN_VIRT_NMI);
1824 	return VMX_TEST_START;
1825 }
1826 
1827 static void nmi_message_thread(void *data)
1828 {
1829     while (vmx_get_test_stage() != 1)
1830         pause();
1831 
1832     delay(NMI_DELAY);
1833     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1834 
1835     while (vmx_get_test_stage() != 2)
1836         pause();
1837 
1838     delay(NMI_DELAY);
1839     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1840 }
1841 
1842 static void nmi_hlt_main(void)
1843 {
1844     long long start;
1845 
1846     if (cpu_count() < 2) {
1847         report_skip(__func__);
1848         vmx_set_test_stage(-1);
1849         return;
1850     }
1851 
1852     vmx_set_test_stage(0);
1853     on_cpu_async(1, nmi_message_thread, NULL);
1854     start = rdtsc();
1855     vmx_set_test_stage(1);
1856     asm volatile ("hlt");
1857     report((rdtsc() - start > NMI_DELAY) && nmi_fired,
1858             "direct NMI + hlt");
1859     if (!nmi_fired)
1860         vmx_set_test_stage(-1);
1861     nmi_fired = false;
1862 
1863     vmcall();
1864 
1865     start = rdtsc();
1866     vmx_set_test_stage(2);
1867     asm volatile ("hlt");
1868     report((rdtsc() - start > NMI_DELAY) && !nmi_fired,
1869             "intercepted NMI + hlt");
1870     if (nmi_fired) {
1871         report(!nmi_fired, "intercepted NMI was dispatched");
1872         vmx_set_test_stage(-1);
1873         return;
1874     }
1875     vmx_set_test_stage(3);
1876 }
1877 
1878 static int nmi_hlt_exit_handler(union exit_reason exit_reason)
1879 {
1880     u64 guest_rip = vmcs_read(GUEST_RIP);
1881     u32 insn_len = vmcs_read(EXI_INST_LEN);
1882 
1883     switch (vmx_get_test_stage()) {
1884     case 1:
1885         if (exit_reason.basic != VMX_VMCALL) {
1886             report(false, "VMEXIT not due to vmcall. Exit reason 0x%x",
1887                    exit_reason.full);
1888             print_vmexit_info(exit_reason);
1889             return VMX_TEST_VMEXIT;
1890         }
1891 
1892         vmcs_write(PIN_CONTROLS,
1893                vmcs_read(PIN_CONTROLS) | PIN_NMI);
1894         vmcs_write(PIN_CONTROLS,
1895                vmcs_read(PIN_CONTROLS) | PIN_VIRT_NMI);
1896         vmcs_write(GUEST_RIP, guest_rip + insn_len);
1897         break;
1898 
1899     case 2:
1900         if (exit_reason.basic != VMX_EXC_NMI) {
1901             report(false, "VMEXIT not due to NMI intercept. Exit reason 0x%x",
1902                    exit_reason.full);
1903             print_vmexit_info(exit_reason);
1904             return VMX_TEST_VMEXIT;
1905         }
1906         report(true, "NMI intercept while running guest");
1907         vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1908         break;
1909 
1910     case 3:
1911         break;
1912 
1913     default:
1914         return VMX_TEST_VMEXIT;
1915     }
1916 
1917     if (vmx_get_test_stage() == 3)
1918         return VMX_TEST_VMEXIT;
1919 
1920     return VMX_TEST_RESUME;
1921 }
1922 
1923 
1924 static int dbgctls_init(struct vmcs *vmcs)
1925 {
1926 	u64 dr7 = 0x402;
1927 	u64 zero = 0;
1928 
1929 	msr_bmp_init();
1930 	asm volatile(
1931 		"mov %0,%%dr0\n\t"
1932 		"mov %0,%%dr1\n\t"
1933 		"mov %0,%%dr2\n\t"
1934 		"mov %1,%%dr7\n\t"
1935 		: : "r" (zero), "r" (dr7));
1936 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1937 	vmcs_write(GUEST_DR7, 0x404);
1938 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1939 
1940 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1941 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1942 
1943 	return VMX_TEST_START;
1944 }
1945 
1946 static void dbgctls_main(void)
1947 {
1948 	u64 dr7, debugctl;
1949 
1950 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1951 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1952 	/* Commented out: KVM does not support DEBUGCTL so far */
1953 	(void)debugctl;
1954 	report(dr7 == 0x404, "Load debug controls" /* && debugctl == 0x2 */);
1955 
1956 	dr7 = 0x408;
1957 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1958 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1959 
1960 	vmx_set_test_stage(0);
1961 	vmcall();
1962 	report(vmx_get_test_stage() == 1, "Save debug controls");
1963 
1964 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1965 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1966 		printf("\tDebug controls are always loaded/saved\n");
1967 		return;
1968 	}
1969 	vmx_set_test_stage(2);
1970 	vmcall();
1971 
1972 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1973 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1974 	/* Commented out: KVM does not support DEBUGCTL so far */
1975 	(void)debugctl;
1976 	report(dr7 == 0x402,
1977 	       "Guest=host debug controls" /* && debugctl == 0x1 */);
1978 
1979 	dr7 = 0x408;
1980 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1981 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1982 
1983 	vmx_set_test_stage(3);
1984 	vmcall();
1985 	report(vmx_get_test_stage() == 4, "Don't save debug controls");
1986 }
1987 
1988 static int dbgctls_exit_handler(union exit_reason exit_reason)
1989 {
1990 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1991 	u64 guest_rip = vmcs_read(GUEST_RIP);
1992 	u64 dr7, debugctl;
1993 
1994 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1995 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1996 
1997 	switch (exit_reason.basic) {
1998 	case VMX_VMCALL:
1999 		switch (vmx_get_test_stage()) {
2000 		case 0:
2001 			if (dr7 == 0x400 && debugctl == 0 &&
2002 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
2003 			    Commented out: KVM does not support DEBUGCTL so far
2004 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
2005 				vmx_inc_test_stage();
2006 			break;
2007 		case 2:
2008 			dr7 = 0x402;
2009 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
2010 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
2011 			vmcs_write(GUEST_DR7, 0x404);
2012 			vmcs_write(GUEST_DEBUGCTL, 0x2);
2013 
2014 			vmcs_write(ENT_CONTROLS,
2015 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
2016 			vmcs_write(EXI_CONTROLS,
2017 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
2018 			break;
2019 		case 3:
2020 			if (dr7 == 0x400 && debugctl == 0 &&
2021 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
2022 			    Commented out: KVM does not support DEBUGCTL so far
2023 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
2024 				vmx_inc_test_stage();
2025 			break;
2026 		}
2027 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
2028 		return VMX_TEST_RESUME;
2029 	default:
2030 		report(false, "Unknown exit reason, %d", exit_reason.full);
2031 		print_vmexit_info(exit_reason);
2032 	}
2033 	return VMX_TEST_VMEXIT;
2034 }
2035 
2036 struct vmx_msr_entry {
2037 	u32 index;
2038 	u32 reserved;
2039 	u64 value;
2040 } __attribute__((packed));
2041 
2042 #define MSR_MAGIC 0x31415926
2043 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
2044 
2045 static int msr_switch_init(struct vmcs *vmcs)
2046 {
2047 	msr_bmp_init();
2048 	exit_msr_store = alloc_page();
2049 	exit_msr_load = alloc_page();
2050 	entry_msr_load = alloc_page();
2051 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
2052 	entry_msr_load[0].value = MSR_MAGIC;
2053 
2054 	vmx_set_test_stage(1);
2055 	vmcs_write(ENT_MSR_LD_CNT, 1);
2056 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
2057 	vmcs_write(EXI_MSR_ST_CNT, 1);
2058 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
2059 	vmcs_write(EXI_MSR_LD_CNT, 1);
2060 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
2061 	return VMX_TEST_START;
2062 }
2063 
2064 static void msr_switch_main(void)
2065 {
2066 	if (vmx_get_test_stage() == 1) {
2067 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC,
2068 		       "VM entry MSR load");
2069 		vmx_set_test_stage(2);
2070 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
2071 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
2072 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
2073 		exit_msr_load[0].value = MSR_MAGIC + 2;
2074 	}
2075 	vmcall();
2076 }
2077 
2078 static int msr_switch_exit_handler(union exit_reason exit_reason)
2079 {
2080 	if (exit_reason.basic == VMX_VMCALL && vmx_get_test_stage() == 2) {
2081 		report(exit_msr_store[0].value == MSR_MAGIC + 1,
2082 		       "VM exit MSR store");
2083 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2,
2084 		       "VM exit MSR load");
2085 		vmx_set_test_stage(3);
2086 		entry_msr_load[0].index = MSR_FS_BASE;
2087 		return VMX_TEST_RESUME;
2088 	}
2089 	printf("ERROR %s: unexpected stage=%u or reason=0x%x\n",
2090 		__func__, vmx_get_test_stage(), exit_reason.full);
2091 	return VMX_TEST_EXIT;
2092 }
2093 
2094 static int msr_switch_entry_failure(struct vmentry_result *result)
2095 {
2096 	if (result->vm_fail) {
2097 		printf("ERROR %s: VM-Fail on %s\n", __func__, result->instr);
2098 		return VMX_TEST_EXIT;
2099 	}
2100 
2101 	if (result->exit_reason.failed_vmentry &&
2102 	    result->exit_reason.basic == VMX_FAIL_MSR &&
2103 	    vmx_get_test_stage() == 3) {
2104 		report(vmcs_read(EXI_QUALIFICATION) == 1,
2105 		       "VM entry MSR load: try to load FS_BASE");
2106 		return VMX_TEST_VMEXIT;
2107 	}
2108 	printf("ERROR %s: unexpected stage=%u or reason=%x\n",
2109 		__func__, vmx_get_test_stage(), result->exit_reason.full);
2110 	return VMX_TEST_EXIT;
2111 }
2112 
2113 static int vmmcall_init(struct vmcs *vmcs)
2114 {
2115 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
2116 	return VMX_TEST_START;
2117 }
2118 
2119 static void vmmcall_main(void)
2120 {
2121 	asm volatile(
2122 		"mov $0xABCD, %%rax\n\t"
2123 		"vmmcall\n\t"
2124 		::: "rax");
2125 
2126 	report(0, "VMMCALL");
2127 }
2128 
2129 static int vmmcall_exit_handler(union exit_reason exit_reason)
2130 {
2131 	switch (exit_reason.basic) {
2132 	case VMX_VMCALL:
2133 		printf("here\n");
2134 		report(0, "VMMCALL triggers #UD");
2135 		break;
2136 	case VMX_EXC_NMI:
2137 		report((vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR,
2138 		       "VMMCALL triggers #UD");
2139 		break;
2140 	default:
2141 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
2142 		print_vmexit_info(exit_reason);
2143 	}
2144 
2145 	return VMX_TEST_VMEXIT;
2146 }
2147 
2148 static int disable_rdtscp_init(struct vmcs *vmcs)
2149 {
2150 	u32 ctrl_cpu1;
2151 
2152 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
2153 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
2154 		ctrl_cpu1 &= ~CPU_RDTSCP;
2155 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
2156 	}
2157 
2158 	return VMX_TEST_START;
2159 }
2160 
2161 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
2162 {
2163 	switch (vmx_get_test_stage()) {
2164 	case 0:
2165 		report(true, "RDTSCP triggers #UD");
2166 		vmx_inc_test_stage();
2167 		regs->rip += 3;
2168 		break;
2169 	case 2:
2170 		report(true, "RDPID triggers #UD");
2171 		vmx_inc_test_stage();
2172 		regs->rip += 4;
2173 		break;
2174 	}
2175 	return;
2176 
2177 }
2178 
2179 static void disable_rdtscp_main(void)
2180 {
2181 	/* Test that #UD is properly injected in L2.  */
2182 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2183 
2184 	vmx_set_test_stage(0);
2185 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2186 	vmcall();
2187 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2188 
2189 	handle_exception(UD_VECTOR, 0);
2190 	vmcall();
2191 }
2192 
2193 static int disable_rdtscp_exit_handler(union exit_reason exit_reason)
2194 {
2195 	switch (exit_reason.basic) {
2196 	case VMX_VMCALL:
2197 		switch (vmx_get_test_stage()) {
2198 		case 0:
2199 			report(false, "RDTSCP triggers #UD");
2200 			vmx_inc_test_stage();
2201 			/* fallthrough */
2202 		case 1:
2203 			vmx_inc_test_stage();
2204 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2205 			return VMX_TEST_RESUME;
2206 		case 2:
2207 			report(false, "RDPID triggers #UD");
2208 			break;
2209 		}
2210 		break;
2211 
2212 	default:
2213 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
2214 		print_vmexit_info(exit_reason);
2215 	}
2216 	return VMX_TEST_VMEXIT;
2217 }
2218 
2219 static int int3_init(struct vmcs *vmcs)
2220 {
2221 	vmcs_write(EXC_BITMAP, ~0u);
2222 	return VMX_TEST_START;
2223 }
2224 
2225 static void int3_guest_main(void)
2226 {
2227 	asm volatile ("int3");
2228 }
2229 
2230 static int int3_exit_handler(union exit_reason exit_reason)
2231 {
2232 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2233 
2234 	report(exit_reason.basic == VMX_EXC_NMI &&
2235 	       (intr_info & INTR_INFO_VALID_MASK) &&
2236 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2237 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2238 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2239 	       "L1 intercepts #BP");
2240 
2241 	return VMX_TEST_VMEXIT;
2242 }
2243 
2244 static int into_init(struct vmcs *vmcs)
2245 {
2246 	vmcs_write(EXC_BITMAP, ~0u);
2247 	return VMX_TEST_START;
2248 }
2249 
2250 static void into_guest_main(void)
2251 {
2252 	struct far_pointer32 fp = {
2253 		.offset = (uintptr_t)&&into,
2254 		.selector = KERNEL_CS32,
2255 	};
2256 	uintptr_t rsp;
2257 
2258 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
2259 
2260 	if (fp.offset != (uintptr_t)&&into) {
2261 		printf("Code address too high.\n");
2262 		return;
2263 	}
2264 	if ((u32)rsp != rsp) {
2265 		printf("Stack address too high.\n");
2266 		return;
2267 	}
2268 
2269 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2270 	return;
2271 into:
2272 	asm volatile (".code32;"
2273 		      "movl $0x7fffffff, %eax;"
2274 		      "addl %eax, %eax;"
2275 		      "into;"
2276 		      "lret;"
2277 		      ".code64");
2278 	__builtin_unreachable();
2279 }
2280 
2281 static int into_exit_handler(union exit_reason exit_reason)
2282 {
2283 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2284 
2285 	report(exit_reason.basic == VMX_EXC_NMI &&
2286 	       (intr_info & INTR_INFO_VALID_MASK) &&
2287 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2288 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2289 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2290 	       "L1 intercepts #OF");
2291 
2292 	return VMX_TEST_VMEXIT;
2293 }
2294 
2295 static void exit_monitor_from_l2_main(void)
2296 {
2297 	printf("Calling exit(0) from l2...\n");
2298 	exit(0);
2299 }
2300 
2301 static int exit_monitor_from_l2_handler(union exit_reason exit_reason)
2302 {
2303 	report(false, "The guest should have killed the VMM");
2304 	return VMX_TEST_EXIT;
2305 }
2306 
2307 static void assert_exit_reason(u64 expected)
2308 {
2309 	u64 actual = vmcs_read(EXI_REASON);
2310 
2311 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2312 			   exit_reason_description(expected),
2313 			   exit_reason_description(actual));
2314 }
2315 
2316 static void skip_exit_insn(void)
2317 {
2318 	u64 guest_rip = vmcs_read(GUEST_RIP);
2319 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2320 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2321 }
2322 
2323 static void skip_exit_vmcall(void)
2324 {
2325 	assert_exit_reason(VMX_VMCALL);
2326 	skip_exit_insn();
2327 }
2328 
2329 static void v2_null_test_guest(void)
2330 {
2331 }
2332 
2333 static void v2_null_test(void)
2334 {
2335 	test_set_guest(v2_null_test_guest);
2336 	enter_guest();
2337 	report(1, __func__);
2338 }
2339 
2340 static void v2_multiple_entries_test_guest(void)
2341 {
2342 	vmx_set_test_stage(1);
2343 	vmcall();
2344 	vmx_set_test_stage(2);
2345 }
2346 
2347 static void v2_multiple_entries_test(void)
2348 {
2349 	test_set_guest(v2_multiple_entries_test_guest);
2350 	enter_guest();
2351 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2352 	skip_exit_vmcall();
2353 	enter_guest();
2354 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2355 	report(1, __func__);
2356 }
2357 
2358 static int fixture_test_data = 1;
2359 
2360 static void fixture_test_teardown(void *data)
2361 {
2362 	*((int *) data) = 1;
2363 }
2364 
2365 static void fixture_test_guest(void)
2366 {
2367 	fixture_test_data++;
2368 }
2369 
2370 
2371 static void fixture_test_setup(void)
2372 {
2373 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2374 			   "fixture_test_teardown didn't run?!");
2375 	fixture_test_data = 2;
2376 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2377 	test_set_guest(fixture_test_guest);
2378 }
2379 
2380 static void fixture_test_case1(void)
2381 {
2382 	fixture_test_setup();
2383 	TEST_ASSERT_EQ(2, fixture_test_data);
2384 	enter_guest();
2385 	TEST_ASSERT_EQ(3, fixture_test_data);
2386 	report(1, __func__);
2387 }
2388 
2389 static void fixture_test_case2(void)
2390 {
2391 	fixture_test_setup();
2392 	TEST_ASSERT_EQ(2, fixture_test_data);
2393 	enter_guest();
2394 	TEST_ASSERT_EQ(3, fixture_test_data);
2395 	report(1, __func__);
2396 }
2397 
2398 enum ept_access_op {
2399 	OP_READ,
2400 	OP_WRITE,
2401 	OP_EXEC,
2402 	OP_FLUSH_TLB,
2403 	OP_EXIT,
2404 };
2405 
2406 static struct ept_access_test_data {
2407 	unsigned long gpa;
2408 	unsigned long *gva;
2409 	unsigned long hpa;
2410 	unsigned long *hva;
2411 	enum ept_access_op op;
2412 } ept_access_test_data;
2413 
2414 extern unsigned char ret42_start;
2415 extern unsigned char ret42_end;
2416 
2417 /* Returns 42. */
2418 asm(
2419 	".align 64\n"
2420 	"ret42_start:\n"
2421 	"mov $42, %eax\n"
2422 	"ret\n"
2423 	"ret42_end:\n"
2424 );
2425 
2426 static void
2427 diagnose_ept_violation_qual(u64 expected, u64 actual)
2428 {
2429 
2430 #define DIAGNOSE(flag)							\
2431 do {									\
2432 	if ((expected & flag) != (actual & flag))			\
2433 		printf(#flag " %sexpected\n",				\
2434 		       (expected & flag) ? "" : "un");			\
2435 } while (0)
2436 
2437 	DIAGNOSE(EPT_VLT_RD);
2438 	DIAGNOSE(EPT_VLT_WR);
2439 	DIAGNOSE(EPT_VLT_FETCH);
2440 	DIAGNOSE(EPT_VLT_PERM_RD);
2441 	DIAGNOSE(EPT_VLT_PERM_WR);
2442 	DIAGNOSE(EPT_VLT_PERM_EX);
2443 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2444 	DIAGNOSE(EPT_VLT_PADDR);
2445 
2446 #undef DIAGNOSE
2447 }
2448 
2449 static void do_ept_access_op(enum ept_access_op op)
2450 {
2451 	ept_access_test_data.op = op;
2452 	enter_guest();
2453 }
2454 
2455 /*
2456  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2457  * needed by tests that modify guest PTEs.
2458  */
2459 static void ept_access_test_guest_flush_tlb(void)
2460 {
2461 	do_ept_access_op(OP_FLUSH_TLB);
2462 	skip_exit_vmcall();
2463 }
2464 
2465 /*
2466  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2467  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2468  * a huge page.
2469  */
2470 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2471 				 unsigned long clear, unsigned long set)
2472 {
2473 	struct ept_access_test_data *data = &ept_access_test_data;
2474 	unsigned long orig_pte;
2475 	unsigned long pte;
2476 
2477 	/* Screw with the mapping at the requested level. */
2478 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2479 	pte = orig_pte;
2480 	if (mkhuge)
2481 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2482 	else
2483 		pte = orig_pte;
2484 	pte = (pte & ~clear) | set;
2485 	set_ept_pte(pml4, gpa, level, pte);
2486 	ept_sync(INVEPT_SINGLE, eptp);
2487 
2488 	return orig_pte;
2489 }
2490 
2491 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2492 {
2493 	set_ept_pte(pml4, gpa, level, orig_pte);
2494 	ept_sync(INVEPT_SINGLE, eptp);
2495 }
2496 
2497 static void do_ept_violation(bool leaf, enum ept_access_op op,
2498 			     u64 expected_qual, u64 expected_paddr)
2499 {
2500 	u64 qual;
2501 
2502 	/* Try the access and observe the violation. */
2503 	do_ept_access_op(op);
2504 
2505 	assert_exit_reason(VMX_EPT_VIOLATION);
2506 
2507 	qual = vmcs_read(EXI_QUALIFICATION);
2508 
2509 	/* Mask undefined bits (which may later be defined in certain cases). */
2510 	qual &= ~(EPT_VLT_GUEST_USER | EPT_VLT_GUEST_RW | EPT_VLT_GUEST_EX |
2511 		 EPT_VLT_PERM_USER_EX);
2512 
2513 	diagnose_ept_violation_qual(expected_qual, qual);
2514 	TEST_EXPECT_EQ(expected_qual, qual);
2515 
2516 	#if 0
2517 	/* Disable for now otherwise every test will fail */
2518 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2519 		       (unsigned long) (
2520 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2521 	#endif
2522 	/*
2523 	 * TODO: tests that probe expected_paddr in pages other than the one at
2524 	 * the beginning of the 1g region.
2525 	 */
2526 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2527 }
2528 
2529 static void
2530 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2531 			      unsigned long set, enum ept_access_op op,
2532 			      u64 expected_qual)
2533 {
2534 	struct ept_access_test_data *data = &ept_access_test_data;
2535 	unsigned long orig_pte;
2536 
2537 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2538 
2539 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2540 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2541 					 data->gpa);
2542 
2543 	/* Fix the violation and resume the op loop. */
2544 	ept_untwiddle(data->gpa, level, orig_pte);
2545 	enter_guest();
2546 	skip_exit_vmcall();
2547 }
2548 
2549 static void
2550 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2551 		       enum ept_access_op op, u64 expected_qual)
2552 {
2553 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2554 				      expected_qual);
2555 	if (ept_huge_pages_supported(level))
2556 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2557 					      expected_qual);
2558 }
2559 
2560 static void ept_violation(unsigned long clear, unsigned long set,
2561 			  enum ept_access_op op, u64 expected_qual)
2562 {
2563 	ept_violation_at_level(1, clear, set, op, expected_qual);
2564 	ept_violation_at_level(2, clear, set, op, expected_qual);
2565 	ept_violation_at_level(3, clear, set, op, expected_qual);
2566 	ept_violation_at_level(4, clear, set, op, expected_qual);
2567 }
2568 
2569 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2570 				       u64 expected_qual)
2571 {
2572 	ept_violation(EPT_PRESENT, access, op,
2573 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2574 }
2575 
2576 /*
2577  * For translations that don't involve a GVA, that is physical address (paddr)
2578  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2579  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2580  * translations don't involve GVAs, such as when the hardware does the guest
2581  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2582  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2583  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2584  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2585  *
2586  * Note that paddr violations can also be triggered by loading PAE page tables
2587  * with wonky addresses. We don't test that yet.
2588  *
2589  * This function modifies the EPT entry that maps the GPA that the guest page
2590  * table entry mapping ept_access_test_data.gva resides on.
2591  *
2592  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2593  *
2594  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2595  *
2596  *	@op		Guest operation to perform with
2597  *			ept_access_test_data.gva.
2598  *
2599  *	@expect_violation
2600  *			Is a violation expected during the paddr access?
2601  *
2602  *	@expected_qual	Expected qualification for the EPT violation.
2603  *			EPT_VLT_PADDR should be clear.
2604  */
2605 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2606 			     enum ept_access_op op, bool expect_violation,
2607 			     u64 expected_qual)
2608 {
2609 	struct ept_access_test_data *data = &ept_access_test_data;
2610 	unsigned long *ptep;
2611 	unsigned long gpa;
2612 	unsigned long orig_epte;
2613 	unsigned long epte;
2614 	int i;
2615 
2616 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2617 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2618 	TEST_ASSERT(ptep);
2619 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2620 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2621 	ept_access_test_guest_flush_tlb();
2622 
2623 	/*
2624 	 * Now modify the access bits on the EPT entry for the GPA that the
2625 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2626 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2627 	 * constructed our test such that those other 511 PTEs aren't used by
2628 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2629 	 * PTE we're modifying is at the beginning of a 4K page and the
2630 	 * following 511 entires are also under our control (and not touched by
2631 	 * the guest).
2632 	 */
2633 	gpa = virt_to_phys(ptep);
2634 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2635 	/*
2636 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2637 	 * otherwise our level=1 twiddling below will fail. We use the
2638 	 * identity map (gpa = gpa) since page tables are shared with the host.
2639 	 */
2640 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2641 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2642 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2643 
2644 	if (expect_violation) {
2645 		do_ept_violation(/*leaf=*/true, op,
2646 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2647 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2648 		do_ept_access_op(op);
2649 	} else {
2650 		do_ept_access_op(op);
2651 		if (ept_ad_enabled()) {
2652 			for (i = EPT_PAGE_LEVEL; i > 0; i--) {
2653 				TEST_ASSERT(get_ept_pte(pml4, gpa, i, &epte));
2654 				TEST_ASSERT(epte & EPT_ACCESS_FLAG);
2655 				if (i == 1)
2656 					TEST_ASSERT(epte & EPT_DIRTY_FLAG);
2657 				else
2658 					TEST_ASSERT_EQ(epte & EPT_DIRTY_FLAG, 0);
2659 			}
2660 		}
2661 
2662 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2663 	}
2664 
2665 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2666 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2667 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2668 
2669 	skip_exit_vmcall();
2670 }
2671 
2672 static void ept_access_allowed_paddr(unsigned long ept_access,
2673 				     unsigned long pte_ad,
2674 				     enum ept_access_op op)
2675 {
2676 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2677 			 /*expected_qual=*/-1);
2678 }
2679 
2680 static void ept_access_violation_paddr(unsigned long ept_access,
2681 				       unsigned long pte_ad,
2682 				       enum ept_access_op op,
2683 				       u64 expected_qual)
2684 {
2685 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2686 			 expected_qual);
2687 }
2688 
2689 
2690 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2691 					unsigned long clear,
2692 					unsigned long set,
2693 					enum ept_access_op op)
2694 {
2695 	struct ept_access_test_data *data = &ept_access_test_data;
2696 	unsigned long orig_pte;
2697 
2698 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2699 
2700 	/* No violation. Should proceed to vmcall. */
2701 	do_ept_access_op(op);
2702 	skip_exit_vmcall();
2703 
2704 	ept_untwiddle(data->gpa, level, orig_pte);
2705 }
2706 
2707 static void ept_allowed_at_level(int level, unsigned long clear,
2708 				 unsigned long set, enum ept_access_op op)
2709 {
2710 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2711 	if (ept_huge_pages_supported(level))
2712 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2713 }
2714 
2715 static void ept_allowed(unsigned long clear, unsigned long set,
2716 			enum ept_access_op op)
2717 {
2718 	ept_allowed_at_level(1, clear, set, op);
2719 	ept_allowed_at_level(2, clear, set, op);
2720 	ept_allowed_at_level(3, clear, set, op);
2721 	ept_allowed_at_level(4, clear, set, op);
2722 }
2723 
2724 static void ept_ignored_bit(int bit)
2725 {
2726 	/* Set the bit. */
2727 	ept_allowed(0, 1ul << bit, OP_READ);
2728 	ept_allowed(0, 1ul << bit, OP_WRITE);
2729 	ept_allowed(0, 1ul << bit, OP_EXEC);
2730 
2731 	/* Clear the bit. */
2732 	ept_allowed(1ul << bit, 0, OP_READ);
2733 	ept_allowed(1ul << bit, 0, OP_WRITE);
2734 	ept_allowed(1ul << bit, 0, OP_EXEC);
2735 }
2736 
2737 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2738 {
2739 	ept_allowed(EPT_PRESENT, access, op);
2740 }
2741 
2742 
2743 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2744 					     unsigned long clear,
2745 					     unsigned long set,
2746 					     enum ept_access_op op)
2747 {
2748 	struct ept_access_test_data *data = &ept_access_test_data;
2749 	unsigned long orig_pte;
2750 
2751 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2752 
2753 	do_ept_access_op(op);
2754 	assert_exit_reason(VMX_EPT_MISCONFIG);
2755 
2756 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2757 	#if 0
2758 	/* broken: */
2759 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2760 	#endif
2761 	#if 0
2762 	/*
2763 	 * broken:
2764 	 * According to description of exit qual for EPT violation,
2765 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2766 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2767 	 * to be set for msiconfig.
2768 	 */
2769 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2770 		       (unsigned long) (
2771 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2772 	#endif
2773 
2774 	/* Fix the violation and resume the op loop. */
2775 	ept_untwiddle(data->gpa, level, orig_pte);
2776 	enter_guest();
2777 	skip_exit_vmcall();
2778 }
2779 
2780 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2781 					  unsigned long clear,
2782 					  unsigned long set)
2783 {
2784 	/* The op shouldn't matter (read, write, exec), so try them all! */
2785 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2786 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2787 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2788 }
2789 
2790 static void ept_misconfig_at_level(int level, unsigned long clear,
2791 				   unsigned long set)
2792 {
2793 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2794 	if (ept_huge_pages_supported(level))
2795 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2796 }
2797 
2798 static void ept_misconfig(unsigned long clear, unsigned long set)
2799 {
2800 	ept_misconfig_at_level(1, clear, set);
2801 	ept_misconfig_at_level(2, clear, set);
2802 	ept_misconfig_at_level(3, clear, set);
2803 	ept_misconfig_at_level(4, clear, set);
2804 }
2805 
2806 static void ept_access_misconfig(unsigned long access)
2807 {
2808 	ept_misconfig(EPT_PRESENT, access);
2809 }
2810 
2811 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2812 {
2813 	/* Setting the bit causes a misconfig. */
2814 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2815 
2816 	/* Making the entry non-present turns reserved bits into ignored. */
2817 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2818 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2819 }
2820 
2821 static void ept_reserved_bit_at_level_huge(int level, int bit)
2822 {
2823 	/* Setting the bit causes a misconfig. */
2824 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2825 
2826 	/* Making the entry non-present turns reserved bits into ignored. */
2827 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2828 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2829 }
2830 
2831 static void ept_reserved_bit_at_level(int level, int bit)
2832 {
2833 	/* Setting the bit causes a misconfig. */
2834 	ept_misconfig_at_level(level, 0, 1ul << bit);
2835 
2836 	/* Making the entry non-present turns reserved bits into ignored. */
2837 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2838 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2839 }
2840 
2841 static void ept_reserved_bit(int bit)
2842 {
2843 	ept_reserved_bit_at_level(1, bit);
2844 	ept_reserved_bit_at_level(2, bit);
2845 	ept_reserved_bit_at_level(3, bit);
2846 	ept_reserved_bit_at_level(4, bit);
2847 }
2848 
2849 #define PAGE_2M_ORDER 9
2850 #define PAGE_1G_ORDER 18
2851 
2852 static void *get_1g_page(void)
2853 {
2854 	static void *alloc;
2855 
2856 	if (!alloc)
2857 		alloc = alloc_pages(PAGE_1G_ORDER);
2858 	return alloc;
2859 }
2860 
2861 static void ept_access_test_teardown(void *unused)
2862 {
2863 	/* Exit the guest cleanly. */
2864 	do_ept_access_op(OP_EXIT);
2865 }
2866 
2867 static void ept_access_test_guest(void)
2868 {
2869 	struct ept_access_test_data *data = &ept_access_test_data;
2870 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2871 
2872 	while (true) {
2873 		switch (data->op) {
2874 		case OP_READ:
2875 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2876 			break;
2877 		case OP_WRITE:
2878 			*data->gva = MAGIC_VAL_2;
2879 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2880 			*data->gva = MAGIC_VAL_1;
2881 			break;
2882 		case OP_EXEC:
2883 			TEST_ASSERT_EQ(42, code());
2884 			break;
2885 		case OP_FLUSH_TLB:
2886 			write_cr3(read_cr3());
2887 			break;
2888 		case OP_EXIT:
2889 			return;
2890 		default:
2891 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2892 		}
2893 		vmcall();
2894 	}
2895 }
2896 
2897 static void ept_access_test_setup(void)
2898 {
2899 	struct ept_access_test_data *data = &ept_access_test_data;
2900 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2901 	unsigned long size = npages * PAGE_SIZE;
2902 	unsigned long *page_table = current_page_table();
2903 	unsigned long pte;
2904 
2905 	if (setup_ept(false))
2906 		test_skip("EPT not supported");
2907 
2908 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2909 	if (cpuid_maxphyaddr() < 40)
2910 		test_skip("Test needs MAXPHYADDR >= 40");
2911 
2912 	test_set_guest(ept_access_test_guest);
2913 	test_add_teardown(ept_access_test_teardown, NULL);
2914 
2915 	data->hva = get_1g_page();
2916 	TEST_ASSERT(data->hva);
2917 	data->hpa = virt_to_phys(data->hva);
2918 
2919 	data->gpa = 1ul << 39;
2920 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2921 				   size);
2922 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2923 	install_pages(page_table, data->gpa, size, data->gva);
2924 
2925 	/*
2926 	 * Make sure nothing's mapped here so the tests that screw with the
2927 	 * pml4 entry don't inadvertently break something.
2928 	 */
2929 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2930 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2931 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2932 
2933 	data->hva[0] = MAGIC_VAL_1;
2934 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2935 }
2936 
2937 static void ept_access_test_not_present(void)
2938 {
2939 	ept_access_test_setup();
2940 	/* --- */
2941 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2942 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2943 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2944 }
2945 
2946 static void ept_access_test_read_only(void)
2947 {
2948 	ept_access_test_setup();
2949 
2950 	/* r-- */
2951 	ept_access_allowed(EPT_RA, OP_READ);
2952 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2953 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2954 }
2955 
2956 static void ept_access_test_write_only(void)
2957 {
2958 	ept_access_test_setup();
2959 	/* -w- */
2960 	ept_access_misconfig(EPT_WA);
2961 }
2962 
2963 static void ept_access_test_read_write(void)
2964 {
2965 	ept_access_test_setup();
2966 	/* rw- */
2967 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2968 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2969 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2970 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2971 }
2972 
2973 
2974 static void ept_access_test_execute_only(void)
2975 {
2976 	ept_access_test_setup();
2977 	/* --x */
2978 	if (ept_execute_only_supported()) {
2979 		ept_access_violation(EPT_EA, OP_READ,
2980 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2981 		ept_access_violation(EPT_EA, OP_WRITE,
2982 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2983 		ept_access_allowed(EPT_EA, OP_EXEC);
2984 	} else {
2985 		ept_access_misconfig(EPT_EA);
2986 	}
2987 }
2988 
2989 static void ept_access_test_read_execute(void)
2990 {
2991 	ept_access_test_setup();
2992 	/* r-x */
2993 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2994 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2995 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2996 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2997 }
2998 
2999 static void ept_access_test_write_execute(void)
3000 {
3001 	ept_access_test_setup();
3002 	/* -wx */
3003 	ept_access_misconfig(EPT_WA | EPT_EA);
3004 }
3005 
3006 static void ept_access_test_read_write_execute(void)
3007 {
3008 	ept_access_test_setup();
3009 	/* rwx */
3010 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
3011 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
3012 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
3013 }
3014 
3015 static void ept_access_test_reserved_bits(void)
3016 {
3017 	int i;
3018 	int maxphyaddr;
3019 
3020 	ept_access_test_setup();
3021 
3022 	/* Reserved bits above maxphyaddr. */
3023 	maxphyaddr = cpuid_maxphyaddr();
3024 	for (i = maxphyaddr; i <= 51; i++) {
3025 		report_prefix_pushf("reserved_bit=%d", i);
3026 		ept_reserved_bit(i);
3027 		report_prefix_pop();
3028 	}
3029 
3030 	/* Level-specific reserved bits. */
3031 	ept_reserved_bit_at_level_nohuge(2, 3);
3032 	ept_reserved_bit_at_level_nohuge(2, 4);
3033 	ept_reserved_bit_at_level_nohuge(2, 5);
3034 	ept_reserved_bit_at_level_nohuge(2, 6);
3035 	/* 2M alignment. */
3036 	for (i = 12; i < 20; i++) {
3037 		report_prefix_pushf("reserved_bit=%d", i);
3038 		ept_reserved_bit_at_level_huge(2, i);
3039 		report_prefix_pop();
3040 	}
3041 	ept_reserved_bit_at_level_nohuge(3, 3);
3042 	ept_reserved_bit_at_level_nohuge(3, 4);
3043 	ept_reserved_bit_at_level_nohuge(3, 5);
3044 	ept_reserved_bit_at_level_nohuge(3, 6);
3045 	/* 1G alignment. */
3046 	for (i = 12; i < 29; i++) {
3047 		report_prefix_pushf("reserved_bit=%d", i);
3048 		ept_reserved_bit_at_level_huge(3, i);
3049 		report_prefix_pop();
3050 	}
3051 	ept_reserved_bit_at_level(4, 3);
3052 	ept_reserved_bit_at_level(4, 4);
3053 	ept_reserved_bit_at_level(4, 5);
3054 	ept_reserved_bit_at_level(4, 6);
3055 	ept_reserved_bit_at_level(4, 7);
3056 }
3057 
3058 static void ept_access_test_ignored_bits(void)
3059 {
3060 	ept_access_test_setup();
3061 	/*
3062 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
3063 	 * far as translation is concerned even if AD bits are enabled in the
3064 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
3065 	 * control is 0.
3066 	 */
3067 	ept_ignored_bit(8);
3068 	ept_ignored_bit(9);
3069 	ept_ignored_bit(10);
3070 	ept_ignored_bit(11);
3071 	ept_ignored_bit(52);
3072 	ept_ignored_bit(53);
3073 	ept_ignored_bit(54);
3074 	ept_ignored_bit(55);
3075 	ept_ignored_bit(56);
3076 	ept_ignored_bit(57);
3077 	ept_ignored_bit(58);
3078 	ept_ignored_bit(59);
3079 	ept_ignored_bit(60);
3080 	ept_ignored_bit(61);
3081 	ept_ignored_bit(62);
3082 	ept_ignored_bit(63);
3083 }
3084 
3085 static void ept_access_test_paddr_not_present_ad_disabled(void)
3086 {
3087 	ept_access_test_setup();
3088 	ept_disable_ad_bits();
3089 
3090 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
3091 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
3092 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
3093 }
3094 
3095 static void ept_access_test_paddr_not_present_ad_enabled(void)
3096 {
3097 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
3098 
3099 	ept_access_test_setup();
3100 	ept_enable_ad_bits_or_skip_test();
3101 
3102 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
3103 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
3104 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
3105 }
3106 
3107 static void ept_access_test_paddr_read_only_ad_disabled(void)
3108 {
3109 	/*
3110 	 * When EPT AD bits are disabled, all accesses to guest paging
3111 	 * structures are reported separately as a read and (after
3112 	 * translation of the GPA to host physical address) a read+write
3113 	 * if the A/D bits have to be set.
3114 	 */
3115 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3116 
3117 	ept_access_test_setup();
3118 	ept_disable_ad_bits();
3119 
3120 	/* Can't update A bit, so all accesses fail. */
3121 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3122 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3123 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3124 	/* AD bits disabled, so only writes try to update the D bit. */
3125 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
3126 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3127 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
3128 	/* Both A and D already set, so read-only is OK. */
3129 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
3130 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
3131 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
3132 }
3133 
3134 static void ept_access_test_paddr_read_only_ad_enabled(void)
3135 {
3136 	/*
3137 	 * When EPT AD bits are enabled, all accesses to guest paging
3138 	 * structures are considered writes as far as EPT translation
3139 	 * is concerned.
3140 	 */
3141 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3142 
3143 	ept_access_test_setup();
3144 	ept_enable_ad_bits_or_skip_test();
3145 
3146 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3147 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3148 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3149 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
3150 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3151 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
3152 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
3153 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
3154 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
3155 }
3156 
3157 static void ept_access_test_paddr_read_write(void)
3158 {
3159 	ept_access_test_setup();
3160 	/* Read-write access to paging structure. */
3161 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
3162 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
3163 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
3164 }
3165 
3166 static void ept_access_test_paddr_read_write_execute(void)
3167 {
3168 	ept_access_test_setup();
3169 	/* RWX access to paging structure. */
3170 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
3171 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
3172 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
3173 }
3174 
3175 static void ept_access_test_paddr_read_execute_ad_disabled(void)
3176 {
3177   	/*
3178 	 * When EPT AD bits are disabled, all accesses to guest paging
3179 	 * structures are reported separately as a read and (after
3180 	 * translation of the GPA to host physical address) a read+write
3181 	 * if the A/D bits have to be set.
3182 	 */
3183 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3184 
3185 	ept_access_test_setup();
3186 	ept_disable_ad_bits();
3187 
3188 	/* Can't update A bit, so all accesses fail. */
3189 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3190 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3191 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3192 	/* AD bits disabled, so only writes try to update the D bit. */
3193 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
3194 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3195 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3196 	/* Both A and D already set, so read-only is OK. */
3197 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3198 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3199 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3200 }
3201 
3202 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3203 {
3204 	/*
3205 	 * When EPT AD bits are enabled, all accesses to guest paging
3206 	 * structures are considered writes as far as EPT translation
3207 	 * is concerned.
3208 	 */
3209 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3210 
3211 	ept_access_test_setup();
3212 	ept_enable_ad_bits_or_skip_test();
3213 
3214 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3215 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3216 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3217 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3218 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3219 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3220 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3221 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3222 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3223 }
3224 
3225 static void ept_access_test_paddr_not_present_page_fault(void)
3226 {
3227 	ept_access_test_setup();
3228 	/*
3229 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3230 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3231 	 * Thus guest page fault before host takes EPT violation for trying to
3232 	 * update A bit.
3233 	 */
3234 }
3235 
3236 static void ept_access_test_force_2m_page(void)
3237 {
3238 	ept_access_test_setup();
3239 
3240 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3241 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3242 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3243 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3244 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3245 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3246 }
3247 
3248 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3249 {
3250 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3251 
3252 	TEST_ASSERT(msr & VPID_CAP_INVVPID);
3253 
3254 	if (type < INVVPID_ADDR || type > INVVPID_CONTEXT_LOCAL)
3255 		return false;
3256 
3257 	if (!(msr & (1ull << (type + VPID_CAP_INVVPID_TYPES_SHIFT))))
3258 		return false;
3259 
3260 	if (vpid >> 16)
3261 		return false;
3262 
3263 	if (type != INVVPID_ALL && !vpid)
3264 		return false;
3265 
3266 	if (type == INVVPID_ADDR && !is_canonical(gla))
3267 		return false;
3268 
3269 	return true;
3270 }
3271 
3272 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3273 {
3274 	int rc;
3275 	bool valid = invvpid_valid(type, vpid, gla);
3276 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3277 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3278 	/*
3279 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3280 	 * that we can tell if it is updated by INVVPID.
3281 	 */
3282 	vmcs_read(~0);
3283 	rc = invvpid(type, vpid, gla);
3284 	report(!rc == valid, "INVVPID type %ld VPID %lx GLA %lx %s", type,
3285 	       vpid, gla,
3286 	       valid ? "passes" : "fails");
3287 	report(vmcs_read(VMX_INST_ERROR) == expected,
3288 	       "After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3289 	       rc ? "failed" : "successful",
3290 	       expected, vmcs_read(VMX_INST_ERROR));
3291 }
3292 
3293 static void ds_invvpid(void *data)
3294 {
3295 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3296 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3297 
3298 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3299 	asm volatile("invvpid %0, %1"
3300 		     :
3301 		     : "m"(*(struct invvpid_operand *)data),
3302 		       "r"(type));
3303 }
3304 
3305 /*
3306  * The SS override is ignored in 64-bit mode, so we use an addressing
3307  * mode with %rsp as the base register to generate an implicit SS
3308  * reference.
3309  */
3310 static void ss_invvpid(void *data)
3311 {
3312 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3313 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3314 
3315 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3316 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3317 		     : "+r"(data)
3318 		     : "r"(type));
3319 }
3320 
3321 static void invvpid_test_gp(void)
3322 {
3323 	bool fault;
3324 
3325 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3326 				   (void *)NONCANONICAL);
3327 	report(fault, "INVVPID with non-canonical DS operand raises #GP");
3328 }
3329 
3330 static void invvpid_test_ss(void)
3331 {
3332 	bool fault;
3333 
3334 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3335 				   (void *)NONCANONICAL);
3336 	report(fault, "INVVPID with non-canonical SS operand raises #SS");
3337 }
3338 
3339 static void invvpid_test_pf(void)
3340 {
3341 	void *vpage = alloc_vpage();
3342 	bool fault;
3343 
3344 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3345 	report(fault, "INVVPID with unmapped operand raises #PF");
3346 }
3347 
3348 static void try_compat_invvpid(void *unused)
3349 {
3350 	struct far_pointer32 fp = {
3351 		.offset = (uintptr_t)&&invvpid,
3352 		.selector = KERNEL_CS32,
3353 	};
3354 	uintptr_t rsp;
3355 
3356 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
3357 
3358 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3359 			"Code address too high.");
3360 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3361 
3362 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3363 	return;
3364 invvpid:
3365 	asm volatile (".code32;"
3366 		      "invvpid (%eax), %eax;"
3367 		      "lret;"
3368 		      ".code64");
3369 	__builtin_unreachable();
3370 }
3371 
3372 static void invvpid_test_compatibility_mode(void)
3373 {
3374 	bool fault;
3375 
3376 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3377 	report(fault, "Compatibility mode INVVPID raises #UD");
3378 }
3379 
3380 static void invvpid_test_not_in_vmx_operation(void)
3381 {
3382 	bool fault;
3383 
3384 	TEST_ASSERT(!vmx_off());
3385 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3386 	report(fault, "INVVPID outside of VMX operation raises #UD");
3387 	TEST_ASSERT(!vmx_on());
3388 }
3389 
3390 /*
3391  * This does not test real-address mode, virtual-8086 mode, protected mode,
3392  * or CPL > 0.
3393  */
3394 static void invvpid_test_v2(void)
3395 {
3396 	u64 msr;
3397 	int i;
3398 	unsigned types = 0;
3399 	unsigned type;
3400 
3401 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
3402 	    !(ctrl_cpu_rev[1].clr & CPU_VPID))
3403 		test_skip("VPID not supported");
3404 
3405 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3406 
3407 	if (!(msr & VPID_CAP_INVVPID))
3408 		test_skip("INVVPID not supported.\n");
3409 
3410 	if (msr & VPID_CAP_INVVPID_ADDR)
3411 		types |= 1u << INVVPID_ADDR;
3412 	if (msr & VPID_CAP_INVVPID_CXTGLB)
3413 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3414 	if (msr & VPID_CAP_INVVPID_ALL)
3415 		types |= 1u << INVVPID_ALL;
3416 	if (msr & VPID_CAP_INVVPID_CXTLOC)
3417 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3418 
3419 	if (!types)
3420 		test_skip("No INVVPID types supported.\n");
3421 
3422 	for (i = -127; i < 128; i++)
3423 		try_invvpid(i, 0xffff, 0);
3424 
3425 	/*
3426 	 * VPID must not be more than 16 bits.
3427 	 */
3428 	for (i = 0; i < 64; i++)
3429 		for (type = 0; type < 4; type++)
3430 			if (types & (1u << type))
3431 				try_invvpid(type, 1ul << i, 0);
3432 
3433 	/*
3434 	 * VPID must not be zero, except for "all contexts."
3435 	 */
3436 	for (type = 0; type < 4; type++)
3437 		if (types & (1u << type))
3438 			try_invvpid(type, 0, 0);
3439 
3440 	/*
3441 	 * The gla operand is only validated for single-address INVVPID.
3442 	 */
3443 	if (types & (1u << INVVPID_ADDR))
3444 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3445 
3446 	invvpid_test_gp();
3447 	invvpid_test_ss();
3448 	invvpid_test_pf();
3449 	invvpid_test_compatibility_mode();
3450 	invvpid_test_not_in_vmx_operation();
3451 }
3452 
3453 /*
3454  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3455  * at least as far as the guest-state checks. Returns false if the
3456  * VMLAUNCH fails early and execution falls through to the next
3457  * instruction.
3458  */
3459 static bool vmlaunch_succeeds(void)
3460 {
3461 	u32 exit_reason;
3462 
3463 	/*
3464 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3465 	 * unsupported VMCS component"). The caller can then check
3466 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3467 	 */
3468 	vmcs_write(~0u, 0);
3469 
3470 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3471 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3472 				   :
3473 				   : "r" ((u64)HOST_RSP)
3474 				   : "cc", "memory"
3475 				   : success);
3476 	return false;
3477 success:
3478 	exit_reason = vmcs_read(EXI_REASON);
3479 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3480 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3481 	return true;
3482 }
3483 
3484 /*
3485  * Try to launch the current VMCS.
3486  */
3487 static void test_vmx_vmlaunch(u32 xerror)
3488 {
3489 	bool success = vmlaunch_succeeds();
3490 	u32 vmx_inst_err;
3491 
3492 	report(success == !xerror, "vmlaunch %s",
3493 	       !xerror ? "succeeds" : "fails");
3494 	if (!success && xerror) {
3495 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3496 		report(vmx_inst_err == xerror,
3497 		       "VMX inst error is %d (actual %d)", xerror,
3498 		       vmx_inst_err);
3499 	}
3500 }
3501 
3502 /*
3503  * Try to launch the current VMCS, and expect one of two possible
3504  * errors (or success) codes.
3505  */
3506 static void test_vmx_vmlaunch2(u32 xerror1, u32 xerror2)
3507 {
3508 	bool success = vmlaunch_succeeds();
3509 	u32 vmx_inst_err;
3510 
3511 	if (!xerror1 == !xerror2)
3512 		report(success == !xerror1, "vmlaunch %s",
3513 		       !xerror1 ? "succeeds" : "fails");
3514 
3515 	if (!success && (xerror1 || xerror2)) {
3516 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3517 		report(vmx_inst_err == xerror1 || vmx_inst_err == xerror2,
3518 		       "VMX inst error is %d or %d (actual %d)", xerror1,
3519 		       xerror2, vmx_inst_err);
3520 	}
3521 }
3522 
3523 static void test_vmx_invalid_controls(void)
3524 {
3525 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3526 }
3527 
3528 static void test_vmx_valid_controls(void)
3529 {
3530 	test_vmx_vmlaunch(0);
3531 }
3532 
3533 /*
3534  * Test a particular value of a VM-execution control bit, if the value
3535  * is required or if the value is zero.
3536  */
3537 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3538 				    enum Encoding encoding, unsigned bit,
3539 				    unsigned val)
3540 {
3541 	u32 mask = 1u << bit;
3542 	bool expected;
3543 	u32 controls;
3544 
3545 	if (msr.set & mask)
3546 		TEST_ASSERT(msr.clr & mask);
3547 
3548 	/*
3549 	 * We can't arbitrarily turn on a control bit, because it may
3550 	 * introduce dependencies on other VMCS fields. So, we only
3551 	 * test turning on bits that have a required setting.
3552 	 */
3553 	if (val && (msr.clr & mask) && !(msr.set & mask))
3554 		return;
3555 
3556 	report_prefix_pushf("%s %s bit %d",
3557 			    val ? "Set" : "Clear", name, bit);
3558 
3559 	controls = vmcs_read(encoding);
3560 	if (val) {
3561 		vmcs_write(encoding, msr.set | mask);
3562 		expected = (msr.clr & mask);
3563 	} else {
3564 		vmcs_write(encoding, msr.set & ~mask);
3565 		expected = !(msr.set & mask);
3566 	}
3567 	if (expected)
3568 		test_vmx_valid_controls();
3569 	else
3570 		test_vmx_invalid_controls();
3571 	vmcs_write(encoding, controls);
3572 	report_prefix_pop();
3573 }
3574 
3575 /*
3576  * Test reserved values of a VM-execution control bit, based on the
3577  * allowed bit settings from the corresponding VMX capability MSR.
3578  */
3579 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3580 			      enum Encoding encoding, unsigned bit)
3581 {
3582 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3583 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3584 }
3585 
3586 /*
3587  * Reserved bits in the pin-based VM-execution controls must be set
3588  * properly. Software may consult the VMX capability MSRs to determine
3589  * the proper settings.
3590  * [Intel SDM]
3591  */
3592 static void test_pin_based_ctls(void)
3593 {
3594 	unsigned bit;
3595 
3596 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3597 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3598 	for (bit = 0; bit < 32; bit++)
3599 		test_rsvd_ctl_bit("pin-based controls",
3600 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3601 }
3602 
3603 /*
3604  * Reserved bits in the primary processor-based VM-execution controls
3605  * must be set properly. Software may consult the VMX capability MSRs
3606  * to determine the proper settings.
3607  * [Intel SDM]
3608  */
3609 static void test_primary_processor_based_ctls(void)
3610 {
3611 	unsigned bit;
3612 
3613 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3614 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3615 	for (bit = 0; bit < 32; bit++)
3616 		test_rsvd_ctl_bit("primary processor-based controls",
3617 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3618 }
3619 
3620 /*
3621  * If the "activate secondary controls" primary processor-based
3622  * VM-execution control is 1, reserved bits in the secondary
3623  * processor-based VM-execution controls must be cleared. Software may
3624  * consult the VMX capability MSRs to determine which bits are
3625  * reserved.
3626  * If the "activate secondary controls" primary processor-based
3627  * VM-execution control is 0 (or if the processor does not support the
3628  * 1-setting of that control), no checks are performed on the
3629  * secondary processor-based VM-execution controls.
3630  * [Intel SDM]
3631  */
3632 static void test_secondary_processor_based_ctls(void)
3633 {
3634 	u32 primary;
3635 	u32 secondary;
3636 	unsigned bit;
3637 
3638 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3639 		return;
3640 
3641 	primary = vmcs_read(CPU_EXEC_CTRL0);
3642 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3643 
3644 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3645 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3646 	for (bit = 0; bit < 32; bit++)
3647 		test_rsvd_ctl_bit("secondary processor-based controls",
3648 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3649 
3650 	/*
3651 	 * When the "activate secondary controls" VM-execution control
3652 	 * is clear, there are no checks on the secondary controls.
3653 	 */
3654 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3655 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3656 	report(vmlaunch_succeeds(),
3657 	       "Secondary processor-based controls ignored");
3658 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3659 	vmcs_write(CPU_EXEC_CTRL0, primary);
3660 }
3661 
3662 static void try_cr3_target_count(unsigned i, unsigned max)
3663 {
3664 	report_prefix_pushf("CR3 target count 0x%x", i);
3665 	vmcs_write(CR3_TARGET_COUNT, i);
3666 	if (i <= max)
3667 		test_vmx_valid_controls();
3668 	else
3669 		test_vmx_invalid_controls();
3670 	report_prefix_pop();
3671 }
3672 
3673 /*
3674  * The CR3-target count must not be greater than 4. Future processors
3675  * may support a different number of CR3-target values. Software
3676  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3677  * number of values supported.
3678  * [Intel SDM]
3679  */
3680 static void test_cr3_targets(void)
3681 {
3682 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3683 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3684 	unsigned i;
3685 
3686 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3687 	TEST_ASSERT(supported_targets <= 256);
3688 
3689 	try_cr3_target_count(-1u, supported_targets);
3690 	try_cr3_target_count(0x80000000, supported_targets);
3691 	try_cr3_target_count(0x7fffffff, supported_targets);
3692 	for (i = 0; i <= supported_targets + 1; i++)
3693 		try_cr3_target_count(i, supported_targets);
3694 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3695 
3696 	/* VMWRITE to nonexistent target fields should fail. */
3697 	for (i = supported_targets; i < 256; i++)
3698 		TEST_ASSERT(vmcs_write(CR3_TARGET_0 + i*2, 0));
3699 }
3700 
3701 /*
3702  * Test a particular address setting in the VMCS
3703  */
3704 static void test_vmcs_addr(const char *name,
3705 			   enum Encoding encoding,
3706 			   u64 align,
3707 			   bool ignored,
3708 			   bool skip_beyond_mapped_ram,
3709 			   u64 addr)
3710 {
3711 	report_prefix_pushf("%s = %lx", name, addr);
3712 	vmcs_write(encoding, addr);
3713 	if (skip_beyond_mapped_ram &&
3714 	    addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3715 	    addr < (1ul << cpuid_maxphyaddr()))
3716 		printf("Skipping physical address beyond mapped RAM\n");
3717 	else if (ignored || (IS_ALIGNED(addr, align) &&
3718 	    addr < (1ul << cpuid_maxphyaddr())))
3719 		test_vmx_valid_controls();
3720 	else
3721 		test_vmx_invalid_controls();
3722 	report_prefix_pop();
3723 }
3724 
3725 /*
3726  * Test interesting values for a VMCS address
3727  */
3728 static void test_vmcs_addr_values(const char *name,
3729 				  enum Encoding encoding,
3730 				  u64 align,
3731 				  bool ignored,
3732 				  bool skip_beyond_mapped_ram,
3733 				  u32 bit_start, u32 bit_end)
3734 {
3735 	unsigned i;
3736 	u64 orig_val = vmcs_read(encoding);
3737 
3738 	for (i = bit_start; i <= bit_end; i++)
3739 		test_vmcs_addr(name, encoding, align, ignored,
3740 			       skip_beyond_mapped_ram, 1ul << i);
3741 
3742 	test_vmcs_addr(name, encoding, align, ignored,
3743 		       skip_beyond_mapped_ram, PAGE_SIZE - 1);
3744 	test_vmcs_addr(name, encoding, align, ignored,
3745 		       skip_beyond_mapped_ram, PAGE_SIZE);
3746 	test_vmcs_addr(name, encoding, align, ignored,
3747 		       skip_beyond_mapped_ram,
3748 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3749 	test_vmcs_addr(name, encoding, align, ignored,
3750 		       skip_beyond_mapped_ram, -1ul);
3751 
3752 	vmcs_write(encoding, orig_val);
3753 }
3754 
3755 /*
3756  * Test a physical address reference in the VMCS, when the corresponding
3757  * feature is enabled and when the corresponding feature is disabled.
3758  */
3759 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3760 				     const char *field_name,
3761 				     const char *control_name, u64 align,
3762 				     bool skip_beyond_mapped_ram,
3763 				     bool control_primary)
3764 {
3765 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3766 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3767 	u64 page_addr;
3768 
3769 	if (control_primary) {
3770 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3771 			return;
3772 	} else {
3773 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3774 			return;
3775 	}
3776 
3777 	page_addr = vmcs_read(field);
3778 
3779 	report_prefix_pushf("%s enabled", control_name);
3780 	if (control_primary) {
3781 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3782 	} else {
3783 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3784 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3785 	}
3786 
3787 	test_vmcs_addr_values(field_name, field, align, false,
3788 			      skip_beyond_mapped_ram, 0, 63);
3789 	report_prefix_pop();
3790 
3791 	report_prefix_pushf("%s disabled", control_name);
3792 	if (control_primary) {
3793 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3794 	} else {
3795 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3796 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3797 	}
3798 
3799 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3800 	report_prefix_pop();
3801 
3802 	vmcs_write(field, page_addr);
3803 	vmcs_write(CPU_EXEC_CTRL0, primary);
3804 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3805 }
3806 
3807 /*
3808  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3809  * each I/O-bitmap address must be 0. Neither address should set any
3810  * bits beyond the processor's physical-address width.
3811  * [Intel SDM]
3812  */
3813 static void test_io_bitmaps(void)
3814 {
3815 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3816 				 "I/O bitmap A", "Use I/O bitmaps",
3817 				 PAGE_SIZE, false, true);
3818 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3819 				 "I/O bitmap B", "Use I/O bitmaps",
3820 				 PAGE_SIZE, false, true);
3821 }
3822 
3823 /*
3824  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3825  * the MSR-bitmap address must be 0. The address should not set any
3826  * bits beyond the processor's physical-address width.
3827  * [Intel SDM]
3828  */
3829 static void test_msr_bitmap(void)
3830 {
3831 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3832 				 "MSR bitmap", "Use MSR bitmaps",
3833 				 PAGE_SIZE, false, true);
3834 }
3835 
3836 /*
3837  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3838  * address must satisfy the following checks:
3839  * - Bits 11:0 of the address must be 0.
3840  * - The address should not set any bits beyond the processor's
3841  *   physical-address width.
3842  * [Intel SDM]
3843  */
3844 static void test_apic_virt_addr(void)
3845 {
3846 	/*
3847 	 * Ensure the processor will never use the virtual-APIC page, since
3848 	 * we will point it to invalid RAM.  Otherwise KVM is puzzled about
3849 	 * what we're trying to achieve and fails vmentry.
3850 	 */
3851 	u32 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
3852 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0 | CPU_CR8_LOAD | CPU_CR8_STORE);
3853 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3854 				 "virtual-APIC address", "Use TPR shadow",
3855 				 PAGE_SIZE, false, true);
3856 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
3857 }
3858 
3859 /*
3860  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3861  * APIC-access address must satisfy the following checks:
3862  *  - Bits 11:0 of the address must be 0.
3863  *  - The address should not set any bits beyond the processor's
3864  *    physical-address width.
3865  * [Intel SDM]
3866  */
3867 static void test_apic_access_addr(void)
3868 {
3869 	void *apic_access_page = alloc_page();
3870 
3871 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3872 
3873 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3874 				 "APIC-access address",
3875 				 "virtualize APIC-accesses", PAGE_SIZE,
3876 				 true, false);
3877 }
3878 
3879 static bool set_bit_pattern(u8 mask, u32 *secondary)
3880 {
3881 	u8 i;
3882 	bool flag = false;
3883 	u32 test_bits[3] = {
3884 		CPU_VIRT_X2APIC,
3885 		CPU_APIC_REG_VIRT,
3886 		CPU_VINTD
3887 	};
3888 
3889         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3890 		if ((mask & (1u << i)) &&
3891 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3892 			*secondary |= test_bits[i];
3893 			flag = true;
3894 		}
3895 	}
3896 
3897 	return (flag);
3898 }
3899 
3900 /*
3901  * If the "use TPR shadow" VM-execution control is 0, the following
3902  * VM-execution controls must also be 0:
3903  * 	- virtualize x2APIC mode
3904  *	- APIC-register virtualization
3905  *	- virtual-interrupt delivery
3906  *    [Intel SDM]
3907  *
3908  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3909  *    "virtualize APIC accesses" VM-execution control must be 0.
3910  *    [Intel SDM]
3911  */
3912 static void test_apic_virtual_ctls(void)
3913 {
3914 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3915 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3916 	u32 primary = saved_primary;
3917 	u32 secondary = saved_secondary;
3918 	bool ctrl = false;
3919 	char str[10] = "disabled";
3920 	u8 i = 0, j;
3921 
3922 	/*
3923 	 * First test
3924 	 */
3925 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3926 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3927 		return;
3928 
3929 	primary |= CPU_SECONDARY;
3930 	primary &= ~CPU_TPR_SHADOW;
3931 	vmcs_write(CPU_EXEC_CTRL0, primary);
3932 
3933 	while (1) {
3934 		for (j = 1; j < 8; j++) {
3935 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3936 			if (primary & CPU_TPR_SHADOW) {
3937 				ctrl = true;
3938 			} else {
3939 				if (! set_bit_pattern(j, &secondary))
3940 					ctrl = true;
3941 				else
3942 					ctrl = false;
3943 			}
3944 
3945 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3946 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3947 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3948 			if (ctrl)
3949 				test_vmx_valid_controls();
3950 			else
3951 				test_vmx_invalid_controls();
3952 			report_prefix_pop();
3953 		}
3954 
3955 		if (i == 1)
3956 			break;
3957 		i++;
3958 
3959 		primary |= CPU_TPR_SHADOW;
3960 		vmcs_write(CPU_EXEC_CTRL0, primary);
3961 		strcpy(str, "enabled");
3962 	}
3963 
3964 	/*
3965 	 * Second test
3966 	 */
3967 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3968 
3969 	primary = saved_primary;
3970 	secondary = saved_secondary;
3971 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3972 		return;
3973 
3974 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3975 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3976 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3977 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3978 	test_vmx_valid_controls();
3979 	report_prefix_pop();
3980 
3981 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3982 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3983 	test_vmx_valid_controls();
3984 	report_prefix_pop();
3985 
3986 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3987 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3988 	test_vmx_invalid_controls();
3989 	report_prefix_pop();
3990 
3991 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3992 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3993 	test_vmx_valid_controls();
3994 	report_prefix_pop();
3995 
3996 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3997 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3998 }
3999 
4000 /*
4001  * If the "virtual-interrupt delivery" VM-execution control is 1, the
4002  * "external-interrupt exiting" VM-execution control must be 1.
4003  * [Intel SDM]
4004  */
4005 static void test_virtual_intr_ctls(void)
4006 {
4007 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4008 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4009 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4010 	u32 primary = saved_primary;
4011 	u32 secondary = saved_secondary;
4012 	u32 pin = saved_pin;
4013 
4014 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
4015 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
4016 		return;
4017 
4018 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
4019 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
4020 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
4021 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
4022 	test_vmx_valid_controls();
4023 	report_prefix_pop();
4024 
4025 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
4026 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
4027 	test_vmx_invalid_controls();
4028 	report_prefix_pop();
4029 
4030 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
4031 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
4032 	test_vmx_valid_controls();
4033 	report_prefix_pop();
4034 
4035 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
4036 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
4037 	test_vmx_invalid_controls();
4038 	report_prefix_pop();
4039 
4040 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4041 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4042 	vmcs_write(PIN_CONTROLS, saved_pin);
4043 }
4044 
4045 static void test_pi_desc_addr(u64 addr, bool ctrl)
4046 {
4047 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
4048 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
4049 	if (ctrl)
4050 		test_vmx_valid_controls();
4051 	else
4052 		test_vmx_invalid_controls();
4053 	report_prefix_pop();
4054 }
4055 
4056 /*
4057  * If the “process posted interrupts†VM-execution control is 1, the
4058  * following must be true:
4059  *
4060  *	- The “virtual-interrupt delivery†VM-execution control is 1.
4061  *	- The “acknowledge interrupt on exit†VM-exit control is 1.
4062  *	- The posted-interrupt notification vector has a value in the
4063  *	- range 0–255 (bits 15:8 are all 0).
4064  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
4065  *	- The posted-interrupt descriptor address does not set any bits
4066  *	  beyond the processor's physical-address width.
4067  * [Intel SDM]
4068  */
4069 static void test_posted_intr(void)
4070 {
4071 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4072 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4073 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4074 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
4075 	u32 primary = saved_primary;
4076 	u32 secondary = saved_secondary;
4077 	u32 pin = saved_pin;
4078 	u32 exit_ctl = exit_ctl_saved;
4079 	u16 vec;
4080 	int i;
4081 
4082 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
4083 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
4084 	    (ctrl_exit_rev.clr & EXI_INTA)))
4085 		return;
4086 
4087 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
4088 
4089 	/*
4090 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
4091 	 */
4092 	pin |= PIN_POST_INTR;
4093 	vmcs_write(PIN_CONTROLS, pin);
4094 	secondary &= ~CPU_VINTD;
4095 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4096 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
4097 	test_vmx_invalid_controls();
4098 	report_prefix_pop();
4099 
4100 	secondary |= CPU_VINTD;
4101 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4102 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
4103 	test_vmx_invalid_controls();
4104 	report_prefix_pop();
4105 
4106 	exit_ctl &= ~EXI_INTA;
4107 	vmcs_write(EXI_CONTROLS, exit_ctl);
4108 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
4109 	test_vmx_invalid_controls();
4110 	report_prefix_pop();
4111 
4112 	exit_ctl |= EXI_INTA;
4113 	vmcs_write(EXI_CONTROLS, exit_ctl);
4114 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4115 	test_vmx_valid_controls();
4116 	report_prefix_pop();
4117 
4118 	secondary &= ~CPU_VINTD;
4119 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4120 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
4121 	test_vmx_invalid_controls();
4122 	report_prefix_pop();
4123 
4124 	secondary |= CPU_VINTD;
4125 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4126 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4127 	test_vmx_valid_controls();
4128 	report_prefix_pop();
4129 
4130 	/*
4131 	 * Test posted-interrupt notification vector
4132 	 */
4133 	for (i = 0; i < 8; i++) {
4134 		vec = (1ul << i);
4135 		vmcs_write(PINV, vec);
4136 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4137 		test_vmx_valid_controls();
4138 		report_prefix_pop();
4139 	}
4140 	for (i = 8; i < 16; i++) {
4141 		vec = (1ul << i);
4142 		vmcs_write(PINV, vec);
4143 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4144 		test_vmx_invalid_controls();
4145 		report_prefix_pop();
4146 	}
4147 
4148 	vec &= ~(0xff << 8);
4149 	vmcs_write(PINV, vec);
4150 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4151 	test_vmx_valid_controls();
4152 	report_prefix_pop();
4153 
4154 	/*
4155 	 * Test posted-interrupt descriptor addresss
4156 	 */
4157 	for (i = 0; i < 6; i++) {
4158 		test_pi_desc_addr(1ul << i, false);
4159 	}
4160 
4161 	test_pi_desc_addr(0xf0, false);
4162 	test_pi_desc_addr(0xff, false);
4163 	test_pi_desc_addr(0x0f, false);
4164 	test_pi_desc_addr(0x8000, true);
4165 	test_pi_desc_addr(0x00, true);
4166 	test_pi_desc_addr(0xc000, true);
4167 
4168 	test_vmcs_addr_values("process-posted interrupts",
4169 			       POSTED_INTR_DESC_ADDR, 64,
4170 			       false, false, 0, 63);
4171 
4172 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4173 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4174 	vmcs_write(PIN_CONTROLS, saved_pin);
4175 }
4176 
4177 static void test_apic_ctls(void)
4178 {
4179 	test_apic_virt_addr();
4180 	test_apic_access_addr();
4181 	test_apic_virtual_ctls();
4182 	test_virtual_intr_ctls();
4183 	test_posted_intr();
4184 }
4185 
4186 /*
4187  * If the “enable VPID†VM-execution control is 1, the value of the
4188  * of the VPID VM-execution control field must not be 0000H.
4189  * [Intel SDM]
4190  */
4191 static void test_vpid(void)
4192 {
4193 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4194 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4195 	u16 vpid = 0x0000;
4196 	int i;
4197 
4198 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4199 	    (ctrl_cpu_rev[1].clr & CPU_VPID))) {
4200 		printf("Secondary controls and/or VPID not supported\n");
4201 		return;
4202 	}
4203 
4204 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
4205 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
4206 	vmcs_write(VPID, vpid);
4207 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
4208 	test_vmx_valid_controls();
4209 	report_prefix_pop();
4210 
4211 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
4212 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4213 	test_vmx_invalid_controls();
4214 	report_prefix_pop();
4215 
4216 	for (i = 0; i < 16; i++) {
4217 		vpid = (short)1 << i;;
4218 		vmcs_write(VPID, vpid);
4219 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4220 		test_vmx_valid_controls();
4221 		report_prefix_pop();
4222 	}
4223 
4224 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4225 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4226 }
4227 
4228 static void set_vtpr(unsigned vtpr)
4229 {
4230 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
4231 }
4232 
4233 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
4234 {
4235 	bool valid = true;
4236 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4237 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4238 
4239 	if ((primary & CPU_TPR_SHADOW) &&
4240 	    (!(primary & CPU_SECONDARY) ||
4241 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
4242 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
4243 
4244 	set_vtpr(vtpr);
4245 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
4246 	    threshold, (vtpr >> 4) & 0xf);
4247 	if (valid)
4248 		test_vmx_valid_controls();
4249 	else
4250 		test_vmx_invalid_controls();
4251 	report_prefix_pop();
4252 }
4253 
4254 static void test_invalid_event_injection(void)
4255 {
4256 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
4257 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
4258 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
4259 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
4260 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
4261 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4262 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4263 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4264 	u32 cnt;
4265 
4266 	/* Setup */
4267 	report_prefix_push("invalid event injection");
4268 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4269 	vmcs_write(ENT_INST_LEN, 0x00000001);
4270 
4271 	/* The field’s interruption type is not set to a reserved value. */
4272 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4273 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4274 			    "RESERVED interruption type invalid [-]",
4275 			    ent_intr_info);
4276 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4277 	test_vmx_invalid_controls();
4278 	report_prefix_pop();
4279 
4280 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4281 			DE_VECTOR;
4282 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4283 			    "RESERVED interruption type invalid [+]",
4284 			    ent_intr_info);
4285 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4286 	test_vmx_valid_controls();
4287 	report_prefix_pop();
4288 
4289 	/* If the interruption type is other event, the vector is 0. */
4290 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4291 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4292 			    "(OTHER EVENT && vector != 0) invalid [-]",
4293 			    ent_intr_info);
4294 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4295 	test_vmx_invalid_controls();
4296 	report_prefix_pop();
4297 
4298 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4299 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4300 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4301 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4302 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4303 	test_vmx_invalid_controls();
4304 	report_prefix_pop();
4305 
4306 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4307 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4308 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4309 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4310 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4311 	test_vmx_valid_controls();
4312 	report_prefix_pop();
4313 
4314 	/*
4315 	 * If the interruption type
4316 	 * is HW exception, the vector is at most 31.
4317 	 */
4318 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4319 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4320 			    "(HW exception && vector > 31) invalid [-]",
4321 			    ent_intr_info);
4322 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4323 	test_vmx_invalid_controls();
4324 	report_prefix_pop();
4325 
4326 	/*
4327 	 * deliver-error-code is 1 iff either
4328 	 * (a) the "unrestricted guest" VM-execution control is 0
4329 	 * (b) CR0.PE is set.
4330 	 */
4331 
4332 	/* Assert that unrestricted guest is disabled or unsupported */
4333 	assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
4334 	       !(secondary_save & CPU_URG));
4335 
4336 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4337 			GP_VECTOR;
4338 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4339 			    "error code <-> (!URG || prot_mode) [-]",
4340 			    ent_intr_info);
4341 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4342 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4343 	test_vmx_invalid_controls();
4344 	report_prefix_pop();
4345 
4346 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4347 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4348 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4349 			    "error code <-> (!URG || prot_mode) [+]",
4350 			    ent_intr_info);
4351 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4352 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4353 	test_vmx_valid_controls();
4354 	report_prefix_pop();
4355 
4356 	if (enable_unrestricted_guest())
4357 		goto skip_unrestricted_guest;
4358 
4359 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4360 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4361 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4362 			    "error code <-> (!URG || prot_mode) [-]",
4363 			    ent_intr_info);
4364 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4365 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4366 	test_vmx_invalid_controls();
4367 	report_prefix_pop();
4368 
4369 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4370 			GP_VECTOR;
4371 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4372 			    "error code <-> (!URG || prot_mode) [-]",
4373 			    ent_intr_info);
4374 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4375 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4376 	test_vmx_invalid_controls();
4377 	report_prefix_pop();
4378 
4379 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4380 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4381 
4382 skip_unrestricted_guest:
4383 	vmcs_write(GUEST_CR0, guest_cr0_save);
4384 
4385 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4386 	report_prefix_push("error code <-> HW exception");
4387 	for (cnt = 0; cnt < 8; cnt++) {
4388 		u32 exception_type_mask = cnt << 8;
4389 		u32 deliver_error_code_mask =
4390 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4391 			INTR_INFO_DELIVER_CODE_MASK : 0;
4392 
4393 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4394 				exception_type_mask | GP_VECTOR;
4395 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4396 				    ent_intr_info);
4397 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4398 		test_vmx_invalid_controls();
4399 		report_prefix_pop();
4400 	}
4401 	report_prefix_pop();
4402 
4403 	/*
4404 	 * deliver-error-code is 1 iff the the vector
4405 	 * indicates an exception that would normally deliver an error code
4406 	 */
4407 	report_prefix_push("error code <-> vector delivers error code");
4408 	for (cnt = 0; cnt < 32; cnt++) {
4409 		bool has_error_code = false;
4410 		u32 deliver_error_code_mask;
4411 
4412 		switch (cnt) {
4413 		case DF_VECTOR:
4414 		case TS_VECTOR:
4415 		case NP_VECTOR:
4416 		case SS_VECTOR:
4417 		case GP_VECTOR:
4418 		case PF_VECTOR:
4419 		case AC_VECTOR:
4420 			has_error_code = true;
4421 		}
4422 
4423 		/* Negative case */
4424 		deliver_error_code_mask = has_error_code ?
4425 						0 :
4426 						INTR_INFO_DELIVER_CODE_MASK;
4427 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4428 				INTR_TYPE_HARD_EXCEPTION | cnt;
4429 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4430 				    ent_intr_info);
4431 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4432 		test_vmx_invalid_controls();
4433 		report_prefix_pop();
4434 
4435 		/* Positive case */
4436 		deliver_error_code_mask = has_error_code ?
4437 						INTR_INFO_DELIVER_CODE_MASK :
4438 						0;
4439 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4440 				INTR_TYPE_HARD_EXCEPTION | cnt;
4441 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4442 				    ent_intr_info);
4443 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4444 		test_vmx_valid_controls();
4445 		report_prefix_pop();
4446 	}
4447 	report_prefix_pop();
4448 
4449 	/* Reserved bits in the field (30:12) are 0. */
4450 	report_prefix_push("reserved bits clear");
4451 	for (cnt = 12; cnt <= 30; cnt++) {
4452 		ent_intr_info = ent_intr_info_base |
4453 				INTR_INFO_DELIVER_CODE_MASK |
4454 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4455 				(1U << cnt);
4456 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4457 				    ent_intr_info);
4458 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4459 		test_vmx_invalid_controls();
4460 		report_prefix_pop();
4461 	}
4462 	report_prefix_pop();
4463 
4464 	/*
4465 	 * If deliver-error-code is 1
4466 	 * bits 31:16 of the VM-entry exception error-code field are 0.
4467 	 */
4468 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4469 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4470 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4471 			    "VM-entry exception error code[31:16] clear",
4472 			    ent_intr_info);
4473 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4474 	for (cnt = 16; cnt <= 31; cnt++) {
4475 		ent_intr_err = 1U << cnt;
4476 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4477 				    ent_intr_err);
4478 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4479 		test_vmx_invalid_controls();
4480 		report_prefix_pop();
4481 	}
4482 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4483 	report_prefix_pop();
4484 
4485 	/*
4486 	 * If the interruption type is software interrupt, software exception,
4487 	 * or privileged software exception, the VM-entry instruction-length
4488 	 * field is in the range 0–15.
4489 	 */
4490 
4491 	for (cnt = 0; cnt < 3; cnt++) {
4492 		switch (cnt) {
4493 		case 0:
4494 			ent_intr_info = ent_intr_info_base |
4495 					INTR_TYPE_SOFT_INTR;
4496 			break;
4497 		case 1:
4498 			ent_intr_info = ent_intr_info_base |
4499 					INTR_TYPE_SOFT_EXCEPTION;
4500 			break;
4501 		case 2:
4502 			ent_intr_info = ent_intr_info_base |
4503 					INTR_TYPE_PRIV_SW_EXCEPTION;
4504 			break;
4505 		}
4506 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4507 				    "VM-entry instruction-length check",
4508 				    ent_intr_info);
4509 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4510 
4511 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4512 		ent_intr_len = -1;
4513 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4514 				    ent_intr_len);
4515 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4516 		test_vmx_invalid_controls();
4517 		report_prefix_pop();
4518 
4519 		/* Instruction length set to 16 should fail */
4520 		ent_intr_len = 0x00000010;
4521 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4522 				    ent_intr_len);
4523 		vmcs_write(ENT_INST_LEN, 0x00000010);
4524 		test_vmx_invalid_controls();
4525 		report_prefix_pop();
4526 
4527 		report_prefix_pop();
4528 	}
4529 
4530 	/* Cleanup */
4531 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4532 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4533 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4534 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4535 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4536 	vmcs_write(GUEST_CR0, guest_cr0_save);
4537 	report_prefix_pop();
4538 }
4539 
4540 /*
4541  * Test interesting vTPR values for a given TPR threshold.
4542  */
4543 static void test_vtpr_values(unsigned threshold)
4544 {
4545 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4546 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4547 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4548 }
4549 
4550 static void try_tpr_threshold(unsigned threshold)
4551 {
4552 	bool valid = true;
4553 
4554 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4555 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4556 
4557 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4558 	    (secondary & CPU_VINTD)))
4559 		valid = !(threshold >> 4);
4560 
4561 	set_vtpr(-1);
4562 	vmcs_write(TPR_THRESHOLD, threshold);
4563 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4564 	if (valid)
4565 		test_vmx_valid_controls();
4566 	else
4567 		test_vmx_invalid_controls();
4568 	report_prefix_pop();
4569 
4570 	if (valid)
4571 		test_vtpr_values(threshold);
4572 }
4573 
4574 /*
4575  * Test interesting TPR threshold values.
4576  */
4577 static void test_tpr_threshold_values(void)
4578 {
4579 	unsigned i;
4580 
4581 	for (i = 0; i < 0x10; i++)
4582 		try_tpr_threshold(i);
4583 	for (i = 4; i < 32; i++)
4584 		try_tpr_threshold(1u << i);
4585 	try_tpr_threshold(-1u);
4586 	try_tpr_threshold(0x7fffffff);
4587 }
4588 
4589 /*
4590  * This test covers the following two VM entry checks:
4591  *
4592  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4593  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4594  *         31:4 of the TPR threshold VM-execution control field must
4595 	   be 0.
4596  *         [Intel SDM]
4597  *
4598  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4599  *          "virtual-interrupt delivery" VM-execution control is 0
4600  *          and the "virtualize APIC accesses" VM-execution control
4601  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4602  *          control field must not be greater than the value of bits
4603  *          7:4 of VTPR.
4604  *          [Intel SDM]
4605  */
4606 static void test_tpr_threshold(void)
4607 {
4608 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4609 	u64 apic_virt_addr = vmcs_read(APIC_VIRT_ADDR);
4610 	u64 threshold = vmcs_read(TPR_THRESHOLD);
4611 	void *virtual_apic_page;
4612 
4613 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4614 		return;
4615 
4616 	virtual_apic_page = alloc_page();
4617 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4618 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4619 
4620 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4621 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4622 	test_tpr_threshold_values();
4623 	report_prefix_pop();
4624 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4625 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4626 	test_tpr_threshold_values();
4627 	report_prefix_pop();
4628 
4629 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4630 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES))))
4631 		goto out;
4632 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4633 
4634 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4635 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4636 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4637 		test_tpr_threshold_values();
4638 		report_prefix_pop();
4639 
4640 		vmcs_write(CPU_EXEC_CTRL0,
4641 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4642 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4643 		test_tpr_threshold_values();
4644 		report_prefix_pop();
4645 	}
4646 
4647 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4648 		vmcs_write(CPU_EXEC_CTRL0,
4649 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4650 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4651 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4652 		test_tpr_threshold_values();
4653 		report_prefix_pop();
4654 
4655 		vmcs_write(CPU_EXEC_CTRL0,
4656 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4657 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4658 		test_tpr_threshold_values();
4659 		report_prefix_pop();
4660 	}
4661 
4662 	if ((ctrl_cpu_rev[1].clr &
4663 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4664 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4665 		vmcs_write(CPU_EXEC_CTRL0,
4666 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4667 		vmcs_write(CPU_EXEC_CTRL1,
4668 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4669 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4670 		test_tpr_threshold_values();
4671 		report_prefix_pop();
4672 
4673 		vmcs_write(CPU_EXEC_CTRL0,
4674 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4675 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4676 		test_tpr_threshold_values();
4677 		report_prefix_pop();
4678 	}
4679 
4680 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4681 out:
4682 	vmcs_write(TPR_THRESHOLD, threshold);
4683 	vmcs_write(APIC_VIRT_ADDR, apic_virt_addr);
4684 	vmcs_write(CPU_EXEC_CTRL0, primary);
4685 }
4686 
4687 /*
4688  * This test verifies the following two vmentry checks:
4689  *
4690  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4691  *  VM-execution control must be 0.
4692  *  [Intel SDM]
4693  *
4694  *  If the “virtual NMIs” VM-execution control is 0, the “NMI-window
4695  *  exiting” VM-execution control must be 0.
4696  *  [Intel SDM]
4697  */
4698 static void test_nmi_ctrls(void)
4699 {
4700 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4701 
4702 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4703 	    (PIN_NMI | PIN_VIRT_NMI)) {
4704 		printf("NMI exiting and Virtual NMIs are not supported !\n");
4705 		return;
4706 	}
4707 
4708 	/* Save the controls so that we can restore them after our tests */
4709 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4710 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4711 
4712 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4713 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4714 
4715 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4716 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4717 	test_vmx_valid_controls();
4718 	report_prefix_pop();
4719 
4720 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4721 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4722 	test_vmx_invalid_controls();
4723 	report_prefix_pop();
4724 
4725 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4726 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4727 	test_vmx_valid_controls();
4728 	report_prefix_pop();
4729 
4730 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4731 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4732 	test_vmx_valid_controls();
4733 	report_prefix_pop();
4734 
4735 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4736 		report_info("NMI-window exiting is not supported, skipping...");
4737 		goto done;
4738 	}
4739 
4740 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4741 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4742 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4743 	test_vmx_invalid_controls();
4744 	report_prefix_pop();
4745 
4746 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4747 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4748 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4749 	test_vmx_valid_controls();
4750 	report_prefix_pop();
4751 
4752 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4753 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4754 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4755 	test_vmx_valid_controls();
4756 	report_prefix_pop();
4757 
4758 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4759 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4760 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4761 	test_vmx_valid_controls();
4762 	report_prefix_pop();
4763 
4764 	/* Restore the controls to their original values */
4765 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4766 done:
4767 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4768 }
4769 
4770 static void test_eptp_ad_bit(u64 eptp, bool ctrl)
4771 {
4772 	vmcs_write(EPTP, eptp);
4773 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4774 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4775 	if (ctrl)
4776 		test_vmx_valid_controls();
4777 	else
4778 		test_vmx_invalid_controls();
4779 	report_prefix_pop();
4780 
4781 }
4782 
4783 /*
4784  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4785  *    control field must satisfy the following checks:
4786  *
4787  *     - The EPT memory type (bits 2:0) must be a value supported by the
4788  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4789  *     - Bits 5:3 (1 less than the EPT page-walk length) must indicate a
4790  *	 supported EPT page-walk length.
4791  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4792  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4793  *	 indicating that the processor does not support accessed and dirty
4794  *	 dirty flags for EPT.
4795  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4796  *	 physical-address width) must all be 0.
4797  *
4798  * 2. If the "unrestricted guest" VM-execution control is 1, the
4799  *    "enable EPT" VM-execution control must also be 1.
4800  */
4801 static void test_ept_eptp(void)
4802 {
4803 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4804 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4805 	u64 eptp_saved = vmcs_read(EPTP);
4806 	u32 primary = primary_saved;
4807 	u32 secondary = secondary_saved;
4808 	u64 msr, eptp = eptp_saved;
4809 	bool un_cache = false;
4810 	bool wr_bk = false;
4811 	bool ctrl;
4812 	u32 i, maxphysaddr;
4813 	u64 j, resv_bits_mask = 0;
4814 
4815 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4816 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4817 		printf("\"CPU secondary\" and/or \"enable EPT\" execution controls are not supported !\n");
4818 		return;
4819 	}
4820 
4821 	/*
4822 	 * Memory type (bits 2:0)
4823 	 */
4824 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
4825 	if (msr & EPT_CAP_UC)
4826 		un_cache = true;
4827 	if (msr & EPT_CAP_WB)
4828 		wr_bk = true;
4829 
4830 	/* Support for 4-level EPT is mandatory. */
4831 	report(msr & EPT_CAP_PWL4, "4-level EPT support check");
4832 
4833 	primary |= CPU_SECONDARY;
4834 	vmcs_write(CPU_EXEC_CTRL0, primary);
4835 	secondary |= CPU_EPT;
4836 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4837 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4838 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4839 	vmcs_write(EPTP, eptp);
4840 
4841 	for (i = 0; i < 8; i++) {
4842 		if (i == 0) {
4843 			if (un_cache) {
4844 				report_info("EPT paging structure memory-type is Un-cacheable\n");
4845 				ctrl = true;
4846 			} else {
4847 				ctrl = false;
4848 			}
4849 		} else if (i == 6) {
4850 			if (wr_bk) {
4851 				report_info("EPT paging structure memory-type is Write-back\n");
4852 				ctrl = true;
4853 			} else {
4854 				ctrl = false;
4855 			}
4856 		} else {
4857 			ctrl = false;
4858 		}
4859 
4860 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4861 		vmcs_write(EPTP, eptp);
4862 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4863 		    eptp & EPT_MEM_TYPE_MASK);
4864 		if (ctrl)
4865 			test_vmx_valid_controls();
4866 		else
4867 			test_vmx_invalid_controls();
4868 		report_prefix_pop();
4869 	}
4870 
4871 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4872 
4873 	/*
4874 	 * Page walk length (bits 5:3).  Note, the value in VMCS.EPTP "is 1
4875 	 * less than the EPT page-walk length".
4876 	 */
4877 	for (i = 0; i < 8; i++) {
4878 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4879 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4880 		if (i == 3 || (i == 4 && (msr & EPT_CAP_PWL5)))
4881 			ctrl = true;
4882 		else
4883 			ctrl = false;
4884 
4885 		vmcs_write(EPTP, eptp);
4886 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4887 		    eptp & EPTP_PG_WALK_LEN_MASK);
4888 		if (ctrl)
4889 			test_vmx_valid_controls();
4890 		else
4891 			test_vmx_invalid_controls();
4892 		report_prefix_pop();
4893 	}
4894 
4895 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4896 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4897 
4898 	/*
4899 	 * Accessed and dirty flag (bit 6)
4900 	 */
4901 	if (msr & EPT_CAP_AD_FLAG) {
4902 		report_info("Processor supports accessed and dirty flag");
4903 		eptp &= ~EPTP_AD_FLAG;
4904 		test_eptp_ad_bit(eptp, true);
4905 
4906 		eptp |= EPTP_AD_FLAG;
4907 		test_eptp_ad_bit(eptp, true);
4908 	} else {
4909 		report_info("Processor does not supports accessed and dirty flag");
4910 		eptp &= ~EPTP_AD_FLAG;
4911 		test_eptp_ad_bit(eptp, true);
4912 
4913 		eptp |= EPTP_AD_FLAG;
4914 		test_eptp_ad_bit(eptp, false);
4915 	}
4916 
4917 	/*
4918 	 * Reserved bits [11:7] and [63:N]
4919 	 */
4920 	for (i = 0; i < 32; i++) {
4921 		eptp = (eptp &
4922 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4923 		    (i << EPTP_RESERV_BITS_SHIFT);
4924 		vmcs_write(EPTP, eptp);
4925 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4926 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4927 		    EPTP_RESERV_BITS_MASK);
4928 		if (i == 0)
4929 			test_vmx_valid_controls();
4930 		else
4931 			test_vmx_invalid_controls();
4932 		report_prefix_pop();
4933 	}
4934 
4935 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4936 
4937 	maxphysaddr = cpuid_maxphyaddr();
4938 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4939 		resv_bits_mask |= 1ul << i;
4940 	}
4941 
4942 	for (j = maxphysaddr - 1; j <= 63; j++) {
4943 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4944 		    (j < maxphysaddr ? 0 : 1ul << j);
4945 		vmcs_write(EPTP, eptp);
4946 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4947 		    (eptp >> maxphysaddr) & resv_bits_mask);
4948 		if (j < maxphysaddr)
4949 			test_vmx_valid_controls();
4950 		else
4951 			test_vmx_invalid_controls();
4952 		report_prefix_pop();
4953 	}
4954 
4955 	secondary &= ~(CPU_EPT | CPU_URG);
4956 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4957 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4958 	test_vmx_valid_controls();
4959 	report_prefix_pop();
4960 
4961 	if (!(ctrl_cpu_rev[1].clr & CPU_URG))
4962 		goto skip_unrestricted_guest;
4963 
4964 	secondary |= CPU_URG;
4965 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4966 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4967 	test_vmx_invalid_controls();
4968 	report_prefix_pop();
4969 
4970 	secondary |= CPU_EPT;
4971 	setup_dummy_ept();
4972 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4973 	test_vmx_valid_controls();
4974 	report_prefix_pop();
4975 
4976 skip_unrestricted_guest:
4977 	secondary &= ~CPU_URG;
4978 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4979 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4980 	test_vmx_valid_controls();
4981 	report_prefix_pop();
4982 
4983 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4984 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4985 	vmcs_write(EPTP, eptp_saved);
4986 }
4987 
4988 /*
4989  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4990  * VM-execution control must also be 1. In addition, the PML address
4991  * must satisfy the following checks:
4992  *
4993  *    * Bits 11:0 of the address must be 0.
4994  *    * The address should not set any bits beyond the processor's
4995  *	physical-address width.
4996  *
4997  *  [Intel SDM]
4998  */
4999 static void test_pml(void)
5000 {
5001 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
5002 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
5003 	u32 primary = primary_saved;
5004 	u32 secondary = secondary_saved;
5005 
5006 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
5007 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
5008 		printf("\"Secondary execution\" control or \"enable EPT\" control or \"enable PML\" control is not supported !\n");
5009 		return;
5010 	}
5011 
5012 	primary |= CPU_SECONDARY;
5013 	vmcs_write(CPU_EXEC_CTRL0, primary);
5014 	secondary &= ~(CPU_PML | CPU_EPT);
5015 	vmcs_write(CPU_EXEC_CTRL1, secondary);
5016 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
5017 	test_vmx_valid_controls();
5018 	report_prefix_pop();
5019 
5020 	secondary |= CPU_PML;
5021 	vmcs_write(CPU_EXEC_CTRL1, secondary);
5022 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
5023 	test_vmx_invalid_controls();
5024 	report_prefix_pop();
5025 
5026 	secondary |= CPU_EPT;
5027 	setup_dummy_ept();
5028 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
5029 	test_vmx_valid_controls();
5030 	report_prefix_pop();
5031 
5032 	secondary &= ~CPU_PML;
5033 	vmcs_write(CPU_EXEC_CTRL1, secondary);
5034 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
5035 	test_vmx_valid_controls();
5036 	report_prefix_pop();
5037 
5038 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
5039 				 PAGE_SIZE, false, false);
5040 
5041 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
5042 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
5043 }
5044 
5045  /*
5046  * If the "activate VMX-preemption timer" VM-execution control is 0, the
5047  * the "save VMX-preemption timer value" VM-exit control must also be 0.
5048  *
5049  *  [Intel SDM]
5050  */
5051 static void test_vmx_preemption_timer(void)
5052 {
5053 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
5054 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
5055 	u32 pin = saved_pin;
5056 	u32 exit = saved_exit;
5057 
5058 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
5059 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
5060 		printf("\"Save-VMX-preemption-timer\" control and/or \"Enable-VMX-preemption-timer\" control is not supported\n");
5061 		return;
5062 	}
5063 
5064 	pin |= PIN_PREEMPT;
5065 	vmcs_write(PIN_CONTROLS, pin);
5066 	exit &= ~EXI_SAVE_PREEMPT;
5067 	vmcs_write(EXI_CONTROLS, exit);
5068 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
5069 	test_vmx_valid_controls();
5070 	report_prefix_pop();
5071 
5072 	exit |= EXI_SAVE_PREEMPT;
5073 	vmcs_write(EXI_CONTROLS, exit);
5074 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
5075 	test_vmx_valid_controls();
5076 	report_prefix_pop();
5077 
5078 	pin &= ~PIN_PREEMPT;
5079 	vmcs_write(PIN_CONTROLS, pin);
5080 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
5081 	test_vmx_invalid_controls();
5082 	report_prefix_pop();
5083 
5084 	exit &= ~EXI_SAVE_PREEMPT;
5085 	vmcs_write(EXI_CONTROLS, exit);
5086 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
5087 	test_vmx_valid_controls();
5088 	report_prefix_pop();
5089 
5090 	vmcs_write(PIN_CONTROLS, saved_pin);
5091 	vmcs_write(EXI_CONTROLS, saved_exit);
5092 }
5093 
5094 extern unsigned char test_mtf1;
5095 extern unsigned char test_mtf2;
5096 extern unsigned char test_mtf3;
5097 extern unsigned char test_mtf4;
5098 
5099 static void test_mtf_guest(void)
5100 {
5101 	asm ("vmcall;\n\t"
5102 	     "out %al, $0x80;\n\t"
5103 	     "test_mtf1:\n\t"
5104 	     "vmcall;\n\t"
5105 	     "out %al, $0x80;\n\t"
5106 	     "test_mtf2:\n\t"
5107 	     /*
5108 	      * Prepare for the 'MOV CR3' test. Attempt to induce a
5109 	      * general-protection fault by moving a non-canonical address into
5110 	      * CR3. The 'MOV CR3' instruction does not take an imm64 operand,
5111 	      * so we must MOV the desired value into a register first.
5112 	      *
5113 	      * MOV RAX is done before the VMCALL such that MTF is only enabled
5114 	      * for the instruction under test.
5115 	      */
5116 	     "mov $0x8000000000000000, %rax;\n\t"
5117 	     "vmcall;\n\t"
5118 	     "mov %rax, %cr3;\n\t"
5119 	     "test_mtf3:\n\t"
5120 	     "vmcall;\n\t"
5121 	     /*
5122 	      * ICEBP/INT1 instruction. Though the instruction is now
5123 	      * documented, don't rely on assemblers enumerating the
5124 	      * instruction. Resort to hand assembly.
5125 	      */
5126 	     ".byte 0xf1;\n\t"
5127 	     "vmcall;\n\t"
5128 	     "test_mtf4:\n\t"
5129 	     "mov $0, %eax;\n\t");
5130 }
5131 
5132 static void test_mtf_gp_handler(struct ex_regs *regs)
5133 {
5134 	regs->rip = (unsigned long) &test_mtf3;
5135 }
5136 
5137 static void test_mtf_db_handler(struct ex_regs *regs)
5138 {
5139 }
5140 
5141 static void enable_mtf(void)
5142 {
5143 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5144 
5145 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 | CPU_MTF);
5146 }
5147 
5148 static void disable_mtf(void)
5149 {
5150 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5151 
5152 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 & ~CPU_MTF);
5153 }
5154 
5155 static void enable_tf(void)
5156 {
5157 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5158 
5159 	vmcs_write(GUEST_RFLAGS, rflags | X86_EFLAGS_TF);
5160 }
5161 
5162 static void disable_tf(void)
5163 {
5164 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5165 
5166 	vmcs_write(GUEST_RFLAGS, rflags & ~X86_EFLAGS_TF);
5167 }
5168 
5169 static void report_mtf(const char *insn_name, unsigned long exp_rip)
5170 {
5171 	unsigned long rip = vmcs_read(GUEST_RIP);
5172 
5173 	assert_exit_reason(VMX_MTF);
5174 	report(rip == exp_rip, "MTF VM-exit after %s. RIP: 0x%lx (expected 0x%lx)",
5175 	       insn_name, rip, exp_rip);
5176 }
5177 
5178 static void vmx_mtf_test(void)
5179 {
5180 	unsigned long pending_dbg;
5181 	handler old_gp, old_db;
5182 
5183 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5184 		printf("CPU does not support the 'monitor trap flag' processor-based VM-execution control.\n");
5185 		return;
5186 	}
5187 
5188 	test_set_guest(test_mtf_guest);
5189 
5190 	/* Expect an MTF VM-exit after OUT instruction */
5191 	enter_guest();
5192 	skip_exit_vmcall();
5193 
5194 	enable_mtf();
5195 	enter_guest();
5196 	report_mtf("OUT", (unsigned long) &test_mtf1);
5197 	disable_mtf();
5198 
5199 	/*
5200 	 * Concurrent #DB trap and MTF on instruction boundary. Expect MTF
5201 	 * VM-exit with populated 'pending debug exceptions' VMCS field.
5202 	 */
5203 	enter_guest();
5204 	skip_exit_vmcall();
5205 
5206 	enable_mtf();
5207 	enable_tf();
5208 
5209 	enter_guest();
5210 	report_mtf("OUT", (unsigned long) &test_mtf2);
5211 	pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
5212 	report(pending_dbg & DR_STEP,
5213 	       "'pending debug exceptions' field after MTF VM-exit: 0x%lx (expected 0x%lx)",
5214 	       pending_dbg, (unsigned long) DR_STEP);
5215 
5216 	disable_mtf();
5217 	disable_tf();
5218 	vmcs_write(GUEST_PENDING_DEBUG, 0);
5219 
5220 	/*
5221 	 * #GP exception takes priority over MTF. Expect MTF VM-exit with RIP
5222 	 * advanced to first instruction of #GP handler.
5223 	 */
5224 	enter_guest();
5225 	skip_exit_vmcall();
5226 
5227 	old_gp = handle_exception(GP_VECTOR, test_mtf_gp_handler);
5228 
5229 	enable_mtf();
5230 	enter_guest();
5231 	report_mtf("MOV CR3", (unsigned long) get_idt_addr(&boot_idt[GP_VECTOR]));
5232 	disable_mtf();
5233 
5234 	/*
5235 	 * Concurrent MTF and privileged software exception (i.e. ICEBP/INT1).
5236 	 * MTF should follow the delivery of #DB trap, though the SDM doesn't
5237 	 * provide clear indication of the relative priority.
5238 	 */
5239 	enter_guest();
5240 	skip_exit_vmcall();
5241 
5242 	handle_exception(GP_VECTOR, old_gp);
5243 	old_db = handle_exception(DB_VECTOR, test_mtf_db_handler);
5244 
5245 	enable_mtf();
5246 	enter_guest();
5247 	report_mtf("INT1", (unsigned long) get_idt_addr(&boot_idt[DB_VECTOR]));
5248 	disable_mtf();
5249 
5250 	enter_guest();
5251 	skip_exit_vmcall();
5252 	handle_exception(DB_VECTOR, old_db);
5253 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK | INTR_TYPE_OTHER_EVENT);
5254 	enter_guest();
5255 	report_mtf("injected MTF", (unsigned long) &test_mtf4);
5256 	enter_guest();
5257 }
5258 
5259 /*
5260  * Tests for VM-execution control fields
5261  */
5262 static void test_vm_execution_ctls(void)
5263 {
5264 	test_pin_based_ctls();
5265 	test_primary_processor_based_ctls();
5266 	test_secondary_processor_based_ctls();
5267 	test_cr3_targets();
5268 	test_io_bitmaps();
5269 	test_msr_bitmap();
5270 	test_apic_ctls();
5271 	test_tpr_threshold();
5272 	test_nmi_ctrls();
5273 	test_pml();
5274 	test_vpid();
5275 	test_ept_eptp();
5276 	test_vmx_preemption_timer();
5277 }
5278 
5279  /*
5280   * The following checks are performed for the VM-entry MSR-load address if
5281   * the VM-entry MSR-load count field is non-zero:
5282   *
5283   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
5284   *      The address should not set any bits beyond the processor’s
5285   *      physical-address width.
5286   *
5287   *    - The address of the last byte in the VM-entry MSR-load area
5288   *      should not set any bits beyond the processor’s physical-address
5289   *      width. The address of this last byte is VM-entry MSR-load address
5290   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5291   *      uses more bits than the processor’s physical-address width.)
5292   *
5293   *
5294   *  [Intel SDM]
5295   */
5296 static void test_entry_msr_load(void)
5297 {
5298 	entry_msr_load = alloc_page();
5299 	u64 tmp;
5300 	u32 entry_msr_ld_cnt = 1;
5301 	int i;
5302 	u32 addr_len = 64;
5303 
5304 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
5305 
5306 	/* Check first 4 bits of VM-entry MSR-load address */
5307 	for (i = 0; i < 4; i++) {
5308 		tmp = (u64)entry_msr_load | 1ull << i;
5309 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
5310 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
5311 				    tmp & 0xf);
5312 		test_vmx_invalid_controls();
5313 		report_prefix_pop();
5314 	}
5315 
5316 	if (basic.val & (1ul << 48))
5317 		addr_len = 32;
5318 
5319 	test_vmcs_addr_values("VM-entry-MSR-load address",
5320 				ENTER_MSR_LD_ADDR, 16, false, false,
5321 				4, addr_len - 1);
5322 
5323 	/*
5324 	 * Check last byte of VM-entry MSR-load address
5325 	 */
5326 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
5327 
5328 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5329 							i < 64; i++) {
5330 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
5331 			1ul << i;
5332 		vmcs_write(ENTER_MSR_LD_ADDR,
5333 			   tmp - (entry_msr_ld_cnt * 16 - 1));
5334 		test_vmx_invalid_controls();
5335 	}
5336 
5337 	vmcs_write(ENT_MSR_LD_CNT, 2);
5338 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5339 	test_vmx_invalid_controls();
5340 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5341 	test_vmx_valid_controls();
5342 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5343 	test_vmx_valid_controls();
5344 }
5345 
5346 static struct vmx_state_area_test_data {
5347 	u32 msr;
5348 	u64 exp;
5349 	bool enabled;
5350 } vmx_state_area_test_data;
5351 
5352 static void guest_state_test_main(void)
5353 {
5354 	u64 obs;
5355 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
5356 
5357 	while (1) {
5358 		if (vmx_get_test_stage() == 2)
5359 			break;
5360 
5361 		if (data->enabled) {
5362 			obs = rdmsr(data->msr);
5363 			report(data->exp == obs,
5364 			       "Guest state is 0x%lx (expected 0x%lx)",
5365 			       obs, data->exp);
5366 		}
5367 
5368 		vmcall();
5369 	}
5370 
5371 	asm volatile("fnop");
5372 }
5373 
5374 static void test_guest_state(const char *test, bool xfail, u64 field,
5375 			     const char * field_name)
5376 {
5377 	struct vmentry_result result;
5378 	u8 abort_flags;
5379 
5380 	abort_flags = ABORT_ON_EARLY_VMENTRY_FAIL;
5381 	if (!xfail)
5382 		abort_flags = ABORT_ON_INVALID_GUEST_STATE;
5383 
5384 	__enter_guest(abort_flags, &result);
5385 
5386 	report(result.exit_reason.failed_vmentry == xfail &&
5387 	       ((xfail && result.exit_reason.basic == VMX_FAIL_STATE) ||
5388 	        (!xfail && result.exit_reason.basic == VMX_VMCALL)) &&
5389 		(!xfail || vmcs_read(EXI_QUALIFICATION) == ENTRY_FAIL_DEFAULT),
5390 	        "%s, %s %lx", test, field_name, field);
5391 
5392 	if (!result.exit_reason.failed_vmentry)
5393 		skip_exit_insn();
5394 }
5395 
5396 /*
5397  * Tests for VM-entry control fields
5398  */
5399 static void test_vm_entry_ctls(void)
5400 {
5401 	test_invalid_event_injection();
5402 	test_entry_msr_load();
5403 }
5404 
5405 /*
5406  * The following checks are performed for the VM-exit MSR-store address if
5407  * the VM-exit MSR-store count field is non-zero:
5408  *
5409  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
5410  *      The address should not set any bits beyond the processor’s
5411  *      physical-address width.
5412  *
5413  *    - The address of the last byte in the VM-exit MSR-store area
5414  *      should not set any bits beyond the processor’s physical-address
5415  *      width. The address of this last byte is VM-exit MSR-store address
5416  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5417  *      uses more bits than the processor’s physical-address width.)
5418  *
5419  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
5420  * in the range 63:32.
5421  *
5422  *  [Intel SDM]
5423  */
5424 static void test_exit_msr_store(void)
5425 {
5426 	exit_msr_store = alloc_page();
5427 	u64 tmp;
5428 	u32 exit_msr_st_cnt = 1;
5429 	int i;
5430 	u32 addr_len = 64;
5431 
5432 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
5433 
5434 	/* Check first 4 bits of VM-exit MSR-store address */
5435 	for (i = 0; i < 4; i++) {
5436 		tmp = (u64)exit_msr_store | 1ull << i;
5437 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
5438 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
5439 				    tmp & 0xf);
5440 		test_vmx_invalid_controls();
5441 		report_prefix_pop();
5442 	}
5443 
5444 	if (basic.val & (1ul << 48))
5445 		addr_len = 32;
5446 
5447 	test_vmcs_addr_values("VM-exit-MSR-store address",
5448 				EXIT_MSR_ST_ADDR, 16, false, false,
5449 				4, addr_len - 1);
5450 
5451 	/*
5452 	 * Check last byte of VM-exit MSR-store address
5453 	 */
5454 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
5455 
5456 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5457 							i < 64; i++) {
5458 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
5459 			1ul << i;
5460 		vmcs_write(EXIT_MSR_ST_ADDR,
5461 			   tmp - (exit_msr_st_cnt * 16 - 1));
5462 		test_vmx_invalid_controls();
5463 	}
5464 
5465 	vmcs_write(EXI_MSR_ST_CNT, 2);
5466 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5467 	test_vmx_invalid_controls();
5468 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5469 	test_vmx_valid_controls();
5470 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5471 	test_vmx_valid_controls();
5472 }
5473 
5474 /*
5475  * Tests for VM-exit controls
5476  */
5477 static void test_vm_exit_ctls(void)
5478 {
5479 	test_exit_msr_store();
5480 }
5481 
5482 /*
5483  * Check that the virtual CPU checks all of the VMX controls as
5484  * documented in the Intel SDM.
5485  */
5486 static void vmx_controls_test(void)
5487 {
5488 	/*
5489 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
5490 	 * fail due to invalid guest state, should we make it that
5491 	 * far.
5492 	 */
5493 	vmcs_write(GUEST_RFLAGS, 0);
5494 
5495 	test_vm_execution_ctls();
5496 	test_vm_exit_ctls();
5497 	test_vm_entry_ctls();
5498 }
5499 
5500 struct apic_reg_virt_config {
5501 	bool apic_register_virtualization;
5502 	bool use_tpr_shadow;
5503 	bool virtualize_apic_accesses;
5504 	bool virtualize_x2apic_mode;
5505 	bool activate_secondary_controls;
5506 };
5507 
5508 struct apic_reg_test {
5509 	const char *name;
5510 	struct apic_reg_virt_config apic_reg_virt_config;
5511 };
5512 
5513 struct apic_reg_virt_expectation {
5514 	enum Reason rd_exit_reason;
5515 	enum Reason wr_exit_reason;
5516 	u32 val;
5517 	u32 (*virt_fn)(u32);
5518 
5519 	/*
5520 	 * If false, accessing the APIC access address from L2 is treated as a
5521 	 * normal memory operation, rather than triggering virtualization.
5522 	 */
5523 	bool virtualize_apic_accesses;
5524 };
5525 
5526 static u32 apic_virt_identity(u32 val)
5527 {
5528 	return val;
5529 }
5530 
5531 static u32 apic_virt_nibble1(u32 val)
5532 {
5533 	return val & 0xf0;
5534 }
5535 
5536 static u32 apic_virt_byte3(u32 val)
5537 {
5538 	return val & (0xff << 24);
5539 }
5540 
5541 static bool apic_reg_virt_exit_expectation(
5542 	u32 reg, struct apic_reg_virt_config *config,
5543 	struct apic_reg_virt_expectation *expectation)
5544 {
5545 	/* Good configs, where some L2 APIC accesses are virtualized. */
5546 	bool virtualize_apic_accesses_only =
5547 		config->virtualize_apic_accesses &&
5548 		!config->use_tpr_shadow &&
5549 		!config->apic_register_virtualization &&
5550 		!config->virtualize_x2apic_mode &&
5551 		config->activate_secondary_controls;
5552 	bool virtualize_apic_accesses_and_use_tpr_shadow =
5553 		config->virtualize_apic_accesses &&
5554 		config->use_tpr_shadow &&
5555 		!config->apic_register_virtualization &&
5556 		!config->virtualize_x2apic_mode &&
5557 		config->activate_secondary_controls;
5558 	bool apic_register_virtualization =
5559 		config->virtualize_apic_accesses &&
5560 		config->use_tpr_shadow &&
5561 		config->apic_register_virtualization &&
5562 		!config->virtualize_x2apic_mode &&
5563 		config->activate_secondary_controls;
5564 
5565 	expectation->val = MAGIC_VAL_1;
5566 	expectation->virt_fn = apic_virt_identity;
5567 	expectation->virtualize_apic_accesses =
5568 		config->virtualize_apic_accesses &&
5569 		config->activate_secondary_controls;
5570 	if (virtualize_apic_accesses_only) {
5571 		expectation->rd_exit_reason = VMX_APIC_ACCESS;
5572 		expectation->wr_exit_reason = VMX_APIC_ACCESS;
5573 	} else if (virtualize_apic_accesses_and_use_tpr_shadow) {
5574 		switch (reg) {
5575 		case APIC_TASKPRI:
5576 			expectation->rd_exit_reason = VMX_VMCALL;
5577 			expectation->wr_exit_reason = VMX_VMCALL;
5578 			expectation->virt_fn = apic_virt_nibble1;
5579 			break;
5580 		default:
5581 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5582 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5583 		}
5584 	} else if (apic_register_virtualization) {
5585 		expectation->rd_exit_reason = VMX_VMCALL;
5586 
5587 		switch (reg) {
5588 		case APIC_ID:
5589 		case APIC_EOI:
5590 		case APIC_LDR:
5591 		case APIC_DFR:
5592 		case APIC_SPIV:
5593 		case APIC_ESR:
5594 		case APIC_ICR:
5595 		case APIC_LVTT:
5596 		case APIC_LVTTHMR:
5597 		case APIC_LVTPC:
5598 		case APIC_LVT0:
5599 		case APIC_LVT1:
5600 		case APIC_LVTERR:
5601 		case APIC_TMICT:
5602 		case APIC_TDCR:
5603 			expectation->wr_exit_reason = VMX_APIC_WRITE;
5604 			break;
5605 		case APIC_LVR:
5606 		case APIC_ISR ... APIC_ISR + 0x70:
5607 		case APIC_TMR ... APIC_TMR + 0x70:
5608 		case APIC_IRR ... APIC_IRR + 0x70:
5609 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5610 			break;
5611 		case APIC_TASKPRI:
5612 			expectation->wr_exit_reason = VMX_VMCALL;
5613 			expectation->virt_fn = apic_virt_nibble1;
5614 			break;
5615 		case APIC_ICR2:
5616 			expectation->wr_exit_reason = VMX_VMCALL;
5617 			expectation->virt_fn = apic_virt_byte3;
5618 			break;
5619 		default:
5620 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5621 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5622 		}
5623 	} else if (!expectation->virtualize_apic_accesses) {
5624 		/*
5625 		 * No APIC registers are directly virtualized. This includes
5626 		 * VTPR, which can be virtualized through MOV to/from CR8 via
5627 		 * the use TPR shadow control, but not through directly
5628 		 * accessing VTPR.
5629 		 */
5630 		expectation->rd_exit_reason = VMX_VMCALL;
5631 		expectation->wr_exit_reason = VMX_VMCALL;
5632 	} else {
5633 		printf("Cannot parse APIC register virtualization config:\n"
5634 		       "\tvirtualize_apic_accesses: %d\n"
5635 		       "\tuse_tpr_shadow: %d\n"
5636 		       "\tapic_register_virtualization: %d\n"
5637 		       "\tvirtualize_x2apic_mode: %d\n"
5638 		       "\tactivate_secondary_controls: %d\n",
5639 		       config->virtualize_apic_accesses,
5640 		       config->use_tpr_shadow,
5641 		       config->apic_register_virtualization,
5642 		       config->virtualize_x2apic_mode,
5643 		       config->activate_secondary_controls);
5644 
5645 		return false;
5646 	}
5647 
5648 	return true;
5649 }
5650 
5651 struct apic_reg_test apic_reg_tests[] = {
5652 	/* Good configs, where some L2 APIC accesses are virtualized. */
5653 	{
5654 		.name = "Virtualize APIC accesses",
5655 		.apic_reg_virt_config = {
5656 			.virtualize_apic_accesses = true,
5657 			.use_tpr_shadow = false,
5658 			.apic_register_virtualization = false,
5659 			.virtualize_x2apic_mode = false,
5660 			.activate_secondary_controls = true,
5661 		},
5662 	},
5663 	{
5664 		.name = "Virtualize APIC accesses + Use TPR shadow",
5665 		.apic_reg_virt_config = {
5666 			.virtualize_apic_accesses = true,
5667 			.use_tpr_shadow = true,
5668 			.apic_register_virtualization = false,
5669 			.virtualize_x2apic_mode = false,
5670 			.activate_secondary_controls = true,
5671 		},
5672 	},
5673 	{
5674 		.name = "APIC-register virtualization",
5675 		.apic_reg_virt_config = {
5676 			.virtualize_apic_accesses = true,
5677 			.use_tpr_shadow = true,
5678 			.apic_register_virtualization = true,
5679 			.virtualize_x2apic_mode = false,
5680 			.activate_secondary_controls = true,
5681 		},
5682 	},
5683 
5684 	/*
5685 	 * Test that the secondary processor-based VM-execution controls are
5686 	 * correctly ignored when "activate secondary controls" is disabled.
5687 	 */
5688 	{
5689 		.name = "Activate secondary controls off",
5690 		.apic_reg_virt_config = {
5691 			.virtualize_apic_accesses = true,
5692 			.use_tpr_shadow = false,
5693 			.apic_register_virtualization = true,
5694 			.virtualize_x2apic_mode = true,
5695 			.activate_secondary_controls = false,
5696 		},
5697 	},
5698 	{
5699 		.name = "Activate secondary controls off + Use TPR shadow",
5700 		.apic_reg_virt_config = {
5701 			.virtualize_apic_accesses = true,
5702 			.use_tpr_shadow = true,
5703 			.apic_register_virtualization = true,
5704 			.virtualize_x2apic_mode = true,
5705 			.activate_secondary_controls = false,
5706 		},
5707 	},
5708 
5709 	/*
5710 	 * Test that the APIC access address is treated like an arbitrary memory
5711 	 * address when "virtualize APIC accesses" is disabled.
5712 	 */
5713 	{
5714 		.name = "Virtualize APIC accesses off + Use TPR shadow",
5715 		.apic_reg_virt_config = {
5716 			.virtualize_apic_accesses = false,
5717 			.use_tpr_shadow = true,
5718 			.apic_register_virtualization = true,
5719 			.virtualize_x2apic_mode = true,
5720 			.activate_secondary_controls = true,
5721 		},
5722 	},
5723 
5724 	/*
5725 	 * Test that VM entry fails due to invalid controls when
5726 	 * "APIC-register virtualization" is enabled while "use TPR shadow" is
5727 	 * disabled.
5728 	 */
5729 	{
5730 		.name = "APIC-register virtualization + Use TPR shadow off",
5731 		.apic_reg_virt_config = {
5732 			.virtualize_apic_accesses = true,
5733 			.use_tpr_shadow = false,
5734 			.apic_register_virtualization = true,
5735 			.virtualize_x2apic_mode = false,
5736 			.activate_secondary_controls = true,
5737 		},
5738 	},
5739 
5740 	/*
5741 	 * Test that VM entry fails due to invalid controls when
5742 	 * "Virtualize x2APIC mode" is enabled while "use TPR shadow" is
5743 	 * disabled.
5744 	 */
5745 	{
5746 		.name = "Virtualize x2APIC mode + Use TPR shadow off",
5747 		.apic_reg_virt_config = {
5748 			.virtualize_apic_accesses = false,
5749 			.use_tpr_shadow = false,
5750 			.apic_register_virtualization = false,
5751 			.virtualize_x2apic_mode = true,
5752 			.activate_secondary_controls = true,
5753 		},
5754 	},
5755 	{
5756 		.name = "Virtualize x2APIC mode + Use TPR shadow off v2",
5757 		.apic_reg_virt_config = {
5758 			.virtualize_apic_accesses = false,
5759 			.use_tpr_shadow = false,
5760 			.apic_register_virtualization = true,
5761 			.virtualize_x2apic_mode = true,
5762 			.activate_secondary_controls = true,
5763 		},
5764 	},
5765 
5766 	/*
5767 	 * Test that VM entry fails due to invalid controls when
5768 	 * "virtualize x2APIC mode" is enabled while "virtualize APIC accesses"
5769 	 * is enabled.
5770 	 */
5771 	{
5772 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses",
5773 		.apic_reg_virt_config = {
5774 			.virtualize_apic_accesses = true,
5775 			.use_tpr_shadow = true,
5776 			.apic_register_virtualization = false,
5777 			.virtualize_x2apic_mode = true,
5778 			.activate_secondary_controls = true,
5779 		},
5780 	},
5781 	{
5782 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses v2",
5783 		.apic_reg_virt_config = {
5784 			.virtualize_apic_accesses = true,
5785 			.use_tpr_shadow = true,
5786 			.apic_register_virtualization = true,
5787 			.virtualize_x2apic_mode = true,
5788 			.activate_secondary_controls = true,
5789 		},
5790 	},
5791 };
5792 
5793 enum Apic_op {
5794 	APIC_OP_XAPIC_RD,
5795 	APIC_OP_XAPIC_WR,
5796 	TERMINATE,
5797 };
5798 
5799 static u32 vmx_xapic_read(u32 *apic_access_address, u32 reg)
5800 {
5801 	return *(volatile u32 *)((uintptr_t)apic_access_address + reg);
5802 }
5803 
5804 static void vmx_xapic_write(u32 *apic_access_address, u32 reg, u32 val)
5805 {
5806 	*(volatile u32 *)((uintptr_t)apic_access_address + reg) = val;
5807 }
5808 
5809 struct apic_reg_virt_guest_args {
5810 	enum Apic_op op;
5811 	u32 *apic_access_address;
5812 	u32 reg;
5813 	u32 val;
5814 	bool check_rd;
5815 	u32 (*virt_fn)(u32);
5816 } apic_reg_virt_guest_args;
5817 
5818 static void apic_reg_virt_guest(void)
5819 {
5820 	volatile struct apic_reg_virt_guest_args *args =
5821 		&apic_reg_virt_guest_args;
5822 
5823 	for (;;) {
5824 		enum Apic_op op = args->op;
5825 		u32 *apic_access_address = args->apic_access_address;
5826 		u32 reg = args->reg;
5827 		u32 val = args->val;
5828 		bool check_rd = args->check_rd;
5829 		u32 (*virt_fn)(u32) = args->virt_fn;
5830 
5831 		if (op == TERMINATE)
5832 			break;
5833 
5834 		if (op == APIC_OP_XAPIC_RD) {
5835 			u32 ret = vmx_xapic_read(apic_access_address, reg);
5836 
5837 			if (check_rd) {
5838 				u32 want = virt_fn(val);
5839 				u32 got = virt_fn(ret);
5840 
5841 				report(got == want,
5842 				       "read 0x%x, expected 0x%x.", got, want);
5843 			}
5844 		} else if (op == APIC_OP_XAPIC_WR) {
5845 			vmx_xapic_write(apic_access_address, reg, val);
5846 		}
5847 
5848 		/*
5849 		 * The L1 should always execute a vmcall after it's done testing
5850 		 * an individual APIC operation. This helps to validate that the
5851 		 * L1 and L2 are in sync with each other, as expected.
5852 		 */
5853 		vmcall();
5854 	}
5855 }
5856 
5857 static void test_xapic_rd(
5858 	u32 reg, struct apic_reg_virt_expectation *expectation,
5859 	u32 *apic_access_address, u32 *virtual_apic_page)
5860 {
5861 	u32 val = expectation->val;
5862 	u32 exit_reason_want = expectation->rd_exit_reason;
5863 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5864 
5865 	report_prefix_pushf("xapic - reading 0x%03x", reg);
5866 
5867 	/* Configure guest to do an xapic read */
5868 	args->op = APIC_OP_XAPIC_RD;
5869 	args->apic_access_address = apic_access_address;
5870 	args->reg = reg;
5871 	args->val = val;
5872 	args->check_rd = exit_reason_want == VMX_VMCALL;
5873 	args->virt_fn = expectation->virt_fn;
5874 
5875 	/* Setup virtual APIC page */
5876 	if (!expectation->virtualize_apic_accesses) {
5877 		apic_access_address[apic_reg_index(reg)] = val;
5878 		virtual_apic_page[apic_reg_index(reg)] = 0;
5879 	} else if (exit_reason_want == VMX_VMCALL) {
5880 		apic_access_address[apic_reg_index(reg)] = 0;
5881 		virtual_apic_page[apic_reg_index(reg)] = val;
5882 	}
5883 
5884 	/* Enter guest */
5885 	enter_guest();
5886 
5887 	/*
5888 	 * Validate the behavior and
5889 	 * pass a magic value back to the guest.
5890 	 */
5891 	if (exit_reason_want == VMX_APIC_ACCESS) {
5892 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5893 
5894 		assert_exit_reason(exit_reason_want);
5895 		report(apic_page_offset == reg,
5896 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
5897 		       apic_page_offset, reg);
5898 		skip_exit_insn();
5899 
5900 		/* Reenter guest so it can consume/check rcx and exit again. */
5901 		enter_guest();
5902 	} else if (exit_reason_want != VMX_VMCALL) {
5903 		report(false, "Oops, bad exit expectation: %u.",
5904 		       exit_reason_want);
5905 	}
5906 
5907 	skip_exit_vmcall();
5908 	report_prefix_pop();
5909 }
5910 
5911 static void test_xapic_wr(
5912 	u32 reg, struct apic_reg_virt_expectation *expectation,
5913 	u32 *apic_access_address, u32 *virtual_apic_page)
5914 {
5915 	u32 val = expectation->val;
5916 	u32 exit_reason_want = expectation->wr_exit_reason;
5917 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5918 	bool virtualized =
5919 		expectation->virtualize_apic_accesses &&
5920 		(exit_reason_want == VMX_APIC_WRITE ||
5921 		 exit_reason_want == VMX_VMCALL);
5922 	bool checked = false;
5923 
5924 	report_prefix_pushf("xapic - writing 0x%x to 0x%03x", val, reg);
5925 
5926 	/* Configure guest to do an xapic read */
5927 	args->op = APIC_OP_XAPIC_WR;
5928 	args->apic_access_address = apic_access_address;
5929 	args->reg = reg;
5930 	args->val = val;
5931 
5932 	/* Setup virtual APIC page */
5933 	if (virtualized || !expectation->virtualize_apic_accesses) {
5934 		apic_access_address[apic_reg_index(reg)] = 0;
5935 		virtual_apic_page[apic_reg_index(reg)] = 0;
5936 	}
5937 
5938 	/* Enter guest */
5939 	enter_guest();
5940 
5941 	/*
5942 	 * Validate the behavior and
5943 	 * pass a magic value back to the guest.
5944 	 */
5945 	if (exit_reason_want == VMX_APIC_ACCESS) {
5946 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5947 
5948 		assert_exit_reason(exit_reason_want);
5949 		report(apic_page_offset == reg,
5950 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
5951 		       apic_page_offset, reg);
5952 		skip_exit_insn();
5953 
5954 		/* Reenter guest so it can consume/check rcx and exit again. */
5955 		enter_guest();
5956 	} else if (exit_reason_want == VMX_APIC_WRITE) {
5957 		assert_exit_reason(exit_reason_want);
5958 		report(virtual_apic_page[apic_reg_index(reg)] == val,
5959 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%x",
5960 		       apic_reg_index(reg),
5961 		       virtual_apic_page[apic_reg_index(reg)], val);
5962 		checked = true;
5963 
5964 		/* Reenter guest so it can consume/check rcx and exit again. */
5965 		enter_guest();
5966 	} else if (exit_reason_want != VMX_VMCALL) {
5967 		report(false, "Oops, bad exit expectation: %u.",
5968 		       exit_reason_want);
5969 	}
5970 
5971 	assert_exit_reason(VMX_VMCALL);
5972 	if (virtualized && !checked) {
5973 		u32 want = expectation->virt_fn(val);
5974 		u32 got = virtual_apic_page[apic_reg_index(reg)];
5975 		got = expectation->virt_fn(got);
5976 
5977 		report(got == want, "exitless write; val is 0x%x, want 0x%x",
5978 		       got, want);
5979 	} else if (!expectation->virtualize_apic_accesses && !checked) {
5980 		u32 got = apic_access_address[apic_reg_index(reg)];
5981 
5982 		report(got == val,
5983 		       "non-virtualized write; val is 0x%x, want 0x%x", got,
5984 		       val);
5985 	} else if (!expectation->virtualize_apic_accesses && checked) {
5986 		report(false,
5987 		       "Non-virtualized write was prematurely checked!");
5988 	}
5989 
5990 	skip_exit_vmcall();
5991 	report_prefix_pop();
5992 }
5993 
5994 enum Config_type {
5995 	CONFIG_TYPE_GOOD,
5996 	CONFIG_TYPE_UNSUPPORTED,
5997 	CONFIG_TYPE_VMENTRY_FAILS_EARLY,
5998 };
5999 
6000 static enum Config_type configure_apic_reg_virt_test(
6001 	struct apic_reg_virt_config *apic_reg_virt_config)
6002 {
6003 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6004 	u32 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6005 	/* Configs where L2 entry fails early, due to invalid controls. */
6006 	bool use_tpr_shadow_incorrectly_off =
6007 		!apic_reg_virt_config->use_tpr_shadow &&
6008 		(apic_reg_virt_config->apic_register_virtualization ||
6009 		 apic_reg_virt_config->virtualize_x2apic_mode) &&
6010 		apic_reg_virt_config->activate_secondary_controls;
6011 	bool virtualize_apic_accesses_incorrectly_on =
6012 		apic_reg_virt_config->virtualize_apic_accesses &&
6013 		apic_reg_virt_config->virtualize_x2apic_mode &&
6014 		apic_reg_virt_config->activate_secondary_controls;
6015 	bool vmentry_fails_early =
6016 		use_tpr_shadow_incorrectly_off ||
6017 		virtualize_apic_accesses_incorrectly_on;
6018 
6019 	if (apic_reg_virt_config->activate_secondary_controls) {
6020 		if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
6021 			printf("VM-execution control \"activate secondary controls\" NOT supported.\n");
6022 			return CONFIG_TYPE_UNSUPPORTED;
6023 		}
6024 		cpu_exec_ctrl0 |= CPU_SECONDARY;
6025 	} else {
6026 		cpu_exec_ctrl0 &= ~CPU_SECONDARY;
6027 	}
6028 
6029 	if (apic_reg_virt_config->virtualize_apic_accesses) {
6030 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES)) {
6031 			printf("VM-execution control \"virtualize APIC accesses\" NOT supported.\n");
6032 			return CONFIG_TYPE_UNSUPPORTED;
6033 		}
6034 		cpu_exec_ctrl1 |= CPU_VIRT_APIC_ACCESSES;
6035 	} else {
6036 		cpu_exec_ctrl1 &= ~CPU_VIRT_APIC_ACCESSES;
6037 	}
6038 
6039 	if (apic_reg_virt_config->use_tpr_shadow) {
6040 		if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6041 			printf("VM-execution control \"use TPR shadow\" NOT supported.\n");
6042 			return CONFIG_TYPE_UNSUPPORTED;
6043 		}
6044 		cpu_exec_ctrl0 |= CPU_TPR_SHADOW;
6045 	} else {
6046 		cpu_exec_ctrl0 &= ~CPU_TPR_SHADOW;
6047 	}
6048 
6049 	if (apic_reg_virt_config->apic_register_virtualization) {
6050 		if (!(ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT)) {
6051 			printf("VM-execution control \"APIC-register virtualization\" NOT supported.\n");
6052 			return CONFIG_TYPE_UNSUPPORTED;
6053 		}
6054 		cpu_exec_ctrl1 |= CPU_APIC_REG_VIRT;
6055 	} else {
6056 		cpu_exec_ctrl1 &= ~CPU_APIC_REG_VIRT;
6057 	}
6058 
6059 	if (apic_reg_virt_config->virtualize_x2apic_mode) {
6060 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_X2APIC)) {
6061 			printf("VM-execution control \"virtualize x2APIC mode\" NOT supported.\n");
6062 			return CONFIG_TYPE_UNSUPPORTED;
6063 		}
6064 		cpu_exec_ctrl1 |= CPU_VIRT_X2APIC;
6065 	} else {
6066 		cpu_exec_ctrl1 &= ~CPU_VIRT_X2APIC;
6067 	}
6068 
6069 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6070 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6071 
6072 	if (vmentry_fails_early)
6073 		return CONFIG_TYPE_VMENTRY_FAILS_EARLY;
6074 
6075 	return CONFIG_TYPE_GOOD;
6076 }
6077 
6078 static bool cpu_has_apicv(void)
6079 {
6080 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
6081 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
6082 		(ctrl_pin_rev.clr & PIN_POST_INTR));
6083 }
6084 
6085 /* Validates APIC register access across valid virtualization configurations. */
6086 static void apic_reg_virt_test(void)
6087 {
6088 	u32 *apic_access_address;
6089 	u32 *virtual_apic_page;
6090 	u64 control;
6091 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6092 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6093 	int i;
6094 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
6095 
6096 	if (!cpu_has_apicv()) {
6097 		report_skip(__func__);
6098 		return;
6099 	}
6100 
6101 	control = cpu_exec_ctrl1;
6102 	control &= ~CPU_VINTD;
6103 	vmcs_write(CPU_EXEC_CTRL1, control);
6104 
6105 	test_set_guest(apic_reg_virt_guest);
6106 
6107 	/*
6108 	 * From the SDM: The 1-setting of the "virtualize APIC accesses"
6109 	 * VM-execution is guaranteed to apply only if translations to the
6110 	 * APIC-access address use a 4-KByte page.
6111 	 */
6112 	apic_access_address = alloc_page();
6113 	force_4k_page(apic_access_address);
6114 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_address));
6115 
6116 	virtual_apic_page = alloc_page();
6117 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6118 
6119 	for (i = 0; i < ARRAY_SIZE(apic_reg_tests); i++) {
6120 		struct apic_reg_test *apic_reg_test = &apic_reg_tests[i];
6121 		struct apic_reg_virt_config *apic_reg_virt_config =
6122 				&apic_reg_test->apic_reg_virt_config;
6123 		enum Config_type config_type;
6124 		u32 reg;
6125 
6126 		printf("--- %s test ---\n", apic_reg_test->name);
6127 		config_type =
6128 			configure_apic_reg_virt_test(apic_reg_virt_config);
6129 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6130 			printf("Skip because of missing features.\n");
6131 			continue;
6132 		}
6133 
6134 		if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6135 			enter_guest_with_bad_controls();
6136 			continue;
6137 		}
6138 
6139 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6140 			struct apic_reg_virt_expectation expectation = {};
6141 			bool ok;
6142 
6143 			ok = apic_reg_virt_exit_expectation(
6144 				reg, apic_reg_virt_config, &expectation);
6145 			if (!ok) {
6146 				report(false, "Malformed test.");
6147 				break;
6148 			}
6149 
6150 			test_xapic_rd(reg, &expectation, apic_access_address,
6151 				      virtual_apic_page);
6152 			test_xapic_wr(reg, &expectation, apic_access_address,
6153 				      virtual_apic_page);
6154 		}
6155 	}
6156 
6157 	/* Terminate the guest */
6158 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6159 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6160 	args->op = TERMINATE;
6161 	enter_guest();
6162 	assert_exit_reason(VMX_VMCALL);
6163 }
6164 
6165 struct virt_x2apic_mode_config {
6166 	struct apic_reg_virt_config apic_reg_virt_config;
6167 	bool virtual_interrupt_delivery;
6168 	bool use_msr_bitmaps;
6169 	bool disable_x2apic_msr_intercepts;
6170 	bool disable_x2apic;
6171 };
6172 
6173 struct virt_x2apic_mode_test_case {
6174 	const char *name;
6175 	struct virt_x2apic_mode_config virt_x2apic_mode_config;
6176 };
6177 
6178 enum Virt_x2apic_mode_behavior_type {
6179 	X2APIC_ACCESS_VIRTUALIZED,
6180 	X2APIC_ACCESS_PASSED_THROUGH,
6181 	X2APIC_ACCESS_TRIGGERS_GP,
6182 };
6183 
6184 struct virt_x2apic_mode_expectation {
6185 	enum Reason rd_exit_reason;
6186 	enum Reason wr_exit_reason;
6187 
6188 	/*
6189 	 * RDMSR and WRMSR handle 64-bit values. However, except for ICR, all of
6190 	 * the x2APIC registers are 32 bits. Notice:
6191 	 *   1. vmx_x2apic_read() clears the upper 32 bits for 32-bit registers.
6192 	 *   2. vmx_x2apic_write() expects the val arg to be well-formed.
6193 	 */
6194 	u64 rd_val;
6195 	u64 wr_val;
6196 
6197 	/*
6198 	 * Compares input to virtualized output;
6199 	 * 1st arg is pointer to return expected virtualization output.
6200 	 */
6201 	u64 (*virt_fn)(u64);
6202 
6203 	enum Virt_x2apic_mode_behavior_type rd_behavior;
6204 	enum Virt_x2apic_mode_behavior_type wr_behavior;
6205 	bool wr_only;
6206 };
6207 
6208 static u64 virt_x2apic_mode_identity(u64 val)
6209 {
6210 	return val;
6211 }
6212 
6213 static u64 virt_x2apic_mode_nibble1(u64 val)
6214 {
6215 	return val & 0xf0;
6216 }
6217 
6218 static void virt_x2apic_mode_rd_expectation(
6219 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6220 	bool apic_register_virtualization, bool virtual_interrupt_delivery,
6221 	struct virt_x2apic_mode_expectation *expectation)
6222 {
6223 	bool readable =
6224 		!x2apic_reg_reserved(reg) &&
6225 		reg != APIC_EOI;
6226 
6227 	expectation->rd_exit_reason = VMX_VMCALL;
6228 	expectation->virt_fn = virt_x2apic_mode_identity;
6229 	if (virt_x2apic_mode_on && apic_register_virtualization) {
6230 		expectation->rd_val = MAGIC_VAL_1;
6231 		if (reg == APIC_PROCPRI && virtual_interrupt_delivery)
6232 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6233 		else if (reg == APIC_TASKPRI)
6234 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6235 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6236 	} else if (virt_x2apic_mode_on && !apic_register_virtualization &&
6237 		   reg == APIC_TASKPRI) {
6238 		expectation->rd_val = MAGIC_VAL_1;
6239 		expectation->virt_fn = virt_x2apic_mode_nibble1;
6240 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6241 	} else if (!disable_x2apic && readable) {
6242 		expectation->rd_val = apic_read(reg);
6243 		expectation->rd_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6244 	} else {
6245 		expectation->rd_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6246 	}
6247 }
6248 
6249 /*
6250  * get_x2apic_wr_val() creates an innocuous write value for an x2APIC register.
6251  *
6252  * For writable registers, get_x2apic_wr_val() deposits the write value into the
6253  * val pointer arg and returns true. For non-writable registers, val is not
6254  * modified and get_x2apic_wr_val() returns false.
6255  */
6256 static bool get_x2apic_wr_val(u32 reg, u64 *val)
6257 {
6258 	switch (reg) {
6259 	case APIC_TASKPRI:
6260 		/* Bits 31:8 are reserved. */
6261 		*val &= 0xff;
6262 		break;
6263 	case APIC_EOI:
6264 	case APIC_ESR:
6265 	case APIC_TMICT:
6266 		/*
6267 		 * EOI, ESR: WRMSR of a non-zero value causes #GP(0).
6268 		 * TMICT: A write of 0 to the initial-count register effectively
6269 		 *        stops the local APIC timer, in both one-shot and
6270 		 *        periodic mode.
6271 		 */
6272 		*val = 0;
6273 		break;
6274 	case APIC_SPIV:
6275 	case APIC_LVTT:
6276 	case APIC_LVTTHMR:
6277 	case APIC_LVTPC:
6278 	case APIC_LVT0:
6279 	case APIC_LVT1:
6280 	case APIC_LVTERR:
6281 	case APIC_TDCR:
6282 		/*
6283 		 * To avoid writing a 1 to a reserved bit or causing some other
6284 		 * unintended side effect, read the current value and use it as
6285 		 * the write value.
6286 		 */
6287 		*val = apic_read(reg);
6288 		break;
6289 	case APIC_CMCI:
6290 		if (!apic_lvt_entry_supported(6))
6291 			return false;
6292 		*val = apic_read(reg);
6293 		break;
6294 	case APIC_ICR:
6295 		*val = 0x40000 | 0xf1;
6296 		break;
6297 	case APIC_SELF_IPI:
6298 		/*
6299 		 * With special processing (i.e., virtualize x2APIC mode +
6300 		 * virtual interrupt delivery), writing zero causes an
6301 		 * APIC-write VM exit. We plan to add a test for enabling
6302 		 * "virtual-interrupt delivery" in VMCS12, and that's where we
6303 		 * will test a self IPI with special processing.
6304 		 */
6305 		*val = 0x0;
6306 		break;
6307 	default:
6308 		return false;
6309 	}
6310 
6311 	return true;
6312 }
6313 
6314 static bool special_processing_applies(u32 reg, u64 *val,
6315 				       bool virt_int_delivery)
6316 {
6317 	bool special_processing =
6318 		(reg == APIC_TASKPRI) ||
6319 		(virt_int_delivery &&
6320 		 (reg == APIC_EOI || reg == APIC_SELF_IPI));
6321 
6322 	if (special_processing) {
6323 		TEST_ASSERT(get_x2apic_wr_val(reg, val));
6324 		return true;
6325 	}
6326 
6327 	return false;
6328 }
6329 
6330 static void virt_x2apic_mode_wr_expectation(
6331 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6332 	bool virt_int_delivery,
6333 	struct virt_x2apic_mode_expectation *expectation)
6334 {
6335 	expectation->wr_exit_reason = VMX_VMCALL;
6336 	expectation->wr_val = MAGIC_VAL_1;
6337 	expectation->wr_only = false;
6338 
6339 	if (virt_x2apic_mode_on &&
6340 	    special_processing_applies(reg, &expectation->wr_val,
6341 				       virt_int_delivery)) {
6342 		expectation->wr_behavior = X2APIC_ACCESS_VIRTUALIZED;
6343 		if (reg == APIC_SELF_IPI)
6344 			expectation->wr_exit_reason = VMX_APIC_WRITE;
6345 	} else if (!disable_x2apic &&
6346 		   get_x2apic_wr_val(reg, &expectation->wr_val)) {
6347 		expectation->wr_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6348 		if (reg == APIC_EOI || reg == APIC_SELF_IPI)
6349 			expectation->wr_only = true;
6350 		if (reg == APIC_ICR)
6351 			expectation->wr_exit_reason = VMX_EXTINT;
6352 	} else {
6353 		expectation->wr_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6354 		/*
6355 		 * Writing 1 to a reserved bit triggers a #GP.
6356 		 * Thus, set the write value to 0, which seems
6357 		 * the most likely to detect a missed #GP.
6358 		 */
6359 		expectation->wr_val = 0;
6360 	}
6361 }
6362 
6363 static void virt_x2apic_mode_exit_expectation(
6364 	u32 reg, struct virt_x2apic_mode_config *config,
6365 	struct virt_x2apic_mode_expectation *expectation)
6366 {
6367 	struct apic_reg_virt_config *base_config =
6368 		&config->apic_reg_virt_config;
6369 	bool virt_x2apic_mode_on =
6370 		base_config->virtualize_x2apic_mode &&
6371 		config->use_msr_bitmaps &&
6372 		config->disable_x2apic_msr_intercepts &&
6373 		base_config->activate_secondary_controls;
6374 
6375 	virt_x2apic_mode_wr_expectation(
6376 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6377 		config->virtual_interrupt_delivery, expectation);
6378 	virt_x2apic_mode_rd_expectation(
6379 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6380 		base_config->apic_register_virtualization,
6381 		config->virtual_interrupt_delivery, expectation);
6382 }
6383 
6384 struct virt_x2apic_mode_test_case virt_x2apic_mode_tests[] = {
6385 	/*
6386 	 * Baseline "virtualize x2APIC mode" configuration:
6387 	 *   - virtualize x2APIC mode
6388 	 *   - virtual-interrupt delivery
6389 	 *   - APIC-register virtualization
6390 	 *   - x2APIC MSR intercepts disabled
6391 	 *
6392 	 * Reads come from virtual APIC page, special processing applies to
6393 	 * VTPR, EOI, and SELF IPI, and all other writes pass through to L1
6394 	 * APIC.
6395 	 */
6396 	{
6397 		.name = "Baseline",
6398 		.virt_x2apic_mode_config = {
6399 			.virtual_interrupt_delivery = true,
6400 			.use_msr_bitmaps = true,
6401 			.disable_x2apic_msr_intercepts = true,
6402 			.disable_x2apic = false,
6403 			.apic_reg_virt_config = {
6404 				.apic_register_virtualization = true,
6405 				.use_tpr_shadow = true,
6406 				.virtualize_apic_accesses = false,
6407 				.virtualize_x2apic_mode = true,
6408 				.activate_secondary_controls = true,
6409 			},
6410 		},
6411 	},
6412 	{
6413 		.name = "Baseline w/ x2apic disabled",
6414 		.virt_x2apic_mode_config = {
6415 			.virtual_interrupt_delivery = true,
6416 			.use_msr_bitmaps = true,
6417 			.disable_x2apic_msr_intercepts = true,
6418 			.disable_x2apic = true,
6419 			.apic_reg_virt_config = {
6420 				.apic_register_virtualization = true,
6421 				.use_tpr_shadow = true,
6422 				.virtualize_apic_accesses = false,
6423 				.virtualize_x2apic_mode = true,
6424 				.activate_secondary_controls = true,
6425 			},
6426 		},
6427 	},
6428 
6429 	/*
6430 	 * Baseline, minus virtual-interrupt delivery. Reads come from virtual
6431 	 * APIC page, special processing applies to VTPR, and all other writes
6432 	 * pass through to L1 APIC.
6433 	 */
6434 	{
6435 		.name = "Baseline - virtual interrupt delivery",
6436 		.virt_x2apic_mode_config = {
6437 			.virtual_interrupt_delivery = false,
6438 			.use_msr_bitmaps = true,
6439 			.disable_x2apic_msr_intercepts = true,
6440 			.disable_x2apic = false,
6441 			.apic_reg_virt_config = {
6442 				.apic_register_virtualization = true,
6443 				.use_tpr_shadow = true,
6444 				.virtualize_apic_accesses = false,
6445 				.virtualize_x2apic_mode = true,
6446 				.activate_secondary_controls = true,
6447 			},
6448 		},
6449 	},
6450 
6451 	/*
6452 	 * Baseline, minus APIC-register virtualization. x2APIC reads pass
6453 	 * through to L1's APIC, unless reading VTPR
6454 	 */
6455 	{
6456 		.name = "Virtualize x2APIC mode, no APIC reg virt",
6457 		.virt_x2apic_mode_config = {
6458 			.virtual_interrupt_delivery = true,
6459 			.use_msr_bitmaps = true,
6460 			.disable_x2apic_msr_intercepts = true,
6461 			.disable_x2apic = false,
6462 			.apic_reg_virt_config = {
6463 				.apic_register_virtualization = false,
6464 				.use_tpr_shadow = true,
6465 				.virtualize_apic_accesses = false,
6466 				.virtualize_x2apic_mode = true,
6467 				.activate_secondary_controls = true,
6468 			},
6469 		},
6470 	},
6471 	{
6472 		.name = "Virtualize x2APIC mode, no APIC reg virt, x2APIC off",
6473 		.virt_x2apic_mode_config = {
6474 			.virtual_interrupt_delivery = true,
6475 			.use_msr_bitmaps = true,
6476 			.disable_x2apic_msr_intercepts = true,
6477 			.disable_x2apic = true,
6478 			.apic_reg_virt_config = {
6479 				.apic_register_virtualization = false,
6480 				.use_tpr_shadow = true,
6481 				.virtualize_apic_accesses = false,
6482 				.virtualize_x2apic_mode = true,
6483 				.activate_secondary_controls = true,
6484 			},
6485 		},
6486 	},
6487 
6488 	/*
6489 	 * Enable "virtualize x2APIC mode" and "APIC-register virtualization",
6490 	 * and disable intercepts for the x2APIC MSRs, but fail to enable
6491 	 * "activate secondary controls" (i.e. L2 gets access to L1's x2APIC
6492 	 * MSRs).
6493 	 */
6494 	{
6495 		.name = "Fail to enable activate secondary controls",
6496 		.virt_x2apic_mode_config = {
6497 			.virtual_interrupt_delivery = true,
6498 			.use_msr_bitmaps = true,
6499 			.disable_x2apic_msr_intercepts = true,
6500 			.disable_x2apic = false,
6501 			.apic_reg_virt_config = {
6502 				.apic_register_virtualization = true,
6503 				.use_tpr_shadow = true,
6504 				.virtualize_apic_accesses = false,
6505 				.virtualize_x2apic_mode = true,
6506 				.activate_secondary_controls = false,
6507 			},
6508 		},
6509 	},
6510 
6511 	/*
6512 	 * Enable "APIC-register virtualization" and enable "activate secondary
6513 	 * controls" and disable intercepts for the x2APIC MSRs, but do not
6514 	 * enable the "virtualize x2APIC mode" VM-execution control (i.e. L2
6515 	 * gets access to L1's x2APIC MSRs).
6516 	 */
6517 	{
6518 		.name = "Fail to enable virtualize x2APIC mode",
6519 		.virt_x2apic_mode_config = {
6520 			.virtual_interrupt_delivery = true,
6521 			.use_msr_bitmaps = true,
6522 			.disable_x2apic_msr_intercepts = true,
6523 			.disable_x2apic = false,
6524 			.apic_reg_virt_config = {
6525 				.apic_register_virtualization = true,
6526 				.use_tpr_shadow = true,
6527 				.virtualize_apic_accesses = false,
6528 				.virtualize_x2apic_mode = false,
6529 				.activate_secondary_controls = true,
6530 			},
6531 		},
6532 	},
6533 
6534 	/*
6535 	 * Disable "Virtualize x2APIC mode", disable x2APIC MSR intercepts, and
6536 	 * enable "APIC-register virtualization" --> L2 gets L1's x2APIC MSRs.
6537 	 */
6538 	{
6539 		.name = "Baseline",
6540 		.virt_x2apic_mode_config = {
6541 			.virtual_interrupt_delivery = true,
6542 			.use_msr_bitmaps = true,
6543 			.disable_x2apic_msr_intercepts = true,
6544 			.disable_x2apic = false,
6545 			.apic_reg_virt_config = {
6546 				.apic_register_virtualization = true,
6547 				.use_tpr_shadow = true,
6548 				.virtualize_apic_accesses = false,
6549 				.virtualize_x2apic_mode = false,
6550 				.activate_secondary_controls = true,
6551 			},
6552 		},
6553 	},
6554 };
6555 
6556 enum X2apic_op {
6557 	X2APIC_OP_RD,
6558 	X2APIC_OP_WR,
6559 	X2APIC_TERMINATE,
6560 };
6561 
6562 static u64 vmx_x2apic_read(u32 reg)
6563 {
6564 	u32 msr_addr = x2apic_msr(reg);
6565 	u64 val;
6566 
6567 	val = rdmsr(msr_addr);
6568 
6569 	return val;
6570 }
6571 
6572 static void vmx_x2apic_write(u32 reg, u64 val)
6573 {
6574 	u32 msr_addr = x2apic_msr(reg);
6575 
6576 	wrmsr(msr_addr, val);
6577 }
6578 
6579 struct virt_x2apic_mode_guest_args {
6580 	enum X2apic_op op;
6581 	u32 reg;
6582 	u64 val;
6583 	bool should_gp;
6584 	u64 (*virt_fn)(u64);
6585 } virt_x2apic_mode_guest_args;
6586 
6587 static volatile bool handle_x2apic_gp_ran;
6588 static volatile u32 handle_x2apic_gp_insn_len;
6589 static void handle_x2apic_gp(struct ex_regs *regs)
6590 {
6591 	handle_x2apic_gp_ran = true;
6592 	regs->rip += handle_x2apic_gp_insn_len;
6593 }
6594 
6595 static handler setup_x2apic_gp_handler(void)
6596 {
6597 	handler old_handler;
6598 
6599 	old_handler = handle_exception(GP_VECTOR, handle_x2apic_gp);
6600 	/* RDMSR and WRMSR are both 2 bytes, assuming no prefixes. */
6601 	handle_x2apic_gp_insn_len = 2;
6602 
6603 	return old_handler;
6604 }
6605 
6606 static void teardown_x2apic_gp_handler(handler old_handler)
6607 {
6608 	handle_exception(GP_VECTOR, old_handler);
6609 
6610 	/*
6611 	 * Defensively reset instruction length, so that if the handler is
6612 	 * incorrectly used, it will loop infinitely, rather than run off into
6613 	 * la la land.
6614 	 */
6615 	handle_x2apic_gp_insn_len = 0;
6616 	handle_x2apic_gp_ran = false;
6617 }
6618 
6619 static void virt_x2apic_mode_guest(void)
6620 {
6621 	volatile struct virt_x2apic_mode_guest_args *args =
6622 		&virt_x2apic_mode_guest_args;
6623 
6624 	for (;;) {
6625 		enum X2apic_op op = args->op;
6626 		u32 reg = args->reg;
6627 		u64 val = args->val;
6628 		bool should_gp = args->should_gp;
6629 		u64 (*virt_fn)(u64) = args->virt_fn;
6630 		handler old_handler;
6631 
6632 		if (op == X2APIC_TERMINATE)
6633 			break;
6634 
6635 		if (should_gp) {
6636 			TEST_ASSERT(!handle_x2apic_gp_ran);
6637 			old_handler = setup_x2apic_gp_handler();
6638 		}
6639 
6640 		if (op == X2APIC_OP_RD) {
6641 			u64 ret = vmx_x2apic_read(reg);
6642 
6643 			if (!should_gp) {
6644 				u64 want = virt_fn(val);
6645 				u64 got = virt_fn(ret);
6646 
6647 				report(got == want,
6648 				       "APIC read; got 0x%lx, want 0x%lx.",
6649 				       got, want);
6650 			}
6651 		} else if (op == X2APIC_OP_WR) {
6652 			vmx_x2apic_write(reg, val);
6653 		}
6654 
6655 		if (should_gp) {
6656 			report(handle_x2apic_gp_ran,
6657 			       "x2APIC op triggered GP.");
6658 			teardown_x2apic_gp_handler(old_handler);
6659 		}
6660 
6661 		/*
6662 		 * The L1 should always execute a vmcall after it's done testing
6663 		 * an individual APIC operation. This helps to validate that the
6664 		 * L1 and L2 are in sync with each other, as expected.
6665 		 */
6666 		vmcall();
6667 	}
6668 }
6669 
6670 static void test_x2apic_rd(
6671 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6672 	u32 *virtual_apic_page)
6673 {
6674 	u64 val = expectation->rd_val;
6675 	u32 exit_reason_want = expectation->rd_exit_reason;
6676 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6677 
6678 	report_prefix_pushf("x2apic - reading 0x%03x", reg);
6679 
6680 	/* Configure guest to do an x2apic read */
6681 	args->op = X2APIC_OP_RD;
6682 	args->reg = reg;
6683 	args->val = val;
6684 	args->should_gp = expectation->rd_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6685 	args->virt_fn = expectation->virt_fn;
6686 
6687 	/* Setup virtual APIC page */
6688 	if (expectation->rd_behavior == X2APIC_ACCESS_VIRTUALIZED)
6689 		virtual_apic_page[apic_reg_index(reg)] = (u32)val;
6690 
6691 	/* Enter guest */
6692 	enter_guest();
6693 
6694 	if (exit_reason_want != VMX_VMCALL) {
6695 		report(false, "Oops, bad exit expectation: %u.",
6696 		       exit_reason_want);
6697 	}
6698 
6699 	skip_exit_vmcall();
6700 	report_prefix_pop();
6701 }
6702 
6703 static volatile bool handle_x2apic_ipi_ran;
6704 static void handle_x2apic_ipi(isr_regs_t *regs)
6705 {
6706 	handle_x2apic_ipi_ran = true;
6707 	eoi();
6708 }
6709 
6710 static void test_x2apic_wr(
6711 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6712 	u32 *virtual_apic_page)
6713 {
6714 	u64 val = expectation->wr_val;
6715 	u32 exit_reason_want = expectation->wr_exit_reason;
6716 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6717 	int ipi_vector = 0xf1;
6718 	u32 restore_val = 0;
6719 
6720 	report_prefix_pushf("x2apic - writing 0x%lx to 0x%03x", val, reg);
6721 
6722 	/* Configure guest to do an x2apic read */
6723 	args->op = X2APIC_OP_WR;
6724 	args->reg = reg;
6725 	args->val = val;
6726 	args->should_gp = expectation->wr_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6727 
6728 	/* Setup virtual APIC page */
6729 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED)
6730 		virtual_apic_page[apic_reg_index(reg)] = 0;
6731 	if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH && !expectation->wr_only)
6732 		restore_val = apic_read(reg);
6733 
6734 	/* Setup IPI handler */
6735 	handle_x2apic_ipi_ran = false;
6736 	handle_irq(ipi_vector, handle_x2apic_ipi);
6737 
6738 	/* Enter guest */
6739 	enter_guest();
6740 
6741 	/*
6742 	 * Validate the behavior and
6743 	 * pass a magic value back to the guest.
6744 	 */
6745 	if (exit_reason_want == VMX_EXTINT) {
6746 		assert_exit_reason(exit_reason_want);
6747 
6748 		/* Clear the external interrupt. */
6749 		irq_enable();
6750 		asm volatile ("nop");
6751 		irq_disable();
6752 		report(handle_x2apic_ipi_ran,
6753 		       "Got pending interrupt after IRQ enabled.");
6754 
6755 		enter_guest();
6756 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6757 		assert_exit_reason(exit_reason_want);
6758 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6759 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%lx",
6760 		       apic_reg_index(reg),
6761 		       virtual_apic_page[apic_reg_index(reg)], val);
6762 
6763 		/* Reenter guest so it can consume/check rcx and exit again. */
6764 		enter_guest();
6765 	} else if (exit_reason_want != VMX_VMCALL) {
6766 		report(false, "Oops, bad exit expectation: %u.",
6767 		       exit_reason_want);
6768 	}
6769 
6770 	assert_exit_reason(VMX_VMCALL);
6771 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) {
6772 		u64 want = val;
6773 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6774 
6775 		report(got == want, "x2APIC write; got 0x%x, want 0x%lx", got,
6776 		       want);
6777 	} else if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH) {
6778 		if (!expectation->wr_only) {
6779 			u32 got = apic_read(reg);
6780 			bool ok;
6781 
6782 			/*
6783 			 * When L1's TPR is passed through to L2, the lower
6784 			 * nibble can be lost. For example, if L2 executes
6785 			 * WRMSR(0x808, 0x78), then, L1 might read 0x70.
6786 			 *
6787 			 * Here's how the lower nibble can get lost:
6788 			 *   1. L2 executes WRMSR(0x808, 0x78).
6789 			 *   2. L2 exits to L0 with a WRMSR exit.
6790 			 *   3. L0 emulates WRMSR, by writing L1's TPR.
6791 			 *   4. L0 re-enters L2.
6792 			 *   5. L2 exits to L0 (reason doesn't matter).
6793 			 *   6. L0 reflects L2's exit to L1.
6794 			 *   7. Before entering L1, L0 exits to user-space
6795 			 *      (e.g., to satisfy TPR access reporting).
6796 			 *   8. User-space executes KVM_SET_REGS ioctl, which
6797 			 *      clears the lower nibble of L1's TPR.
6798 			 */
6799 			if (reg == APIC_TASKPRI) {
6800 				got = apic_virt_nibble1(got);
6801 				val = apic_virt_nibble1(val);
6802 			}
6803 
6804 			ok = got == val;
6805 			report(ok,
6806 			       "non-virtualized write; val is 0x%x, want 0x%lx",
6807 			       got, val);
6808 			apic_write(reg, restore_val);
6809 		} else {
6810 			report(true, "non-virtualized and write-only OK");
6811 		}
6812 	}
6813 	skip_exit_insn();
6814 
6815 	report_prefix_pop();
6816 }
6817 
6818 static enum Config_type configure_virt_x2apic_mode_test(
6819 	struct virt_x2apic_mode_config *virt_x2apic_mode_config,
6820 	u8 *msr_bitmap_page)
6821 {
6822 	int msr;
6823 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6824 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6825 
6826 	/* x2apic-specific VMCS config */
6827 	if (virt_x2apic_mode_config->use_msr_bitmaps) {
6828 		/* virt_x2apic_mode_test() checks for MSR bitmaps support */
6829 		cpu_exec_ctrl0 |= CPU_MSR_BITMAP;
6830 	} else {
6831 		cpu_exec_ctrl0 &= ~CPU_MSR_BITMAP;
6832 	}
6833 
6834 	if (virt_x2apic_mode_config->virtual_interrupt_delivery) {
6835 		if (!(ctrl_cpu_rev[1].clr & CPU_VINTD)) {
6836 			report_skip("VM-execution control \"virtual-interrupt delivery\" NOT supported.\n");
6837 			return CONFIG_TYPE_UNSUPPORTED;
6838 		}
6839 		cpu_exec_ctrl1 |= CPU_VINTD;
6840 	} else {
6841 		cpu_exec_ctrl1 &= ~CPU_VINTD;
6842 	}
6843 
6844 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6845 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6846 
6847 	/* x2APIC MSR intercepts are usually off for "Virtualize x2APIC mode" */
6848 	for (msr = 0x800; msr <= 0x8ff; msr++) {
6849 		if (virt_x2apic_mode_config->disable_x2apic_msr_intercepts) {
6850 			clear_bit(msr, msr_bitmap_page + 0x000);
6851 			clear_bit(msr, msr_bitmap_page + 0x800);
6852 		} else {
6853 			set_bit(msr, msr_bitmap_page + 0x000);
6854 			set_bit(msr, msr_bitmap_page + 0x800);
6855 		}
6856 	}
6857 
6858 	/* x2APIC mode can impact virtualization */
6859 	reset_apic();
6860 	if (!virt_x2apic_mode_config->disable_x2apic)
6861 		enable_x2apic();
6862 
6863 	return configure_apic_reg_virt_test(
6864 		&virt_x2apic_mode_config->apic_reg_virt_config);
6865 }
6866 
6867 static void virt_x2apic_mode_test(void)
6868 {
6869 	u32 *virtual_apic_page;
6870 	u8 *msr_bitmap_page;
6871 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6872 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6873 	int i;
6874 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6875 
6876 	if (!cpu_has_apicv()) {
6877 		report_skip(__func__);
6878 		return;
6879 	}
6880 
6881 	/*
6882 	 * This is to exercise an issue in KVM's logic to merge L0's and L1's
6883 	 * MSR bitmaps. Previously, an L1 could get at L0's x2APIC MSRs by
6884 	 * writing the IA32_SPEC_CTRL MSR or the IA32_PRED_CMD MSRs. KVM would
6885 	 * then proceed to manipulate the MSR bitmaps, as if VMCS12 had the
6886 	 * "Virtualize x2APIC mod" control set, even when it didn't.
6887 	 */
6888 	if (has_spec_ctrl())
6889 		wrmsr(MSR_IA32_SPEC_CTRL, 1);
6890 
6891 	/*
6892 	 * Check that VMCS12 supports:
6893 	 *   - "Virtual-APIC address", indicated by "use TPR shadow"
6894 	 *   - "MSR-bitmap address", indicated by "use MSR bitmaps"
6895 	 */
6896 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6897 		report_skip("VM-execution control \"use TPR shadow\" NOT supported.\n");
6898 		return;
6899 	} else if (!(ctrl_cpu_rev[0].clr & CPU_MSR_BITMAP)) {
6900 		report_skip("VM-execution control \"use MSR bitmaps\" NOT supported.\n");
6901 		return;
6902 	}
6903 
6904 	test_set_guest(virt_x2apic_mode_guest);
6905 
6906 	virtual_apic_page = alloc_page();
6907 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6908 
6909 	msr_bitmap_page = alloc_page();
6910 	memset(msr_bitmap_page, 0xff, PAGE_SIZE);
6911 	vmcs_write(MSR_BITMAP, virt_to_phys(msr_bitmap_page));
6912 
6913 	for (i = 0; i < ARRAY_SIZE(virt_x2apic_mode_tests); i++) {
6914 		struct virt_x2apic_mode_test_case *virt_x2apic_mode_test_case =
6915 			&virt_x2apic_mode_tests[i];
6916 		struct virt_x2apic_mode_config *virt_x2apic_mode_config =
6917 			&virt_x2apic_mode_test_case->virt_x2apic_mode_config;
6918 		enum Config_type config_type;
6919 		u32 reg;
6920 
6921 		printf("--- %s test ---\n", virt_x2apic_mode_test_case->name);
6922 		config_type =
6923 			configure_virt_x2apic_mode_test(virt_x2apic_mode_config,
6924 							msr_bitmap_page);
6925 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6926 			report_skip("Skip because of missing features.\n");
6927 			continue;
6928 		} else if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6929 			enter_guest_with_bad_controls();
6930 			continue;
6931 		}
6932 
6933 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6934 			struct virt_x2apic_mode_expectation expectation;
6935 
6936 			virt_x2apic_mode_exit_expectation(
6937 				reg, virt_x2apic_mode_config, &expectation);
6938 
6939 			test_x2apic_rd(reg, &expectation, virtual_apic_page);
6940 			test_x2apic_wr(reg, &expectation, virtual_apic_page);
6941 		}
6942 	}
6943 
6944 
6945 	/* Terminate the guest */
6946 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6947 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6948 	args->op = X2APIC_TERMINATE;
6949 	enter_guest();
6950 	assert_exit_reason(VMX_VMCALL);
6951 }
6952 
6953 static void test_ctl_reg(const char *cr_name, u64 cr, u64 fixed0, u64 fixed1)
6954 {
6955 	u64 val;
6956 	u64 cr_saved = vmcs_read(cr);
6957 	int i;
6958 
6959 	val = fixed0 & fixed1;
6960 	if (cr == HOST_CR4)
6961 		vmcs_write(cr, val | X86_CR4_PAE);
6962 	else
6963 		vmcs_write(cr, val);
6964 	report_prefix_pushf("%s %lx", cr_name, val);
6965 	if (val == fixed0)
6966 		test_vmx_vmlaunch(0);
6967 	else
6968 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6969 	report_prefix_pop();
6970 
6971 	for (i = 0; i < 64; i++) {
6972 
6973 		/* Set a bit when the corresponding bit in fixed1 is 0 */
6974 		if ((fixed1 & (1ull << i)) == 0) {
6975 			if (cr == HOST_CR4 && ((1ull << i) & X86_CR4_SMEP ||
6976 					       (1ull << i) & X86_CR4_SMAP))
6977 				continue;
6978 
6979 			vmcs_write(cr, cr_saved | (1ull << i));
6980 			report_prefix_pushf("%s %llx", cr_name,
6981 						cr_saved | (1ull << i));
6982 			test_vmx_vmlaunch(
6983 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6984 			report_prefix_pop();
6985 		}
6986 
6987 		/* Unset a bit when the corresponding bit in fixed0 is 1 */
6988 		if (fixed0 & (1ull << i)) {
6989 			vmcs_write(cr, cr_saved & ~(1ull << i));
6990 			report_prefix_pushf("%s %llx", cr_name,
6991 						cr_saved & ~(1ull << i));
6992 			test_vmx_vmlaunch(
6993 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6994 			report_prefix_pop();
6995 		}
6996 	}
6997 
6998 	vmcs_write(cr, cr_saved);
6999 }
7000 
7001 /*
7002  * 1. The CR0 field must not set any bit to a value not supported in VMX
7003  *    operation.
7004  * 2. The CR4 field must not set any bit to a value not supported in VMX
7005  *    operation.
7006  * 3. On processors that support Intel 64 architecture, the CR3 field must
7007  *    be such that bits 63:52 and bits in the range 51:32 beyond the
7008  *    processor’s physical-address width must be 0.
7009  *
7010  *  [Intel SDM]
7011  */
7012 static void test_host_ctl_regs(void)
7013 {
7014 	u64 fixed0, fixed1, cr3, cr3_saved;
7015 	int i;
7016 
7017 	/* Test CR0 */
7018 	fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0);
7019 	fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1);
7020 	test_ctl_reg("HOST_CR0", HOST_CR0, fixed0, fixed1);
7021 
7022 	/* Test CR4 */
7023 	fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0);
7024 	fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1) &
7025 		 ~(X86_CR4_SMEP | X86_CR4_SMAP);
7026 	test_ctl_reg("HOST_CR4", HOST_CR4, fixed0, fixed1);
7027 
7028 	/* Test CR3 */
7029 	cr3_saved = vmcs_read(HOST_CR3);
7030 	for (i = cpuid_maxphyaddr(); i < 64; i++) {
7031 		cr3 = cr3_saved | (1ul << i);
7032 		vmcs_write(HOST_CR3, cr3);
7033 		report_prefix_pushf("HOST_CR3 %lx", cr3);
7034 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7035 		report_prefix_pop();
7036 	}
7037 
7038 	vmcs_write(HOST_CR3, cr3_saved);
7039 }
7040 
7041 static void test_efer_vmlaunch(u32 fld, bool ok)
7042 {
7043 	if (fld == HOST_EFER) {
7044 		if (ok)
7045 			test_vmx_vmlaunch(0);
7046 		else
7047 			test_vmx_vmlaunch2(VMXERR_ENTRY_INVALID_CONTROL_FIELD,
7048 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7049 	} else {
7050 		test_guest_state("EFER test", !ok, GUEST_EFER, "GUEST_EFER");
7051 	}
7052 }
7053 
7054 static void test_efer_one(u32 fld, const char * fld_name, u64 efer,
7055 			  u32 ctrl_fld, u64 ctrl,
7056 			  int i, const char *efer_bit_name)
7057 {
7058 	bool ok;
7059 
7060 	ok = true;
7061 	if (ctrl_fld == EXI_CONTROLS && (ctrl & EXI_LOAD_EFER)) {
7062 		if (!!(efer & EFER_LMA) != !!(ctrl & EXI_HOST_64))
7063 			ok = false;
7064 		if (!!(efer & EFER_LME) != !!(ctrl & EXI_HOST_64))
7065 			ok = false;
7066 	}
7067 	if (ctrl_fld == ENT_CONTROLS && (ctrl & ENT_LOAD_EFER)) {
7068 		/* Check LMA too since CR0.PG is set.  */
7069 		if (!!(efer & EFER_LMA) != !!(ctrl & ENT_GUEST_64))
7070 			ok = false;
7071 		if (!!(efer & EFER_LME) != !!(ctrl & ENT_GUEST_64))
7072 			ok = false;
7073 	}
7074 
7075 	/*
7076 	 * Skip the test if it would enter the guest in 32-bit mode.
7077 	 * Perhaps write the test in assembly and make sure it
7078 	 * can be run in either mode?
7079 	 */
7080 	if (fld == GUEST_EFER && ok && !(ctrl & ENT_GUEST_64))
7081 		return;
7082 
7083 	vmcs_write(ctrl_fld, ctrl);
7084 	vmcs_write(fld, efer);
7085 	report_prefix_pushf("%s %s bit turned %s, controls %s",
7086 			    fld_name, efer_bit_name,
7087 			    (i & 1) ? "on" : "off",
7088 			    (i & 2) ? "on" : "off");
7089 
7090 	test_efer_vmlaunch(fld, ok);
7091 	report_prefix_pop();
7092 }
7093 
7094 static void test_efer_bit(u32 fld, const char * fld_name,
7095 			  u32 ctrl_fld, u64 ctrl_bit, u64 efer_bit,
7096 			  const char *efer_bit_name)
7097 {
7098 	u64 efer_saved = vmcs_read(fld);
7099 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7100 	int i;
7101 
7102 	for (i = 0; i < 4; i++) {
7103 		u64 efer = efer_saved & ~efer_bit;
7104 		u64 ctrl = ctrl_saved & ~ctrl_bit;
7105 
7106 		if (i & 1)
7107 			efer |= efer_bit;
7108 		if (i & 2)
7109 			ctrl |= ctrl_bit;
7110 
7111 		test_efer_one(fld, fld_name, efer, ctrl_fld, ctrl,
7112 			      i, efer_bit_name);
7113 	}
7114 
7115 	vmcs_write(ctrl_fld, ctrl_saved);
7116 	vmcs_write(fld, efer_saved);
7117 }
7118 
7119 static void test_efer(u32 fld, const char * fld_name, u32 ctrl_fld,
7120 		      u64 ctrl_bit1, u64 ctrl_bit2)
7121 {
7122 	u64 efer_saved = vmcs_read(fld);
7123 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7124 	u64 efer_reserved_bits =  ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
7125 	u64 i;
7126 	u64 efer;
7127 
7128 	if (cpu_has_efer_nx())
7129 		efer_reserved_bits &= ~EFER_NX;
7130 
7131 	if (!ctrl_bit1) {
7132 		printf("\"Load-IA32-EFER\" exit control not supported\n");
7133 		goto test_entry_exit_mode;
7134 	}
7135 
7136 	report_prefix_pushf("%s %lx", fld_name, efer_saved);
7137 	test_efer_vmlaunch(fld, true);
7138 	report_prefix_pop();
7139 
7140 	/*
7141 	 * Check reserved bits
7142 	 */
7143 	vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit1);
7144 	for (i = 0; i < 64; i++) {
7145 		if ((1ull << i) & efer_reserved_bits) {
7146 			efer = efer_saved | (1ull << i);
7147 			vmcs_write(fld, efer);
7148 			report_prefix_pushf("%s %lx", fld_name, efer);
7149 			test_efer_vmlaunch(fld, true);
7150 			report_prefix_pop();
7151 		}
7152 	}
7153 
7154 	vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit1);
7155 	for (i = 0; i < 64; i++) {
7156 		if ((1ull << i) & efer_reserved_bits) {
7157 			efer = efer_saved | (1ull << i);
7158 			vmcs_write(fld, efer);
7159 			report_prefix_pushf("%s %lx", fld_name, efer);
7160 			test_efer_vmlaunch(fld, false);
7161 			report_prefix_pop();
7162 		}
7163 	}
7164 
7165 	vmcs_write(ctrl_fld, ctrl_saved);
7166 	vmcs_write(fld, efer_saved);
7167 
7168 	/*
7169 	 * Check LMA and LME bits
7170 	 */
7171 	test_efer_bit(fld, fld_name,
7172 		      ctrl_fld, ctrl_bit1,
7173 		      EFER_LMA,
7174 		      "EFER_LMA");
7175 	test_efer_bit(fld, fld_name,
7176 		      ctrl_fld, ctrl_bit1,
7177 		      EFER_LME,
7178 		      "EFER_LME");
7179 
7180 test_entry_exit_mode:
7181 	test_efer_bit(fld, fld_name,
7182 		      ctrl_fld, ctrl_bit2,
7183 		      EFER_LMA,
7184 		      "EFER_LMA");
7185 	test_efer_bit(fld, fld_name,
7186 		      ctrl_fld, ctrl_bit2,
7187 		      EFER_LME,
7188 		      "EFER_LME");
7189 }
7190 
7191 /*
7192  * If the 'load IA32_EFER' VM-exit control is 1, bits reserved in the
7193  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7194  * the values of the LMA and LME bits in the field must each be that of
7195  * the 'host address-space size' VM-exit control.
7196  *
7197  *  [Intel SDM]
7198  */
7199 static void test_host_efer(void)
7200 {
7201 	test_efer(HOST_EFER, "HOST_EFER", EXI_CONTROLS,
7202 		  ctrl_exit_rev.clr & EXI_LOAD_EFER,
7203 		  EXI_HOST_64);
7204 }
7205 
7206 /*
7207  * If the 'load IA32_EFER' VM-enter control is 1, bits reserved in the
7208  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7209  * the values of the LMA and LME bits in the field must each be that of
7210  * the 'IA32e-mode guest' VM-exit control.
7211  */
7212 static void test_guest_efer(void)
7213 {
7214 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER)) {
7215 		printf("\"Load-IA32-EFER\" entry control not supported\n");
7216 		return;
7217 	}
7218 
7219 	vmcs_write(GUEST_EFER, rdmsr(MSR_EFER));
7220 	test_efer(GUEST_EFER, "GUEST_EFER", ENT_CONTROLS,
7221 		  ctrl_enter_rev.clr & ENT_LOAD_EFER,
7222 		  ENT_GUEST_64);
7223 }
7224 
7225 /*
7226  * PAT values higher than 8 are uninteresting since they're likely lumped
7227  * in with "8". We only test values above 8 one bit at a time,
7228  * in order to reduce the number of VM-Entries and keep the runtime reasonable.
7229  */
7230 #define	PAT_VAL_LIMIT	8
7231 
7232 static void test_pat(u32 field, const char * field_name, u32 ctrl_field,
7233 		     u64 ctrl_bit)
7234 {
7235 	u32 ctrl_saved = vmcs_read(ctrl_field);
7236 	u64 pat_saved = vmcs_read(field);
7237 	u64 i, val;
7238 	u32 j;
7239 	int error;
7240 
7241 	vmcs_clear_bits(ctrl_field, ctrl_bit);
7242 
7243 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7244 		/* Test PAT0..PAT7 fields */
7245 		for (j = 0; j < (i ? 8 : 1); j++) {
7246 			val = i << j * 8;
7247 			vmcs_write(field, val);
7248 			if (field == HOST_PAT) {
7249 				report_prefix_pushf("%s %lx", field_name, val);
7250 				test_vmx_vmlaunch(0);
7251 				report_prefix_pop();
7252 
7253 			} else {	// GUEST_PAT
7254 				test_guest_state("ENT_LOAD_PAT enabled", false,
7255 						 val, "GUEST_PAT");
7256 			}
7257 		}
7258 	}
7259 
7260 	vmcs_set_bits(ctrl_field, ctrl_bit);
7261 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7262 		/* Test PAT0..PAT7 fields */
7263 		for (j = 0; j < (i ? 8 : 1); j++) {
7264 			val = i << j * 8;
7265 			vmcs_write(field, val);
7266 
7267 			if (field == HOST_PAT) {
7268 				report_prefix_pushf("%s %lx", field_name, val);
7269 				if (i == 0x2 || i == 0x3 || i >= 0x8)
7270 					error =
7271 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
7272 				else
7273 					error = 0;
7274 
7275 				test_vmx_vmlaunch(error);
7276 				report_prefix_pop();
7277 
7278 			} else {	// GUEST_PAT
7279 				error = (i == 0x2 || i == 0x3 || i >= 0x8);
7280 				test_guest_state("ENT_LOAD_PAT enabled", !!error,
7281 						 val, "GUEST_PAT");
7282 			}
7283 
7284 		}
7285 	}
7286 
7287 	vmcs_write(ctrl_field, ctrl_saved);
7288 	vmcs_write(field, pat_saved);
7289 }
7290 
7291 /*
7292  *  If the "load IA32_PAT" VM-exit control is 1, the value of the field
7293  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7294  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7295  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7296  *  6 (WB), or 7 (UC-).
7297  *
7298  *  [Intel SDM]
7299  */
7300 static void test_load_host_pat(void)
7301 {
7302 	/*
7303 	 * "load IA32_PAT" VM-exit control
7304 	 */
7305 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) {
7306 		printf("\"Load-IA32-PAT\" exit control not supported\n");
7307 		return;
7308 	}
7309 
7310 	test_pat(HOST_PAT, "HOST_PAT", EXI_CONTROLS, EXI_LOAD_PAT);
7311 }
7312 
7313 union cpuidA_eax {
7314 	struct {
7315 		unsigned int version_id:8;
7316 		unsigned int num_counters_gp:8;
7317 		unsigned int bit_width:8;
7318 		unsigned int mask_length:8;
7319 	} split;
7320 	unsigned int full;
7321 };
7322 
7323 union cpuidA_edx {
7324 	struct {
7325 		unsigned int num_counters_fixed:5;
7326 		unsigned int bit_width_fixed:8;
7327 		unsigned int reserved:9;
7328 	} split;
7329 	unsigned int full;
7330 };
7331 
7332 static bool valid_pgc(u64 val)
7333 {
7334 	struct cpuid id;
7335 	union cpuidA_eax eax;
7336 	union cpuidA_edx edx;
7337 	u64 mask;
7338 
7339 	id = cpuid(0xA);
7340 	eax.full = id.a;
7341 	edx.full = id.d;
7342 	mask = ~(((1ull << eax.split.num_counters_gp) - 1) |
7343 		 (((1ull << edx.split.num_counters_fixed) - 1) << 32));
7344 
7345 	return !(val & mask);
7346 }
7347 
7348 static void test_pgc_vmlaunch(u32 xerror, u32 xreason, bool xfail, bool host)
7349 {
7350 	u32 inst_err;
7351 	u64 obs;
7352 	bool success;
7353 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7354 
7355 	if (host) {
7356 		success = vmlaunch_succeeds();
7357 		obs = rdmsr(data->msr);
7358 		if (!success) {
7359 			inst_err = vmcs_read(VMX_INST_ERROR);
7360 			report(xerror == inst_err, "vmlaunch failed, "
7361 			       "VMX Inst Error is %d (expected %d)",
7362 			       inst_err, xerror);
7363 		} else {
7364 			report(!data->enabled || data->exp == obs,
7365 			       "Host state is 0x%lx (expected 0x%lx)",
7366 			       obs, data->exp);
7367 			report(success != xfail, "vmlaunch succeeded");
7368 		}
7369 	} else {
7370 		test_guest_state("load GUEST_PERF_GLOBAL_CTRL", xfail,
7371 				 GUEST_PERF_GLOBAL_CTRL,
7372 				 "GUEST_PERF_GLOBAL_CTRL");
7373 	}
7374 }
7375 
7376 /*
7377  * test_load_perf_global_ctrl is a generic function for testing the
7378  * "load IA32_PERF_GLOBAL_CTRL" VM-{Entry,Exit} controls. This test function
7379  * tests the provided ctrl_val when disabled and enabled.
7380  *
7381  * @nr: VMCS field number corresponding to the host/guest state field
7382  * @name: Name of the above VMCS field for printing in test report
7383  * @ctrl_nr: VMCS field number corresponding to the VM-{Entry,Exit} control
7384  * @ctrl_val: Bit to set on the ctrl_field
7385  */
7386 static void test_perf_global_ctrl(u32 nr, const char *name, u32 ctrl_nr,
7387 				  const char *ctrl_name, u64 ctrl_val)
7388 {
7389 	u64 ctrl_saved = vmcs_read(ctrl_nr);
7390 	u64 pgc_saved = vmcs_read(nr);
7391 	u64 i, val;
7392 	bool host = nr == HOST_PERF_GLOBAL_CTRL;
7393 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7394 
7395 	data->msr = MSR_CORE_PERF_GLOBAL_CTRL;
7396 	msr_bmp_init();
7397 	vmcs_write(ctrl_nr, ctrl_saved & ~ctrl_val);
7398 	data->enabled = false;
7399 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=0 on %s",
7400 			    ctrl_name);
7401 
7402 	for (i = 0; i < 64; i++) {
7403 		val = 1ull << i;
7404 		vmcs_write(nr, val);
7405 		report_prefix_pushf("%s = 0x%lx", name, val);
7406 		test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7407 		report_prefix_pop();
7408 	}
7409 	report_prefix_pop();
7410 
7411 	vmcs_write(ctrl_nr, ctrl_saved | ctrl_val);
7412 	data->enabled = true;
7413 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=1 on %s",
7414 			    ctrl_name);
7415 	for (i = 0; i < 64; i++) {
7416 		val = 1ull << i;
7417 		data->exp = val;
7418 		vmcs_write(nr, val);
7419 		report_prefix_pushf("%s = 0x%lx", name, val);
7420 		if (valid_pgc(val)) {
7421 			test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7422 		} else {
7423 			if (host)
7424 				test_pgc_vmlaunch(
7425 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
7426 					0,
7427 					true,
7428 					host);
7429 			else
7430 				test_pgc_vmlaunch(
7431 					0,
7432 					VMX_ENTRY_FAILURE | VMX_FAIL_STATE,
7433 					true,
7434 					host);
7435 		}
7436 		report_prefix_pop();
7437 	}
7438 
7439 	data->enabled = false;
7440 	report_prefix_pop();
7441 	vmcs_write(ctrl_nr, ctrl_saved);
7442 	vmcs_write(nr, pgc_saved);
7443 }
7444 
7445 static void test_load_host_perf_global_ctrl(void)
7446 {
7447 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PERF)) {
7448 		printf("\"load IA32_PERF_GLOBAL_CTRL\" exit control not supported\n");
7449 		return;
7450 	}
7451 
7452 	test_perf_global_ctrl(HOST_PERF_GLOBAL_CTRL, "HOST_PERF_GLOBAL_CTRL",
7453 				   EXI_CONTROLS, "EXI_CONTROLS", EXI_LOAD_PERF);
7454 }
7455 
7456 
7457 static void test_load_guest_perf_global_ctrl(void)
7458 {
7459 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PERF)) {
7460 		printf("\"load IA32_PERF_GLOBAL_CTRL\" entry control not supported\n");
7461 		return;
7462 	}
7463 
7464 	test_perf_global_ctrl(GUEST_PERF_GLOBAL_CTRL, "GUEST_PERF_GLOBAL_CTRL",
7465 				   ENT_CONTROLS, "ENT_CONTROLS", ENT_LOAD_PERF);
7466 }
7467 
7468 
7469 /*
7470  * test_vmcs_field - test a value for the given VMCS field
7471  * @field: VMCS field
7472  * @field_name: string name of VMCS field
7473  * @bit_start: starting bit
7474  * @bit_end: ending bit
7475  * @val: value that the bit range must or must not contain
7476  * @valid_val: whether value given in 'val' must be valid or not
7477  * @error: expected VMCS error when vmentry fails for an invalid value
7478  */
7479 static void test_vmcs_field(u64 field, const char *field_name, u32 bit_start,
7480 			    u32 bit_end, u64 val, bool valid_val, u32 error)
7481 {
7482 	u64 field_saved = vmcs_read(field);
7483 	u32 i;
7484 	u64 tmp;
7485 	u32 bit_on;
7486 	u64 mask = ~0ull;
7487 
7488 	mask = (mask >> bit_end) << bit_end;
7489 	mask = mask | ((1 << bit_start) - 1);
7490 	tmp = (field_saved & mask) | (val << bit_start);
7491 
7492 	vmcs_write(field, tmp);
7493 	report_prefix_pushf("%s %lx", field_name, tmp);
7494 	if (valid_val)
7495 		test_vmx_vmlaunch(0);
7496 	else
7497 		test_vmx_vmlaunch(error);
7498 	report_prefix_pop();
7499 
7500 	for (i = bit_start; i <= bit_end; i = i + 2) {
7501 		bit_on = ((1ull < i) & (val << bit_start)) ? 0 : 1;
7502 		if (bit_on)
7503 			tmp = field_saved | (1ull << i);
7504 		else
7505 			tmp = field_saved & ~(1ull << i);
7506 		vmcs_write(field, tmp);
7507 		report_prefix_pushf("%s %lx", field_name, tmp);
7508 		if (valid_val)
7509 			test_vmx_vmlaunch(error);
7510 		else
7511 			test_vmx_vmlaunch(0);
7512 		report_prefix_pop();
7513 	}
7514 
7515 	vmcs_write(field, field_saved);
7516 }
7517 
7518 static void test_canonical(u64 field, const char * field_name, bool host)
7519 {
7520 	u64 addr_saved = vmcs_read(field);
7521 
7522 	/*
7523 	 * Use the existing value if possible.  Writing a random canonical
7524 	 * value is not an option as doing so would corrupt the field being
7525 	 * tested and likely hose the test.
7526 	 */
7527 	if (is_canonical(addr_saved)) {
7528 		if (host) {
7529 			report_prefix_pushf("%s %lx", field_name, addr_saved);
7530 			test_vmx_vmlaunch(0);
7531 			report_prefix_pop();
7532 		} else {
7533 			test_guest_state("Test canonical address", false,
7534 					 addr_saved, field_name);
7535 		}
7536 	}
7537 
7538 	vmcs_write(field, NONCANONICAL);
7539 
7540 	if (host) {
7541 		report_prefix_pushf("%s %llx", field_name, NONCANONICAL);
7542 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7543 		report_prefix_pop();
7544 	} else {
7545 		test_guest_state("Test non-canonical address", true,
7546 				 NONCANONICAL, field_name);
7547 	}
7548 
7549 	vmcs_write(field, addr_saved);
7550 }
7551 
7552 #define TEST_RPL_TI_FLAGS(reg, name)				\
7553 	test_vmcs_field(reg, name, 0, 2, 0x0, true,		\
7554 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7555 
7556 #define TEST_CS_TR_FLAGS(reg, name)				\
7557 	test_vmcs_field(reg, name, 3, 15, 0x0000, false,	\
7558 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7559 
7560 /*
7561  * 1. In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the
7562  *    RPL (bits 1:0) and the TI flag (bit 2) must be 0.
7563  * 2. The selector fields for CS and TR cannot be 0000H.
7564  * 3. The selector field for SS cannot be 0000H if the "host address-space
7565  *    size" VM-exit control is 0.
7566  * 4. On processors that support Intel 64 architecture, the base-address
7567  *    fields for FS, GS and TR must contain canonical addresses.
7568  */
7569 static void test_host_segment_regs(void)
7570 {
7571 	u16 selector_saved;
7572 
7573 	/*
7574 	 * Test RPL and TI flags
7575 	 */
7576 	TEST_RPL_TI_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7577 	TEST_RPL_TI_FLAGS(HOST_SEL_SS, "HOST_SEL_SS");
7578 	TEST_RPL_TI_FLAGS(HOST_SEL_DS, "HOST_SEL_DS");
7579 	TEST_RPL_TI_FLAGS(HOST_SEL_ES, "HOST_SEL_ES");
7580 	TEST_RPL_TI_FLAGS(HOST_SEL_FS, "HOST_SEL_FS");
7581 	TEST_RPL_TI_FLAGS(HOST_SEL_GS, "HOST_SEL_GS");
7582 	TEST_RPL_TI_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7583 
7584 	/*
7585 	 * Test that CS and TR fields can not be 0x0000
7586 	 */
7587 	TEST_CS_TR_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7588 	TEST_CS_TR_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7589 
7590 	/*
7591 	 * SS field can not be 0x0000 if "host address-space size" VM-exit
7592 	 * control is 0
7593 	 */
7594 	selector_saved = vmcs_read(HOST_SEL_SS);
7595 	vmcs_write(HOST_SEL_SS, 0);
7596 	report_prefix_pushf("HOST_SEL_SS 0");
7597 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7598 		test_vmx_vmlaunch(0);
7599 	} else {
7600 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7601 	}
7602 	report_prefix_pop();
7603 
7604 	vmcs_write(HOST_SEL_SS, selector_saved);
7605 
7606 #ifdef __x86_64__
7607 	/*
7608 	 * Base address for FS, GS and TR must be canonical
7609 	 */
7610 	test_canonical(HOST_BASE_FS, "HOST_BASE_FS", true);
7611 	test_canonical(HOST_BASE_GS, "HOST_BASE_GS", true);
7612 	test_canonical(HOST_BASE_TR, "HOST_BASE_TR", true);
7613 #endif
7614 }
7615 
7616 /*
7617  *  On processors that support Intel 64 architecture, the base-address
7618  *  fields for GDTR and IDTR must contain canonical addresses.
7619  */
7620 static void test_host_desc_tables(void)
7621 {
7622 #ifdef __x86_64__
7623 	test_canonical(HOST_BASE_GDTR, "HOST_BASE_GDTR", true);
7624 	test_canonical(HOST_BASE_IDTR, "HOST_BASE_IDTR", true);
7625 #endif
7626 }
7627 
7628 /*
7629  * If the "host address-space size" VM-exit control is 0, the following must
7630  * hold:
7631  *    - The "IA-32e mode guest" VM-entry control is 0.
7632  *    - Bit 17 of the CR4 field (corresponding to CR4.PCIDE) is 0.
7633  *    - Bits 63:32 in the RIP field are 0.
7634  *
7635  * If the "host address-space size" VM-exit control is 1, the following must
7636  * hold:
7637  *    - Bit 5 of the CR4 field (corresponding to CR4.PAE) is 1.
7638  *    - The RIP field contains a canonical address.
7639  *
7640  */
7641 static void test_host_addr_size(void)
7642 {
7643 	u64 cr4_saved = vmcs_read(HOST_CR4);
7644 	u64 rip_saved = vmcs_read(HOST_RIP);
7645 	u64 entry_ctrl_saved = vmcs_read(ENT_CONTROLS);
7646 	int i;
7647 	u64 tmp;
7648 
7649 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7650 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7651 		report_prefix_pushf("\"IA-32e mode guest\" enabled");
7652 		test_vmx_vmlaunch(0);
7653 		report_prefix_pop();
7654 
7655 		vmcs_write(HOST_CR4, cr4_saved | X86_CR4_PCIDE);
7656 		report_prefix_pushf("\"CR4.PCIDE\" set");
7657 		test_vmx_vmlaunch(0);
7658 		report_prefix_pop();
7659 
7660 		for (i = 32; i <= 63; i = i + 4) {
7661 			tmp = rip_saved | 1ull << i;
7662 			vmcs_write(HOST_RIP, tmp);
7663 			report_prefix_pushf("HOST_RIP %lx", tmp);
7664 			test_vmx_vmlaunch(0);
7665 			report_prefix_pop();
7666 		}
7667 
7668 		if (cr4_saved & X86_CR4_PAE) {
7669 			vmcs_write(HOST_CR4, cr4_saved  & ~X86_CR4_PAE);
7670 			report_prefix_pushf("\"CR4.PAE\" unset");
7671 			test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7672 		} else {
7673 			report_prefix_pushf("\"CR4.PAE\" set");
7674 			test_vmx_vmlaunch(0);
7675 		}
7676 		report_prefix_pop();
7677 
7678 		vmcs_write(HOST_RIP, NONCANONICAL);
7679 		report_prefix_pushf("HOST_RIP %llx", NONCANONICAL);
7680 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7681 		report_prefix_pop();
7682 
7683 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7684 		vmcs_write(HOST_RIP, rip_saved);
7685 		vmcs_write(HOST_CR4, cr4_saved);
7686 	}
7687 }
7688 
7689 /*
7690  * Check that the virtual CPU checks the VMX Host State Area as
7691  * documented in the Intel SDM.
7692  */
7693 static void vmx_host_state_area_test(void)
7694 {
7695 	/*
7696 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
7697 	 * fail due to invalid guest state, should we make it that
7698 	 * far.
7699 	 */
7700 	vmcs_write(GUEST_RFLAGS, 0);
7701 
7702 	test_host_ctl_regs();
7703 
7704 	test_canonical(HOST_SYSENTER_ESP, "HOST_SYSENTER_ESP", true);
7705 	test_canonical(HOST_SYSENTER_EIP, "HOST_SYSENTER_EIP", true);
7706 
7707 	test_host_efer();
7708 	test_load_host_pat();
7709 	test_host_segment_regs();
7710 	test_host_desc_tables();
7711 	test_host_addr_size();
7712 	test_load_host_perf_global_ctrl();
7713 }
7714 
7715 /*
7716  * If the "load debug controls" VM-entry control is 1, bits 63:32 in
7717  * the DR7 field must be 0.
7718  *
7719  * [Intel SDM]
7720  */
7721 static void test_guest_dr7(void)
7722 {
7723 	u32 ent_saved = vmcs_read(ENT_CONTROLS);
7724 	u64 dr7_saved = vmcs_read(GUEST_DR7);
7725 	u64 val;
7726 	int i;
7727 
7728 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS) {
7729 		vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7730 		for (i = 0; i < 64; i++) {
7731 			val = 1ull << i;
7732 			vmcs_write(GUEST_DR7, val);
7733 			test_guest_state("ENT_LOAD_DBGCTLS disabled", false,
7734 					 val, "GUEST_DR7");
7735 		}
7736 	}
7737 	if (ctrl_enter_rev.clr & ENT_LOAD_DBGCTLS) {
7738 		vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7739 		for (i = 0; i < 64; i++) {
7740 			val = 1ull << i;
7741 			vmcs_write(GUEST_DR7, val);
7742 			test_guest_state("ENT_LOAD_DBGCTLS enabled", i >= 32,
7743 					 val, "GUEST_DR7");
7744 		}
7745 	}
7746 	vmcs_write(GUEST_DR7, dr7_saved);
7747 	vmcs_write(ENT_CONTROLS, ent_saved);
7748 }
7749 
7750 /*
7751  *  If the "load IA32_PAT" VM-entry control is 1, the value of the field
7752  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7753  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7754  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7755  *  6 (WB), or 7 (UC-).
7756  *
7757  *  [Intel SDM]
7758  */
7759 static void test_load_guest_pat(void)
7760 {
7761 	/*
7762 	 * "load IA32_PAT" VM-entry control
7763 	 */
7764 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
7765 		printf("\"Load-IA32-PAT\" entry control not supported\n");
7766 		return;
7767 	}
7768 
7769 	test_pat(GUEST_PAT, "GUEST_PAT", ENT_CONTROLS, ENT_LOAD_PAT);
7770 }
7771 
7772 #define MSR_IA32_BNDCFGS_RSVD_MASK	0x00000ffc
7773 
7774 /*
7775  * If the “load IA32_BNDCFGS†VM-entry control is 1, the following
7776  * checks are performed on the field for the IA32_BNDCFGS MSR:
7777  *
7778  *   —  Bits reserved in the IA32_BNDCFGS MSR must be 0.
7779  *   —  The linear address in bits 63:12 must be canonical.
7780  *
7781  *  [Intel SDM]
7782  */
7783 static void test_load_guest_bndcfgs(void)
7784 {
7785 	u64 bndcfgs_saved = vmcs_read(GUEST_BNDCFGS);
7786 	u64 bndcfgs;
7787 
7788 	if (!(ctrl_enter_rev.clr & ENT_LOAD_BNDCFGS)) {
7789 		printf("\"Load-IA32-BNDCFGS\" entry control not supported\n");
7790 		return;
7791 	}
7792 
7793 	vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7794 
7795 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7796 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7797 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7798 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7799 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7800 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7801 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7802 
7803 	vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7804 
7805 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7806 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7807 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7808 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7809 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7810 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7811 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7812 
7813 	vmcs_write(GUEST_BNDCFGS, bndcfgs_saved);
7814 }
7815 
7816 /*
7817  * Check that the virtual CPU checks the VMX Guest State Area as
7818  * documented in the Intel SDM.
7819  */
7820 static void vmx_guest_state_area_test(void)
7821 {
7822 	vmx_set_test_stage(1);
7823 	test_set_guest(guest_state_test_main);
7824 
7825 	/*
7826 	 * The IA32_SYSENTER_ESP field and the IA32_SYSENTER_EIP field
7827 	 * must each contain a canonical address.
7828 	 */
7829 	test_canonical(GUEST_SYSENTER_ESP, "GUEST_SYSENTER_ESP", false);
7830 	test_canonical(GUEST_SYSENTER_EIP, "GUEST_SYSENTER_EIP", false);
7831 
7832 	test_guest_dr7();
7833 	test_load_guest_pat();
7834 	test_guest_efer();
7835 	test_load_guest_perf_global_ctrl();
7836 	test_load_guest_bndcfgs();
7837 
7838 	/*
7839 	 * Let the guest finish execution
7840 	 */
7841 	vmx_set_test_stage(2);
7842 	enter_guest();
7843 }
7844 
7845 static bool valid_vmcs_for_vmentry(void)
7846 {
7847 	struct vmcs *current_vmcs = NULL;
7848 
7849 	if (vmcs_save(&current_vmcs))
7850 		return false;
7851 
7852 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
7853 }
7854 
7855 static void try_vmentry_in_movss_shadow(void)
7856 {
7857 	u32 vm_inst_err;
7858 	u32 flags;
7859 	bool early_failure = false;
7860 	u32 expected_flags = X86_EFLAGS_FIXED;
7861 	bool valid_vmcs = valid_vmcs_for_vmentry();
7862 
7863 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
7864 
7865 	/*
7866 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
7867 	 * unsupported VMCS component").
7868 	 */
7869 	vmcs_write(~0u, 0);
7870 
7871 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
7872 			      "vmwrite %%rsp, %%rdx;"
7873 			      "mov 0f, %%rax;"
7874 			      "mov %[host_rip], %%edx;"
7875 			      "vmwrite %%rax, %%rdx;"
7876 			      "mov $-1, %%ah;"
7877 			      "sahf;"
7878 			      "mov %%ss, %%ax;"
7879 			      "mov %%ax, %%ss;"
7880 			      "vmlaunch;"
7881 			      "mov $1, %[early_failure];"
7882 			      "0: lahf;"
7883 			      "movzbl %%ah, %[flags]"
7884 			      : [early_failure] "+r" (early_failure),
7885 				[flags] "=&a" (flags)
7886 			      : [host_rsp] "i" (HOST_RSP),
7887 				[host_rip] "i" (HOST_RIP)
7888 			      : "rdx", "cc", "memory");
7889 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
7890 
7891 	report(early_failure, "Early VM-entry failure");
7892 	report(flags == expected_flags, "RFLAGS[8:0] is %x (actual %x)",
7893 	       expected_flags, flags);
7894 	if (valid_vmcs)
7895 		report(vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
7896 		       "VM-instruction error is %d (actual %d)",
7897 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
7898 }
7899 
7900 static void vmentry_movss_shadow_test(void)
7901 {
7902 	struct vmcs *orig_vmcs;
7903 
7904 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
7905 
7906 	/*
7907 	 * Set the launched flag on the current VMCS to verify the correct
7908 	 * error priority, below.
7909 	 */
7910 	test_set_guest(v2_null_test_guest);
7911 	enter_guest();
7912 
7913 	/*
7914 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
7915 	 * fail due to invalid guest state (if we make it that far).
7916 	 */
7917 	vmcs_write(GUEST_RFLAGS, 0);
7918 
7919 	/*
7920 	 * "VM entry with events blocked by MOV SS" takes precedence over
7921 	 * "VMLAUNCH with non-clear VMCS."
7922 	 */
7923 	report_prefix_push("valid current-VMCS");
7924 	try_vmentry_in_movss_shadow();
7925 	report_prefix_pop();
7926 
7927 	/*
7928 	 * VMfailInvalid takes precedence over "VM entry with events
7929 	 * blocked by MOV SS."
7930 	 */
7931 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
7932 	report_prefix_push("no current-VMCS");
7933 	try_vmentry_in_movss_shadow();
7934 	report_prefix_pop();
7935 
7936 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
7937 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
7938 }
7939 
7940 static void vmx_cr_load_test(void)
7941 {
7942 	unsigned long cr3, cr4, orig_cr3, orig_cr4;
7943 
7944 	orig_cr4 = read_cr4();
7945 	orig_cr3 = read_cr3();
7946 
7947 	if (!this_cpu_has(X86_FEATURE_PCID)) {
7948 		report_skip("PCID not detected");
7949 		return;
7950 	}
7951 	if (!this_cpu_has(X86_FEATURE_MCE)) {
7952 		report_skip("MCE not detected");
7953 		return;
7954 	}
7955 
7956 	TEST_ASSERT(!(orig_cr3 & X86_CR3_PCID_MASK));
7957 
7958 	/* Enable PCID for L1. */
7959 	cr4 = orig_cr4 | X86_CR4_PCIDE;
7960 	cr3 = orig_cr3 | 0x1;
7961 	TEST_ASSERT(!write_cr4_checking(cr4));
7962 	write_cr3(cr3);
7963 
7964 	test_set_guest(v2_null_test_guest);
7965 	vmcs_write(HOST_CR4, cr4);
7966 	vmcs_write(HOST_CR3, cr3);
7967 	enter_guest();
7968 
7969 	/*
7970 	 * No exception is expected.
7971 	 *
7972 	 * NB. KVM loads the last guest write to CR4 into CR4 read
7973 	 *     shadow. In order to trigger an exit to KVM, we can toggle a
7974 	 *     bit that is owned by KVM. We use CR4.MCE, which shall
7975 	 *     have no side effect because normally no guest MCE (e.g., as the
7976 	 *     result of bad memory) would happen during this test.
7977 	 */
7978 	TEST_ASSERT(!write_cr4_checking(cr4 ^ X86_CR4_MCE));
7979 
7980 	/* Cleanup L1 state. */
7981 	write_cr3(orig_cr3);
7982 	TEST_ASSERT(!write_cr4_checking(orig_cr4));
7983 }
7984 
7985 static void vmx_nm_test_guest(void)
7986 {
7987 	write_cr0(read_cr0() | X86_CR0_TS);
7988 	asm volatile("fnop");
7989 }
7990 
7991 static void check_nm_exit(const char *test)
7992 {
7993 	u32 reason = vmcs_read(EXI_REASON);
7994 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
7995 	const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
7996 		NM_VECTOR;
7997 
7998 	report(reason == VMX_EXC_NMI && intr_info == expected, "%s", test);
7999 }
8000 
8001 /*
8002  * This test checks that:
8003  *
8004  * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then
8005  *     a subsequent #NM VM-exit is reflected to L1.
8006  *
8007  * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a
8008  *     subsequent #NM VM-exit is reflected to L1.
8009  */
8010 static void vmx_nm_test(void)
8011 {
8012 	unsigned long cr0 = read_cr0();
8013 
8014 	test_set_guest(vmx_nm_test_guest);
8015 
8016 	/*
8017 	 * L1 wants to intercept #NM exceptions encountered in L2.
8018 	 */
8019 	vmcs_write(EXC_BITMAP, 1 << NM_VECTOR);
8020 
8021 	/*
8022 	 * Launch L2 with CR0.TS clear, but don't claim host ownership of
8023 	 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop,
8024 	 * which will raise #NM. L0 should reflect the #NM VM-exit to L1.
8025 	 */
8026 	vmcs_write(CR0_MASK, 0);
8027 	vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS);
8028 	enter_guest();
8029 	check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1");
8030 
8031 	/*
8032 	 * Re-enter L2 at the fnop instruction, with CR0.TS clear but
8033 	 * CR0.EM set. The fnop will still raise #NM, and L0 should
8034 	 * reflect the #NM VM-exit to L1.
8035 	 */
8036 	vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM);
8037 	enter_guest();
8038 	check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1");
8039 
8040 	/*
8041 	 * Re-enter L2 at the fnop instruction, with both CR0.TS and
8042 	 * CR0.EM clear. There will be no #NM, and the L2 guest should
8043 	 * exit normally.
8044 	 */
8045 	vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM));
8046 	enter_guest();
8047 }
8048 
8049 bool vmx_pending_event_ipi_fired;
8050 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
8051 {
8052 	vmx_pending_event_ipi_fired = true;
8053 	eoi();
8054 }
8055 
8056 bool vmx_pending_event_guest_run;
8057 static void vmx_pending_event_guest(void)
8058 {
8059 	vmcall();
8060 	vmx_pending_event_guest_run = true;
8061 }
8062 
8063 static void vmx_pending_event_test_core(bool guest_hlt)
8064 {
8065 	int ipi_vector = 0xf1;
8066 
8067 	vmx_pending_event_ipi_fired = false;
8068 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
8069 
8070 	vmx_pending_event_guest_run = false;
8071 	test_set_guest(vmx_pending_event_guest);
8072 
8073 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
8074 
8075 	enter_guest();
8076 	skip_exit_vmcall();
8077 
8078 	if (guest_hlt)
8079 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8080 
8081 	irq_disable();
8082 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
8083 				   APIC_DM_FIXED | ipi_vector,
8084 				   0);
8085 
8086 	enter_guest();
8087 
8088 	assert_exit_reason(VMX_EXTINT);
8089 	report(!vmx_pending_event_guest_run,
8090 	       "Guest did not run before host received IPI");
8091 
8092 	irq_enable();
8093 	asm volatile ("nop");
8094 	irq_disable();
8095 	report(vmx_pending_event_ipi_fired,
8096 	       "Got pending interrupt after IRQ enabled");
8097 
8098 	if (guest_hlt)
8099 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8100 
8101 	enter_guest();
8102 	report(vmx_pending_event_guest_run,
8103 	       "Guest finished running when no interrupt");
8104 }
8105 
8106 static void vmx_pending_event_test(void)
8107 {
8108 	vmx_pending_event_test_core(false);
8109 }
8110 
8111 static void vmx_pending_event_hlt_test(void)
8112 {
8113 	vmx_pending_event_test_core(true);
8114 }
8115 
8116 static int vmx_window_test_db_count;
8117 
8118 static void vmx_window_test_db_handler(struct ex_regs *regs)
8119 {
8120 	vmx_window_test_db_count++;
8121 }
8122 
8123 static void vmx_nmi_window_test_guest(void)
8124 {
8125 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8126 
8127 	asm volatile("vmcall\n\t"
8128 		     "nop\n\t");
8129 
8130 	handle_exception(DB_VECTOR, NULL);
8131 }
8132 
8133 static void verify_nmi_window_exit(u64 rip)
8134 {
8135 	u32 exit_reason = vmcs_read(EXI_REASON);
8136 
8137 	report(exit_reason == VMX_NMI_WINDOW,
8138 	       "Exit reason (%d) is 'NMI window'", exit_reason);
8139 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8140 	       vmcs_read(GUEST_RIP), rip);
8141 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8142 }
8143 
8144 static void vmx_nmi_window_test(void)
8145 {
8146 	u64 nop_addr;
8147 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8148 
8149 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
8150 		report_skip("CPU does not support the \"Virtual NMIs\" VM-execution control.");
8151 		return;
8152 	}
8153 
8154 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
8155 		report_skip("CPU does not support the \"NMI-window exiting\" VM-execution control.");
8156 		return;
8157 	}
8158 
8159 	vmx_window_test_db_count = 0;
8160 
8161 	report_prefix_push("NMI-window");
8162 	test_set_guest(vmx_nmi_window_test_guest);
8163 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
8164 	enter_guest();
8165 	skip_exit_vmcall();
8166 	nop_addr = vmcs_read(GUEST_RIP);
8167 
8168 	/*
8169 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
8170 	 * RIP will not advance.
8171 	 */
8172 	report_prefix_push("active, no blocking");
8173 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8174 	enter_guest();
8175 	verify_nmi_window_exit(nop_addr);
8176 	report_prefix_pop();
8177 
8178 	/*
8179 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
8180 	 * a VM-exit on the next instruction after the nop. (The nop
8181 	 * is one byte.)
8182 	 */
8183 	report_prefix_push("active, blocking by MOV-SS");
8184 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8185 	enter_guest();
8186 	verify_nmi_window_exit(nop_addr + 1);
8187 	report_prefix_pop();
8188 
8189 	/*
8190 	 * Ask for "NMI-window exiting" (with event injection), and
8191 	 * expect a VM-exit after the event is injected. (RIP should
8192 	 * be at the address specified in the IDT entry for #DB.)
8193 	 */
8194 	report_prefix_push("active, no blocking, injecting #DB");
8195 	vmcs_write(ENT_INTR_INFO,
8196 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8197 	enter_guest();
8198 	verify_nmi_window_exit((u64)db_fault_addr);
8199 	report_prefix_pop();
8200 
8201 	/*
8202 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
8203 	 * a VM-exit after the next IRET (i.e. after the #DB handler
8204 	 * returns). So, RIP should be back at one byte past the nop.
8205 	 */
8206 	report_prefix_push("active, blocking by NMI");
8207 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
8208 	enter_guest();
8209 	verify_nmi_window_exit(nop_addr + 1);
8210 	report(vmx_window_test_db_count == 1,
8211 	       "#DB handler executed once (actual %d times)",
8212 	       vmx_window_test_db_count);
8213 	report_prefix_pop();
8214 
8215 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8216 		report_skip("CPU does not support activity state HLT.");
8217 	} else {
8218 		/*
8219 		 * Ask for "NMI-window exiting" when entering activity
8220 		 * state HLT, and expect an immediate VM-exit. RIP is
8221 		 * still one byte past the nop.
8222 		 */
8223 		report_prefix_push("halted, no blocking");
8224 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8225 		enter_guest();
8226 		verify_nmi_window_exit(nop_addr + 1);
8227 		report_prefix_pop();
8228 
8229 		/*
8230 		 * Ask for "NMI-window exiting" when entering activity
8231 		 * state HLT (with event injection), and expect a
8232 		 * VM-exit after the event is injected. (RIP should be
8233 		 * at the address specified in the IDT entry for #DB.)
8234 		 */
8235 		report_prefix_push("halted, no blocking, injecting #DB");
8236 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8237 		vmcs_write(ENT_INTR_INFO,
8238 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8239 			   DB_VECTOR);
8240 		enter_guest();
8241 		verify_nmi_window_exit((u64)db_fault_addr);
8242 		report_prefix_pop();
8243 	}
8244 
8245 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8246 	enter_guest();
8247 	report_prefix_pop();
8248 }
8249 
8250 static void vmx_intr_window_test_guest(void)
8251 {
8252 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8253 
8254 	/*
8255 	 * The two consecutive STIs are to ensure that only the first
8256 	 * one has a shadow. Note that NOP and STI are one byte
8257 	 * instructions.
8258 	 */
8259 	asm volatile("vmcall\n\t"
8260 		     "nop\n\t"
8261 		     "sti\n\t"
8262 		     "sti\n\t");
8263 
8264 	handle_exception(DB_VECTOR, NULL);
8265 }
8266 
8267 static void verify_intr_window_exit(u64 rip)
8268 {
8269 	u32 exit_reason = vmcs_read(EXI_REASON);
8270 
8271 	report(exit_reason == VMX_INTR_WINDOW,
8272 	       "Exit reason (%d) is 'interrupt window'", exit_reason);
8273 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8274 	       vmcs_read(GUEST_RIP), rip);
8275 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8276 }
8277 
8278 static void vmx_intr_window_test(void)
8279 {
8280 	u64 vmcall_addr;
8281 	u64 nop_addr;
8282 	unsigned int orig_db_gate_type;
8283 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8284 
8285 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
8286 		report_skip("CPU does not support the \"interrupt-window exiting\" VM-execution control.");
8287 		return;
8288 	}
8289 
8290 	/*
8291 	 * Change the IDT entry for #DB from interrupt gate to trap gate,
8292 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
8293 	 * be disabled after vectoring a #DB.
8294 	 */
8295 	orig_db_gate_type = boot_idt[DB_VECTOR].type;
8296 	boot_idt[DB_VECTOR].type = 15;
8297 
8298 	report_prefix_push("interrupt-window");
8299 	test_set_guest(vmx_intr_window_test_guest);
8300 	enter_guest();
8301 	assert_exit_reason(VMX_VMCALL);
8302 	vmcall_addr = vmcs_read(GUEST_RIP);
8303 
8304 	/*
8305 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
8306 	 * no blocking; expect an immediate VM-exit. Note that we have
8307 	 * not advanced past the vmcall instruction yet, so RIP should
8308 	 * point to the vmcall instruction.
8309 	 */
8310 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
8311 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8312 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
8313 	enter_guest();
8314 	verify_intr_window_exit(vmcall_addr);
8315 	report_prefix_pop();
8316 
8317 	/*
8318 	 * Ask for "interrupt-window exiting" (with event injection)
8319 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
8320 	 * the event is injected. That is, RIP should should be at the
8321 	 * address specified in the IDT entry for #DB.
8322 	 */
8323 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #DB");
8324 	vmcs_write(ENT_INTR_INFO,
8325 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8326 	vmcall_addr = vmcs_read(GUEST_RIP);
8327 	enter_guest();
8328 	verify_intr_window_exit((u64)db_fault_addr);
8329 	report_prefix_pop();
8330 
8331 	/*
8332 	 * Let the L2 guest run through the IRET, back to the VMCALL.
8333 	 * We have to clear the "interrupt-window exiting"
8334 	 * VM-execution control, or it would just keep causing
8335 	 * VM-exits. Then, advance past the VMCALL and set the
8336 	 * "interrupt-window exiting" VM-execution control again.
8337 	 */
8338 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8339 	enter_guest();
8340 	skip_exit_vmcall();
8341 	nop_addr = vmcs_read(GUEST_RIP);
8342 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8343 
8344 	/*
8345 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
8346 	 * RFLAGS.IF set, and expect a VM-exit on the next
8347 	 * instruction. (NOP is one byte.)
8348 	 */
8349 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
8350 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8351 	enter_guest();
8352 	verify_intr_window_exit(nop_addr + 1);
8353 	report_prefix_pop();
8354 
8355 	/*
8356 	 * Back up to the NOP and ask for "interrupt-window exiting"
8357 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
8358 	 * on the next instruction. (NOP is one byte.)
8359 	 */
8360 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
8361 	vmcs_write(GUEST_RIP, nop_addr);
8362 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
8363 	enter_guest();
8364 	verify_intr_window_exit(nop_addr + 1);
8365 	report_prefix_pop();
8366 
8367 	/*
8368 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
8369 	 * and expect a VM-exit on the instruction following the STI
8370 	 * shadow. Only the first STI (which is one byte past the NOP)
8371 	 * should have a shadow. The second STI (which is two bytes
8372 	 * past the NOP) has no shadow. Therefore, the interrupt
8373 	 * window opens at three bytes past the NOP.
8374 	 */
8375 	report_prefix_push("active, RFLAGS.IF = 0");
8376 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8377 	enter_guest();
8378 	verify_intr_window_exit(nop_addr + 3);
8379 	report_prefix_pop();
8380 
8381 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8382 		report_skip("CPU does not support activity state HLT.");
8383 	} else {
8384 		/*
8385 		 * Ask for "interrupt-window exiting" when entering
8386 		 * activity state HLT, and expect an immediate
8387 		 * VM-exit. RIP is still three bytes past the nop.
8388 		 */
8389 		report_prefix_push("halted, no blocking");
8390 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8391 		enter_guest();
8392 		verify_intr_window_exit(nop_addr + 3);
8393 		report_prefix_pop();
8394 
8395 		/*
8396 		 * Ask for "interrupt-window exiting" when entering
8397 		 * activity state HLT (with event injection), and
8398 		 * expect a VM-exit after the event is injected. That
8399 		 * is, RIP should should be at the address specified
8400 		 * in the IDT entry for #DB.
8401 		 */
8402 		report_prefix_push("halted, no blocking, injecting #DB");
8403 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8404 		vmcs_write(ENT_INTR_INFO,
8405 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8406 			   DB_VECTOR);
8407 		enter_guest();
8408 		verify_intr_window_exit((u64)db_fault_addr);
8409 		report_prefix_pop();
8410 	}
8411 
8412 	boot_idt[DB_VECTOR].type = orig_db_gate_type;
8413 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8414 	enter_guest();
8415 	report_prefix_pop();
8416 }
8417 
8418 #define GUEST_TSC_OFFSET (1u << 30)
8419 
8420 static u64 guest_tsc;
8421 
8422 static void vmx_store_tsc_test_guest(void)
8423 {
8424 	guest_tsc = rdtsc();
8425 }
8426 
8427 /*
8428  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
8429  * list, the value saved is not subject to the TSC offset that is
8430  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
8431  */
8432 static void vmx_store_tsc_test(void)
8433 {
8434 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
8435 	u64 low, high;
8436 
8437 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
8438 		report_skip("'Use TSC offsetting' not supported");
8439 		return;
8440 	}
8441 
8442 	test_set_guest(vmx_store_tsc_test_guest);
8443 
8444 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
8445 	vmcs_write(EXI_MSR_ST_CNT, 1);
8446 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
8447 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
8448 
8449 	low = rdtsc();
8450 	enter_guest();
8451 	high = rdtsc();
8452 
8453 	report(low + GUEST_TSC_OFFSET <= guest_tsc &&
8454 	       guest_tsc <= high + GUEST_TSC_OFFSET,
8455 	       "RDTSC value in the guest (%lu) is in range [%lu, %lu]",
8456 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
8457 	report(low <= msr_entry.value && msr_entry.value <= high,
8458 	       "IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
8459 	       msr_entry.value, low, high);
8460 }
8461 
8462 static void vmx_preemption_timer_zero_test_db_handler(struct ex_regs *regs)
8463 {
8464 }
8465 
8466 static void vmx_preemption_timer_zero_test_guest(void)
8467 {
8468 	while (vmx_get_test_stage() < 3)
8469 		vmcall();
8470 }
8471 
8472 static void vmx_preemption_timer_zero_activate_preemption_timer(void)
8473 {
8474 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
8475 	vmcs_write(PREEMPT_TIMER_VALUE, 0);
8476 }
8477 
8478 static void vmx_preemption_timer_zero_advance_past_vmcall(void)
8479 {
8480 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
8481 	enter_guest();
8482 	skip_exit_vmcall();
8483 }
8484 
8485 static void vmx_preemption_timer_zero_inject_db(bool intercept_db)
8486 {
8487 	vmx_preemption_timer_zero_activate_preemption_timer();
8488 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK |
8489 		   INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8490 	vmcs_write(EXC_BITMAP, intercept_db ? 1 << DB_VECTOR : 0);
8491 	enter_guest();
8492 }
8493 
8494 static void vmx_preemption_timer_zero_set_pending_dbg(u32 exception_bitmap)
8495 {
8496 	vmx_preemption_timer_zero_activate_preemption_timer();
8497 	vmcs_write(GUEST_PENDING_DEBUG, BIT(12) | DR_TRAP1);
8498 	vmcs_write(EXC_BITMAP, exception_bitmap);
8499 	enter_guest();
8500 }
8501 
8502 static void vmx_preemption_timer_zero_expect_preempt_at_rip(u64 expected_rip)
8503 {
8504 	u32 reason = (u32)vmcs_read(EXI_REASON);
8505 	u64 guest_rip = vmcs_read(GUEST_RIP);
8506 
8507 	report(reason == VMX_PREEMPT && guest_rip == expected_rip,
8508 	       "Exit reason is 0x%x (expected 0x%x) and guest RIP is %lx (0x%lx expected).",
8509 	       reason, VMX_PREEMPT, guest_rip, expected_rip);
8510 }
8511 
8512 /*
8513  * This test ensures that when the VMX preemption timer is zero at
8514  * VM-entry, a VM-exit occurs after any event injection and after any
8515  * pending debug exceptions are raised, but before execution of any
8516  * guest instructions.
8517  */
8518 static void vmx_preemption_timer_zero_test(void)
8519 {
8520 	u64 db_fault_address = (u64)get_idt_addr(&boot_idt[DB_VECTOR]);
8521 	handler old_db;
8522 	u32 reason;
8523 
8524 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
8525 		report_skip("'Activate VMX-preemption timer' not supported");
8526 		return;
8527 	}
8528 
8529 	/*
8530 	 * Install a custom #DB handler that doesn't abort.
8531 	 */
8532 	old_db = handle_exception(DB_VECTOR,
8533 				  vmx_preemption_timer_zero_test_db_handler);
8534 
8535 	test_set_guest(vmx_preemption_timer_zero_test_guest);
8536 
8537 	/*
8538 	 * VMX-preemption timer should fire after event injection.
8539 	 */
8540 	vmx_set_test_stage(0);
8541 	vmx_preemption_timer_zero_inject_db(0);
8542 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8543 	vmx_preemption_timer_zero_advance_past_vmcall();
8544 
8545 	/*
8546 	 * VMX-preemption timer should fire after event injection.
8547 	 * Exception bitmap is irrelevant, since you can't intercept
8548 	 * an event that you injected.
8549 	 */
8550 	vmx_set_test_stage(1);
8551 	vmx_preemption_timer_zero_inject_db(true);
8552 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8553 	vmx_preemption_timer_zero_advance_past_vmcall();
8554 
8555 	/*
8556 	 * VMX-preemption timer should fire after pending debug exceptions
8557 	 * have delivered a #DB trap.
8558 	 */
8559 	vmx_set_test_stage(2);
8560 	vmx_preemption_timer_zero_set_pending_dbg(0);
8561 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8562 	vmx_preemption_timer_zero_advance_past_vmcall();
8563 
8564 	/*
8565 	 * VMX-preemption timer would fire after pending debug exceptions
8566 	 * have delivered a #DB trap, but in this case, the #DB trap is
8567 	 * intercepted.
8568 	 */
8569 	vmx_set_test_stage(3);
8570 	vmx_preemption_timer_zero_set_pending_dbg(1 << DB_VECTOR);
8571 	reason = (u32)vmcs_read(EXI_REASON);
8572 	report(reason == VMX_EXC_NMI, "Exit reason is 0x%x (expected 0x%x)",
8573 	       reason, VMX_EXC_NMI);
8574 
8575 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
8576 	enter_guest();
8577 
8578 	handle_exception(DB_VECTOR, old_db);
8579 }
8580 
8581 static u64 vmx_preemption_timer_tf_test_prev_rip;
8582 
8583 static void vmx_preemption_timer_tf_test_db_handler(struct ex_regs *regs)
8584 {
8585 	extern char vmx_preemption_timer_tf_test_endloop;
8586 
8587 	if (vmx_get_test_stage() == 2) {
8588 		/*
8589 		 * Stage 2 means that we're done, one way or another.
8590 		 * Arrange for the iret to drop us out of the wbinvd
8591 		 * loop and stop single-stepping.
8592 		 */
8593 		regs->rip = (u64)&vmx_preemption_timer_tf_test_endloop;
8594 		regs->rflags &= ~X86_EFLAGS_TF;
8595 	} else if (regs->rip == vmx_preemption_timer_tf_test_prev_rip) {
8596 		/*
8597 		 * The RIP should alternate between the wbinvd and the
8598 		 * jmp instruction in the code below. If we ever see
8599 		 * the same instruction twice in a row, that means a
8600 		 * single-step trap has been dropped. Let the
8601 		 * hypervisor know about the failure by executing a
8602 		 * VMCALL.
8603 		 */
8604 		vmcall();
8605 	}
8606 	vmx_preemption_timer_tf_test_prev_rip = regs->rip;
8607 }
8608 
8609 static void vmx_preemption_timer_tf_test_guest(void)
8610 {
8611 	/*
8612 	 * The hypervisor doesn't intercept WBINVD, so the loop below
8613 	 * shouldn't be a problem--it's just two instructions
8614 	 * executing in VMX non-root mode. However, when the
8615 	 * hypervisor is running in a virtual environment, the parent
8616 	 * hypervisor might intercept WBINVD and emulate it. If the
8617 	 * parent hypervisor is broken, the single-step trap after the
8618 	 * WBINVD might be lost.
8619 	 */
8620 	asm volatile("vmcall\n\t"
8621 		     "0: wbinvd\n\t"
8622 		     "1: jmp 0b\n\t"
8623 		     "vmx_preemption_timer_tf_test_endloop:");
8624 }
8625 
8626 /*
8627  * Ensure that the delivery of a "VMX-preemption timer expired"
8628  * VM-exit doesn't disrupt single-stepping in the guest. Note that
8629  * passing this test doesn't ensure correctness, because the test will
8630  * only fail if the VMX-preemtion timer fires at the right time (or
8631  * the wrong time, as it were).
8632  */
8633 static void vmx_preemption_timer_tf_test(void)
8634 {
8635 	handler old_db;
8636 	u32 reason;
8637 	int i;
8638 
8639 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
8640 		report_skip("'Activate VMX-preemption timer' not supported");
8641 		return;
8642 	}
8643 
8644 	old_db = handle_exception(DB_VECTOR,
8645 				  vmx_preemption_timer_tf_test_db_handler);
8646 
8647 	test_set_guest(vmx_preemption_timer_tf_test_guest);
8648 
8649 	enter_guest();
8650 	skip_exit_vmcall();
8651 
8652 	vmx_set_test_stage(1);
8653 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
8654 	vmcs_write(PREEMPT_TIMER_VALUE, 50000);
8655 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
8656 
8657 	/*
8658 	 * The only exit we should see is "VMX-preemption timer
8659 	 * expired."  If we get a VMCALL exit, that means the #DB
8660 	 * handler has detected a missing single-step trap. It doesn't
8661 	 * matter where the guest RIP is when the VMX-preemption timer
8662 	 * expires (whether it's in the WBINVD loop or in the #DB
8663 	 * handler)--a single-step trap should never be discarded.
8664 	 */
8665 	for (i = 0; i < 10000; i++) {
8666 		enter_guest();
8667 		reason = (u32)vmcs_read(EXI_REASON);
8668 		if (reason == VMX_PREEMPT)
8669 			continue;
8670 		TEST_ASSERT(reason == VMX_VMCALL);
8671 		skip_exit_insn();
8672 		break;
8673 	}
8674 
8675 	report(reason == VMX_PREEMPT, "No single-step traps skipped");
8676 
8677 	vmx_set_test_stage(2);
8678 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
8679 	enter_guest();
8680 
8681 	handle_exception(DB_VECTOR, old_db);
8682 }
8683 
8684 #define VMX_PREEMPTION_TIMER_EXPIRY_CYCLES 1000000
8685 
8686 static u64 vmx_preemption_timer_expiry_start;
8687 static u64 vmx_preemption_timer_expiry_finish;
8688 
8689 static void vmx_preemption_timer_expiry_test_guest(void)
8690 {
8691 	vmcall();
8692 	vmx_preemption_timer_expiry_start = fenced_rdtsc();
8693 
8694 	while (vmx_get_test_stage() == 0)
8695 		vmx_preemption_timer_expiry_finish = fenced_rdtsc();
8696 }
8697 
8698 /*
8699  * Test that the VMX-preemption timer is not excessively delayed.
8700  *
8701  * Per the SDM, volume 3, VM-entry starts the VMX-preemption timer
8702  * with the unsigned value in the VMX-preemption timer-value field,
8703  * and the VMX-preemption timer counts down by 1 every time bit X in
8704  * the TSC changes due to a TSC increment (where X is
8705  * IA32_VMX_MISC[4:0]). If the timer counts down to zero in any state
8706  * other than the wait-for-SIPI state, the logical processor
8707  * transitions to the C0 C-state and causes a VM-exit.
8708  *
8709  * The guest code above reads the starting TSC after VM-entry. At this
8710  * point, the VMX-preemption timer has already been activated. Next,
8711  * the guest code reads the current TSC in a loop, storing the value
8712  * read to memory.
8713  *
8714  * If the RDTSC in the loop reads a value past the VMX-preemption
8715  * timer deadline, then the VMX-preemption timer VM-exit must be
8716  * delivered before the next instruction retires. Even if a higher
8717  * priority SMI is delivered first, the VMX-preemption timer VM-exit
8718  * must be delivered before the next instruction retires. Hence, a TSC
8719  * value past the VMX-preemption timer deadline might be read, but it
8720  * cannot be stored. If a TSC value past the deadline *is* stored,
8721  * then the architectural specification has been violated.
8722  */
8723 static void vmx_preemption_timer_expiry_test(void)
8724 {
8725 	u32 preemption_timer_value;
8726 	union vmx_misc misc;
8727 	u64 tsc_deadline;
8728 	u32 reason;
8729 
8730 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
8731 		report_skip("'Activate VMX-preemption timer' not supported");
8732 		return;
8733 	}
8734 
8735 	test_set_guest(vmx_preemption_timer_expiry_test_guest);
8736 
8737 	enter_guest();
8738 	skip_exit_vmcall();
8739 
8740 	misc.val = rdmsr(MSR_IA32_VMX_MISC);
8741 	preemption_timer_value =
8742 		VMX_PREEMPTION_TIMER_EXPIRY_CYCLES >> misc.pt_bit;
8743 
8744 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
8745 	vmcs_write(PREEMPT_TIMER_VALUE, preemption_timer_value);
8746 	vmx_set_test_stage(0);
8747 
8748 	enter_guest();
8749 	reason = (u32)vmcs_read(EXI_REASON);
8750 	TEST_ASSERT(reason == VMX_PREEMPT);
8751 
8752 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
8753 	vmx_set_test_stage(1);
8754 	enter_guest();
8755 
8756 	tsc_deadline = ((vmx_preemption_timer_expiry_start >> misc.pt_bit) <<
8757 			misc.pt_bit) + (preemption_timer_value << misc.pt_bit);
8758 
8759 	report(vmx_preemption_timer_expiry_finish < tsc_deadline,
8760 	       "Last stored guest TSC (%lu) < TSC deadline (%lu)",
8761 	       vmx_preemption_timer_expiry_finish, tsc_deadline);
8762 }
8763 
8764 static void vmx_db_test_guest(void)
8765 {
8766 	/*
8767 	 * For a hardware generated single-step #DB.
8768 	 */
8769 	asm volatile("vmcall;"
8770 		     "nop;"
8771 		     ".Lpost_nop:");
8772 	/*
8773 	 * ...in a MOVSS shadow, with pending debug exceptions.
8774 	 */
8775 	asm volatile("vmcall;"
8776 		     "nop;"
8777 		     ".Lpost_movss_nop:");
8778 	/*
8779 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
8780 	 * emulates it in software.)
8781 	 */
8782 	asm volatile("vmcall;"
8783 		     "wbinvd;"
8784 		     ".Lpost_wbinvd:");
8785 	/*
8786 	 * ...in a MOVSS shadow, with pending debug exceptions.
8787 	 */
8788 	asm volatile("vmcall;"
8789 		     "wbinvd;"
8790 		     ".Lpost_movss_wbinvd:");
8791 	/*
8792 	 * For a hardware generated single-step #DB in a transactional region.
8793 	 */
8794 	asm volatile("vmcall;"
8795 		     ".Lxbegin: xbegin .Lskip_rtm;"
8796 		     "xend;"
8797 		     ".Lskip_rtm:");
8798 }
8799 
8800 /*
8801  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
8802  * L2. No #DB is delivered and L2 continues to the next point of
8803  * interest.
8804  */
8805 static void dismiss_db(void)
8806 {
8807 	vmcs_write(GUEST_PENDING_DEBUG, 0);
8808 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8809 	enter_guest();
8810 }
8811 
8812 /*
8813  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
8814  * Then throw away the #DB exception and resume L2.
8815  */
8816 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
8817 			  void *expected_rip, u64 expected_exit_qual,
8818 			  u64 expected_dr6)
8819 {
8820 	u32 reason = vmcs_read(EXI_REASON);
8821 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
8822 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
8823 	u64 guest_rip = vmcs_read(GUEST_RIP);
8824 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
8825 	u64 dr6 = read_dr6();
8826 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
8827 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
8828 
8829 	report(reason == VMX_EXC_NMI && intr_info == expected_intr_info,
8830 	       "Expected #DB VM-exit");
8831 	report((u64)expected_rip == guest_rip, "Expected RIP %p (actual %lx)",
8832 	       expected_rip, guest_rip);
8833 	report_xfail(xfail_pdbg, 0 == guest_pending_dbg,
8834 		     "Expected pending debug exceptions 0 (actual %lx)",
8835 		     guest_pending_dbg);
8836 	report_xfail(xfail_qual, expected_exit_qual == exit_qual,
8837 		     "Expected exit qualification %lx (actual %lx)",
8838 		     expected_exit_qual, exit_qual);
8839 	report_xfail(xfail_dr6, expected_dr6 == dr6,
8840 		     "Expected DR6 %lx (actual %lx)", expected_dr6, dr6);
8841 	dismiss_db();
8842 }
8843 
8844 /*
8845  * Assuming the guest has just exited on a VMCALL instruction, skip
8846  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
8847  * pending debug exceptions are non-zero, set the VMCS up as if the
8848  * previous instruction was a MOVSS that generated the indicated
8849  * pending debug exceptions. Then enter L2.
8850  */
8851 static void single_step_guest(const char *test_name, u64 starting_dr6,
8852 			      u64 pending_debug_exceptions)
8853 {
8854 	printf("\n%s\n", test_name);
8855 	skip_exit_vmcall();
8856 	write_dr6(starting_dr6);
8857 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
8858 	if (pending_debug_exceptions) {
8859 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
8860 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8861 	}
8862 	enter_guest();
8863 }
8864 
8865 /*
8866  * When L1 intercepts #DB, verify that a single-step trap clears
8867  * pending debug exceptions, populates the exit qualification field
8868  * properly, and that DR6 is not prematurely clobbered. In a
8869  * (simulated) MOVSS shadow, make sure that the pending debug
8870  * exception bits are properly accumulated into the exit qualification
8871  * field.
8872  */
8873 static void vmx_db_test(void)
8874 {
8875 	/*
8876 	 * We are going to set a few arbitrary bits in DR6 to verify that
8877 	 * (a) DR6 is not modified by an intercepted #DB, and
8878 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
8879          *     the exit qualification field for a subsequent #DB exception.
8880 	 */
8881 	const u64 starting_dr6 = DR6_RESERVED | BIT(13) | DR_TRAP3 | DR_TRAP1;
8882 	extern char post_nop asm(".Lpost_nop");
8883 	extern char post_movss_nop asm(".Lpost_movss_nop");
8884 	extern char post_wbinvd asm(".Lpost_wbinvd");
8885 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
8886 	extern char xbegin asm(".Lxbegin");
8887 	extern char skip_rtm asm(".Lskip_rtm");
8888 
8889 	/*
8890 	 * L1 wants to intercept #DB exceptions encountered in L2.
8891 	 */
8892 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
8893 
8894 	/*
8895 	 * Start L2 and run it up to the first point of interest.
8896 	 */
8897 	test_set_guest(vmx_db_test_guest);
8898 	enter_guest();
8899 
8900 	/*
8901 	 * Hardware-delivered #DB trap for single-step sets the
8902 	 * standard that L0 has to follow for emulated instructions.
8903 	 */
8904 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
8905 	check_db_exit(false, false, false, &post_nop, DR_STEP, starting_dr6);
8906 
8907 	/*
8908 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
8909 	 * also sets the standard that L0 has to follow for emulated
8910 	 * instructions. Here, we establish the VMCS pending debug
8911 	 * exceptions to indicate that the simulated MOVSS triggered a
8912 	 * data breakpoint as well as the single-step trap.
8913 	 */
8914 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
8915 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0 );
8916 	check_db_exit(false, false, false, &post_movss_nop, DR_STEP | DR_TRAP0,
8917 		      starting_dr6);
8918 
8919 	/*
8920 	 * L0 synthesized #DB trap for single-step is buggy, because
8921 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
8922 	 * reconstitute the exit qualification from the prematurely
8923 	 * modified DR6, but fails miserably.
8924 	 */
8925 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
8926 	check_db_exit(false, false, false, &post_wbinvd, DR_STEP, starting_dr6);
8927 
8928 	/*
8929 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
8930 	 * even worse, because L0 also leaves the pending debug
8931 	 * exceptions in the VMCS instead of accumulating them into
8932 	 * the exit qualification field for the #DB exception.
8933 	 */
8934 	single_step_guest("Software synthesized single-step in MOVSS shadow",
8935 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0);
8936 	check_db_exit(true, false, true, &post_movss_wbinvd, DR_STEP | DR_TRAP0,
8937 		      starting_dr6);
8938 
8939 	/*
8940 	 * Optional RTM test for hardware that supports RTM, to
8941 	 * demonstrate that the current volume 3 of the SDM
8942 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
8943 	 * qualification for debug exceptions is not reserved. It is
8944 	 * set to 1 if a debug exception (#DB) or a breakpoint
8945 	 * exception (#BP) occurs inside an RTM region while advanced
8946 	 * debugging of RTM transactional regions is enabled.
8947 	 */
8948 	if (this_cpu_has(X86_FEATURE_RTM)) {
8949 		vmcs_write(ENT_CONTROLS,
8950 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
8951 		/*
8952 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
8953 		 * in the guest to enable advanced debugging of RTM
8954 		 * transactional regions.
8955 		 */
8956 		vmcs_write(GUEST_DR7, BIT(11));
8957 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
8958 		single_step_guest("Hardware delivered single-step in "
8959 				  "transactional region", starting_dr6, 0);
8960 		check_db_exit(false, false, false, &xbegin, BIT(16),
8961 			      starting_dr6);
8962 	} else {
8963 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
8964 		enter_guest();
8965 	}
8966 }
8967 
8968 static void enable_vid(void)
8969 {
8970 	void *virtual_apic_page;
8971 
8972 	assert(cpu_has_apicv());
8973 
8974 	disable_intercept_for_x2apic_msrs();
8975 
8976 	virtual_apic_page = alloc_page();
8977 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
8978 
8979 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
8980 
8981 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
8982 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
8983 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
8984 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
8985 
8986 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
8987 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
8988 }
8989 
8990 static void trigger_ioapic_scan_thread(void *data)
8991 {
8992 	/* Wait until other CPU entered L2 */
8993 	while (vmx_get_test_stage() != 1)
8994 		;
8995 
8996 	/* Trigger ioapic scan */
8997 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
8998 	vmx_set_test_stage(2);
8999 }
9000 
9001 static void irq_79_handler_guest(isr_regs_t *regs)
9002 {
9003 	eoi();
9004 
9005 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
9006 	vmcall();
9007 }
9008 
9009 /*
9010  * Constant for num of busy-loop iterations after which
9011  * a timer interrupt should have happened in host
9012  */
9013 #define TIMER_INTERRUPT_DELAY 100000000
9014 
9015 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
9016 {
9017 	handle_irq(0x79, irq_79_handler_guest);
9018 	irq_enable();
9019 
9020 	/* Signal to L1 CPU to trigger ioapic scan */
9021 	vmx_set_test_stage(1);
9022 	/* Wait until L1 CPU to trigger ioapic scan */
9023 	while (vmx_get_test_stage() != 2)
9024 		;
9025 
9026 	/*
9027 	 * Wait for L0 timer interrupt to be raised while we run in L2
9028 	 * such that L0 will process the IOAPIC scan request before
9029 	 * resuming L2
9030 	 */
9031 	delay(TIMER_INTERRUPT_DELAY);
9032 
9033 	asm volatile ("int $0x79");
9034 }
9035 
9036 static void vmx_eoi_bitmap_ioapic_scan_test(void)
9037 {
9038 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
9039 		report_skip(__func__);
9040 		return;
9041 	}
9042 
9043 	enable_vid();
9044 
9045 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
9046 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
9047 
9048 	/*
9049 	 * Launch L2.
9050 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
9051 	 * In case the reason isn't VMX_VMCALL, the asserion inside
9052 	 * skip_exit_vmcall() will fail.
9053 	 */
9054 	enter_guest();
9055 	skip_exit_vmcall();
9056 
9057 	/* Let L2 finish */
9058 	enter_guest();
9059 	report(1, __func__);
9060 }
9061 
9062 #define HLT_WITH_RVI_VECTOR		(0xf1)
9063 
9064 bool vmx_hlt_with_rvi_guest_isr_fired;
9065 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
9066 {
9067 	vmx_hlt_with_rvi_guest_isr_fired = true;
9068 	eoi();
9069 }
9070 
9071 static void vmx_hlt_with_rvi_guest(void)
9072 {
9073 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
9074 
9075 	irq_enable();
9076 	asm volatile ("nop");
9077 
9078 	vmcall();
9079 }
9080 
9081 static void vmx_hlt_with_rvi_test(void)
9082 {
9083 	if (!cpu_has_apicv()) {
9084 		report_skip(__func__);
9085 		return;
9086 	}
9087 
9088 	enable_vid();
9089 
9090 	vmx_hlt_with_rvi_guest_isr_fired = false;
9091 	test_set_guest(vmx_hlt_with_rvi_guest);
9092 
9093 	enter_guest();
9094 	skip_exit_vmcall();
9095 
9096 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
9097 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
9098 	enter_guest();
9099 
9100 	report(vmx_hlt_with_rvi_guest_isr_fired, "Interrupt raised in guest");
9101 }
9102 
9103 static void set_irq_line_thread(void *data)
9104 {
9105 	/* Wait until other CPU entered L2 */
9106 	while (vmx_get_test_stage() != 1)
9107 		;
9108 
9109 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
9110 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9111 	vmx_set_test_stage(2);
9112 }
9113 
9114 static bool irq_78_handler_vmcall_before_eoi;
9115 static void irq_78_handler_guest(isr_regs_t *regs)
9116 {
9117 	set_irq_line(0xf, 0);
9118 	if (irq_78_handler_vmcall_before_eoi)
9119 		vmcall();
9120 	eoi();
9121 	vmcall();
9122 }
9123 
9124 static void vmx_apic_passthrough_guest(void)
9125 {
9126 	handle_irq(0x78, irq_78_handler_guest);
9127 	irq_enable();
9128 
9129 	/* If requested, wait for other CPU to trigger ioapic scan */
9130 	if (vmx_get_test_stage() < 1) {
9131 		vmx_set_test_stage(1);
9132 		while (vmx_get_test_stage() != 2)
9133 			;
9134 	}
9135 
9136 	set_irq_line(0xf, 1);
9137 }
9138 
9139 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
9140 {
9141 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
9142 		report_skip(__func__);
9143 		return;
9144 	}
9145 
9146 	/* Test device is required for generating IRQs */
9147 	if (!test_device_enabled()) {
9148 		report_skip(__func__);
9149 		return;
9150 	}
9151 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9152 	u64 cpu_ctrl_1 = 0;
9153 
9154 	disable_intercept_for_x2apic_msrs();
9155 
9156 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9157 
9158 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9159 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9160 
9161 	if (set_irq_line_from_thread) {
9162 		irq_78_handler_vmcall_before_eoi = false;
9163 		on_cpu_async(1, set_irq_line_thread, NULL);
9164 	} else {
9165 		irq_78_handler_vmcall_before_eoi = true;
9166 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9167 		vmx_set_test_stage(2);
9168 	}
9169 	test_set_guest(vmx_apic_passthrough_guest);
9170 
9171 	if (irq_78_handler_vmcall_before_eoi) {
9172 		/* Before EOI remote_irr should still be set */
9173 		enter_guest();
9174 		skip_exit_vmcall();
9175 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
9176 			"IOAPIC pass-through: remote_irr=1 before EOI");
9177 	}
9178 
9179 	/* After EOI remote_irr should be cleared */
9180 	enter_guest();
9181 	skip_exit_vmcall();
9182 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
9183 		"IOAPIC pass-through: remote_irr=0 after EOI");
9184 
9185 	/* Let L2 finish */
9186 	enter_guest();
9187 	report(1, __func__);
9188 }
9189 
9190 static void vmx_apic_passthrough_test(void)
9191 {
9192 	vmx_apic_passthrough(false);
9193 }
9194 
9195 static void vmx_apic_passthrough_thread_test(void)
9196 {
9197 	vmx_apic_passthrough(true);
9198 }
9199 
9200 static void vmx_apic_passthrough_tpr_threshold_guest(void)
9201 {
9202 	cli();
9203 	apic_set_tpr(0);
9204 }
9205 
9206 static bool vmx_apic_passthrough_tpr_threshold_ipi_isr_fired;
9207 static void vmx_apic_passthrough_tpr_threshold_ipi_isr(isr_regs_t *regs)
9208 {
9209 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = true;
9210 	eoi();
9211 }
9212 
9213 static void vmx_apic_passthrough_tpr_threshold_test(void)
9214 {
9215 	int ipi_vector = 0xe1;
9216 
9217 	disable_intercept_for_x2apic_msrs();
9218 	vmcs_clear_bits(PIN_CONTROLS, PIN_EXTINT);
9219 
9220 	/* Raise L0 TPR-threshold by queueing vector in LAPIC IRR */
9221 	cli();
9222 	apic_set_tpr((ipi_vector >> 4) + 1);
9223 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
9224 			APIC_DM_FIXED | ipi_vector,
9225 			0);
9226 
9227 	test_set_guest(vmx_apic_passthrough_tpr_threshold_guest);
9228 	enter_guest();
9229 
9230 	report(apic_get_tpr() == 0, "TPR was zero by guest");
9231 
9232 	/* Clean pending self-IPI */
9233 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = false;
9234 	handle_irq(ipi_vector, vmx_apic_passthrough_tpr_threshold_ipi_isr);
9235 	sti();
9236 	asm volatile ("nop");
9237 	report(vmx_apic_passthrough_tpr_threshold_ipi_isr_fired, "self-IPI fired");
9238 
9239 	report(1, __func__);
9240 }
9241 
9242 static u64 init_signal_test_exit_reason;
9243 static bool init_signal_test_thread_continued;
9244 
9245 static void init_signal_test_thread(void *data)
9246 {
9247 	struct vmcs *test_vmcs = data;
9248 
9249 	/* Enter VMX operation (i.e. exec VMXON) */
9250 	u64 *ap_vmxon_region = alloc_page();
9251 	enable_vmx();
9252 	init_vmx(ap_vmxon_region);
9253 	_vmx_on(ap_vmxon_region);
9254 
9255 	/* Signal CPU have entered VMX operation */
9256 	vmx_set_test_stage(1);
9257 
9258 	/* Wait for BSP CPU to send INIT signal */
9259 	while (vmx_get_test_stage() != 2)
9260 		;
9261 
9262 	/*
9263 	 * Signal that we continue as usual as INIT signal
9264 	 * should be blocked while CPU is in VMX operation
9265 	 */
9266 	vmx_set_test_stage(3);
9267 
9268 	/* Wait for signal to enter VMX non-root mode */
9269 	while (vmx_get_test_stage() != 4)
9270 		;
9271 
9272 	/* Enter VMX non-root mode */
9273 	test_set_guest(v2_null_test_guest);
9274 	make_vmcs_current(test_vmcs);
9275 	enter_guest();
9276 	/* Save exit reason for BSP CPU to compare to expected result */
9277 	init_signal_test_exit_reason = vmcs_read(EXI_REASON);
9278 	/* VMCLEAR test-vmcs so it could be loaded by BSP CPU */
9279 	vmcs_clear(test_vmcs);
9280 	launched = false;
9281 	/* Signal that CPU exited to VMX root mode */
9282 	vmx_set_test_stage(5);
9283 
9284 	/* Wait for BSP CPU to signal to exit VMX operation */
9285 	while (vmx_get_test_stage() != 6)
9286 		;
9287 
9288 	/* Exit VMX operation (i.e. exec VMXOFF) */
9289 	vmx_off();
9290 
9291 	/*
9292 	 * Signal to BSP CPU that we continue as usual as INIT signal
9293 	 * should have been consumed by VMX_INIT exit from guest
9294 	 */
9295 	vmx_set_test_stage(7);
9296 
9297 	/* Wait for BSP CPU to signal to enter VMX operation */
9298 	while (vmx_get_test_stage() != 8)
9299 		;
9300 	/* Enter VMX operation (i.e. exec VMXON) */
9301 	_vmx_on(ap_vmxon_region);
9302 	/* Signal to BSP we are in VMX operation */
9303 	vmx_set_test_stage(9);
9304 
9305 	/* Wait for BSP CPU to send INIT signal */
9306 	while (vmx_get_test_stage() != 10)
9307 		;
9308 
9309 	/* Exit VMX operation (i.e. exec VMXOFF) */
9310 	vmx_off();
9311 
9312 	/*
9313 	 * Exiting VMX operation should result in latched
9314 	 * INIT signal being processed. Therefore, we should
9315 	 * never reach the below code. Thus, signal to BSP
9316 	 * CPU if we have reached here so it is able to
9317 	 * report an issue if it happens.
9318 	 */
9319 	init_signal_test_thread_continued = true;
9320 }
9321 
9322 #define INIT_SIGNAL_TEST_DELAY	100000000ULL
9323 
9324 static void vmx_init_signal_test(void)
9325 {
9326 	struct vmcs *test_vmcs;
9327 
9328 	if (cpu_count() < 2) {
9329 		report_skip(__func__);
9330 		return;
9331 	}
9332 
9333 	/* VMCLEAR test-vmcs so it could be loaded by other CPU */
9334 	vmcs_save(&test_vmcs);
9335 	vmcs_clear(test_vmcs);
9336 
9337 	vmx_set_test_stage(0);
9338 	on_cpu_async(1, init_signal_test_thread, test_vmcs);
9339 
9340 	/* Wait for other CPU to enter VMX operation */
9341 	while (vmx_get_test_stage() != 1)
9342 		;
9343 
9344 	/* Send INIT signal to other CPU */
9345 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9346 				   id_map[1]);
9347 	/* Signal other CPU we have sent INIT signal */
9348 	vmx_set_test_stage(2);
9349 
9350 	/*
9351 	 * Wait reasonable amount of time for INIT signal to
9352 	 * be received on other CPU and verify that other CPU
9353 	 * have proceed as usual to next test stage as INIT
9354 	 * signal should be blocked while other CPU in
9355 	 * VMX operation
9356 	 */
9357 	delay(INIT_SIGNAL_TEST_DELAY);
9358 	report(vmx_get_test_stage() == 3,
9359 	       "INIT signal blocked when CPU in VMX operation");
9360 	/* No point to continue if we failed at this point */
9361 	if (vmx_get_test_stage() != 3)
9362 		return;
9363 
9364 	/* Signal other CPU to enter VMX non-root mode */
9365 	init_signal_test_exit_reason = -1ull;
9366 	vmx_set_test_stage(4);
9367 	/*
9368 	 * Wait reasonable amont of time for other CPU
9369 	 * to exit to VMX root mode
9370 	 */
9371 	delay(INIT_SIGNAL_TEST_DELAY);
9372 	if (vmx_get_test_stage() != 5) {
9373 		report(false, "Pending INIT signal didn't result in VMX exit");
9374 		return;
9375 	}
9376 	report(init_signal_test_exit_reason == VMX_INIT,
9377 			"INIT signal during VMX non-root mode result in exit-reason %s (%lu)",
9378 			exit_reason_description(init_signal_test_exit_reason),
9379 			init_signal_test_exit_reason);
9380 
9381 	/* Run guest to completion */
9382 	make_vmcs_current(test_vmcs);
9383 	enter_guest();
9384 
9385 	/* Signal other CPU to exit VMX operation */
9386 	init_signal_test_thread_continued = false;
9387 	vmx_set_test_stage(6);
9388 
9389 	/* Wait reasonable amount of time for other CPU to exit VMX operation */
9390 	delay(INIT_SIGNAL_TEST_DELAY);
9391 	report(vmx_get_test_stage() == 7,
9392 	       "INIT signal consumed on VMX_INIT exit");
9393 	/* No point to continue if we failed at this point */
9394 	if (vmx_get_test_stage() != 7)
9395 		return;
9396 
9397 	/* Signal other CPU to enter VMX operation */
9398 	vmx_set_test_stage(8);
9399 	/* Wait for other CPU to enter VMX operation */
9400 	while (vmx_get_test_stage() != 9)
9401 		;
9402 
9403 	/* Send INIT signal to other CPU */
9404 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9405 				   id_map[1]);
9406 	/* Signal other CPU we have sent INIT signal */
9407 	vmx_set_test_stage(10);
9408 
9409 	/*
9410 	 * Wait reasonable amount of time for other CPU
9411 	 * to exit VMX operation and process INIT signal
9412 	 */
9413 	delay(INIT_SIGNAL_TEST_DELAY);
9414 	report(!init_signal_test_thread_continued,
9415 	       "INIT signal processed after exit VMX operation");
9416 
9417 	/*
9418 	 * TODO: Send SIPI to other CPU to sipi_entry (See x86/cstart64.S)
9419 	 * to re-init it to kvm-unit-tests standard environment.
9420 	 * Somehow (?) verify that SIPI was indeed received.
9421 	 */
9422 }
9423 
9424 enum vmcs_access {
9425 	ACCESS_VMREAD,
9426 	ACCESS_VMWRITE,
9427 	ACCESS_NONE,
9428 };
9429 
9430 struct vmcs_shadow_test_common {
9431 	enum vmcs_access op;
9432 	enum Reason reason;
9433 	u64 field;
9434 	u64 value;
9435 	u64 flags;
9436 	u64 time;
9437 } l1_l2_common;
9438 
9439 static inline u64 vmread_flags(u64 field, u64 *val)
9440 {
9441 	u64 flags;
9442 
9443 	asm volatile ("vmread %2, %1; pushf; pop %0"
9444 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
9445 	return flags & X86_EFLAGS_ALU;
9446 }
9447 
9448 static inline u64 vmwrite_flags(u64 field, u64 val)
9449 {
9450 	u64 flags;
9451 
9452 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
9453 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
9454 	return flags & X86_EFLAGS_ALU;
9455 }
9456 
9457 static void vmx_vmcs_shadow_test_guest(void)
9458 {
9459 	struct vmcs_shadow_test_common *c = &l1_l2_common;
9460 	u64 start;
9461 
9462 	while (c->op != ACCESS_NONE) {
9463 		start = rdtsc();
9464 		switch (c->op) {
9465 		default:
9466 			c->flags = -1ull;
9467 			break;
9468 		case ACCESS_VMREAD:
9469 			c->flags = vmread_flags(c->field, &c->value);
9470 			break;
9471 		case ACCESS_VMWRITE:
9472 			c->flags = vmwrite_flags(c->field, 0);
9473 			break;
9474 		}
9475 		c->time = rdtsc() - start;
9476 		vmcall();
9477 	}
9478 }
9479 
9480 static u64 vmread_from_shadow(u64 field)
9481 {
9482 	struct vmcs *primary;
9483 	struct vmcs *shadow;
9484 	u64 value;
9485 
9486 	TEST_ASSERT(!vmcs_save(&primary));
9487 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
9488 	TEST_ASSERT(!make_vmcs_current(shadow));
9489 	value = vmcs_read(field);
9490 	TEST_ASSERT(!make_vmcs_current(primary));
9491 	return value;
9492 }
9493 
9494 static u64 vmwrite_to_shadow(u64 field, u64 value)
9495 {
9496 	struct vmcs *primary;
9497 	struct vmcs *shadow;
9498 
9499 	TEST_ASSERT(!vmcs_save(&primary));
9500 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
9501 	TEST_ASSERT(!make_vmcs_current(shadow));
9502 	vmcs_write(field, value);
9503 	value = vmcs_read(field);
9504 	TEST_ASSERT(!make_vmcs_current(primary));
9505 	return value;
9506 }
9507 
9508 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
9509 {
9510 	struct vmcs_shadow_test_common *c = &l1_l2_common;
9511 
9512 	c->op = access;
9513 	vmcs_write(VMX_INST_ERROR, 0);
9514 	enter_guest();
9515 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
9516 	if (c->reason != VMX_VMCALL) {
9517 		skip_exit_insn();
9518 		enter_guest();
9519 	}
9520 	skip_exit_vmcall();
9521 }
9522 
9523 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
9524 {
9525 	struct vmcs_shadow_test_common *c = &l1_l2_common;
9526 	struct vmcs *shadow;
9527 	u64 value;
9528 	uintptr_t flags[2];
9529 	bool good_shadow;
9530 	u32 vmx_inst_error;
9531 
9532 	report_prefix_pushf("field %lx", field);
9533 	c->field = field;
9534 
9535 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
9536 	if (shadow != (struct vmcs *)-1ull) {
9537 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
9538 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
9539 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
9540 	} else {
9541 		/*
9542 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
9543 		 * shadowed-fields should fail with setting RFLAGS.CF.
9544 		 */
9545 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
9546 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
9547 		good_shadow = false;
9548 	}
9549 
9550 	/* Intercept both VMREAD and VMWRITE. */
9551 	report_prefix_push("no VMREAD/VMWRITE permission");
9552 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
9553 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
9554 		set_bit(field, bitmap[ACCESS_VMREAD]);
9555 		set_bit(field, bitmap[ACCESS_VMWRITE]);
9556 	}
9557 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
9558 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
9559 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
9560 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
9561 	report_prefix_pop();
9562 
9563 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
9564 		goto out;
9565 
9566 	/* Permit shadowed VMREAD. */
9567 	report_prefix_push("VMREAD permission only");
9568 	clear_bit(field, bitmap[ACCESS_VMREAD]);
9569 	set_bit(field, bitmap[ACCESS_VMWRITE]);
9570 	if (good_shadow)
9571 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
9572 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
9573 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
9574 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
9575 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
9576 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
9577 	       c->time);
9578 	report(c->flags == flags[ACCESS_VMREAD],
9579 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
9580 	       c->flags, flags[ACCESS_VMREAD]);
9581 	if (good_shadow)
9582 		report(c->value == value,
9583 		       "value read from shadow (%lx) is as expected (%lx)",
9584 		       c->value, value);
9585 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
9586 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
9587 		       "VMX_INST_ERROR (%d) is as expected (%d)",
9588 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
9589 	report_prefix_pop();
9590 
9591 	/* Permit shadowed VMWRITE. */
9592 	report_prefix_push("VMWRITE permission only");
9593 	set_bit(field, bitmap[ACCESS_VMREAD]);
9594 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
9595 	if (good_shadow)
9596 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
9597 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
9598 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
9599 	report(c->reason == VMX_VMCALL,
9600 		"shadowed for VMWRITE (in %ld cycles)",
9601 		c->time);
9602 	report(c->flags == flags[ACCESS_VMREAD],
9603 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
9604 	       c->flags, flags[ACCESS_VMREAD]);
9605 	if (good_shadow) {
9606 		value = vmread_from_shadow(field);
9607 		report(value == 0,
9608 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
9609 		       0ul);
9610 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
9611 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
9612 		       "VMX_INST_ERROR (%d) is as expected (%d)",
9613 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
9614 	}
9615 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
9616 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
9617 	report_prefix_pop();
9618 
9619 	/* Permit shadowed VMREAD and VMWRITE. */
9620 	report_prefix_push("VMREAD and VMWRITE permission");
9621 	clear_bit(field, bitmap[ACCESS_VMREAD]);
9622 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
9623 	if (good_shadow)
9624 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
9625 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
9626 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
9627 	report(c->reason == VMX_VMCALL,
9628 		"shadowed for VMWRITE (in %ld cycles)",
9629 		c->time);
9630 	report(c->flags == flags[ACCESS_VMREAD],
9631 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
9632 	       c->flags, flags[ACCESS_VMREAD]);
9633 	if (good_shadow) {
9634 		value = vmread_from_shadow(field);
9635 		report(value == 0,
9636 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
9637 		       0ul);
9638 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
9639 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
9640 		       "VMX_INST_ERROR (%d) is as expected (%d)",
9641 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
9642 	}
9643 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
9644 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
9645 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
9646 	       c->time);
9647 	report(c->flags == flags[ACCESS_VMREAD],
9648 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
9649 	       c->flags, flags[ACCESS_VMREAD]);
9650 	if (good_shadow)
9651 		report(c->value == 0,
9652 		       "value read from shadow (%lx) is as expected (%lx)",
9653 		       c->value, 0ul);
9654 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
9655 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
9656 		       "VMX_INST_ERROR (%d) is as expected (%d)",
9657 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
9658 	report_prefix_pop();
9659 
9660 out:
9661 	report_prefix_pop();
9662 }
9663 
9664 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
9665 {
9666 	unsigned base;
9667 	unsigned index;
9668 	unsigned bit;
9669 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
9670 
9671 	/* Run test on all possible valid VMCS fields */
9672 	for (base = 0;
9673 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
9674 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
9675 		for (index = 0; index <= highest_index; index++)
9676 			vmcs_shadow_test_field(bitmap, base + index);
9677 
9678 	/*
9679 	 * Run tests on some invalid VMCS fields
9680 	 * (Have reserved bit set).
9681 	 */
9682 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
9683 		vmcs_shadow_test_field(bitmap, (1ull << bit));
9684 }
9685 
9686 static void vmx_vmcs_shadow_test(void)
9687 {
9688 	u8 *bitmap[2];
9689 	struct vmcs *shadow;
9690 
9691 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
9692 		printf("\t'Activate secondary controls' not supported.\n");
9693 		return;
9694 	}
9695 
9696 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
9697 		printf("\t'VMCS shadowing' not supported.\n");
9698 		return;
9699 	}
9700 
9701 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
9702 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
9703 		printf("\tVMWRITE can't modify VM-exit information fields.\n");
9704 		return;
9705 	}
9706 
9707 	test_set_guest(vmx_vmcs_shadow_test_guest);
9708 
9709 	bitmap[ACCESS_VMREAD] = alloc_page();
9710 	bitmap[ACCESS_VMWRITE] = alloc_page();
9711 
9712 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
9713 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
9714 
9715 	shadow = alloc_page();
9716 	shadow->hdr.revision_id = basic.revision;
9717 	shadow->hdr.shadow_vmcs = 1;
9718 	TEST_ASSERT(!vmcs_clear(shadow));
9719 
9720 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
9721 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
9722 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
9723 
9724 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
9725 	report_prefix_push("valid link pointer");
9726 	vmx_vmcs_shadow_test_body(bitmap);
9727 	report_prefix_pop();
9728 
9729 	vmcs_write(VMCS_LINK_PTR, -1ull);
9730 	report_prefix_push("invalid link pointer");
9731 	vmx_vmcs_shadow_test_body(bitmap);
9732 	report_prefix_pop();
9733 
9734 	l1_l2_common.op = ACCESS_NONE;
9735 	enter_guest();
9736 }
9737 
9738 /*
9739  * This test monitors the difference between a guest RDTSC instruction
9740  * and the IA32_TIME_STAMP_COUNTER MSR value stored in the VMCS12
9741  * VM-exit MSR-store list when taking a VM-exit on the instruction
9742  * following RDTSC.
9743  */
9744 #define RDTSC_DIFF_ITERS 100000
9745 #define RDTSC_DIFF_FAILS 100
9746 #define HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD 750
9747 
9748 /*
9749  * Set 'use TSC offsetting' and set the guest offset to the
9750  * inverse of the host's current TSC value, so that the guest starts running
9751  * with an effective TSC value of 0.
9752  */
9753 static void reset_guest_tsc_to_zero(void)
9754 {
9755 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
9756 	vmcs_write(TSC_OFFSET, -rdtsc());
9757 }
9758 
9759 static void rdtsc_vmexit_diff_test_guest(void)
9760 {
9761 	int i;
9762 
9763 	for (i = 0; i < RDTSC_DIFF_ITERS; i++)
9764 		/* Ensure rdtsc is the last instruction before the vmcall. */
9765 		asm volatile("rdtsc; vmcall" : : : "eax", "edx");
9766 }
9767 
9768 /*
9769  * This function only considers the "use TSC offsetting" VM-execution
9770  * control.  It does not handle "use TSC scaling" (because the latter
9771  * isn't available to the host today.)
9772  */
9773 static unsigned long long host_time_to_guest_time(unsigned long long t)
9774 {
9775 	TEST_ASSERT(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
9776 		    !(vmcs_read(CPU_EXEC_CTRL1) & CPU_USE_TSC_SCALING));
9777 
9778 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_USE_TSC_OFFSET)
9779 		t += vmcs_read(TSC_OFFSET);
9780 
9781 	return t;
9782 }
9783 
9784 static unsigned long long rdtsc_vmexit_diff_test_iteration(void)
9785 {
9786 	unsigned long long guest_tsc, host_to_guest_tsc;
9787 
9788 	enter_guest();
9789 	skip_exit_vmcall();
9790 	guest_tsc = (u32) regs.rax + (regs.rdx << 32);
9791 	host_to_guest_tsc = host_time_to_guest_time(exit_msr_store[0].value);
9792 
9793 	return host_to_guest_tsc - guest_tsc;
9794 }
9795 
9796 static void rdtsc_vmexit_diff_test(void)
9797 {
9798 	unsigned long long delta;
9799 	int fail = 0;
9800 	int i;
9801 
9802 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET))
9803 		test_skip("CPU doesn't support the 'use TSC offsetting' processor-based VM-execution control.\n");
9804 
9805 	test_set_guest(rdtsc_vmexit_diff_test_guest);
9806 
9807 	reset_guest_tsc_to_zero();
9808 
9809 	/*
9810 	 * Set up the VMCS12 VM-exit MSR-store list to store just one
9811 	 * MSR: IA32_TIME_STAMP_COUNTER. Note that the value stored is
9812 	 * in the host time domain (i.e., it is not adjusted according
9813 	 * to the TSC multiplier and TSC offset fields in the VMCS12,
9814 	 * as a guest RDTSC would be.)
9815 	 */
9816 	exit_msr_store = alloc_page();
9817 	exit_msr_store[0].index = MSR_IA32_TSC;
9818 	vmcs_write(EXI_MSR_ST_CNT, 1);
9819 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(exit_msr_store));
9820 
9821 	for (i = 0; i < RDTSC_DIFF_ITERS && fail < RDTSC_DIFF_FAILS; i++) {
9822 		delta = rdtsc_vmexit_diff_test_iteration();
9823 		if (delta >= HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD)
9824 			fail++;
9825 	}
9826 
9827 	enter_guest();
9828 
9829 	report(fail < RDTSC_DIFF_FAILS,
9830 	       "RDTSC to VM-exit delta too high in %d of %d iterations, last = %llu",
9831 	       fail, i, delta);
9832 }
9833 
9834 static int invalid_msr_init(struct vmcs *vmcs)
9835 {
9836 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9837 		printf("\tPreemption timer is not supported\n");
9838 		return VMX_TEST_EXIT;
9839 	}
9840 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
9841 	preempt_val = 10000000;
9842 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
9843 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
9844 
9845 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
9846 		printf("\tSave preemption value is not supported\n");
9847 
9848 	vmcs_write(ENT_MSR_LD_CNT, 1);
9849 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
9850 
9851 	return VMX_TEST_START;
9852 }
9853 
9854 
9855 static void invalid_msr_main(void)
9856 {
9857 	report(0, "Invalid MSR load");
9858 }
9859 
9860 static int invalid_msr_exit_handler(union exit_reason exit_reason)
9861 {
9862 	report(0, "Invalid MSR load");
9863 	print_vmexit_info(exit_reason);
9864 	return VMX_TEST_EXIT;
9865 }
9866 
9867 static int invalid_msr_entry_failure(struct vmentry_result *result)
9868 {
9869 	report(result->exit_reason.failed_vmentry &&
9870 	       result->exit_reason.basic == VMX_FAIL_MSR, "Invalid MSR load");
9871 	return VMX_TEST_VMEXIT;
9872 }
9873 
9874 /*
9875  * The max number of MSRs in an atomic switch MSR list is:
9876  * (111B + 1) * 512 = 4096
9877  *
9878  * Each list entry consumes:
9879  * 4-byte MSR index + 4 bytes reserved + 8-byte data = 16 bytes
9880  *
9881  * Allocate 128 kB to cover max_msr_list_size (i.e., 64 kB) and then some.
9882  */
9883 static const u32 msr_list_page_order = 5;
9884 
9885 static void atomic_switch_msr_limit_test_guest(void)
9886 {
9887 	vmcall();
9888 }
9889 
9890 static void populate_msr_list(struct vmx_msr_entry *msr_list,
9891 			      size_t byte_capacity, int count)
9892 {
9893 	int i;
9894 
9895 	for (i = 0; i < count; i++) {
9896 		msr_list[i].index = MSR_IA32_TSC;
9897 		msr_list[i].reserved = 0;
9898 		msr_list[i].value = 0x1234567890abcdef;
9899 	}
9900 
9901 	memset(msr_list + count, 0xff,
9902 	       byte_capacity - count * sizeof(*msr_list));
9903 }
9904 
9905 static int max_msr_list_size(void)
9906 {
9907 	u32 vmx_misc = rdmsr(MSR_IA32_VMX_MISC);
9908 	u32 factor = ((vmx_misc & GENMASK(27, 25)) >> 25) + 1;
9909 
9910 	return factor * 512;
9911 }
9912 
9913 static void atomic_switch_msrs_test(int count)
9914 {
9915 	struct vmx_msr_entry *vm_enter_load;
9916         struct vmx_msr_entry *vm_exit_load;
9917         struct vmx_msr_entry *vm_exit_store;
9918 	int max_allowed = max_msr_list_size();
9919 	int byte_capacity = 1ul << (msr_list_page_order + PAGE_SHIFT);
9920 	/* Exceeding the max MSR list size at exit trigers KVM to abort. */
9921 	int exit_count = count > max_allowed ? max_allowed : count;
9922 	int cleanup_count = count > max_allowed ? 2 : 1;
9923 	int i;
9924 
9925 	/*
9926 	 * Check for the IA32_TSC MSR,
9927 	 * available with the "TSC flag" and used to populate the MSR lists.
9928 	 */
9929 	if (!(cpuid(1).d & (1 << 4))) {
9930 		report_skip(__func__);
9931 		return;
9932 	}
9933 
9934 	/* Set L2 guest. */
9935 	test_set_guest(atomic_switch_msr_limit_test_guest);
9936 
9937 	/* Setup atomic MSR switch lists. */
9938 	vm_enter_load = alloc_pages(msr_list_page_order);
9939 	vm_exit_load = alloc_pages(msr_list_page_order);
9940 	vm_exit_store = alloc_pages(msr_list_page_order);
9941 
9942 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)vm_enter_load);
9943 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)vm_exit_load);
9944 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)vm_exit_store);
9945 
9946 	/*
9947 	 * VM-Enter should succeed up to the max number of MSRs per list, and
9948 	 * should not consume junk beyond the last entry.
9949 	 */
9950 	populate_msr_list(vm_enter_load, byte_capacity, count);
9951 	populate_msr_list(vm_exit_load, byte_capacity, exit_count);
9952 	populate_msr_list(vm_exit_store, byte_capacity, exit_count);
9953 
9954 	vmcs_write(ENT_MSR_LD_CNT, count);
9955 	vmcs_write(EXI_MSR_LD_CNT, exit_count);
9956 	vmcs_write(EXI_MSR_ST_CNT, exit_count);
9957 
9958 	if (count <= max_allowed) {
9959 		enter_guest();
9960 		assert_exit_reason(VMX_VMCALL);
9961 		skip_exit_vmcall();
9962 	} else {
9963 		u32 exit_qual;
9964 
9965 		test_guest_state("Invalid MSR Load Count", true, count,
9966 				 "ENT_MSR_LD_CNT");
9967 
9968 		exit_qual = vmcs_read(EXI_QUALIFICATION);
9969 		report(exit_qual == max_allowed + 1, "exit_qual, %u, is %u.",
9970 		       exit_qual, max_allowed + 1);
9971 	}
9972 
9973 	/* Cleanup. */
9974 	vmcs_write(ENT_MSR_LD_CNT, 0);
9975 	vmcs_write(EXI_MSR_LD_CNT, 0);
9976 	vmcs_write(EXI_MSR_ST_CNT, 0);
9977 	for (i = 0; i < cleanup_count; i++) {
9978 		enter_guest();
9979 		skip_exit_vmcall();
9980 	}
9981 	free_pages_by_order(vm_enter_load, msr_list_page_order);
9982 	free_pages_by_order(vm_exit_load, msr_list_page_order);
9983 	free_pages_by_order(vm_exit_store, msr_list_page_order);
9984 }
9985 
9986 static void atomic_switch_max_msrs_test(void)
9987 {
9988 	atomic_switch_msrs_test(max_msr_list_size());
9989 }
9990 
9991 static void atomic_switch_overflow_msrs_test(void)
9992 {
9993 	if (test_device_enabled())
9994 		atomic_switch_msrs_test(max_msr_list_size() + 1);
9995 	else
9996 		test_skip("Test is only supported on KVM");
9997 }
9998 
9999 #define TEST(name) { #name, .v2 = name }
10000 
10001 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
10002 struct vmx_test vmx_tests[] = {
10003 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
10004 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
10005 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
10006 		preemption_timer_exit_handler, NULL, {0} },
10007 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
10008 		test_ctrl_pat_exit_handler, NULL, {0} },
10009 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
10010 		test_ctrl_efer_exit_handler, NULL, {0} },
10011 	{ "CR shadowing", NULL, cr_shadowing_main,
10012 		cr_shadowing_exit_handler, NULL, {0} },
10013 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
10014 		NULL, {0} },
10015 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
10016 		insn_intercept_exit_handler, NULL, {0} },
10017 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
10018 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
10019 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
10020 	{ "VPID", vpid_init, vpid_main, vpid_exit_handler, NULL, {0} },
10021 	{ "interrupt", interrupt_init, interrupt_main,
10022 		interrupt_exit_handler, NULL, {0} },
10023 	{ "nmi_hlt", nmi_hlt_init, nmi_hlt_main,
10024 		nmi_hlt_exit_handler, NULL, {0} },
10025 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
10026 		NULL, {0} },
10027 	{ "MSR switch", msr_switch_init, msr_switch_main,
10028 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
10029 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
10030 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
10031 		disable_rdtscp_exit_handler, NULL, {0} },
10032 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
10033 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
10034 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
10035 		exit_monitor_from_l2_handler, NULL, {0} },
10036 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
10037 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
10038 	/* Basic V2 tests. */
10039 	TEST(v2_null_test),
10040 	TEST(v2_multiple_entries_test),
10041 	TEST(fixture_test_case1),
10042 	TEST(fixture_test_case2),
10043 	/* Opcode tests. */
10044 	TEST(invvpid_test_v2),
10045 	/* VM-entry tests */
10046 	TEST(vmx_controls_test),
10047 	TEST(vmx_host_state_area_test),
10048 	TEST(vmx_guest_state_area_test),
10049 	TEST(vmentry_movss_shadow_test),
10050 	/* APICv tests */
10051 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
10052 	TEST(vmx_hlt_with_rvi_test),
10053 	TEST(apic_reg_virt_test),
10054 	TEST(virt_x2apic_mode_test),
10055 	/* APIC pass-through tests */
10056 	TEST(vmx_apic_passthrough_test),
10057 	TEST(vmx_apic_passthrough_thread_test),
10058 	TEST(vmx_apic_passthrough_tpr_threshold_test),
10059 	TEST(vmx_init_signal_test),
10060 	/* VMCS Shadowing tests */
10061 	TEST(vmx_vmcs_shadow_test),
10062 	/* Regression tests */
10063 	TEST(vmx_cr_load_test),
10064 	TEST(vmx_nm_test),
10065 	TEST(vmx_db_test),
10066 	TEST(vmx_nmi_window_test),
10067 	TEST(vmx_intr_window_test),
10068 	TEST(vmx_pending_event_test),
10069 	TEST(vmx_pending_event_hlt_test),
10070 	TEST(vmx_store_tsc_test),
10071 	TEST(vmx_preemption_timer_zero_test),
10072 	TEST(vmx_preemption_timer_tf_test),
10073 	TEST(vmx_preemption_timer_expiry_test),
10074 	/* EPT access tests. */
10075 	TEST(ept_access_test_not_present),
10076 	TEST(ept_access_test_read_only),
10077 	TEST(ept_access_test_write_only),
10078 	TEST(ept_access_test_read_write),
10079 	TEST(ept_access_test_execute_only),
10080 	TEST(ept_access_test_read_execute),
10081 	TEST(ept_access_test_write_execute),
10082 	TEST(ept_access_test_read_write_execute),
10083 	TEST(ept_access_test_reserved_bits),
10084 	TEST(ept_access_test_ignored_bits),
10085 	TEST(ept_access_test_paddr_not_present_ad_disabled),
10086 	TEST(ept_access_test_paddr_not_present_ad_enabled),
10087 	TEST(ept_access_test_paddr_read_only_ad_disabled),
10088 	TEST(ept_access_test_paddr_read_only_ad_enabled),
10089 	TEST(ept_access_test_paddr_read_write),
10090 	TEST(ept_access_test_paddr_read_write_execute),
10091 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
10092 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
10093 	TEST(ept_access_test_paddr_not_present_page_fault),
10094 	TEST(ept_access_test_force_2m_page),
10095 	/* Atomic MSR switch tests. */
10096 	TEST(atomic_switch_max_msrs_test),
10097 	TEST(atomic_switch_overflow_msrs_test),
10098 	TEST(rdtsc_vmexit_diff_test),
10099 	TEST(vmx_mtf_test),
10100 	{ NULL, NULL, NULL, NULL, NULL, {0} },
10101 };
10102