xref: /kvm-unit-tests/x86/vmx_tests.c (revision e282100a1f0e5404425acf3fa62d67648fb2c664)
1 /*
2  * All test cases of nested virtualization should be in this file
3  *
4  * Author : Arthur Chunqi Li <yzt356@gmail.com>
5  */
6 #include "vmx.h"
7 #include "msr.h"
8 #include "processor.h"
9 #include "vm.h"
10 #include "pci.h"
11 #include "fwcfg.h"
12 #include "isr.h"
13 #include "desc.h"
14 #include "apic.h"
15 #include "types.h"
16 #include "vmalloc.h"
17 #include "alloc_page.h"
18 #include "smp.h"
19 #include "delay.h"
20 
21 #define NONCANONICAL            0xaaaaaaaaaaaaaaaaull
22 
23 #define VPID_CAP_INVVPID_TYPES_SHIFT 40
24 
25 u64 ia32_pat;
26 u64 ia32_efer;
27 void *io_bitmap_a, *io_bitmap_b;
28 u16 ioport;
29 
30 unsigned long *pml4;
31 u64 eptp;
32 void *data_page1, *data_page2;
33 
34 phys_addr_t pci_physaddr;
35 
36 void *pml_log;
37 #define PML_INDEX 512
38 
39 static inline unsigned ffs(unsigned x)
40 {
41 	int pos = -1;
42 
43 	__asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0"
44 			     : "+r"(pos) : "rm"(x) : "eax");
45 	return pos + 1;
46 }
47 
48 static inline void vmcall()
49 {
50 	asm volatile("vmcall");
51 }
52 
53 void basic_guest_main()
54 {
55 	report("Basic VMX test", 1);
56 }
57 
58 int basic_exit_handler()
59 {
60 	report("Basic VMX test", 0);
61 	print_vmexit_info();
62 	return VMX_TEST_EXIT;
63 }
64 
65 void vmenter_main()
66 {
67 	u64 rax;
68 	u64 rsp, resume_rsp;
69 
70 	report("test vmlaunch", 1);
71 
72 	asm volatile(
73 		"mov %%rsp, %0\n\t"
74 		"mov %3, %%rax\n\t"
75 		"vmcall\n\t"
76 		"mov %%rax, %1\n\t"
77 		"mov %%rsp, %2\n\t"
78 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
79 		: "g"(0xABCD));
80 	report("test vmresume", (rax == 0xFFFF) && (rsp == resume_rsp));
81 }
82 
83 int vmenter_exit_handler()
84 {
85 	u64 guest_rip;
86 	ulong reason;
87 
88 	guest_rip = vmcs_read(GUEST_RIP);
89 	reason = vmcs_read(EXI_REASON) & 0xff;
90 	switch (reason) {
91 	case VMX_VMCALL:
92 		if (regs.rax != 0xABCD) {
93 			report("test vmresume", 0);
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("test vmresume", 0);
101 		print_vmexit_info();
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 int preemption_timer_init()
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 void preemption_timer_main()
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 int preemption_timer_exit_handler()
153 {
154 	bool guest_halted;
155 	u64 guest_rip;
156 	ulong reason;
157 	u32 insn_len;
158 	u32 ctrl_exit;
159 
160 	guest_rip = vmcs_read(GUEST_RIP);
161 	reason = vmcs_read(EXI_REASON) & 0xff;
162 	insn_len = vmcs_read(EXI_INST_LEN);
163 	switch (reason) {
164 	case VMX_PREEMPT:
165 		switch (vmx_get_test_stage()) {
166 		case 1:
167 		case 2:
168 			report("busy-wait for preemption timer",
169 			       ((rdtsc() - tsc_val) >> preempt_scale) >=
170 			       preempt_val);
171 			vmx_set_test_stage(3);
172 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
173 			return VMX_TEST_RESUME;
174 		case 3:
175 			guest_halted =
176 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
177 			report("preemption timer during hlt",
178 			       ((rdtsc() - tsc_val) >> preempt_scale) >=
179 			       preempt_val && guest_halted);
180 			vmx_set_test_stage(4);
181 			vmcs_write(PIN_CONTROLS,
182 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
183 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
184 			return VMX_TEST_RESUME;
185 		case 4:
186 			report("preemption timer with 0 value",
187 			       saved_rip == guest_rip);
188 			break;
189 		default:
190 			report("Invalid stage.", false);
191 			print_vmexit_info();
192 			break;
193 		}
194 		break;
195 	case VMX_VMCALL:
196 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
197 		switch (vmx_get_test_stage()) {
198 		case 0:
199 			report("Keep preemption value",
200 			       vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val);
201 			vmx_set_test_stage(1);
202 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
203 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
204 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
205 			vmcs_write(EXI_CONTROLS, ctrl_exit);
206 			return VMX_TEST_RESUME;
207 		case 1:
208 			report("Save preemption value",
209 			       vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val);
210 			return VMX_TEST_RESUME;
211 		case 2:
212 			report("busy-wait for preemption timer", 0);
213 			vmx_set_test_stage(3);
214 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
215 			return VMX_TEST_RESUME;
216 		case 3:
217 			report("preemption timer during hlt", 0);
218 			vmx_set_test_stage(4);
219 			/* fall through */
220 		case 4:
221 			vmcs_write(PIN_CONTROLS,
222 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
223 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
224 			saved_rip = guest_rip + insn_len;
225 			return VMX_TEST_RESUME;
226 		case 5:
227 			report("preemption timer with 0 value (vmcall stage 5)", 0);
228 			break;
229 		default:
230 			// Should not reach here
231 			report("unexpected stage, %d", false,
232 			       vmx_get_test_stage());
233 			print_vmexit_info();
234 			return VMX_TEST_VMEXIT;
235 		}
236 		break;
237 	default:
238 		report("Unknown exit reason, %ld", false, reason);
239 		print_vmexit_info();
240 	}
241 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
242 	return VMX_TEST_VMEXIT;
243 }
244 
245 void msr_bmp_init()
246 {
247 	void *msr_bitmap;
248 	u32 ctrl_cpu0;
249 
250 	msr_bitmap = alloc_page();
251 	memset(msr_bitmap, 0x0, PAGE_SIZE);
252 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
253 	ctrl_cpu0 |= CPU_MSR_BITMAP;
254 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
255 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
256 }
257 
258 static int test_ctrl_pat_init()
259 {
260 	u64 ctrl_ent;
261 	u64 ctrl_exi;
262 
263 	msr_bmp_init();
264 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
265 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
266 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
267 		printf("\tSave/load PAT is not supported\n");
268 		return 1;
269 	}
270 
271 	ctrl_ent = vmcs_read(ENT_CONTROLS);
272 	ctrl_exi = vmcs_read(EXI_CONTROLS);
273 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
274 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
275 	vmcs_write(ENT_CONTROLS, ctrl_ent);
276 	vmcs_write(EXI_CONTROLS, ctrl_exi);
277 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
278 	vmcs_write(GUEST_PAT, 0x0);
279 	vmcs_write(HOST_PAT, ia32_pat);
280 	return VMX_TEST_START;
281 }
282 
283 static void test_ctrl_pat_main()
284 {
285 	u64 guest_ia32_pat;
286 
287 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
288 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
289 		printf("\tENT_LOAD_PAT is not supported.\n");
290 	else {
291 		if (guest_ia32_pat != 0) {
292 			report("Entry load PAT", 0);
293 			return;
294 		}
295 	}
296 	wrmsr(MSR_IA32_CR_PAT, 0x6);
297 	vmcall();
298 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
299 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
300 		report("Entry load PAT", guest_ia32_pat == ia32_pat);
301 }
302 
303 static int test_ctrl_pat_exit_handler()
304 {
305 	u64 guest_rip;
306 	ulong reason;
307 	u64 guest_pat;
308 
309 	guest_rip = vmcs_read(GUEST_RIP);
310 	reason = vmcs_read(EXI_REASON) & 0xff;
311 	switch (reason) {
312 	case VMX_VMCALL:
313 		guest_pat = vmcs_read(GUEST_PAT);
314 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
315 			printf("\tEXI_SAVE_PAT is not supported\n");
316 			vmcs_write(GUEST_PAT, 0x6);
317 		} else {
318 			report("Exit save PAT", guest_pat == 0x6);
319 		}
320 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
321 			printf("\tEXI_LOAD_PAT is not supported\n");
322 		else
323 			report("Exit load PAT", rdmsr(MSR_IA32_CR_PAT) == ia32_pat);
324 		vmcs_write(GUEST_PAT, ia32_pat);
325 		vmcs_write(GUEST_RIP, guest_rip + 3);
326 		return VMX_TEST_RESUME;
327 	default:
328 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
329 		break;
330 	}
331 	return VMX_TEST_VMEXIT;
332 }
333 
334 static int test_ctrl_efer_init()
335 {
336 	u64 ctrl_ent;
337 	u64 ctrl_exi;
338 
339 	msr_bmp_init();
340 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
341 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
342 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
343 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
344 	ia32_efer = rdmsr(MSR_EFER);
345 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
346 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
347 	return VMX_TEST_START;
348 }
349 
350 static void test_ctrl_efer_main()
351 {
352 	u64 guest_ia32_efer;
353 
354 	guest_ia32_efer = rdmsr(MSR_EFER);
355 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
356 		printf("\tENT_LOAD_EFER is not supported.\n");
357 	else {
358 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
359 			report("Entry load EFER", 0);
360 			return;
361 		}
362 	}
363 	wrmsr(MSR_EFER, ia32_efer);
364 	vmcall();
365 	guest_ia32_efer = rdmsr(MSR_EFER);
366 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
367 		report("Entry load EFER", guest_ia32_efer == ia32_efer);
368 }
369 
370 static int test_ctrl_efer_exit_handler()
371 {
372 	u64 guest_rip;
373 	ulong reason;
374 	u64 guest_efer;
375 
376 	guest_rip = vmcs_read(GUEST_RIP);
377 	reason = vmcs_read(EXI_REASON) & 0xff;
378 	switch (reason) {
379 	case VMX_VMCALL:
380 		guest_efer = vmcs_read(GUEST_EFER);
381 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
382 			printf("\tEXI_SAVE_EFER is not supported\n");
383 			vmcs_write(GUEST_EFER, ia32_efer);
384 		} else {
385 			report("Exit save EFER", guest_efer == ia32_efer);
386 		}
387 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
388 			printf("\tEXI_LOAD_EFER is not supported\n");
389 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
390 		} else {
391 			report("Exit load EFER", rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX));
392 		}
393 		vmcs_write(GUEST_PAT, ia32_efer);
394 		vmcs_write(GUEST_RIP, guest_rip + 3);
395 		return VMX_TEST_RESUME;
396 	default:
397 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
398 		break;
399 	}
400 	return VMX_TEST_VMEXIT;
401 }
402 
403 u32 guest_cr0, guest_cr4;
404 
405 static void cr_shadowing_main()
406 {
407 	u32 cr0, cr4, tmp;
408 
409 	// Test read through
410 	vmx_set_test_stage(0);
411 	guest_cr0 = read_cr0();
412 	if (vmx_get_test_stage() == 1)
413 		report("Read through CR0", 0);
414 	else
415 		vmcall();
416 	vmx_set_test_stage(1);
417 	guest_cr4 = read_cr4();
418 	if (vmx_get_test_stage() == 2)
419 		report("Read through CR4", 0);
420 	else
421 		vmcall();
422 	// Test write through
423 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
424 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
425 	vmx_set_test_stage(2);
426 	write_cr0(guest_cr0);
427 	if (vmx_get_test_stage() == 3)
428 		report("Write throuth CR0", 0);
429 	else
430 		vmcall();
431 	vmx_set_test_stage(3);
432 	write_cr4(guest_cr4);
433 	if (vmx_get_test_stage() == 4)
434 		report("Write through CR4", 0);
435 	else
436 		vmcall();
437 	// Test read shadow
438 	vmx_set_test_stage(4);
439 	vmcall();
440 	cr0 = read_cr0();
441 	if (vmx_get_test_stage() != 5)
442 		report("Read shadowing CR0", cr0 == guest_cr0);
443 	vmx_set_test_stage(5);
444 	cr4 = read_cr4();
445 	if (vmx_get_test_stage() != 6)
446 		report("Read shadowing CR4", cr4 == guest_cr4);
447 	// Test write shadow (same value with shadow)
448 	vmx_set_test_stage(6);
449 	write_cr0(guest_cr0);
450 	if (vmx_get_test_stage() == 7)
451 		report("Write shadowing CR0 (same value with shadow)", 0);
452 	else
453 		vmcall();
454 	vmx_set_test_stage(7);
455 	write_cr4(guest_cr4);
456 	if (vmx_get_test_stage() == 8)
457 		report("Write shadowing CR4 (same value with shadow)", 0);
458 	else
459 		vmcall();
460 	// Test write shadow (different value)
461 	vmx_set_test_stage(8);
462 	tmp = guest_cr0 ^ X86_CR0_TS;
463 	asm volatile("mov %0, %%rsi\n\t"
464 		"mov %%rsi, %%cr0\n\t"
465 		::"m"(tmp)
466 		:"rsi", "memory", "cc");
467 	report("Write shadowing different X86_CR0_TS", vmx_get_test_stage() == 9);
468 	vmx_set_test_stage(9);
469 	tmp = guest_cr0 ^ X86_CR0_MP;
470 	asm volatile("mov %0, %%rsi\n\t"
471 		"mov %%rsi, %%cr0\n\t"
472 		::"m"(tmp)
473 		:"rsi", "memory", "cc");
474 	report("Write shadowing different X86_CR0_MP", vmx_get_test_stage() == 10);
475 	vmx_set_test_stage(10);
476 	tmp = guest_cr4 ^ X86_CR4_TSD;
477 	asm volatile("mov %0, %%rsi\n\t"
478 		"mov %%rsi, %%cr4\n\t"
479 		::"m"(tmp)
480 		:"rsi", "memory", "cc");
481 	report("Write shadowing different X86_CR4_TSD", vmx_get_test_stage() == 11);
482 	vmx_set_test_stage(11);
483 	tmp = guest_cr4 ^ X86_CR4_DE;
484 	asm volatile("mov %0, %%rsi\n\t"
485 		"mov %%rsi, %%cr4\n\t"
486 		::"m"(tmp)
487 		:"rsi", "memory", "cc");
488 	report("Write shadowing different X86_CR4_DE", vmx_get_test_stage() == 12);
489 }
490 
491 static int cr_shadowing_exit_handler()
492 {
493 	u64 guest_rip;
494 	ulong reason;
495 	u32 insn_len;
496 	u32 exit_qual;
497 
498 	guest_rip = vmcs_read(GUEST_RIP);
499 	reason = vmcs_read(EXI_REASON) & 0xff;
500 	insn_len = vmcs_read(EXI_INST_LEN);
501 	exit_qual = vmcs_read(EXI_QUALIFICATION);
502 	switch (reason) {
503 	case VMX_VMCALL:
504 		switch (vmx_get_test_stage()) {
505 		case 0:
506 			report("Read through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
507 			break;
508 		case 1:
509 			report("Read through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
510 			break;
511 		case 2:
512 			report("Write through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
513 			break;
514 		case 3:
515 			report("Write through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
516 			break;
517 		case 4:
518 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
519 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
520 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
521 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
522 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
523 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
524 			break;
525 		case 6:
526 			report("Write shadowing CR0 (same value)",
527 					guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)));
528 			break;
529 		case 7:
530 			report("Write shadowing CR4 (same value)",
531 					guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)));
532 			break;
533 		default:
534 			// Should not reach here
535 			report("unexpected stage, %d", false,
536 			       vmx_get_test_stage());
537 			print_vmexit_info();
538 			return VMX_TEST_VMEXIT;
539 		}
540 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
541 		return VMX_TEST_RESUME;
542 	case VMX_CR:
543 		switch (vmx_get_test_stage()) {
544 		case 4:
545 			report("Read shadowing CR0", 0);
546 			vmx_inc_test_stage();
547 			break;
548 		case 5:
549 			report("Read shadowing CR4", 0);
550 			vmx_inc_test_stage();
551 			break;
552 		case 6:
553 			report("Write shadowing CR0 (same value)", 0);
554 			vmx_inc_test_stage();
555 			break;
556 		case 7:
557 			report("Write shadowing CR4 (same value)", 0);
558 			vmx_inc_test_stage();
559 			break;
560 		case 8:
561 		case 9:
562 			// 0x600 encodes "mov %esi, %cr0"
563 			if (exit_qual == 0x600)
564 				vmx_inc_test_stage();
565 			break;
566 		case 10:
567 		case 11:
568 			// 0x604 encodes "mov %esi, %cr4"
569 			if (exit_qual == 0x604)
570 				vmx_inc_test_stage();
571 			break;
572 		default:
573 			// Should not reach here
574 			report("unexpected stage, %d", false,
575 			       vmx_get_test_stage());
576 			print_vmexit_info();
577 			return VMX_TEST_VMEXIT;
578 		}
579 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
580 		return VMX_TEST_RESUME;
581 	default:
582 		report("Unknown exit reason, %ld", false, reason);
583 		print_vmexit_info();
584 	}
585 	return VMX_TEST_VMEXIT;
586 }
587 
588 static int iobmp_init()
589 {
590 	u32 ctrl_cpu0;
591 
592 	io_bitmap_a = alloc_page();
593 	io_bitmap_b = alloc_page();
594 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
595 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
596 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
597 	ctrl_cpu0 |= CPU_IO_BITMAP;
598 	ctrl_cpu0 &= (~CPU_IO);
599 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
600 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
601 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
602 	return VMX_TEST_START;
603 }
604 
605 static void iobmp_main()
606 {
607 	// stage 0, test IO pass
608 	vmx_set_test_stage(0);
609 	inb(0x5000);
610 	outb(0x0, 0x5000);
611 	report("I/O bitmap - I/O pass", vmx_get_test_stage() == 0);
612 	// test IO width, in/out
613 	((u8 *)io_bitmap_a)[0] = 0xFF;
614 	vmx_set_test_stage(2);
615 	inb(0x0);
616 	report("I/O bitmap - trap in", vmx_get_test_stage() == 3);
617 	vmx_set_test_stage(3);
618 	outw(0x0, 0x0);
619 	report("I/O bitmap - trap out", vmx_get_test_stage() == 4);
620 	vmx_set_test_stage(4);
621 	inl(0x0);
622 	report("I/O bitmap - I/O width, long", vmx_get_test_stage() == 5);
623 	// test low/high IO port
624 	vmx_set_test_stage(5);
625 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
626 	inb(0x5000);
627 	report("I/O bitmap - I/O port, low part", vmx_get_test_stage() == 6);
628 	vmx_set_test_stage(6);
629 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
630 	inb(0x9000);
631 	report("I/O bitmap - I/O port, high part", vmx_get_test_stage() == 7);
632 	// test partial pass
633 	vmx_set_test_stage(7);
634 	inl(0x4FFF);
635 	report("I/O bitmap - partial pass", vmx_get_test_stage() == 8);
636 	// test overrun
637 	vmx_set_test_stage(8);
638 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
639 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
640 	inl(0xFFFF);
641 	report("I/O bitmap - overrun", vmx_get_test_stage() == 9);
642 	vmx_set_test_stage(9);
643 	vmcall();
644 	outb(0x0, 0x0);
645 	report("I/O bitmap - ignore unconditional exiting",
646 	       vmx_get_test_stage() == 9);
647 	vmx_set_test_stage(10);
648 	vmcall();
649 	outb(0x0, 0x0);
650 	report("I/O bitmap - unconditional exiting",
651 	       vmx_get_test_stage() == 11);
652 }
653 
654 static int iobmp_exit_handler()
655 {
656 	u64 guest_rip;
657 	ulong reason, exit_qual;
658 	u32 insn_len, ctrl_cpu0;
659 
660 	guest_rip = vmcs_read(GUEST_RIP);
661 	reason = vmcs_read(EXI_REASON) & 0xff;
662 	exit_qual = vmcs_read(EXI_QUALIFICATION);
663 	insn_len = vmcs_read(EXI_INST_LEN);
664 	switch (reason) {
665 	case VMX_IO:
666 		switch (vmx_get_test_stage()) {
667 		case 0:
668 		case 1:
669 			vmx_inc_test_stage();
670 			break;
671 		case 2:
672 			report("I/O bitmap - I/O width, byte",
673 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE);
674 			report("I/O bitmap - I/O direction, in", exit_qual & VMX_IO_IN);
675 			vmx_inc_test_stage();
676 			break;
677 		case 3:
678 			report("I/O bitmap - I/O width, word",
679 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD);
680 			report("I/O bitmap - I/O direction, out",
681 					!(exit_qual & VMX_IO_IN));
682 			vmx_inc_test_stage();
683 			break;
684 		case 4:
685 			report("I/O bitmap - I/O width, long",
686 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG);
687 			vmx_inc_test_stage();
688 			break;
689 		case 5:
690 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
691 				vmx_inc_test_stage();
692 			break;
693 		case 6:
694 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
695 				vmx_inc_test_stage();
696 			break;
697 		case 7:
698 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
699 				vmx_inc_test_stage();
700 			break;
701 		case 8:
702 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
703 				vmx_inc_test_stage();
704 			break;
705 		case 9:
706 		case 10:
707 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
708 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
709 			vmx_inc_test_stage();
710 			break;
711 		default:
712 			// Should not reach here
713 			report("unexpected stage, %d", false,
714 			       vmx_get_test_stage());
715 			print_vmexit_info();
716 			return VMX_TEST_VMEXIT;
717 		}
718 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
719 		return VMX_TEST_RESUME;
720 	case VMX_VMCALL:
721 		switch (vmx_get_test_stage()) {
722 		case 9:
723 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
724 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
725 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
726 			break;
727 		case 10:
728 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
729 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
730 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
731 			break;
732 		default:
733 			// Should not reach here
734 			report("unexpected stage, %d", false,
735 			       vmx_get_test_stage());
736 			print_vmexit_info();
737 			return VMX_TEST_VMEXIT;
738 		}
739 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
740 		return VMX_TEST_RESUME;
741 	default:
742 		printf("guest_rip = %#lx\n", guest_rip);
743 		printf("\tERROR : Undefined exit reason, reason = %ld.\n", reason);
744 		break;
745 	}
746 	return VMX_TEST_VMEXIT;
747 }
748 
749 #define INSN_CPU0		0
750 #define INSN_CPU1		1
751 #define INSN_ALWAYS_TRAP	2
752 
753 #define FIELD_EXIT_QUAL		(1 << 0)
754 #define FIELD_INSN_INFO		(1 << 1)
755 
756 asm(
757 	"insn_hlt: hlt;ret\n\t"
758 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
759 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
760 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
761 	"insn_rdtsc: rdtsc;ret\n\t"
762 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
763 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
764 #ifdef __x86_64__
765 	"insn_cr8_load: mov %rax,%cr8;ret\n\t"
766 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
767 #endif
768 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
769 	"insn_pause: pause;ret\n\t"
770 	"insn_wbinvd: wbinvd;ret\n\t"
771 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
772 	"insn_invd: invd;ret\n\t"
773 	"insn_sgdt: sgdt gdt64_desc;ret\n\t"
774 	"insn_lgdt: lgdt gdt64_desc;ret\n\t"
775 	"insn_sidt: sidt idt_descr;ret\n\t"
776 	"insn_lidt: lidt idt_descr;ret\n\t"
777 	"insn_sldt: sldt %ax;ret\n\t"
778 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
779 	"insn_str: str %ax;ret\n\t"
780 	"insn_rdrand: rdrand %rax;ret\n\t"
781 	"insn_rdseed: rdseed %rax;ret\n\t"
782 );
783 extern void insn_hlt();
784 extern void insn_invlpg();
785 extern void insn_mwait();
786 extern void insn_rdpmc();
787 extern void insn_rdtsc();
788 extern void insn_cr3_load();
789 extern void insn_cr3_store();
790 #ifdef __x86_64__
791 extern void insn_cr8_load();
792 extern void insn_cr8_store();
793 #endif
794 extern void insn_monitor();
795 extern void insn_pause();
796 extern void insn_wbinvd();
797 extern void insn_sgdt();
798 extern void insn_lgdt();
799 extern void insn_sidt();
800 extern void insn_lidt();
801 extern void insn_sldt();
802 extern void insn_lldt();
803 extern void insn_str();
804 extern void insn_cpuid();
805 extern void insn_invd();
806 extern void insn_rdrand();
807 extern void insn_rdseed();
808 
809 u32 cur_insn;
810 u64 cr3;
811 
812 struct insn_table {
813 	const char *name;
814 	u32 flag;
815 	void (*insn_func)();
816 	u32 type;
817 	u32 reason;
818 	ulong exit_qual;
819 	u32 insn_info;
820 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
821 	// which field need to be tested, reason is always tested
822 	u32 test_field;
823 };
824 
825 /*
826  * Add more test cases of instruction intercept here. Elements in this
827  * table is:
828  *	name/control flag/insn function/type/exit reason/exit qulification/
829  *	instruction info/field to test
830  * The last field defines which fields (exit_qual and insn_info) need to be
831  * tested in exit handler. If set to 0, only "reason" is checked.
832  */
833 static struct insn_table insn_table[] = {
834 	// Flags for Primary Processor-Based VM-Execution Controls
835 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
836 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
837 		0x12345678, 0, FIELD_EXIT_QUAL},
838 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0},
839 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0},
840 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
841 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
842 		FIELD_EXIT_QUAL},
843 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
844 		FIELD_EXIT_QUAL},
845 #ifdef __x86_64__
846 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
847 		FIELD_EXIT_QUAL},
848 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
849 		FIELD_EXIT_QUAL},
850 #endif
851 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0},
852 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
853 	// Flags for Secondary Processor-Based VM-Execution Controls
854 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
855 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
856 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
857 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
858 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
859 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
860 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
861 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
862 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
863 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
864 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
865 	// Instructions always trap
866 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
867 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
868 	// Instructions never trap
869 	{NULL},
870 };
871 
872 static int insn_intercept_init()
873 {
874 	u32 ctrl_cpu;
875 
876 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
877 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
878 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
879 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
880 	cr3 = read_cr3();
881 	return VMX_TEST_START;
882 }
883 
884 static void insn_intercept_main()
885 {
886 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
887 		vmx_set_test_stage(cur_insn * 2);
888 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
889 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
890 		    (insn_table[cur_insn].type == INSN_CPU1 &&
891 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
892 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
893 			       insn_table[cur_insn].type - INSN_CPU0,
894 			       insn_table[cur_insn].name);
895 			continue;
896 		}
897 
898 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
899 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
900 		    (insn_table[cur_insn].type == INSN_CPU1 &&
901 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
902 			/* skip hlt, it stalls the guest and is tested below */
903 			if (insn_table[cur_insn].insn_func != insn_hlt)
904 				insn_table[cur_insn].insn_func();
905 			report("execute %s", vmx_get_test_stage() == cur_insn * 2,
906 					insn_table[cur_insn].name);
907 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
908 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
909 			       insn_table[cur_insn].type - INSN_CPU0,
910 			       insn_table[cur_insn].name);
911 
912 		vmcall();
913 
914 		insn_table[cur_insn].insn_func();
915 		report("intercept %s", vmx_get_test_stage() == cur_insn * 2 + 1,
916 				insn_table[cur_insn].name);
917 
918 		vmx_set_test_stage(cur_insn * 2 + 1);
919 		vmcall();
920 	}
921 }
922 
923 static int insn_intercept_exit_handler()
924 {
925 	u64 guest_rip;
926 	u32 reason;
927 	ulong exit_qual;
928 	u32 insn_len;
929 	u32 insn_info;
930 	bool pass;
931 
932 	guest_rip = vmcs_read(GUEST_RIP);
933 	reason = vmcs_read(EXI_REASON) & 0xff;
934 	exit_qual = vmcs_read(EXI_QUALIFICATION);
935 	insn_len = vmcs_read(EXI_INST_LEN);
936 	insn_info = vmcs_read(EXI_INST_INFO);
937 
938 	if (reason == VMX_VMCALL) {
939 		u32 val = 0;
940 
941 		if (insn_table[cur_insn].type == INSN_CPU0)
942 			val = vmcs_read(CPU_EXEC_CTRL0);
943 		else if (insn_table[cur_insn].type == INSN_CPU1)
944 			val = vmcs_read(CPU_EXEC_CTRL1);
945 
946 		if (vmx_get_test_stage() & 1)
947 			val &= ~insn_table[cur_insn].flag;
948 		else
949 			val |= insn_table[cur_insn].flag;
950 
951 		if (insn_table[cur_insn].type == INSN_CPU0)
952 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
953 		else if (insn_table[cur_insn].type == INSN_CPU1)
954 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
955 	} else {
956 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
957 			insn_table[cur_insn].reason == reason;
958 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
959 		    insn_table[cur_insn].exit_qual != exit_qual)
960 			pass = false;
961 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
962 		    insn_table[cur_insn].insn_info != insn_info)
963 			pass = false;
964 		if (pass)
965 			vmx_inc_test_stage();
966 	}
967 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
968 	return VMX_TEST_RESUME;
969 }
970 
971 
972 /* Enables EPT and sets up the identity map. */
973 static int setup_ept(bool enable_ad)
974 {
975 	unsigned long end_of_memory;
976 	u32 ctrl_cpu[2];
977 
978 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
979 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
980 		printf("\tEPT is not supported");
981 		return 1;
982 	}
983 
984 
985 	if (!(ept_vpid.val & EPT_CAP_UC) &&
986 			!(ept_vpid.val & EPT_CAP_WB)) {
987 		printf("\tEPT paging-structure memory type "
988 				"UC&WB are not supported\n");
989 		return 1;
990 	}
991 	if (ept_vpid.val & EPT_CAP_UC)
992 		eptp = EPT_MEM_TYPE_UC;
993 	else
994 		eptp = EPT_MEM_TYPE_WB;
995 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
996 		printf("\tPWL4 is not supported\n");
997 		return 1;
998 	}
999 	ctrl_cpu[0] = vmcs_read(CPU_EXEC_CTRL0);
1000 	ctrl_cpu[1] = vmcs_read(CPU_EXEC_CTRL1);
1001 	ctrl_cpu[0] = (ctrl_cpu[0] | CPU_SECONDARY)
1002 		& ctrl_cpu_rev[0].clr;
1003 	ctrl_cpu[1] = (ctrl_cpu[1] | CPU_EPT)
1004 		& ctrl_cpu_rev[1].clr;
1005 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu[0]);
1006 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu[1]);
1007 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1008 	pml4 = alloc_page();
1009 	memset(pml4, 0, PAGE_SIZE);
1010 	eptp |= virt_to_phys(pml4);
1011 	if (enable_ad)
1012 		eptp |= EPTP_AD_FLAG;
1013 	vmcs_write(EPTP, eptp);
1014 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1015 	if (end_of_memory < (1ul << 32))
1016 		end_of_memory = (1ul << 32);
1017 	/* Cannot use large EPT pages if we need to track EPT
1018 	 * accessed/dirty bits at 4K granularity.
1019 	 */
1020 	setup_ept_range(pml4, 0, end_of_memory, 0,
1021 			!enable_ad && ept_2m_supported(),
1022 			EPT_WA | EPT_RA | EPT_EA);
1023 	return 0;
1024 }
1025 
1026 static void ept_enable_ad_bits(void)
1027 {
1028 	eptp |= EPTP_AD_FLAG;
1029 	vmcs_write(EPTP, eptp);
1030 }
1031 
1032 static void ept_disable_ad_bits(void)
1033 {
1034 	eptp &= ~EPTP_AD_FLAG;
1035 	vmcs_write(EPTP, eptp);
1036 }
1037 
1038 static void ept_enable_ad_bits_or_skip_test(void)
1039 {
1040 	if (!ept_ad_bits_supported())
1041 		test_skip("EPT AD bits not supported.");
1042 	ept_enable_ad_bits();
1043 }
1044 
1045 static int apic_version;
1046 
1047 static int ept_init_common(bool have_ad)
1048 {
1049 	int ret;
1050 	struct pci_dev pcidev;
1051 
1052 	if (setup_ept(have_ad))
1053 		return VMX_TEST_EXIT;
1054 	data_page1 = alloc_page();
1055 	data_page2 = alloc_page();
1056 	memset(data_page1, 0x0, PAGE_SIZE);
1057 	memset(data_page2, 0x0, PAGE_SIZE);
1058 	*((u32 *)data_page1) = MAGIC_VAL_1;
1059 	*((u32 *)data_page2) = MAGIC_VAL_2;
1060 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1061 			EPT_RA | EPT_WA | EPT_EA);
1062 
1063 	apic_version = apic_read(APIC_LVR);
1064 
1065 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1066 	if (ret != PCIDEVADDR_INVALID) {
1067 		pci_dev_init(&pcidev, ret);
1068 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1069 	}
1070 
1071 	return VMX_TEST_START;
1072 }
1073 
1074 static int ept_init()
1075 {
1076 	return ept_init_common(false);
1077 }
1078 
1079 static void ept_common()
1080 {
1081 	vmx_set_test_stage(0);
1082 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1083 			*((u32 *)data_page1) != MAGIC_VAL_1)
1084 		report("EPT basic framework - read", 0);
1085 	else {
1086 		*((u32 *)data_page2) = MAGIC_VAL_3;
1087 		vmcall();
1088 		if (vmx_get_test_stage() == 1) {
1089 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1090 					*((u32 *)data_page2) == MAGIC_VAL_2)
1091 				report("EPT basic framework", 1);
1092 			else
1093 				report("EPT basic framework - remap", 1);
1094 		}
1095 	}
1096 	// Test EPT Misconfigurations
1097 	vmx_set_test_stage(1);
1098 	vmcall();
1099 	*((u32 *)data_page1) = MAGIC_VAL_1;
1100 	if (vmx_get_test_stage() != 2) {
1101 		report("EPT misconfigurations", 0);
1102 		goto t1;
1103 	}
1104 	vmx_set_test_stage(2);
1105 	vmcall();
1106 	*((u32 *)data_page1) = MAGIC_VAL_1;
1107 	report("EPT misconfigurations", vmx_get_test_stage() == 3);
1108 t1:
1109 	// Test EPT violation
1110 	vmx_set_test_stage(3);
1111 	vmcall();
1112 	*((u32 *)data_page1) = MAGIC_VAL_1;
1113 	report("EPT violation - page permission", vmx_get_test_stage() == 4);
1114 	// Violation caused by EPT paging structure
1115 	vmx_set_test_stage(4);
1116 	vmcall();
1117 	*((u32 *)data_page1) = MAGIC_VAL_2;
1118 	report("EPT violation - paging structure", vmx_get_test_stage() == 5);
1119 
1120 	// MMIO Read/Write
1121 	vmx_set_test_stage(5);
1122 	vmcall();
1123 
1124 	*(u32 volatile *)pci_physaddr;
1125 	report("MMIO EPT violation - read", vmx_get_test_stage() == 6);
1126 
1127 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1128 	report("MMIO EPT violation - write", vmx_get_test_stage() == 7);
1129 }
1130 
1131 static void ept_main()
1132 {
1133 	ept_common();
1134 
1135 	// Test EPT access to L1 MMIO
1136 	vmx_set_test_stage(7);
1137 	report("EPT - MMIO access", *((u32 *)0xfee00030UL) == apic_version);
1138 
1139 	// Test invalid operand for INVEPT
1140 	vmcall();
1141 	report("EPT - unsupported INVEPT", vmx_get_test_stage() == 8);
1142 }
1143 
1144 bool invept_test(int type, u64 eptp)
1145 {
1146 	bool ret, supported;
1147 
1148 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1149 	ret = invept(type, eptp);
1150 
1151 	if (ret == !supported)
1152 		return false;
1153 
1154 	if (!supported)
1155 		printf("WARNING: unsupported invept passed!\n");
1156 	else
1157 		printf("WARNING: invept failed!\n");
1158 
1159 	return true;
1160 }
1161 
1162 static int pml_exit_handler(void)
1163 {
1164 	u16 index, count;
1165 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1166 	u64 *pmlbuf = pml_log;
1167 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1168 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1169 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1170 
1171 	switch (reason) {
1172 	case VMX_VMCALL:
1173 		switch (vmx_get_test_stage()) {
1174 		case 0:
1175 			index = vmcs_read(GUEST_PML_INDEX);
1176 			for (count = index + 1; count < PML_INDEX; count++) {
1177 				if (pmlbuf[count] == (u64)data_page2) {
1178 					vmx_inc_test_stage();
1179 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1180 					break;
1181 				}
1182 			}
1183 			break;
1184 		case 1:
1185 			index = vmcs_read(GUEST_PML_INDEX);
1186 			/* Keep clearing the dirty bit till a overflow */
1187 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1188 			break;
1189 		default:
1190 			report("unexpected stage, %d.", false,
1191 			       vmx_get_test_stage());
1192 			print_vmexit_info();
1193 			return VMX_TEST_VMEXIT;
1194 		}
1195 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1196 		return VMX_TEST_RESUME;
1197 	case VMX_PML_FULL:
1198 		vmx_inc_test_stage();
1199 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1200 		return VMX_TEST_RESUME;
1201 	default:
1202 		report("Unknown exit reason, %ld", false, reason);
1203 		print_vmexit_info();
1204 	}
1205 	return VMX_TEST_VMEXIT;
1206 }
1207 
1208 static int ept_exit_handler_common(bool have_ad)
1209 {
1210 	u64 guest_rip;
1211 	u64 guest_cr3;
1212 	ulong reason;
1213 	u32 insn_len;
1214 	u32 exit_qual;
1215 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte;
1216 
1217 	guest_rip = vmcs_read(GUEST_RIP);
1218 	guest_cr3 = vmcs_read(GUEST_CR3);
1219 	reason = vmcs_read(EXI_REASON) & 0xff;
1220 	insn_len = vmcs_read(EXI_INST_LEN);
1221 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1222 	switch (reason) {
1223 	case VMX_VMCALL:
1224 		switch (vmx_get_test_stage()) {
1225 		case 0:
1226 			check_ept_ad(pml4, guest_cr3,
1227 				     (unsigned long)data_page1,
1228 				     have_ad ? EPT_ACCESS_FLAG : 0,
1229 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1230 			check_ept_ad(pml4, guest_cr3,
1231 				     (unsigned long)data_page2,
1232 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1233 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1234 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1235 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1236 			if (have_ad)
1237 				ept_sync(INVEPT_SINGLE, eptp);;
1238 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1239 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1240 				vmx_inc_test_stage();
1241 				install_ept(pml4, (unsigned long)data_page2,
1242 						(unsigned long)data_page2,
1243 						EPT_RA | EPT_WA | EPT_EA);
1244 			} else
1245 				report("EPT basic framework - write", 0);
1246 			break;
1247 		case 1:
1248 			install_ept(pml4, (unsigned long)data_page1,
1249  				(unsigned long)data_page1, EPT_WA);
1250 			ept_sync(INVEPT_SINGLE, eptp);
1251 			break;
1252 		case 2:
1253 			install_ept(pml4, (unsigned long)data_page1,
1254  				(unsigned long)data_page1,
1255  				EPT_RA | EPT_WA | EPT_EA |
1256  				(2 << EPT_MEM_TYPE_SHIFT));
1257 			ept_sync(INVEPT_SINGLE, eptp);
1258 			break;
1259 		case 3:
1260 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1261 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1262 						1, &data_page1_pte));
1263 			set_ept_pte(pml4, (unsigned long)data_page1,
1264 				1, data_page1_pte & ~EPT_PRESENT);
1265 			ept_sync(INVEPT_SINGLE, eptp);
1266 			break;
1267 		case 4:
1268 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1269 						2, &data_page1_pte));
1270 			data_page1_pte &= PAGE_MASK;
1271 			TEST_ASSERT(get_ept_pte(pml4, data_page1_pte,
1272 						2, &data_page1_pte_pte));
1273 			set_ept_pte(pml4, data_page1_pte, 2,
1274 				data_page1_pte_pte & ~EPT_PRESENT);
1275 			ept_sync(INVEPT_SINGLE, eptp);
1276 			break;
1277 		case 5:
1278 			install_ept(pml4, (unsigned long)pci_physaddr,
1279 				(unsigned long)pci_physaddr, 0);
1280 			ept_sync(INVEPT_SINGLE, eptp);
1281 			break;
1282 		case 7:
1283 			if (!invept_test(0, eptp))
1284 				vmx_inc_test_stage();
1285 			break;
1286 		// Should not reach here
1287 		default:
1288 			report("ERROR - unexpected stage, %d.", false,
1289 			       vmx_get_test_stage());
1290 			print_vmexit_info();
1291 			return VMX_TEST_VMEXIT;
1292 		}
1293 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1294 		return VMX_TEST_RESUME;
1295 	case VMX_EPT_MISCONFIG:
1296 		switch (vmx_get_test_stage()) {
1297 		case 1:
1298 		case 2:
1299 			vmx_inc_test_stage();
1300 			install_ept(pml4, (unsigned long)data_page1,
1301  				(unsigned long)data_page1,
1302  				EPT_RA | EPT_WA | EPT_EA);
1303 			ept_sync(INVEPT_SINGLE, eptp);
1304 			break;
1305 		// Should not reach here
1306 		default:
1307 			report("ERROR - unexpected stage, %d.", false,
1308 			       vmx_get_test_stage());
1309 			print_vmexit_info();
1310 			return VMX_TEST_VMEXIT;
1311 		}
1312 		return VMX_TEST_RESUME;
1313 	case VMX_EPT_VIOLATION:
1314 		switch(vmx_get_test_stage()) {
1315 		case 3:
1316 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1317 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1318 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1319 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1320 					EPT_VLT_PADDR))
1321 				vmx_inc_test_stage();
1322 			set_ept_pte(pml4, (unsigned long)data_page1,
1323 				1, data_page1_pte | (EPT_PRESENT));
1324 			ept_sync(INVEPT_SINGLE, eptp);
1325 			break;
1326 		case 4:
1327 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1328 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1329 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1330 			if (exit_qual == (EPT_VLT_RD |
1331 					  (have_ad ? EPT_VLT_WR : 0) |
1332 					  EPT_VLT_LADDR_VLD))
1333 				vmx_inc_test_stage();
1334 			set_ept_pte(pml4, data_page1_pte, 2,
1335 				data_page1_pte_pte | (EPT_PRESENT));
1336 			ept_sync(INVEPT_SINGLE, eptp);
1337 			break;
1338 		case 5:
1339 			if (exit_qual & EPT_VLT_RD)
1340 				vmx_inc_test_stage();
1341 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1342 						1, &memaddr_pte));
1343 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1344 			ept_sync(INVEPT_SINGLE, eptp);
1345 			break;
1346 		case 6:
1347 			if (exit_qual & EPT_VLT_WR)
1348 				vmx_inc_test_stage();
1349 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1350 						1, &memaddr_pte));
1351 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1352 			ept_sync(INVEPT_SINGLE, eptp);
1353 			break;
1354 		default:
1355 			// Should not reach here
1356 			report("ERROR : unexpected stage, %d", false,
1357 			       vmx_get_test_stage());
1358 			print_vmexit_info();
1359 			return VMX_TEST_VMEXIT;
1360 		}
1361 		return VMX_TEST_RESUME;
1362 	default:
1363 		report("Unknown exit reason, %ld", false, reason);
1364 		print_vmexit_info();
1365 	}
1366 	return VMX_TEST_VMEXIT;
1367 }
1368 
1369 static int ept_exit_handler()
1370 {
1371 	return ept_exit_handler_common(false);
1372 }
1373 
1374 static int eptad_init()
1375 {
1376 	int r = ept_init_common(true);
1377 
1378 	if (r == VMX_TEST_EXIT)
1379 		return r;
1380 
1381 	if ((rdmsr(MSR_IA32_VMX_EPT_VPID_CAP) & EPT_CAP_AD_FLAG) == 0) {
1382 		printf("\tEPT A/D bits are not supported");
1383 		return VMX_TEST_EXIT;
1384 	}
1385 
1386 	return r;
1387 }
1388 
1389 static int pml_init()
1390 {
1391 	u32 ctrl_cpu;
1392 	int r = eptad_init();
1393 
1394 	if (r == VMX_TEST_EXIT)
1395 		return r;
1396 
1397 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1398 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1399 		printf("\tPML is not supported");
1400 		return VMX_TEST_EXIT;
1401 	}
1402 
1403 	pml_log = alloc_page();
1404 	memset(pml_log, 0x0, PAGE_SIZE);
1405 	vmcs_write(PMLADDR, (u64)pml_log);
1406 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1407 
1408 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1409 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1410 
1411 	return VMX_TEST_START;
1412 }
1413 
1414 static void pml_main()
1415 {
1416 	int count = 0;
1417 
1418 	vmx_set_test_stage(0);
1419 	*((u32 *)data_page2) = 0x1;
1420 	vmcall();
1421 	report("PML - Dirty GPA Logging", vmx_get_test_stage() == 1);
1422 
1423 	while (vmx_get_test_stage() == 1) {
1424 		vmcall();
1425 		*((u32 *)data_page2) = 0x1;
1426 		if (count++ > PML_INDEX)
1427 			break;
1428 	}
1429 	report("PML Full Event", vmx_get_test_stage() == 2);
1430 }
1431 
1432 static void eptad_main()
1433 {
1434 	ept_common();
1435 }
1436 
1437 static int eptad_exit_handler()
1438 {
1439 	return ept_exit_handler_common(true);
1440 }
1441 
1442 bool invvpid_test(int type, u16 vpid)
1443 {
1444 	bool ret, supported;
1445 
1446 	supported = ept_vpid.val &
1447 		(VPID_CAP_INVVPID_ADDR >> INVVPID_ADDR << type);
1448 	ret = invvpid(type, vpid, 0);
1449 
1450 	if (ret == !supported)
1451 		return false;
1452 
1453 	if (!supported)
1454 		printf("WARNING: unsupported invvpid passed!\n");
1455 	else
1456 		printf("WARNING: invvpid failed!\n");
1457 
1458 	return true;
1459 }
1460 
1461 static int vpid_init()
1462 {
1463 	u32 ctrl_cpu1;
1464 
1465 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1466 		!(ctrl_cpu_rev[1].clr & CPU_VPID)) {
1467 		printf("\tVPID is not supported");
1468 		return VMX_TEST_EXIT;
1469 	}
1470 
1471 	ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
1472 	ctrl_cpu1 |= CPU_VPID;
1473 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
1474 	return VMX_TEST_START;
1475 }
1476 
1477 static void vpid_main()
1478 {
1479 	vmx_set_test_stage(0);
1480 	vmcall();
1481 	report("INVVPID SINGLE ADDRESS", vmx_get_test_stage() == 1);
1482 	vmx_set_test_stage(2);
1483 	vmcall();
1484 	report("INVVPID SINGLE", vmx_get_test_stage() == 3);
1485 	vmx_set_test_stage(4);
1486 	vmcall();
1487 	report("INVVPID ALL", vmx_get_test_stage() == 5);
1488 }
1489 
1490 static int vpid_exit_handler()
1491 {
1492 	u64 guest_rip;
1493 	ulong reason;
1494 	u32 insn_len;
1495 
1496 	guest_rip = vmcs_read(GUEST_RIP);
1497 	reason = vmcs_read(EXI_REASON) & 0xff;
1498 	insn_len = vmcs_read(EXI_INST_LEN);
1499 
1500 	switch (reason) {
1501 	case VMX_VMCALL:
1502 		switch(vmx_get_test_stage()) {
1503 		case 0:
1504 			if (!invvpid_test(INVVPID_ADDR, 1))
1505 				vmx_inc_test_stage();
1506 			break;
1507 		case 2:
1508 			if (!invvpid_test(INVVPID_CONTEXT_GLOBAL, 1))
1509 				vmx_inc_test_stage();
1510 			break;
1511 		case 4:
1512 			if (!invvpid_test(INVVPID_ALL, 1))
1513 				vmx_inc_test_stage();
1514 			break;
1515 		default:
1516 			report("ERROR: unexpected stage, %d", false,
1517 					vmx_get_test_stage());
1518 			print_vmexit_info();
1519 			return VMX_TEST_VMEXIT;
1520 		}
1521 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1522 		return VMX_TEST_RESUME;
1523 	default:
1524 		report("Unknown exit reason, %ld", false, reason);
1525 		print_vmexit_info();
1526 	}
1527 	return VMX_TEST_VMEXIT;
1528 }
1529 
1530 #define TIMER_VECTOR	222
1531 
1532 static volatile bool timer_fired;
1533 
1534 static void timer_isr(isr_regs_t *regs)
1535 {
1536 	timer_fired = true;
1537 	apic_write(APIC_EOI, 0);
1538 }
1539 
1540 static int interrupt_init(struct vmcs *vmcs)
1541 {
1542 	msr_bmp_init();
1543 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1544 	handle_irq(TIMER_VECTOR, timer_isr);
1545 	return VMX_TEST_START;
1546 }
1547 
1548 static void interrupt_main(void)
1549 {
1550 	long long start, loops;
1551 
1552 	vmx_set_test_stage(0);
1553 
1554 	apic_write(APIC_LVTT, TIMER_VECTOR);
1555 	irq_enable();
1556 
1557 	apic_write(APIC_TMICT, 1);
1558 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1559 		asm volatile ("nop");
1560 	report("direct interrupt while running guest", timer_fired);
1561 
1562 	apic_write(APIC_TMICT, 0);
1563 	irq_disable();
1564 	vmcall();
1565 	timer_fired = false;
1566 	apic_write(APIC_TMICT, 1);
1567 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1568 		asm volatile ("nop");
1569 	report("intercepted interrupt while running guest", timer_fired);
1570 
1571 	irq_enable();
1572 	apic_write(APIC_TMICT, 0);
1573 	irq_disable();
1574 	vmcall();
1575 	timer_fired = false;
1576 	start = rdtsc();
1577 	apic_write(APIC_TMICT, 1000000);
1578 
1579 	asm volatile ("sti; hlt");
1580 
1581 	report("direct interrupt + hlt",
1582 	       rdtsc() - start > 1000000 && timer_fired);
1583 
1584 	apic_write(APIC_TMICT, 0);
1585 	irq_disable();
1586 	vmcall();
1587 	timer_fired = false;
1588 	start = rdtsc();
1589 	apic_write(APIC_TMICT, 1000000);
1590 
1591 	asm volatile ("sti; hlt");
1592 
1593 	report("intercepted interrupt + hlt",
1594 	       rdtsc() - start > 10000 && timer_fired);
1595 
1596 	apic_write(APIC_TMICT, 0);
1597 	irq_disable();
1598 	vmcall();
1599 	timer_fired = false;
1600 	start = rdtsc();
1601 	apic_write(APIC_TMICT, 1000000);
1602 
1603 	irq_enable();
1604 	asm volatile ("nop");
1605 	vmcall();
1606 
1607 	report("direct interrupt + activity state hlt",
1608 	       rdtsc() - start > 10000 && timer_fired);
1609 
1610 	apic_write(APIC_TMICT, 0);
1611 	irq_disable();
1612 	vmcall();
1613 	timer_fired = false;
1614 	start = rdtsc();
1615 	apic_write(APIC_TMICT, 1000000);
1616 
1617 	irq_enable();
1618 	asm volatile ("nop");
1619 	vmcall();
1620 
1621 	report("intercepted interrupt + activity state hlt",
1622 	       rdtsc() - start > 10000 && timer_fired);
1623 
1624 	apic_write(APIC_TMICT, 0);
1625 	irq_disable();
1626 	vmx_set_test_stage(7);
1627 	vmcall();
1628 	timer_fired = false;
1629 	apic_write(APIC_TMICT, 1);
1630 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1631 		asm volatile ("nop");
1632 	report("running a guest with interrupt acknowledgement set", timer_fired);
1633 
1634 	apic_write(APIC_TMICT, 0);
1635 	irq_enable();
1636 	timer_fired = false;
1637 	vmcall();
1638 	report("Inject an event to a halted guest", timer_fired);
1639 }
1640 
1641 static int interrupt_exit_handler(void)
1642 {
1643 	u64 guest_rip = vmcs_read(GUEST_RIP);
1644 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1645 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1646 
1647 	switch (reason) {
1648 	case VMX_VMCALL:
1649 		switch (vmx_get_test_stage()) {
1650 		case 0:
1651 		case 2:
1652 		case 5:
1653 			vmcs_write(PIN_CONTROLS,
1654 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1655 			break;
1656 		case 7:
1657 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1658 			vmcs_write(PIN_CONTROLS,
1659 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1660 			break;
1661 		case 1:
1662 		case 3:
1663 			vmcs_write(PIN_CONTROLS,
1664 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1665 			break;
1666 		case 4:
1667 		case 6:
1668 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1669 			break;
1670 
1671 		case 8:
1672 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1673 			vmcs_write(ENT_INTR_INFO,
1674 				   TIMER_VECTOR |
1675 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1676 				   INTR_INFO_VALID_MASK);
1677 			break;
1678 		}
1679 		vmx_inc_test_stage();
1680 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1681 		return VMX_TEST_RESUME;
1682 	case VMX_EXTINT:
1683 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1684 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1685 			handle_external_interrupt(vector);
1686 		} else {
1687 			irq_enable();
1688 			asm volatile ("nop");
1689 			irq_disable();
1690 		}
1691 		if (vmx_get_test_stage() >= 2)
1692 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1693 		return VMX_TEST_RESUME;
1694 	default:
1695 		report("Unknown exit reason, %ld", false, reason);
1696 		print_vmexit_info();
1697 	}
1698 
1699 	return VMX_TEST_VMEXIT;
1700 }
1701 
1702 static int dbgctls_init(struct vmcs *vmcs)
1703 {
1704 	u64 dr7 = 0x402;
1705 	u64 zero = 0;
1706 
1707 	msr_bmp_init();
1708 	asm volatile(
1709 		"mov %0,%%dr0\n\t"
1710 		"mov %0,%%dr1\n\t"
1711 		"mov %0,%%dr2\n\t"
1712 		"mov %1,%%dr7\n\t"
1713 		: : "r" (zero), "r" (dr7));
1714 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1715 	vmcs_write(GUEST_DR7, 0x404);
1716 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1717 
1718 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1719 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1720 
1721 	return VMX_TEST_START;
1722 }
1723 
1724 static void dbgctls_main(void)
1725 {
1726 	u64 dr7, debugctl;
1727 
1728 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1729 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1730 	/* Commented out: KVM does not support DEBUGCTL so far */
1731 	(void)debugctl;
1732 	report("Load debug controls", dr7 == 0x404 /* && debugctl == 0x2 */);
1733 
1734 	dr7 = 0x408;
1735 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1736 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1737 
1738 	vmx_set_test_stage(0);
1739 	vmcall();
1740 	report("Save debug controls", vmx_get_test_stage() == 1);
1741 
1742 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1743 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1744 		printf("\tDebug controls are always loaded/saved\n");
1745 		return;
1746 	}
1747 	vmx_set_test_stage(2);
1748 	vmcall();
1749 
1750 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1751 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1752 	/* Commented out: KVM does not support DEBUGCTL so far */
1753 	(void)debugctl;
1754 	report("Guest=host debug controls", dr7 == 0x402 /* && debugctl == 0x1 */);
1755 
1756 	dr7 = 0x408;
1757 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1758 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1759 
1760 	vmx_set_test_stage(3);
1761 	vmcall();
1762 	report("Don't save debug controls", vmx_get_test_stage() == 4);
1763 }
1764 
1765 static int dbgctls_exit_handler(void)
1766 {
1767 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
1768 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1769 	u64 guest_rip = vmcs_read(GUEST_RIP);
1770 	u64 dr7, debugctl;
1771 
1772 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1773 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1774 
1775 	switch (reason) {
1776 	case VMX_VMCALL:
1777 		switch (vmx_get_test_stage()) {
1778 		case 0:
1779 			if (dr7 == 0x400 && debugctl == 0 &&
1780 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1781 			    Commented out: KVM does not support DEBUGCTL so far
1782 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1783 				vmx_inc_test_stage();
1784 			break;
1785 		case 2:
1786 			dr7 = 0x402;
1787 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1788 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1789 			vmcs_write(GUEST_DR7, 0x404);
1790 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1791 
1792 			vmcs_write(ENT_CONTROLS,
1793 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1794 			vmcs_write(EXI_CONTROLS,
1795 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1796 			break;
1797 		case 3:
1798 			if (dr7 == 0x400 && debugctl == 0 &&
1799 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1800 			    Commented out: KVM does not support DEBUGCTL so far
1801 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1802 				vmx_inc_test_stage();
1803 			break;
1804 		}
1805 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1806 		return VMX_TEST_RESUME;
1807 	default:
1808 		report("Unknown exit reason, %d", false, reason);
1809 		print_vmexit_info();
1810 	}
1811 	return VMX_TEST_VMEXIT;
1812 }
1813 
1814 struct vmx_msr_entry {
1815 	u32 index;
1816 	u32 reserved;
1817 	u64 value;
1818 } __attribute__((packed));
1819 
1820 #define MSR_MAGIC 0x31415926
1821 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1822 
1823 static int msr_switch_init(struct vmcs *vmcs)
1824 {
1825 	msr_bmp_init();
1826 	exit_msr_store = alloc_page();
1827 	exit_msr_load = alloc_page();
1828 	entry_msr_load = alloc_page();
1829 	memset(exit_msr_store, 0, PAGE_SIZE);
1830 	memset(exit_msr_load, 0, PAGE_SIZE);
1831 	memset(entry_msr_load, 0, PAGE_SIZE);
1832 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1833 	entry_msr_load[0].value = MSR_MAGIC;
1834 
1835 	vmx_set_test_stage(1);
1836 	vmcs_write(ENT_MSR_LD_CNT, 1);
1837 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1838 	vmcs_write(EXI_MSR_ST_CNT, 1);
1839 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1840 	vmcs_write(EXI_MSR_LD_CNT, 1);
1841 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1842 	return VMX_TEST_START;
1843 }
1844 
1845 static void msr_switch_main()
1846 {
1847 	if (vmx_get_test_stage() == 1) {
1848 		report("VM entry MSR load",
1849 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC);
1850 		vmx_set_test_stage(2);
1851 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
1852 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
1853 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
1854 		exit_msr_load[0].value = MSR_MAGIC + 2;
1855 	}
1856 	vmcall();
1857 }
1858 
1859 static int msr_switch_exit_handler()
1860 {
1861 	ulong reason;
1862 
1863 	reason = vmcs_read(EXI_REASON);
1864 	if (reason == VMX_VMCALL && vmx_get_test_stage() == 2) {
1865 		report("VM exit MSR store",
1866 			exit_msr_store[0].value == MSR_MAGIC + 1);
1867 		report("VM exit MSR load",
1868 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2);
1869 		vmx_set_test_stage(3);
1870 		entry_msr_load[0].index = MSR_FS_BASE;
1871 		return VMX_TEST_RESUME;
1872 	}
1873 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1874 		__func__, vmx_get_test_stage(), reason);
1875 	return VMX_TEST_EXIT;
1876 }
1877 
1878 static int msr_switch_entry_failure(struct vmentry_failure *failure)
1879 {
1880 	ulong reason;
1881 
1882 	if (failure->early) {
1883 		printf("ERROR %s: early exit\n", __func__);
1884 		return VMX_TEST_EXIT;
1885 	}
1886 
1887 	reason = vmcs_read(EXI_REASON);
1888 	if (reason == (VMX_ENTRY_FAILURE | VMX_FAIL_MSR) &&
1889 	    vmx_get_test_stage() == 3) {
1890 		report("VM entry MSR load: try to load FS_BASE",
1891 			vmcs_read(EXI_QUALIFICATION) == 1);
1892 		return VMX_TEST_VMEXIT;
1893 	}
1894 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1895 		__func__, vmx_get_test_stage(), reason);
1896 	return VMX_TEST_EXIT;
1897 }
1898 
1899 static int vmmcall_init(struct vmcs *vmcs	)
1900 {
1901 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
1902 	return VMX_TEST_START;
1903 }
1904 
1905 static void vmmcall_main(void)
1906 {
1907 	asm volatile(
1908 		"mov $0xABCD, %%rax\n\t"
1909 		"vmmcall\n\t"
1910 		::: "rax");
1911 
1912 	report("VMMCALL", 0);
1913 }
1914 
1915 static int vmmcall_exit_handler()
1916 {
1917 	ulong reason;
1918 
1919 	reason = vmcs_read(EXI_REASON);
1920 	switch (reason) {
1921 	case VMX_VMCALL:
1922 		printf("here\n");
1923 		report("VMMCALL triggers #UD", 0);
1924 		break;
1925 	case VMX_EXC_NMI:
1926 		report("VMMCALL triggers #UD",
1927 		       (vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR);
1928 		break;
1929 	default:
1930 		report("Unknown exit reason, %ld", false, reason);
1931 		print_vmexit_info();
1932 	}
1933 
1934 	return VMX_TEST_VMEXIT;
1935 }
1936 
1937 static int disable_rdtscp_init(struct vmcs *vmcs)
1938 {
1939 	u32 ctrl_cpu1;
1940 
1941 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
1942 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
1943 		ctrl_cpu1 &= ~CPU_RDTSCP;
1944 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
1945 	}
1946 
1947 	return VMX_TEST_START;
1948 }
1949 
1950 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
1951 {
1952 	switch (vmx_get_test_stage()) {
1953 	case 0:
1954 		report("RDTSCP triggers #UD", true);
1955 		vmx_inc_test_stage();
1956 		regs->rip += 3;
1957 		break;
1958 	case 2:
1959 		report("RDPID triggers #UD", true);
1960 		vmx_inc_test_stage();
1961 		regs->rip += 4;
1962 		break;
1963 	}
1964 	return;
1965 
1966 }
1967 
1968 static void disable_rdtscp_main(void)
1969 {
1970 	/* Test that #UD is properly injected in L2.  */
1971 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
1972 
1973 	vmx_set_test_stage(0);
1974 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
1975 	vmcall();
1976 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
1977 	vmcall();
1978 }
1979 
1980 static int disable_rdtscp_exit_handler(void)
1981 {
1982 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
1983 
1984 	switch (reason) {
1985 	case VMX_VMCALL:
1986 		switch (vmx_get_test_stage()) {
1987 		case 0:
1988 			report("RDTSCP triggers #UD", false);
1989 			vmx_inc_test_stage();
1990 			/* fallthrough */
1991 		case 1:
1992 			vmx_inc_test_stage();
1993 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
1994 			return VMX_TEST_RESUME;
1995 		case 2:
1996 			report("RDPID triggers #UD", false);
1997 			break;
1998 		}
1999 		break;
2000 
2001 	default:
2002 		report("Unknown exit reason, %d", false, reason);
2003 		print_vmexit_info();
2004 	}
2005 	return VMX_TEST_VMEXIT;
2006 }
2007 
2008 int int3_init()
2009 {
2010 	vmcs_write(EXC_BITMAP, ~0u);
2011 	return VMX_TEST_START;
2012 }
2013 
2014 void int3_guest_main()
2015 {
2016 	asm volatile ("int3");
2017 }
2018 
2019 int int3_exit_handler()
2020 {
2021 	u32 reason = vmcs_read(EXI_REASON);
2022 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2023 
2024 	report("L1 intercepts #BP", reason == VMX_EXC_NMI &&
2025 	       (intr_info & INTR_INFO_VALID_MASK) &&
2026 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2027 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2028 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2029 
2030 	return VMX_TEST_VMEXIT;
2031 }
2032 
2033 int into_init()
2034 {
2035 	vmcs_write(EXC_BITMAP, ~0u);
2036 	return VMX_TEST_START;
2037 }
2038 
2039 void into_guest_main()
2040 {
2041 	struct far_pointer32 fp = {
2042 		.offset = (uintptr_t)&&into,
2043 		.selector = KERNEL_CS32,
2044 	};
2045 	register uintptr_t rsp asm("rsp");
2046 
2047 	if (fp.offset != (uintptr_t)&&into) {
2048 		printf("Code address too high.\n");
2049 		return;
2050 	}
2051 	if ((u32)rsp != rsp) {
2052 		printf("Stack address too high.\n");
2053 		return;
2054 	}
2055 
2056 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2057 	return;
2058 into:
2059 	asm volatile (".code32;"
2060 		      "movl $0x7fffffff, %eax;"
2061 		      "addl %eax, %eax;"
2062 		      "into;"
2063 		      "lret;"
2064 		      ".code64");
2065 	__builtin_unreachable();
2066 }
2067 
2068 int into_exit_handler()
2069 {
2070 	u32 reason = vmcs_read(EXI_REASON);
2071 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2072 
2073 	report("L1 intercepts #OF", reason == VMX_EXC_NMI &&
2074 	       (intr_info & INTR_INFO_VALID_MASK) &&
2075 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2076 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2077 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2078 
2079 	return VMX_TEST_VMEXIT;
2080 }
2081 
2082 static void exit_monitor_from_l2_main(void)
2083 {
2084 	printf("Calling exit(0) from l2...\n");
2085 	exit(0);
2086 }
2087 
2088 static int exit_monitor_from_l2_handler(void)
2089 {
2090 	report("The guest should have killed the VMM", false);
2091 	return VMX_TEST_EXIT;
2092 }
2093 
2094 static void assert_exit_reason(u64 expected)
2095 {
2096 	u64 actual = vmcs_read(EXI_REASON);
2097 
2098 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2099 			   exit_reason_description(expected),
2100 			   exit_reason_description(actual));
2101 }
2102 
2103 static void skip_exit_vmcall()
2104 {
2105 	u64 guest_rip = vmcs_read(GUEST_RIP);
2106 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2107 
2108 	assert_exit_reason(VMX_VMCALL);
2109 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2110 }
2111 
2112 static void v2_null_test_guest(void)
2113 {
2114 }
2115 
2116 static void v2_null_test(void)
2117 {
2118 	test_set_guest(v2_null_test_guest);
2119 	enter_guest();
2120 	report(__func__, 1);
2121 }
2122 
2123 static void v2_multiple_entries_test_guest(void)
2124 {
2125 	vmx_set_test_stage(1);
2126 	vmcall();
2127 	vmx_set_test_stage(2);
2128 }
2129 
2130 static void v2_multiple_entries_test(void)
2131 {
2132 	test_set_guest(v2_multiple_entries_test_guest);
2133 	enter_guest();
2134 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2135 	skip_exit_vmcall();
2136 	enter_guest();
2137 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2138 	report(__func__, 1);
2139 }
2140 
2141 static int fixture_test_data = 1;
2142 
2143 static void fixture_test_teardown(void *data)
2144 {
2145 	*((int *) data) = 1;
2146 }
2147 
2148 static void fixture_test_guest(void)
2149 {
2150 	fixture_test_data++;
2151 }
2152 
2153 
2154 static void fixture_test_setup(void)
2155 {
2156 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2157 			   "fixture_test_teardown didn't run?!");
2158 	fixture_test_data = 2;
2159 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2160 	test_set_guest(fixture_test_guest);
2161 }
2162 
2163 static void fixture_test_case1(void)
2164 {
2165 	fixture_test_setup();
2166 	TEST_ASSERT_EQ(2, fixture_test_data);
2167 	enter_guest();
2168 	TEST_ASSERT_EQ(3, fixture_test_data);
2169 	report(__func__, 1);
2170 }
2171 
2172 static void fixture_test_case2(void)
2173 {
2174 	fixture_test_setup();
2175 	TEST_ASSERT_EQ(2, fixture_test_data);
2176 	enter_guest();
2177 	TEST_ASSERT_EQ(3, fixture_test_data);
2178 	report(__func__, 1);
2179 }
2180 
2181 enum ept_access_op {
2182 	OP_READ,
2183 	OP_WRITE,
2184 	OP_EXEC,
2185 	OP_FLUSH_TLB,
2186 	OP_EXIT,
2187 };
2188 
2189 static struct ept_access_test_data {
2190 	unsigned long gpa;
2191 	unsigned long *gva;
2192 	unsigned long hpa;
2193 	unsigned long *hva;
2194 	enum ept_access_op op;
2195 } ept_access_test_data;
2196 
2197 extern unsigned char ret42_start;
2198 extern unsigned char ret42_end;
2199 
2200 /* Returns 42. */
2201 asm(
2202 	".align 64\n"
2203 	"ret42_start:\n"
2204 	"mov $42, %eax\n"
2205 	"ret\n"
2206 	"ret42_end:\n"
2207 );
2208 
2209 static void
2210 diagnose_ept_violation_qual(u64 expected, u64 actual)
2211 {
2212 
2213 #define DIAGNOSE(flag)							\
2214 do {									\
2215 	if ((expected & flag) != (actual & flag))			\
2216 		printf(#flag " %sexpected\n",				\
2217 		       (expected & flag) ? "" : "un");			\
2218 } while (0)
2219 
2220 	DIAGNOSE(EPT_VLT_RD);
2221 	DIAGNOSE(EPT_VLT_WR);
2222 	DIAGNOSE(EPT_VLT_FETCH);
2223 	DIAGNOSE(EPT_VLT_PERM_RD);
2224 	DIAGNOSE(EPT_VLT_PERM_WR);
2225 	DIAGNOSE(EPT_VLT_PERM_EX);
2226 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2227 	DIAGNOSE(EPT_VLT_PADDR);
2228 
2229 #undef DIAGNOSE
2230 }
2231 
2232 static void do_ept_access_op(enum ept_access_op op)
2233 {
2234 	ept_access_test_data.op = op;
2235 	enter_guest();
2236 }
2237 
2238 /*
2239  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2240  * needed by tests that modify guest PTEs.
2241  */
2242 static void ept_access_test_guest_flush_tlb(void)
2243 {
2244 	do_ept_access_op(OP_FLUSH_TLB);
2245 	skip_exit_vmcall();
2246 }
2247 
2248 /*
2249  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2250  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2251  * a huge page.
2252  */
2253 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2254 				 unsigned long clear, unsigned long set)
2255 {
2256 	struct ept_access_test_data *data = &ept_access_test_data;
2257 	unsigned long orig_pte;
2258 	unsigned long pte;
2259 
2260 	/* Screw with the mapping at the requested level. */
2261 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2262 	pte = orig_pte;
2263 	if (mkhuge)
2264 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2265 	else
2266 		pte = orig_pte;
2267 	pte = (pte & ~clear) | set;
2268 	set_ept_pte(pml4, gpa, level, pte);
2269 	ept_sync(INVEPT_SINGLE, eptp);
2270 
2271 	return orig_pte;
2272 }
2273 
2274 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2275 {
2276 	set_ept_pte(pml4, gpa, level, orig_pte);
2277 }
2278 
2279 static void do_ept_violation(bool leaf, enum ept_access_op op,
2280 			     u64 expected_qual, u64 expected_paddr)
2281 {
2282 	u64 qual;
2283 
2284 	/* Try the access and observe the violation. */
2285 	do_ept_access_op(op);
2286 
2287 	assert_exit_reason(VMX_EPT_VIOLATION);
2288 
2289 	qual = vmcs_read(EXI_QUALIFICATION);
2290 
2291 	diagnose_ept_violation_qual(expected_qual, qual);
2292 	TEST_EXPECT_EQ(expected_qual, qual);
2293 
2294 	#if 0
2295 	/* Disable for now otherwise every test will fail */
2296 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2297 		       (unsigned long) (
2298 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2299 	#endif
2300 	/*
2301 	 * TODO: tests that probe expected_paddr in pages other than the one at
2302 	 * the beginning of the 1g region.
2303 	 */
2304 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2305 }
2306 
2307 static void
2308 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2309 			      unsigned long set, enum ept_access_op op,
2310 			      u64 expected_qual)
2311 {
2312 	struct ept_access_test_data *data = &ept_access_test_data;
2313 	unsigned long orig_pte;
2314 
2315 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2316 
2317 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2318 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2319 					 data->gpa);
2320 
2321 	/* Fix the violation and resume the op loop. */
2322 	ept_untwiddle(data->gpa, level, orig_pte);
2323 	enter_guest();
2324 	skip_exit_vmcall();
2325 }
2326 
2327 static void
2328 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2329 		       enum ept_access_op op, u64 expected_qual)
2330 {
2331 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2332 				      expected_qual);
2333 	if (ept_huge_pages_supported(level))
2334 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2335 					      expected_qual);
2336 }
2337 
2338 static void ept_violation(unsigned long clear, unsigned long set,
2339 			  enum ept_access_op op, u64 expected_qual)
2340 {
2341 	ept_violation_at_level(1, clear, set, op, expected_qual);
2342 	ept_violation_at_level(2, clear, set, op, expected_qual);
2343 	ept_violation_at_level(3, clear, set, op, expected_qual);
2344 	ept_violation_at_level(4, clear, set, op, expected_qual);
2345 }
2346 
2347 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2348 				       u64 expected_qual)
2349 {
2350 	ept_violation(EPT_PRESENT, access, op,
2351 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2352 }
2353 
2354 /*
2355  * For translations that don't involve a GVA, that is physical address (paddr)
2356  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2357  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2358  * translations don't involve GVAs, such as when the hardware does the guest
2359  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2360  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2361  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2362  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2363  *
2364  * Note that paddr violations can also be triggered by loading PAE page tables
2365  * with wonky addresses. We don't test that yet.
2366  *
2367  * This function modifies the EPT entry that maps the GPA that the guest page
2368  * table entry mapping ept_access_data.gva resides on.
2369  *
2370  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2371  *
2372  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2373  *
2374  *	@op		Guest operation to perform with ept_access_data.gva.
2375  *
2376  *	@expect_violation
2377  *			Is a violation expected during the paddr access?
2378  *
2379  *	@expected_qual	Expected qualification for the EPT violation.
2380  *			EPT_VLT_PADDR should be clear.
2381  */
2382 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2383 			     enum ept_access_op op, bool expect_violation,
2384 			     u64 expected_qual)
2385 {
2386 	struct ept_access_test_data *data = &ept_access_test_data;
2387 	unsigned long *ptep;
2388 	unsigned long gpa;
2389 	unsigned long orig_epte;
2390 
2391 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2392 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2393 	TEST_ASSERT(ptep);
2394 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2395 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2396 	ept_access_test_guest_flush_tlb();
2397 
2398 	/*
2399 	 * Now modify the access bits on the EPT entry for the GPA that the
2400 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2401 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2402 	 * constructed our test such that those other 511 PTEs aren't used by
2403 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2404 	 * PTE we're modifying is at the beginning of a 4K page and the
2405 	 * following 511 entires are also under our control (and not touched by
2406 	 * the guest).
2407 	 */
2408 	gpa = virt_to_phys(ptep);
2409 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2410 	/*
2411 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2412 	 * otherwise our level=1 twiddling below will fail. We use the
2413 	 * identity map (gpa = gpa) since page tables are shared with the host.
2414 	 */
2415 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2416 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2417 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2418 
2419 	if (expect_violation) {
2420 		do_ept_violation(/*leaf=*/true, op,
2421 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2422 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2423 		do_ept_access_op(op);
2424 	} else {
2425 		do_ept_access_op(op);
2426 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2427 	}
2428 
2429 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2430 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2431 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2432 
2433 	skip_exit_vmcall();
2434 }
2435 
2436 static void ept_access_allowed_paddr(unsigned long ept_access,
2437 				     unsigned long pte_ad,
2438 				     enum ept_access_op op)
2439 {
2440 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2441 			 /*expected_qual=*/-1);
2442 }
2443 
2444 static void ept_access_violation_paddr(unsigned long ept_access,
2445 				       unsigned long pte_ad,
2446 				       enum ept_access_op op,
2447 				       u64 expected_qual)
2448 {
2449 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2450 			 expected_qual);
2451 }
2452 
2453 
2454 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2455 					unsigned long clear,
2456 					unsigned long set,
2457 					enum ept_access_op op)
2458 {
2459 	struct ept_access_test_data *data = &ept_access_test_data;
2460 	unsigned long orig_pte;
2461 
2462 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2463 
2464 	/* No violation. Should proceed to vmcall. */
2465 	do_ept_access_op(op);
2466 	skip_exit_vmcall();
2467 
2468 	ept_untwiddle(data->gpa, level, orig_pte);
2469 }
2470 
2471 static void ept_allowed_at_level(int level, unsigned long clear,
2472 				 unsigned long set, enum ept_access_op op)
2473 {
2474 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2475 	if (ept_huge_pages_supported(level))
2476 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2477 }
2478 
2479 static void ept_allowed(unsigned long clear, unsigned long set,
2480 			enum ept_access_op op)
2481 {
2482 	ept_allowed_at_level(1, clear, set, op);
2483 	ept_allowed_at_level(2, clear, set, op);
2484 	ept_allowed_at_level(3, clear, set, op);
2485 	ept_allowed_at_level(4, clear, set, op);
2486 }
2487 
2488 static void ept_ignored_bit(int bit)
2489 {
2490 	/* Set the bit. */
2491 	ept_allowed(0, 1ul << bit, OP_READ);
2492 	ept_allowed(0, 1ul << bit, OP_WRITE);
2493 	ept_allowed(0, 1ul << bit, OP_EXEC);
2494 
2495 	/* Clear the bit. */
2496 	ept_allowed(1ul << bit, 0, OP_READ);
2497 	ept_allowed(1ul << bit, 0, OP_WRITE);
2498 	ept_allowed(1ul << bit, 0, OP_EXEC);
2499 }
2500 
2501 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2502 {
2503 	ept_allowed(EPT_PRESENT, access, op);
2504 }
2505 
2506 
2507 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2508 					     unsigned long clear,
2509 					     unsigned long set,
2510 					     enum ept_access_op op)
2511 {
2512 	struct ept_access_test_data *data = &ept_access_test_data;
2513 	unsigned long orig_pte;
2514 
2515 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2516 
2517 	do_ept_access_op(op);
2518 	assert_exit_reason(VMX_EPT_MISCONFIG);
2519 
2520 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2521 	#if 0
2522 	/* broken: */
2523 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2524 	#endif
2525 	#if 0
2526 	/*
2527 	 * broken:
2528 	 * According to description of exit qual for EPT violation,
2529 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2530 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2531 	 * to be set for msiconfig.
2532 	 */
2533 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2534 		       (unsigned long) (
2535 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2536 	#endif
2537 
2538 	/* Fix the violation and resume the op loop. */
2539 	ept_untwiddle(data->gpa, level, orig_pte);
2540 	enter_guest();
2541 	skip_exit_vmcall();
2542 }
2543 
2544 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2545 					  unsigned long clear,
2546 					  unsigned long set)
2547 {
2548 	/* The op shouldn't matter (read, write, exec), so try them all! */
2549 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2550 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2551 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2552 }
2553 
2554 static void ept_misconfig_at_level(int level, unsigned long clear,
2555 				   unsigned long set)
2556 {
2557 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2558 	if (ept_huge_pages_supported(level))
2559 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2560 }
2561 
2562 static void ept_misconfig(unsigned long clear, unsigned long set)
2563 {
2564 	ept_misconfig_at_level(1, clear, set);
2565 	ept_misconfig_at_level(2, clear, set);
2566 	ept_misconfig_at_level(3, clear, set);
2567 	ept_misconfig_at_level(4, clear, set);
2568 }
2569 
2570 static void ept_access_misconfig(unsigned long access)
2571 {
2572 	ept_misconfig(EPT_PRESENT, access);
2573 }
2574 
2575 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2576 {
2577 	/* Setting the bit causes a misconfig. */
2578 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2579 
2580 	/* Making the entry non-present turns reserved bits into ignored. */
2581 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2582 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2583 }
2584 
2585 static void ept_reserved_bit_at_level_huge(int level, int bit)
2586 {
2587 	/* Setting the bit causes a misconfig. */
2588 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2589 
2590 	/* Making the entry non-present turns reserved bits into ignored. */
2591 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2592 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2593 }
2594 
2595 static void ept_reserved_bit_at_level(int level, int bit)
2596 {
2597 	/* Setting the bit causes a misconfig. */
2598 	ept_misconfig_at_level(level, 0, 1ul << bit);
2599 
2600 	/* Making the entry non-present turns reserved bits into ignored. */
2601 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2602 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2603 }
2604 
2605 static void ept_reserved_bit(int bit)
2606 {
2607 	ept_reserved_bit_at_level(1, bit);
2608 	ept_reserved_bit_at_level(2, bit);
2609 	ept_reserved_bit_at_level(3, bit);
2610 	ept_reserved_bit_at_level(4, bit);
2611 }
2612 
2613 #define PAGE_2M_ORDER 9
2614 #define PAGE_1G_ORDER 18
2615 
2616 static void *get_1g_page(void)
2617 {
2618 	static void *alloc;
2619 
2620 	if (!alloc)
2621 		alloc = alloc_pages(PAGE_1G_ORDER);
2622 	return alloc;
2623 }
2624 
2625 static void ept_access_test_teardown(void *unused)
2626 {
2627 	/* Exit the guest cleanly. */
2628 	do_ept_access_op(OP_EXIT);
2629 }
2630 
2631 static void ept_access_test_guest(void)
2632 {
2633 	struct ept_access_test_data *data = &ept_access_test_data;
2634 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2635 
2636 	while (true) {
2637 		switch (data->op) {
2638 		case OP_READ:
2639 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2640 			break;
2641 		case OP_WRITE:
2642 			*data->gva = MAGIC_VAL_2;
2643 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2644 			*data->gva = MAGIC_VAL_1;
2645 			break;
2646 		case OP_EXEC:
2647 			TEST_ASSERT_EQ(42, code());
2648 			break;
2649 		case OP_FLUSH_TLB:
2650 			write_cr3(read_cr3());
2651 			break;
2652 		case OP_EXIT:
2653 			return;
2654 		default:
2655 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2656 		}
2657 		vmcall();
2658 	}
2659 }
2660 
2661 static void ept_access_test_setup(void)
2662 {
2663 	struct ept_access_test_data *data = &ept_access_test_data;
2664 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2665 	unsigned long size = npages * PAGE_SIZE;
2666 	unsigned long *page_table = current_page_table();
2667 	unsigned long pte;
2668 
2669 	if (setup_ept(false))
2670 		test_skip("EPT not supported");
2671 
2672 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2673 	if (cpuid_maxphyaddr() < 40)
2674 		test_skip("Test needs MAXPHYADDR >= 40");
2675 
2676 	test_set_guest(ept_access_test_guest);
2677 	test_add_teardown(ept_access_test_teardown, NULL);
2678 
2679 	data->hva = get_1g_page();
2680 	TEST_ASSERT(data->hva);
2681 	data->hpa = virt_to_phys(data->hva);
2682 
2683 	data->gpa = 1ul << 39;
2684 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2685 				   size);
2686 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2687 	install_pages(page_table, data->gpa, size, data->gva);
2688 
2689 	/*
2690 	 * Make sure nothing's mapped here so the tests that screw with the
2691 	 * pml4 entry don't inadvertently break something.
2692 	 */
2693 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2694 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2695 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2696 
2697 	data->hva[0] = MAGIC_VAL_1;
2698 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2699 }
2700 
2701 static void ept_access_test_not_present(void)
2702 {
2703 	ept_access_test_setup();
2704 	/* --- */
2705 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2706 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2707 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2708 }
2709 
2710 static void ept_access_test_read_only(void)
2711 {
2712 	ept_access_test_setup();
2713 
2714 	/* r-- */
2715 	ept_access_allowed(EPT_RA, OP_READ);
2716 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2717 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2718 }
2719 
2720 static void ept_access_test_write_only(void)
2721 {
2722 	ept_access_test_setup();
2723 	/* -w- */
2724 	ept_access_misconfig(EPT_WA);
2725 }
2726 
2727 static void ept_access_test_read_write(void)
2728 {
2729 	ept_access_test_setup();
2730 	/* rw- */
2731 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2732 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2733 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2734 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2735 }
2736 
2737 
2738 static void ept_access_test_execute_only(void)
2739 {
2740 	ept_access_test_setup();
2741 	/* --x */
2742 	if (ept_execute_only_supported()) {
2743 		ept_access_violation(EPT_EA, OP_READ,
2744 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2745 		ept_access_violation(EPT_EA, OP_WRITE,
2746 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2747 		ept_access_allowed(EPT_EA, OP_EXEC);
2748 	} else {
2749 		ept_access_misconfig(EPT_EA);
2750 	}
2751 }
2752 
2753 static void ept_access_test_read_execute(void)
2754 {
2755 	ept_access_test_setup();
2756 	/* r-x */
2757 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2758 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2759 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2760 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2761 }
2762 
2763 static void ept_access_test_write_execute(void)
2764 {
2765 	ept_access_test_setup();
2766 	/* -wx */
2767 	ept_access_misconfig(EPT_WA | EPT_EA);
2768 }
2769 
2770 static void ept_access_test_read_write_execute(void)
2771 {
2772 	ept_access_test_setup();
2773 	/* rwx */
2774 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2775 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2776 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2777 }
2778 
2779 static void ept_access_test_reserved_bits(void)
2780 {
2781 	int i;
2782 	int maxphyaddr;
2783 
2784 	ept_access_test_setup();
2785 
2786 	/* Reserved bits above maxphyaddr. */
2787 	maxphyaddr = cpuid_maxphyaddr();
2788 	for (i = maxphyaddr; i <= 51; i++) {
2789 		report_prefix_pushf("reserved_bit=%d", i);
2790 		ept_reserved_bit(i);
2791 		report_prefix_pop();
2792 	}
2793 
2794 	/* Level-specific reserved bits. */
2795 	ept_reserved_bit_at_level_nohuge(2, 3);
2796 	ept_reserved_bit_at_level_nohuge(2, 4);
2797 	ept_reserved_bit_at_level_nohuge(2, 5);
2798 	ept_reserved_bit_at_level_nohuge(2, 6);
2799 	/* 2M alignment. */
2800 	for (i = 12; i < 20; i++) {
2801 		report_prefix_pushf("reserved_bit=%d", i);
2802 		ept_reserved_bit_at_level_huge(2, i);
2803 		report_prefix_pop();
2804 	}
2805 	ept_reserved_bit_at_level_nohuge(3, 3);
2806 	ept_reserved_bit_at_level_nohuge(3, 4);
2807 	ept_reserved_bit_at_level_nohuge(3, 5);
2808 	ept_reserved_bit_at_level_nohuge(3, 6);
2809 	/* 1G alignment. */
2810 	for (i = 12; i < 29; i++) {
2811 		report_prefix_pushf("reserved_bit=%d", i);
2812 		ept_reserved_bit_at_level_huge(3, i);
2813 		report_prefix_pop();
2814 	}
2815 	ept_reserved_bit_at_level(4, 3);
2816 	ept_reserved_bit_at_level(4, 4);
2817 	ept_reserved_bit_at_level(4, 5);
2818 	ept_reserved_bit_at_level(4, 6);
2819 	ept_reserved_bit_at_level(4, 7);
2820 }
2821 
2822 static void ept_access_test_ignored_bits(void)
2823 {
2824 	ept_access_test_setup();
2825 	/*
2826 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2827 	 * far as translation is concerned even if AD bits are enabled in the
2828 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2829 	 * control is 0.
2830 	 */
2831 	ept_ignored_bit(8);
2832 	ept_ignored_bit(9);
2833 	ept_ignored_bit(10);
2834 	ept_ignored_bit(11);
2835 	ept_ignored_bit(52);
2836 	ept_ignored_bit(53);
2837 	ept_ignored_bit(54);
2838 	ept_ignored_bit(55);
2839 	ept_ignored_bit(56);
2840 	ept_ignored_bit(57);
2841 	ept_ignored_bit(58);
2842 	ept_ignored_bit(59);
2843 	ept_ignored_bit(60);
2844 	ept_ignored_bit(61);
2845 	ept_ignored_bit(62);
2846 	ept_ignored_bit(63);
2847 }
2848 
2849 static void ept_access_test_paddr_not_present_ad_disabled(void)
2850 {
2851 	ept_access_test_setup();
2852 	ept_disable_ad_bits();
2853 
2854 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
2855 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
2856 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
2857 }
2858 
2859 static void ept_access_test_paddr_not_present_ad_enabled(void)
2860 {
2861 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
2862 
2863 	ept_access_test_setup();
2864 	ept_enable_ad_bits_or_skip_test();
2865 
2866 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
2867 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
2868 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
2869 }
2870 
2871 static void ept_access_test_paddr_read_only_ad_disabled(void)
2872 {
2873 	/*
2874 	 * When EPT AD bits are disabled, all accesses to guest paging
2875 	 * structures are reported separately as a read and (after
2876 	 * translation of the GPA to host physical address) a read+write
2877 	 * if the A/D bits have to be set.
2878 	 */
2879 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2880 
2881 	ept_access_test_setup();
2882 	ept_disable_ad_bits();
2883 
2884 	/* Can't update A bit, so all accesses fail. */
2885 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2886 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
2887 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
2888 	/* AD bits disabled, so only writes try to update the D bit. */
2889 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
2890 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
2891 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
2892 	/* Both A and D already set, so read-only is OK. */
2893 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
2894 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
2895 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
2896 }
2897 
2898 static void ept_access_test_paddr_read_only_ad_enabled(void)
2899 {
2900 	/*
2901 	 * When EPT AD bits are enabled, all accesses to guest paging
2902 	 * structures are considered writes as far as EPT translation
2903 	 * is concerned.
2904 	 */
2905 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2906 
2907 	ept_access_test_setup();
2908 	ept_enable_ad_bits_or_skip_test();
2909 
2910 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2911 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
2912 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
2913 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
2914 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
2915 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
2916 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
2917 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
2918 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
2919 }
2920 
2921 static void ept_access_test_paddr_read_write(void)
2922 {
2923 	ept_access_test_setup();
2924 	/* Read-write access to paging structure. */
2925 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
2926 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
2927 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
2928 }
2929 
2930 static void ept_access_test_paddr_read_write_execute(void)
2931 {
2932 	ept_access_test_setup();
2933 	/* RWX access to paging structure. */
2934 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
2935 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
2936 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
2937 }
2938 
2939 static void ept_access_test_paddr_read_execute_ad_disabled(void)
2940 {
2941   	/*
2942 	 * When EPT AD bits are disabled, all accesses to guest paging
2943 	 * structures are reported separately as a read and (after
2944 	 * translation of the GPA to host physical address) a read+write
2945 	 * if the A/D bits have to be set.
2946 	 */
2947 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
2948 
2949 	ept_access_test_setup();
2950 	ept_disable_ad_bits();
2951 
2952 	/* Can't update A bit, so all accesses fail. */
2953 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
2954 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
2955 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
2956 	/* AD bits disabled, so only writes try to update the D bit. */
2957 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
2958 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
2959 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
2960 	/* Both A and D already set, so read-only is OK. */
2961 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
2962 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
2963 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
2964 }
2965 
2966 static void ept_access_test_paddr_read_execute_ad_enabled(void)
2967 {
2968 	/*
2969 	 * When EPT AD bits are enabled, all accesses to guest paging
2970 	 * structures are considered writes as far as EPT translation
2971 	 * is concerned.
2972 	 */
2973 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
2974 
2975 	ept_access_test_setup();
2976 	ept_enable_ad_bits_or_skip_test();
2977 
2978 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
2979 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
2980 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
2981 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
2982 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
2983 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
2984 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
2985 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
2986 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
2987 }
2988 
2989 static void ept_access_test_paddr_not_present_page_fault(void)
2990 {
2991 	ept_access_test_setup();
2992 	/*
2993 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
2994 	 * page is read-only in EPT but GVA is also mapped read only in PT.
2995 	 * Thus guest page fault before host takes EPT violation for trying to
2996 	 * update A bit.
2997 	 */
2998 }
2999 
3000 static void ept_access_test_force_2m_page(void)
3001 {
3002 	ept_access_test_setup();
3003 
3004 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3005 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3006 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3007 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3008 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3009 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3010 }
3011 
3012 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3013 {
3014 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3015 
3016 	TEST_ASSERT(msr & VPID_CAP_INVVPID);
3017 
3018 	if (type < INVVPID_ADDR || type > INVVPID_CONTEXT_LOCAL)
3019 		return false;
3020 
3021 	if (!(msr & (1ull << (type + VPID_CAP_INVVPID_TYPES_SHIFT))))
3022 		return false;
3023 
3024 	if (vpid >> 16)
3025 		return false;
3026 
3027 	if (type != INVVPID_ALL && !vpid)
3028 		return false;
3029 
3030 	if (type == INVVPID_ADDR && !is_canonical(gla))
3031 		return false;
3032 
3033 	return true;
3034 }
3035 
3036 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3037 {
3038 	int rc;
3039 	bool valid = invvpid_valid(type, vpid, gla);
3040 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3041 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3042 	/*
3043 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3044 	 * that we can tell if it is updated by INVVPID.
3045 	 */
3046 	vmcs_read(~0);
3047 	rc = invvpid(type, vpid, gla);
3048 	report("INVVPID type %ld VPID %lx GLA %lx %s",
3049 	       !rc == valid, type, vpid, gla,
3050 	       valid ? "passes" : "fails");
3051 	report("After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3052 	       vmcs_read(VMX_INST_ERROR) == expected,
3053 	       rc ? "failed" : "successful",
3054 	       expected, vmcs_read(VMX_INST_ERROR));
3055 }
3056 
3057 static void ds_invvpid(void *data)
3058 {
3059 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3060 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3061 
3062 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3063 	asm volatile("invvpid %0, %1"
3064 		     :
3065 		     : "m"(*(struct invvpid_operand *)data),
3066 		       "r"(type));
3067 }
3068 
3069 /*
3070  * The SS override is ignored in 64-bit mode, so we use an addressing
3071  * mode with %rsp as the base register to generate an implicit SS
3072  * reference.
3073  */
3074 static void ss_invvpid(void *data)
3075 {
3076 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3077 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3078 
3079 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3080 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3081 		     : "+r"(data)
3082 		     : "r"(type));
3083 }
3084 
3085 static void invvpid_test_gp(void)
3086 {
3087 	bool fault;
3088 
3089 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3090 				   (void *)NONCANONICAL);
3091 	report("INVVPID with non-canonical DS operand raises #GP", fault);
3092 }
3093 
3094 static void invvpid_test_ss(void)
3095 {
3096 	bool fault;
3097 
3098 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3099 				   (void *)NONCANONICAL);
3100 	report("INVVPID with non-canonical SS operand raises #SS", fault);
3101 }
3102 
3103 static void invvpid_test_pf(void)
3104 {
3105 	void *vpage = alloc_vpage();
3106 	bool fault;
3107 
3108 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3109 	report("INVVPID with unmapped operand raises #PF", fault);
3110 }
3111 
3112 static void try_compat_invvpid(void *unused)
3113 {
3114 	struct far_pointer32 fp = {
3115 		.offset = (uintptr_t)&&invvpid,
3116 		.selector = KERNEL_CS32,
3117 	};
3118 	register uintptr_t rsp asm("rsp");
3119 
3120 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3121 			"Code address too high.");
3122 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3123 
3124 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3125 	return;
3126 invvpid:
3127 	asm volatile (".code32;"
3128 		      "invvpid (%eax), %eax;"
3129 		      "lret;"
3130 		      ".code64");
3131 	__builtin_unreachable();
3132 }
3133 
3134 static void invvpid_test_compatibility_mode(void)
3135 {
3136 	bool fault;
3137 
3138 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3139 	report("Compatibility mode INVVPID raises #UD", fault);
3140 }
3141 
3142 static void invvpid_test_not_in_vmx_operation(void)
3143 {
3144 	bool fault;
3145 
3146 	TEST_ASSERT(!vmx_off());
3147 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3148 	report("INVVPID outside of VMX operation raises #UD", fault);
3149 	TEST_ASSERT(!vmx_on());
3150 }
3151 
3152 /*
3153  * This does not test real-address mode, virtual-8086 mode, protected mode,
3154  * or CPL > 0.
3155  */
3156 static void invvpid_test_v2(void)
3157 {
3158 	u64 msr;
3159 	int i;
3160 	unsigned types = 0;
3161 	unsigned type;
3162 
3163 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
3164 	    !(ctrl_cpu_rev[1].clr & CPU_VPID))
3165 		test_skip("VPID not supported");
3166 
3167 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3168 
3169 	if (!(msr & VPID_CAP_INVVPID))
3170 		test_skip("INVVPID not supported.\n");
3171 
3172 	if (msr & VPID_CAP_INVVPID_ADDR)
3173 		types |= 1u << INVVPID_ADDR;
3174 	if (msr & VPID_CAP_INVVPID_CXTGLB)
3175 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3176 	if (msr & VPID_CAP_INVVPID_ALL)
3177 		types |= 1u << INVVPID_ALL;
3178 	if (msr & VPID_CAP_INVVPID_CXTLOC)
3179 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3180 
3181 	if (!types)
3182 		test_skip("No INVVPID types supported.\n");
3183 
3184 	for (i = -127; i < 128; i++)
3185 		try_invvpid(i, 0xffff, 0);
3186 
3187 	/*
3188 	 * VPID must not be more than 16 bits.
3189 	 */
3190 	for (i = 0; i < 64; i++)
3191 		for (type = 0; type < 4; type++)
3192 			if (types & (1u << type))
3193 				try_invvpid(type, 1ul << i, 0);
3194 
3195 	/*
3196 	 * VPID must not be zero, except for "all contexts."
3197 	 */
3198 	for (type = 0; type < 4; type++)
3199 		if (types & (1u << type))
3200 			try_invvpid(type, 0, 0);
3201 
3202 	/*
3203 	 * The gla operand is only validated for single-address INVVPID.
3204 	 */
3205 	if (types & (1u << INVVPID_ADDR))
3206 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3207 
3208 	invvpid_test_gp();
3209 	invvpid_test_ss();
3210 	invvpid_test_pf();
3211 	invvpid_test_compatibility_mode();
3212 	invvpid_test_not_in_vmx_operation();
3213 }
3214 
3215 /*
3216  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3217  * at least as far as the guest-state checks. Returns false if the
3218  * VMLAUNCH fails early and execution falls through to the next
3219  * instruction.
3220  */
3221 static bool vmlaunch_succeeds(void)
3222 {
3223 	/*
3224 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3225 	 * unsupported VMCS component"). The caller can then check
3226 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3227 	 */
3228 	vmcs_write(~0u, 0);
3229 
3230 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3231 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3232 				   :
3233 				   : "r" ((u64)HOST_RSP)
3234 				   : "cc", "memory"
3235 				   : success);
3236 	return false;
3237 success:
3238 	TEST_ASSERT(vmcs_read(EXI_REASON) ==
3239 		    (VMX_FAIL_STATE | VMX_ENTRY_FAILURE));
3240 	return true;
3241 }
3242 
3243 /*
3244  * Try to launch the current VMCS.
3245  */
3246 static void test_vmx_controls(bool controls_valid, bool xfail)
3247 {
3248 	bool success = vmlaunch_succeeds();
3249 	u32 vmx_inst_err;
3250 
3251 	report_xfail("vmlaunch %s", xfail, success == controls_valid,
3252 		     controls_valid ? "succeeds" : "fails");
3253 	if (!success) {
3254 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3255 		report("VMX inst error is %d (actual %d)",
3256 		       vmx_inst_err == VMXERR_ENTRY_INVALID_CONTROL_FIELD,
3257 		       VMXERR_ENTRY_INVALID_CONTROL_FIELD, vmx_inst_err);
3258 	}
3259 }
3260 
3261 /*
3262  * Test a particular value of a VM-execution control bit, if the value
3263  * is required or if the value is zero.
3264  */
3265 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3266 				    enum Encoding encoding, unsigned bit,
3267 				    unsigned val)
3268 {
3269 	u32 mask = 1u << bit;
3270 	bool expected;
3271 	u32 controls;
3272 
3273 	if (msr.set & mask)
3274 		TEST_ASSERT(msr.clr & mask);
3275 
3276 	/*
3277 	 * We can't arbitrarily turn on a control bit, because it may
3278 	 * introduce dependencies on other VMCS fields. So, we only
3279 	 * test turning on bits that have a required setting.
3280 	 */
3281 	if (val && (msr.clr & mask) && !(msr.set & mask))
3282 		return;
3283 
3284 	report_prefix_pushf("%s %s bit %d",
3285 			    val ? "Set" : "Clear", name, bit);
3286 
3287 	controls = vmcs_read(encoding);
3288 	if (val) {
3289 		vmcs_write(encoding, msr.set | mask);
3290 		expected = (msr.clr & mask);
3291 	} else {
3292 		vmcs_write(encoding, msr.set & ~mask);
3293 		expected = !(msr.set & mask);
3294 	}
3295 	test_vmx_controls(expected, false);
3296 	vmcs_write(encoding, controls);
3297 	report_prefix_pop();
3298 }
3299 
3300 /*
3301  * Test reserved values of a VM-execution control bit, based on the
3302  * allowed bit settings from the corresponding VMX capability MSR.
3303  */
3304 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3305 			      enum Encoding encoding, unsigned bit)
3306 {
3307 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3308 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3309 }
3310 
3311 /*
3312  * Reserved bits in the pin-based VM-execution controls must be set
3313  * properly. Software may consult the VMX capability MSRs to determine
3314  * the proper settings.
3315  * [Intel SDM]
3316  */
3317 static void test_pin_based_ctls(void)
3318 {
3319 	unsigned bit;
3320 
3321 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3322 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3323 	for (bit = 0; bit < 32; bit++)
3324 		test_rsvd_ctl_bit("pin-based controls",
3325 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3326 }
3327 
3328 /*
3329  * Reserved bits in the primary processor-based VM-execution controls
3330  * must be set properly. Software may consult the VMX capability MSRs
3331  * to determine the proper settings.
3332  * [Intel SDM]
3333  */
3334 static void test_primary_processor_based_ctls(void)
3335 {
3336 	unsigned bit;
3337 
3338 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3339 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3340 	for (bit = 0; bit < 32; bit++)
3341 		test_rsvd_ctl_bit("primary processor-based controls",
3342 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3343 }
3344 
3345 /*
3346  * If the "activate secondary controls" primary processor-based
3347  * VM-execution control is 1, reserved bits in the secondary
3348  * processor-based VM-execution controls must be cleared. Software may
3349  * consult the VMX capability MSRs to determine which bits are
3350  * reserved.
3351  * If the "activate secondary controls" primary processor-based
3352  * VM-execution control is 0 (or if the processor does not support the
3353  * 1-setting of that control), no checks are performed on the
3354  * secondary processor-based VM-execution controls.
3355  * [Intel SDM]
3356  */
3357 static void test_secondary_processor_based_ctls(void)
3358 {
3359 	u32 primary;
3360 	u32 secondary;
3361 	unsigned bit;
3362 
3363 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3364 		return;
3365 
3366 	primary = vmcs_read(CPU_EXEC_CTRL0);
3367 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3368 
3369 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3370 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3371 	for (bit = 0; bit < 32; bit++)
3372 		test_rsvd_ctl_bit("secondary processor-based controls",
3373 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3374 
3375 	/*
3376 	 * When the "activate secondary controls" VM-execution control
3377 	 * is clear, there are no checks on the secondary controls.
3378 	 */
3379 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3380 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3381 	report("Secondary processor-based controls ignored",
3382 	       vmlaunch_succeeds());
3383 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3384 	vmcs_write(CPU_EXEC_CTRL0, primary);
3385 }
3386 
3387 static void try_cr3_target_count(unsigned i, unsigned max)
3388 {
3389 	report_prefix_pushf("CR3 target count 0x%x", i);
3390 	vmcs_write(CR3_TARGET_COUNT, i);
3391 	test_vmx_controls(i <= max, false);
3392 	report_prefix_pop();
3393 }
3394 
3395 /*
3396  * The CR3-target count must not be greater than 4. Future processors
3397  * may support a different number of CR3-target values. Software
3398  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3399  * number of values supported.
3400  * [Intel SDM]
3401  */
3402 static void test_cr3_targets(void)
3403 {
3404 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3405 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3406 	unsigned i;
3407 
3408 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3409 	TEST_ASSERT(supported_targets <= 256);
3410 
3411 	try_cr3_target_count(-1u, supported_targets);
3412 	try_cr3_target_count(0x80000000, supported_targets);
3413 	try_cr3_target_count(0x7fffffff, supported_targets);
3414 	for (i = 0; i <= supported_targets + 1; i++)
3415 		try_cr3_target_count(i, supported_targets);
3416 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3417 }
3418 
3419 /*
3420  * Test a particular address setting for a physical page reference in
3421  * the VMCS.
3422  */
3423 static void test_vmcs_page_addr(const char *name,
3424 				enum Encoding encoding,
3425 				bool ignored,
3426 				bool xfail_beyond_mapped_ram,
3427 				u64 addr)
3428 {
3429 	bool xfail =
3430 		(xfail_beyond_mapped_ram &&
3431 		 addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - PAGE_SIZE &&
3432 		 addr < (1ul << cpuid_maxphyaddr()));
3433 
3434 	report_prefix_pushf("%s = %lx", name, addr);
3435 	vmcs_write(encoding, addr);
3436 	test_vmx_controls(ignored || (IS_ALIGNED(addr, PAGE_SIZE) &&
3437 				  addr < (1ul << cpuid_maxphyaddr())),
3438 			  xfail);
3439 	report_prefix_pop();
3440 	xfail = false;
3441 }
3442 
3443 /*
3444  * Test interesting values for a physical page reference in the VMCS.
3445  */
3446 static void test_vmcs_page_values(const char *name,
3447 				  enum Encoding encoding,
3448 				  bool ignored,
3449 				  bool xfail_beyond_mapped_ram)
3450 {
3451 	unsigned i;
3452 	u64 orig_val = vmcs_read(encoding);
3453 
3454 	for (i = 0; i < 64; i++)
3455 		test_vmcs_page_addr(name, encoding, ignored,
3456 				    xfail_beyond_mapped_ram, 1ul << i);
3457 
3458 	test_vmcs_page_addr(name, encoding, ignored,
3459 			    xfail_beyond_mapped_ram, PAGE_SIZE - 1);
3460 	test_vmcs_page_addr(name, encoding, ignored,
3461 			    xfail_beyond_mapped_ram, PAGE_SIZE);
3462 	test_vmcs_page_addr(name, encoding, ignored,
3463 			    xfail_beyond_mapped_ram,
3464 			    (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3465 	test_vmcs_page_addr(name, encoding, ignored,
3466 			    xfail_beyond_mapped_ram,
3467 			    -1ul);
3468 
3469 	vmcs_write(encoding, orig_val);
3470 }
3471 
3472 /*
3473  * Test a physical page reference in the VMCS, when the corresponding
3474  * feature is enabled and when the corresponding feature is disabled.
3475  */
3476 static void test_vmcs_page_reference(u32 control_bit, enum Encoding field,
3477 				     const char *field_name,
3478 				     const char *control_name,
3479 				     bool xfail_beyond_mapped_ram,
3480 				     bool control_primary)
3481 {
3482 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3483 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3484 	u64 page_addr;
3485 
3486 	if (control_primary) {
3487 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3488 			return;
3489 	} else {
3490 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3491 			return;
3492 	}
3493 
3494 	page_addr = vmcs_read(field);
3495 
3496 	report_prefix_pushf("%s enabled", control_name);
3497 	if (control_primary) {
3498 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3499 	} else {
3500 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3501 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3502 	}
3503 	test_vmcs_page_values(field_name, field, false, xfail_beyond_mapped_ram);
3504 	report_prefix_pop();
3505 
3506 	report_prefix_pushf("%s disabled", control_name);
3507 	if (control_primary) {
3508 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3509 	} else {
3510 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3511 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3512 	}
3513 	test_vmcs_page_values(field_name, field, true, false);
3514 	report_prefix_pop();
3515 
3516 	vmcs_write(field, page_addr);
3517 	vmcs_write(CPU_EXEC_CTRL0, primary);
3518 }
3519 
3520 /*
3521  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3522  * each I/O-bitmap address must be 0. Neither address should set any
3523  * bits beyond the processor's physical-address width.
3524  * [Intel SDM]
3525  */
3526 static void test_io_bitmaps(void)
3527 {
3528 	test_vmcs_page_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3529 				 "I/O bitmap A", "Use I/O bitmaps", false,
3530 				 true);
3531 	test_vmcs_page_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3532 				 "I/O bitmap B", "Use I/O bitmaps", false,
3533 				 true);
3534 }
3535 
3536 /*
3537  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3538  * the MSR-bitmap address must be 0. The address should not set any
3539  * bits beyond the processor's physical-address width.
3540  * [Intel SDM]
3541  */
3542 static void test_msr_bitmap(void)
3543 {
3544 	test_vmcs_page_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3545 				 "MSR bitmap", "Use MSR bitmaps", false,
3546 				 true);
3547 }
3548 
3549 /*
3550  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3551  * address must satisfy the following checks:
3552  * - Bits 11:0 of the address must be 0.
3553  * - The address should not set any bits beyond the processor's
3554  *   physical-address width.
3555  * [Intel SDM]
3556  */
3557 static void test_apic_virt_addr(void)
3558 {
3559 	test_vmcs_page_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3560 				 "virtual-APIC address", "Use TPR shadow",
3561 				 true, true);
3562 }
3563 
3564 /*
3565  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3566  * APIC-access address must satisfy the following checks:
3567  *  - Bits 11:0 of the address must be 0.
3568  *  - The address should not set any bits beyond the processor's
3569  *    physical-address width.
3570  * [Intel SDM]
3571  */
3572 static void test_apic_access_addr(void)
3573 {
3574 	void *apic_access_page = alloc_page();
3575 
3576 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3577 
3578 	test_vmcs_page_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3579 				 "APIC-access address",
3580 				 "virtualize APIC-accesses", true, false);
3581 }
3582 
3583 static void set_vtpr(unsigned vtpr)
3584 {
3585 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
3586 }
3587 
3588 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
3589 {
3590 	bool valid = true;
3591 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3592 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3593 
3594 	if ((primary & CPU_TPR_SHADOW) &&
3595 	    (!(primary & CPU_SECONDARY) ||
3596 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
3597 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
3598 
3599 	set_vtpr(vtpr);
3600 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
3601 	    threshold, (vtpr >> 4) & 0xf);
3602 	test_vmx_controls(valid, false);
3603 	report_prefix_pop();
3604 }
3605 
3606 /*
3607  * Test interesting vTPR values for a given TPR threshold.
3608  */
3609 static void test_vtpr_values(unsigned threshold)
3610 {
3611 	try_tpr_threshold_and_vtpr(threshold, threshold - 1);
3612 	try_tpr_threshold_and_vtpr(threshold, threshold);
3613 	try_tpr_threshold_and_vtpr(threshold, threshold + 1);
3614 }
3615 
3616 static void try_tpr_threshold(unsigned threshold)
3617 {
3618 	bool valid = true;
3619 
3620 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3621 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3622 
3623 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
3624 	    (secondary & CPU_VINTD)))
3625 		valid = !(threshold >> 4);
3626 
3627 	set_vtpr(-1);
3628 	vmcs_write(TPR_THRESHOLD, threshold);
3629 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
3630 	test_vmx_controls(valid, false);
3631 	report_prefix_pop();
3632 
3633 	if (valid)
3634 		test_vtpr_values(threshold);
3635 }
3636 
3637 /*
3638  * Test interesting TPR threshold values.
3639  */
3640 static void test_tpr_threshold_values(void)
3641 {
3642 	unsigned i;
3643 
3644 	for (i = 0; i < 0x10; i++)
3645 		try_tpr_threshold(i);
3646 	for (i = 4; i < 32; i++)
3647 		try_tpr_threshold(1u << i);
3648 	try_tpr_threshold(-1u);
3649 	try_tpr_threshold(0x7fffffff);
3650 }
3651 
3652 /*
3653  * This test covers the following two VM entry checks:
3654  *
3655  *      i) If the "use TPR shadow" VM-execution control is 1 and the
3656  *         "virtual-interrupt delivery" VM-execution control is 0, bits
3657  *         31:4 of the TPR threshold VM-execution control field must
3658 	   be 0.
3659  *         [Intel SDM]
3660  *
3661  *      ii) If the "use TPR shadow" VM-execution control is 1, the
3662  *          "virtual-interrupt delivery" VM-execution control is 0
3663  *          and the "virtualize APIC accesses" VM-execution control
3664  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
3665  *          control field must not be greater than the value of bits
3666  *          7:4 of VTPR.
3667  *          [Intel SDM]
3668  */
3669 static void test_tpr_threshold(void)
3670 {
3671 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3672 	void *virtual_apic_page;
3673 
3674 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
3675 		return;
3676 
3677 	virtual_apic_page = alloc_page();
3678 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
3679 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
3680 
3681 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
3682 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
3683 	test_tpr_threshold_values();
3684 	report_prefix_pop();
3685 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
3686 	report_prefix_pushf("Use TPR shadow enabled");
3687 	test_tpr_threshold_values();
3688 	report_prefix_pop();
3689 
3690 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
3691 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES))))
3692 		return;
3693 
3694 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3695 
3696 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
3697 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
3698 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
3699 		test_tpr_threshold_values();
3700 		report_prefix_pop();
3701 
3702 		vmcs_write(CPU_EXEC_CTRL0,
3703 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
3704 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
3705 		test_tpr_threshold_values();
3706 		report_prefix_pop();
3707 	}
3708 
3709 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
3710 		vmcs_write(CPU_EXEC_CTRL0,
3711 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
3712 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
3713 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
3714 		test_tpr_threshold_values();
3715 		report_prefix_pop();
3716 
3717 		vmcs_write(CPU_EXEC_CTRL0,
3718 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
3719 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
3720 		test_tpr_threshold_values();
3721 		report_prefix_pop();
3722 	}
3723 
3724 	if ((ctrl_cpu_rev[1].clr &
3725 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
3726 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
3727 		vmcs_write(CPU_EXEC_CTRL0,
3728 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
3729 		vmcs_write(CPU_EXEC_CTRL1,
3730 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
3731 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
3732 		test_tpr_threshold_values();
3733 		report_prefix_pop();
3734 
3735 		vmcs_write(CPU_EXEC_CTRL0,
3736 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
3737 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
3738 		test_tpr_threshold_values();
3739 		report_prefix_pop();
3740 	}
3741 
3742 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3743 	vmcs_write(CPU_EXEC_CTRL0, primary);
3744 }
3745 
3746 /*
3747  * This test verifies the following two vmentry checks:
3748  *
3749  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
3750  *  VM-execution control must be 0.
3751  *  [Intel SDM]
3752  *
3753  *  If the “virtual NMIs” VM-execution control is 0, the “NMI-window
3754  *  exiting” VM-execution control must be 0.
3755  *  [Intel SDM]
3756  */
3757 static void test_nmi_ctrls(void)
3758 {
3759 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
3760 
3761 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
3762 	    (PIN_NMI | PIN_VIRT_NMI)) {
3763 		test_skip("NMI exiting and Virtual NMIs are not supported !");
3764 		return;
3765 	}
3766 
3767 	/* Save the controls so that we can restore them after our tests */
3768 	pin_ctrls = vmcs_read(PIN_CONTROLS);
3769 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
3770 
3771 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
3772 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
3773 
3774 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
3775 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
3776 	test_vmx_controls(true, false);
3777 	report_prefix_pop();
3778 
3779 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
3780 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
3781 	test_vmx_controls(false, false);
3782 	report_prefix_pop();
3783 
3784 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
3785 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
3786 	test_vmx_controls(true, false);
3787 	report_prefix_pop();
3788 
3789 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
3790 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
3791 	test_vmx_controls(true, false);
3792 	report_prefix_pop();
3793 
3794 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
3795 		report_info("NMI-window exiting is not supported, skipping...");
3796 		goto done;
3797 	}
3798 
3799 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
3800 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
3801 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
3802 	test_vmx_controls(false, false);
3803 	report_prefix_pop();
3804 
3805 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
3806 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
3807 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
3808 	test_vmx_controls(true, false);
3809 	report_prefix_pop();
3810 
3811 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
3812 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
3813 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
3814 	test_vmx_controls(true, false);
3815 	report_prefix_pop();
3816 
3817 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
3818 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
3819 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
3820 	test_vmx_controls(true, false);
3821 	report_prefix_pop();
3822 
3823 	/* Restore the controls to their original values */
3824 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
3825 done:
3826 	vmcs_write(PIN_CONTROLS, pin_ctrls);
3827 }
3828 
3829 
3830 /*
3831  * Check that the virtual CPU checks all of the VMX controls as
3832  * documented in the Intel SDM.
3833  */
3834 static void vmx_controls_test(void)
3835 {
3836 	/*
3837 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
3838 	 * fail due to invalid guest state, should we make it that
3839 	 * far.
3840 	 */
3841 	vmcs_write(GUEST_RFLAGS, 0);
3842 
3843 	test_pin_based_ctls();
3844 	test_primary_processor_based_ctls();
3845 	test_secondary_processor_based_ctls();
3846 	test_cr3_targets();
3847 	test_io_bitmaps();
3848 	test_msr_bitmap();
3849 	test_apic_virt_addr();
3850 	test_apic_access_addr();
3851 	test_tpr_threshold();
3852 	test_nmi_ctrls();
3853 }
3854 
3855 static bool valid_vmcs_for_vmentry(void)
3856 {
3857 	struct vmcs *current_vmcs = NULL;
3858 
3859 	if (vmcs_save(&current_vmcs))
3860 		return false;
3861 
3862 	return current_vmcs && !(current_vmcs->revision_id >> 31);
3863 }
3864 
3865 static void try_vmentry_in_movss_shadow(void)
3866 {
3867 	u32 vm_inst_err;
3868 	u32 flags;
3869 	bool early_failure = false;
3870 	u32 expected_flags = X86_EFLAGS_FIXED;
3871 	bool valid_vmcs = valid_vmcs_for_vmentry();
3872 
3873 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
3874 
3875 	/*
3876 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3877 	 * unsupported VMCS component").
3878 	 */
3879 	vmcs_write(~0u, 0);
3880 
3881 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
3882 			      "vmwrite %%rsp, %%rdx;"
3883 			      "mov 0f, %%rax;"
3884 			      "mov %[host_rip], %%edx;"
3885 			      "vmwrite %%rax, %%rdx;"
3886 			      "mov $-1, %%ah;"
3887 			      "sahf;"
3888 			      "mov %%ss, %%ax;"
3889 			      "mov %%ax, %%ss;"
3890 			      "vmlaunch;"
3891 			      "mov $1, %[early_failure];"
3892 			      "0: lahf;"
3893 			      "movzbl %%ah, %[flags]"
3894 			      : [early_failure] "+r" (early_failure),
3895 				[flags] "=&a" (flags)
3896 			      : [host_rsp] "i" (HOST_RSP),
3897 				[host_rip] "i" (HOST_RIP)
3898 			      : "rdx", "cc", "memory");
3899 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
3900 
3901 	report("Early VM-entry failure", early_failure);
3902 	report("RFLAGS[8:0] is %x (actual %x)", flags == expected_flags,
3903 	       expected_flags, flags);
3904 	if (valid_vmcs)
3905 		report("VM-instruction error is %d (actual %d)",
3906 		       vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
3907 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
3908 }
3909 
3910 static void vmentry_movss_shadow_test(void)
3911 {
3912 	struct vmcs *orig_vmcs;
3913 
3914 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
3915 
3916 	/*
3917 	 * Set the launched flag on the current VMCS to verify the correct
3918 	 * error priority, below.
3919 	 */
3920 	test_set_guest(v2_null_test_guest);
3921 	enter_guest();
3922 
3923 	/*
3924 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
3925 	 * fail due to invalid guest state (if we make it that far).
3926 	 */
3927 	vmcs_write(GUEST_RFLAGS, 0);
3928 
3929 	/*
3930 	 * "VM entry with events blocked by MOV SS" takes precedence over
3931 	 * "VMLAUNCH with non-clear VMCS."
3932 	 */
3933 	report_prefix_push("valid current-VMCS");
3934 	try_vmentry_in_movss_shadow();
3935 	report_prefix_pop();
3936 
3937 	/*
3938 	 * VMfailInvalid takes precedence over "VM entry with events
3939 	 * blocked by MOV SS."
3940 	 */
3941 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
3942 	report_prefix_push("no current-VMCS");
3943 	try_vmentry_in_movss_shadow();
3944 	report_prefix_pop();
3945 
3946 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
3947 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
3948 }
3949 
3950 #define X86_FEATURE_PCID       (1 << 17)
3951 #define X86_FEATURE_MCE        (1 << 7)
3952 
3953 static int write_cr4_checking(unsigned long val)
3954 {
3955 	asm volatile(ASM_TRY("1f")
3956 		     "mov %0, %%cr4\n\t"
3957 		     "1:": : "r" (val));
3958 	return exception_vector();
3959 }
3960 
3961 static void vmx_cr_load_test(void)
3962 {
3963 	struct cpuid _cpuid = cpuid(1);
3964 	unsigned long cr4 = read_cr4(), cr3 = read_cr3();
3965 
3966 	if (!(_cpuid.c & X86_FEATURE_PCID)) {
3967 		report_skip("PCID not detected");
3968 		return;
3969 	}
3970 	if (!(_cpuid.d & X86_FEATURE_MCE)) {
3971 		report_skip("MCE not detected");
3972 		return;
3973 	}
3974 
3975 	TEST_ASSERT(!(cr4 & (X86_CR4_PCIDE | X86_CR4_MCE)));
3976 	TEST_ASSERT(!(cr3 & X86_CR3_PCID_MASK));
3977 
3978 	/* Enable PCID for L1. */
3979 	cr4 |= X86_CR4_PCIDE;
3980 	cr3 |= 0x1;
3981 	TEST_ASSERT(!write_cr4_checking(cr4));
3982 	write_cr3(cr3);
3983 
3984 	test_set_guest(v2_null_test_guest);
3985 	vmcs_write(HOST_CR4, cr4);
3986 	vmcs_write(HOST_CR3, cr3);
3987 	enter_guest();
3988 
3989 	/*
3990 	 * No exception is expected.
3991 	 *
3992 	 * NB. KVM loads the last guest write to CR4 into CR4 read
3993 	 *     shadow. In order to trigger an exit to KVM, we can set a
3994 	 *     bit that was zero in the above CR4 write and is owned by
3995 	 *     KVM. We choose to set CR4.MCE, which shall have no side
3996 	 *     effect because normally no guest MCE (e.g., as the result
3997 	 *     of bad memory) would happen during this test.
3998 	 */
3999 	TEST_ASSERT(!write_cr4_checking(cr4 | X86_CR4_MCE));
4000 
4001 	/* Cleanup L1 state: disable PCID. */
4002 	write_cr3(cr3 & ~X86_CR3_PCID_MASK);
4003 	TEST_ASSERT(!write_cr4_checking(cr4 & ~X86_CR4_PCIDE));
4004 }
4005 
4006 static bool cpu_has_apicv(void)
4007 {
4008 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
4009 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
4010 		(ctrl_pin_rev.clr & PIN_POST_INTR));
4011 }
4012 
4013 static void trigger_ioapic_scan_thread(void *data)
4014 {
4015 	/* Wait until other CPU entered L2 */
4016 	while (vmx_get_test_stage() != 1)
4017 		;
4018 
4019 	/* Trigger ioapic scan */
4020 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
4021 	vmx_set_test_stage(2);
4022 }
4023 
4024 static void irq_79_handler_guest(isr_regs_t *regs)
4025 {
4026 	eoi();
4027 
4028 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
4029 	vmcall();
4030 }
4031 
4032 /*
4033  * Constant for num of busy-loop iterations after which
4034  * a timer interrupt should have happened in host
4035  */
4036 #define TIMER_INTERRUPT_DELAY 100000000
4037 
4038 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
4039 {
4040 	handle_irq(0x79, irq_79_handler_guest);
4041 	irq_enable();
4042 
4043 	/* Signal to L1 CPU to trigger ioapic scan */
4044 	vmx_set_test_stage(1);
4045 	/* Wait until L1 CPU to trigger ioapic scan */
4046 	while (vmx_get_test_stage() != 2)
4047 		;
4048 
4049 	/*
4050 	 * Wait for L0 timer interrupt to be raised while we run in L2
4051 	 * such that L0 will process the IOAPIC scan request before
4052 	 * resuming L2
4053 	 */
4054 	delay(TIMER_INTERRUPT_DELAY);
4055 
4056 	asm volatile ("int $0x79");
4057 }
4058 
4059 static void vmx_eoi_bitmap_ioapic_scan_test(void)
4060 {
4061 	void *msr_bitmap;
4062 	void *virtual_apic_page;
4063 
4064 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
4065 		report_skip(__func__);
4066 		return;
4067 	}
4068 
4069 	msr_bitmap = alloc_page();
4070 	virtual_apic_page = alloc_page();
4071 
4072 	u64 cpu_ctrl_0 = CPU_SECONDARY | CPU_TPR_SHADOW | CPU_MSR_BITMAP;
4073 	u64 cpu_ctrl_1 = CPU_VINTD | CPU_VIRT_X2APIC;
4074 
4075 	memset(msr_bitmap, 0x0, PAGE_SIZE);
4076 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
4077 
4078 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
4079 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
4080 
4081 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
4082 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
4083 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
4084 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
4085 
4086 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
4087 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
4088 
4089 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
4090 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
4091 
4092 	/*
4093 	 * Launch L2.
4094 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
4095 	 * In case the reason isn't VMX_VMCALL, the asserion inside
4096 	 * skip_exit_vmcall() will fail.
4097 	 */
4098 	enter_guest();
4099 	skip_exit_vmcall();
4100 
4101 	/* Let L2 finish */
4102 	enter_guest();
4103 	report(__func__, 1);
4104 }
4105 
4106 static void set_irq_line_thread(void *data)
4107 {
4108 	/* Wait until other CPU entered L2 */
4109 	while (vmx_get_test_stage() != 1)
4110 		;
4111 
4112 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
4113 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
4114 	vmx_set_test_stage(2);
4115 }
4116 
4117 static bool irq_78_handler_vmcall_before_eoi;
4118 static void irq_78_handler_guest(isr_regs_t *regs)
4119 {
4120 	set_irq_line(0xf, 0);
4121 	if (irq_78_handler_vmcall_before_eoi)
4122 		vmcall();
4123 	eoi();
4124 	vmcall();
4125 }
4126 
4127 static void vmx_apic_passthrough_guest(void)
4128 {
4129 	handle_irq(0x78, irq_78_handler_guest);
4130 	irq_enable();
4131 
4132 	/* If requested, wait for other CPU to trigger ioapic scan */
4133 	if (vmx_get_test_stage() < 1) {
4134 		vmx_set_test_stage(1);
4135 		while (vmx_get_test_stage() != 2)
4136 			;
4137 	}
4138 
4139 	set_irq_line(0xf, 1);
4140 }
4141 
4142 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
4143 {
4144 	void *msr_bitmap;
4145 
4146 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
4147 		report_skip(__func__);
4148 		return;
4149 	}
4150 
4151 	msr_bitmap = alloc_page();
4152 
4153 	u64 cpu_ctrl_0 = CPU_SECONDARY | CPU_MSR_BITMAP;
4154 	u64 cpu_ctrl_1 = 0;
4155 
4156 	memset(msr_bitmap, 0x0, PAGE_SIZE);
4157 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
4158 
4159 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
4160 
4161 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
4162 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
4163 
4164 	if (set_irq_line_from_thread) {
4165 		irq_78_handler_vmcall_before_eoi = false;
4166 		on_cpu_async(1, set_irq_line_thread, NULL);
4167 	} else {
4168 		irq_78_handler_vmcall_before_eoi = true;
4169 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
4170 		vmx_set_test_stage(2);
4171 	}
4172 	test_set_guest(vmx_apic_passthrough_guest);
4173 
4174 	if (irq_78_handler_vmcall_before_eoi) {
4175 		/* Before EOI remote_irr should still be set */
4176 		enter_guest();
4177 		skip_exit_vmcall();
4178 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
4179 			"IOAPIC pass-through: remote_irr=1 before EOI");
4180 	}
4181 
4182 	/* After EOI remote_irr should be cleared */
4183 	enter_guest();
4184 	skip_exit_vmcall();
4185 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
4186 		"IOAPIC pass-through: remote_irr=0 after EOI");
4187 
4188 	/* Let L2 finish */
4189 	enter_guest();
4190 	report(__func__, 1);
4191 }
4192 
4193 static void vmx_apic_passthrough_test(void)
4194 {
4195 	vmx_apic_passthrough(false);
4196 }
4197 
4198 static void vmx_apic_passthrough_thread_test(void)
4199 {
4200 	vmx_apic_passthrough(true);
4201 }
4202 
4203 #define TEST(name) { #name, .v2 = name }
4204 
4205 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
4206 struct vmx_test vmx_tests[] = {
4207 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
4208 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
4209 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
4210 		preemption_timer_exit_handler, NULL, {0} },
4211 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
4212 		test_ctrl_pat_exit_handler, NULL, {0} },
4213 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
4214 		test_ctrl_efer_exit_handler, NULL, {0} },
4215 	{ "CR shadowing", NULL, cr_shadowing_main,
4216 		cr_shadowing_exit_handler, NULL, {0} },
4217 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
4218 		NULL, {0} },
4219 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
4220 		insn_intercept_exit_handler, NULL, {0} },
4221 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
4222 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
4223 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
4224 	{ "VPID", vpid_init, vpid_main, vpid_exit_handler, NULL, {0} },
4225 	{ "interrupt", interrupt_init, interrupt_main,
4226 		interrupt_exit_handler, NULL, {0} },
4227 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
4228 		NULL, {0} },
4229 	{ "MSR switch", msr_switch_init, msr_switch_main,
4230 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
4231 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
4232 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
4233 		disable_rdtscp_exit_handler, NULL, {0} },
4234 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
4235 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
4236 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
4237 		exit_monitor_from_l2_handler, NULL, {0} },
4238 	/* Basic V2 tests. */
4239 	TEST(v2_null_test),
4240 	TEST(v2_multiple_entries_test),
4241 	TEST(fixture_test_case1),
4242 	TEST(fixture_test_case2),
4243 	/* EPT access tests. */
4244 	TEST(ept_access_test_not_present),
4245 	TEST(ept_access_test_read_only),
4246 	TEST(ept_access_test_write_only),
4247 	TEST(ept_access_test_read_write),
4248 	TEST(ept_access_test_execute_only),
4249 	TEST(ept_access_test_read_execute),
4250 	TEST(ept_access_test_write_execute),
4251 	TEST(ept_access_test_read_write_execute),
4252 	TEST(ept_access_test_reserved_bits),
4253 	TEST(ept_access_test_ignored_bits),
4254 	TEST(ept_access_test_paddr_not_present_ad_disabled),
4255 	TEST(ept_access_test_paddr_not_present_ad_enabled),
4256 	TEST(ept_access_test_paddr_read_only_ad_disabled),
4257 	TEST(ept_access_test_paddr_read_only_ad_enabled),
4258 	TEST(ept_access_test_paddr_read_write),
4259 	TEST(ept_access_test_paddr_read_write_execute),
4260 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
4261 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
4262 	TEST(ept_access_test_paddr_not_present_page_fault),
4263 	TEST(ept_access_test_force_2m_page),
4264 	/* Opcode tests. */
4265 	TEST(invvpid_test_v2),
4266 	/* VM-entry tests */
4267 	TEST(vmx_controls_test),
4268 	TEST(vmentry_movss_shadow_test),
4269 	/* APICv tests */
4270 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
4271 	/* APIC pass-through tests */
4272 	TEST(vmx_apic_passthrough_test),
4273 	TEST(vmx_apic_passthrough_thread_test),
4274 	/* Regression tests */
4275 	TEST(vmx_cr_load_test),
4276 	{ NULL, NULL, NULL, NULL, NULL, {0} },
4277 };
4278