xref: /kvm-unit-tests/x86/vmx_tests.c (revision 47cc3d85c2fe6fe4618e513b39046abe73f2803c)
1 /*
2  * All test cases of nested virtualization should be in this file
3  *
4  * Author : Arthur Chunqi Li <yzt356@gmail.com>
5  */
6 
7 #include <asm/debugreg.h>
8 
9 #include "vmx.h"
10 #include "msr.h"
11 #include "processor.h"
12 #include "vm.h"
13 #include "pci.h"
14 #include "fwcfg.h"
15 #include "isr.h"
16 #include "desc.h"
17 #include "apic.h"
18 #include "types.h"
19 #include "vmalloc.h"
20 #include "alloc_page.h"
21 #include "smp.h"
22 #include "delay.h"
23 
24 #define NONCANONICAL            0xaaaaaaaaaaaaaaaaull
25 
26 #define VPID_CAP_INVVPID_TYPES_SHIFT 40
27 
28 u64 ia32_pat;
29 u64 ia32_efer;
30 void *io_bitmap_a, *io_bitmap_b;
31 u16 ioport;
32 
33 unsigned long *pml4;
34 u64 eptp;
35 void *data_page1, *data_page2;
36 
37 phys_addr_t pci_physaddr;
38 
39 void *pml_log;
40 #define PML_INDEX 512
41 
42 static inline unsigned ffs(unsigned x)
43 {
44 	int pos = -1;
45 
46 	__asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0"
47 			     : "+r"(pos) : "rm"(x) : "eax");
48 	return pos + 1;
49 }
50 
51 static inline void vmcall(void)
52 {
53 	asm volatile("vmcall");
54 }
55 
56 static void basic_guest_main(void)
57 {
58 	report("Basic VMX test", 1);
59 }
60 
61 static int basic_exit_handler(void)
62 {
63 	report("Basic VMX test", 0);
64 	print_vmexit_info();
65 	return VMX_TEST_EXIT;
66 }
67 
68 static void vmenter_main(void)
69 {
70 	u64 rax;
71 	u64 rsp, resume_rsp;
72 
73 	report("test vmlaunch", 1);
74 
75 	asm volatile(
76 		"mov %%rsp, %0\n\t"
77 		"mov %3, %%rax\n\t"
78 		"vmcall\n\t"
79 		"mov %%rax, %1\n\t"
80 		"mov %%rsp, %2\n\t"
81 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
82 		: "g"(0xABCD));
83 	report("test vmresume", (rax == 0xFFFF) && (rsp == resume_rsp));
84 }
85 
86 static int vmenter_exit_handler(void)
87 {
88 	u64 guest_rip;
89 	ulong reason;
90 
91 	guest_rip = vmcs_read(GUEST_RIP);
92 	reason = vmcs_read(EXI_REASON) & 0xff;
93 	switch (reason) {
94 	case VMX_VMCALL:
95 		if (regs.rax != 0xABCD) {
96 			report("test vmresume", 0);
97 			return VMX_TEST_VMEXIT;
98 		}
99 		regs.rax = 0xFFFF;
100 		vmcs_write(GUEST_RIP, guest_rip + 3);
101 		return VMX_TEST_RESUME;
102 	default:
103 		report("test vmresume", 0);
104 		print_vmexit_info();
105 	}
106 	return VMX_TEST_VMEXIT;
107 }
108 
109 u32 preempt_scale;
110 volatile unsigned long long tsc_val;
111 volatile u32 preempt_val;
112 u64 saved_rip;
113 
114 static int preemption_timer_init(struct vmcs *vmcs)
115 {
116 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
117 		printf("\tPreemption timer is not supported\n");
118 		return VMX_TEST_EXIT;
119 	}
120 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
121 	preempt_val = 10000000;
122 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
123 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
124 
125 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
126 		printf("\tSave preemption value is not supported\n");
127 
128 	return VMX_TEST_START;
129 }
130 
131 static void preemption_timer_main(void)
132 {
133 	tsc_val = rdtsc();
134 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
135 		vmx_set_test_stage(0);
136 		vmcall();
137 		if (vmx_get_test_stage() == 1)
138 			vmcall();
139 	}
140 	vmx_set_test_stage(1);
141 	while (vmx_get_test_stage() == 1) {
142 		if (((rdtsc() - tsc_val) >> preempt_scale)
143 				> 10 * preempt_val) {
144 			vmx_set_test_stage(2);
145 			vmcall();
146 		}
147 	}
148 	tsc_val = rdtsc();
149 	asm volatile ("hlt");
150 	vmcall();
151 	vmx_set_test_stage(5);
152 	vmcall();
153 }
154 
155 static int preemption_timer_exit_handler(void)
156 {
157 	bool guest_halted;
158 	u64 guest_rip;
159 	ulong reason;
160 	u32 insn_len;
161 	u32 ctrl_exit;
162 
163 	guest_rip = vmcs_read(GUEST_RIP);
164 	reason = vmcs_read(EXI_REASON) & 0xff;
165 	insn_len = vmcs_read(EXI_INST_LEN);
166 	switch (reason) {
167 	case VMX_PREEMPT:
168 		switch (vmx_get_test_stage()) {
169 		case 1:
170 		case 2:
171 			report("busy-wait for preemption timer",
172 			       ((rdtsc() - tsc_val) >> preempt_scale) >=
173 			       preempt_val);
174 			vmx_set_test_stage(3);
175 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
176 			return VMX_TEST_RESUME;
177 		case 3:
178 			guest_halted =
179 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
180 			report("preemption timer during hlt",
181 			       ((rdtsc() - tsc_val) >> preempt_scale) >=
182 			       preempt_val && guest_halted);
183 			vmx_set_test_stage(4);
184 			vmcs_write(PIN_CONTROLS,
185 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
186 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
187 			return VMX_TEST_RESUME;
188 		case 4:
189 			report("preemption timer with 0 value",
190 			       saved_rip == guest_rip);
191 			break;
192 		default:
193 			report("Invalid stage.", false);
194 			print_vmexit_info();
195 			break;
196 		}
197 		break;
198 	case VMX_VMCALL:
199 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
200 		switch (vmx_get_test_stage()) {
201 		case 0:
202 			report("Keep preemption value",
203 			       vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val);
204 			vmx_set_test_stage(1);
205 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
206 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
207 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
208 			vmcs_write(EXI_CONTROLS, ctrl_exit);
209 			return VMX_TEST_RESUME;
210 		case 1:
211 			report("Save preemption value",
212 			       vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val);
213 			return VMX_TEST_RESUME;
214 		case 2:
215 			report("busy-wait for preemption timer", 0);
216 			vmx_set_test_stage(3);
217 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
218 			return VMX_TEST_RESUME;
219 		case 3:
220 			report("preemption timer during hlt", 0);
221 			vmx_set_test_stage(4);
222 			/* fall through */
223 		case 4:
224 			vmcs_write(PIN_CONTROLS,
225 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
226 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
227 			saved_rip = guest_rip + insn_len;
228 			return VMX_TEST_RESUME;
229 		case 5:
230 			report("preemption timer with 0 value (vmcall stage 5)", 0);
231 			break;
232 		default:
233 			// Should not reach here
234 			report("unexpected stage, %d", false,
235 			       vmx_get_test_stage());
236 			print_vmexit_info();
237 			return VMX_TEST_VMEXIT;
238 		}
239 		break;
240 	default:
241 		report("Unknown exit reason, %ld", false, reason);
242 		print_vmexit_info();
243 	}
244 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
245 	return VMX_TEST_VMEXIT;
246 }
247 
248 static void msr_bmp_init(void)
249 {
250 	void *msr_bitmap;
251 	u32 ctrl_cpu0;
252 
253 	msr_bitmap = alloc_page();
254 	memset(msr_bitmap, 0x0, PAGE_SIZE);
255 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
256 	ctrl_cpu0 |= CPU_MSR_BITMAP;
257 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
258 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
259 }
260 
261 static void *get_msr_bitmap(void)
262 {
263 	void *msr_bitmap;
264 
265 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
266 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
267 	} else {
268 		msr_bitmap = alloc_page();
269 		memset(msr_bitmap, 0xff, PAGE_SIZE);
270 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
271 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
272 	}
273 
274 	return msr_bitmap;
275 }
276 
277 static void disable_intercept_for_x2apic_msrs(void)
278 {
279 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
280 	u32 msr;
281 
282 	for (msr = APIC_BASE_MSR;
283 		 msr < (APIC_BASE_MSR+0xff);
284 		 msr += BITS_PER_LONG) {
285 		unsigned int word = msr / BITS_PER_LONG;
286 
287 		msr_bitmap[word] = 0;
288 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
289 	}
290 }
291 
292 static int test_ctrl_pat_init(struct vmcs *vmcs)
293 {
294 	u64 ctrl_ent;
295 	u64 ctrl_exi;
296 
297 	msr_bmp_init();
298 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
299 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
300 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
301 		printf("\tSave/load PAT is not supported\n");
302 		return 1;
303 	}
304 
305 	ctrl_ent = vmcs_read(ENT_CONTROLS);
306 	ctrl_exi = vmcs_read(EXI_CONTROLS);
307 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
308 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
309 	vmcs_write(ENT_CONTROLS, ctrl_ent);
310 	vmcs_write(EXI_CONTROLS, ctrl_exi);
311 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
312 	vmcs_write(GUEST_PAT, 0x0);
313 	vmcs_write(HOST_PAT, ia32_pat);
314 	return VMX_TEST_START;
315 }
316 
317 static void test_ctrl_pat_main(void)
318 {
319 	u64 guest_ia32_pat;
320 
321 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
322 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
323 		printf("\tENT_LOAD_PAT is not supported.\n");
324 	else {
325 		if (guest_ia32_pat != 0) {
326 			report("Entry load PAT", 0);
327 			return;
328 		}
329 	}
330 	wrmsr(MSR_IA32_CR_PAT, 0x6);
331 	vmcall();
332 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
333 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
334 		report("Entry load PAT", guest_ia32_pat == ia32_pat);
335 }
336 
337 static int test_ctrl_pat_exit_handler(void)
338 {
339 	u64 guest_rip;
340 	ulong reason;
341 	u64 guest_pat;
342 
343 	guest_rip = vmcs_read(GUEST_RIP);
344 	reason = vmcs_read(EXI_REASON) & 0xff;
345 	switch (reason) {
346 	case VMX_VMCALL:
347 		guest_pat = vmcs_read(GUEST_PAT);
348 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
349 			printf("\tEXI_SAVE_PAT is not supported\n");
350 			vmcs_write(GUEST_PAT, 0x6);
351 		} else {
352 			report("Exit save PAT", guest_pat == 0x6);
353 		}
354 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
355 			printf("\tEXI_LOAD_PAT is not supported\n");
356 		else
357 			report("Exit load PAT", rdmsr(MSR_IA32_CR_PAT) == ia32_pat);
358 		vmcs_write(GUEST_PAT, ia32_pat);
359 		vmcs_write(GUEST_RIP, guest_rip + 3);
360 		return VMX_TEST_RESUME;
361 	default:
362 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
363 		break;
364 	}
365 	return VMX_TEST_VMEXIT;
366 }
367 
368 static int test_ctrl_efer_init(struct vmcs *vmcs)
369 {
370 	u64 ctrl_ent;
371 	u64 ctrl_exi;
372 
373 	msr_bmp_init();
374 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
375 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
376 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
377 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
378 	ia32_efer = rdmsr(MSR_EFER);
379 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
380 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
381 	return VMX_TEST_START;
382 }
383 
384 static void test_ctrl_efer_main(void)
385 {
386 	u64 guest_ia32_efer;
387 
388 	guest_ia32_efer = rdmsr(MSR_EFER);
389 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
390 		printf("\tENT_LOAD_EFER is not supported.\n");
391 	else {
392 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
393 			report("Entry load EFER", 0);
394 			return;
395 		}
396 	}
397 	wrmsr(MSR_EFER, ia32_efer);
398 	vmcall();
399 	guest_ia32_efer = rdmsr(MSR_EFER);
400 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
401 		report("Entry load EFER", guest_ia32_efer == ia32_efer);
402 }
403 
404 static int test_ctrl_efer_exit_handler(void)
405 {
406 	u64 guest_rip;
407 	ulong reason;
408 	u64 guest_efer;
409 
410 	guest_rip = vmcs_read(GUEST_RIP);
411 	reason = vmcs_read(EXI_REASON) & 0xff;
412 	switch (reason) {
413 	case VMX_VMCALL:
414 		guest_efer = vmcs_read(GUEST_EFER);
415 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
416 			printf("\tEXI_SAVE_EFER is not supported\n");
417 			vmcs_write(GUEST_EFER, ia32_efer);
418 		} else {
419 			report("Exit save EFER", guest_efer == ia32_efer);
420 		}
421 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
422 			printf("\tEXI_LOAD_EFER is not supported\n");
423 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
424 		} else {
425 			report("Exit load EFER", rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX));
426 		}
427 		vmcs_write(GUEST_PAT, ia32_efer);
428 		vmcs_write(GUEST_RIP, guest_rip + 3);
429 		return VMX_TEST_RESUME;
430 	default:
431 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
432 		break;
433 	}
434 	return VMX_TEST_VMEXIT;
435 }
436 
437 u32 guest_cr0, guest_cr4;
438 
439 static void cr_shadowing_main(void)
440 {
441 	u32 cr0, cr4, tmp;
442 
443 	// Test read through
444 	vmx_set_test_stage(0);
445 	guest_cr0 = read_cr0();
446 	if (vmx_get_test_stage() == 1)
447 		report("Read through CR0", 0);
448 	else
449 		vmcall();
450 	vmx_set_test_stage(1);
451 	guest_cr4 = read_cr4();
452 	if (vmx_get_test_stage() == 2)
453 		report("Read through CR4", 0);
454 	else
455 		vmcall();
456 	// Test write through
457 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
458 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
459 	vmx_set_test_stage(2);
460 	write_cr0(guest_cr0);
461 	if (vmx_get_test_stage() == 3)
462 		report("Write throuth CR0", 0);
463 	else
464 		vmcall();
465 	vmx_set_test_stage(3);
466 	write_cr4(guest_cr4);
467 	if (vmx_get_test_stage() == 4)
468 		report("Write through CR4", 0);
469 	else
470 		vmcall();
471 	// Test read shadow
472 	vmx_set_test_stage(4);
473 	vmcall();
474 	cr0 = read_cr0();
475 	if (vmx_get_test_stage() != 5)
476 		report("Read shadowing CR0", cr0 == guest_cr0);
477 	vmx_set_test_stage(5);
478 	cr4 = read_cr4();
479 	if (vmx_get_test_stage() != 6)
480 		report("Read shadowing CR4", cr4 == guest_cr4);
481 	// Test write shadow (same value with shadow)
482 	vmx_set_test_stage(6);
483 	write_cr0(guest_cr0);
484 	if (vmx_get_test_stage() == 7)
485 		report("Write shadowing CR0 (same value with shadow)", 0);
486 	else
487 		vmcall();
488 	vmx_set_test_stage(7);
489 	write_cr4(guest_cr4);
490 	if (vmx_get_test_stage() == 8)
491 		report("Write shadowing CR4 (same value with shadow)", 0);
492 	else
493 		vmcall();
494 	// Test write shadow (different value)
495 	vmx_set_test_stage(8);
496 	tmp = guest_cr0 ^ X86_CR0_TS;
497 	asm volatile("mov %0, %%rsi\n\t"
498 		"mov %%rsi, %%cr0\n\t"
499 		::"m"(tmp)
500 		:"rsi", "memory", "cc");
501 	report("Write shadowing different X86_CR0_TS", vmx_get_test_stage() == 9);
502 	vmx_set_test_stage(9);
503 	tmp = guest_cr0 ^ X86_CR0_MP;
504 	asm volatile("mov %0, %%rsi\n\t"
505 		"mov %%rsi, %%cr0\n\t"
506 		::"m"(tmp)
507 		:"rsi", "memory", "cc");
508 	report("Write shadowing different X86_CR0_MP", vmx_get_test_stage() == 10);
509 	vmx_set_test_stage(10);
510 	tmp = guest_cr4 ^ X86_CR4_TSD;
511 	asm volatile("mov %0, %%rsi\n\t"
512 		"mov %%rsi, %%cr4\n\t"
513 		::"m"(tmp)
514 		:"rsi", "memory", "cc");
515 	report("Write shadowing different X86_CR4_TSD", vmx_get_test_stage() == 11);
516 	vmx_set_test_stage(11);
517 	tmp = guest_cr4 ^ X86_CR4_DE;
518 	asm volatile("mov %0, %%rsi\n\t"
519 		"mov %%rsi, %%cr4\n\t"
520 		::"m"(tmp)
521 		:"rsi", "memory", "cc");
522 	report("Write shadowing different X86_CR4_DE", vmx_get_test_stage() == 12);
523 }
524 
525 static int cr_shadowing_exit_handler(void)
526 {
527 	u64 guest_rip;
528 	ulong reason;
529 	u32 insn_len;
530 	u32 exit_qual;
531 
532 	guest_rip = vmcs_read(GUEST_RIP);
533 	reason = vmcs_read(EXI_REASON) & 0xff;
534 	insn_len = vmcs_read(EXI_INST_LEN);
535 	exit_qual = vmcs_read(EXI_QUALIFICATION);
536 	switch (reason) {
537 	case VMX_VMCALL:
538 		switch (vmx_get_test_stage()) {
539 		case 0:
540 			report("Read through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
541 			break;
542 		case 1:
543 			report("Read through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
544 			break;
545 		case 2:
546 			report("Write through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
547 			break;
548 		case 3:
549 			report("Write through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
550 			break;
551 		case 4:
552 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
553 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
554 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
555 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
556 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
557 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
558 			break;
559 		case 6:
560 			report("Write shadowing CR0 (same value)",
561 					guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)));
562 			break;
563 		case 7:
564 			report("Write shadowing CR4 (same value)",
565 					guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)));
566 			break;
567 		default:
568 			// Should not reach here
569 			report("unexpected stage, %d", false,
570 			       vmx_get_test_stage());
571 			print_vmexit_info();
572 			return VMX_TEST_VMEXIT;
573 		}
574 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
575 		return VMX_TEST_RESUME;
576 	case VMX_CR:
577 		switch (vmx_get_test_stage()) {
578 		case 4:
579 			report("Read shadowing CR0", 0);
580 			vmx_inc_test_stage();
581 			break;
582 		case 5:
583 			report("Read shadowing CR4", 0);
584 			vmx_inc_test_stage();
585 			break;
586 		case 6:
587 			report("Write shadowing CR0 (same value)", 0);
588 			vmx_inc_test_stage();
589 			break;
590 		case 7:
591 			report("Write shadowing CR4 (same value)", 0);
592 			vmx_inc_test_stage();
593 			break;
594 		case 8:
595 		case 9:
596 			// 0x600 encodes "mov %esi, %cr0"
597 			if (exit_qual == 0x600)
598 				vmx_inc_test_stage();
599 			break;
600 		case 10:
601 		case 11:
602 			// 0x604 encodes "mov %esi, %cr4"
603 			if (exit_qual == 0x604)
604 				vmx_inc_test_stage();
605 			break;
606 		default:
607 			// Should not reach here
608 			report("unexpected stage, %d", false,
609 			       vmx_get_test_stage());
610 			print_vmexit_info();
611 			return VMX_TEST_VMEXIT;
612 		}
613 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
614 		return VMX_TEST_RESUME;
615 	default:
616 		report("Unknown exit reason, %ld", false, reason);
617 		print_vmexit_info();
618 	}
619 	return VMX_TEST_VMEXIT;
620 }
621 
622 static int iobmp_init(struct vmcs *vmcs)
623 {
624 	u32 ctrl_cpu0;
625 
626 	io_bitmap_a = alloc_page();
627 	io_bitmap_b = alloc_page();
628 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
629 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
630 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
631 	ctrl_cpu0 |= CPU_IO_BITMAP;
632 	ctrl_cpu0 &= (~CPU_IO);
633 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
634 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
635 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
636 	return VMX_TEST_START;
637 }
638 
639 static void iobmp_main(void)
640 {
641 	// stage 0, test IO pass
642 	vmx_set_test_stage(0);
643 	inb(0x5000);
644 	outb(0x0, 0x5000);
645 	report("I/O bitmap - I/O pass", vmx_get_test_stage() == 0);
646 	// test IO width, in/out
647 	((u8 *)io_bitmap_a)[0] = 0xFF;
648 	vmx_set_test_stage(2);
649 	inb(0x0);
650 	report("I/O bitmap - trap in", vmx_get_test_stage() == 3);
651 	vmx_set_test_stage(3);
652 	outw(0x0, 0x0);
653 	report("I/O bitmap - trap out", vmx_get_test_stage() == 4);
654 	vmx_set_test_stage(4);
655 	inl(0x0);
656 	report("I/O bitmap - I/O width, long", vmx_get_test_stage() == 5);
657 	// test low/high IO port
658 	vmx_set_test_stage(5);
659 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
660 	inb(0x5000);
661 	report("I/O bitmap - I/O port, low part", vmx_get_test_stage() == 6);
662 	vmx_set_test_stage(6);
663 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
664 	inb(0x9000);
665 	report("I/O bitmap - I/O port, high part", vmx_get_test_stage() == 7);
666 	// test partial pass
667 	vmx_set_test_stage(7);
668 	inl(0x4FFF);
669 	report("I/O bitmap - partial pass", vmx_get_test_stage() == 8);
670 	// test overrun
671 	vmx_set_test_stage(8);
672 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
673 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
674 	inl(0xFFFF);
675 	report("I/O bitmap - overrun", vmx_get_test_stage() == 9);
676 	vmx_set_test_stage(9);
677 	vmcall();
678 	outb(0x0, 0x0);
679 	report("I/O bitmap - ignore unconditional exiting",
680 	       vmx_get_test_stage() == 9);
681 	vmx_set_test_stage(10);
682 	vmcall();
683 	outb(0x0, 0x0);
684 	report("I/O bitmap - unconditional exiting",
685 	       vmx_get_test_stage() == 11);
686 }
687 
688 static int iobmp_exit_handler(void)
689 {
690 	u64 guest_rip;
691 	ulong reason, exit_qual;
692 	u32 insn_len, ctrl_cpu0;
693 
694 	guest_rip = vmcs_read(GUEST_RIP);
695 	reason = vmcs_read(EXI_REASON) & 0xff;
696 	exit_qual = vmcs_read(EXI_QUALIFICATION);
697 	insn_len = vmcs_read(EXI_INST_LEN);
698 	switch (reason) {
699 	case VMX_IO:
700 		switch (vmx_get_test_stage()) {
701 		case 0:
702 		case 1:
703 			vmx_inc_test_stage();
704 			break;
705 		case 2:
706 			report("I/O bitmap - I/O width, byte",
707 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE);
708 			report("I/O bitmap - I/O direction, in", exit_qual & VMX_IO_IN);
709 			vmx_inc_test_stage();
710 			break;
711 		case 3:
712 			report("I/O bitmap - I/O width, word",
713 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD);
714 			report("I/O bitmap - I/O direction, out",
715 					!(exit_qual & VMX_IO_IN));
716 			vmx_inc_test_stage();
717 			break;
718 		case 4:
719 			report("I/O bitmap - I/O width, long",
720 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG);
721 			vmx_inc_test_stage();
722 			break;
723 		case 5:
724 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
725 				vmx_inc_test_stage();
726 			break;
727 		case 6:
728 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
729 				vmx_inc_test_stage();
730 			break;
731 		case 7:
732 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
733 				vmx_inc_test_stage();
734 			break;
735 		case 8:
736 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
737 				vmx_inc_test_stage();
738 			break;
739 		case 9:
740 		case 10:
741 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
742 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
743 			vmx_inc_test_stage();
744 			break;
745 		default:
746 			// Should not reach here
747 			report("unexpected stage, %d", false,
748 			       vmx_get_test_stage());
749 			print_vmexit_info();
750 			return VMX_TEST_VMEXIT;
751 		}
752 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
753 		return VMX_TEST_RESUME;
754 	case VMX_VMCALL:
755 		switch (vmx_get_test_stage()) {
756 		case 9:
757 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
758 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
759 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
760 			break;
761 		case 10:
762 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
763 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
764 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
765 			break;
766 		default:
767 			// Should not reach here
768 			report("unexpected stage, %d", false,
769 			       vmx_get_test_stage());
770 			print_vmexit_info();
771 			return VMX_TEST_VMEXIT;
772 		}
773 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
774 		return VMX_TEST_RESUME;
775 	default:
776 		printf("guest_rip = %#lx\n", guest_rip);
777 		printf("\tERROR : Undefined exit reason, reason = %ld.\n", reason);
778 		break;
779 	}
780 	return VMX_TEST_VMEXIT;
781 }
782 
783 #define INSN_CPU0		0
784 #define INSN_CPU1		1
785 #define INSN_ALWAYS_TRAP	2
786 
787 #define FIELD_EXIT_QUAL		(1 << 0)
788 #define FIELD_INSN_INFO		(1 << 1)
789 
790 asm(
791 	"insn_hlt: hlt;ret\n\t"
792 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
793 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
794 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
795 	"insn_rdtsc: rdtsc;ret\n\t"
796 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
797 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
798 #ifdef __x86_64__
799 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
800 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
801 #endif
802 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
803 	"insn_pause: pause;ret\n\t"
804 	"insn_wbinvd: wbinvd;ret\n\t"
805 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
806 	"insn_invd: invd;ret\n\t"
807 	"insn_sgdt: sgdt gdt64_desc;ret\n\t"
808 	"insn_lgdt: lgdt gdt64_desc;ret\n\t"
809 	"insn_sidt: sidt idt_descr;ret\n\t"
810 	"insn_lidt: lidt idt_descr;ret\n\t"
811 	"insn_sldt: sldt %ax;ret\n\t"
812 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
813 	"insn_str: str %ax;ret\n\t"
814 	"insn_rdrand: rdrand %rax;ret\n\t"
815 	"insn_rdseed: rdseed %rax;ret\n\t"
816 );
817 extern void insn_hlt(void);
818 extern void insn_invlpg(void);
819 extern void insn_mwait(void);
820 extern void insn_rdpmc(void);
821 extern void insn_rdtsc(void);
822 extern void insn_cr3_load(void);
823 extern void insn_cr3_store(void);
824 #ifdef __x86_64__
825 extern void insn_cr8_load(void);
826 extern void insn_cr8_store(void);
827 #endif
828 extern void insn_monitor(void);
829 extern void insn_pause(void);
830 extern void insn_wbinvd(void);
831 extern void insn_sgdt(void);
832 extern void insn_lgdt(void);
833 extern void insn_sidt(void);
834 extern void insn_lidt(void);
835 extern void insn_sldt(void);
836 extern void insn_lldt(void);
837 extern void insn_str(void);
838 extern void insn_cpuid(void);
839 extern void insn_invd(void);
840 extern void insn_rdrand(void);
841 extern void insn_rdseed(void);
842 
843 u32 cur_insn;
844 u64 cr3;
845 
846 struct insn_table {
847 	const char *name;
848 	u32 flag;
849 	void (*insn_func)(void);
850 	u32 type;
851 	u32 reason;
852 	ulong exit_qual;
853 	u32 insn_info;
854 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
855 	// which field need to be tested, reason is always tested
856 	u32 test_field;
857 };
858 
859 /*
860  * Add more test cases of instruction intercept here. Elements in this
861  * table is:
862  *	name/control flag/insn function/type/exit reason/exit qulification/
863  *	instruction info/field to test
864  * The last field defines which fields (exit_qual and insn_info) need to be
865  * tested in exit handler. If set to 0, only "reason" is checked.
866  */
867 static struct insn_table insn_table[] = {
868 	// Flags for Primary Processor-Based VM-Execution Controls
869 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
870 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
871 		0x12345678, 0, FIELD_EXIT_QUAL},
872 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0},
873 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0},
874 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
875 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
876 		FIELD_EXIT_QUAL},
877 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
878 		FIELD_EXIT_QUAL},
879 #ifdef __x86_64__
880 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
881 		FIELD_EXIT_QUAL},
882 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
883 		FIELD_EXIT_QUAL},
884 #endif
885 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0},
886 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
887 	// Flags for Secondary Processor-Based VM-Execution Controls
888 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
889 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
890 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
891 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
892 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
893 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
894 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
895 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
896 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
897 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
898 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
899 	// Instructions always trap
900 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
901 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
902 	// Instructions never trap
903 	{NULL},
904 };
905 
906 static int insn_intercept_init(struct vmcs *vmcs)
907 {
908 	u32 ctrl_cpu;
909 
910 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
911 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
912 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
913 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
914 	cr3 = read_cr3();
915 	return VMX_TEST_START;
916 }
917 
918 static void insn_intercept_main(void)
919 {
920 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
921 		vmx_set_test_stage(cur_insn * 2);
922 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
923 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
924 		    (insn_table[cur_insn].type == INSN_CPU1 &&
925 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
926 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
927 			       insn_table[cur_insn].type - INSN_CPU0,
928 			       insn_table[cur_insn].name);
929 			continue;
930 		}
931 
932 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
933 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
934 		    (insn_table[cur_insn].type == INSN_CPU1 &&
935 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
936 			/* skip hlt, it stalls the guest and is tested below */
937 			if (insn_table[cur_insn].insn_func != insn_hlt)
938 				insn_table[cur_insn].insn_func();
939 			report("execute %s", vmx_get_test_stage() == cur_insn * 2,
940 					insn_table[cur_insn].name);
941 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
942 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
943 			       insn_table[cur_insn].type - INSN_CPU0,
944 			       insn_table[cur_insn].name);
945 
946 		vmcall();
947 
948 		insn_table[cur_insn].insn_func();
949 		report("intercept %s", vmx_get_test_stage() == cur_insn * 2 + 1,
950 				insn_table[cur_insn].name);
951 
952 		vmx_set_test_stage(cur_insn * 2 + 1);
953 		vmcall();
954 	}
955 }
956 
957 static int insn_intercept_exit_handler(void)
958 {
959 	u64 guest_rip;
960 	u32 reason;
961 	ulong exit_qual;
962 	u32 insn_len;
963 	u32 insn_info;
964 	bool pass;
965 
966 	guest_rip = vmcs_read(GUEST_RIP);
967 	reason = vmcs_read(EXI_REASON) & 0xff;
968 	exit_qual = vmcs_read(EXI_QUALIFICATION);
969 	insn_len = vmcs_read(EXI_INST_LEN);
970 	insn_info = vmcs_read(EXI_INST_INFO);
971 
972 	if (reason == VMX_VMCALL) {
973 		u32 val = 0;
974 
975 		if (insn_table[cur_insn].type == INSN_CPU0)
976 			val = vmcs_read(CPU_EXEC_CTRL0);
977 		else if (insn_table[cur_insn].type == INSN_CPU1)
978 			val = vmcs_read(CPU_EXEC_CTRL1);
979 
980 		if (vmx_get_test_stage() & 1)
981 			val &= ~insn_table[cur_insn].flag;
982 		else
983 			val |= insn_table[cur_insn].flag;
984 
985 		if (insn_table[cur_insn].type == INSN_CPU0)
986 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
987 		else if (insn_table[cur_insn].type == INSN_CPU1)
988 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
989 	} else {
990 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
991 			insn_table[cur_insn].reason == reason;
992 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
993 		    insn_table[cur_insn].exit_qual != exit_qual)
994 			pass = false;
995 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
996 		    insn_table[cur_insn].insn_info != insn_info)
997 			pass = false;
998 		if (pass)
999 			vmx_inc_test_stage();
1000 	}
1001 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1002 	return VMX_TEST_RESUME;
1003 }
1004 
1005 
1006 /* Enables EPT and sets up the identity map. */
1007 static int setup_ept(bool enable_ad)
1008 {
1009 	unsigned long end_of_memory;
1010 	u32 ctrl_cpu[2];
1011 
1012 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1013 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1014 		printf("\tEPT is not supported");
1015 		return 1;
1016 	}
1017 
1018 
1019 	if (!(ept_vpid.val & EPT_CAP_UC) &&
1020 			!(ept_vpid.val & EPT_CAP_WB)) {
1021 		printf("\tEPT paging-structure memory type "
1022 				"UC&WB are not supported\n");
1023 		return 1;
1024 	}
1025 	if (ept_vpid.val & EPT_CAP_UC)
1026 		eptp = EPT_MEM_TYPE_UC;
1027 	else
1028 		eptp = EPT_MEM_TYPE_WB;
1029 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
1030 		printf("\tPWL4 is not supported\n");
1031 		return 1;
1032 	}
1033 	ctrl_cpu[0] = vmcs_read(CPU_EXEC_CTRL0);
1034 	ctrl_cpu[1] = vmcs_read(CPU_EXEC_CTRL1);
1035 	ctrl_cpu[0] = (ctrl_cpu[0] | CPU_SECONDARY)
1036 		& ctrl_cpu_rev[0].clr;
1037 	ctrl_cpu[1] = (ctrl_cpu[1] | CPU_EPT)
1038 		& ctrl_cpu_rev[1].clr;
1039 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu[0]);
1040 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu[1]);
1041 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1042 	pml4 = alloc_page();
1043 	memset(pml4, 0, PAGE_SIZE);
1044 	eptp |= virt_to_phys(pml4);
1045 	if (enable_ad)
1046 		eptp |= EPTP_AD_FLAG;
1047 	vmcs_write(EPTP, eptp);
1048 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1049 	if (end_of_memory < (1ul << 32))
1050 		end_of_memory = (1ul << 32);
1051 	/* Cannot use large EPT pages if we need to track EPT
1052 	 * accessed/dirty bits at 4K granularity.
1053 	 */
1054 	setup_ept_range(pml4, 0, end_of_memory, 0,
1055 			!enable_ad && ept_2m_supported(),
1056 			EPT_WA | EPT_RA | EPT_EA);
1057 	return 0;
1058 }
1059 
1060 static void ept_enable_ad_bits(void)
1061 {
1062 	eptp |= EPTP_AD_FLAG;
1063 	vmcs_write(EPTP, eptp);
1064 }
1065 
1066 static void ept_disable_ad_bits(void)
1067 {
1068 	eptp &= ~EPTP_AD_FLAG;
1069 	vmcs_write(EPTP, eptp);
1070 }
1071 
1072 static void ept_enable_ad_bits_or_skip_test(void)
1073 {
1074 	if (!ept_ad_bits_supported())
1075 		test_skip("EPT AD bits not supported.");
1076 	ept_enable_ad_bits();
1077 }
1078 
1079 static int apic_version;
1080 
1081 static int ept_init_common(bool have_ad)
1082 {
1083 	int ret;
1084 	struct pci_dev pcidev;
1085 
1086 	if (setup_ept(have_ad))
1087 		return VMX_TEST_EXIT;
1088 	data_page1 = alloc_page();
1089 	data_page2 = alloc_page();
1090 	memset(data_page1, 0x0, PAGE_SIZE);
1091 	memset(data_page2, 0x0, PAGE_SIZE);
1092 	*((u32 *)data_page1) = MAGIC_VAL_1;
1093 	*((u32 *)data_page2) = MAGIC_VAL_2;
1094 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1095 			EPT_RA | EPT_WA | EPT_EA);
1096 
1097 	apic_version = apic_read(APIC_LVR);
1098 
1099 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1100 	if (ret != PCIDEVADDR_INVALID) {
1101 		pci_dev_init(&pcidev, ret);
1102 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1103 	}
1104 
1105 	return VMX_TEST_START;
1106 }
1107 
1108 static int ept_init(struct vmcs *vmcs)
1109 {
1110 	return ept_init_common(false);
1111 }
1112 
1113 static void ept_common(void)
1114 {
1115 	vmx_set_test_stage(0);
1116 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1117 			*((u32 *)data_page1) != MAGIC_VAL_1)
1118 		report("EPT basic framework - read", 0);
1119 	else {
1120 		*((u32 *)data_page2) = MAGIC_VAL_3;
1121 		vmcall();
1122 		if (vmx_get_test_stage() == 1) {
1123 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1124 					*((u32 *)data_page2) == MAGIC_VAL_2)
1125 				report("EPT basic framework", 1);
1126 			else
1127 				report("EPT basic framework - remap", 1);
1128 		}
1129 	}
1130 	// Test EPT Misconfigurations
1131 	vmx_set_test_stage(1);
1132 	vmcall();
1133 	*((u32 *)data_page1) = MAGIC_VAL_1;
1134 	if (vmx_get_test_stage() != 2) {
1135 		report("EPT misconfigurations", 0);
1136 		goto t1;
1137 	}
1138 	vmx_set_test_stage(2);
1139 	vmcall();
1140 	*((u32 *)data_page1) = MAGIC_VAL_1;
1141 	report("EPT misconfigurations", vmx_get_test_stage() == 3);
1142 t1:
1143 	// Test EPT violation
1144 	vmx_set_test_stage(3);
1145 	vmcall();
1146 	*((u32 *)data_page1) = MAGIC_VAL_1;
1147 	report("EPT violation - page permission", vmx_get_test_stage() == 4);
1148 	// Violation caused by EPT paging structure
1149 	vmx_set_test_stage(4);
1150 	vmcall();
1151 	*((u32 *)data_page1) = MAGIC_VAL_2;
1152 	report("EPT violation - paging structure", vmx_get_test_stage() == 5);
1153 
1154 	// MMIO Read/Write
1155 	vmx_set_test_stage(5);
1156 	vmcall();
1157 
1158 	*(u32 volatile *)pci_physaddr;
1159 	report("MMIO EPT violation - read", vmx_get_test_stage() == 6);
1160 
1161 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1162 	report("MMIO EPT violation - write", vmx_get_test_stage() == 7);
1163 }
1164 
1165 static void ept_main(void)
1166 {
1167 	ept_common();
1168 
1169 	// Test EPT access to L1 MMIO
1170 	vmx_set_test_stage(7);
1171 	report("EPT - MMIO access", *((u32 *)0xfee00030UL) == apic_version);
1172 
1173 	// Test invalid operand for INVEPT
1174 	vmcall();
1175 	report("EPT - unsupported INVEPT", vmx_get_test_stage() == 8);
1176 }
1177 
1178 static bool invept_test(int type, u64 eptp)
1179 {
1180 	bool ret, supported;
1181 
1182 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1183 	ret = invept(type, eptp);
1184 
1185 	if (ret == !supported)
1186 		return false;
1187 
1188 	if (!supported)
1189 		printf("WARNING: unsupported invept passed!\n");
1190 	else
1191 		printf("WARNING: invept failed!\n");
1192 
1193 	return true;
1194 }
1195 
1196 static int pml_exit_handler(void)
1197 {
1198 	u16 index, count;
1199 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1200 	u64 *pmlbuf = pml_log;
1201 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1202 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1203 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1204 
1205 	switch (reason) {
1206 	case VMX_VMCALL:
1207 		switch (vmx_get_test_stage()) {
1208 		case 0:
1209 			index = vmcs_read(GUEST_PML_INDEX);
1210 			for (count = index + 1; count < PML_INDEX; count++) {
1211 				if (pmlbuf[count] == (u64)data_page2) {
1212 					vmx_inc_test_stage();
1213 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1214 					break;
1215 				}
1216 			}
1217 			break;
1218 		case 1:
1219 			index = vmcs_read(GUEST_PML_INDEX);
1220 			/* Keep clearing the dirty bit till a overflow */
1221 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1222 			break;
1223 		default:
1224 			report("unexpected stage, %d.", false,
1225 			       vmx_get_test_stage());
1226 			print_vmexit_info();
1227 			return VMX_TEST_VMEXIT;
1228 		}
1229 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1230 		return VMX_TEST_RESUME;
1231 	case VMX_PML_FULL:
1232 		vmx_inc_test_stage();
1233 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1234 		return VMX_TEST_RESUME;
1235 	default:
1236 		report("Unknown exit reason, %ld", false, reason);
1237 		print_vmexit_info();
1238 	}
1239 	return VMX_TEST_VMEXIT;
1240 }
1241 
1242 static int ept_exit_handler_common(bool have_ad)
1243 {
1244 	u64 guest_rip;
1245 	u64 guest_cr3;
1246 	ulong reason;
1247 	u32 insn_len;
1248 	u32 exit_qual;
1249 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte;
1250 
1251 	guest_rip = vmcs_read(GUEST_RIP);
1252 	guest_cr3 = vmcs_read(GUEST_CR3);
1253 	reason = vmcs_read(EXI_REASON) & 0xff;
1254 	insn_len = vmcs_read(EXI_INST_LEN);
1255 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1256 	switch (reason) {
1257 	case VMX_VMCALL:
1258 		switch (vmx_get_test_stage()) {
1259 		case 0:
1260 			check_ept_ad(pml4, guest_cr3,
1261 				     (unsigned long)data_page1,
1262 				     have_ad ? EPT_ACCESS_FLAG : 0,
1263 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1264 			check_ept_ad(pml4, guest_cr3,
1265 				     (unsigned long)data_page2,
1266 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1267 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1268 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1269 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1270 			if (have_ad)
1271 				ept_sync(INVEPT_SINGLE, eptp);;
1272 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1273 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1274 				vmx_inc_test_stage();
1275 				install_ept(pml4, (unsigned long)data_page2,
1276 						(unsigned long)data_page2,
1277 						EPT_RA | EPT_WA | EPT_EA);
1278 			} else
1279 				report("EPT basic framework - write", 0);
1280 			break;
1281 		case 1:
1282 			install_ept(pml4, (unsigned long)data_page1,
1283  				(unsigned long)data_page1, EPT_WA);
1284 			ept_sync(INVEPT_SINGLE, eptp);
1285 			break;
1286 		case 2:
1287 			install_ept(pml4, (unsigned long)data_page1,
1288  				(unsigned long)data_page1,
1289  				EPT_RA | EPT_WA | EPT_EA |
1290  				(2 << EPT_MEM_TYPE_SHIFT));
1291 			ept_sync(INVEPT_SINGLE, eptp);
1292 			break;
1293 		case 3:
1294 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1295 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1296 						1, &data_page1_pte));
1297 			set_ept_pte(pml4, (unsigned long)data_page1,
1298 				1, data_page1_pte & ~EPT_PRESENT);
1299 			ept_sync(INVEPT_SINGLE, eptp);
1300 			break;
1301 		case 4:
1302 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1303 						2, &data_page1_pte));
1304 			data_page1_pte &= PAGE_MASK;
1305 			TEST_ASSERT(get_ept_pte(pml4, data_page1_pte,
1306 						2, &data_page1_pte_pte));
1307 			set_ept_pte(pml4, data_page1_pte, 2,
1308 				data_page1_pte_pte & ~EPT_PRESENT);
1309 			ept_sync(INVEPT_SINGLE, eptp);
1310 			break;
1311 		case 5:
1312 			install_ept(pml4, (unsigned long)pci_physaddr,
1313 				(unsigned long)pci_physaddr, 0);
1314 			ept_sync(INVEPT_SINGLE, eptp);
1315 			break;
1316 		case 7:
1317 			if (!invept_test(0, eptp))
1318 				vmx_inc_test_stage();
1319 			break;
1320 		// Should not reach here
1321 		default:
1322 			report("ERROR - unexpected stage, %d.", false,
1323 			       vmx_get_test_stage());
1324 			print_vmexit_info();
1325 			return VMX_TEST_VMEXIT;
1326 		}
1327 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1328 		return VMX_TEST_RESUME;
1329 	case VMX_EPT_MISCONFIG:
1330 		switch (vmx_get_test_stage()) {
1331 		case 1:
1332 		case 2:
1333 			vmx_inc_test_stage();
1334 			install_ept(pml4, (unsigned long)data_page1,
1335  				(unsigned long)data_page1,
1336  				EPT_RA | EPT_WA | EPT_EA);
1337 			ept_sync(INVEPT_SINGLE, eptp);
1338 			break;
1339 		// Should not reach here
1340 		default:
1341 			report("ERROR - unexpected stage, %d.", false,
1342 			       vmx_get_test_stage());
1343 			print_vmexit_info();
1344 			return VMX_TEST_VMEXIT;
1345 		}
1346 		return VMX_TEST_RESUME;
1347 	case VMX_EPT_VIOLATION:
1348 		switch(vmx_get_test_stage()) {
1349 		case 3:
1350 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1351 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1352 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1353 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1354 					EPT_VLT_PADDR))
1355 				vmx_inc_test_stage();
1356 			set_ept_pte(pml4, (unsigned long)data_page1,
1357 				1, data_page1_pte | (EPT_PRESENT));
1358 			ept_sync(INVEPT_SINGLE, eptp);
1359 			break;
1360 		case 4:
1361 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1362 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1363 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1364 			if (exit_qual == (EPT_VLT_RD |
1365 					  (have_ad ? EPT_VLT_WR : 0) |
1366 					  EPT_VLT_LADDR_VLD))
1367 				vmx_inc_test_stage();
1368 			set_ept_pte(pml4, data_page1_pte, 2,
1369 				data_page1_pte_pte | (EPT_PRESENT));
1370 			ept_sync(INVEPT_SINGLE, eptp);
1371 			break;
1372 		case 5:
1373 			if (exit_qual & EPT_VLT_RD)
1374 				vmx_inc_test_stage();
1375 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1376 						1, &memaddr_pte));
1377 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1378 			ept_sync(INVEPT_SINGLE, eptp);
1379 			break;
1380 		case 6:
1381 			if (exit_qual & EPT_VLT_WR)
1382 				vmx_inc_test_stage();
1383 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1384 						1, &memaddr_pte));
1385 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1386 			ept_sync(INVEPT_SINGLE, eptp);
1387 			break;
1388 		default:
1389 			// Should not reach here
1390 			report("ERROR : unexpected stage, %d", false,
1391 			       vmx_get_test_stage());
1392 			print_vmexit_info();
1393 			return VMX_TEST_VMEXIT;
1394 		}
1395 		return VMX_TEST_RESUME;
1396 	default:
1397 		report("Unknown exit reason, %ld", false, reason);
1398 		print_vmexit_info();
1399 	}
1400 	return VMX_TEST_VMEXIT;
1401 }
1402 
1403 static int ept_exit_handler(void)
1404 {
1405 	return ept_exit_handler_common(false);
1406 }
1407 
1408 static int eptad_init(struct vmcs *vmcs)
1409 {
1410 	int r = ept_init_common(true);
1411 
1412 	if (r == VMX_TEST_EXIT)
1413 		return r;
1414 
1415 	if ((rdmsr(MSR_IA32_VMX_EPT_VPID_CAP) & EPT_CAP_AD_FLAG) == 0) {
1416 		printf("\tEPT A/D bits are not supported");
1417 		return VMX_TEST_EXIT;
1418 	}
1419 
1420 	return r;
1421 }
1422 
1423 static int pml_init(struct vmcs *vmcs)
1424 {
1425 	u32 ctrl_cpu;
1426 	int r = eptad_init(vmcs);
1427 
1428 	if (r == VMX_TEST_EXIT)
1429 		return r;
1430 
1431 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1432 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1433 		printf("\tPML is not supported");
1434 		return VMX_TEST_EXIT;
1435 	}
1436 
1437 	pml_log = alloc_page();
1438 	memset(pml_log, 0x0, PAGE_SIZE);
1439 	vmcs_write(PMLADDR, (u64)pml_log);
1440 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1441 
1442 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1443 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1444 
1445 	return VMX_TEST_START;
1446 }
1447 
1448 static void pml_main(void)
1449 {
1450 	int count = 0;
1451 
1452 	vmx_set_test_stage(0);
1453 	*((u32 *)data_page2) = 0x1;
1454 	vmcall();
1455 	report("PML - Dirty GPA Logging", vmx_get_test_stage() == 1);
1456 
1457 	while (vmx_get_test_stage() == 1) {
1458 		vmcall();
1459 		*((u32 *)data_page2) = 0x1;
1460 		if (count++ > PML_INDEX)
1461 			break;
1462 	}
1463 	report("PML Full Event", vmx_get_test_stage() == 2);
1464 }
1465 
1466 static void eptad_main(void)
1467 {
1468 	ept_common();
1469 }
1470 
1471 static int eptad_exit_handler(void)
1472 {
1473 	return ept_exit_handler_common(true);
1474 }
1475 
1476 static bool invvpid_test(int type, u16 vpid)
1477 {
1478 	bool ret, supported;
1479 
1480 	supported = ept_vpid.val &
1481 		(VPID_CAP_INVVPID_ADDR >> INVVPID_ADDR << type);
1482 	ret = invvpid(type, vpid, 0);
1483 
1484 	if (ret == !supported)
1485 		return false;
1486 
1487 	if (!supported)
1488 		printf("WARNING: unsupported invvpid passed!\n");
1489 	else
1490 		printf("WARNING: invvpid failed!\n");
1491 
1492 	return true;
1493 }
1494 
1495 static int vpid_init(struct vmcs *vmcs)
1496 {
1497 	u32 ctrl_cpu1;
1498 
1499 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1500 		!(ctrl_cpu_rev[1].clr & CPU_VPID)) {
1501 		printf("\tVPID is not supported");
1502 		return VMX_TEST_EXIT;
1503 	}
1504 
1505 	ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
1506 	ctrl_cpu1 |= CPU_VPID;
1507 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
1508 	return VMX_TEST_START;
1509 }
1510 
1511 static void vpid_main(void)
1512 {
1513 	vmx_set_test_stage(0);
1514 	vmcall();
1515 	report("INVVPID SINGLE ADDRESS", vmx_get_test_stage() == 1);
1516 	vmx_set_test_stage(2);
1517 	vmcall();
1518 	report("INVVPID SINGLE", vmx_get_test_stage() == 3);
1519 	vmx_set_test_stage(4);
1520 	vmcall();
1521 	report("INVVPID ALL", vmx_get_test_stage() == 5);
1522 }
1523 
1524 static int vpid_exit_handler(void)
1525 {
1526 	u64 guest_rip;
1527 	ulong reason;
1528 	u32 insn_len;
1529 
1530 	guest_rip = vmcs_read(GUEST_RIP);
1531 	reason = vmcs_read(EXI_REASON) & 0xff;
1532 	insn_len = vmcs_read(EXI_INST_LEN);
1533 
1534 	switch (reason) {
1535 	case VMX_VMCALL:
1536 		switch(vmx_get_test_stage()) {
1537 		case 0:
1538 			if (!invvpid_test(INVVPID_ADDR, 1))
1539 				vmx_inc_test_stage();
1540 			break;
1541 		case 2:
1542 			if (!invvpid_test(INVVPID_CONTEXT_GLOBAL, 1))
1543 				vmx_inc_test_stage();
1544 			break;
1545 		case 4:
1546 			if (!invvpid_test(INVVPID_ALL, 1))
1547 				vmx_inc_test_stage();
1548 			break;
1549 		default:
1550 			report("ERROR: unexpected stage, %d", false,
1551 					vmx_get_test_stage());
1552 			print_vmexit_info();
1553 			return VMX_TEST_VMEXIT;
1554 		}
1555 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1556 		return VMX_TEST_RESUME;
1557 	default:
1558 		report("Unknown exit reason, %ld", false, reason);
1559 		print_vmexit_info();
1560 	}
1561 	return VMX_TEST_VMEXIT;
1562 }
1563 
1564 #define TIMER_VECTOR	222
1565 
1566 static volatile bool timer_fired;
1567 
1568 static void timer_isr(isr_regs_t *regs)
1569 {
1570 	timer_fired = true;
1571 	apic_write(APIC_EOI, 0);
1572 }
1573 
1574 static int interrupt_init(struct vmcs *vmcs)
1575 {
1576 	msr_bmp_init();
1577 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1578 	handle_irq(TIMER_VECTOR, timer_isr);
1579 	return VMX_TEST_START;
1580 }
1581 
1582 static void interrupt_main(void)
1583 {
1584 	long long start, loops;
1585 
1586 	vmx_set_test_stage(0);
1587 
1588 	apic_write(APIC_LVTT, TIMER_VECTOR);
1589 	irq_enable();
1590 
1591 	apic_write(APIC_TMICT, 1);
1592 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1593 		asm volatile ("nop");
1594 	report("direct interrupt while running guest", timer_fired);
1595 
1596 	apic_write(APIC_TMICT, 0);
1597 	irq_disable();
1598 	vmcall();
1599 	timer_fired = false;
1600 	apic_write(APIC_TMICT, 1);
1601 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1602 		asm volatile ("nop");
1603 	report("intercepted interrupt while running guest", timer_fired);
1604 
1605 	irq_enable();
1606 	apic_write(APIC_TMICT, 0);
1607 	irq_disable();
1608 	vmcall();
1609 	timer_fired = false;
1610 	start = rdtsc();
1611 	apic_write(APIC_TMICT, 1000000);
1612 
1613 	asm volatile ("sti; hlt");
1614 
1615 	report("direct interrupt + hlt",
1616 	       rdtsc() - start > 1000000 && timer_fired);
1617 
1618 	apic_write(APIC_TMICT, 0);
1619 	irq_disable();
1620 	vmcall();
1621 	timer_fired = false;
1622 	start = rdtsc();
1623 	apic_write(APIC_TMICT, 1000000);
1624 
1625 	asm volatile ("sti; hlt");
1626 
1627 	report("intercepted interrupt + hlt",
1628 	       rdtsc() - start > 10000 && timer_fired);
1629 
1630 	apic_write(APIC_TMICT, 0);
1631 	irq_disable();
1632 	vmcall();
1633 	timer_fired = false;
1634 	start = rdtsc();
1635 	apic_write(APIC_TMICT, 1000000);
1636 
1637 	irq_enable();
1638 	asm volatile ("nop");
1639 	vmcall();
1640 
1641 	report("direct interrupt + activity state hlt",
1642 	       rdtsc() - start > 10000 && timer_fired);
1643 
1644 	apic_write(APIC_TMICT, 0);
1645 	irq_disable();
1646 	vmcall();
1647 	timer_fired = false;
1648 	start = rdtsc();
1649 	apic_write(APIC_TMICT, 1000000);
1650 
1651 	irq_enable();
1652 	asm volatile ("nop");
1653 	vmcall();
1654 
1655 	report("intercepted interrupt + activity state hlt",
1656 	       rdtsc() - start > 10000 && timer_fired);
1657 
1658 	apic_write(APIC_TMICT, 0);
1659 	irq_disable();
1660 	vmx_set_test_stage(7);
1661 	vmcall();
1662 	timer_fired = false;
1663 	apic_write(APIC_TMICT, 1);
1664 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1665 		asm volatile ("nop");
1666 	report("running a guest with interrupt acknowledgement set", timer_fired);
1667 
1668 	apic_write(APIC_TMICT, 0);
1669 	irq_enable();
1670 	timer_fired = false;
1671 	vmcall();
1672 	report("Inject an event to a halted guest", timer_fired);
1673 }
1674 
1675 static int interrupt_exit_handler(void)
1676 {
1677 	u64 guest_rip = vmcs_read(GUEST_RIP);
1678 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1679 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1680 
1681 	switch (reason) {
1682 	case VMX_VMCALL:
1683 		switch (vmx_get_test_stage()) {
1684 		case 0:
1685 		case 2:
1686 		case 5:
1687 			vmcs_write(PIN_CONTROLS,
1688 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1689 			break;
1690 		case 7:
1691 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1692 			vmcs_write(PIN_CONTROLS,
1693 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1694 			break;
1695 		case 1:
1696 		case 3:
1697 			vmcs_write(PIN_CONTROLS,
1698 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1699 			break;
1700 		case 4:
1701 		case 6:
1702 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1703 			break;
1704 
1705 		case 8:
1706 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1707 			vmcs_write(ENT_INTR_INFO,
1708 				   TIMER_VECTOR |
1709 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1710 				   INTR_INFO_VALID_MASK);
1711 			break;
1712 		}
1713 		vmx_inc_test_stage();
1714 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1715 		return VMX_TEST_RESUME;
1716 	case VMX_EXTINT:
1717 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1718 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1719 			handle_external_interrupt(vector);
1720 		} else {
1721 			irq_enable();
1722 			asm volatile ("nop");
1723 			irq_disable();
1724 		}
1725 		if (vmx_get_test_stage() >= 2)
1726 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1727 		return VMX_TEST_RESUME;
1728 	default:
1729 		report("Unknown exit reason, %ld", false, reason);
1730 		print_vmexit_info();
1731 	}
1732 
1733 	return VMX_TEST_VMEXIT;
1734 }
1735 
1736 static int dbgctls_init(struct vmcs *vmcs)
1737 {
1738 	u64 dr7 = 0x402;
1739 	u64 zero = 0;
1740 
1741 	msr_bmp_init();
1742 	asm volatile(
1743 		"mov %0,%%dr0\n\t"
1744 		"mov %0,%%dr1\n\t"
1745 		"mov %0,%%dr2\n\t"
1746 		"mov %1,%%dr7\n\t"
1747 		: : "r" (zero), "r" (dr7));
1748 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1749 	vmcs_write(GUEST_DR7, 0x404);
1750 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1751 
1752 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1753 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1754 
1755 	return VMX_TEST_START;
1756 }
1757 
1758 static void dbgctls_main(void)
1759 {
1760 	u64 dr7, debugctl;
1761 
1762 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1763 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1764 	/* Commented out: KVM does not support DEBUGCTL so far */
1765 	(void)debugctl;
1766 	report("Load debug controls", dr7 == 0x404 /* && debugctl == 0x2 */);
1767 
1768 	dr7 = 0x408;
1769 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1770 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1771 
1772 	vmx_set_test_stage(0);
1773 	vmcall();
1774 	report("Save debug controls", vmx_get_test_stage() == 1);
1775 
1776 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1777 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1778 		printf("\tDebug controls are always loaded/saved\n");
1779 		return;
1780 	}
1781 	vmx_set_test_stage(2);
1782 	vmcall();
1783 
1784 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1785 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1786 	/* Commented out: KVM does not support DEBUGCTL so far */
1787 	(void)debugctl;
1788 	report("Guest=host debug controls", dr7 == 0x402 /* && debugctl == 0x1 */);
1789 
1790 	dr7 = 0x408;
1791 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1792 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1793 
1794 	vmx_set_test_stage(3);
1795 	vmcall();
1796 	report("Don't save debug controls", vmx_get_test_stage() == 4);
1797 }
1798 
1799 static int dbgctls_exit_handler(void)
1800 {
1801 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
1802 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1803 	u64 guest_rip = vmcs_read(GUEST_RIP);
1804 	u64 dr7, debugctl;
1805 
1806 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1807 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1808 
1809 	switch (reason) {
1810 	case VMX_VMCALL:
1811 		switch (vmx_get_test_stage()) {
1812 		case 0:
1813 			if (dr7 == 0x400 && debugctl == 0 &&
1814 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1815 			    Commented out: KVM does not support DEBUGCTL so far
1816 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1817 				vmx_inc_test_stage();
1818 			break;
1819 		case 2:
1820 			dr7 = 0x402;
1821 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1822 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1823 			vmcs_write(GUEST_DR7, 0x404);
1824 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1825 
1826 			vmcs_write(ENT_CONTROLS,
1827 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1828 			vmcs_write(EXI_CONTROLS,
1829 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1830 			break;
1831 		case 3:
1832 			if (dr7 == 0x400 && debugctl == 0 &&
1833 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1834 			    Commented out: KVM does not support DEBUGCTL so far
1835 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1836 				vmx_inc_test_stage();
1837 			break;
1838 		}
1839 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1840 		return VMX_TEST_RESUME;
1841 	default:
1842 		report("Unknown exit reason, %d", false, reason);
1843 		print_vmexit_info();
1844 	}
1845 	return VMX_TEST_VMEXIT;
1846 }
1847 
1848 struct vmx_msr_entry {
1849 	u32 index;
1850 	u32 reserved;
1851 	u64 value;
1852 } __attribute__((packed));
1853 
1854 #define MSR_MAGIC 0x31415926
1855 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1856 
1857 static int msr_switch_init(struct vmcs *vmcs)
1858 {
1859 	msr_bmp_init();
1860 	exit_msr_store = alloc_page();
1861 	exit_msr_load = alloc_page();
1862 	entry_msr_load = alloc_page();
1863 	memset(exit_msr_store, 0, PAGE_SIZE);
1864 	memset(exit_msr_load, 0, PAGE_SIZE);
1865 	memset(entry_msr_load, 0, PAGE_SIZE);
1866 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1867 	entry_msr_load[0].value = MSR_MAGIC;
1868 
1869 	vmx_set_test_stage(1);
1870 	vmcs_write(ENT_MSR_LD_CNT, 1);
1871 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1872 	vmcs_write(EXI_MSR_ST_CNT, 1);
1873 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1874 	vmcs_write(EXI_MSR_LD_CNT, 1);
1875 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1876 	return VMX_TEST_START;
1877 }
1878 
1879 static void msr_switch_main(void)
1880 {
1881 	if (vmx_get_test_stage() == 1) {
1882 		report("VM entry MSR load",
1883 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC);
1884 		vmx_set_test_stage(2);
1885 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
1886 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
1887 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
1888 		exit_msr_load[0].value = MSR_MAGIC + 2;
1889 	}
1890 	vmcall();
1891 }
1892 
1893 static int msr_switch_exit_handler(void)
1894 {
1895 	ulong reason;
1896 
1897 	reason = vmcs_read(EXI_REASON);
1898 	if (reason == VMX_VMCALL && vmx_get_test_stage() == 2) {
1899 		report("VM exit MSR store",
1900 			exit_msr_store[0].value == MSR_MAGIC + 1);
1901 		report("VM exit MSR load",
1902 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2);
1903 		vmx_set_test_stage(3);
1904 		entry_msr_load[0].index = MSR_FS_BASE;
1905 		return VMX_TEST_RESUME;
1906 	}
1907 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1908 		__func__, vmx_get_test_stage(), reason);
1909 	return VMX_TEST_EXIT;
1910 }
1911 
1912 static int msr_switch_entry_failure(struct vmentry_failure *failure)
1913 {
1914 	ulong reason;
1915 
1916 	if (failure->early) {
1917 		printf("ERROR %s: early exit\n", __func__);
1918 		return VMX_TEST_EXIT;
1919 	}
1920 
1921 	reason = vmcs_read(EXI_REASON);
1922 	if (reason == (VMX_ENTRY_FAILURE | VMX_FAIL_MSR) &&
1923 	    vmx_get_test_stage() == 3) {
1924 		report("VM entry MSR load: try to load FS_BASE",
1925 			vmcs_read(EXI_QUALIFICATION) == 1);
1926 		return VMX_TEST_VMEXIT;
1927 	}
1928 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1929 		__func__, vmx_get_test_stage(), reason);
1930 	return VMX_TEST_EXIT;
1931 }
1932 
1933 static int vmmcall_init(struct vmcs *vmcs)
1934 {
1935 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
1936 	return VMX_TEST_START;
1937 }
1938 
1939 static void vmmcall_main(void)
1940 {
1941 	asm volatile(
1942 		"mov $0xABCD, %%rax\n\t"
1943 		"vmmcall\n\t"
1944 		::: "rax");
1945 
1946 	report("VMMCALL", 0);
1947 }
1948 
1949 static int vmmcall_exit_handler(void)
1950 {
1951 	ulong reason;
1952 
1953 	reason = vmcs_read(EXI_REASON);
1954 	switch (reason) {
1955 	case VMX_VMCALL:
1956 		printf("here\n");
1957 		report("VMMCALL triggers #UD", 0);
1958 		break;
1959 	case VMX_EXC_NMI:
1960 		report("VMMCALL triggers #UD",
1961 		       (vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR);
1962 		break;
1963 	default:
1964 		report("Unknown exit reason, %ld", false, reason);
1965 		print_vmexit_info();
1966 	}
1967 
1968 	return VMX_TEST_VMEXIT;
1969 }
1970 
1971 static int disable_rdtscp_init(struct vmcs *vmcs)
1972 {
1973 	u32 ctrl_cpu1;
1974 
1975 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
1976 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
1977 		ctrl_cpu1 &= ~CPU_RDTSCP;
1978 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
1979 	}
1980 
1981 	return VMX_TEST_START;
1982 }
1983 
1984 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
1985 {
1986 	switch (vmx_get_test_stage()) {
1987 	case 0:
1988 		report("RDTSCP triggers #UD", true);
1989 		vmx_inc_test_stage();
1990 		regs->rip += 3;
1991 		break;
1992 	case 2:
1993 		report("RDPID triggers #UD", true);
1994 		vmx_inc_test_stage();
1995 		regs->rip += 4;
1996 		break;
1997 	}
1998 	return;
1999 
2000 }
2001 
2002 static void disable_rdtscp_main(void)
2003 {
2004 	/* Test that #UD is properly injected in L2.  */
2005 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2006 
2007 	vmx_set_test_stage(0);
2008 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2009 	vmcall();
2010 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2011 
2012 	handle_exception(UD_VECTOR, 0);
2013 	vmcall();
2014 }
2015 
2016 static int disable_rdtscp_exit_handler(void)
2017 {
2018 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
2019 
2020 	switch (reason) {
2021 	case VMX_VMCALL:
2022 		switch (vmx_get_test_stage()) {
2023 		case 0:
2024 			report("RDTSCP triggers #UD", false);
2025 			vmx_inc_test_stage();
2026 			/* fallthrough */
2027 		case 1:
2028 			vmx_inc_test_stage();
2029 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2030 			return VMX_TEST_RESUME;
2031 		case 2:
2032 			report("RDPID triggers #UD", false);
2033 			break;
2034 		}
2035 		break;
2036 
2037 	default:
2038 		report("Unknown exit reason, %d", false, reason);
2039 		print_vmexit_info();
2040 	}
2041 	return VMX_TEST_VMEXIT;
2042 }
2043 
2044 static int int3_init(struct vmcs *vmcs)
2045 {
2046 	vmcs_write(EXC_BITMAP, ~0u);
2047 	return VMX_TEST_START;
2048 }
2049 
2050 static void int3_guest_main(void)
2051 {
2052 	asm volatile ("int3");
2053 }
2054 
2055 static int int3_exit_handler(void)
2056 {
2057 	u32 reason = vmcs_read(EXI_REASON);
2058 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2059 
2060 	report("L1 intercepts #BP", reason == VMX_EXC_NMI &&
2061 	       (intr_info & INTR_INFO_VALID_MASK) &&
2062 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2063 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2064 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2065 
2066 	return VMX_TEST_VMEXIT;
2067 }
2068 
2069 static int into_init(struct vmcs *vmcs)
2070 {
2071 	vmcs_write(EXC_BITMAP, ~0u);
2072 	return VMX_TEST_START;
2073 }
2074 
2075 static void into_guest_main(void)
2076 {
2077 	struct far_pointer32 fp = {
2078 		.offset = (uintptr_t)&&into,
2079 		.selector = KERNEL_CS32,
2080 	};
2081 	register uintptr_t rsp asm("rsp");
2082 
2083 	if (fp.offset != (uintptr_t)&&into) {
2084 		printf("Code address too high.\n");
2085 		return;
2086 	}
2087 	if ((u32)rsp != rsp) {
2088 		printf("Stack address too high.\n");
2089 		return;
2090 	}
2091 
2092 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2093 	return;
2094 into:
2095 	asm volatile (".code32;"
2096 		      "movl $0x7fffffff, %eax;"
2097 		      "addl %eax, %eax;"
2098 		      "into;"
2099 		      "lret;"
2100 		      ".code64");
2101 	__builtin_unreachable();
2102 }
2103 
2104 static int into_exit_handler(void)
2105 {
2106 	u32 reason = vmcs_read(EXI_REASON);
2107 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2108 
2109 	report("L1 intercepts #OF", reason == VMX_EXC_NMI &&
2110 	       (intr_info & INTR_INFO_VALID_MASK) &&
2111 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2112 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2113 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2114 
2115 	return VMX_TEST_VMEXIT;
2116 }
2117 
2118 static void exit_monitor_from_l2_main(void)
2119 {
2120 	printf("Calling exit(0) from l2...\n");
2121 	exit(0);
2122 }
2123 
2124 static int exit_monitor_from_l2_handler(void)
2125 {
2126 	report("The guest should have killed the VMM", false);
2127 	return VMX_TEST_EXIT;
2128 }
2129 
2130 static void assert_exit_reason(u64 expected)
2131 {
2132 	u64 actual = vmcs_read(EXI_REASON);
2133 
2134 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2135 			   exit_reason_description(expected),
2136 			   exit_reason_description(actual));
2137 }
2138 
2139 static void skip_exit_insn(void)
2140 {
2141 	u64 guest_rip = vmcs_read(GUEST_RIP);
2142 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2143 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2144 }
2145 
2146 static void skip_exit_vmcall(void)
2147 {
2148 	assert_exit_reason(VMX_VMCALL);
2149 	skip_exit_insn();
2150 }
2151 
2152 static void v2_null_test_guest(void)
2153 {
2154 }
2155 
2156 static void v2_null_test(void)
2157 {
2158 	test_set_guest(v2_null_test_guest);
2159 	enter_guest();
2160 	report(__func__, 1);
2161 }
2162 
2163 static void v2_multiple_entries_test_guest(void)
2164 {
2165 	vmx_set_test_stage(1);
2166 	vmcall();
2167 	vmx_set_test_stage(2);
2168 }
2169 
2170 static void v2_multiple_entries_test(void)
2171 {
2172 	test_set_guest(v2_multiple_entries_test_guest);
2173 	enter_guest();
2174 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2175 	skip_exit_vmcall();
2176 	enter_guest();
2177 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2178 	report(__func__, 1);
2179 }
2180 
2181 static int fixture_test_data = 1;
2182 
2183 static void fixture_test_teardown(void *data)
2184 {
2185 	*((int *) data) = 1;
2186 }
2187 
2188 static void fixture_test_guest(void)
2189 {
2190 	fixture_test_data++;
2191 }
2192 
2193 
2194 static void fixture_test_setup(void)
2195 {
2196 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2197 			   "fixture_test_teardown didn't run?!");
2198 	fixture_test_data = 2;
2199 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2200 	test_set_guest(fixture_test_guest);
2201 }
2202 
2203 static void fixture_test_case1(void)
2204 {
2205 	fixture_test_setup();
2206 	TEST_ASSERT_EQ(2, fixture_test_data);
2207 	enter_guest();
2208 	TEST_ASSERT_EQ(3, fixture_test_data);
2209 	report(__func__, 1);
2210 }
2211 
2212 static void fixture_test_case2(void)
2213 {
2214 	fixture_test_setup();
2215 	TEST_ASSERT_EQ(2, fixture_test_data);
2216 	enter_guest();
2217 	TEST_ASSERT_EQ(3, fixture_test_data);
2218 	report(__func__, 1);
2219 }
2220 
2221 enum ept_access_op {
2222 	OP_READ,
2223 	OP_WRITE,
2224 	OP_EXEC,
2225 	OP_FLUSH_TLB,
2226 	OP_EXIT,
2227 };
2228 
2229 static struct ept_access_test_data {
2230 	unsigned long gpa;
2231 	unsigned long *gva;
2232 	unsigned long hpa;
2233 	unsigned long *hva;
2234 	enum ept_access_op op;
2235 } ept_access_test_data;
2236 
2237 extern unsigned char ret42_start;
2238 extern unsigned char ret42_end;
2239 
2240 /* Returns 42. */
2241 asm(
2242 	".align 64\n"
2243 	"ret42_start:\n"
2244 	"mov $42, %eax\n"
2245 	"ret\n"
2246 	"ret42_end:\n"
2247 );
2248 
2249 static void
2250 diagnose_ept_violation_qual(u64 expected, u64 actual)
2251 {
2252 
2253 #define DIAGNOSE(flag)							\
2254 do {									\
2255 	if ((expected & flag) != (actual & flag))			\
2256 		printf(#flag " %sexpected\n",				\
2257 		       (expected & flag) ? "" : "un");			\
2258 } while (0)
2259 
2260 	DIAGNOSE(EPT_VLT_RD);
2261 	DIAGNOSE(EPT_VLT_WR);
2262 	DIAGNOSE(EPT_VLT_FETCH);
2263 	DIAGNOSE(EPT_VLT_PERM_RD);
2264 	DIAGNOSE(EPT_VLT_PERM_WR);
2265 	DIAGNOSE(EPT_VLT_PERM_EX);
2266 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2267 	DIAGNOSE(EPT_VLT_PADDR);
2268 
2269 #undef DIAGNOSE
2270 }
2271 
2272 static void do_ept_access_op(enum ept_access_op op)
2273 {
2274 	ept_access_test_data.op = op;
2275 	enter_guest();
2276 }
2277 
2278 /*
2279  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2280  * needed by tests that modify guest PTEs.
2281  */
2282 static void ept_access_test_guest_flush_tlb(void)
2283 {
2284 	do_ept_access_op(OP_FLUSH_TLB);
2285 	skip_exit_vmcall();
2286 }
2287 
2288 /*
2289  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2290  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2291  * a huge page.
2292  */
2293 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2294 				 unsigned long clear, unsigned long set)
2295 {
2296 	struct ept_access_test_data *data = &ept_access_test_data;
2297 	unsigned long orig_pte;
2298 	unsigned long pte;
2299 
2300 	/* Screw with the mapping at the requested level. */
2301 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2302 	pte = orig_pte;
2303 	if (mkhuge)
2304 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2305 	else
2306 		pte = orig_pte;
2307 	pte = (pte & ~clear) | set;
2308 	set_ept_pte(pml4, gpa, level, pte);
2309 	ept_sync(INVEPT_SINGLE, eptp);
2310 
2311 	return orig_pte;
2312 }
2313 
2314 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2315 {
2316 	set_ept_pte(pml4, gpa, level, orig_pte);
2317 }
2318 
2319 static void do_ept_violation(bool leaf, enum ept_access_op op,
2320 			     u64 expected_qual, u64 expected_paddr)
2321 {
2322 	u64 qual;
2323 
2324 	/* Try the access and observe the violation. */
2325 	do_ept_access_op(op);
2326 
2327 	assert_exit_reason(VMX_EPT_VIOLATION);
2328 
2329 	qual = vmcs_read(EXI_QUALIFICATION);
2330 
2331 	diagnose_ept_violation_qual(expected_qual, qual);
2332 	TEST_EXPECT_EQ(expected_qual, qual);
2333 
2334 	#if 0
2335 	/* Disable for now otherwise every test will fail */
2336 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2337 		       (unsigned long) (
2338 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2339 	#endif
2340 	/*
2341 	 * TODO: tests that probe expected_paddr in pages other than the one at
2342 	 * the beginning of the 1g region.
2343 	 */
2344 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2345 }
2346 
2347 static void
2348 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2349 			      unsigned long set, enum ept_access_op op,
2350 			      u64 expected_qual)
2351 {
2352 	struct ept_access_test_data *data = &ept_access_test_data;
2353 	unsigned long orig_pte;
2354 
2355 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2356 
2357 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2358 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2359 					 data->gpa);
2360 
2361 	/* Fix the violation and resume the op loop. */
2362 	ept_untwiddle(data->gpa, level, orig_pte);
2363 	enter_guest();
2364 	skip_exit_vmcall();
2365 }
2366 
2367 static void
2368 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2369 		       enum ept_access_op op, u64 expected_qual)
2370 {
2371 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2372 				      expected_qual);
2373 	if (ept_huge_pages_supported(level))
2374 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2375 					      expected_qual);
2376 }
2377 
2378 static void ept_violation(unsigned long clear, unsigned long set,
2379 			  enum ept_access_op op, u64 expected_qual)
2380 {
2381 	ept_violation_at_level(1, clear, set, op, expected_qual);
2382 	ept_violation_at_level(2, clear, set, op, expected_qual);
2383 	ept_violation_at_level(3, clear, set, op, expected_qual);
2384 	ept_violation_at_level(4, clear, set, op, expected_qual);
2385 }
2386 
2387 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2388 				       u64 expected_qual)
2389 {
2390 	ept_violation(EPT_PRESENT, access, op,
2391 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2392 }
2393 
2394 /*
2395  * For translations that don't involve a GVA, that is physical address (paddr)
2396  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2397  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2398  * translations don't involve GVAs, such as when the hardware does the guest
2399  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2400  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2401  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2402  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2403  *
2404  * Note that paddr violations can also be triggered by loading PAE page tables
2405  * with wonky addresses. We don't test that yet.
2406  *
2407  * This function modifies the EPT entry that maps the GPA that the guest page
2408  * table entry mapping ept_access_data.gva resides on.
2409  *
2410  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2411  *
2412  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2413  *
2414  *	@op		Guest operation to perform with ept_access_data.gva.
2415  *
2416  *	@expect_violation
2417  *			Is a violation expected during the paddr access?
2418  *
2419  *	@expected_qual	Expected qualification for the EPT violation.
2420  *			EPT_VLT_PADDR should be clear.
2421  */
2422 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2423 			     enum ept_access_op op, bool expect_violation,
2424 			     u64 expected_qual)
2425 {
2426 	struct ept_access_test_data *data = &ept_access_test_data;
2427 	unsigned long *ptep;
2428 	unsigned long gpa;
2429 	unsigned long orig_epte;
2430 
2431 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2432 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2433 	TEST_ASSERT(ptep);
2434 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2435 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2436 	ept_access_test_guest_flush_tlb();
2437 
2438 	/*
2439 	 * Now modify the access bits on the EPT entry for the GPA that the
2440 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2441 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2442 	 * constructed our test such that those other 511 PTEs aren't used by
2443 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2444 	 * PTE we're modifying is at the beginning of a 4K page and the
2445 	 * following 511 entires are also under our control (and not touched by
2446 	 * the guest).
2447 	 */
2448 	gpa = virt_to_phys(ptep);
2449 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2450 	/*
2451 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2452 	 * otherwise our level=1 twiddling below will fail. We use the
2453 	 * identity map (gpa = gpa) since page tables are shared with the host.
2454 	 */
2455 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2456 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2457 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2458 
2459 	if (expect_violation) {
2460 		do_ept_violation(/*leaf=*/true, op,
2461 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2462 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2463 		do_ept_access_op(op);
2464 	} else {
2465 		do_ept_access_op(op);
2466 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2467 	}
2468 
2469 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2470 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2471 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2472 
2473 	skip_exit_vmcall();
2474 }
2475 
2476 static void ept_access_allowed_paddr(unsigned long ept_access,
2477 				     unsigned long pte_ad,
2478 				     enum ept_access_op op)
2479 {
2480 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2481 			 /*expected_qual=*/-1);
2482 }
2483 
2484 static void ept_access_violation_paddr(unsigned long ept_access,
2485 				       unsigned long pte_ad,
2486 				       enum ept_access_op op,
2487 				       u64 expected_qual)
2488 {
2489 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2490 			 expected_qual);
2491 }
2492 
2493 
2494 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2495 					unsigned long clear,
2496 					unsigned long set,
2497 					enum ept_access_op op)
2498 {
2499 	struct ept_access_test_data *data = &ept_access_test_data;
2500 	unsigned long orig_pte;
2501 
2502 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2503 
2504 	/* No violation. Should proceed to vmcall. */
2505 	do_ept_access_op(op);
2506 	skip_exit_vmcall();
2507 
2508 	ept_untwiddle(data->gpa, level, orig_pte);
2509 }
2510 
2511 static void ept_allowed_at_level(int level, unsigned long clear,
2512 				 unsigned long set, enum ept_access_op op)
2513 {
2514 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2515 	if (ept_huge_pages_supported(level))
2516 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2517 }
2518 
2519 static void ept_allowed(unsigned long clear, unsigned long set,
2520 			enum ept_access_op op)
2521 {
2522 	ept_allowed_at_level(1, clear, set, op);
2523 	ept_allowed_at_level(2, clear, set, op);
2524 	ept_allowed_at_level(3, clear, set, op);
2525 	ept_allowed_at_level(4, clear, set, op);
2526 }
2527 
2528 static void ept_ignored_bit(int bit)
2529 {
2530 	/* Set the bit. */
2531 	ept_allowed(0, 1ul << bit, OP_READ);
2532 	ept_allowed(0, 1ul << bit, OP_WRITE);
2533 	ept_allowed(0, 1ul << bit, OP_EXEC);
2534 
2535 	/* Clear the bit. */
2536 	ept_allowed(1ul << bit, 0, OP_READ);
2537 	ept_allowed(1ul << bit, 0, OP_WRITE);
2538 	ept_allowed(1ul << bit, 0, OP_EXEC);
2539 }
2540 
2541 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2542 {
2543 	ept_allowed(EPT_PRESENT, access, op);
2544 }
2545 
2546 
2547 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2548 					     unsigned long clear,
2549 					     unsigned long set,
2550 					     enum ept_access_op op)
2551 {
2552 	struct ept_access_test_data *data = &ept_access_test_data;
2553 	unsigned long orig_pte;
2554 
2555 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2556 
2557 	do_ept_access_op(op);
2558 	assert_exit_reason(VMX_EPT_MISCONFIG);
2559 
2560 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2561 	#if 0
2562 	/* broken: */
2563 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2564 	#endif
2565 	#if 0
2566 	/*
2567 	 * broken:
2568 	 * According to description of exit qual for EPT violation,
2569 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2570 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2571 	 * to be set for msiconfig.
2572 	 */
2573 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2574 		       (unsigned long) (
2575 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2576 	#endif
2577 
2578 	/* Fix the violation and resume the op loop. */
2579 	ept_untwiddle(data->gpa, level, orig_pte);
2580 	enter_guest();
2581 	skip_exit_vmcall();
2582 }
2583 
2584 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2585 					  unsigned long clear,
2586 					  unsigned long set)
2587 {
2588 	/* The op shouldn't matter (read, write, exec), so try them all! */
2589 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2590 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2591 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2592 }
2593 
2594 static void ept_misconfig_at_level(int level, unsigned long clear,
2595 				   unsigned long set)
2596 {
2597 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2598 	if (ept_huge_pages_supported(level))
2599 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2600 }
2601 
2602 static void ept_misconfig(unsigned long clear, unsigned long set)
2603 {
2604 	ept_misconfig_at_level(1, clear, set);
2605 	ept_misconfig_at_level(2, clear, set);
2606 	ept_misconfig_at_level(3, clear, set);
2607 	ept_misconfig_at_level(4, clear, set);
2608 }
2609 
2610 static void ept_access_misconfig(unsigned long access)
2611 {
2612 	ept_misconfig(EPT_PRESENT, access);
2613 }
2614 
2615 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2616 {
2617 	/* Setting the bit causes a misconfig. */
2618 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2619 
2620 	/* Making the entry non-present turns reserved bits into ignored. */
2621 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2622 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2623 }
2624 
2625 static void ept_reserved_bit_at_level_huge(int level, int bit)
2626 {
2627 	/* Setting the bit causes a misconfig. */
2628 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2629 
2630 	/* Making the entry non-present turns reserved bits into ignored. */
2631 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2632 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2633 }
2634 
2635 static void ept_reserved_bit_at_level(int level, int bit)
2636 {
2637 	/* Setting the bit causes a misconfig. */
2638 	ept_misconfig_at_level(level, 0, 1ul << bit);
2639 
2640 	/* Making the entry non-present turns reserved bits into ignored. */
2641 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2642 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2643 }
2644 
2645 static void ept_reserved_bit(int bit)
2646 {
2647 	ept_reserved_bit_at_level(1, bit);
2648 	ept_reserved_bit_at_level(2, bit);
2649 	ept_reserved_bit_at_level(3, bit);
2650 	ept_reserved_bit_at_level(4, bit);
2651 }
2652 
2653 #define PAGE_2M_ORDER 9
2654 #define PAGE_1G_ORDER 18
2655 
2656 static void *get_1g_page(void)
2657 {
2658 	static void *alloc;
2659 
2660 	if (!alloc)
2661 		alloc = alloc_pages(PAGE_1G_ORDER);
2662 	return alloc;
2663 }
2664 
2665 static void ept_access_test_teardown(void *unused)
2666 {
2667 	/* Exit the guest cleanly. */
2668 	do_ept_access_op(OP_EXIT);
2669 }
2670 
2671 static void ept_access_test_guest(void)
2672 {
2673 	struct ept_access_test_data *data = &ept_access_test_data;
2674 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2675 
2676 	while (true) {
2677 		switch (data->op) {
2678 		case OP_READ:
2679 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2680 			break;
2681 		case OP_WRITE:
2682 			*data->gva = MAGIC_VAL_2;
2683 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2684 			*data->gva = MAGIC_VAL_1;
2685 			break;
2686 		case OP_EXEC:
2687 			TEST_ASSERT_EQ(42, code());
2688 			break;
2689 		case OP_FLUSH_TLB:
2690 			write_cr3(read_cr3());
2691 			break;
2692 		case OP_EXIT:
2693 			return;
2694 		default:
2695 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2696 		}
2697 		vmcall();
2698 	}
2699 }
2700 
2701 static void ept_access_test_setup(void)
2702 {
2703 	struct ept_access_test_data *data = &ept_access_test_data;
2704 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2705 	unsigned long size = npages * PAGE_SIZE;
2706 	unsigned long *page_table = current_page_table();
2707 	unsigned long pte;
2708 
2709 	if (setup_ept(false))
2710 		test_skip("EPT not supported");
2711 
2712 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2713 	if (cpuid_maxphyaddr() < 40)
2714 		test_skip("Test needs MAXPHYADDR >= 40");
2715 
2716 	test_set_guest(ept_access_test_guest);
2717 	test_add_teardown(ept_access_test_teardown, NULL);
2718 
2719 	data->hva = get_1g_page();
2720 	TEST_ASSERT(data->hva);
2721 	data->hpa = virt_to_phys(data->hva);
2722 
2723 	data->gpa = 1ul << 39;
2724 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2725 				   size);
2726 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2727 	install_pages(page_table, data->gpa, size, data->gva);
2728 
2729 	/*
2730 	 * Make sure nothing's mapped here so the tests that screw with the
2731 	 * pml4 entry don't inadvertently break something.
2732 	 */
2733 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2734 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2735 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2736 
2737 	data->hva[0] = MAGIC_VAL_1;
2738 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2739 }
2740 
2741 static void ept_access_test_not_present(void)
2742 {
2743 	ept_access_test_setup();
2744 	/* --- */
2745 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2746 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2747 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2748 }
2749 
2750 static void ept_access_test_read_only(void)
2751 {
2752 	ept_access_test_setup();
2753 
2754 	/* r-- */
2755 	ept_access_allowed(EPT_RA, OP_READ);
2756 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2757 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2758 }
2759 
2760 static void ept_access_test_write_only(void)
2761 {
2762 	ept_access_test_setup();
2763 	/* -w- */
2764 	ept_access_misconfig(EPT_WA);
2765 }
2766 
2767 static void ept_access_test_read_write(void)
2768 {
2769 	ept_access_test_setup();
2770 	/* rw- */
2771 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2772 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2773 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2774 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2775 }
2776 
2777 
2778 static void ept_access_test_execute_only(void)
2779 {
2780 	ept_access_test_setup();
2781 	/* --x */
2782 	if (ept_execute_only_supported()) {
2783 		ept_access_violation(EPT_EA, OP_READ,
2784 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2785 		ept_access_violation(EPT_EA, OP_WRITE,
2786 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2787 		ept_access_allowed(EPT_EA, OP_EXEC);
2788 	} else {
2789 		ept_access_misconfig(EPT_EA);
2790 	}
2791 }
2792 
2793 static void ept_access_test_read_execute(void)
2794 {
2795 	ept_access_test_setup();
2796 	/* r-x */
2797 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2798 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2799 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2800 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2801 }
2802 
2803 static void ept_access_test_write_execute(void)
2804 {
2805 	ept_access_test_setup();
2806 	/* -wx */
2807 	ept_access_misconfig(EPT_WA | EPT_EA);
2808 }
2809 
2810 static void ept_access_test_read_write_execute(void)
2811 {
2812 	ept_access_test_setup();
2813 	/* rwx */
2814 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2815 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2816 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2817 }
2818 
2819 static void ept_access_test_reserved_bits(void)
2820 {
2821 	int i;
2822 	int maxphyaddr;
2823 
2824 	ept_access_test_setup();
2825 
2826 	/* Reserved bits above maxphyaddr. */
2827 	maxphyaddr = cpuid_maxphyaddr();
2828 	for (i = maxphyaddr; i <= 51; i++) {
2829 		report_prefix_pushf("reserved_bit=%d", i);
2830 		ept_reserved_bit(i);
2831 		report_prefix_pop();
2832 	}
2833 
2834 	/* Level-specific reserved bits. */
2835 	ept_reserved_bit_at_level_nohuge(2, 3);
2836 	ept_reserved_bit_at_level_nohuge(2, 4);
2837 	ept_reserved_bit_at_level_nohuge(2, 5);
2838 	ept_reserved_bit_at_level_nohuge(2, 6);
2839 	/* 2M alignment. */
2840 	for (i = 12; i < 20; i++) {
2841 		report_prefix_pushf("reserved_bit=%d", i);
2842 		ept_reserved_bit_at_level_huge(2, i);
2843 		report_prefix_pop();
2844 	}
2845 	ept_reserved_bit_at_level_nohuge(3, 3);
2846 	ept_reserved_bit_at_level_nohuge(3, 4);
2847 	ept_reserved_bit_at_level_nohuge(3, 5);
2848 	ept_reserved_bit_at_level_nohuge(3, 6);
2849 	/* 1G alignment. */
2850 	for (i = 12; i < 29; i++) {
2851 		report_prefix_pushf("reserved_bit=%d", i);
2852 		ept_reserved_bit_at_level_huge(3, i);
2853 		report_prefix_pop();
2854 	}
2855 	ept_reserved_bit_at_level(4, 3);
2856 	ept_reserved_bit_at_level(4, 4);
2857 	ept_reserved_bit_at_level(4, 5);
2858 	ept_reserved_bit_at_level(4, 6);
2859 	ept_reserved_bit_at_level(4, 7);
2860 }
2861 
2862 static void ept_access_test_ignored_bits(void)
2863 {
2864 	ept_access_test_setup();
2865 	/*
2866 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2867 	 * far as translation is concerned even if AD bits are enabled in the
2868 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2869 	 * control is 0.
2870 	 */
2871 	ept_ignored_bit(8);
2872 	ept_ignored_bit(9);
2873 	ept_ignored_bit(10);
2874 	ept_ignored_bit(11);
2875 	ept_ignored_bit(52);
2876 	ept_ignored_bit(53);
2877 	ept_ignored_bit(54);
2878 	ept_ignored_bit(55);
2879 	ept_ignored_bit(56);
2880 	ept_ignored_bit(57);
2881 	ept_ignored_bit(58);
2882 	ept_ignored_bit(59);
2883 	ept_ignored_bit(60);
2884 	ept_ignored_bit(61);
2885 	ept_ignored_bit(62);
2886 	ept_ignored_bit(63);
2887 }
2888 
2889 static void ept_access_test_paddr_not_present_ad_disabled(void)
2890 {
2891 	ept_access_test_setup();
2892 	ept_disable_ad_bits();
2893 
2894 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
2895 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
2896 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
2897 }
2898 
2899 static void ept_access_test_paddr_not_present_ad_enabled(void)
2900 {
2901 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
2902 
2903 	ept_access_test_setup();
2904 	ept_enable_ad_bits_or_skip_test();
2905 
2906 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
2907 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
2908 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
2909 }
2910 
2911 static void ept_access_test_paddr_read_only_ad_disabled(void)
2912 {
2913 	/*
2914 	 * When EPT AD bits are disabled, all accesses to guest paging
2915 	 * structures are reported separately as a read and (after
2916 	 * translation of the GPA to host physical address) a read+write
2917 	 * if the A/D bits have to be set.
2918 	 */
2919 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2920 
2921 	ept_access_test_setup();
2922 	ept_disable_ad_bits();
2923 
2924 	/* Can't update A bit, so all accesses fail. */
2925 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2926 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
2927 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
2928 	/* AD bits disabled, so only writes try to update the D bit. */
2929 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
2930 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
2931 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
2932 	/* Both A and D already set, so read-only is OK. */
2933 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
2934 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
2935 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
2936 }
2937 
2938 static void ept_access_test_paddr_read_only_ad_enabled(void)
2939 {
2940 	/*
2941 	 * When EPT AD bits are enabled, all accesses to guest paging
2942 	 * structures are considered writes as far as EPT translation
2943 	 * is concerned.
2944 	 */
2945 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2946 
2947 	ept_access_test_setup();
2948 	ept_enable_ad_bits_or_skip_test();
2949 
2950 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2951 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
2952 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
2953 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
2954 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
2955 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
2956 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
2957 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
2958 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
2959 }
2960 
2961 static void ept_access_test_paddr_read_write(void)
2962 {
2963 	ept_access_test_setup();
2964 	/* Read-write access to paging structure. */
2965 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
2966 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
2967 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
2968 }
2969 
2970 static void ept_access_test_paddr_read_write_execute(void)
2971 {
2972 	ept_access_test_setup();
2973 	/* RWX access to paging structure. */
2974 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
2975 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
2976 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
2977 }
2978 
2979 static void ept_access_test_paddr_read_execute_ad_disabled(void)
2980 {
2981   	/*
2982 	 * When EPT AD bits are disabled, all accesses to guest paging
2983 	 * structures are reported separately as a read and (after
2984 	 * translation of the GPA to host physical address) a read+write
2985 	 * if the A/D bits have to be set.
2986 	 */
2987 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
2988 
2989 	ept_access_test_setup();
2990 	ept_disable_ad_bits();
2991 
2992 	/* Can't update A bit, so all accesses fail. */
2993 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
2994 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
2995 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
2996 	/* AD bits disabled, so only writes try to update the D bit. */
2997 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
2998 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
2999 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3000 	/* Both A and D already set, so read-only is OK. */
3001 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3002 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3003 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3004 }
3005 
3006 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3007 {
3008 	/*
3009 	 * When EPT AD bits are enabled, all accesses to guest paging
3010 	 * structures are considered writes as far as EPT translation
3011 	 * is concerned.
3012 	 */
3013 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3014 
3015 	ept_access_test_setup();
3016 	ept_enable_ad_bits_or_skip_test();
3017 
3018 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3019 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3020 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3021 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3022 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3023 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3024 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3025 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3026 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3027 }
3028 
3029 static void ept_access_test_paddr_not_present_page_fault(void)
3030 {
3031 	ept_access_test_setup();
3032 	/*
3033 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3034 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3035 	 * Thus guest page fault before host takes EPT violation for trying to
3036 	 * update A bit.
3037 	 */
3038 }
3039 
3040 static void ept_access_test_force_2m_page(void)
3041 {
3042 	ept_access_test_setup();
3043 
3044 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3045 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3046 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3047 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3048 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3049 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3050 }
3051 
3052 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3053 {
3054 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3055 
3056 	TEST_ASSERT(msr & VPID_CAP_INVVPID);
3057 
3058 	if (type < INVVPID_ADDR || type > INVVPID_CONTEXT_LOCAL)
3059 		return false;
3060 
3061 	if (!(msr & (1ull << (type + VPID_CAP_INVVPID_TYPES_SHIFT))))
3062 		return false;
3063 
3064 	if (vpid >> 16)
3065 		return false;
3066 
3067 	if (type != INVVPID_ALL && !vpid)
3068 		return false;
3069 
3070 	if (type == INVVPID_ADDR && !is_canonical(gla))
3071 		return false;
3072 
3073 	return true;
3074 }
3075 
3076 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3077 {
3078 	int rc;
3079 	bool valid = invvpid_valid(type, vpid, gla);
3080 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3081 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3082 	/*
3083 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3084 	 * that we can tell if it is updated by INVVPID.
3085 	 */
3086 	vmcs_read(~0);
3087 	rc = invvpid(type, vpid, gla);
3088 	report("INVVPID type %ld VPID %lx GLA %lx %s",
3089 	       !rc == valid, type, vpid, gla,
3090 	       valid ? "passes" : "fails");
3091 	report("After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3092 	       vmcs_read(VMX_INST_ERROR) == expected,
3093 	       rc ? "failed" : "successful",
3094 	       expected, vmcs_read(VMX_INST_ERROR));
3095 }
3096 
3097 static void ds_invvpid(void *data)
3098 {
3099 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3100 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3101 
3102 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3103 	asm volatile("invvpid %0, %1"
3104 		     :
3105 		     : "m"(*(struct invvpid_operand *)data),
3106 		       "r"(type));
3107 }
3108 
3109 /*
3110  * The SS override is ignored in 64-bit mode, so we use an addressing
3111  * mode with %rsp as the base register to generate an implicit SS
3112  * reference.
3113  */
3114 static void ss_invvpid(void *data)
3115 {
3116 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3117 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3118 
3119 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3120 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3121 		     : "+r"(data)
3122 		     : "r"(type));
3123 }
3124 
3125 static void invvpid_test_gp(void)
3126 {
3127 	bool fault;
3128 
3129 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3130 				   (void *)NONCANONICAL);
3131 	report("INVVPID with non-canonical DS operand raises #GP", fault);
3132 }
3133 
3134 static void invvpid_test_ss(void)
3135 {
3136 	bool fault;
3137 
3138 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3139 				   (void *)NONCANONICAL);
3140 	report("INVVPID with non-canonical SS operand raises #SS", fault);
3141 }
3142 
3143 static void invvpid_test_pf(void)
3144 {
3145 	void *vpage = alloc_vpage();
3146 	bool fault;
3147 
3148 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3149 	report("INVVPID with unmapped operand raises #PF", fault);
3150 }
3151 
3152 static void try_compat_invvpid(void *unused)
3153 {
3154 	struct far_pointer32 fp = {
3155 		.offset = (uintptr_t)&&invvpid,
3156 		.selector = KERNEL_CS32,
3157 	};
3158 	register uintptr_t rsp asm("rsp");
3159 
3160 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3161 			"Code address too high.");
3162 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3163 
3164 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3165 	return;
3166 invvpid:
3167 	asm volatile (".code32;"
3168 		      "invvpid (%eax), %eax;"
3169 		      "lret;"
3170 		      ".code64");
3171 	__builtin_unreachable();
3172 }
3173 
3174 static void invvpid_test_compatibility_mode(void)
3175 {
3176 	bool fault;
3177 
3178 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3179 	report("Compatibility mode INVVPID raises #UD", fault);
3180 }
3181 
3182 static void invvpid_test_not_in_vmx_operation(void)
3183 {
3184 	bool fault;
3185 
3186 	TEST_ASSERT(!vmx_off());
3187 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3188 	report("INVVPID outside of VMX operation raises #UD", fault);
3189 	TEST_ASSERT(!vmx_on());
3190 }
3191 
3192 /*
3193  * This does not test real-address mode, virtual-8086 mode, protected mode,
3194  * or CPL > 0.
3195  */
3196 static void invvpid_test_v2(void)
3197 {
3198 	u64 msr;
3199 	int i;
3200 	unsigned types = 0;
3201 	unsigned type;
3202 
3203 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
3204 	    !(ctrl_cpu_rev[1].clr & CPU_VPID))
3205 		test_skip("VPID not supported");
3206 
3207 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3208 
3209 	if (!(msr & VPID_CAP_INVVPID))
3210 		test_skip("INVVPID not supported.\n");
3211 
3212 	if (msr & VPID_CAP_INVVPID_ADDR)
3213 		types |= 1u << INVVPID_ADDR;
3214 	if (msr & VPID_CAP_INVVPID_CXTGLB)
3215 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3216 	if (msr & VPID_CAP_INVVPID_ALL)
3217 		types |= 1u << INVVPID_ALL;
3218 	if (msr & VPID_CAP_INVVPID_CXTLOC)
3219 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3220 
3221 	if (!types)
3222 		test_skip("No INVVPID types supported.\n");
3223 
3224 	for (i = -127; i < 128; i++)
3225 		try_invvpid(i, 0xffff, 0);
3226 
3227 	/*
3228 	 * VPID must not be more than 16 bits.
3229 	 */
3230 	for (i = 0; i < 64; i++)
3231 		for (type = 0; type < 4; type++)
3232 			if (types & (1u << type))
3233 				try_invvpid(type, 1ul << i, 0);
3234 
3235 	/*
3236 	 * VPID must not be zero, except for "all contexts."
3237 	 */
3238 	for (type = 0; type < 4; type++)
3239 		if (types & (1u << type))
3240 			try_invvpid(type, 0, 0);
3241 
3242 	/*
3243 	 * The gla operand is only validated for single-address INVVPID.
3244 	 */
3245 	if (types & (1u << INVVPID_ADDR))
3246 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3247 
3248 	invvpid_test_gp();
3249 	invvpid_test_ss();
3250 	invvpid_test_pf();
3251 	invvpid_test_compatibility_mode();
3252 	invvpid_test_not_in_vmx_operation();
3253 }
3254 
3255 /*
3256  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3257  * at least as far as the guest-state checks. Returns false if the
3258  * VMLAUNCH fails early and execution falls through to the next
3259  * instruction.
3260  */
3261 static bool vmlaunch_succeeds(void)
3262 {
3263 	/*
3264 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3265 	 * unsupported VMCS component"). The caller can then check
3266 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3267 	 */
3268 	vmcs_write(~0u, 0);
3269 
3270 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3271 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3272 				   :
3273 				   : "r" ((u64)HOST_RSP)
3274 				   : "cc", "memory"
3275 				   : success);
3276 	return false;
3277 success:
3278 	TEST_ASSERT(vmcs_read(EXI_REASON) ==
3279 		    (VMX_FAIL_STATE | VMX_ENTRY_FAILURE));
3280 	return true;
3281 }
3282 
3283 /*
3284  * Try to launch the current VMCS.
3285  */
3286 static void test_vmx_controls(bool controls_valid, bool xfail)
3287 {
3288 	bool success = vmlaunch_succeeds();
3289 	u32 vmx_inst_err;
3290 
3291 	report_xfail("vmlaunch %s", xfail, success == controls_valid,
3292 		     controls_valid ? "succeeds" : "fails");
3293 	if (!success) {
3294 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3295 		report("VMX inst error is %d (actual %d)",
3296 		       vmx_inst_err == VMXERR_ENTRY_INVALID_CONTROL_FIELD,
3297 		       VMXERR_ENTRY_INVALID_CONTROL_FIELD, vmx_inst_err);
3298 	}
3299 }
3300 
3301 /*
3302  * Test a particular value of a VM-execution control bit, if the value
3303  * is required or if the value is zero.
3304  */
3305 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3306 				    enum Encoding encoding, unsigned bit,
3307 				    unsigned val)
3308 {
3309 	u32 mask = 1u << bit;
3310 	bool expected;
3311 	u32 controls;
3312 
3313 	if (msr.set & mask)
3314 		TEST_ASSERT(msr.clr & mask);
3315 
3316 	/*
3317 	 * We can't arbitrarily turn on a control bit, because it may
3318 	 * introduce dependencies on other VMCS fields. So, we only
3319 	 * test turning on bits that have a required setting.
3320 	 */
3321 	if (val && (msr.clr & mask) && !(msr.set & mask))
3322 		return;
3323 
3324 	report_prefix_pushf("%s %s bit %d",
3325 			    val ? "Set" : "Clear", name, bit);
3326 
3327 	controls = vmcs_read(encoding);
3328 	if (val) {
3329 		vmcs_write(encoding, msr.set | mask);
3330 		expected = (msr.clr & mask);
3331 	} else {
3332 		vmcs_write(encoding, msr.set & ~mask);
3333 		expected = !(msr.set & mask);
3334 	}
3335 	test_vmx_controls(expected, false);
3336 	vmcs_write(encoding, controls);
3337 	report_prefix_pop();
3338 }
3339 
3340 /*
3341  * Test reserved values of a VM-execution control bit, based on the
3342  * allowed bit settings from the corresponding VMX capability MSR.
3343  */
3344 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3345 			      enum Encoding encoding, unsigned bit)
3346 {
3347 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3348 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3349 }
3350 
3351 /*
3352  * Reserved bits in the pin-based VM-execution controls must be set
3353  * properly. Software may consult the VMX capability MSRs to determine
3354  * the proper settings.
3355  * [Intel SDM]
3356  */
3357 static void test_pin_based_ctls(void)
3358 {
3359 	unsigned bit;
3360 
3361 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3362 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3363 	for (bit = 0; bit < 32; bit++)
3364 		test_rsvd_ctl_bit("pin-based controls",
3365 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3366 }
3367 
3368 /*
3369  * Reserved bits in the primary processor-based VM-execution controls
3370  * must be set properly. Software may consult the VMX capability MSRs
3371  * to determine the proper settings.
3372  * [Intel SDM]
3373  */
3374 static void test_primary_processor_based_ctls(void)
3375 {
3376 	unsigned bit;
3377 
3378 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3379 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3380 	for (bit = 0; bit < 32; bit++)
3381 		test_rsvd_ctl_bit("primary processor-based controls",
3382 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3383 }
3384 
3385 /*
3386  * If the "activate secondary controls" primary processor-based
3387  * VM-execution control is 1, reserved bits in the secondary
3388  * processor-based VM-execution controls must be cleared. Software may
3389  * consult the VMX capability MSRs to determine which bits are
3390  * reserved.
3391  * If the "activate secondary controls" primary processor-based
3392  * VM-execution control is 0 (or if the processor does not support the
3393  * 1-setting of that control), no checks are performed on the
3394  * secondary processor-based VM-execution controls.
3395  * [Intel SDM]
3396  */
3397 static void test_secondary_processor_based_ctls(void)
3398 {
3399 	u32 primary;
3400 	u32 secondary;
3401 	unsigned bit;
3402 
3403 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3404 		return;
3405 
3406 	primary = vmcs_read(CPU_EXEC_CTRL0);
3407 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3408 
3409 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3410 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3411 	for (bit = 0; bit < 32; bit++)
3412 		test_rsvd_ctl_bit("secondary processor-based controls",
3413 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3414 
3415 	/*
3416 	 * When the "activate secondary controls" VM-execution control
3417 	 * is clear, there are no checks on the secondary controls.
3418 	 */
3419 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3420 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3421 	report("Secondary processor-based controls ignored",
3422 	       vmlaunch_succeeds());
3423 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3424 	vmcs_write(CPU_EXEC_CTRL0, primary);
3425 }
3426 
3427 static void try_cr3_target_count(unsigned i, unsigned max)
3428 {
3429 	report_prefix_pushf("CR3 target count 0x%x", i);
3430 	vmcs_write(CR3_TARGET_COUNT, i);
3431 	test_vmx_controls(i <= max, false);
3432 	report_prefix_pop();
3433 }
3434 
3435 /*
3436  * The CR3-target count must not be greater than 4. Future processors
3437  * may support a different number of CR3-target values. Software
3438  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3439  * number of values supported.
3440  * [Intel SDM]
3441  */
3442 static void test_cr3_targets(void)
3443 {
3444 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3445 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3446 	unsigned i;
3447 
3448 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3449 	TEST_ASSERT(supported_targets <= 256);
3450 
3451 	try_cr3_target_count(-1u, supported_targets);
3452 	try_cr3_target_count(0x80000000, supported_targets);
3453 	try_cr3_target_count(0x7fffffff, supported_targets);
3454 	for (i = 0; i <= supported_targets + 1; i++)
3455 		try_cr3_target_count(i, supported_targets);
3456 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3457 }
3458 
3459 /*
3460  * Test a particular address setting for a physical page reference in
3461  * the VMCS.
3462  */
3463 static void test_vmcs_page_addr(const char *name,
3464 				enum Encoding encoding,
3465 				bool ignored,
3466 				bool xfail_beyond_mapped_ram,
3467 				u64 addr)
3468 {
3469 	bool xfail =
3470 		(xfail_beyond_mapped_ram &&
3471 		 addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - PAGE_SIZE &&
3472 		 addr < (1ul << cpuid_maxphyaddr()));
3473 
3474 	report_prefix_pushf("%s = %lx", name, addr);
3475 	vmcs_write(encoding, addr);
3476 	test_vmx_controls(ignored || (IS_ALIGNED(addr, PAGE_SIZE) &&
3477 				  addr < (1ul << cpuid_maxphyaddr())),
3478 			  xfail);
3479 	report_prefix_pop();
3480 	xfail = false;
3481 }
3482 
3483 /*
3484  * Test interesting values for a physical page reference in the VMCS.
3485  */
3486 static void test_vmcs_page_values(const char *name,
3487 				  enum Encoding encoding,
3488 				  bool ignored,
3489 				  bool xfail_beyond_mapped_ram)
3490 {
3491 	unsigned i;
3492 	u64 orig_val = vmcs_read(encoding);
3493 
3494 	for (i = 0; i < 64; i++)
3495 		test_vmcs_page_addr(name, encoding, ignored,
3496 				    xfail_beyond_mapped_ram, 1ul << i);
3497 
3498 	test_vmcs_page_addr(name, encoding, ignored,
3499 			    xfail_beyond_mapped_ram, PAGE_SIZE - 1);
3500 	test_vmcs_page_addr(name, encoding, ignored,
3501 			    xfail_beyond_mapped_ram, PAGE_SIZE);
3502 	test_vmcs_page_addr(name, encoding, ignored,
3503 			    xfail_beyond_mapped_ram,
3504 			    (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3505 	test_vmcs_page_addr(name, encoding, ignored,
3506 			    xfail_beyond_mapped_ram,
3507 			    -1ul);
3508 
3509 	vmcs_write(encoding, orig_val);
3510 }
3511 
3512 /*
3513  * Test a physical page reference in the VMCS, when the corresponding
3514  * feature is enabled and when the corresponding feature is disabled.
3515  */
3516 static void test_vmcs_page_reference(u32 control_bit, enum Encoding field,
3517 				     const char *field_name,
3518 				     const char *control_name,
3519 				     bool xfail_beyond_mapped_ram,
3520 				     bool control_primary)
3521 {
3522 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3523 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3524 	u64 page_addr;
3525 
3526 	if (control_primary) {
3527 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3528 			return;
3529 	} else {
3530 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3531 			return;
3532 	}
3533 
3534 	page_addr = vmcs_read(field);
3535 
3536 	report_prefix_pushf("%s enabled", control_name);
3537 	if (control_primary) {
3538 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3539 	} else {
3540 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3541 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3542 	}
3543 	test_vmcs_page_values(field_name, field, false, xfail_beyond_mapped_ram);
3544 	report_prefix_pop();
3545 
3546 	report_prefix_pushf("%s disabled", control_name);
3547 	if (control_primary) {
3548 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3549 	} else {
3550 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3551 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3552 	}
3553 	test_vmcs_page_values(field_name, field, true, false);
3554 	report_prefix_pop();
3555 
3556 	vmcs_write(field, page_addr);
3557 	vmcs_write(CPU_EXEC_CTRL0, primary);
3558 }
3559 
3560 /*
3561  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3562  * each I/O-bitmap address must be 0. Neither address should set any
3563  * bits beyond the processor's physical-address width.
3564  * [Intel SDM]
3565  */
3566 static void test_io_bitmaps(void)
3567 {
3568 	test_vmcs_page_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3569 				 "I/O bitmap A", "Use I/O bitmaps", false,
3570 				 true);
3571 	test_vmcs_page_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3572 				 "I/O bitmap B", "Use I/O bitmaps", false,
3573 				 true);
3574 }
3575 
3576 /*
3577  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3578  * the MSR-bitmap address must be 0. The address should not set any
3579  * bits beyond the processor's physical-address width.
3580  * [Intel SDM]
3581  */
3582 static void test_msr_bitmap(void)
3583 {
3584 	test_vmcs_page_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3585 				 "MSR bitmap", "Use MSR bitmaps", false,
3586 				 true);
3587 }
3588 
3589 /*
3590  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3591  * address must satisfy the following checks:
3592  * - Bits 11:0 of the address must be 0.
3593  * - The address should not set any bits beyond the processor's
3594  *   physical-address width.
3595  * [Intel SDM]
3596  */
3597 static void test_apic_virt_addr(void)
3598 {
3599 	test_vmcs_page_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3600 				 "virtual-APIC address", "Use TPR shadow",
3601 				 true, true);
3602 }
3603 
3604 /*
3605  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3606  * APIC-access address must satisfy the following checks:
3607  *  - Bits 11:0 of the address must be 0.
3608  *  - The address should not set any bits beyond the processor's
3609  *    physical-address width.
3610  * [Intel SDM]
3611  */
3612 static void test_apic_access_addr(void)
3613 {
3614 	void *apic_access_page = alloc_page();
3615 
3616 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3617 
3618 	test_vmcs_page_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3619 				 "APIC-access address",
3620 				 "virtualize APIC-accesses", false, false);
3621 }
3622 
3623 static bool set_bit_pattern(u8 mask, u32 *secondary)
3624 {
3625 	u8 i;
3626 	bool flag = false;
3627 	u32 test_bits[3] = {
3628 		CPU_VIRT_X2APIC,
3629 		CPU_APIC_REG_VIRT,
3630 		CPU_VINTD
3631 	};
3632 
3633         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3634 		if ((mask & (1u << i)) &&
3635 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3636 			*secondary |= test_bits[i];
3637 			flag = true;
3638 		}
3639 	}
3640 
3641 	return (flag);
3642 }
3643 
3644 /*
3645  * If the "use TPR shadow" VM-execution control is 0, the following
3646  * VM-execution controls must also be 0:
3647  * 	- virtualize x2APIC mode
3648  *	- APIC-register virtualization
3649  *	- virtual-interrupt delivery
3650  *    [Intel SDM]
3651  *
3652  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3653  *    "virtualize APIC accesses" VM-execution control must be 0.
3654  *    [Intel SDM]
3655  */
3656 static void test_apic_virtual_ctls(void)
3657 {
3658 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3659 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3660 	u32 primary = saved_primary;
3661 	u32 secondary = saved_secondary;
3662 	bool ctrl = false;
3663 	char str[10] = "disabled";
3664 	u8 i = 0, j;
3665 
3666 	/*
3667 	 * First test
3668 	 */
3669 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3670 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3671 		return;
3672 
3673 	primary |= CPU_SECONDARY;
3674 	primary &= ~CPU_TPR_SHADOW;
3675 	vmcs_write(CPU_EXEC_CTRL0, primary);
3676 
3677 	while (1) {
3678 		for (j = 1; j < 8; j++) {
3679 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3680 			if (primary & CPU_TPR_SHADOW) {
3681 				ctrl = true;
3682 			} else {
3683 				if (! set_bit_pattern(j, &secondary))
3684 					ctrl = true;
3685 				else
3686 					ctrl = false;
3687 			}
3688 
3689 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3690 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3691 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3692 			test_vmx_controls(ctrl, false);
3693 			report_prefix_pop();
3694 		}
3695 
3696 		if (i == 1)
3697 			break;
3698 		i++;
3699 
3700 		primary |= CPU_TPR_SHADOW;
3701 		vmcs_write(CPU_EXEC_CTRL0, primary);
3702 		strcpy(str, "enabled");
3703 	}
3704 
3705 	/*
3706 	 * Second test
3707 	 */
3708 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3709 
3710 	primary = saved_primary;
3711 	secondary = saved_secondary;
3712 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3713 		return;
3714 
3715 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3716 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3717 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3718 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3719 	test_vmx_controls(true, false);
3720 	report_prefix_pop();
3721 
3722 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3723 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3724 	test_vmx_controls(true, false);
3725 	report_prefix_pop();
3726 
3727 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3728 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3729 	test_vmx_controls(false, false);
3730 	report_prefix_pop();
3731 
3732 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3733 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3734 	test_vmx_controls(true, false);
3735 	report_prefix_pop();
3736 
3737 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3738 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3739 }
3740 
3741 /*
3742  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3743  * "external-interrupt exiting" VM-execution control must be 1.
3744  * [Intel SDM]
3745  */
3746 static void test_virtual_intr_ctls(void)
3747 {
3748 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3749 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3750 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3751 	u32 primary = saved_primary;
3752 	u32 secondary = saved_secondary;
3753 	u32 pin = saved_pin;
3754 
3755 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3756 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
3757 		return;
3758 
3759 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3760 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
3761 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3762 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
3763 	test_vmx_controls(true, false);
3764 	report_prefix_pop();
3765 
3766 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
3767 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3768 	test_vmx_controls(false, false);
3769 	report_prefix_pop();
3770 
3771 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
3772 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
3773 	test_vmx_controls(true, false);
3774 	report_prefix_pop();
3775 
3776 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3777 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3778 	test_vmx_controls(false, false);
3779 	report_prefix_pop();
3780 
3781 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3782 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3783 	vmcs_write(PIN_CONTROLS, saved_pin);
3784 }
3785 
3786 static void test_pi_desc_addr(u64 addr, bool ctrl)
3787 {
3788 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
3789 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
3790 	test_vmx_controls(ctrl, false);
3791 	report_prefix_pop();
3792 }
3793 
3794 /*
3795  * If the “process posted interrupts†VM-execution control is 1, the
3796  * following must be true:
3797  *
3798  *	- The “virtual-interrupt delivery†VM-execution control is 1.
3799  *	- The “acknowledge interrupt on exit†VM-exit control is 1.
3800  *	- The posted-interrupt notification vector has a value in the
3801  *	- range 0–255 (bits 15:8 are all 0).
3802  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
3803  *	- The posted-interrupt descriptor address does not set any bits
3804  *	  beyond the processor's physical-address width.
3805  * [Intel SDM]
3806  */
3807 static void test_posted_intr(void)
3808 {
3809 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3810 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3811 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3812 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
3813 	u32 primary = saved_primary;
3814 	u32 secondary = saved_secondary;
3815 	u32 pin = saved_pin;
3816 	u32 exit_ctl = exit_ctl_saved;
3817 	u16 vec;
3818 	int i;
3819 
3820 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
3821 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3822 	    (ctrl_exit_rev.clr & EXI_INTA)))
3823 		return;
3824 
3825 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3826 
3827 	/*
3828 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
3829 	 */
3830 	pin |= PIN_POST_INTR;
3831 	vmcs_write(PIN_CONTROLS, pin);
3832 	secondary &= ~CPU_VINTD;
3833 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3834 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
3835 	test_vmx_controls(false, false);
3836 	report_prefix_pop();
3837 
3838 	secondary |= CPU_VINTD;
3839 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3840 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
3841 	test_vmx_controls(false, false);
3842 	report_prefix_pop();
3843 
3844 	exit_ctl &= ~EXI_INTA;
3845 	vmcs_write(EXI_CONTROLS, exit_ctl);
3846 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
3847 	test_vmx_controls(false, false);
3848 	report_prefix_pop();
3849 
3850 	exit_ctl |= EXI_INTA;
3851 	vmcs_write(EXI_CONTROLS, exit_ctl);
3852 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3853 	test_vmx_controls(true, false);
3854 	report_prefix_pop();
3855 
3856 	secondary &= ~CPU_VINTD;
3857 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3858 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
3859 	test_vmx_controls(false, false);
3860 	report_prefix_pop();
3861 
3862 	secondary |= CPU_VINTD;
3863 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3864 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3865 	test_vmx_controls(true, false);
3866 	report_prefix_pop();
3867 
3868 	/*
3869 	 * Test posted-interrupt notification vector
3870 	 */
3871 	for (i = 0; i < 8; i++) {
3872 		vec = (1ul << i);
3873 		vmcs_write(PINV, vec);
3874 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3875 		test_vmx_controls(true, false);
3876 		report_prefix_pop();
3877 	}
3878 	for (i = 8; i < 16; i++) {
3879 		vec = (1ul << i);
3880 		vmcs_write(PINV, vec);
3881 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3882 		test_vmx_controls(false, false);
3883 		report_prefix_pop();
3884 	}
3885 
3886 	vec &= ~(0xff << 8);
3887 	vmcs_write(PINV, vec);
3888 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3889 	test_vmx_controls(true, false);
3890 	report_prefix_pop();
3891 
3892 	/*
3893 	 * Test posted-interrupt descriptor addresss
3894 	 */
3895 	for (i = 0; i < 6; i++) {
3896 		test_pi_desc_addr(1ul << i, false);
3897 	}
3898 
3899 	test_pi_desc_addr(0xf0, false);
3900 	test_pi_desc_addr(0xff, false);
3901 	test_pi_desc_addr(0x0f, false);
3902 	test_pi_desc_addr(0x8000, true);
3903 	test_pi_desc_addr(0x00, true);
3904 	test_pi_desc_addr(0xc000, true);
3905 
3906 	test_vmcs_page_values("process-posted interrupts",
3907 	    POSTED_INTR_DESC_ADDR, false, false);
3908 
3909 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3910 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3911 	vmcs_write(PIN_CONTROLS, saved_pin);
3912 }
3913 
3914 static void test_apic_ctls(void)
3915 {
3916 	test_apic_virt_addr();
3917 	test_apic_access_addr();
3918 	test_apic_virtual_ctls();
3919 	test_virtual_intr_ctls();
3920 	test_posted_intr();
3921 }
3922 
3923 /*
3924  * If the “enable VPID†VM-execution control is 1, the value of the
3925  * of the VPID VM-execution control field must not be 0000H.
3926  * [Intel SDM]
3927  */
3928 static void test_vpid(void)
3929 {
3930 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3931 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3932 	u16 vpid = 0x0000;
3933 	int i;
3934 
3935 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
3936 	    (ctrl_cpu_rev[1].clr & CPU_VPID))) {
3937 		test_skip("Secondary controls and/or VPID not supported");
3938 		return;
3939 	}
3940 
3941 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
3942 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
3943 	vmcs_write(VPID, vpid);
3944 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
3945 	test_vmx_controls(true, false);
3946 	report_prefix_pop();
3947 
3948 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
3949 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
3950 	test_vmx_controls(false, false);
3951 	report_prefix_pop();
3952 
3953 	for (i = 0; i < 16; i++) {
3954 		vpid = (short)1 << i;;
3955 		vmcs_write(VPID, vpid);
3956 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
3957 		test_vmx_controls(true, false);
3958 		report_prefix_pop();
3959 	}
3960 
3961 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3962 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3963 }
3964 
3965 static void set_vtpr(unsigned vtpr)
3966 {
3967 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
3968 }
3969 
3970 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
3971 {
3972 	bool valid = true;
3973 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3974 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3975 
3976 	if ((primary & CPU_TPR_SHADOW) &&
3977 	    (!(primary & CPU_SECONDARY) ||
3978 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
3979 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
3980 
3981 	set_vtpr(vtpr);
3982 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
3983 	    threshold, (vtpr >> 4) & 0xf);
3984 	test_vmx_controls(valid, false);
3985 	report_prefix_pop();
3986 }
3987 
3988 static void test_invalid_event_injection(void)
3989 {
3990 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
3991 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
3992 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
3993 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
3994 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
3995 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
3996 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
3997 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
3998 	u32 cnt;
3999 
4000 	/* Setup */
4001 	report_prefix_push("invalid event injection");
4002 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4003 	vmcs_write(ENT_INST_LEN, 0x00000001);
4004 
4005 	/* The field’s interruption type is not set to a reserved value. */
4006 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4007 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4008 			    "RESERVED interruption type invalid [-]",
4009 			    ent_intr_info);
4010 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4011 	test_vmx_controls(false, false);
4012 	report_prefix_pop();
4013 
4014 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4015 			DE_VECTOR;
4016 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4017 			    "RESERVED interruption type invalid [+]",
4018 			    ent_intr_info);
4019 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4020 	test_vmx_controls(true, false);
4021 	report_prefix_pop();
4022 
4023 	/* If the interruption type is other event, the vector is 0. */
4024 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4025 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4026 			    "(OTHER EVENT && vector != 0) invalid [-]",
4027 			    ent_intr_info);
4028 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4029 	test_vmx_controls(false, false);
4030 	report_prefix_pop();
4031 
4032 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4033 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4034 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4035 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4036 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4037 	test_vmx_controls(false, false);
4038 	report_prefix_pop();
4039 
4040 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4041 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4042 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4043 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4044 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4045 	test_vmx_controls(true, false);
4046 	report_prefix_pop();
4047 
4048 	/*
4049 	 * If the interruption type
4050 	 * is HW exception, the vector is at most 31.
4051 	 */
4052 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4053 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4054 			    "(HW exception && vector > 31) invalid [-]",
4055 			    ent_intr_info);
4056 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4057 	test_vmx_controls(false, false);
4058 	report_prefix_pop();
4059 
4060 	/*
4061 	 * deliver-error-code is 1 iff either
4062 	 * (a) the "unrestricted guest" VM-execution control is 0
4063 	 * (b) CR0.PE is set.
4064 	 */
4065 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4066 			GP_VECTOR;
4067 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4068 			    "error code <-> (!URG || prot_mode) [-]",
4069 			    ent_intr_info);
4070 	disable_unrestricted_guest();
4071 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4072 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4073 	test_vmx_controls(false, false);
4074 	report_prefix_pop();
4075 
4076 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4077 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4078 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4079 			    "error code <-> (!URG || prot_mode) [+]",
4080 			    ent_intr_info);
4081 	disable_unrestricted_guest();
4082 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4083 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4084 	test_vmx_controls(true, false);
4085 	report_prefix_pop();
4086 
4087 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4088 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4089 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4090 			    "error code <-> (!URG || prot_mode) [-]",
4091 			    ent_intr_info);
4092 	enable_unrestricted_guest();
4093 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4094 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4095 	test_vmx_controls(false, false);
4096 	report_prefix_pop();
4097 
4098 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4099 			GP_VECTOR;
4100 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4101 			    "error code <-> (!URG || prot_mode) [-]",
4102 			    ent_intr_info);
4103 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4104 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4105 	test_vmx_controls(false, false);
4106 	report_prefix_pop();
4107 
4108 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4109 	report_prefix_push("error code <-> HW exception");
4110 	for (cnt = 0; cnt < 8; cnt++) {
4111 		u32 exception_type_mask = cnt << 8;
4112 		u32 deliver_error_code_mask =
4113 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4114 			INTR_INFO_DELIVER_CODE_MASK : 0;
4115 
4116 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4117 				exception_type_mask | GP_VECTOR;
4118 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4119 				    ent_intr_info);
4120 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4121 		test_vmx_controls(false, false);
4122 		report_prefix_pop();
4123 	}
4124 	report_prefix_pop();
4125 
4126 	/*
4127 	 * deliver-error-code is 1 iff the the vector
4128 	 * indicates an exception that would normally deliver an error code
4129 	 */
4130 	report_prefix_push("error code <-> vector delivers error code");
4131 	for (cnt = 0; cnt < 32; cnt++) {
4132 		bool has_error_code = false;
4133 		u32 deliver_error_code_mask;
4134 
4135 		switch (cnt) {
4136 		case DF_VECTOR:
4137 		case TS_VECTOR:
4138 		case NP_VECTOR:
4139 		case SS_VECTOR:
4140 		case GP_VECTOR:
4141 		case PF_VECTOR:
4142 		case AC_VECTOR:
4143 			has_error_code = true;
4144 		}
4145 
4146 		/* Negative case */
4147 		deliver_error_code_mask = has_error_code ?
4148 						0 :
4149 						INTR_INFO_DELIVER_CODE_MASK;
4150 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4151 				INTR_TYPE_HARD_EXCEPTION | cnt;
4152 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4153 				    ent_intr_info);
4154 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4155 		test_vmx_controls(false, false);
4156 		report_prefix_pop();
4157 
4158 		/* Positive case */
4159 		deliver_error_code_mask = has_error_code ?
4160 						INTR_INFO_DELIVER_CODE_MASK :
4161 						0;
4162 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4163 				INTR_TYPE_HARD_EXCEPTION | cnt;
4164 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4165 				    ent_intr_info);
4166 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4167 		test_vmx_controls(true, false);
4168 		report_prefix_pop();
4169 	}
4170 	report_prefix_pop();
4171 
4172 	/* Reserved bits in the field (30:12) are 0. */
4173 	report_prefix_push("reserved bits clear");
4174 	for (cnt = 12; cnt <= 30; cnt++) {
4175 		ent_intr_info = ent_intr_info_base |
4176 				INTR_INFO_DELIVER_CODE_MASK |
4177 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4178 				(1U << cnt);
4179 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4180 				    ent_intr_info);
4181 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4182 		test_vmx_controls(false, false);
4183 		report_prefix_pop();
4184 	}
4185 	report_prefix_pop();
4186 
4187 	/*
4188 	 * If deliver-error-code is 1
4189 	 * bits 31:15 of the VM-entry exception error-code field are 0.
4190 	 */
4191 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4192 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4193 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4194 			    "VM-entry exception error code[31:15] clear",
4195 			    ent_intr_info);
4196 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4197 	for (cnt = 15; cnt <= 31; cnt++) {
4198 		ent_intr_err = 1U << cnt;
4199 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4200 				    ent_intr_err);
4201 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4202 		test_vmx_controls(false, false);
4203 		report_prefix_pop();
4204 	}
4205 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4206 	report_prefix_pop();
4207 
4208 	/*
4209 	 * If the interruption type is software interrupt, software exception,
4210 	 * or privileged software exception, the VM-entry instruction-length
4211 	 * field is in the range 0–15.
4212 	 */
4213 
4214 	for (cnt = 0; cnt < 3; cnt++) {
4215 		switch (cnt) {
4216 		case 0:
4217 			ent_intr_info = ent_intr_info_base |
4218 					INTR_TYPE_SOFT_INTR;
4219 			break;
4220 		case 1:
4221 			ent_intr_info = ent_intr_info_base |
4222 					INTR_TYPE_SOFT_EXCEPTION;
4223 			break;
4224 		case 2:
4225 			ent_intr_info = ent_intr_info_base |
4226 					INTR_TYPE_PRIV_SW_EXCEPTION;
4227 			break;
4228 		}
4229 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4230 				    "VM-entry instruction-length check",
4231 				    ent_intr_info);
4232 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4233 
4234 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4235 		ent_intr_len = -1;
4236 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4237 				    ent_intr_len);
4238 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4239 		test_vmx_controls(false, false);
4240 		report_prefix_pop();
4241 
4242 		/* Instruction length set to 16 should fail */
4243 		ent_intr_len = 0x00000010;
4244 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4245 				    ent_intr_len);
4246 		vmcs_write(ENT_INST_LEN, 0x00000010);
4247 		test_vmx_controls(false, false);
4248 		report_prefix_pop();
4249 
4250 		report_prefix_pop();
4251 	}
4252 
4253 	/* Cleanup */
4254 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4255 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4256 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4257 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4258 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4259 	vmcs_write(GUEST_CR0, guest_cr0_save);
4260 	report_prefix_pop();
4261 }
4262 
4263 /*
4264  * Test interesting vTPR values for a given TPR threshold.
4265  */
4266 static void test_vtpr_values(unsigned threshold)
4267 {
4268 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4269 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4270 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4271 }
4272 
4273 static void try_tpr_threshold(unsigned threshold)
4274 {
4275 	bool valid = true;
4276 
4277 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4278 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4279 
4280 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4281 	    (secondary & CPU_VINTD)))
4282 		valid = !(threshold >> 4);
4283 
4284 	set_vtpr(-1);
4285 	vmcs_write(TPR_THRESHOLD, threshold);
4286 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4287 	test_vmx_controls(valid, false);
4288 	report_prefix_pop();
4289 
4290 	if (valid)
4291 		test_vtpr_values(threshold);
4292 }
4293 
4294 /*
4295  * Test interesting TPR threshold values.
4296  */
4297 static void test_tpr_threshold_values(void)
4298 {
4299 	unsigned i;
4300 
4301 	for (i = 0; i < 0x10; i++)
4302 		try_tpr_threshold(i);
4303 	for (i = 4; i < 32; i++)
4304 		try_tpr_threshold(1u << i);
4305 	try_tpr_threshold(-1u);
4306 	try_tpr_threshold(0x7fffffff);
4307 }
4308 
4309 /*
4310  * This test covers the following two VM entry checks:
4311  *
4312  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4313  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4314  *         31:4 of the TPR threshold VM-execution control field must
4315 	   be 0.
4316  *         [Intel SDM]
4317  *
4318  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4319  *          "virtual-interrupt delivery" VM-execution control is 0
4320  *          and the "virtualize APIC accesses" VM-execution control
4321  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4322  *          control field must not be greater than the value of bits
4323  *          7:4 of VTPR.
4324  *          [Intel SDM]
4325  */
4326 static void test_tpr_threshold(void)
4327 {
4328 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4329 	void *virtual_apic_page;
4330 
4331 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4332 		return;
4333 
4334 	virtual_apic_page = alloc_page();
4335 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4336 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4337 
4338 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4339 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4340 	test_tpr_threshold_values();
4341 	report_prefix_pop();
4342 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4343 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4344 	test_tpr_threshold_values();
4345 	report_prefix_pop();
4346 
4347 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4348 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES)))) {
4349 		vmcs_write(CPU_EXEC_CTRL0, primary);
4350 		return;
4351 	}
4352 
4353 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4354 
4355 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4356 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4357 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4358 		test_tpr_threshold_values();
4359 		report_prefix_pop();
4360 
4361 		vmcs_write(CPU_EXEC_CTRL0,
4362 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4363 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4364 		test_tpr_threshold_values();
4365 		report_prefix_pop();
4366 	}
4367 
4368 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4369 		vmcs_write(CPU_EXEC_CTRL0,
4370 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4371 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4372 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4373 		test_tpr_threshold_values();
4374 		report_prefix_pop();
4375 
4376 		vmcs_write(CPU_EXEC_CTRL0,
4377 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4378 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4379 		test_tpr_threshold_values();
4380 		report_prefix_pop();
4381 	}
4382 
4383 	if ((ctrl_cpu_rev[1].clr &
4384 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4385 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4386 		vmcs_write(CPU_EXEC_CTRL0,
4387 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4388 		vmcs_write(CPU_EXEC_CTRL1,
4389 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4390 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4391 		test_tpr_threshold_values();
4392 		report_prefix_pop();
4393 
4394 		vmcs_write(CPU_EXEC_CTRL0,
4395 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4396 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4397 		test_tpr_threshold_values();
4398 		report_prefix_pop();
4399 	}
4400 
4401 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4402 	vmcs_write(CPU_EXEC_CTRL0, primary);
4403 }
4404 
4405 /*
4406  * This test verifies the following two vmentry checks:
4407  *
4408  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4409  *  VM-execution control must be 0.
4410  *  [Intel SDM]
4411  *
4412  *  If the “virtual NMIs” VM-execution control is 0, the “NMI-window
4413  *  exiting” VM-execution control must be 0.
4414  *  [Intel SDM]
4415  */
4416 static void test_nmi_ctrls(void)
4417 {
4418 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4419 
4420 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4421 	    (PIN_NMI | PIN_VIRT_NMI)) {
4422 		test_skip("NMI exiting and Virtual NMIs are not supported !");
4423 		return;
4424 	}
4425 
4426 	/* Save the controls so that we can restore them after our tests */
4427 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4428 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4429 
4430 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4431 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4432 
4433 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4434 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4435 	test_vmx_controls(true, false);
4436 	report_prefix_pop();
4437 
4438 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4439 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4440 	test_vmx_controls(false, false);
4441 	report_prefix_pop();
4442 
4443 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4444 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4445 	test_vmx_controls(true, false);
4446 	report_prefix_pop();
4447 
4448 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4449 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4450 	test_vmx_controls(true, false);
4451 	report_prefix_pop();
4452 
4453 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4454 		report_info("NMI-window exiting is not supported, skipping...");
4455 		goto done;
4456 	}
4457 
4458 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4459 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4460 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4461 	test_vmx_controls(false, false);
4462 	report_prefix_pop();
4463 
4464 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4465 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4466 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4467 	test_vmx_controls(true, false);
4468 	report_prefix_pop();
4469 
4470 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4471 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4472 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4473 	test_vmx_controls(true, false);
4474 	report_prefix_pop();
4475 
4476 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4477 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4478 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4479 	test_vmx_controls(true, false);
4480 	report_prefix_pop();
4481 
4482 	/* Restore the controls to their original values */
4483 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4484 done:
4485 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4486 }
4487 
4488 static void test_eptp_ad_bit(u64 eptp, bool ctrl)
4489 {
4490 	vmcs_write(EPTP, eptp);
4491 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4492 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4493 	test_vmx_controls(ctrl, false);
4494 	report_prefix_pop();
4495 
4496 }
4497 /*
4498  * If the “enable EPT†VM-execution control is 1, the EPTP VM-execution
4499  * control field must satisfy the following checks:
4500  *
4501  *   —  The EPT memory type (bits 2:0) must be a value supported by the
4502  *	processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4503  *   —  Bits 5:3 (1 less than the EPT page-walk length) must be 3,
4504  *	indicating an EPT page-walk length of 4.
4505  *   —  Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4506  *	0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4507  *	indicating that the processor does not support accessed and dirty
4508  *	dirty flags for EPT.
4509  *   —  Reserved bits 11:7 and 63:N (where N is the processor’s
4510  *	physical-address width) must all be 0.
4511  *
4512  *  [Intel SDM]
4513  */
4514 static void test_eptp(void)
4515 {
4516 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4517 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4518 	u64 eptp_saved = vmcs_read(EPTP);
4519 	u32 primary = primary_saved;
4520 	u32 secondary = secondary_saved;
4521 	u64 msr, eptp = eptp_saved;
4522 	bool un_cache = false;
4523 	bool wr_bk = false;
4524 	bool ctrl;
4525 	u32 i, maxphysaddr;
4526 	u64 j, resv_bits_mask = 0;
4527 
4528 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4529 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4530 		test_skip("\"CPU secondary\" and/or \"enable EPT\" execution controls are not supported !");
4531 		return;
4532 	}
4533 
4534 	/*
4535 	 * Memory type (bits 2:0)
4536 	 */
4537 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
4538 	if (msr & EPT_CAP_UC)
4539 		un_cache = true;
4540 	if (msr & EPT_CAP_WB)
4541 		wr_bk = true;
4542 
4543 	primary |= CPU_SECONDARY;
4544 	vmcs_write(CPU_EXEC_CTRL0, primary);
4545 	secondary |= CPU_EPT;
4546 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4547 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4548 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4549 	vmcs_write(EPTP, eptp);
4550 
4551 	for (i = 0; i < 8; i++) {
4552 		if (i == 0) {
4553 			if (un_cache) {
4554 				report_info("EPT paging structure memory-type is Un-cacheable\n");
4555 				ctrl = true;
4556 			} else {
4557 				ctrl = false;
4558 			}
4559 		} else if (i == 6) {
4560 			if (wr_bk) {
4561 				report_info("EPT paging structure memory-type is Write-back\n");
4562 				ctrl = true;
4563 			} else {
4564 				ctrl = false;
4565 			}
4566 		} else {
4567 			ctrl = false;
4568 		}
4569 
4570 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4571 		vmcs_write(EPTP, eptp);
4572 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4573 		    eptp & EPT_MEM_TYPE_MASK);
4574 		test_vmx_controls(ctrl, false);
4575 		report_prefix_pop();
4576 	}
4577 
4578 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4579 
4580 	/*
4581 	 * Page walk length (bits 5:3)
4582 	 */
4583 	for (i = 0; i < 8; i++) {
4584 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4585 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4586 		if (i == 3)
4587 			ctrl = true;
4588 		else
4589 			ctrl = false;
4590 
4591 		vmcs_write(EPTP, eptp);
4592 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4593 		    eptp & EPTP_PG_WALK_LEN_MASK);
4594 		test_vmx_controls(ctrl, false);
4595 		report_prefix_pop();
4596 	}
4597 
4598 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4599 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4600 
4601 	/*
4602 	 * Accessed and dirty flag (bit 6)
4603 	 */
4604 	if (msr & EPT_CAP_AD_FLAG) {
4605 		report_info("Processor supports accessed and dirty flag");
4606 		eptp &= ~EPTP_AD_FLAG;
4607 		test_eptp_ad_bit(eptp, true);
4608 
4609 		eptp |= EPTP_AD_FLAG;
4610 		test_eptp_ad_bit(eptp, true);
4611 	} else {
4612 		report_info("Processor does not supports accessed and dirty flag");
4613 		eptp &= ~EPTP_AD_FLAG;
4614 		test_eptp_ad_bit(eptp, true);
4615 
4616 		eptp |= EPTP_AD_FLAG;
4617 		test_eptp_ad_bit(eptp, false);
4618 	}
4619 
4620 	/*
4621 	 * Reserved bits [11:7] and [63:N]
4622 	 */
4623 	for (i = 0; i < 32; i++) {
4624 		if (i  == 0)
4625 			ctrl = true;
4626 		else
4627 			ctrl = false;
4628 
4629 		eptp = (eptp &
4630 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4631 		    (i << EPTP_RESERV_BITS_SHIFT);
4632 		vmcs_write(EPTP, eptp);
4633 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4634 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4635 		    EPTP_RESERV_BITS_MASK);
4636 		test_vmx_controls(ctrl, false);
4637 		report_prefix_pop();
4638 	}
4639 
4640 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4641 
4642 	maxphysaddr = cpuid_maxphyaddr();
4643 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4644 		resv_bits_mask |= 1ul << i;
4645 	}
4646 
4647 	for (j = 0; j < (63 - maxphysaddr + 1); j++) {
4648 		if (j  == 0)
4649 			ctrl = true;
4650 		else
4651 			ctrl = false;
4652 
4653 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4654 		    (j << maxphysaddr);
4655 		vmcs_write(EPTP, eptp);
4656 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4657 		    (eptp >> maxphysaddr) & resv_bits_mask);
4658 		test_vmx_controls(ctrl, false);
4659 		report_prefix_pop();
4660 	}
4661 
4662 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4663 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4664 	vmcs_write(EPTP, eptp_saved);
4665 }
4666 
4667 /*
4668  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4669  * VM-execution control must also be 1. In addition, the PML address
4670  * must satisfy the following checks:
4671  *
4672  *    * Bits 11:0 of the address must be 0.
4673  *    * The address should not set any bits beyond the processor's
4674  *	physical-address width.
4675  *
4676  *  [Intel SDM]
4677  */
4678 static void test_pml(void)
4679 {
4680 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4681 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4682 	u32 primary = primary_saved;
4683 	u32 secondary = secondary_saved;
4684 
4685 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4686 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
4687 		test_skip("\"Secondary execution\" control or \"enable EPT\" control or \"enable PML\" control is not supported !");
4688 		return;
4689 	}
4690 
4691 	primary |= CPU_SECONDARY;
4692 	vmcs_write(CPU_EXEC_CTRL0, primary);
4693 	secondary &= ~(CPU_PML | CPU_EPT);
4694 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4695 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
4696 	test_vmx_controls(true, false);
4697 	report_prefix_pop();
4698 
4699 	secondary |= CPU_PML;
4700 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4701 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
4702 	test_vmx_controls(false, false);
4703 	report_prefix_pop();
4704 
4705 	secondary |= CPU_EPT;
4706 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4707 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
4708 	test_vmx_controls(true, false);
4709 	report_prefix_pop();
4710 
4711 	secondary &= ~CPU_PML;
4712 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4713 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
4714 	test_vmx_controls(true, false);
4715 	report_prefix_pop();
4716 
4717 	test_vmcs_page_reference(CPU_PML, PMLADDR, "PML address",
4718 	    "PML", false, false);
4719 
4720 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4721 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4722 }
4723 
4724 /*
4725  * Check that the virtual CPU checks all of the VMX controls as
4726  * documented in the Intel SDM.
4727  */
4728 static void vmx_controls_test(void)
4729 {
4730 	/*
4731 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
4732 	 * fail due to invalid guest state, should we make it that
4733 	 * far.
4734 	 */
4735 	vmcs_write(GUEST_RFLAGS, 0);
4736 
4737 	test_pin_based_ctls();
4738 	test_primary_processor_based_ctls();
4739 	test_secondary_processor_based_ctls();
4740 	test_cr3_targets();
4741 	test_io_bitmaps();
4742 	test_msr_bitmap();
4743 	test_apic_ctls();
4744 	test_tpr_threshold();
4745 	test_nmi_ctrls();
4746 	test_pml();
4747 	test_invalid_event_injection();
4748 	test_vpid();
4749 	test_eptp();
4750 }
4751 
4752 static bool valid_vmcs_for_vmentry(void)
4753 {
4754 	struct vmcs *current_vmcs = NULL;
4755 
4756 	if (vmcs_save(&current_vmcs))
4757 		return false;
4758 
4759 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
4760 }
4761 
4762 static void try_vmentry_in_movss_shadow(void)
4763 {
4764 	u32 vm_inst_err;
4765 	u32 flags;
4766 	bool early_failure = false;
4767 	u32 expected_flags = X86_EFLAGS_FIXED;
4768 	bool valid_vmcs = valid_vmcs_for_vmentry();
4769 
4770 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
4771 
4772 	/*
4773 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
4774 	 * unsupported VMCS component").
4775 	 */
4776 	vmcs_write(~0u, 0);
4777 
4778 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
4779 			      "vmwrite %%rsp, %%rdx;"
4780 			      "mov 0f, %%rax;"
4781 			      "mov %[host_rip], %%edx;"
4782 			      "vmwrite %%rax, %%rdx;"
4783 			      "mov $-1, %%ah;"
4784 			      "sahf;"
4785 			      "mov %%ss, %%ax;"
4786 			      "mov %%ax, %%ss;"
4787 			      "vmlaunch;"
4788 			      "mov $1, %[early_failure];"
4789 			      "0: lahf;"
4790 			      "movzbl %%ah, %[flags]"
4791 			      : [early_failure] "+r" (early_failure),
4792 				[flags] "=&a" (flags)
4793 			      : [host_rsp] "i" (HOST_RSP),
4794 				[host_rip] "i" (HOST_RIP)
4795 			      : "rdx", "cc", "memory");
4796 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
4797 
4798 	report("Early VM-entry failure", early_failure);
4799 	report("RFLAGS[8:0] is %x (actual %x)", flags == expected_flags,
4800 	       expected_flags, flags);
4801 	if (valid_vmcs)
4802 		report("VM-instruction error is %d (actual %d)",
4803 		       vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
4804 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
4805 }
4806 
4807 static void vmentry_movss_shadow_test(void)
4808 {
4809 	struct vmcs *orig_vmcs;
4810 
4811 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
4812 
4813 	/*
4814 	 * Set the launched flag on the current VMCS to verify the correct
4815 	 * error priority, below.
4816 	 */
4817 	test_set_guest(v2_null_test_guest);
4818 	enter_guest();
4819 
4820 	/*
4821 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
4822 	 * fail due to invalid guest state (if we make it that far).
4823 	 */
4824 	vmcs_write(GUEST_RFLAGS, 0);
4825 
4826 	/*
4827 	 * "VM entry with events blocked by MOV SS" takes precedence over
4828 	 * "VMLAUNCH with non-clear VMCS."
4829 	 */
4830 	report_prefix_push("valid current-VMCS");
4831 	try_vmentry_in_movss_shadow();
4832 	report_prefix_pop();
4833 
4834 	/*
4835 	 * VMfailInvalid takes precedence over "VM entry with events
4836 	 * blocked by MOV SS."
4837 	 */
4838 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
4839 	report_prefix_push("no current-VMCS");
4840 	try_vmentry_in_movss_shadow();
4841 	report_prefix_pop();
4842 
4843 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
4844 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
4845 }
4846 
4847 #define X86_FEATURE_PCID       (1 << 17)
4848 #define X86_FEATURE_MCE        (1 << 7)
4849 
4850 static int write_cr4_checking(unsigned long val)
4851 {
4852 	asm volatile(ASM_TRY("1f")
4853 		     "mov %0, %%cr4\n\t"
4854 		     "1:": : "r" (val));
4855 	return exception_vector();
4856 }
4857 
4858 static void vmx_cr_load_test(void)
4859 {
4860 	struct cpuid _cpuid = cpuid(1);
4861 	unsigned long cr4 = read_cr4(), cr3 = read_cr3();
4862 
4863 	if (!(_cpuid.c & X86_FEATURE_PCID)) {
4864 		report_skip("PCID not detected");
4865 		return;
4866 	}
4867 	if (!(_cpuid.d & X86_FEATURE_MCE)) {
4868 		report_skip("MCE not detected");
4869 		return;
4870 	}
4871 
4872 	TEST_ASSERT(!(cr4 & (X86_CR4_PCIDE | X86_CR4_MCE)));
4873 	TEST_ASSERT(!(cr3 & X86_CR3_PCID_MASK));
4874 
4875 	/* Enable PCID for L1. */
4876 	cr4 |= X86_CR4_PCIDE;
4877 	cr3 |= 0x1;
4878 	TEST_ASSERT(!write_cr4_checking(cr4));
4879 	write_cr3(cr3);
4880 
4881 	test_set_guest(v2_null_test_guest);
4882 	vmcs_write(HOST_CR4, cr4);
4883 	vmcs_write(HOST_CR3, cr3);
4884 	enter_guest();
4885 
4886 	/*
4887 	 * No exception is expected.
4888 	 *
4889 	 * NB. KVM loads the last guest write to CR4 into CR4 read
4890 	 *     shadow. In order to trigger an exit to KVM, we can set a
4891 	 *     bit that was zero in the above CR4 write and is owned by
4892 	 *     KVM. We choose to set CR4.MCE, which shall have no side
4893 	 *     effect because normally no guest MCE (e.g., as the result
4894 	 *     of bad memory) would happen during this test.
4895 	 */
4896 	TEST_ASSERT(!write_cr4_checking(cr4 | X86_CR4_MCE));
4897 
4898 	/* Cleanup L1 state: disable PCID. */
4899 	write_cr3(cr3 & ~X86_CR3_PCID_MASK);
4900 	TEST_ASSERT(!write_cr4_checking(cr4 & ~X86_CR4_PCIDE));
4901 }
4902 
4903 static void vmx_nm_test_guest(void)
4904 {
4905 	write_cr0(read_cr0() | X86_CR0_TS);
4906 	asm volatile("fnop");
4907 }
4908 
4909 static void check_nm_exit(const char *test)
4910 {
4911 	u32 reason = vmcs_read(EXI_REASON);
4912 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
4913 	const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
4914 		NM_VECTOR;
4915 
4916 	report("%s", reason == VMX_EXC_NMI && intr_info == expected, test);
4917 }
4918 
4919 /*
4920  * This test checks that:
4921  *
4922  * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then
4923  *     a subsequent #NM VM-exit is reflected to L1.
4924  *
4925  * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a
4926  *     subsequent #NM VM-exit is reflected to L1.
4927  */
4928 static void vmx_nm_test(void)
4929 {
4930 	unsigned long cr0 = read_cr0();
4931 
4932 	test_set_guest(vmx_nm_test_guest);
4933 
4934 	/*
4935 	 * L1 wants to intercept #NM exceptions encountered in L2.
4936 	 */
4937 	vmcs_write(EXC_BITMAP, 1 << NM_VECTOR);
4938 
4939 	/*
4940 	 * Launch L2 with CR0.TS clear, but don't claim host ownership of
4941 	 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop,
4942 	 * which will raise #NM. L0 should reflect the #NM VM-exit to L1.
4943 	 */
4944 	vmcs_write(CR0_MASK, 0);
4945 	vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS);
4946 	enter_guest();
4947 	check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1");
4948 
4949 	/*
4950 	 * Re-enter L2 at the fnop instruction, with CR0.TS clear but
4951 	 * CR0.EM set. The fnop will still raise #NM, and L0 should
4952 	 * reflect the #NM VM-exit to L1.
4953 	 */
4954 	vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM);
4955 	enter_guest();
4956 	check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1");
4957 
4958 	/*
4959 	 * Re-enter L2 at the fnop instruction, with both CR0.TS and
4960 	 * CR0.EM clear. There will be no #NM, and the L2 guest should
4961 	 * exit normally.
4962 	 */
4963 	vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM));
4964 	enter_guest();
4965 }
4966 
4967 bool vmx_pending_event_ipi_fired;
4968 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
4969 {
4970 	vmx_pending_event_ipi_fired = true;
4971 	eoi();
4972 }
4973 
4974 bool vmx_pending_event_guest_run;
4975 static void vmx_pending_event_guest(void)
4976 {
4977 	vmcall();
4978 	vmx_pending_event_guest_run = true;
4979 }
4980 
4981 static void vmx_pending_event_test_core(bool guest_hlt)
4982 {
4983 	int ipi_vector = 0xf1;
4984 
4985 	vmx_pending_event_ipi_fired = false;
4986 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
4987 
4988 	vmx_pending_event_guest_run = false;
4989 	test_set_guest(vmx_pending_event_guest);
4990 
4991 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
4992 
4993 	enter_guest();
4994 	skip_exit_vmcall();
4995 
4996 	if (guest_hlt)
4997 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
4998 
4999 	irq_disable();
5000 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
5001 				   APIC_DM_FIXED | ipi_vector,
5002 				   0);
5003 
5004 	enter_guest();
5005 
5006 	assert_exit_reason(VMX_EXTINT);
5007 	report("Guest did not run before host received IPI",
5008 		   !vmx_pending_event_guest_run);
5009 
5010 	irq_enable();
5011 	asm volatile ("nop");
5012 	irq_disable();
5013 	report("Got pending interrupt after IRQ enabled",
5014 		   vmx_pending_event_ipi_fired);
5015 
5016 	if (guest_hlt)
5017 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
5018 
5019 	enter_guest();
5020 	report("Guest finished running when no interrupt",
5021 		   vmx_pending_event_guest_run);
5022 }
5023 
5024 static void vmx_pending_event_test(void)
5025 {
5026 	vmx_pending_event_test_core(false);
5027 }
5028 
5029 static void vmx_pending_event_hlt_test(void)
5030 {
5031 	vmx_pending_event_test_core(true);
5032 }
5033 
5034 static void vmx_db_test_guest(void)
5035 {
5036 	/*
5037 	 * For a hardware generated single-step #DB.
5038 	 */
5039 	asm volatile("vmcall;"
5040 		     "nop;"
5041 		     ".Lpost_nop:");
5042 	/*
5043 	 * ...in a MOVSS shadow, with pending debug exceptions.
5044 	 */
5045 	asm volatile("vmcall;"
5046 		     "nop;"
5047 		     ".Lpost_movss_nop:");
5048 	/*
5049 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
5050 	 * emulates it in software.)
5051 	 */
5052 	asm volatile("vmcall;"
5053 		     "wbinvd;"
5054 		     ".Lpost_wbinvd:");
5055 	/*
5056 	 * ...in a MOVSS shadow, with pending debug exceptions.
5057 	 */
5058 	asm volatile("vmcall;"
5059 		     "wbinvd;"
5060 		     ".Lpost_movss_wbinvd:");
5061 	/*
5062 	 * For a hardware generated single-step #DB in a transactional region.
5063 	 */
5064 	asm volatile("vmcall;"
5065 		     ".Lxbegin: xbegin .Lskip_rtm;"
5066 		     "xend;"
5067 		     ".Lskip_rtm:");
5068 }
5069 
5070 /*
5071  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
5072  * L2. No #DB is delivered and L2 continues to the next point of
5073  * interest.
5074  */
5075 static void dismiss_db(void)
5076 {
5077 	vmcs_write(GUEST_PENDING_DEBUG, 0);
5078 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
5079 	enter_guest();
5080 }
5081 
5082 /*
5083  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
5084  * Then throw away the #DB exception and resume L2.
5085  */
5086 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
5087 			  void *expected_rip, u64 expected_exit_qual,
5088 			  u64 expected_dr6)
5089 {
5090 	u32 reason = vmcs_read(EXI_REASON);
5091 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
5092 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
5093 	u64 guest_rip = vmcs_read(GUEST_RIP);
5094 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
5095 	u64 dr6 = read_dr6();
5096 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
5097 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
5098 
5099 	report("Expected #DB VM-exit",
5100 	       reason == VMX_EXC_NMI && intr_info == expected_intr_info);
5101 	report("Expected RIP %p (actual %lx)", (u64)expected_rip == guest_rip,
5102 	       expected_rip, guest_rip);
5103 	report_xfail("Expected pending debug exceptions 0 (actual %lx)",
5104 		     xfail_pdbg, 0 == guest_pending_dbg, guest_pending_dbg);
5105 	report_xfail("Expected exit qualification %lx (actual %lx)", xfail_qual,
5106 		     expected_exit_qual == exit_qual,
5107 		     expected_exit_qual, exit_qual);
5108 	report_xfail("Expected DR6 %lx (actual %lx)", xfail_dr6,
5109 		     expected_dr6 == dr6, expected_dr6, dr6);
5110 	dismiss_db();
5111 }
5112 
5113 /*
5114  * Assuming the guest has just exited on a VMCALL instruction, skip
5115  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
5116  * pending debug exceptions are non-zero, set the VMCS up as if the
5117  * previous instruction was a MOVSS that generated the indicated
5118  * pending debug exceptions. Then enter L2.
5119  */
5120 static void single_step_guest(const char *test_name, u64 starting_dr6,
5121 			      u64 pending_debug_exceptions)
5122 {
5123 	printf("\n%s\n", test_name);
5124 	skip_exit_vmcall();
5125 	write_dr6(starting_dr6);
5126 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
5127 	if (pending_debug_exceptions) {
5128 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
5129 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
5130 	}
5131 	enter_guest();
5132 }
5133 
5134 /*
5135  * When L1 intercepts #DB, verify that a single-step trap clears
5136  * pending debug exceptions, populates the exit qualification field
5137  * properly, and that DR6 is not prematurely clobbered. In a
5138  * (simulated) MOVSS shadow, make sure that the pending debug
5139  * exception bits are properly accumulated into the exit qualification
5140  * field.
5141  */
5142 static void vmx_db_test(void)
5143 {
5144 	/*
5145 	 * We are going to set a few arbitrary bits in DR6 to verify that
5146 	 * (a) DR6 is not modified by an intercepted #DB, and
5147 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
5148          *     the exit qualification field for a subsequent #DB exception.
5149 	 */
5150 	const u64 starting_dr6 = DR6_RESERVED | BIT(13) | DR_TRAP3 | DR_TRAP1;
5151 	extern char post_nop asm(".Lpost_nop");
5152 	extern char post_movss_nop asm(".Lpost_movss_nop");
5153 	extern char post_wbinvd asm(".Lpost_wbinvd");
5154 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
5155 	extern char xbegin asm(".Lxbegin");
5156 	extern char skip_rtm asm(".Lskip_rtm");
5157 
5158 	/*
5159 	 * L1 wants to intercept #DB exceptions encountered in L2.
5160 	 */
5161 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
5162 
5163 	/*
5164 	 * Start L2 and run it up to the first point of interest.
5165 	 */
5166 	test_set_guest(vmx_db_test_guest);
5167 	enter_guest();
5168 
5169 	/*
5170 	 * Hardware-delivered #DB trap for single-step sets the
5171 	 * standard that L0 has to follow for emulated instructions.
5172 	 */
5173 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
5174 	check_db_exit(false, false, false, &post_nop, DR_STEP, starting_dr6);
5175 
5176 	/*
5177 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
5178 	 * also sets the standard that L0 has to follow for emulated
5179 	 * instructions. Here, we establish the VMCS pending debug
5180 	 * exceptions to indicate that the simulated MOVSS triggered a
5181 	 * data breakpoint as well as the single-step trap.
5182 	 */
5183 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
5184 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0 );
5185 	check_db_exit(false, false, false, &post_movss_nop, DR_STEP | DR_TRAP0,
5186 		      starting_dr6);
5187 
5188 	/*
5189 	 * L0 synthesized #DB trap for single-step is buggy, because
5190 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
5191 	 * reconstitute the exit qualification from the prematurely
5192 	 * modified DR6, but fails miserably.
5193 	 */
5194 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
5195 	check_db_exit(true, true, false, &post_wbinvd, DR_STEP, starting_dr6);
5196 
5197 	/*
5198 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
5199 	 * even worse, because L0 also leaves the pending debug
5200 	 * exceptions in the VMCS instead of accumulating them into
5201 	 * the exit qualification field for the #DB exception.
5202 	 */
5203 	single_step_guest("Software synthesized single-step in MOVSS shadow",
5204 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0);
5205 	check_db_exit(true, true, true, &post_movss_wbinvd, DR_STEP | DR_TRAP0,
5206 		      starting_dr6);
5207 
5208 	/*
5209 	 * Optional RTM test for hardware that supports RTM, to
5210 	 * demonstrate that the current volume 3 of the SDM
5211 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
5212 	 * qualification for debug exceptions is not reserved. It is
5213 	 * set to 1 if a debug exception (#DB) or a breakpoint
5214 	 * exception (#BP) occurs inside an RTM region while advanced
5215 	 * debugging of RTM transactional regions is enabled.
5216 	 */
5217 	if (cpuid(7).b & BIT(11)) {
5218 		vmcs_write(ENT_CONTROLS,
5219 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
5220 		/*
5221 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
5222 		 * in the guest to enable advanced debugging of RTM
5223 		 * transactional regions.
5224 		 */
5225 		vmcs_write(GUEST_DR7, BIT(11));
5226 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
5227 		single_step_guest("Hardware delivered single-step in "
5228 				  "transactional region", starting_dr6, 0);
5229 		check_db_exit(false, false, false, &xbegin, BIT(16),
5230 			      starting_dr6);
5231 	} else {
5232 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
5233 		enter_guest();
5234 	}
5235 }
5236 
5237 static bool cpu_has_apicv(void)
5238 {
5239 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
5240 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
5241 		(ctrl_pin_rev.clr & PIN_POST_INTR));
5242 }
5243 
5244 static void enable_vid(void)
5245 {
5246 	void *virtual_apic_page;
5247 
5248 	assert(cpu_has_apicv());
5249 
5250 	disable_intercept_for_x2apic_msrs();
5251 
5252 	virtual_apic_page = alloc_page();
5253 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
5254 
5255 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
5256 
5257 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
5258 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
5259 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
5260 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
5261 
5262 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
5263 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
5264 }
5265 
5266 static void trigger_ioapic_scan_thread(void *data)
5267 {
5268 	/* Wait until other CPU entered L2 */
5269 	while (vmx_get_test_stage() != 1)
5270 		;
5271 
5272 	/* Trigger ioapic scan */
5273 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
5274 	vmx_set_test_stage(2);
5275 }
5276 
5277 static void irq_79_handler_guest(isr_regs_t *regs)
5278 {
5279 	eoi();
5280 
5281 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
5282 	vmcall();
5283 }
5284 
5285 /*
5286  * Constant for num of busy-loop iterations after which
5287  * a timer interrupt should have happened in host
5288  */
5289 #define TIMER_INTERRUPT_DELAY 100000000
5290 
5291 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
5292 {
5293 	handle_irq(0x79, irq_79_handler_guest);
5294 	irq_enable();
5295 
5296 	/* Signal to L1 CPU to trigger ioapic scan */
5297 	vmx_set_test_stage(1);
5298 	/* Wait until L1 CPU to trigger ioapic scan */
5299 	while (vmx_get_test_stage() != 2)
5300 		;
5301 
5302 	/*
5303 	 * Wait for L0 timer interrupt to be raised while we run in L2
5304 	 * such that L0 will process the IOAPIC scan request before
5305 	 * resuming L2
5306 	 */
5307 	delay(TIMER_INTERRUPT_DELAY);
5308 
5309 	asm volatile ("int $0x79");
5310 }
5311 
5312 static void vmx_eoi_bitmap_ioapic_scan_test(void)
5313 {
5314 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
5315 		report_skip(__func__);
5316 		return;
5317 	}
5318 
5319 	enable_vid();
5320 
5321 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
5322 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
5323 
5324 	/*
5325 	 * Launch L2.
5326 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
5327 	 * In case the reason isn't VMX_VMCALL, the asserion inside
5328 	 * skip_exit_vmcall() will fail.
5329 	 */
5330 	enter_guest();
5331 	skip_exit_vmcall();
5332 
5333 	/* Let L2 finish */
5334 	enter_guest();
5335 	report(__func__, 1);
5336 }
5337 
5338 #define HLT_WITH_RVI_VECTOR		(0xf1)
5339 
5340 bool vmx_hlt_with_rvi_guest_isr_fired;
5341 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
5342 {
5343 	vmx_hlt_with_rvi_guest_isr_fired = true;
5344 	eoi();
5345 }
5346 
5347 static void vmx_hlt_with_rvi_guest(void)
5348 {
5349 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
5350 
5351 	irq_enable();
5352 	asm volatile ("nop");
5353 
5354 	vmcall();
5355 }
5356 
5357 static void vmx_hlt_with_rvi_test(void)
5358 {
5359 	if (!cpu_has_apicv()) {
5360 		report_skip(__func__);
5361 		return;
5362 	}
5363 
5364 	enable_vid();
5365 
5366 	vmx_hlt_with_rvi_guest_isr_fired = false;
5367 	test_set_guest(vmx_hlt_with_rvi_guest);
5368 
5369 	enter_guest();
5370 	skip_exit_vmcall();
5371 
5372 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
5373 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
5374 	enter_guest();
5375 
5376 	report("Interrupt raised in guest", vmx_hlt_with_rvi_guest_isr_fired);
5377 }
5378 
5379 static void set_irq_line_thread(void *data)
5380 {
5381 	/* Wait until other CPU entered L2 */
5382 	while (vmx_get_test_stage() != 1)
5383 		;
5384 
5385 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
5386 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
5387 	vmx_set_test_stage(2);
5388 }
5389 
5390 static bool irq_78_handler_vmcall_before_eoi;
5391 static void irq_78_handler_guest(isr_regs_t *regs)
5392 {
5393 	set_irq_line(0xf, 0);
5394 	if (irq_78_handler_vmcall_before_eoi)
5395 		vmcall();
5396 	eoi();
5397 	vmcall();
5398 }
5399 
5400 static void vmx_apic_passthrough_guest(void)
5401 {
5402 	handle_irq(0x78, irq_78_handler_guest);
5403 	irq_enable();
5404 
5405 	/* If requested, wait for other CPU to trigger ioapic scan */
5406 	if (vmx_get_test_stage() < 1) {
5407 		vmx_set_test_stage(1);
5408 		while (vmx_get_test_stage() != 2)
5409 			;
5410 	}
5411 
5412 	set_irq_line(0xf, 1);
5413 }
5414 
5415 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
5416 {
5417 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
5418 		report_skip(__func__);
5419 		return;
5420 	}
5421 
5422 	u64 cpu_ctrl_0 = CPU_SECONDARY;
5423 	u64 cpu_ctrl_1 = 0;
5424 
5425 	disable_intercept_for_x2apic_msrs();
5426 
5427 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
5428 
5429 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
5430 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
5431 
5432 	if (set_irq_line_from_thread) {
5433 		irq_78_handler_vmcall_before_eoi = false;
5434 		on_cpu_async(1, set_irq_line_thread, NULL);
5435 	} else {
5436 		irq_78_handler_vmcall_before_eoi = true;
5437 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
5438 		vmx_set_test_stage(2);
5439 	}
5440 	test_set_guest(vmx_apic_passthrough_guest);
5441 
5442 	if (irq_78_handler_vmcall_before_eoi) {
5443 		/* Before EOI remote_irr should still be set */
5444 		enter_guest();
5445 		skip_exit_vmcall();
5446 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
5447 			"IOAPIC pass-through: remote_irr=1 before EOI");
5448 	}
5449 
5450 	/* After EOI remote_irr should be cleared */
5451 	enter_guest();
5452 	skip_exit_vmcall();
5453 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
5454 		"IOAPIC pass-through: remote_irr=0 after EOI");
5455 
5456 	/* Let L2 finish */
5457 	enter_guest();
5458 	report(__func__, 1);
5459 }
5460 
5461 static void vmx_apic_passthrough_test(void)
5462 {
5463 	vmx_apic_passthrough(false);
5464 }
5465 
5466 static void vmx_apic_passthrough_thread_test(void)
5467 {
5468 	vmx_apic_passthrough(true);
5469 }
5470 
5471 enum vmcs_access {
5472 	ACCESS_VMREAD,
5473 	ACCESS_VMWRITE,
5474 	ACCESS_NONE,
5475 };
5476 
5477 struct vmcs_shadow_test_common {
5478 	enum vmcs_access op;
5479 	enum Reason reason;
5480 	u64 field;
5481 	u64 value;
5482 	u64 flags;
5483 	u64 time;
5484 } l1_l2_common;
5485 
5486 static inline u64 vmread_flags(u64 field, u64 *val)
5487 {
5488 	u64 flags;
5489 
5490 	asm volatile ("vmread %2, %1; pushf; pop %0"
5491 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
5492 	return flags & X86_EFLAGS_ALU;
5493 }
5494 
5495 static inline u64 vmwrite_flags(u64 field, u64 val)
5496 {
5497 	u64 flags;
5498 
5499 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
5500 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
5501 	return flags & X86_EFLAGS_ALU;
5502 }
5503 
5504 static void vmx_vmcs_shadow_test_guest(void)
5505 {
5506 	struct vmcs_shadow_test_common *c = &l1_l2_common;
5507 	u64 start;
5508 
5509 	while (c->op != ACCESS_NONE) {
5510 		start = rdtsc();
5511 		switch (c->op) {
5512 		default:
5513 			c->flags = -1ull;
5514 			break;
5515 		case ACCESS_VMREAD:
5516 			c->flags = vmread_flags(c->field, &c->value);
5517 			break;
5518 		case ACCESS_VMWRITE:
5519 			c->flags = vmwrite_flags(c->field, 0);
5520 			break;
5521 		}
5522 		c->time = rdtsc() - start;
5523 		vmcall();
5524 	}
5525 }
5526 
5527 static u64 vmread_from_shadow(u64 field)
5528 {
5529 	struct vmcs *primary;
5530 	struct vmcs *shadow;
5531 	u64 value;
5532 
5533 	TEST_ASSERT(!vmcs_save(&primary));
5534 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
5535 	TEST_ASSERT(!make_vmcs_current(shadow));
5536 	value = vmcs_read(field);
5537 	TEST_ASSERT(!make_vmcs_current(primary));
5538 	return value;
5539 }
5540 
5541 static u64 vmwrite_to_shadow(u64 field, u64 value)
5542 {
5543 	struct vmcs *primary;
5544 	struct vmcs *shadow;
5545 
5546 	TEST_ASSERT(!vmcs_save(&primary));
5547 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
5548 	TEST_ASSERT(!make_vmcs_current(shadow));
5549 	vmcs_write(field, value);
5550 	value = vmcs_read(field);
5551 	TEST_ASSERT(!make_vmcs_current(primary));
5552 	return value;
5553 }
5554 
5555 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
5556 {
5557 	struct vmcs_shadow_test_common *c = &l1_l2_common;
5558 
5559 	c->op = access;
5560 	vmcs_write(VMX_INST_ERROR, 0);
5561 	enter_guest();
5562 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
5563 	if (c->reason != VMX_VMCALL) {
5564 		skip_exit_insn();
5565 		enter_guest();
5566 	}
5567 	skip_exit_vmcall();
5568 }
5569 
5570 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
5571 {
5572 	struct vmcs_shadow_test_common *c = &l1_l2_common;
5573 	struct vmcs *shadow;
5574 	u64 value;
5575 	uintptr_t flags[2];
5576 	bool good_shadow;
5577 	u32 vmx_inst_error;
5578 
5579 	report_prefix_pushf("field %lx", field);
5580 	c->field = field;
5581 
5582 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
5583 	if (shadow != (struct vmcs *)-1ull) {
5584 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
5585 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
5586 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
5587 	} else {
5588 		/*
5589 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
5590 		 * shadowed-fields should fail with setting RFLAGS.CF.
5591 		 */
5592 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
5593 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
5594 		good_shadow = false;
5595 	}
5596 
5597 	/* Intercept both VMREAD and VMWRITE. */
5598 	report_prefix_push("no VMREAD/VMWRITE permission");
5599 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
5600 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
5601 		set_bit(field, bitmap[ACCESS_VMREAD]);
5602 		set_bit(field, bitmap[ACCESS_VMWRITE]);
5603 	}
5604 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
5605 	report("not shadowed for VMWRITE", c->reason == VMX_VMWRITE);
5606 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
5607 	report("not shadowed for VMREAD", c->reason == VMX_VMREAD);
5608 	report_prefix_pop();
5609 
5610 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
5611 		goto out;
5612 
5613 	/* Permit shadowed VMREAD. */
5614 	report_prefix_push("VMREAD permission only");
5615 	clear_bit(field, bitmap[ACCESS_VMREAD]);
5616 	set_bit(field, bitmap[ACCESS_VMWRITE]);
5617 	if (good_shadow)
5618 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
5619 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
5620 	report("not shadowed for VMWRITE", c->reason == VMX_VMWRITE);
5621 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
5622 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
5623 	report("shadowed for VMREAD (in %ld cycles)", c->reason == VMX_VMCALL,
5624 	       c->time);
5625 	report("ALU flags after VMREAD (%lx) are as expected (%lx)",
5626 	       c->flags == flags[ACCESS_VMREAD],
5627 	       c->flags, flags[ACCESS_VMREAD]);
5628 	if (good_shadow)
5629 		report("value read from shadow (%lx) is as expected (%lx)",
5630 		       c->value == value, c->value, value);
5631 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
5632 		report("VMX_INST_ERROR (%d) is as expected (%d)",
5633 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
5634 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5635 	report_prefix_pop();
5636 
5637 	/* Permit shadowed VMWRITE. */
5638 	report_prefix_push("VMWRITE permission only");
5639 	set_bit(field, bitmap[ACCESS_VMREAD]);
5640 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
5641 	if (good_shadow)
5642 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
5643 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
5644 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
5645 	report("shadowed for VMWRITE (in %ld cycles)", c->reason == VMX_VMCALL,
5646 		c->time);
5647 	report("ALU flags after VMWRITE (%lx) are as expected (%lx)",
5648 	       c->flags == flags[ACCESS_VMREAD],
5649 	       c->flags, flags[ACCESS_VMREAD]);
5650 	if (good_shadow) {
5651 		value = vmread_from_shadow(field);
5652 		report("shadow VMCS value (%lx) is as expected (%lx)",
5653 		       value == 0, value, 0ul);
5654 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
5655 		report("VMX_INST_ERROR (%d) is as expected (%d)",
5656 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
5657 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5658 	}
5659 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
5660 	report("not shadowed for VMREAD", c->reason == VMX_VMREAD);
5661 	report_prefix_pop();
5662 
5663 	/* Permit shadowed VMREAD and VMWRITE. */
5664 	report_prefix_push("VMREAD and VMWRITE permission");
5665 	clear_bit(field, bitmap[ACCESS_VMREAD]);
5666 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
5667 	if (good_shadow)
5668 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
5669 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
5670 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
5671 	report("shadowed for VMWRITE (in %ld cycles)", c->reason == VMX_VMCALL,
5672 		c->time);
5673 	report("ALU flags after VMWRITE (%lx) are as expected (%lx)",
5674 	       c->flags == flags[ACCESS_VMREAD],
5675 	       c->flags, flags[ACCESS_VMREAD]);
5676 	if (good_shadow) {
5677 		value = vmread_from_shadow(field);
5678 		report("shadow VMCS value (%lx) is as expected (%lx)",
5679 		       value == 0, value, 0ul);
5680 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
5681 		report("VMX_INST_ERROR (%d) is as expected (%d)",
5682 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
5683 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5684 	}
5685 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
5686 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
5687 	report("shadowed for VMREAD (in %ld cycles)", c->reason == VMX_VMCALL,
5688 	       c->time);
5689 	report("ALU flags after VMREAD (%lx) are as expected (%lx)",
5690 	       c->flags == flags[ACCESS_VMREAD],
5691 	       c->flags, flags[ACCESS_VMREAD]);
5692 	if (good_shadow)
5693 		report("value read from shadow (%lx) is as expected (%lx)",
5694 		       c->value == 0, c->value, 0ul);
5695 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
5696 		report("VMX_INST_ERROR (%d) is as expected (%d)",
5697 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
5698 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5699 	report_prefix_pop();
5700 
5701 out:
5702 	report_prefix_pop();
5703 }
5704 
5705 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
5706 {
5707 	unsigned base;
5708 	unsigned index;
5709 	unsigned bit;
5710 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
5711 
5712 	/* Run test on all possible valid VMCS fields */
5713 	for (base = 0;
5714 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
5715 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
5716 		for (index = 0; index <= highest_index; index++)
5717 			vmcs_shadow_test_field(bitmap, base + index);
5718 
5719 	/*
5720 	 * Run tests on some invalid VMCS fields
5721 	 * (Have reserved bit set).
5722 	 */
5723 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
5724 		vmcs_shadow_test_field(bitmap, (1ull << bit));
5725 }
5726 
5727 static void vmx_vmcs_shadow_test(void)
5728 {
5729 	u8 *bitmap[2];
5730 	struct vmcs *shadow;
5731 
5732 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
5733 		printf("\t'Activate secondary controls' not supported.\n");
5734 		return;
5735 	}
5736 
5737 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
5738 		printf("\t'VMCS shadowing' not supported.\n");
5739 		return;
5740 	}
5741 
5742 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
5743 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
5744 		printf("\tVMWRITE can't modify VM-exit information fields.\n");
5745 		return;
5746 	}
5747 
5748 	test_set_guest(vmx_vmcs_shadow_test_guest);
5749 
5750 	bitmap[ACCESS_VMREAD] = alloc_page();
5751 	bitmap[ACCESS_VMWRITE] = alloc_page();
5752 
5753 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
5754 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
5755 
5756 	shadow = alloc_page();
5757 	shadow->hdr.revision_id = basic.revision;
5758 	shadow->hdr.shadow_vmcs = 1;
5759 	TEST_ASSERT(!vmcs_clear(shadow));
5760 
5761 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
5762 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
5763 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
5764 
5765 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
5766 	report_prefix_push("valid link pointer");
5767 	vmx_vmcs_shadow_test_body(bitmap);
5768 	report_prefix_pop();
5769 
5770 	vmcs_write(VMCS_LINK_PTR, -1ull);
5771 	report_prefix_push("invalid link pointer");
5772 	vmx_vmcs_shadow_test_body(bitmap);
5773 	report_prefix_pop();
5774 
5775 	l1_l2_common.op = ACCESS_NONE;
5776 	enter_guest();
5777 }
5778 
5779 #define TEST(name) { #name, .v2 = name }
5780 
5781 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
5782 struct vmx_test vmx_tests[] = {
5783 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
5784 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
5785 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
5786 		preemption_timer_exit_handler, NULL, {0} },
5787 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
5788 		test_ctrl_pat_exit_handler, NULL, {0} },
5789 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
5790 		test_ctrl_efer_exit_handler, NULL, {0} },
5791 	{ "CR shadowing", NULL, cr_shadowing_main,
5792 		cr_shadowing_exit_handler, NULL, {0} },
5793 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
5794 		NULL, {0} },
5795 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
5796 		insn_intercept_exit_handler, NULL, {0} },
5797 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
5798 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
5799 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
5800 	{ "VPID", vpid_init, vpid_main, vpid_exit_handler, NULL, {0} },
5801 	{ "interrupt", interrupt_init, interrupt_main,
5802 		interrupt_exit_handler, NULL, {0} },
5803 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
5804 		NULL, {0} },
5805 	{ "MSR switch", msr_switch_init, msr_switch_main,
5806 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
5807 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
5808 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
5809 		disable_rdtscp_exit_handler, NULL, {0} },
5810 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
5811 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
5812 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
5813 		exit_monitor_from_l2_handler, NULL, {0} },
5814 	/* Basic V2 tests. */
5815 	TEST(v2_null_test),
5816 	TEST(v2_multiple_entries_test),
5817 	TEST(fixture_test_case1),
5818 	TEST(fixture_test_case2),
5819 	/* Opcode tests. */
5820 	TEST(invvpid_test_v2),
5821 	/* VM-entry tests */
5822 	TEST(vmx_controls_test),
5823 	TEST(vmentry_movss_shadow_test),
5824 	/* APICv tests */
5825 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
5826 	TEST(vmx_hlt_with_rvi_test),
5827 	/* APIC pass-through tests */
5828 	TEST(vmx_apic_passthrough_test),
5829 	TEST(vmx_apic_passthrough_thread_test),
5830 	/* VMCS Shadowing tests */
5831 	TEST(vmx_vmcs_shadow_test),
5832 	/* Regression tests */
5833 	TEST(vmx_cr_load_test),
5834 	TEST(vmx_nm_test),
5835 	TEST(vmx_db_test),
5836 	TEST(vmx_pending_event_test),
5837 	TEST(vmx_pending_event_hlt_test),
5838 	/* EPT access tests. */
5839 	TEST(ept_access_test_not_present),
5840 	TEST(ept_access_test_read_only),
5841 	TEST(ept_access_test_write_only),
5842 	TEST(ept_access_test_read_write),
5843 	TEST(ept_access_test_execute_only),
5844 	TEST(ept_access_test_read_execute),
5845 	TEST(ept_access_test_write_execute),
5846 	TEST(ept_access_test_read_write_execute),
5847 	TEST(ept_access_test_reserved_bits),
5848 	TEST(ept_access_test_ignored_bits),
5849 	TEST(ept_access_test_paddr_not_present_ad_disabled),
5850 	TEST(ept_access_test_paddr_not_present_ad_enabled),
5851 	TEST(ept_access_test_paddr_read_only_ad_disabled),
5852 	TEST(ept_access_test_paddr_read_only_ad_enabled),
5853 	TEST(ept_access_test_paddr_read_write),
5854 	TEST(ept_access_test_paddr_read_write_execute),
5855 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
5856 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
5857 	TEST(ept_access_test_paddr_not_present_page_fault),
5858 	TEST(ept_access_test_force_2m_page),
5859 	{ NULL, NULL, NULL, NULL, NULL, {0} },
5860 };
5861