xref: /kvm-unit-tests/x86/vmx_tests.c (revision e329c328f3693f2791b678481120004c5ce1ac2d)
1 /*
2  * All test cases of nested virtualization should be in this file
3  *
4  * Author : Arthur Chunqi Li <yzt356@gmail.com>
5  */
6 
7 #include <asm/debugreg.h>
8 
9 #include "vmx.h"
10 #include "msr.h"
11 #include "processor.h"
12 #include "vm.h"
13 #include "pci.h"
14 #include "fwcfg.h"
15 #include "isr.h"
16 #include "desc.h"
17 #include "apic.h"
18 #include "types.h"
19 #include "vmalloc.h"
20 #include "alloc_page.h"
21 #include "smp.h"
22 #include "delay.h"
23 
24 #define NONCANONICAL            0xaaaaaaaaaaaaaaaaull
25 
26 #define VPID_CAP_INVVPID_TYPES_SHIFT 40
27 
28 u64 ia32_pat;
29 u64 ia32_efer;
30 void *io_bitmap_a, *io_bitmap_b;
31 u16 ioport;
32 
33 unsigned long *pml4;
34 u64 eptp;
35 void *data_page1, *data_page2;
36 
37 phys_addr_t pci_physaddr;
38 
39 void *pml_log;
40 #define PML_INDEX 512
41 
42 static inline unsigned ffs(unsigned x)
43 {
44 	int pos = -1;
45 
46 	__asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0"
47 			     : "+r"(pos) : "rm"(x) : "eax");
48 	return pos + 1;
49 }
50 
51 static inline void vmcall(void)
52 {
53 	asm volatile("vmcall");
54 }
55 
56 static void basic_guest_main(void)
57 {
58 	report("Basic VMX test", 1);
59 }
60 
61 static int basic_exit_handler(void)
62 {
63 	report("Basic VMX test", 0);
64 	print_vmexit_info();
65 	return VMX_TEST_EXIT;
66 }
67 
68 static void vmenter_main(void)
69 {
70 	u64 rax;
71 	u64 rsp, resume_rsp;
72 
73 	report("test vmlaunch", 1);
74 
75 	asm volatile(
76 		"mov %%rsp, %0\n\t"
77 		"mov %3, %%rax\n\t"
78 		"vmcall\n\t"
79 		"mov %%rax, %1\n\t"
80 		"mov %%rsp, %2\n\t"
81 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
82 		: "g"(0xABCD));
83 	report("test vmresume", (rax == 0xFFFF) && (rsp == resume_rsp));
84 }
85 
86 static int vmenter_exit_handler(void)
87 {
88 	u64 guest_rip;
89 	ulong reason;
90 
91 	guest_rip = vmcs_read(GUEST_RIP);
92 	reason = vmcs_read(EXI_REASON) & 0xff;
93 	switch (reason) {
94 	case VMX_VMCALL:
95 		if (regs.rax != 0xABCD) {
96 			report("test vmresume", 0);
97 			return VMX_TEST_VMEXIT;
98 		}
99 		regs.rax = 0xFFFF;
100 		vmcs_write(GUEST_RIP, guest_rip + 3);
101 		return VMX_TEST_RESUME;
102 	default:
103 		report("test vmresume", 0);
104 		print_vmexit_info();
105 	}
106 	return VMX_TEST_VMEXIT;
107 }
108 
109 u32 preempt_scale;
110 volatile unsigned long long tsc_val;
111 volatile u32 preempt_val;
112 u64 saved_rip;
113 
114 static int preemption_timer_init(struct vmcs *vmcs)
115 {
116 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
117 		printf("\tPreemption timer is not supported\n");
118 		return VMX_TEST_EXIT;
119 	}
120 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
121 	preempt_val = 10000000;
122 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
123 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
124 
125 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
126 		printf("\tSave preemption value is not supported\n");
127 
128 	return VMX_TEST_START;
129 }
130 
131 static void preemption_timer_main(void)
132 {
133 	tsc_val = rdtsc();
134 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
135 		vmx_set_test_stage(0);
136 		vmcall();
137 		if (vmx_get_test_stage() == 1)
138 			vmcall();
139 	}
140 	vmx_set_test_stage(1);
141 	while (vmx_get_test_stage() == 1) {
142 		if (((rdtsc() - tsc_val) >> preempt_scale)
143 				> 10 * preempt_val) {
144 			vmx_set_test_stage(2);
145 			vmcall();
146 		}
147 	}
148 	tsc_val = rdtsc();
149 	asm volatile ("hlt");
150 	vmcall();
151 	vmx_set_test_stage(5);
152 	vmcall();
153 }
154 
155 static int preemption_timer_exit_handler(void)
156 {
157 	bool guest_halted;
158 	u64 guest_rip;
159 	ulong reason;
160 	u32 insn_len;
161 	u32 ctrl_exit;
162 
163 	guest_rip = vmcs_read(GUEST_RIP);
164 	reason = vmcs_read(EXI_REASON) & 0xff;
165 	insn_len = vmcs_read(EXI_INST_LEN);
166 	switch (reason) {
167 	case VMX_PREEMPT:
168 		switch (vmx_get_test_stage()) {
169 		case 1:
170 		case 2:
171 			report("busy-wait for preemption timer",
172 			       ((rdtsc() - tsc_val) >> preempt_scale) >=
173 			       preempt_val);
174 			vmx_set_test_stage(3);
175 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
176 			return VMX_TEST_RESUME;
177 		case 3:
178 			guest_halted =
179 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
180 			report("preemption timer during hlt",
181 			       ((rdtsc() - tsc_val) >> preempt_scale) >=
182 			       preempt_val && guest_halted);
183 			vmx_set_test_stage(4);
184 			vmcs_write(PIN_CONTROLS,
185 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
186 			vmcs_write(EXI_CONTROLS,
187 				   vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT);
188 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
189 			return VMX_TEST_RESUME;
190 		case 4:
191 			report("preemption timer with 0 value",
192 			       saved_rip == guest_rip);
193 			break;
194 		default:
195 			report("Invalid stage.", false);
196 			print_vmexit_info();
197 			break;
198 		}
199 		break;
200 	case VMX_VMCALL:
201 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
202 		switch (vmx_get_test_stage()) {
203 		case 0:
204 			report("Keep preemption value",
205 			       vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val);
206 			vmx_set_test_stage(1);
207 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
208 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
209 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
210 			vmcs_write(EXI_CONTROLS, ctrl_exit);
211 			return VMX_TEST_RESUME;
212 		case 1:
213 			report("Save preemption value",
214 			       vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val);
215 			return VMX_TEST_RESUME;
216 		case 2:
217 			report("busy-wait for preemption timer", 0);
218 			vmx_set_test_stage(3);
219 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
220 			return VMX_TEST_RESUME;
221 		case 3:
222 			report("preemption timer during hlt", 0);
223 			vmx_set_test_stage(4);
224 			/* fall through */
225 		case 4:
226 			vmcs_write(PIN_CONTROLS,
227 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
228 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
229 			saved_rip = guest_rip + insn_len;
230 			return VMX_TEST_RESUME;
231 		case 5:
232 			report("preemption timer with 0 value (vmcall stage 5)", 0);
233 			break;
234 		default:
235 			// Should not reach here
236 			report("unexpected stage, %d", false,
237 			       vmx_get_test_stage());
238 			print_vmexit_info();
239 			return VMX_TEST_VMEXIT;
240 		}
241 		break;
242 	default:
243 		report("Unknown exit reason, %ld", false, reason);
244 		print_vmexit_info();
245 	}
246 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
247 	return VMX_TEST_VMEXIT;
248 }
249 
250 static void msr_bmp_init(void)
251 {
252 	void *msr_bitmap;
253 	u32 ctrl_cpu0;
254 
255 	msr_bitmap = alloc_page();
256 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
257 	ctrl_cpu0 |= CPU_MSR_BITMAP;
258 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
259 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
260 }
261 
262 static void *get_msr_bitmap(void)
263 {
264 	void *msr_bitmap;
265 
266 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
267 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
268 	} else {
269 		msr_bitmap = alloc_page();
270 		memset(msr_bitmap, 0xff, PAGE_SIZE);
271 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
272 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
273 	}
274 
275 	return msr_bitmap;
276 }
277 
278 static void disable_intercept_for_x2apic_msrs(void)
279 {
280 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
281 	u32 msr;
282 
283 	for (msr = APIC_BASE_MSR;
284 		 msr < (APIC_BASE_MSR+0xff);
285 		 msr += BITS_PER_LONG) {
286 		unsigned int word = msr / BITS_PER_LONG;
287 
288 		msr_bitmap[word] = 0;
289 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
290 	}
291 }
292 
293 static int test_ctrl_pat_init(struct vmcs *vmcs)
294 {
295 	u64 ctrl_ent;
296 	u64 ctrl_exi;
297 
298 	msr_bmp_init();
299 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
300 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
301 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
302 		printf("\tSave/load PAT is not supported\n");
303 		return 1;
304 	}
305 
306 	ctrl_ent = vmcs_read(ENT_CONTROLS);
307 	ctrl_exi = vmcs_read(EXI_CONTROLS);
308 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
309 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
310 	vmcs_write(ENT_CONTROLS, ctrl_ent);
311 	vmcs_write(EXI_CONTROLS, ctrl_exi);
312 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
313 	vmcs_write(GUEST_PAT, 0x0);
314 	vmcs_write(HOST_PAT, ia32_pat);
315 	return VMX_TEST_START;
316 }
317 
318 static void test_ctrl_pat_main(void)
319 {
320 	u64 guest_ia32_pat;
321 
322 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
323 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
324 		printf("\tENT_LOAD_PAT is not supported.\n");
325 	else {
326 		if (guest_ia32_pat != 0) {
327 			report("Entry load PAT", 0);
328 			return;
329 		}
330 	}
331 	wrmsr(MSR_IA32_CR_PAT, 0x6);
332 	vmcall();
333 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
334 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
335 		report("Entry load PAT", guest_ia32_pat == ia32_pat);
336 }
337 
338 static int test_ctrl_pat_exit_handler(void)
339 {
340 	u64 guest_rip;
341 	ulong reason;
342 	u64 guest_pat;
343 
344 	guest_rip = vmcs_read(GUEST_RIP);
345 	reason = vmcs_read(EXI_REASON) & 0xff;
346 	switch (reason) {
347 	case VMX_VMCALL:
348 		guest_pat = vmcs_read(GUEST_PAT);
349 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
350 			printf("\tEXI_SAVE_PAT is not supported\n");
351 			vmcs_write(GUEST_PAT, 0x6);
352 		} else {
353 			report("Exit save PAT", guest_pat == 0x6);
354 		}
355 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
356 			printf("\tEXI_LOAD_PAT is not supported\n");
357 		else
358 			report("Exit load PAT", rdmsr(MSR_IA32_CR_PAT) == ia32_pat);
359 		vmcs_write(GUEST_PAT, ia32_pat);
360 		vmcs_write(GUEST_RIP, guest_rip + 3);
361 		return VMX_TEST_RESUME;
362 	default:
363 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
364 		break;
365 	}
366 	return VMX_TEST_VMEXIT;
367 }
368 
369 static int test_ctrl_efer_init(struct vmcs *vmcs)
370 {
371 	u64 ctrl_ent;
372 	u64 ctrl_exi;
373 
374 	msr_bmp_init();
375 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
376 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
377 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
378 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
379 	ia32_efer = rdmsr(MSR_EFER);
380 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
381 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
382 	return VMX_TEST_START;
383 }
384 
385 static void test_ctrl_efer_main(void)
386 {
387 	u64 guest_ia32_efer;
388 
389 	guest_ia32_efer = rdmsr(MSR_EFER);
390 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
391 		printf("\tENT_LOAD_EFER is not supported.\n");
392 	else {
393 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
394 			report("Entry load EFER", 0);
395 			return;
396 		}
397 	}
398 	wrmsr(MSR_EFER, ia32_efer);
399 	vmcall();
400 	guest_ia32_efer = rdmsr(MSR_EFER);
401 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
402 		report("Entry load EFER", guest_ia32_efer == ia32_efer);
403 }
404 
405 static int test_ctrl_efer_exit_handler(void)
406 {
407 	u64 guest_rip;
408 	ulong reason;
409 	u64 guest_efer;
410 
411 	guest_rip = vmcs_read(GUEST_RIP);
412 	reason = vmcs_read(EXI_REASON) & 0xff;
413 	switch (reason) {
414 	case VMX_VMCALL:
415 		guest_efer = vmcs_read(GUEST_EFER);
416 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
417 			printf("\tEXI_SAVE_EFER is not supported\n");
418 			vmcs_write(GUEST_EFER, ia32_efer);
419 		} else {
420 			report("Exit save EFER", guest_efer == ia32_efer);
421 		}
422 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
423 			printf("\tEXI_LOAD_EFER is not supported\n");
424 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
425 		} else {
426 			report("Exit load EFER", rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX));
427 		}
428 		vmcs_write(GUEST_PAT, ia32_efer);
429 		vmcs_write(GUEST_RIP, guest_rip + 3);
430 		return VMX_TEST_RESUME;
431 	default:
432 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
433 		break;
434 	}
435 	return VMX_TEST_VMEXIT;
436 }
437 
438 u32 guest_cr0, guest_cr4;
439 
440 static void cr_shadowing_main(void)
441 {
442 	u32 cr0, cr4, tmp;
443 
444 	// Test read through
445 	vmx_set_test_stage(0);
446 	guest_cr0 = read_cr0();
447 	if (vmx_get_test_stage() == 1)
448 		report("Read through CR0", 0);
449 	else
450 		vmcall();
451 	vmx_set_test_stage(1);
452 	guest_cr4 = read_cr4();
453 	if (vmx_get_test_stage() == 2)
454 		report("Read through CR4", 0);
455 	else
456 		vmcall();
457 	// Test write through
458 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
459 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
460 	vmx_set_test_stage(2);
461 	write_cr0(guest_cr0);
462 	if (vmx_get_test_stage() == 3)
463 		report("Write throuth CR0", 0);
464 	else
465 		vmcall();
466 	vmx_set_test_stage(3);
467 	write_cr4(guest_cr4);
468 	if (vmx_get_test_stage() == 4)
469 		report("Write through CR4", 0);
470 	else
471 		vmcall();
472 	// Test read shadow
473 	vmx_set_test_stage(4);
474 	vmcall();
475 	cr0 = read_cr0();
476 	if (vmx_get_test_stage() != 5)
477 		report("Read shadowing CR0", cr0 == guest_cr0);
478 	vmx_set_test_stage(5);
479 	cr4 = read_cr4();
480 	if (vmx_get_test_stage() != 6)
481 		report("Read shadowing CR4", cr4 == guest_cr4);
482 	// Test write shadow (same value with shadow)
483 	vmx_set_test_stage(6);
484 	write_cr0(guest_cr0);
485 	if (vmx_get_test_stage() == 7)
486 		report("Write shadowing CR0 (same value with shadow)", 0);
487 	else
488 		vmcall();
489 	vmx_set_test_stage(7);
490 	write_cr4(guest_cr4);
491 	if (vmx_get_test_stage() == 8)
492 		report("Write shadowing CR4 (same value with shadow)", 0);
493 	else
494 		vmcall();
495 	// Test write shadow (different value)
496 	vmx_set_test_stage(8);
497 	tmp = guest_cr0 ^ X86_CR0_TS;
498 	asm volatile("mov %0, %%rsi\n\t"
499 		"mov %%rsi, %%cr0\n\t"
500 		::"m"(tmp)
501 		:"rsi", "memory", "cc");
502 	report("Write shadowing different X86_CR0_TS", vmx_get_test_stage() == 9);
503 	vmx_set_test_stage(9);
504 	tmp = guest_cr0 ^ X86_CR0_MP;
505 	asm volatile("mov %0, %%rsi\n\t"
506 		"mov %%rsi, %%cr0\n\t"
507 		::"m"(tmp)
508 		:"rsi", "memory", "cc");
509 	report("Write shadowing different X86_CR0_MP", vmx_get_test_stage() == 10);
510 	vmx_set_test_stage(10);
511 	tmp = guest_cr4 ^ X86_CR4_TSD;
512 	asm volatile("mov %0, %%rsi\n\t"
513 		"mov %%rsi, %%cr4\n\t"
514 		::"m"(tmp)
515 		:"rsi", "memory", "cc");
516 	report("Write shadowing different X86_CR4_TSD", vmx_get_test_stage() == 11);
517 	vmx_set_test_stage(11);
518 	tmp = guest_cr4 ^ X86_CR4_DE;
519 	asm volatile("mov %0, %%rsi\n\t"
520 		"mov %%rsi, %%cr4\n\t"
521 		::"m"(tmp)
522 		:"rsi", "memory", "cc");
523 	report("Write shadowing different X86_CR4_DE", vmx_get_test_stage() == 12);
524 }
525 
526 static int cr_shadowing_exit_handler(void)
527 {
528 	u64 guest_rip;
529 	ulong reason;
530 	u32 insn_len;
531 	u32 exit_qual;
532 
533 	guest_rip = vmcs_read(GUEST_RIP);
534 	reason = vmcs_read(EXI_REASON) & 0xff;
535 	insn_len = vmcs_read(EXI_INST_LEN);
536 	exit_qual = vmcs_read(EXI_QUALIFICATION);
537 	switch (reason) {
538 	case VMX_VMCALL:
539 		switch (vmx_get_test_stage()) {
540 		case 0:
541 			report("Read through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
542 			break;
543 		case 1:
544 			report("Read through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
545 			break;
546 		case 2:
547 			report("Write through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
548 			break;
549 		case 3:
550 			report("Write through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
551 			break;
552 		case 4:
553 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
554 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
555 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
556 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
557 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
558 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
559 			break;
560 		case 6:
561 			report("Write shadowing CR0 (same value)",
562 					guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)));
563 			break;
564 		case 7:
565 			report("Write shadowing CR4 (same value)",
566 					guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)));
567 			break;
568 		default:
569 			// Should not reach here
570 			report("unexpected stage, %d", false,
571 			       vmx_get_test_stage());
572 			print_vmexit_info();
573 			return VMX_TEST_VMEXIT;
574 		}
575 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
576 		return VMX_TEST_RESUME;
577 	case VMX_CR:
578 		switch (vmx_get_test_stage()) {
579 		case 4:
580 			report("Read shadowing CR0", 0);
581 			vmx_inc_test_stage();
582 			break;
583 		case 5:
584 			report("Read shadowing CR4", 0);
585 			vmx_inc_test_stage();
586 			break;
587 		case 6:
588 			report("Write shadowing CR0 (same value)", 0);
589 			vmx_inc_test_stage();
590 			break;
591 		case 7:
592 			report("Write shadowing CR4 (same value)", 0);
593 			vmx_inc_test_stage();
594 			break;
595 		case 8:
596 		case 9:
597 			// 0x600 encodes "mov %esi, %cr0"
598 			if (exit_qual == 0x600)
599 				vmx_inc_test_stage();
600 			break;
601 		case 10:
602 		case 11:
603 			// 0x604 encodes "mov %esi, %cr4"
604 			if (exit_qual == 0x604)
605 				vmx_inc_test_stage();
606 			break;
607 		default:
608 			// Should not reach here
609 			report("unexpected stage, %d", false,
610 			       vmx_get_test_stage());
611 			print_vmexit_info();
612 			return VMX_TEST_VMEXIT;
613 		}
614 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
615 		return VMX_TEST_RESUME;
616 	default:
617 		report("Unknown exit reason, %ld", false, reason);
618 		print_vmexit_info();
619 	}
620 	return VMX_TEST_VMEXIT;
621 }
622 
623 static int iobmp_init(struct vmcs *vmcs)
624 {
625 	u32 ctrl_cpu0;
626 
627 	io_bitmap_a = alloc_page();
628 	io_bitmap_b = alloc_page();
629 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
630 	ctrl_cpu0 |= CPU_IO_BITMAP;
631 	ctrl_cpu0 &= (~CPU_IO);
632 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
633 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
634 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
635 	return VMX_TEST_START;
636 }
637 
638 static void iobmp_main(void)
639 {
640 	// stage 0, test IO pass
641 	vmx_set_test_stage(0);
642 	inb(0x5000);
643 	outb(0x0, 0x5000);
644 	report("I/O bitmap - I/O pass", vmx_get_test_stage() == 0);
645 	// test IO width, in/out
646 	((u8 *)io_bitmap_a)[0] = 0xFF;
647 	vmx_set_test_stage(2);
648 	inb(0x0);
649 	report("I/O bitmap - trap in", vmx_get_test_stage() == 3);
650 	vmx_set_test_stage(3);
651 	outw(0x0, 0x0);
652 	report("I/O bitmap - trap out", vmx_get_test_stage() == 4);
653 	vmx_set_test_stage(4);
654 	inl(0x0);
655 	report("I/O bitmap - I/O width, long", vmx_get_test_stage() == 5);
656 	// test low/high IO port
657 	vmx_set_test_stage(5);
658 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
659 	inb(0x5000);
660 	report("I/O bitmap - I/O port, low part", vmx_get_test_stage() == 6);
661 	vmx_set_test_stage(6);
662 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
663 	inb(0x9000);
664 	report("I/O bitmap - I/O port, high part", vmx_get_test_stage() == 7);
665 	// test partial pass
666 	vmx_set_test_stage(7);
667 	inl(0x4FFF);
668 	report("I/O bitmap - partial pass", vmx_get_test_stage() == 8);
669 	// test overrun
670 	vmx_set_test_stage(8);
671 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
672 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
673 	inl(0xFFFF);
674 	report("I/O bitmap - overrun", vmx_get_test_stage() == 9);
675 	vmx_set_test_stage(9);
676 	vmcall();
677 	outb(0x0, 0x0);
678 	report("I/O bitmap - ignore unconditional exiting",
679 	       vmx_get_test_stage() == 9);
680 	vmx_set_test_stage(10);
681 	vmcall();
682 	outb(0x0, 0x0);
683 	report("I/O bitmap - unconditional exiting",
684 	       vmx_get_test_stage() == 11);
685 }
686 
687 static int iobmp_exit_handler(void)
688 {
689 	u64 guest_rip;
690 	ulong reason, exit_qual;
691 	u32 insn_len, ctrl_cpu0;
692 
693 	guest_rip = vmcs_read(GUEST_RIP);
694 	reason = vmcs_read(EXI_REASON) & 0xff;
695 	exit_qual = vmcs_read(EXI_QUALIFICATION);
696 	insn_len = vmcs_read(EXI_INST_LEN);
697 	switch (reason) {
698 	case VMX_IO:
699 		switch (vmx_get_test_stage()) {
700 		case 0:
701 		case 1:
702 			vmx_inc_test_stage();
703 			break;
704 		case 2:
705 			report("I/O bitmap - I/O width, byte",
706 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE);
707 			report("I/O bitmap - I/O direction, in", exit_qual & VMX_IO_IN);
708 			vmx_inc_test_stage();
709 			break;
710 		case 3:
711 			report("I/O bitmap - I/O width, word",
712 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD);
713 			report("I/O bitmap - I/O direction, out",
714 					!(exit_qual & VMX_IO_IN));
715 			vmx_inc_test_stage();
716 			break;
717 		case 4:
718 			report("I/O bitmap - I/O width, long",
719 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG);
720 			vmx_inc_test_stage();
721 			break;
722 		case 5:
723 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
724 				vmx_inc_test_stage();
725 			break;
726 		case 6:
727 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
728 				vmx_inc_test_stage();
729 			break;
730 		case 7:
731 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
732 				vmx_inc_test_stage();
733 			break;
734 		case 8:
735 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
736 				vmx_inc_test_stage();
737 			break;
738 		case 9:
739 		case 10:
740 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
741 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
742 			vmx_inc_test_stage();
743 			break;
744 		default:
745 			// Should not reach here
746 			report("unexpected stage, %d", false,
747 			       vmx_get_test_stage());
748 			print_vmexit_info();
749 			return VMX_TEST_VMEXIT;
750 		}
751 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
752 		return VMX_TEST_RESUME;
753 	case VMX_VMCALL:
754 		switch (vmx_get_test_stage()) {
755 		case 9:
756 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
757 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
758 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
759 			break;
760 		case 10:
761 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
762 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
763 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
764 			break;
765 		default:
766 			// Should not reach here
767 			report("unexpected stage, %d", false,
768 			       vmx_get_test_stage());
769 			print_vmexit_info();
770 			return VMX_TEST_VMEXIT;
771 		}
772 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
773 		return VMX_TEST_RESUME;
774 	default:
775 		printf("guest_rip = %#lx\n", guest_rip);
776 		printf("\tERROR : Undefined exit reason, reason = %ld.\n", reason);
777 		break;
778 	}
779 	return VMX_TEST_VMEXIT;
780 }
781 
782 #define INSN_CPU0		0
783 #define INSN_CPU1		1
784 #define INSN_ALWAYS_TRAP	2
785 
786 #define FIELD_EXIT_QUAL		(1 << 0)
787 #define FIELD_INSN_INFO		(1 << 1)
788 
789 asm(
790 	"insn_hlt: hlt;ret\n\t"
791 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
792 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
793 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
794 	"insn_rdtsc: rdtsc;ret\n\t"
795 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
796 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
797 #ifdef __x86_64__
798 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
799 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
800 #endif
801 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
802 	"insn_pause: pause;ret\n\t"
803 	"insn_wbinvd: wbinvd;ret\n\t"
804 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
805 	"insn_invd: invd;ret\n\t"
806 	"insn_sgdt: sgdt gdt64_desc;ret\n\t"
807 	"insn_lgdt: lgdt gdt64_desc;ret\n\t"
808 	"insn_sidt: sidt idt_descr;ret\n\t"
809 	"insn_lidt: lidt idt_descr;ret\n\t"
810 	"insn_sldt: sldt %ax;ret\n\t"
811 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
812 	"insn_str: str %ax;ret\n\t"
813 	"insn_rdrand: rdrand %rax;ret\n\t"
814 	"insn_rdseed: rdseed %rax;ret\n\t"
815 );
816 extern void insn_hlt(void);
817 extern void insn_invlpg(void);
818 extern void insn_mwait(void);
819 extern void insn_rdpmc(void);
820 extern void insn_rdtsc(void);
821 extern void insn_cr3_load(void);
822 extern void insn_cr3_store(void);
823 #ifdef __x86_64__
824 extern void insn_cr8_load(void);
825 extern void insn_cr8_store(void);
826 #endif
827 extern void insn_monitor(void);
828 extern void insn_pause(void);
829 extern void insn_wbinvd(void);
830 extern void insn_sgdt(void);
831 extern void insn_lgdt(void);
832 extern void insn_sidt(void);
833 extern void insn_lidt(void);
834 extern void insn_sldt(void);
835 extern void insn_lldt(void);
836 extern void insn_str(void);
837 extern void insn_cpuid(void);
838 extern void insn_invd(void);
839 extern void insn_rdrand(void);
840 extern void insn_rdseed(void);
841 
842 u32 cur_insn;
843 u64 cr3;
844 
845 #define X86_FEATURE_MONITOR	(1 << 3)
846 
847 typedef bool (*supported_fn)(void);
848 
849 static bool monitor_supported(void)
850 {
851 	return this_cpu_has(X86_FEATURE_MWAIT);
852 }
853 
854 struct insn_table {
855 	const char *name;
856 	u32 flag;
857 	void (*insn_func)(void);
858 	u32 type;
859 	u32 reason;
860 	ulong exit_qual;
861 	u32 insn_info;
862 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
863 	// which field need to be tested, reason is always tested
864 	u32 test_field;
865 	const supported_fn supported_fn;
866 	u8 disabled;
867 };
868 
869 /*
870  * Add more test cases of instruction intercept here. Elements in this
871  * table is:
872  *	name/control flag/insn function/type/exit reason/exit qulification/
873  *	instruction info/field to test
874  * The last field defines which fields (exit_qual and insn_info) need to be
875  * tested in exit handler. If set to 0, only "reason" is checked.
876  */
877 static struct insn_table insn_table[] = {
878 	// Flags for Primary Processor-Based VM-Execution Controls
879 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
880 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
881 		0x12345678, 0, FIELD_EXIT_QUAL},
882 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0, &monitor_supported},
883 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0},
884 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
885 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
886 		FIELD_EXIT_QUAL},
887 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
888 		FIELD_EXIT_QUAL},
889 #ifdef __x86_64__
890 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
891 		FIELD_EXIT_QUAL},
892 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
893 		FIELD_EXIT_QUAL},
894 #endif
895 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0, &monitor_supported},
896 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
897 	// Flags for Secondary Processor-Based VM-Execution Controls
898 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
899 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
900 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
901 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
902 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
903 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
904 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
905 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
906 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
907 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
908 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
909 	// Instructions always trap
910 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
911 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
912 	// Instructions never trap
913 	{NULL},
914 };
915 
916 static int insn_intercept_init(struct vmcs *vmcs)
917 {
918 	u32 ctrl_cpu, cur_insn;
919 
920 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
921 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
922 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
923 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
924 	cr3 = read_cr3();
925 
926 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
927 		if (insn_table[cur_insn].supported_fn == NULL)
928 			continue;
929 		insn_table[cur_insn].disabled = !insn_table[cur_insn].supported_fn();
930 	}
931 	return VMX_TEST_START;
932 }
933 
934 static void insn_intercept_main(void)
935 {
936 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
937 		vmx_set_test_stage(cur_insn * 2);
938 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
939 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
940 		    (insn_table[cur_insn].type == INSN_CPU1 &&
941 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
942 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
943 			       insn_table[cur_insn].type - INSN_CPU0,
944 			       insn_table[cur_insn].name);
945 			continue;
946 		}
947 
948 		if (insn_table[cur_insn].disabled) {
949 			printf("\tFeature required for %s is not supported.\n",
950 			       insn_table[cur_insn].name);
951 			continue;
952 		}
953 
954 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
955 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
956 		    (insn_table[cur_insn].type == INSN_CPU1 &&
957 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
958 			/* skip hlt, it stalls the guest and is tested below */
959 			if (insn_table[cur_insn].insn_func != insn_hlt)
960 				insn_table[cur_insn].insn_func();
961 			report("execute %s", vmx_get_test_stage() == cur_insn * 2,
962 					insn_table[cur_insn].name);
963 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
964 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
965 			       insn_table[cur_insn].type - INSN_CPU0,
966 			       insn_table[cur_insn].name);
967 
968 		vmcall();
969 
970 		insn_table[cur_insn].insn_func();
971 		report("intercept %s", vmx_get_test_stage() == cur_insn * 2 + 1,
972 				insn_table[cur_insn].name);
973 
974 		vmx_set_test_stage(cur_insn * 2 + 1);
975 		vmcall();
976 	}
977 }
978 
979 static int insn_intercept_exit_handler(void)
980 {
981 	u64 guest_rip;
982 	u32 reason;
983 	ulong exit_qual;
984 	u32 insn_len;
985 	u32 insn_info;
986 	bool pass;
987 
988 	guest_rip = vmcs_read(GUEST_RIP);
989 	reason = vmcs_read(EXI_REASON) & 0xff;
990 	exit_qual = vmcs_read(EXI_QUALIFICATION);
991 	insn_len = vmcs_read(EXI_INST_LEN);
992 	insn_info = vmcs_read(EXI_INST_INFO);
993 
994 	if (reason == VMX_VMCALL) {
995 		u32 val = 0;
996 
997 		if (insn_table[cur_insn].type == INSN_CPU0)
998 			val = vmcs_read(CPU_EXEC_CTRL0);
999 		else if (insn_table[cur_insn].type == INSN_CPU1)
1000 			val = vmcs_read(CPU_EXEC_CTRL1);
1001 
1002 		if (vmx_get_test_stage() & 1)
1003 			val &= ~insn_table[cur_insn].flag;
1004 		else
1005 			val |= insn_table[cur_insn].flag;
1006 
1007 		if (insn_table[cur_insn].type == INSN_CPU0)
1008 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
1009 		else if (insn_table[cur_insn].type == INSN_CPU1)
1010 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
1011 	} else {
1012 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
1013 			insn_table[cur_insn].reason == reason;
1014 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
1015 		    insn_table[cur_insn].exit_qual != exit_qual)
1016 			pass = false;
1017 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
1018 		    insn_table[cur_insn].insn_info != insn_info)
1019 			pass = false;
1020 		if (pass)
1021 			vmx_inc_test_stage();
1022 	}
1023 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1024 	return VMX_TEST_RESUME;
1025 }
1026 
1027 /**
1028  * __setup_ept - Setup the VMCS fields to enable Extended Page Tables (EPT)
1029  * @hpa:	Host physical address of the top-level, a.k.a. root, EPT table
1030  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1031  *
1032  * Returns 0 on success, 1 on failure.
1033  *
1034  * Note that @hpa doesn't need to point at actual memory if VM-Launch is
1035  * expected to fail, e.g. setup_dummy_ept() arbitrarily passes '0' to satisfy
1036  * the various EPTP consistency checks, but doesn't ensure backing for HPA '0'.
1037  */
1038 static int __setup_ept(u64 hpa, bool enable_ad)
1039 {
1040 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1041 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1042 		printf("\tEPT is not supported");
1043 		return 1;
1044 	}
1045 	if (!(ept_vpid.val & EPT_CAP_WB)) {
1046 		printf("WB memtype for EPT walks not supported\n");
1047 		return 1;
1048 	}
1049 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
1050 		printf("\tPWL4 is not supported\n");
1051 		return 1;
1052 	}
1053 
1054 	eptp = EPT_MEM_TYPE_WB;
1055 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1056 	eptp |= hpa;
1057 	if (enable_ad)
1058 		eptp |= EPTP_AD_FLAG;
1059 
1060 	vmcs_write(EPTP, eptp);
1061 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY);
1062 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT);
1063 
1064 	return 0;
1065 }
1066 
1067 /**
1068  * setup_ept - Enable Extended Page Tables (EPT) and setup an identity map
1069  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1070  *
1071  * Returns 0 on success, 1 on failure.
1072  *
1073  * This is the "real" function for setting up EPT tables, i.e. use this for
1074  * tests that need to run code in the guest with EPT enabled.
1075  */
1076 static int setup_ept(bool enable_ad)
1077 {
1078 	unsigned long end_of_memory;
1079 
1080 	pml4 = alloc_page();
1081 
1082 	if (__setup_ept(virt_to_phys(pml4), enable_ad))
1083 		return 1;
1084 
1085 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1086 	if (end_of_memory < (1ul << 32))
1087 		end_of_memory = (1ul << 32);
1088 	/* Cannot use large EPT pages if we need to track EPT
1089 	 * accessed/dirty bits at 4K granularity.
1090 	 */
1091 	setup_ept_range(pml4, 0, end_of_memory, 0,
1092 			!enable_ad && ept_2m_supported(),
1093 			EPT_WA | EPT_RA | EPT_EA);
1094 	return 0;
1095 }
1096 
1097 /**
1098  * setup_dummy_ept - Enable Extended Page Tables (EPT) with a dummy root HPA
1099  *
1100  * Setup EPT using a semi-arbitrary dummy root HPA.  This function is intended
1101  * for use by tests that need EPT enabled to verify dependent VMCS controls
1102  * but never expect to fully enter the guest, i.e. don't need setup the actual
1103  * EPT tables.
1104  */
1105 static void setup_dummy_ept(void)
1106 {
1107 	if (__setup_ept(0, false))
1108 		report_abort("EPT setup unexpectedly failed");
1109 }
1110 
1111 static int enable_unrestricted_guest(void)
1112 {
1113 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1114 	    !(ctrl_cpu_rev[1].clr & CPU_URG) ||
1115 	    !(ctrl_cpu_rev[1].clr & CPU_EPT))
1116 		return 1;
1117 
1118 	setup_dummy_ept();
1119 
1120 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
1121 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
1122 
1123 	return 0;
1124 }
1125 
1126 static void ept_enable_ad_bits(void)
1127 {
1128 	eptp |= EPTP_AD_FLAG;
1129 	vmcs_write(EPTP, eptp);
1130 }
1131 
1132 static void ept_disable_ad_bits(void)
1133 {
1134 	eptp &= ~EPTP_AD_FLAG;
1135 	vmcs_write(EPTP, eptp);
1136 }
1137 
1138 static void ept_enable_ad_bits_or_skip_test(void)
1139 {
1140 	if (!ept_ad_bits_supported())
1141 		test_skip("EPT AD bits not supported.");
1142 	ept_enable_ad_bits();
1143 }
1144 
1145 static int apic_version;
1146 
1147 static int ept_init_common(bool have_ad)
1148 {
1149 	int ret;
1150 	struct pci_dev pcidev;
1151 
1152 	if (setup_ept(have_ad))
1153 		return VMX_TEST_EXIT;
1154 	data_page1 = alloc_page();
1155 	data_page2 = alloc_page();
1156 	*((u32 *)data_page1) = MAGIC_VAL_1;
1157 	*((u32 *)data_page2) = MAGIC_VAL_2;
1158 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1159 			EPT_RA | EPT_WA | EPT_EA);
1160 
1161 	apic_version = apic_read(APIC_LVR);
1162 
1163 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1164 	if (ret != PCIDEVADDR_INVALID) {
1165 		pci_dev_init(&pcidev, ret);
1166 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1167 	}
1168 
1169 	return VMX_TEST_START;
1170 }
1171 
1172 static int ept_init(struct vmcs *vmcs)
1173 {
1174 	return ept_init_common(false);
1175 }
1176 
1177 static void ept_common(void)
1178 {
1179 	vmx_set_test_stage(0);
1180 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1181 			*((u32 *)data_page1) != MAGIC_VAL_1)
1182 		report("EPT basic framework - read", 0);
1183 	else {
1184 		*((u32 *)data_page2) = MAGIC_VAL_3;
1185 		vmcall();
1186 		if (vmx_get_test_stage() == 1) {
1187 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1188 					*((u32 *)data_page2) == MAGIC_VAL_2)
1189 				report("EPT basic framework", 1);
1190 			else
1191 				report("EPT basic framework - remap", 1);
1192 		}
1193 	}
1194 	// Test EPT Misconfigurations
1195 	vmx_set_test_stage(1);
1196 	vmcall();
1197 	*((u32 *)data_page1) = MAGIC_VAL_1;
1198 	if (vmx_get_test_stage() != 2) {
1199 		report("EPT misconfigurations", 0);
1200 		goto t1;
1201 	}
1202 	vmx_set_test_stage(2);
1203 	vmcall();
1204 	*((u32 *)data_page1) = MAGIC_VAL_1;
1205 	report("EPT misconfigurations", vmx_get_test_stage() == 3);
1206 t1:
1207 	// Test EPT violation
1208 	vmx_set_test_stage(3);
1209 	vmcall();
1210 	*((u32 *)data_page1) = MAGIC_VAL_1;
1211 	report("EPT violation - page permission", vmx_get_test_stage() == 4);
1212 	// Violation caused by EPT paging structure
1213 	vmx_set_test_stage(4);
1214 	vmcall();
1215 	*((u32 *)data_page1) = MAGIC_VAL_2;
1216 	report("EPT violation - paging structure", vmx_get_test_stage() == 5);
1217 
1218 	// MMIO Read/Write
1219 	vmx_set_test_stage(5);
1220 	vmcall();
1221 
1222 	*(u32 volatile *)pci_physaddr;
1223 	report("MMIO EPT violation - read", vmx_get_test_stage() == 6);
1224 
1225 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1226 	report("MMIO EPT violation - write", vmx_get_test_stage() == 7);
1227 }
1228 
1229 static void ept_main(void)
1230 {
1231 	ept_common();
1232 
1233 	// Test EPT access to L1 MMIO
1234 	vmx_set_test_stage(7);
1235 	report("EPT - MMIO access", *((u32 *)0xfee00030UL) == apic_version);
1236 
1237 	// Test invalid operand for INVEPT
1238 	vmcall();
1239 	report("EPT - unsupported INVEPT", vmx_get_test_stage() == 8);
1240 }
1241 
1242 static bool invept_test(int type, u64 eptp)
1243 {
1244 	bool ret, supported;
1245 
1246 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1247 	ret = invept(type, eptp);
1248 
1249 	if (ret == !supported)
1250 		return false;
1251 
1252 	if (!supported)
1253 		printf("WARNING: unsupported invept passed!\n");
1254 	else
1255 		printf("WARNING: invept failed!\n");
1256 
1257 	return true;
1258 }
1259 
1260 static int pml_exit_handler(void)
1261 {
1262 	u16 index, count;
1263 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1264 	u64 *pmlbuf = pml_log;
1265 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1266 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1267 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1268 
1269 	switch (reason) {
1270 	case VMX_VMCALL:
1271 		switch (vmx_get_test_stage()) {
1272 		case 0:
1273 			index = vmcs_read(GUEST_PML_INDEX);
1274 			for (count = index + 1; count < PML_INDEX; count++) {
1275 				if (pmlbuf[count] == (u64)data_page2) {
1276 					vmx_inc_test_stage();
1277 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1278 					break;
1279 				}
1280 			}
1281 			break;
1282 		case 1:
1283 			index = vmcs_read(GUEST_PML_INDEX);
1284 			/* Keep clearing the dirty bit till a overflow */
1285 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1286 			break;
1287 		default:
1288 			report("unexpected stage, %d.", false,
1289 			       vmx_get_test_stage());
1290 			print_vmexit_info();
1291 			return VMX_TEST_VMEXIT;
1292 		}
1293 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1294 		return VMX_TEST_RESUME;
1295 	case VMX_PML_FULL:
1296 		vmx_inc_test_stage();
1297 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1298 		return VMX_TEST_RESUME;
1299 	default:
1300 		report("Unknown exit reason, %ld", false, reason);
1301 		print_vmexit_info();
1302 	}
1303 	return VMX_TEST_VMEXIT;
1304 }
1305 
1306 static int ept_exit_handler_common(bool have_ad)
1307 {
1308 	u64 guest_rip;
1309 	u64 guest_cr3;
1310 	ulong reason;
1311 	u32 insn_len;
1312 	u32 exit_qual;
1313 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte;
1314 
1315 	guest_rip = vmcs_read(GUEST_RIP);
1316 	guest_cr3 = vmcs_read(GUEST_CR3);
1317 	reason = vmcs_read(EXI_REASON) & 0xff;
1318 	insn_len = vmcs_read(EXI_INST_LEN);
1319 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1320 	switch (reason) {
1321 	case VMX_VMCALL:
1322 		switch (vmx_get_test_stage()) {
1323 		case 0:
1324 			check_ept_ad(pml4, guest_cr3,
1325 				     (unsigned long)data_page1,
1326 				     have_ad ? EPT_ACCESS_FLAG : 0,
1327 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1328 			check_ept_ad(pml4, guest_cr3,
1329 				     (unsigned long)data_page2,
1330 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1331 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1332 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1333 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1334 			if (have_ad)
1335 				ept_sync(INVEPT_SINGLE, eptp);;
1336 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1337 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1338 				vmx_inc_test_stage();
1339 				install_ept(pml4, (unsigned long)data_page2,
1340 						(unsigned long)data_page2,
1341 						EPT_RA | EPT_WA | EPT_EA);
1342 			} else
1343 				report("EPT basic framework - write", 0);
1344 			break;
1345 		case 1:
1346 			install_ept(pml4, (unsigned long)data_page1,
1347  				(unsigned long)data_page1, EPT_WA);
1348 			ept_sync(INVEPT_SINGLE, eptp);
1349 			break;
1350 		case 2:
1351 			install_ept(pml4, (unsigned long)data_page1,
1352  				(unsigned long)data_page1,
1353  				EPT_RA | EPT_WA | EPT_EA |
1354  				(2 << EPT_MEM_TYPE_SHIFT));
1355 			ept_sync(INVEPT_SINGLE, eptp);
1356 			break;
1357 		case 3:
1358 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1359 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1360 						1, &data_page1_pte));
1361 			set_ept_pte(pml4, (unsigned long)data_page1,
1362 				1, data_page1_pte & ~EPT_PRESENT);
1363 			ept_sync(INVEPT_SINGLE, eptp);
1364 			break;
1365 		case 4:
1366 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1367 						2, &data_page1_pte));
1368 			data_page1_pte &= PAGE_MASK;
1369 			TEST_ASSERT(get_ept_pte(pml4, data_page1_pte,
1370 						2, &data_page1_pte_pte));
1371 			set_ept_pte(pml4, data_page1_pte, 2,
1372 				data_page1_pte_pte & ~EPT_PRESENT);
1373 			ept_sync(INVEPT_SINGLE, eptp);
1374 			break;
1375 		case 5:
1376 			install_ept(pml4, (unsigned long)pci_physaddr,
1377 				(unsigned long)pci_physaddr, 0);
1378 			ept_sync(INVEPT_SINGLE, eptp);
1379 			break;
1380 		case 7:
1381 			if (!invept_test(0, eptp))
1382 				vmx_inc_test_stage();
1383 			break;
1384 		// Should not reach here
1385 		default:
1386 			report("ERROR - unexpected stage, %d.", false,
1387 			       vmx_get_test_stage());
1388 			print_vmexit_info();
1389 			return VMX_TEST_VMEXIT;
1390 		}
1391 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1392 		return VMX_TEST_RESUME;
1393 	case VMX_EPT_MISCONFIG:
1394 		switch (vmx_get_test_stage()) {
1395 		case 1:
1396 		case 2:
1397 			vmx_inc_test_stage();
1398 			install_ept(pml4, (unsigned long)data_page1,
1399  				(unsigned long)data_page1,
1400  				EPT_RA | EPT_WA | EPT_EA);
1401 			ept_sync(INVEPT_SINGLE, eptp);
1402 			break;
1403 		// Should not reach here
1404 		default:
1405 			report("ERROR - unexpected stage, %d.", false,
1406 			       vmx_get_test_stage());
1407 			print_vmexit_info();
1408 			return VMX_TEST_VMEXIT;
1409 		}
1410 		return VMX_TEST_RESUME;
1411 	case VMX_EPT_VIOLATION:
1412 		switch(vmx_get_test_stage()) {
1413 		case 3:
1414 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1415 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1416 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1417 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1418 					EPT_VLT_PADDR))
1419 				vmx_inc_test_stage();
1420 			set_ept_pte(pml4, (unsigned long)data_page1,
1421 				1, data_page1_pte | (EPT_PRESENT));
1422 			ept_sync(INVEPT_SINGLE, eptp);
1423 			break;
1424 		case 4:
1425 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1426 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1427 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1428 			if (exit_qual == (EPT_VLT_RD |
1429 					  (have_ad ? EPT_VLT_WR : 0) |
1430 					  EPT_VLT_LADDR_VLD))
1431 				vmx_inc_test_stage();
1432 			set_ept_pte(pml4, data_page1_pte, 2,
1433 				data_page1_pte_pte | (EPT_PRESENT));
1434 			ept_sync(INVEPT_SINGLE, eptp);
1435 			break;
1436 		case 5:
1437 			if (exit_qual & EPT_VLT_RD)
1438 				vmx_inc_test_stage();
1439 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1440 						1, &memaddr_pte));
1441 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1442 			ept_sync(INVEPT_SINGLE, eptp);
1443 			break;
1444 		case 6:
1445 			if (exit_qual & EPT_VLT_WR)
1446 				vmx_inc_test_stage();
1447 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1448 						1, &memaddr_pte));
1449 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1450 			ept_sync(INVEPT_SINGLE, eptp);
1451 			break;
1452 		default:
1453 			// Should not reach here
1454 			report("ERROR : unexpected stage, %d", false,
1455 			       vmx_get_test_stage());
1456 			print_vmexit_info();
1457 			return VMX_TEST_VMEXIT;
1458 		}
1459 		return VMX_TEST_RESUME;
1460 	default:
1461 		report("Unknown exit reason, %ld", false, reason);
1462 		print_vmexit_info();
1463 	}
1464 	return VMX_TEST_VMEXIT;
1465 }
1466 
1467 static int ept_exit_handler(void)
1468 {
1469 	return ept_exit_handler_common(false);
1470 }
1471 
1472 static int eptad_init(struct vmcs *vmcs)
1473 {
1474 	int r = ept_init_common(true);
1475 
1476 	if (r == VMX_TEST_EXIT)
1477 		return r;
1478 
1479 	if ((rdmsr(MSR_IA32_VMX_EPT_VPID_CAP) & EPT_CAP_AD_FLAG) == 0) {
1480 		printf("\tEPT A/D bits are not supported");
1481 		return VMX_TEST_EXIT;
1482 	}
1483 
1484 	return r;
1485 }
1486 
1487 static int pml_init(struct vmcs *vmcs)
1488 {
1489 	u32 ctrl_cpu;
1490 	int r = eptad_init(vmcs);
1491 
1492 	if (r == VMX_TEST_EXIT)
1493 		return r;
1494 
1495 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1496 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1497 		printf("\tPML is not supported");
1498 		return VMX_TEST_EXIT;
1499 	}
1500 
1501 	pml_log = alloc_page();
1502 	vmcs_write(PMLADDR, (u64)pml_log);
1503 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1504 
1505 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1506 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1507 
1508 	return VMX_TEST_START;
1509 }
1510 
1511 static void pml_main(void)
1512 {
1513 	int count = 0;
1514 
1515 	vmx_set_test_stage(0);
1516 	*((u32 *)data_page2) = 0x1;
1517 	vmcall();
1518 	report("PML - Dirty GPA Logging", vmx_get_test_stage() == 1);
1519 
1520 	while (vmx_get_test_stage() == 1) {
1521 		vmcall();
1522 		*((u32 *)data_page2) = 0x1;
1523 		if (count++ > PML_INDEX)
1524 			break;
1525 	}
1526 	report("PML Full Event", vmx_get_test_stage() == 2);
1527 }
1528 
1529 static void eptad_main(void)
1530 {
1531 	ept_common();
1532 }
1533 
1534 static int eptad_exit_handler(void)
1535 {
1536 	return ept_exit_handler_common(true);
1537 }
1538 
1539 static bool invvpid_test(int type, u16 vpid)
1540 {
1541 	bool ret, supported;
1542 
1543 	supported = ept_vpid.val &
1544 		(VPID_CAP_INVVPID_ADDR >> INVVPID_ADDR << type);
1545 	ret = invvpid(type, vpid, 0);
1546 
1547 	if (ret == !supported)
1548 		return false;
1549 
1550 	if (!supported)
1551 		printf("WARNING: unsupported invvpid passed!\n");
1552 	else
1553 		printf("WARNING: invvpid failed!\n");
1554 
1555 	return true;
1556 }
1557 
1558 static int vpid_init(struct vmcs *vmcs)
1559 {
1560 	u32 ctrl_cpu1;
1561 
1562 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1563 		!(ctrl_cpu_rev[1].clr & CPU_VPID)) {
1564 		printf("\tVPID is not supported");
1565 		return VMX_TEST_EXIT;
1566 	}
1567 
1568 	ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
1569 	ctrl_cpu1 |= CPU_VPID;
1570 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
1571 	return VMX_TEST_START;
1572 }
1573 
1574 static void vpid_main(void)
1575 {
1576 	vmx_set_test_stage(0);
1577 	vmcall();
1578 	report("INVVPID SINGLE ADDRESS", vmx_get_test_stage() == 1);
1579 	vmx_set_test_stage(2);
1580 	vmcall();
1581 	report("INVVPID SINGLE", vmx_get_test_stage() == 3);
1582 	vmx_set_test_stage(4);
1583 	vmcall();
1584 	report("INVVPID ALL", vmx_get_test_stage() == 5);
1585 }
1586 
1587 static int vpid_exit_handler(void)
1588 {
1589 	u64 guest_rip;
1590 	ulong reason;
1591 	u32 insn_len;
1592 
1593 	guest_rip = vmcs_read(GUEST_RIP);
1594 	reason = vmcs_read(EXI_REASON) & 0xff;
1595 	insn_len = vmcs_read(EXI_INST_LEN);
1596 
1597 	switch (reason) {
1598 	case VMX_VMCALL:
1599 		switch(vmx_get_test_stage()) {
1600 		case 0:
1601 			if (!invvpid_test(INVVPID_ADDR, 1))
1602 				vmx_inc_test_stage();
1603 			break;
1604 		case 2:
1605 			if (!invvpid_test(INVVPID_CONTEXT_GLOBAL, 1))
1606 				vmx_inc_test_stage();
1607 			break;
1608 		case 4:
1609 			if (!invvpid_test(INVVPID_ALL, 1))
1610 				vmx_inc_test_stage();
1611 			break;
1612 		default:
1613 			report("ERROR: unexpected stage, %d", false,
1614 					vmx_get_test_stage());
1615 			print_vmexit_info();
1616 			return VMX_TEST_VMEXIT;
1617 		}
1618 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1619 		return VMX_TEST_RESUME;
1620 	default:
1621 		report("Unknown exit reason, %ld", false, reason);
1622 		print_vmexit_info();
1623 	}
1624 	return VMX_TEST_VMEXIT;
1625 }
1626 
1627 #define TIMER_VECTOR	222
1628 
1629 static volatile bool timer_fired;
1630 
1631 static void timer_isr(isr_regs_t *regs)
1632 {
1633 	timer_fired = true;
1634 	apic_write(APIC_EOI, 0);
1635 }
1636 
1637 static int interrupt_init(struct vmcs *vmcs)
1638 {
1639 	msr_bmp_init();
1640 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1641 	handle_irq(TIMER_VECTOR, timer_isr);
1642 	return VMX_TEST_START;
1643 }
1644 
1645 static void interrupt_main(void)
1646 {
1647 	long long start, loops;
1648 
1649 	vmx_set_test_stage(0);
1650 
1651 	apic_write(APIC_LVTT, TIMER_VECTOR);
1652 	irq_enable();
1653 
1654 	apic_write(APIC_TMICT, 1);
1655 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1656 		asm volatile ("nop");
1657 	report("direct interrupt while running guest", timer_fired);
1658 
1659 	apic_write(APIC_TMICT, 0);
1660 	irq_disable();
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("intercepted interrupt while running guest", timer_fired);
1667 
1668 	irq_enable();
1669 	apic_write(APIC_TMICT, 0);
1670 	irq_disable();
1671 	vmcall();
1672 	timer_fired = false;
1673 	start = rdtsc();
1674 	apic_write(APIC_TMICT, 1000000);
1675 
1676 	asm volatile ("sti; hlt");
1677 
1678 	report("direct interrupt + hlt",
1679 	       rdtsc() - start > 1000000 && timer_fired);
1680 
1681 	apic_write(APIC_TMICT, 0);
1682 	irq_disable();
1683 	vmcall();
1684 	timer_fired = false;
1685 	start = rdtsc();
1686 	apic_write(APIC_TMICT, 1000000);
1687 
1688 	asm volatile ("sti; hlt");
1689 
1690 	report("intercepted interrupt + hlt",
1691 	       rdtsc() - start > 10000 && timer_fired);
1692 
1693 	apic_write(APIC_TMICT, 0);
1694 	irq_disable();
1695 	vmcall();
1696 	timer_fired = false;
1697 	start = rdtsc();
1698 	apic_write(APIC_TMICT, 1000000);
1699 
1700 	irq_enable();
1701 	asm volatile ("nop");
1702 	vmcall();
1703 
1704 	report("direct interrupt + activity state hlt",
1705 	       rdtsc() - start > 10000 && timer_fired);
1706 
1707 	apic_write(APIC_TMICT, 0);
1708 	irq_disable();
1709 	vmcall();
1710 	timer_fired = false;
1711 	start = rdtsc();
1712 	apic_write(APIC_TMICT, 1000000);
1713 
1714 	irq_enable();
1715 	asm volatile ("nop");
1716 	vmcall();
1717 
1718 	report("intercepted interrupt + activity state hlt",
1719 	       rdtsc() - start > 10000 && timer_fired);
1720 
1721 	apic_write(APIC_TMICT, 0);
1722 	irq_disable();
1723 	vmx_set_test_stage(7);
1724 	vmcall();
1725 	timer_fired = false;
1726 	apic_write(APIC_TMICT, 1);
1727 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1728 		asm volatile ("nop");
1729 	report("running a guest with interrupt acknowledgement set", timer_fired);
1730 
1731 	apic_write(APIC_TMICT, 0);
1732 	irq_enable();
1733 	timer_fired = false;
1734 	vmcall();
1735 	report("Inject an event to a halted guest", timer_fired);
1736 }
1737 
1738 static int interrupt_exit_handler(void)
1739 {
1740 	u64 guest_rip = vmcs_read(GUEST_RIP);
1741 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1742 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1743 
1744 	switch (reason) {
1745 	case VMX_VMCALL:
1746 		switch (vmx_get_test_stage()) {
1747 		case 0:
1748 		case 2:
1749 		case 5:
1750 			vmcs_write(PIN_CONTROLS,
1751 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1752 			break;
1753 		case 7:
1754 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1755 			vmcs_write(PIN_CONTROLS,
1756 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1757 			break;
1758 		case 1:
1759 		case 3:
1760 			vmcs_write(PIN_CONTROLS,
1761 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1762 			break;
1763 		case 4:
1764 		case 6:
1765 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1766 			break;
1767 
1768 		case 8:
1769 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1770 			vmcs_write(ENT_INTR_INFO,
1771 				   TIMER_VECTOR |
1772 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1773 				   INTR_INFO_VALID_MASK);
1774 			break;
1775 		}
1776 		vmx_inc_test_stage();
1777 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1778 		return VMX_TEST_RESUME;
1779 	case VMX_EXTINT:
1780 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1781 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1782 			handle_external_interrupt(vector);
1783 		} else {
1784 			irq_enable();
1785 			asm volatile ("nop");
1786 			irq_disable();
1787 		}
1788 		if (vmx_get_test_stage() >= 2)
1789 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1790 		return VMX_TEST_RESUME;
1791 	default:
1792 		report("Unknown exit reason, %ld", false, reason);
1793 		print_vmexit_info();
1794 	}
1795 
1796 	return VMX_TEST_VMEXIT;
1797 }
1798 
1799 static int dbgctls_init(struct vmcs *vmcs)
1800 {
1801 	u64 dr7 = 0x402;
1802 	u64 zero = 0;
1803 
1804 	msr_bmp_init();
1805 	asm volatile(
1806 		"mov %0,%%dr0\n\t"
1807 		"mov %0,%%dr1\n\t"
1808 		"mov %0,%%dr2\n\t"
1809 		"mov %1,%%dr7\n\t"
1810 		: : "r" (zero), "r" (dr7));
1811 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1812 	vmcs_write(GUEST_DR7, 0x404);
1813 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1814 
1815 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1816 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1817 
1818 	return VMX_TEST_START;
1819 }
1820 
1821 static void dbgctls_main(void)
1822 {
1823 	u64 dr7, debugctl;
1824 
1825 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1826 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1827 	/* Commented out: KVM does not support DEBUGCTL so far */
1828 	(void)debugctl;
1829 	report("Load debug controls", dr7 == 0x404 /* && debugctl == 0x2 */);
1830 
1831 	dr7 = 0x408;
1832 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1833 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1834 
1835 	vmx_set_test_stage(0);
1836 	vmcall();
1837 	report("Save debug controls", vmx_get_test_stage() == 1);
1838 
1839 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1840 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1841 		printf("\tDebug controls are always loaded/saved\n");
1842 		return;
1843 	}
1844 	vmx_set_test_stage(2);
1845 	vmcall();
1846 
1847 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1848 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1849 	/* Commented out: KVM does not support DEBUGCTL so far */
1850 	(void)debugctl;
1851 	report("Guest=host debug controls", dr7 == 0x402 /* && debugctl == 0x1 */);
1852 
1853 	dr7 = 0x408;
1854 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1855 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1856 
1857 	vmx_set_test_stage(3);
1858 	vmcall();
1859 	report("Don't save debug controls", vmx_get_test_stage() == 4);
1860 }
1861 
1862 static int dbgctls_exit_handler(void)
1863 {
1864 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
1865 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1866 	u64 guest_rip = vmcs_read(GUEST_RIP);
1867 	u64 dr7, debugctl;
1868 
1869 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1870 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1871 
1872 	switch (reason) {
1873 	case VMX_VMCALL:
1874 		switch (vmx_get_test_stage()) {
1875 		case 0:
1876 			if (dr7 == 0x400 && debugctl == 0 &&
1877 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1878 			    Commented out: KVM does not support DEBUGCTL so far
1879 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1880 				vmx_inc_test_stage();
1881 			break;
1882 		case 2:
1883 			dr7 = 0x402;
1884 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1885 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1886 			vmcs_write(GUEST_DR7, 0x404);
1887 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1888 
1889 			vmcs_write(ENT_CONTROLS,
1890 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1891 			vmcs_write(EXI_CONTROLS,
1892 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1893 			break;
1894 		case 3:
1895 			if (dr7 == 0x400 && debugctl == 0 &&
1896 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1897 			    Commented out: KVM does not support DEBUGCTL so far
1898 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1899 				vmx_inc_test_stage();
1900 			break;
1901 		}
1902 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1903 		return VMX_TEST_RESUME;
1904 	default:
1905 		report("Unknown exit reason, %d", false, reason);
1906 		print_vmexit_info();
1907 	}
1908 	return VMX_TEST_VMEXIT;
1909 }
1910 
1911 struct vmx_msr_entry {
1912 	u32 index;
1913 	u32 reserved;
1914 	u64 value;
1915 } __attribute__((packed));
1916 
1917 #define MSR_MAGIC 0x31415926
1918 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1919 
1920 static int msr_switch_init(struct vmcs *vmcs)
1921 {
1922 	msr_bmp_init();
1923 	exit_msr_store = alloc_page();
1924 	exit_msr_load = alloc_page();
1925 	entry_msr_load = alloc_page();
1926 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1927 	entry_msr_load[0].value = MSR_MAGIC;
1928 
1929 	vmx_set_test_stage(1);
1930 	vmcs_write(ENT_MSR_LD_CNT, 1);
1931 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1932 	vmcs_write(EXI_MSR_ST_CNT, 1);
1933 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1934 	vmcs_write(EXI_MSR_LD_CNT, 1);
1935 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1936 	return VMX_TEST_START;
1937 }
1938 
1939 static void msr_switch_main(void)
1940 {
1941 	if (vmx_get_test_stage() == 1) {
1942 		report("VM entry MSR load",
1943 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC);
1944 		vmx_set_test_stage(2);
1945 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
1946 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
1947 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
1948 		exit_msr_load[0].value = MSR_MAGIC + 2;
1949 	}
1950 	vmcall();
1951 }
1952 
1953 static int msr_switch_exit_handler(void)
1954 {
1955 	ulong reason;
1956 
1957 	reason = vmcs_read(EXI_REASON);
1958 	if (reason == VMX_VMCALL && vmx_get_test_stage() == 2) {
1959 		report("VM exit MSR store",
1960 			exit_msr_store[0].value == MSR_MAGIC + 1);
1961 		report("VM exit MSR load",
1962 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2);
1963 		vmx_set_test_stage(3);
1964 		entry_msr_load[0].index = MSR_FS_BASE;
1965 		return VMX_TEST_RESUME;
1966 	}
1967 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1968 		__func__, vmx_get_test_stage(), reason);
1969 	return VMX_TEST_EXIT;
1970 }
1971 
1972 static int msr_switch_entry_failure(struct vmentry_failure *failure)
1973 {
1974 	ulong reason;
1975 
1976 	if (failure->early) {
1977 		printf("ERROR %s: early exit\n", __func__);
1978 		return VMX_TEST_EXIT;
1979 	}
1980 
1981 	reason = vmcs_read(EXI_REASON);
1982 	if (reason == (VMX_ENTRY_FAILURE | VMX_FAIL_MSR) &&
1983 	    vmx_get_test_stage() == 3) {
1984 		report("VM entry MSR load: try to load FS_BASE",
1985 			vmcs_read(EXI_QUALIFICATION) == 1);
1986 		return VMX_TEST_VMEXIT;
1987 	}
1988 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1989 		__func__, vmx_get_test_stage(), reason);
1990 	return VMX_TEST_EXIT;
1991 }
1992 
1993 static int vmmcall_init(struct vmcs *vmcs)
1994 {
1995 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
1996 	return VMX_TEST_START;
1997 }
1998 
1999 static void vmmcall_main(void)
2000 {
2001 	asm volatile(
2002 		"mov $0xABCD, %%rax\n\t"
2003 		"vmmcall\n\t"
2004 		::: "rax");
2005 
2006 	report("VMMCALL", 0);
2007 }
2008 
2009 static int vmmcall_exit_handler(void)
2010 {
2011 	ulong reason;
2012 
2013 	reason = vmcs_read(EXI_REASON);
2014 	switch (reason) {
2015 	case VMX_VMCALL:
2016 		printf("here\n");
2017 		report("VMMCALL triggers #UD", 0);
2018 		break;
2019 	case VMX_EXC_NMI:
2020 		report("VMMCALL triggers #UD",
2021 		       (vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR);
2022 		break;
2023 	default:
2024 		report("Unknown exit reason, %ld", false, reason);
2025 		print_vmexit_info();
2026 	}
2027 
2028 	return VMX_TEST_VMEXIT;
2029 }
2030 
2031 static int disable_rdtscp_init(struct vmcs *vmcs)
2032 {
2033 	u32 ctrl_cpu1;
2034 
2035 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
2036 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
2037 		ctrl_cpu1 &= ~CPU_RDTSCP;
2038 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
2039 	}
2040 
2041 	return VMX_TEST_START;
2042 }
2043 
2044 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
2045 {
2046 	switch (vmx_get_test_stage()) {
2047 	case 0:
2048 		report("RDTSCP triggers #UD", true);
2049 		vmx_inc_test_stage();
2050 		regs->rip += 3;
2051 		break;
2052 	case 2:
2053 		report("RDPID triggers #UD", true);
2054 		vmx_inc_test_stage();
2055 		regs->rip += 4;
2056 		break;
2057 	}
2058 	return;
2059 
2060 }
2061 
2062 static void disable_rdtscp_main(void)
2063 {
2064 	/* Test that #UD is properly injected in L2.  */
2065 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2066 
2067 	vmx_set_test_stage(0);
2068 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2069 	vmcall();
2070 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2071 
2072 	handle_exception(UD_VECTOR, 0);
2073 	vmcall();
2074 }
2075 
2076 static int disable_rdtscp_exit_handler(void)
2077 {
2078 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
2079 
2080 	switch (reason) {
2081 	case VMX_VMCALL:
2082 		switch (vmx_get_test_stage()) {
2083 		case 0:
2084 			report("RDTSCP triggers #UD", false);
2085 			vmx_inc_test_stage();
2086 			/* fallthrough */
2087 		case 1:
2088 			vmx_inc_test_stage();
2089 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2090 			return VMX_TEST_RESUME;
2091 		case 2:
2092 			report("RDPID triggers #UD", false);
2093 			break;
2094 		}
2095 		break;
2096 
2097 	default:
2098 		report("Unknown exit reason, %d", false, reason);
2099 		print_vmexit_info();
2100 	}
2101 	return VMX_TEST_VMEXIT;
2102 }
2103 
2104 static int int3_init(struct vmcs *vmcs)
2105 {
2106 	vmcs_write(EXC_BITMAP, ~0u);
2107 	return VMX_TEST_START;
2108 }
2109 
2110 static void int3_guest_main(void)
2111 {
2112 	asm volatile ("int3");
2113 }
2114 
2115 static int int3_exit_handler(void)
2116 {
2117 	u32 reason = vmcs_read(EXI_REASON);
2118 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2119 
2120 	report("L1 intercepts #BP", reason == VMX_EXC_NMI &&
2121 	       (intr_info & INTR_INFO_VALID_MASK) &&
2122 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2123 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2124 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2125 
2126 	return VMX_TEST_VMEXIT;
2127 }
2128 
2129 static int into_init(struct vmcs *vmcs)
2130 {
2131 	vmcs_write(EXC_BITMAP, ~0u);
2132 	return VMX_TEST_START;
2133 }
2134 
2135 static void into_guest_main(void)
2136 {
2137 	struct far_pointer32 fp = {
2138 		.offset = (uintptr_t)&&into,
2139 		.selector = KERNEL_CS32,
2140 	};
2141 	register uintptr_t rsp asm("rsp");
2142 
2143 	if (fp.offset != (uintptr_t)&&into) {
2144 		printf("Code address too high.\n");
2145 		return;
2146 	}
2147 	if ((u32)rsp != rsp) {
2148 		printf("Stack address too high.\n");
2149 		return;
2150 	}
2151 
2152 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2153 	return;
2154 into:
2155 	asm volatile (".code32;"
2156 		      "movl $0x7fffffff, %eax;"
2157 		      "addl %eax, %eax;"
2158 		      "into;"
2159 		      "lret;"
2160 		      ".code64");
2161 	__builtin_unreachable();
2162 }
2163 
2164 static int into_exit_handler(void)
2165 {
2166 	u32 reason = vmcs_read(EXI_REASON);
2167 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2168 
2169 	report("L1 intercepts #OF", reason == VMX_EXC_NMI &&
2170 	       (intr_info & INTR_INFO_VALID_MASK) &&
2171 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2172 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2173 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2174 
2175 	return VMX_TEST_VMEXIT;
2176 }
2177 
2178 static void exit_monitor_from_l2_main(void)
2179 {
2180 	printf("Calling exit(0) from l2...\n");
2181 	exit(0);
2182 }
2183 
2184 static int exit_monitor_from_l2_handler(void)
2185 {
2186 	report("The guest should have killed the VMM", false);
2187 	return VMX_TEST_EXIT;
2188 }
2189 
2190 static void assert_exit_reason(u64 expected)
2191 {
2192 	u64 actual = vmcs_read(EXI_REASON);
2193 
2194 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2195 			   exit_reason_description(expected),
2196 			   exit_reason_description(actual));
2197 }
2198 
2199 static void skip_exit_insn(void)
2200 {
2201 	u64 guest_rip = vmcs_read(GUEST_RIP);
2202 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2203 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2204 }
2205 
2206 static void skip_exit_vmcall(void)
2207 {
2208 	assert_exit_reason(VMX_VMCALL);
2209 	skip_exit_insn();
2210 }
2211 
2212 static void v2_null_test_guest(void)
2213 {
2214 }
2215 
2216 static void v2_null_test(void)
2217 {
2218 	test_set_guest(v2_null_test_guest);
2219 	enter_guest();
2220 	report(__func__, 1);
2221 }
2222 
2223 static void v2_multiple_entries_test_guest(void)
2224 {
2225 	vmx_set_test_stage(1);
2226 	vmcall();
2227 	vmx_set_test_stage(2);
2228 }
2229 
2230 static void v2_multiple_entries_test(void)
2231 {
2232 	test_set_guest(v2_multiple_entries_test_guest);
2233 	enter_guest();
2234 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2235 	skip_exit_vmcall();
2236 	enter_guest();
2237 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2238 	report(__func__, 1);
2239 }
2240 
2241 static int fixture_test_data = 1;
2242 
2243 static void fixture_test_teardown(void *data)
2244 {
2245 	*((int *) data) = 1;
2246 }
2247 
2248 static void fixture_test_guest(void)
2249 {
2250 	fixture_test_data++;
2251 }
2252 
2253 
2254 static void fixture_test_setup(void)
2255 {
2256 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2257 			   "fixture_test_teardown didn't run?!");
2258 	fixture_test_data = 2;
2259 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2260 	test_set_guest(fixture_test_guest);
2261 }
2262 
2263 static void fixture_test_case1(void)
2264 {
2265 	fixture_test_setup();
2266 	TEST_ASSERT_EQ(2, fixture_test_data);
2267 	enter_guest();
2268 	TEST_ASSERT_EQ(3, fixture_test_data);
2269 	report(__func__, 1);
2270 }
2271 
2272 static void fixture_test_case2(void)
2273 {
2274 	fixture_test_setup();
2275 	TEST_ASSERT_EQ(2, fixture_test_data);
2276 	enter_guest();
2277 	TEST_ASSERT_EQ(3, fixture_test_data);
2278 	report(__func__, 1);
2279 }
2280 
2281 enum ept_access_op {
2282 	OP_READ,
2283 	OP_WRITE,
2284 	OP_EXEC,
2285 	OP_FLUSH_TLB,
2286 	OP_EXIT,
2287 };
2288 
2289 static struct ept_access_test_data {
2290 	unsigned long gpa;
2291 	unsigned long *gva;
2292 	unsigned long hpa;
2293 	unsigned long *hva;
2294 	enum ept_access_op op;
2295 } ept_access_test_data;
2296 
2297 extern unsigned char ret42_start;
2298 extern unsigned char ret42_end;
2299 
2300 /* Returns 42. */
2301 asm(
2302 	".align 64\n"
2303 	"ret42_start:\n"
2304 	"mov $42, %eax\n"
2305 	"ret\n"
2306 	"ret42_end:\n"
2307 );
2308 
2309 static void
2310 diagnose_ept_violation_qual(u64 expected, u64 actual)
2311 {
2312 
2313 #define DIAGNOSE(flag)							\
2314 do {									\
2315 	if ((expected & flag) != (actual & flag))			\
2316 		printf(#flag " %sexpected\n",				\
2317 		       (expected & flag) ? "" : "un");			\
2318 } while (0)
2319 
2320 	DIAGNOSE(EPT_VLT_RD);
2321 	DIAGNOSE(EPT_VLT_WR);
2322 	DIAGNOSE(EPT_VLT_FETCH);
2323 	DIAGNOSE(EPT_VLT_PERM_RD);
2324 	DIAGNOSE(EPT_VLT_PERM_WR);
2325 	DIAGNOSE(EPT_VLT_PERM_EX);
2326 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2327 	DIAGNOSE(EPT_VLT_PADDR);
2328 
2329 #undef DIAGNOSE
2330 }
2331 
2332 static void do_ept_access_op(enum ept_access_op op)
2333 {
2334 	ept_access_test_data.op = op;
2335 	enter_guest();
2336 }
2337 
2338 /*
2339  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2340  * needed by tests that modify guest PTEs.
2341  */
2342 static void ept_access_test_guest_flush_tlb(void)
2343 {
2344 	do_ept_access_op(OP_FLUSH_TLB);
2345 	skip_exit_vmcall();
2346 }
2347 
2348 /*
2349  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2350  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2351  * a huge page.
2352  */
2353 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2354 				 unsigned long clear, unsigned long set)
2355 {
2356 	struct ept_access_test_data *data = &ept_access_test_data;
2357 	unsigned long orig_pte;
2358 	unsigned long pte;
2359 
2360 	/* Screw with the mapping at the requested level. */
2361 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2362 	pte = orig_pte;
2363 	if (mkhuge)
2364 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2365 	else
2366 		pte = orig_pte;
2367 	pte = (pte & ~clear) | set;
2368 	set_ept_pte(pml4, gpa, level, pte);
2369 	ept_sync(INVEPT_SINGLE, eptp);
2370 
2371 	return orig_pte;
2372 }
2373 
2374 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2375 {
2376 	set_ept_pte(pml4, gpa, level, orig_pte);
2377 	ept_sync(INVEPT_SINGLE, eptp);
2378 }
2379 
2380 static void do_ept_violation(bool leaf, enum ept_access_op op,
2381 			     u64 expected_qual, u64 expected_paddr)
2382 {
2383 	u64 qual;
2384 
2385 	/* Try the access and observe the violation. */
2386 	do_ept_access_op(op);
2387 
2388 	assert_exit_reason(VMX_EPT_VIOLATION);
2389 
2390 	qual = vmcs_read(EXI_QUALIFICATION);
2391 
2392 	/* Mask undefined bits (which may later be defined in certain cases). */
2393 	qual &= ~(EPT_VLT_GUEST_USER | EPT_VLT_GUEST_RW | EPT_VLT_GUEST_EX |
2394 		 EPT_VLT_PERM_USER_EX);
2395 
2396 	diagnose_ept_violation_qual(expected_qual, qual);
2397 	TEST_EXPECT_EQ(expected_qual, qual);
2398 
2399 	#if 0
2400 	/* Disable for now otherwise every test will fail */
2401 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2402 		       (unsigned long) (
2403 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2404 	#endif
2405 	/*
2406 	 * TODO: tests that probe expected_paddr in pages other than the one at
2407 	 * the beginning of the 1g region.
2408 	 */
2409 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2410 }
2411 
2412 static void
2413 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2414 			      unsigned long set, enum ept_access_op op,
2415 			      u64 expected_qual)
2416 {
2417 	struct ept_access_test_data *data = &ept_access_test_data;
2418 	unsigned long orig_pte;
2419 
2420 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2421 
2422 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2423 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2424 					 data->gpa);
2425 
2426 	/* Fix the violation and resume the op loop. */
2427 	ept_untwiddle(data->gpa, level, orig_pte);
2428 	enter_guest();
2429 	skip_exit_vmcall();
2430 }
2431 
2432 static void
2433 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2434 		       enum ept_access_op op, u64 expected_qual)
2435 {
2436 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2437 				      expected_qual);
2438 	if (ept_huge_pages_supported(level))
2439 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2440 					      expected_qual);
2441 }
2442 
2443 static void ept_violation(unsigned long clear, unsigned long set,
2444 			  enum ept_access_op op, u64 expected_qual)
2445 {
2446 	ept_violation_at_level(1, clear, set, op, expected_qual);
2447 	ept_violation_at_level(2, clear, set, op, expected_qual);
2448 	ept_violation_at_level(3, clear, set, op, expected_qual);
2449 	ept_violation_at_level(4, clear, set, op, expected_qual);
2450 }
2451 
2452 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2453 				       u64 expected_qual)
2454 {
2455 	ept_violation(EPT_PRESENT, access, op,
2456 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2457 }
2458 
2459 /*
2460  * For translations that don't involve a GVA, that is physical address (paddr)
2461  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2462  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2463  * translations don't involve GVAs, such as when the hardware does the guest
2464  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2465  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2466  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2467  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2468  *
2469  * Note that paddr violations can also be triggered by loading PAE page tables
2470  * with wonky addresses. We don't test that yet.
2471  *
2472  * This function modifies the EPT entry that maps the GPA that the guest page
2473  * table entry mapping ept_access_test_data.gva resides on.
2474  *
2475  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2476  *
2477  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2478  *
2479  *	@op		Guest operation to perform with
2480  *			ept_access_test_data.gva.
2481  *
2482  *	@expect_violation
2483  *			Is a violation expected during the paddr access?
2484  *
2485  *	@expected_qual	Expected qualification for the EPT violation.
2486  *			EPT_VLT_PADDR should be clear.
2487  */
2488 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2489 			     enum ept_access_op op, bool expect_violation,
2490 			     u64 expected_qual)
2491 {
2492 	struct ept_access_test_data *data = &ept_access_test_data;
2493 	unsigned long *ptep;
2494 	unsigned long gpa;
2495 	unsigned long orig_epte;
2496 
2497 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2498 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2499 	TEST_ASSERT(ptep);
2500 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2501 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2502 	ept_access_test_guest_flush_tlb();
2503 
2504 	/*
2505 	 * Now modify the access bits on the EPT entry for the GPA that the
2506 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2507 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2508 	 * constructed our test such that those other 511 PTEs aren't used by
2509 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2510 	 * PTE we're modifying is at the beginning of a 4K page and the
2511 	 * following 511 entires are also under our control (and not touched by
2512 	 * the guest).
2513 	 */
2514 	gpa = virt_to_phys(ptep);
2515 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2516 	/*
2517 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2518 	 * otherwise our level=1 twiddling below will fail. We use the
2519 	 * identity map (gpa = gpa) since page tables are shared with the host.
2520 	 */
2521 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2522 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2523 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2524 
2525 	if (expect_violation) {
2526 		do_ept_violation(/*leaf=*/true, op,
2527 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2528 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2529 		do_ept_access_op(op);
2530 	} else {
2531 		do_ept_access_op(op);
2532 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2533 	}
2534 
2535 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2536 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2537 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2538 
2539 	skip_exit_vmcall();
2540 }
2541 
2542 static void ept_access_allowed_paddr(unsigned long ept_access,
2543 				     unsigned long pte_ad,
2544 				     enum ept_access_op op)
2545 {
2546 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2547 			 /*expected_qual=*/-1);
2548 }
2549 
2550 static void ept_access_violation_paddr(unsigned long ept_access,
2551 				       unsigned long pte_ad,
2552 				       enum ept_access_op op,
2553 				       u64 expected_qual)
2554 {
2555 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2556 			 expected_qual);
2557 }
2558 
2559 
2560 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2561 					unsigned long clear,
2562 					unsigned long set,
2563 					enum ept_access_op op)
2564 {
2565 	struct ept_access_test_data *data = &ept_access_test_data;
2566 	unsigned long orig_pte;
2567 
2568 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2569 
2570 	/* No violation. Should proceed to vmcall. */
2571 	do_ept_access_op(op);
2572 	skip_exit_vmcall();
2573 
2574 	ept_untwiddle(data->gpa, level, orig_pte);
2575 }
2576 
2577 static void ept_allowed_at_level(int level, unsigned long clear,
2578 				 unsigned long set, enum ept_access_op op)
2579 {
2580 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2581 	if (ept_huge_pages_supported(level))
2582 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2583 }
2584 
2585 static void ept_allowed(unsigned long clear, unsigned long set,
2586 			enum ept_access_op op)
2587 {
2588 	ept_allowed_at_level(1, clear, set, op);
2589 	ept_allowed_at_level(2, clear, set, op);
2590 	ept_allowed_at_level(3, clear, set, op);
2591 	ept_allowed_at_level(4, clear, set, op);
2592 }
2593 
2594 static void ept_ignored_bit(int bit)
2595 {
2596 	/* Set the bit. */
2597 	ept_allowed(0, 1ul << bit, OP_READ);
2598 	ept_allowed(0, 1ul << bit, OP_WRITE);
2599 	ept_allowed(0, 1ul << bit, OP_EXEC);
2600 
2601 	/* Clear the bit. */
2602 	ept_allowed(1ul << bit, 0, OP_READ);
2603 	ept_allowed(1ul << bit, 0, OP_WRITE);
2604 	ept_allowed(1ul << bit, 0, OP_EXEC);
2605 }
2606 
2607 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2608 {
2609 	ept_allowed(EPT_PRESENT, access, op);
2610 }
2611 
2612 
2613 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2614 					     unsigned long clear,
2615 					     unsigned long set,
2616 					     enum ept_access_op op)
2617 {
2618 	struct ept_access_test_data *data = &ept_access_test_data;
2619 	unsigned long orig_pte;
2620 
2621 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2622 
2623 	do_ept_access_op(op);
2624 	assert_exit_reason(VMX_EPT_MISCONFIG);
2625 
2626 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2627 	#if 0
2628 	/* broken: */
2629 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2630 	#endif
2631 	#if 0
2632 	/*
2633 	 * broken:
2634 	 * According to description of exit qual for EPT violation,
2635 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2636 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2637 	 * to be set for msiconfig.
2638 	 */
2639 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2640 		       (unsigned long) (
2641 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2642 	#endif
2643 
2644 	/* Fix the violation and resume the op loop. */
2645 	ept_untwiddle(data->gpa, level, orig_pte);
2646 	enter_guest();
2647 	skip_exit_vmcall();
2648 }
2649 
2650 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2651 					  unsigned long clear,
2652 					  unsigned long set)
2653 {
2654 	/* The op shouldn't matter (read, write, exec), so try them all! */
2655 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2656 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2657 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2658 }
2659 
2660 static void ept_misconfig_at_level(int level, unsigned long clear,
2661 				   unsigned long set)
2662 {
2663 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2664 	if (ept_huge_pages_supported(level))
2665 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2666 }
2667 
2668 static void ept_misconfig(unsigned long clear, unsigned long set)
2669 {
2670 	ept_misconfig_at_level(1, clear, set);
2671 	ept_misconfig_at_level(2, clear, set);
2672 	ept_misconfig_at_level(3, clear, set);
2673 	ept_misconfig_at_level(4, clear, set);
2674 }
2675 
2676 static void ept_access_misconfig(unsigned long access)
2677 {
2678 	ept_misconfig(EPT_PRESENT, access);
2679 }
2680 
2681 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2682 {
2683 	/* Setting the bit causes a misconfig. */
2684 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2685 
2686 	/* Making the entry non-present turns reserved bits into ignored. */
2687 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2688 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2689 }
2690 
2691 static void ept_reserved_bit_at_level_huge(int level, int bit)
2692 {
2693 	/* Setting the bit causes a misconfig. */
2694 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2695 
2696 	/* Making the entry non-present turns reserved bits into ignored. */
2697 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2698 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2699 }
2700 
2701 static void ept_reserved_bit_at_level(int level, int bit)
2702 {
2703 	/* Setting the bit causes a misconfig. */
2704 	ept_misconfig_at_level(level, 0, 1ul << bit);
2705 
2706 	/* Making the entry non-present turns reserved bits into ignored. */
2707 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2708 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2709 }
2710 
2711 static void ept_reserved_bit(int bit)
2712 {
2713 	ept_reserved_bit_at_level(1, bit);
2714 	ept_reserved_bit_at_level(2, bit);
2715 	ept_reserved_bit_at_level(3, bit);
2716 	ept_reserved_bit_at_level(4, bit);
2717 }
2718 
2719 #define PAGE_2M_ORDER 9
2720 #define PAGE_1G_ORDER 18
2721 
2722 static void *get_1g_page(void)
2723 {
2724 	static void *alloc;
2725 
2726 	if (!alloc)
2727 		alloc = alloc_pages(PAGE_1G_ORDER);
2728 	return alloc;
2729 }
2730 
2731 static void ept_access_test_teardown(void *unused)
2732 {
2733 	/* Exit the guest cleanly. */
2734 	do_ept_access_op(OP_EXIT);
2735 }
2736 
2737 static void ept_access_test_guest(void)
2738 {
2739 	struct ept_access_test_data *data = &ept_access_test_data;
2740 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2741 
2742 	while (true) {
2743 		switch (data->op) {
2744 		case OP_READ:
2745 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2746 			break;
2747 		case OP_WRITE:
2748 			*data->gva = MAGIC_VAL_2;
2749 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2750 			*data->gva = MAGIC_VAL_1;
2751 			break;
2752 		case OP_EXEC:
2753 			TEST_ASSERT_EQ(42, code());
2754 			break;
2755 		case OP_FLUSH_TLB:
2756 			write_cr3(read_cr3());
2757 			break;
2758 		case OP_EXIT:
2759 			return;
2760 		default:
2761 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2762 		}
2763 		vmcall();
2764 	}
2765 }
2766 
2767 static void ept_access_test_setup(void)
2768 {
2769 	struct ept_access_test_data *data = &ept_access_test_data;
2770 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2771 	unsigned long size = npages * PAGE_SIZE;
2772 	unsigned long *page_table = current_page_table();
2773 	unsigned long pte;
2774 
2775 	if (setup_ept(false))
2776 		test_skip("EPT not supported");
2777 
2778 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2779 	if (cpuid_maxphyaddr() < 40)
2780 		test_skip("Test needs MAXPHYADDR >= 40");
2781 
2782 	test_set_guest(ept_access_test_guest);
2783 	test_add_teardown(ept_access_test_teardown, NULL);
2784 
2785 	data->hva = get_1g_page();
2786 	TEST_ASSERT(data->hva);
2787 	data->hpa = virt_to_phys(data->hva);
2788 
2789 	data->gpa = 1ul << 39;
2790 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2791 				   size);
2792 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2793 	install_pages(page_table, data->gpa, size, data->gva);
2794 
2795 	/*
2796 	 * Make sure nothing's mapped here so the tests that screw with the
2797 	 * pml4 entry don't inadvertently break something.
2798 	 */
2799 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2800 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2801 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2802 
2803 	data->hva[0] = MAGIC_VAL_1;
2804 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2805 }
2806 
2807 static void ept_access_test_not_present(void)
2808 {
2809 	ept_access_test_setup();
2810 	/* --- */
2811 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2812 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2813 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2814 }
2815 
2816 static void ept_access_test_read_only(void)
2817 {
2818 	ept_access_test_setup();
2819 
2820 	/* r-- */
2821 	ept_access_allowed(EPT_RA, OP_READ);
2822 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2823 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2824 }
2825 
2826 static void ept_access_test_write_only(void)
2827 {
2828 	ept_access_test_setup();
2829 	/* -w- */
2830 	ept_access_misconfig(EPT_WA);
2831 }
2832 
2833 static void ept_access_test_read_write(void)
2834 {
2835 	ept_access_test_setup();
2836 	/* rw- */
2837 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2838 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2839 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2840 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2841 }
2842 
2843 
2844 static void ept_access_test_execute_only(void)
2845 {
2846 	ept_access_test_setup();
2847 	/* --x */
2848 	if (ept_execute_only_supported()) {
2849 		ept_access_violation(EPT_EA, OP_READ,
2850 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2851 		ept_access_violation(EPT_EA, OP_WRITE,
2852 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2853 		ept_access_allowed(EPT_EA, OP_EXEC);
2854 	} else {
2855 		ept_access_misconfig(EPT_EA);
2856 	}
2857 }
2858 
2859 static void ept_access_test_read_execute(void)
2860 {
2861 	ept_access_test_setup();
2862 	/* r-x */
2863 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2864 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2865 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2866 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2867 }
2868 
2869 static void ept_access_test_write_execute(void)
2870 {
2871 	ept_access_test_setup();
2872 	/* -wx */
2873 	ept_access_misconfig(EPT_WA | EPT_EA);
2874 }
2875 
2876 static void ept_access_test_read_write_execute(void)
2877 {
2878 	ept_access_test_setup();
2879 	/* rwx */
2880 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2881 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2882 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2883 }
2884 
2885 static void ept_access_test_reserved_bits(void)
2886 {
2887 	int i;
2888 	int maxphyaddr;
2889 
2890 	ept_access_test_setup();
2891 
2892 	/* Reserved bits above maxphyaddr. */
2893 	maxphyaddr = cpuid_maxphyaddr();
2894 	for (i = maxphyaddr; i <= 51; i++) {
2895 		report_prefix_pushf("reserved_bit=%d", i);
2896 		ept_reserved_bit(i);
2897 		report_prefix_pop();
2898 	}
2899 
2900 	/* Level-specific reserved bits. */
2901 	ept_reserved_bit_at_level_nohuge(2, 3);
2902 	ept_reserved_bit_at_level_nohuge(2, 4);
2903 	ept_reserved_bit_at_level_nohuge(2, 5);
2904 	ept_reserved_bit_at_level_nohuge(2, 6);
2905 	/* 2M alignment. */
2906 	for (i = 12; i < 20; i++) {
2907 		report_prefix_pushf("reserved_bit=%d", i);
2908 		ept_reserved_bit_at_level_huge(2, i);
2909 		report_prefix_pop();
2910 	}
2911 	ept_reserved_bit_at_level_nohuge(3, 3);
2912 	ept_reserved_bit_at_level_nohuge(3, 4);
2913 	ept_reserved_bit_at_level_nohuge(3, 5);
2914 	ept_reserved_bit_at_level_nohuge(3, 6);
2915 	/* 1G alignment. */
2916 	for (i = 12; i < 29; i++) {
2917 		report_prefix_pushf("reserved_bit=%d", i);
2918 		ept_reserved_bit_at_level_huge(3, i);
2919 		report_prefix_pop();
2920 	}
2921 	ept_reserved_bit_at_level(4, 3);
2922 	ept_reserved_bit_at_level(4, 4);
2923 	ept_reserved_bit_at_level(4, 5);
2924 	ept_reserved_bit_at_level(4, 6);
2925 	ept_reserved_bit_at_level(4, 7);
2926 }
2927 
2928 static void ept_access_test_ignored_bits(void)
2929 {
2930 	ept_access_test_setup();
2931 	/*
2932 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2933 	 * far as translation is concerned even if AD bits are enabled in the
2934 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2935 	 * control is 0.
2936 	 */
2937 	ept_ignored_bit(8);
2938 	ept_ignored_bit(9);
2939 	ept_ignored_bit(10);
2940 	ept_ignored_bit(11);
2941 	ept_ignored_bit(52);
2942 	ept_ignored_bit(53);
2943 	ept_ignored_bit(54);
2944 	ept_ignored_bit(55);
2945 	ept_ignored_bit(56);
2946 	ept_ignored_bit(57);
2947 	ept_ignored_bit(58);
2948 	ept_ignored_bit(59);
2949 	ept_ignored_bit(60);
2950 	ept_ignored_bit(61);
2951 	ept_ignored_bit(62);
2952 	ept_ignored_bit(63);
2953 }
2954 
2955 static void ept_access_test_paddr_not_present_ad_disabled(void)
2956 {
2957 	ept_access_test_setup();
2958 	ept_disable_ad_bits();
2959 
2960 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
2961 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
2962 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
2963 }
2964 
2965 static void ept_access_test_paddr_not_present_ad_enabled(void)
2966 {
2967 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
2968 
2969 	ept_access_test_setup();
2970 	ept_enable_ad_bits_or_skip_test();
2971 
2972 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
2973 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
2974 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
2975 }
2976 
2977 static void ept_access_test_paddr_read_only_ad_disabled(void)
2978 {
2979 	/*
2980 	 * When EPT AD bits are disabled, all accesses to guest paging
2981 	 * structures are reported separately as a read and (after
2982 	 * translation of the GPA to host physical address) a read+write
2983 	 * if the A/D bits have to be set.
2984 	 */
2985 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2986 
2987 	ept_access_test_setup();
2988 	ept_disable_ad_bits();
2989 
2990 	/* Can't update A bit, so all accesses fail. */
2991 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2992 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
2993 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
2994 	/* AD bits disabled, so only writes try to update the D bit. */
2995 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
2996 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
2997 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
2998 	/* Both A and D already set, so read-only is OK. */
2999 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
3000 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
3001 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
3002 }
3003 
3004 static void ept_access_test_paddr_read_only_ad_enabled(void)
3005 {
3006 	/*
3007 	 * When EPT AD bits are enabled, all accesses to guest paging
3008 	 * structures are considered writes as far as EPT translation
3009 	 * is concerned.
3010 	 */
3011 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3012 
3013 	ept_access_test_setup();
3014 	ept_enable_ad_bits_or_skip_test();
3015 
3016 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3017 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3018 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3019 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
3020 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3021 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
3022 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
3023 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
3024 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
3025 }
3026 
3027 static void ept_access_test_paddr_read_write(void)
3028 {
3029 	ept_access_test_setup();
3030 	/* Read-write access to paging structure. */
3031 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
3032 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
3033 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
3034 }
3035 
3036 static void ept_access_test_paddr_read_write_execute(void)
3037 {
3038 	ept_access_test_setup();
3039 	/* RWX access to paging structure. */
3040 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
3041 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
3042 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
3043 }
3044 
3045 static void ept_access_test_paddr_read_execute_ad_disabled(void)
3046 {
3047   	/*
3048 	 * When EPT AD bits are disabled, all accesses to guest paging
3049 	 * structures are reported separately as a read and (after
3050 	 * translation of the GPA to host physical address) a read+write
3051 	 * if the A/D bits have to be set.
3052 	 */
3053 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3054 
3055 	ept_access_test_setup();
3056 	ept_disable_ad_bits();
3057 
3058 	/* Can't update A bit, so all accesses fail. */
3059 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3060 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3061 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3062 	/* AD bits disabled, so only writes try to update the D bit. */
3063 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
3064 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3065 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3066 	/* Both A and D already set, so read-only is OK. */
3067 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3068 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3069 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3070 }
3071 
3072 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3073 {
3074 	/*
3075 	 * When EPT AD bits are enabled, all accesses to guest paging
3076 	 * structures are considered writes as far as EPT translation
3077 	 * is concerned.
3078 	 */
3079 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3080 
3081 	ept_access_test_setup();
3082 	ept_enable_ad_bits_or_skip_test();
3083 
3084 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3085 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3086 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3087 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3088 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3089 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3090 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3091 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3092 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3093 }
3094 
3095 static void ept_access_test_paddr_not_present_page_fault(void)
3096 {
3097 	ept_access_test_setup();
3098 	/*
3099 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3100 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3101 	 * Thus guest page fault before host takes EPT violation for trying to
3102 	 * update A bit.
3103 	 */
3104 }
3105 
3106 static void ept_access_test_force_2m_page(void)
3107 {
3108 	ept_access_test_setup();
3109 
3110 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3111 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3112 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3113 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3114 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3115 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3116 }
3117 
3118 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3119 {
3120 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3121 
3122 	TEST_ASSERT(msr & VPID_CAP_INVVPID);
3123 
3124 	if (type < INVVPID_ADDR || type > INVVPID_CONTEXT_LOCAL)
3125 		return false;
3126 
3127 	if (!(msr & (1ull << (type + VPID_CAP_INVVPID_TYPES_SHIFT))))
3128 		return false;
3129 
3130 	if (vpid >> 16)
3131 		return false;
3132 
3133 	if (type != INVVPID_ALL && !vpid)
3134 		return false;
3135 
3136 	if (type == INVVPID_ADDR && !is_canonical(gla))
3137 		return false;
3138 
3139 	return true;
3140 }
3141 
3142 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3143 {
3144 	int rc;
3145 	bool valid = invvpid_valid(type, vpid, gla);
3146 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3147 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3148 	/*
3149 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3150 	 * that we can tell if it is updated by INVVPID.
3151 	 */
3152 	vmcs_read(~0);
3153 	rc = invvpid(type, vpid, gla);
3154 	report("INVVPID type %ld VPID %lx GLA %lx %s",
3155 	       !rc == valid, type, vpid, gla,
3156 	       valid ? "passes" : "fails");
3157 	report("After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3158 	       vmcs_read(VMX_INST_ERROR) == expected,
3159 	       rc ? "failed" : "successful",
3160 	       expected, vmcs_read(VMX_INST_ERROR));
3161 }
3162 
3163 static void ds_invvpid(void *data)
3164 {
3165 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3166 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3167 
3168 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3169 	asm volatile("invvpid %0, %1"
3170 		     :
3171 		     : "m"(*(struct invvpid_operand *)data),
3172 		       "r"(type));
3173 }
3174 
3175 /*
3176  * The SS override is ignored in 64-bit mode, so we use an addressing
3177  * mode with %rsp as the base register to generate an implicit SS
3178  * reference.
3179  */
3180 static void ss_invvpid(void *data)
3181 {
3182 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3183 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3184 
3185 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3186 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3187 		     : "+r"(data)
3188 		     : "r"(type));
3189 }
3190 
3191 static void invvpid_test_gp(void)
3192 {
3193 	bool fault;
3194 
3195 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3196 				   (void *)NONCANONICAL);
3197 	report("INVVPID with non-canonical DS operand raises #GP", fault);
3198 }
3199 
3200 static void invvpid_test_ss(void)
3201 {
3202 	bool fault;
3203 
3204 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3205 				   (void *)NONCANONICAL);
3206 	report("INVVPID with non-canonical SS operand raises #SS", fault);
3207 }
3208 
3209 static void invvpid_test_pf(void)
3210 {
3211 	void *vpage = alloc_vpage();
3212 	bool fault;
3213 
3214 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3215 	report("INVVPID with unmapped operand raises #PF", fault);
3216 }
3217 
3218 static void try_compat_invvpid(void *unused)
3219 {
3220 	struct far_pointer32 fp = {
3221 		.offset = (uintptr_t)&&invvpid,
3222 		.selector = KERNEL_CS32,
3223 	};
3224 	register uintptr_t rsp asm("rsp");
3225 
3226 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3227 			"Code address too high.");
3228 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3229 
3230 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3231 	return;
3232 invvpid:
3233 	asm volatile (".code32;"
3234 		      "invvpid (%eax), %eax;"
3235 		      "lret;"
3236 		      ".code64");
3237 	__builtin_unreachable();
3238 }
3239 
3240 static void invvpid_test_compatibility_mode(void)
3241 {
3242 	bool fault;
3243 
3244 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3245 	report("Compatibility mode INVVPID raises #UD", fault);
3246 }
3247 
3248 static void invvpid_test_not_in_vmx_operation(void)
3249 {
3250 	bool fault;
3251 
3252 	TEST_ASSERT(!vmx_off());
3253 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3254 	report("INVVPID outside of VMX operation raises #UD", fault);
3255 	TEST_ASSERT(!vmx_on());
3256 }
3257 
3258 /*
3259  * This does not test real-address mode, virtual-8086 mode, protected mode,
3260  * or CPL > 0.
3261  */
3262 static void invvpid_test_v2(void)
3263 {
3264 	u64 msr;
3265 	int i;
3266 	unsigned types = 0;
3267 	unsigned type;
3268 
3269 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
3270 	    !(ctrl_cpu_rev[1].clr & CPU_VPID))
3271 		test_skip("VPID not supported");
3272 
3273 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3274 
3275 	if (!(msr & VPID_CAP_INVVPID))
3276 		test_skip("INVVPID not supported.\n");
3277 
3278 	if (msr & VPID_CAP_INVVPID_ADDR)
3279 		types |= 1u << INVVPID_ADDR;
3280 	if (msr & VPID_CAP_INVVPID_CXTGLB)
3281 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3282 	if (msr & VPID_CAP_INVVPID_ALL)
3283 		types |= 1u << INVVPID_ALL;
3284 	if (msr & VPID_CAP_INVVPID_CXTLOC)
3285 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3286 
3287 	if (!types)
3288 		test_skip("No INVVPID types supported.\n");
3289 
3290 	for (i = -127; i < 128; i++)
3291 		try_invvpid(i, 0xffff, 0);
3292 
3293 	/*
3294 	 * VPID must not be more than 16 bits.
3295 	 */
3296 	for (i = 0; i < 64; i++)
3297 		for (type = 0; type < 4; type++)
3298 			if (types & (1u << type))
3299 				try_invvpid(type, 1ul << i, 0);
3300 
3301 	/*
3302 	 * VPID must not be zero, except for "all contexts."
3303 	 */
3304 	for (type = 0; type < 4; type++)
3305 		if (types & (1u << type))
3306 			try_invvpid(type, 0, 0);
3307 
3308 	/*
3309 	 * The gla operand is only validated for single-address INVVPID.
3310 	 */
3311 	if (types & (1u << INVVPID_ADDR))
3312 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3313 
3314 	invvpid_test_gp();
3315 	invvpid_test_ss();
3316 	invvpid_test_pf();
3317 	invvpid_test_compatibility_mode();
3318 	invvpid_test_not_in_vmx_operation();
3319 }
3320 
3321 /*
3322  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3323  * at least as far as the guest-state checks. Returns false if the
3324  * VMLAUNCH fails early and execution falls through to the next
3325  * instruction.
3326  */
3327 static bool vmlaunch_succeeds(void)
3328 {
3329 	u32 exit_reason;
3330 
3331 	/*
3332 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3333 	 * unsupported VMCS component"). The caller can then check
3334 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3335 	 */
3336 	vmcs_write(~0u, 0);
3337 
3338 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3339 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3340 				   :
3341 				   : "r" ((u64)HOST_RSP)
3342 				   : "cc", "memory"
3343 				   : success);
3344 	return false;
3345 success:
3346 	exit_reason = vmcs_read(EXI_REASON);
3347 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3348 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3349 	return true;
3350 }
3351 
3352 /*
3353  * Try to launch the current VMCS.
3354  */
3355 static void test_vmx_vmlaunch(u32 xerror)
3356 {
3357 	bool success = vmlaunch_succeeds();
3358 	u32 vmx_inst_err;
3359 
3360 	report("vmlaunch %s", success == !xerror,
3361 	       !xerror ? "succeeds" : "fails");
3362 	if (!success && xerror) {
3363 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3364 		report("VMX inst error is %d (actual %d)",
3365 		       vmx_inst_err == xerror, xerror, vmx_inst_err);
3366 	}
3367 }
3368 
3369 static void test_vmx_invalid_controls(void)
3370 {
3371 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3372 }
3373 
3374 static void test_vmx_valid_controls(void)
3375 {
3376 	test_vmx_vmlaunch(0);
3377 }
3378 
3379 /*
3380  * Test a particular value of a VM-execution control bit, if the value
3381  * is required or if the value is zero.
3382  */
3383 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3384 				    enum Encoding encoding, unsigned bit,
3385 				    unsigned val)
3386 {
3387 	u32 mask = 1u << bit;
3388 	bool expected;
3389 	u32 controls;
3390 
3391 	if (msr.set & mask)
3392 		TEST_ASSERT(msr.clr & mask);
3393 
3394 	/*
3395 	 * We can't arbitrarily turn on a control bit, because it may
3396 	 * introduce dependencies on other VMCS fields. So, we only
3397 	 * test turning on bits that have a required setting.
3398 	 */
3399 	if (val && (msr.clr & mask) && !(msr.set & mask))
3400 		return;
3401 
3402 	report_prefix_pushf("%s %s bit %d",
3403 			    val ? "Set" : "Clear", name, bit);
3404 
3405 	controls = vmcs_read(encoding);
3406 	if (val) {
3407 		vmcs_write(encoding, msr.set | mask);
3408 		expected = (msr.clr & mask);
3409 	} else {
3410 		vmcs_write(encoding, msr.set & ~mask);
3411 		expected = !(msr.set & mask);
3412 	}
3413 	if (expected)
3414 		test_vmx_valid_controls();
3415 	else
3416 		test_vmx_invalid_controls();
3417 	vmcs_write(encoding, controls);
3418 	report_prefix_pop();
3419 }
3420 
3421 /*
3422  * Test reserved values of a VM-execution control bit, based on the
3423  * allowed bit settings from the corresponding VMX capability MSR.
3424  */
3425 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3426 			      enum Encoding encoding, unsigned bit)
3427 {
3428 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3429 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3430 }
3431 
3432 /*
3433  * Reserved bits in the pin-based VM-execution controls must be set
3434  * properly. Software may consult the VMX capability MSRs to determine
3435  * the proper settings.
3436  * [Intel SDM]
3437  */
3438 static void test_pin_based_ctls(void)
3439 {
3440 	unsigned bit;
3441 
3442 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3443 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3444 	for (bit = 0; bit < 32; bit++)
3445 		test_rsvd_ctl_bit("pin-based controls",
3446 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3447 }
3448 
3449 /*
3450  * Reserved bits in the primary processor-based VM-execution controls
3451  * must be set properly. Software may consult the VMX capability MSRs
3452  * to determine the proper settings.
3453  * [Intel SDM]
3454  */
3455 static void test_primary_processor_based_ctls(void)
3456 {
3457 	unsigned bit;
3458 
3459 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3460 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3461 	for (bit = 0; bit < 32; bit++)
3462 		test_rsvd_ctl_bit("primary processor-based controls",
3463 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3464 }
3465 
3466 /*
3467  * If the "activate secondary controls" primary processor-based
3468  * VM-execution control is 1, reserved bits in the secondary
3469  * processor-based VM-execution controls must be cleared. Software may
3470  * consult the VMX capability MSRs to determine which bits are
3471  * reserved.
3472  * If the "activate secondary controls" primary processor-based
3473  * VM-execution control is 0 (or if the processor does not support the
3474  * 1-setting of that control), no checks are performed on the
3475  * secondary processor-based VM-execution controls.
3476  * [Intel SDM]
3477  */
3478 static void test_secondary_processor_based_ctls(void)
3479 {
3480 	u32 primary;
3481 	u32 secondary;
3482 	unsigned bit;
3483 
3484 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3485 		return;
3486 
3487 	primary = vmcs_read(CPU_EXEC_CTRL0);
3488 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3489 
3490 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3491 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3492 	for (bit = 0; bit < 32; bit++)
3493 		test_rsvd_ctl_bit("secondary processor-based controls",
3494 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3495 
3496 	/*
3497 	 * When the "activate secondary controls" VM-execution control
3498 	 * is clear, there are no checks on the secondary controls.
3499 	 */
3500 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3501 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3502 	report("Secondary processor-based controls ignored",
3503 	       vmlaunch_succeeds());
3504 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3505 	vmcs_write(CPU_EXEC_CTRL0, primary);
3506 }
3507 
3508 static void try_cr3_target_count(unsigned i, unsigned max)
3509 {
3510 	report_prefix_pushf("CR3 target count 0x%x", i);
3511 	vmcs_write(CR3_TARGET_COUNT, i);
3512 	if (i <= max)
3513 		test_vmx_valid_controls();
3514 	else
3515 		test_vmx_invalid_controls();
3516 	report_prefix_pop();
3517 }
3518 
3519 /*
3520  * The CR3-target count must not be greater than 4. Future processors
3521  * may support a different number of CR3-target values. Software
3522  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3523  * number of values supported.
3524  * [Intel SDM]
3525  */
3526 static void test_cr3_targets(void)
3527 {
3528 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3529 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3530 	unsigned i;
3531 
3532 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3533 	TEST_ASSERT(supported_targets <= 256);
3534 
3535 	try_cr3_target_count(-1u, supported_targets);
3536 	try_cr3_target_count(0x80000000, supported_targets);
3537 	try_cr3_target_count(0x7fffffff, supported_targets);
3538 	for (i = 0; i <= supported_targets + 1; i++)
3539 		try_cr3_target_count(i, supported_targets);
3540 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3541 }
3542 
3543 /*
3544  * Test a particular address setting in the VMCS
3545  */
3546 static void test_vmcs_addr(const char *name,
3547 			   enum Encoding encoding,
3548 			   u64 align,
3549 			   bool ignored,
3550 			   bool skip_beyond_mapped_ram,
3551 			   u64 addr)
3552 {
3553 	report_prefix_pushf("%s = %lx", name, addr);
3554 	vmcs_write(encoding, addr);
3555 	if (skip_beyond_mapped_ram &&
3556 	    addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3557 	    addr < (1ul << cpuid_maxphyaddr()))
3558 		printf("Skipping physical address beyond mapped RAM\n");
3559 	else if (ignored || (IS_ALIGNED(addr, align) &&
3560 	    addr < (1ul << cpuid_maxphyaddr())))
3561 		test_vmx_valid_controls();
3562 	else
3563 		test_vmx_invalid_controls();
3564 	report_prefix_pop();
3565 }
3566 
3567 /*
3568  * Test interesting values for a VMCS address
3569  */
3570 static void test_vmcs_addr_values(const char *name,
3571 				  enum Encoding encoding,
3572 				  u64 align,
3573 				  bool ignored,
3574 				  bool skip_beyond_mapped_ram,
3575 				  u32 bit_start, u32 bit_end)
3576 {
3577 	unsigned i;
3578 	u64 orig_val = vmcs_read(encoding);
3579 
3580 	for (i = bit_start; i <= bit_end; i++)
3581 		test_vmcs_addr(name, encoding, align, ignored,
3582 			       skip_beyond_mapped_ram, 1ul << i);
3583 
3584 	test_vmcs_addr(name, encoding, align, ignored,
3585 		       skip_beyond_mapped_ram, PAGE_SIZE - 1);
3586 	test_vmcs_addr(name, encoding, align, ignored,
3587 		       skip_beyond_mapped_ram, PAGE_SIZE);
3588 	test_vmcs_addr(name, encoding, align, ignored,
3589 		       skip_beyond_mapped_ram,
3590 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3591 	test_vmcs_addr(name, encoding, align, ignored,
3592 		       skip_beyond_mapped_ram, -1ul);
3593 
3594 	vmcs_write(encoding, orig_val);
3595 }
3596 
3597 /*
3598  * Test a physical address reference in the VMCS, when the corresponding
3599  * feature is enabled and when the corresponding feature is disabled.
3600  */
3601 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3602 				     const char *field_name,
3603 				     const char *control_name, u64 align,
3604 				     bool skip_beyond_mapped_ram,
3605 				     bool control_primary)
3606 {
3607 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3608 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3609 	u64 page_addr;
3610 
3611 	if (control_primary) {
3612 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3613 			return;
3614 	} else {
3615 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3616 			return;
3617 	}
3618 
3619 	page_addr = vmcs_read(field);
3620 
3621 	report_prefix_pushf("%s enabled", control_name);
3622 	if (control_primary) {
3623 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3624 	} else {
3625 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3626 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3627 	}
3628 
3629 	test_vmcs_addr_values(field_name, field, align, false,
3630 			      skip_beyond_mapped_ram, 0, 63);
3631 	report_prefix_pop();
3632 
3633 	report_prefix_pushf("%s disabled", control_name);
3634 	if (control_primary) {
3635 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3636 	} else {
3637 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3638 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3639 	}
3640 
3641 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3642 	report_prefix_pop();
3643 
3644 	vmcs_write(field, page_addr);
3645 	vmcs_write(CPU_EXEC_CTRL0, primary);
3646 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3647 }
3648 
3649 /*
3650  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3651  * each I/O-bitmap address must be 0. Neither address should set any
3652  * bits beyond the processor's physical-address width.
3653  * [Intel SDM]
3654  */
3655 static void test_io_bitmaps(void)
3656 {
3657 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3658 				 "I/O bitmap A", "Use I/O bitmaps",
3659 				 PAGE_SIZE, false, true);
3660 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3661 				 "I/O bitmap B", "Use I/O bitmaps",
3662 				 PAGE_SIZE, false, true);
3663 }
3664 
3665 /*
3666  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3667  * the MSR-bitmap address must be 0. The address should not set any
3668  * bits beyond the processor's physical-address width.
3669  * [Intel SDM]
3670  */
3671 static void test_msr_bitmap(void)
3672 {
3673 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3674 				 "MSR bitmap", "Use MSR bitmaps",
3675 				 PAGE_SIZE, false, true);
3676 }
3677 
3678 /*
3679  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3680  * address must satisfy the following checks:
3681  * - Bits 11:0 of the address must be 0.
3682  * - The address should not set any bits beyond the processor's
3683  *   physical-address width.
3684  * [Intel SDM]
3685  */
3686 static void test_apic_virt_addr(void)
3687 {
3688 	/*
3689 	 * Ensure the processor will never use the virtual-APIC page, since
3690 	 * we will point it to invalid RAM.  Otherwise KVM is puzzled about
3691 	 * what we're trying to achieve and fails vmentry.
3692 	 */
3693 	u32 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
3694 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0 | CPU_CR8_LOAD | CPU_CR8_STORE);
3695 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3696 				 "virtual-APIC address", "Use TPR shadow",
3697 				 PAGE_SIZE, false, true);
3698 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
3699 }
3700 
3701 /*
3702  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3703  * APIC-access address must satisfy the following checks:
3704  *  - Bits 11:0 of the address must be 0.
3705  *  - The address should not set any bits beyond the processor's
3706  *    physical-address width.
3707  * [Intel SDM]
3708  */
3709 static void test_apic_access_addr(void)
3710 {
3711 	void *apic_access_page = alloc_page();
3712 
3713 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3714 
3715 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3716 				 "APIC-access address",
3717 				 "virtualize APIC-accesses", PAGE_SIZE,
3718 				 true, false);
3719 }
3720 
3721 static bool set_bit_pattern(u8 mask, u32 *secondary)
3722 {
3723 	u8 i;
3724 	bool flag = false;
3725 	u32 test_bits[3] = {
3726 		CPU_VIRT_X2APIC,
3727 		CPU_APIC_REG_VIRT,
3728 		CPU_VINTD
3729 	};
3730 
3731         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3732 		if ((mask & (1u << i)) &&
3733 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3734 			*secondary |= test_bits[i];
3735 			flag = true;
3736 		}
3737 	}
3738 
3739 	return (flag);
3740 }
3741 
3742 /*
3743  * If the "use TPR shadow" VM-execution control is 0, the following
3744  * VM-execution controls must also be 0:
3745  * 	- virtualize x2APIC mode
3746  *	- APIC-register virtualization
3747  *	- virtual-interrupt delivery
3748  *    [Intel SDM]
3749  *
3750  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3751  *    "virtualize APIC accesses" VM-execution control must be 0.
3752  *    [Intel SDM]
3753  */
3754 static void test_apic_virtual_ctls(void)
3755 {
3756 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3757 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3758 	u32 primary = saved_primary;
3759 	u32 secondary = saved_secondary;
3760 	bool ctrl = false;
3761 	char str[10] = "disabled";
3762 	u8 i = 0, j;
3763 
3764 	/*
3765 	 * First test
3766 	 */
3767 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3768 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3769 		return;
3770 
3771 	primary |= CPU_SECONDARY;
3772 	primary &= ~CPU_TPR_SHADOW;
3773 	vmcs_write(CPU_EXEC_CTRL0, primary);
3774 
3775 	while (1) {
3776 		for (j = 1; j < 8; j++) {
3777 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3778 			if (primary & CPU_TPR_SHADOW) {
3779 				ctrl = true;
3780 			} else {
3781 				if (! set_bit_pattern(j, &secondary))
3782 					ctrl = true;
3783 				else
3784 					ctrl = false;
3785 			}
3786 
3787 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3788 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3789 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3790 			if (ctrl)
3791 				test_vmx_valid_controls();
3792 			else
3793 				test_vmx_invalid_controls();
3794 			report_prefix_pop();
3795 		}
3796 
3797 		if (i == 1)
3798 			break;
3799 		i++;
3800 
3801 		primary |= CPU_TPR_SHADOW;
3802 		vmcs_write(CPU_EXEC_CTRL0, primary);
3803 		strcpy(str, "enabled");
3804 	}
3805 
3806 	/*
3807 	 * Second test
3808 	 */
3809 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3810 
3811 	primary = saved_primary;
3812 	secondary = saved_secondary;
3813 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3814 		return;
3815 
3816 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3817 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3818 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3819 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3820 	test_vmx_valid_controls();
3821 	report_prefix_pop();
3822 
3823 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3824 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3825 	test_vmx_valid_controls();
3826 	report_prefix_pop();
3827 
3828 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3829 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3830 	test_vmx_invalid_controls();
3831 	report_prefix_pop();
3832 
3833 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3834 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3835 	test_vmx_valid_controls();
3836 	report_prefix_pop();
3837 
3838 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3839 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3840 }
3841 
3842 /*
3843  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3844  * "external-interrupt exiting" VM-execution control must be 1.
3845  * [Intel SDM]
3846  */
3847 static void test_virtual_intr_ctls(void)
3848 {
3849 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3850 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3851 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3852 	u32 primary = saved_primary;
3853 	u32 secondary = saved_secondary;
3854 	u32 pin = saved_pin;
3855 
3856 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3857 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
3858 		return;
3859 
3860 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3861 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
3862 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3863 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
3864 	test_vmx_valid_controls();
3865 	report_prefix_pop();
3866 
3867 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
3868 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3869 	test_vmx_invalid_controls();
3870 	report_prefix_pop();
3871 
3872 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
3873 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
3874 	test_vmx_valid_controls();
3875 	report_prefix_pop();
3876 
3877 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3878 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3879 	test_vmx_invalid_controls();
3880 	report_prefix_pop();
3881 
3882 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3883 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3884 	vmcs_write(PIN_CONTROLS, saved_pin);
3885 }
3886 
3887 static void test_pi_desc_addr(u64 addr, bool ctrl)
3888 {
3889 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
3890 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
3891 	if (ctrl)
3892 		test_vmx_valid_controls();
3893 	else
3894 		test_vmx_invalid_controls();
3895 	report_prefix_pop();
3896 }
3897 
3898 /*
3899  * If the “process posted interrupts†VM-execution control is 1, the
3900  * following must be true:
3901  *
3902  *	- The “virtual-interrupt delivery†VM-execution control is 1.
3903  *	- The “acknowledge interrupt on exit†VM-exit control is 1.
3904  *	- The posted-interrupt notification vector has a value in the
3905  *	- range 0–255 (bits 15:8 are all 0).
3906  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
3907  *	- The posted-interrupt descriptor address does not set any bits
3908  *	  beyond the processor's physical-address width.
3909  * [Intel SDM]
3910  */
3911 static void test_posted_intr(void)
3912 {
3913 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3914 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3915 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3916 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
3917 	u32 primary = saved_primary;
3918 	u32 secondary = saved_secondary;
3919 	u32 pin = saved_pin;
3920 	u32 exit_ctl = exit_ctl_saved;
3921 	u16 vec;
3922 	int i;
3923 
3924 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
3925 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3926 	    (ctrl_exit_rev.clr & EXI_INTA)))
3927 		return;
3928 
3929 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3930 
3931 	/*
3932 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
3933 	 */
3934 	pin |= PIN_POST_INTR;
3935 	vmcs_write(PIN_CONTROLS, pin);
3936 	secondary &= ~CPU_VINTD;
3937 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3938 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
3939 	test_vmx_invalid_controls();
3940 	report_prefix_pop();
3941 
3942 	secondary |= CPU_VINTD;
3943 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3944 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
3945 	test_vmx_invalid_controls();
3946 	report_prefix_pop();
3947 
3948 	exit_ctl &= ~EXI_INTA;
3949 	vmcs_write(EXI_CONTROLS, exit_ctl);
3950 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
3951 	test_vmx_invalid_controls();
3952 	report_prefix_pop();
3953 
3954 	exit_ctl |= EXI_INTA;
3955 	vmcs_write(EXI_CONTROLS, exit_ctl);
3956 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3957 	test_vmx_valid_controls();
3958 	report_prefix_pop();
3959 
3960 	secondary &= ~CPU_VINTD;
3961 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3962 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
3963 	test_vmx_invalid_controls();
3964 	report_prefix_pop();
3965 
3966 	secondary |= CPU_VINTD;
3967 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3968 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3969 	test_vmx_valid_controls();
3970 	report_prefix_pop();
3971 
3972 	/*
3973 	 * Test posted-interrupt notification vector
3974 	 */
3975 	for (i = 0; i < 8; i++) {
3976 		vec = (1ul << i);
3977 		vmcs_write(PINV, vec);
3978 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3979 		test_vmx_valid_controls();
3980 		report_prefix_pop();
3981 	}
3982 	for (i = 8; i < 16; i++) {
3983 		vec = (1ul << i);
3984 		vmcs_write(PINV, vec);
3985 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3986 		test_vmx_invalid_controls();
3987 		report_prefix_pop();
3988 	}
3989 
3990 	vec &= ~(0xff << 8);
3991 	vmcs_write(PINV, vec);
3992 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3993 	test_vmx_valid_controls();
3994 	report_prefix_pop();
3995 
3996 	/*
3997 	 * Test posted-interrupt descriptor addresss
3998 	 */
3999 	for (i = 0; i < 6; i++) {
4000 		test_pi_desc_addr(1ul << i, false);
4001 	}
4002 
4003 	test_pi_desc_addr(0xf0, false);
4004 	test_pi_desc_addr(0xff, false);
4005 	test_pi_desc_addr(0x0f, false);
4006 	test_pi_desc_addr(0x8000, true);
4007 	test_pi_desc_addr(0x00, true);
4008 	test_pi_desc_addr(0xc000, true);
4009 
4010 	test_vmcs_addr_values("process-posted interrupts",
4011 			       POSTED_INTR_DESC_ADDR, 64,
4012 			       false, false, 0, 63);
4013 
4014 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4015 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4016 	vmcs_write(PIN_CONTROLS, saved_pin);
4017 }
4018 
4019 static void test_apic_ctls(void)
4020 {
4021 	test_apic_virt_addr();
4022 	test_apic_access_addr();
4023 	test_apic_virtual_ctls();
4024 	test_virtual_intr_ctls();
4025 	test_posted_intr();
4026 }
4027 
4028 /*
4029  * If the “enable VPID†VM-execution control is 1, the value of the
4030  * of the VPID VM-execution control field must not be 0000H.
4031  * [Intel SDM]
4032  */
4033 static void test_vpid(void)
4034 {
4035 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4036 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4037 	u16 vpid = 0x0000;
4038 	int i;
4039 
4040 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4041 	    (ctrl_cpu_rev[1].clr & CPU_VPID))) {
4042 		printf("Secondary controls and/or VPID not supported\n");
4043 		return;
4044 	}
4045 
4046 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
4047 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
4048 	vmcs_write(VPID, vpid);
4049 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
4050 	test_vmx_valid_controls();
4051 	report_prefix_pop();
4052 
4053 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
4054 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4055 	test_vmx_invalid_controls();
4056 	report_prefix_pop();
4057 
4058 	for (i = 0; i < 16; i++) {
4059 		vpid = (short)1 << i;;
4060 		vmcs_write(VPID, vpid);
4061 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4062 		test_vmx_valid_controls();
4063 		report_prefix_pop();
4064 	}
4065 
4066 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4067 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4068 }
4069 
4070 static void set_vtpr(unsigned vtpr)
4071 {
4072 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
4073 }
4074 
4075 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
4076 {
4077 	bool valid = true;
4078 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4079 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4080 
4081 	if ((primary & CPU_TPR_SHADOW) &&
4082 	    (!(primary & CPU_SECONDARY) ||
4083 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
4084 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
4085 
4086 	set_vtpr(vtpr);
4087 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
4088 	    threshold, (vtpr >> 4) & 0xf);
4089 	if (valid)
4090 		test_vmx_valid_controls();
4091 	else
4092 		test_vmx_invalid_controls();
4093 	report_prefix_pop();
4094 }
4095 
4096 static void test_invalid_event_injection(void)
4097 {
4098 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
4099 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
4100 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
4101 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
4102 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
4103 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4104 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4105 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4106 	u32 cnt;
4107 
4108 	/* Setup */
4109 	report_prefix_push("invalid event injection");
4110 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4111 	vmcs_write(ENT_INST_LEN, 0x00000001);
4112 
4113 	/* The field’s interruption type is not set to a reserved value. */
4114 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4115 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4116 			    "RESERVED interruption type invalid [-]",
4117 			    ent_intr_info);
4118 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4119 	test_vmx_invalid_controls();
4120 	report_prefix_pop();
4121 
4122 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4123 			DE_VECTOR;
4124 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4125 			    "RESERVED interruption type invalid [+]",
4126 			    ent_intr_info);
4127 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4128 	test_vmx_valid_controls();
4129 	report_prefix_pop();
4130 
4131 	/* If the interruption type is other event, the vector is 0. */
4132 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4133 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4134 			    "(OTHER EVENT && vector != 0) invalid [-]",
4135 			    ent_intr_info);
4136 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4137 	test_vmx_invalid_controls();
4138 	report_prefix_pop();
4139 
4140 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4141 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4142 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4143 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4144 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4145 	test_vmx_invalid_controls();
4146 	report_prefix_pop();
4147 
4148 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4149 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4150 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4151 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4152 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4153 	test_vmx_valid_controls();
4154 	report_prefix_pop();
4155 
4156 	/*
4157 	 * If the interruption type
4158 	 * is HW exception, the vector is at most 31.
4159 	 */
4160 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4161 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4162 			    "(HW exception && vector > 31) invalid [-]",
4163 			    ent_intr_info);
4164 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4165 	test_vmx_invalid_controls();
4166 	report_prefix_pop();
4167 
4168 	/*
4169 	 * deliver-error-code is 1 iff either
4170 	 * (a) the "unrestricted guest" VM-execution control is 0
4171 	 * (b) CR0.PE is set.
4172 	 */
4173 
4174 	/* Assert that unrestricted guest is disabled or unsupported */
4175 	assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
4176 	       !(secondary_save & CPU_URG));
4177 
4178 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4179 			GP_VECTOR;
4180 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4181 			    "error code <-> (!URG || prot_mode) [-]",
4182 			    ent_intr_info);
4183 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4184 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4185 	test_vmx_invalid_controls();
4186 	report_prefix_pop();
4187 
4188 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4189 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4190 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4191 			    "error code <-> (!URG || prot_mode) [+]",
4192 			    ent_intr_info);
4193 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4194 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4195 	test_vmx_valid_controls();
4196 	report_prefix_pop();
4197 
4198 	if (enable_unrestricted_guest())
4199 		goto skip_unrestricted_guest;
4200 
4201 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4202 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4203 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4204 			    "error code <-> (!URG || prot_mode) [-]",
4205 			    ent_intr_info);
4206 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4207 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4208 	test_vmx_invalid_controls();
4209 	report_prefix_pop();
4210 
4211 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4212 			GP_VECTOR;
4213 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4214 			    "error code <-> (!URG || prot_mode) [-]",
4215 			    ent_intr_info);
4216 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4217 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4218 	test_vmx_invalid_controls();
4219 	report_prefix_pop();
4220 
4221 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4222 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4223 
4224 skip_unrestricted_guest:
4225 	vmcs_write(GUEST_CR0, guest_cr0_save);
4226 
4227 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4228 	report_prefix_push("error code <-> HW exception");
4229 	for (cnt = 0; cnt < 8; cnt++) {
4230 		u32 exception_type_mask = cnt << 8;
4231 		u32 deliver_error_code_mask =
4232 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4233 			INTR_INFO_DELIVER_CODE_MASK : 0;
4234 
4235 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4236 				exception_type_mask | GP_VECTOR;
4237 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4238 				    ent_intr_info);
4239 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4240 		test_vmx_invalid_controls();
4241 		report_prefix_pop();
4242 	}
4243 	report_prefix_pop();
4244 
4245 	/*
4246 	 * deliver-error-code is 1 iff the the vector
4247 	 * indicates an exception that would normally deliver an error code
4248 	 */
4249 	report_prefix_push("error code <-> vector delivers error code");
4250 	for (cnt = 0; cnt < 32; cnt++) {
4251 		bool has_error_code = false;
4252 		u32 deliver_error_code_mask;
4253 
4254 		switch (cnt) {
4255 		case DF_VECTOR:
4256 		case TS_VECTOR:
4257 		case NP_VECTOR:
4258 		case SS_VECTOR:
4259 		case GP_VECTOR:
4260 		case PF_VECTOR:
4261 		case AC_VECTOR:
4262 			has_error_code = true;
4263 		}
4264 
4265 		/* Negative case */
4266 		deliver_error_code_mask = has_error_code ?
4267 						0 :
4268 						INTR_INFO_DELIVER_CODE_MASK;
4269 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4270 				INTR_TYPE_HARD_EXCEPTION | cnt;
4271 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4272 				    ent_intr_info);
4273 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4274 		test_vmx_invalid_controls();
4275 		report_prefix_pop();
4276 
4277 		/* Positive case */
4278 		deliver_error_code_mask = has_error_code ?
4279 						INTR_INFO_DELIVER_CODE_MASK :
4280 						0;
4281 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4282 				INTR_TYPE_HARD_EXCEPTION | cnt;
4283 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4284 				    ent_intr_info);
4285 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4286 		test_vmx_valid_controls();
4287 		report_prefix_pop();
4288 	}
4289 	report_prefix_pop();
4290 
4291 	/* Reserved bits in the field (30:12) are 0. */
4292 	report_prefix_push("reserved bits clear");
4293 	for (cnt = 12; cnt <= 30; cnt++) {
4294 		ent_intr_info = ent_intr_info_base |
4295 				INTR_INFO_DELIVER_CODE_MASK |
4296 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4297 				(1U << cnt);
4298 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4299 				    ent_intr_info);
4300 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4301 		test_vmx_invalid_controls();
4302 		report_prefix_pop();
4303 	}
4304 	report_prefix_pop();
4305 
4306 	/*
4307 	 * If deliver-error-code is 1
4308 	 * bits 31:16 of the VM-entry exception error-code field are 0.
4309 	 */
4310 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4311 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4312 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4313 			    "VM-entry exception error code[31:16] clear",
4314 			    ent_intr_info);
4315 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4316 	for (cnt = 16; cnt <= 31; cnt++) {
4317 		ent_intr_err = 1U << cnt;
4318 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4319 				    ent_intr_err);
4320 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4321 		test_vmx_invalid_controls();
4322 		report_prefix_pop();
4323 	}
4324 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4325 	report_prefix_pop();
4326 
4327 	/*
4328 	 * If the interruption type is software interrupt, software exception,
4329 	 * or privileged software exception, the VM-entry instruction-length
4330 	 * field is in the range 0–15.
4331 	 */
4332 
4333 	for (cnt = 0; cnt < 3; cnt++) {
4334 		switch (cnt) {
4335 		case 0:
4336 			ent_intr_info = ent_intr_info_base |
4337 					INTR_TYPE_SOFT_INTR;
4338 			break;
4339 		case 1:
4340 			ent_intr_info = ent_intr_info_base |
4341 					INTR_TYPE_SOFT_EXCEPTION;
4342 			break;
4343 		case 2:
4344 			ent_intr_info = ent_intr_info_base |
4345 					INTR_TYPE_PRIV_SW_EXCEPTION;
4346 			break;
4347 		}
4348 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4349 				    "VM-entry instruction-length check",
4350 				    ent_intr_info);
4351 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4352 
4353 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4354 		ent_intr_len = -1;
4355 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4356 				    ent_intr_len);
4357 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4358 		test_vmx_invalid_controls();
4359 		report_prefix_pop();
4360 
4361 		/* Instruction length set to 16 should fail */
4362 		ent_intr_len = 0x00000010;
4363 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4364 				    ent_intr_len);
4365 		vmcs_write(ENT_INST_LEN, 0x00000010);
4366 		test_vmx_invalid_controls();
4367 		report_prefix_pop();
4368 
4369 		report_prefix_pop();
4370 	}
4371 
4372 	/* Cleanup */
4373 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4374 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4375 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4376 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4377 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4378 	vmcs_write(GUEST_CR0, guest_cr0_save);
4379 	report_prefix_pop();
4380 }
4381 
4382 /*
4383  * Test interesting vTPR values for a given TPR threshold.
4384  */
4385 static void test_vtpr_values(unsigned threshold)
4386 {
4387 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4388 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4389 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4390 }
4391 
4392 static void try_tpr_threshold(unsigned threshold)
4393 {
4394 	bool valid = true;
4395 
4396 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4397 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4398 
4399 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4400 	    (secondary & CPU_VINTD)))
4401 		valid = !(threshold >> 4);
4402 
4403 	set_vtpr(-1);
4404 	vmcs_write(TPR_THRESHOLD, threshold);
4405 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4406 	if (valid)
4407 		test_vmx_valid_controls();
4408 	else
4409 		test_vmx_invalid_controls();
4410 	report_prefix_pop();
4411 
4412 	if (valid)
4413 		test_vtpr_values(threshold);
4414 }
4415 
4416 /*
4417  * Test interesting TPR threshold values.
4418  */
4419 static void test_tpr_threshold_values(void)
4420 {
4421 	unsigned i;
4422 
4423 	for (i = 0; i < 0x10; i++)
4424 		try_tpr_threshold(i);
4425 	for (i = 4; i < 32; i++)
4426 		try_tpr_threshold(1u << i);
4427 	try_tpr_threshold(-1u);
4428 	try_tpr_threshold(0x7fffffff);
4429 }
4430 
4431 /*
4432  * This test covers the following two VM entry checks:
4433  *
4434  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4435  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4436  *         31:4 of the TPR threshold VM-execution control field must
4437 	   be 0.
4438  *         [Intel SDM]
4439  *
4440  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4441  *          "virtual-interrupt delivery" VM-execution control is 0
4442  *          and the "virtualize APIC accesses" VM-execution control
4443  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4444  *          control field must not be greater than the value of bits
4445  *          7:4 of VTPR.
4446  *          [Intel SDM]
4447  */
4448 static void test_tpr_threshold(void)
4449 {
4450 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4451 	u64 apic_virt_addr = vmcs_read(APIC_VIRT_ADDR);
4452 	u64 threshold = vmcs_read(TPR_THRESHOLD);
4453 	void *virtual_apic_page;
4454 
4455 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4456 		return;
4457 
4458 	virtual_apic_page = alloc_page();
4459 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4460 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4461 
4462 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4463 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4464 	test_tpr_threshold_values();
4465 	report_prefix_pop();
4466 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4467 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4468 	test_tpr_threshold_values();
4469 	report_prefix_pop();
4470 
4471 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4472 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES))))
4473 		goto out;
4474 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4475 
4476 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4477 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4478 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4479 		test_tpr_threshold_values();
4480 		report_prefix_pop();
4481 
4482 		vmcs_write(CPU_EXEC_CTRL0,
4483 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4484 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4485 		test_tpr_threshold_values();
4486 		report_prefix_pop();
4487 	}
4488 
4489 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4490 		vmcs_write(CPU_EXEC_CTRL0,
4491 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4492 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4493 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4494 		test_tpr_threshold_values();
4495 		report_prefix_pop();
4496 
4497 		vmcs_write(CPU_EXEC_CTRL0,
4498 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4499 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4500 		test_tpr_threshold_values();
4501 		report_prefix_pop();
4502 	}
4503 
4504 	if ((ctrl_cpu_rev[1].clr &
4505 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4506 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4507 		vmcs_write(CPU_EXEC_CTRL0,
4508 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4509 		vmcs_write(CPU_EXEC_CTRL1,
4510 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4511 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4512 		test_tpr_threshold_values();
4513 		report_prefix_pop();
4514 
4515 		vmcs_write(CPU_EXEC_CTRL0,
4516 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4517 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4518 		test_tpr_threshold_values();
4519 		report_prefix_pop();
4520 	}
4521 
4522 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4523 out:
4524 	vmcs_write(TPR_THRESHOLD, threshold);
4525 	vmcs_write(APIC_VIRT_ADDR, apic_virt_addr);
4526 	vmcs_write(CPU_EXEC_CTRL0, primary);
4527 }
4528 
4529 /*
4530  * This test verifies the following two vmentry checks:
4531  *
4532  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4533  *  VM-execution control must be 0.
4534  *  [Intel SDM]
4535  *
4536  *  If the “virtual NMIs” VM-execution control is 0, the “NMI-window
4537  *  exiting” VM-execution control must be 0.
4538  *  [Intel SDM]
4539  */
4540 static void test_nmi_ctrls(void)
4541 {
4542 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4543 
4544 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4545 	    (PIN_NMI | PIN_VIRT_NMI)) {
4546 		printf("NMI exiting and Virtual NMIs are not supported !\n");
4547 		return;
4548 	}
4549 
4550 	/* Save the controls so that we can restore them after our tests */
4551 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4552 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4553 
4554 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4555 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4556 
4557 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4558 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4559 	test_vmx_valid_controls();
4560 	report_prefix_pop();
4561 
4562 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4563 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4564 	test_vmx_invalid_controls();
4565 	report_prefix_pop();
4566 
4567 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4568 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4569 	test_vmx_valid_controls();
4570 	report_prefix_pop();
4571 
4572 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4573 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4574 	test_vmx_valid_controls();
4575 	report_prefix_pop();
4576 
4577 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4578 		report_info("NMI-window exiting is not supported, skipping...");
4579 		goto done;
4580 	}
4581 
4582 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4583 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4584 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4585 	test_vmx_invalid_controls();
4586 	report_prefix_pop();
4587 
4588 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4589 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4590 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4591 	test_vmx_valid_controls();
4592 	report_prefix_pop();
4593 
4594 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4595 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4596 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4597 	test_vmx_valid_controls();
4598 	report_prefix_pop();
4599 
4600 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4601 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4602 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4603 	test_vmx_valid_controls();
4604 	report_prefix_pop();
4605 
4606 	/* Restore the controls to their original values */
4607 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4608 done:
4609 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4610 }
4611 
4612 static void test_eptp_ad_bit(u64 eptp, bool ctrl)
4613 {
4614 	vmcs_write(EPTP, eptp);
4615 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4616 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4617 	if (ctrl)
4618 		test_vmx_valid_controls();
4619 	else
4620 		test_vmx_invalid_controls();
4621 	report_prefix_pop();
4622 
4623 }
4624 
4625 /*
4626  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4627  *    control field must satisfy the following checks:
4628  *
4629  *     - The EPT memory type (bits 2:0) must be a value supported by the
4630  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4631  *     - Bits 5:3 (1 less than the EPT page-walk length) must be 3,
4632  *	 indicating an EPT page-walk length of 4.
4633  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4634  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4635  *	 indicating that the processor does not support accessed and dirty
4636  *	 dirty flags for EPT.
4637  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4638  *	 physical-address width) must all be 0.
4639  *
4640  * 2. If the "unrestricted guest" VM-execution control is 1, the
4641  *    "enable EPT" VM-execution control must also be 1.
4642  */
4643 static void test_ept_eptp(void)
4644 {
4645 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4646 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4647 	u64 eptp_saved = vmcs_read(EPTP);
4648 	u32 primary = primary_saved;
4649 	u32 secondary = secondary_saved;
4650 	u64 msr, eptp = eptp_saved;
4651 	bool un_cache = false;
4652 	bool wr_bk = false;
4653 	bool ctrl;
4654 	u32 i, maxphysaddr;
4655 	u64 j, resv_bits_mask = 0;
4656 
4657 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4658 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4659 		printf("\"CPU secondary\" and/or \"enable EPT\" execution controls are not supported !\n");
4660 		return;
4661 	}
4662 
4663 	/*
4664 	 * Memory type (bits 2:0)
4665 	 */
4666 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
4667 	if (msr & EPT_CAP_UC)
4668 		un_cache = true;
4669 	if (msr & EPT_CAP_WB)
4670 		wr_bk = true;
4671 
4672 	primary |= CPU_SECONDARY;
4673 	vmcs_write(CPU_EXEC_CTRL0, primary);
4674 	secondary |= CPU_EPT;
4675 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4676 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4677 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4678 	vmcs_write(EPTP, eptp);
4679 
4680 	for (i = 0; i < 8; i++) {
4681 		if (i == 0) {
4682 			if (un_cache) {
4683 				report_info("EPT paging structure memory-type is Un-cacheable\n");
4684 				ctrl = true;
4685 			} else {
4686 				ctrl = false;
4687 			}
4688 		} else if (i == 6) {
4689 			if (wr_bk) {
4690 				report_info("EPT paging structure memory-type is Write-back\n");
4691 				ctrl = true;
4692 			} else {
4693 				ctrl = false;
4694 			}
4695 		} else {
4696 			ctrl = false;
4697 		}
4698 
4699 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4700 		vmcs_write(EPTP, eptp);
4701 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4702 		    eptp & EPT_MEM_TYPE_MASK);
4703 		if (ctrl)
4704 			test_vmx_valid_controls();
4705 		else
4706 			test_vmx_invalid_controls();
4707 		report_prefix_pop();
4708 	}
4709 
4710 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4711 
4712 	/*
4713 	 * Page walk length (bits 5:3)
4714 	 */
4715 	for (i = 0; i < 8; i++) {
4716 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4717 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4718 		if (i == 3)
4719 			ctrl = true;
4720 		else
4721 			ctrl = false;
4722 
4723 		vmcs_write(EPTP, eptp);
4724 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4725 		    eptp & EPTP_PG_WALK_LEN_MASK);
4726 		if (ctrl)
4727 			test_vmx_valid_controls();
4728 		else
4729 			test_vmx_invalid_controls();
4730 		report_prefix_pop();
4731 	}
4732 
4733 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4734 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4735 
4736 	/*
4737 	 * Accessed and dirty flag (bit 6)
4738 	 */
4739 	if (msr & EPT_CAP_AD_FLAG) {
4740 		report_info("Processor supports accessed and dirty flag");
4741 		eptp &= ~EPTP_AD_FLAG;
4742 		test_eptp_ad_bit(eptp, true);
4743 
4744 		eptp |= EPTP_AD_FLAG;
4745 		test_eptp_ad_bit(eptp, true);
4746 	} else {
4747 		report_info("Processor does not supports accessed and dirty flag");
4748 		eptp &= ~EPTP_AD_FLAG;
4749 		test_eptp_ad_bit(eptp, true);
4750 
4751 		eptp |= EPTP_AD_FLAG;
4752 		test_eptp_ad_bit(eptp, false);
4753 	}
4754 
4755 	/*
4756 	 * Reserved bits [11:7] and [63:N]
4757 	 */
4758 	for (i = 0; i < 32; i++) {
4759 		eptp = (eptp &
4760 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4761 		    (i << EPTP_RESERV_BITS_SHIFT);
4762 		vmcs_write(EPTP, eptp);
4763 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4764 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4765 		    EPTP_RESERV_BITS_MASK);
4766 		if (i == 0)
4767 			test_vmx_valid_controls();
4768 		else
4769 			test_vmx_invalid_controls();
4770 		report_prefix_pop();
4771 	}
4772 
4773 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4774 
4775 	maxphysaddr = cpuid_maxphyaddr();
4776 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4777 		resv_bits_mask |= 1ul << i;
4778 	}
4779 
4780 	for (j = maxphysaddr - 1; j <= 63; j++) {
4781 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4782 		    (j < maxphysaddr ? 0 : 1ul << j);
4783 		vmcs_write(EPTP, eptp);
4784 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4785 		    (eptp >> maxphysaddr) & resv_bits_mask);
4786 		if (j < maxphysaddr)
4787 			test_vmx_valid_controls();
4788 		else
4789 			test_vmx_invalid_controls();
4790 		report_prefix_pop();
4791 	}
4792 
4793 	secondary &= ~(CPU_EPT | CPU_URG);
4794 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4795 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4796 	test_vmx_valid_controls();
4797 	report_prefix_pop();
4798 
4799 	if (!(ctrl_cpu_rev[1].clr & CPU_URG))
4800 		goto skip_unrestricted_guest;
4801 
4802 	secondary |= CPU_URG;
4803 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4804 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4805 	test_vmx_invalid_controls();
4806 	report_prefix_pop();
4807 
4808 	secondary |= CPU_EPT;
4809 	setup_dummy_ept();
4810 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4811 	test_vmx_valid_controls();
4812 	report_prefix_pop();
4813 
4814 skip_unrestricted_guest:
4815 	secondary &= ~CPU_URG;
4816 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4817 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4818 	test_vmx_valid_controls();
4819 	report_prefix_pop();
4820 
4821 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4822 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4823 	vmcs_write(EPTP, eptp_saved);
4824 }
4825 
4826 /*
4827  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4828  * VM-execution control must also be 1. In addition, the PML address
4829  * must satisfy the following checks:
4830  *
4831  *    * Bits 11:0 of the address must be 0.
4832  *    * The address should not set any bits beyond the processor's
4833  *	physical-address width.
4834  *
4835  *  [Intel SDM]
4836  */
4837 static void test_pml(void)
4838 {
4839 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4840 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4841 	u32 primary = primary_saved;
4842 	u32 secondary = secondary_saved;
4843 
4844 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4845 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
4846 		printf("\"Secondary execution\" control or \"enable EPT\" control or \"enable PML\" control is not supported !\n");
4847 		return;
4848 	}
4849 
4850 	primary |= CPU_SECONDARY;
4851 	vmcs_write(CPU_EXEC_CTRL0, primary);
4852 	secondary &= ~(CPU_PML | CPU_EPT);
4853 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4854 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
4855 	test_vmx_valid_controls();
4856 	report_prefix_pop();
4857 
4858 	secondary |= CPU_PML;
4859 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4860 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
4861 	test_vmx_invalid_controls();
4862 	report_prefix_pop();
4863 
4864 	secondary |= CPU_EPT;
4865 	setup_dummy_ept();
4866 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
4867 	test_vmx_valid_controls();
4868 	report_prefix_pop();
4869 
4870 	secondary &= ~CPU_PML;
4871 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4872 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
4873 	test_vmx_valid_controls();
4874 	report_prefix_pop();
4875 
4876 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
4877 				 PAGE_SIZE, false, false);
4878 
4879 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4880 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4881 }
4882 
4883  /*
4884  * If the "activate VMX-preemption timer" VM-execution control is 0, the
4885  * the "save VMX-preemption timer value" VM-exit control must also be 0.
4886  *
4887  *  [Intel SDM]
4888  */
4889 static void test_vmx_preemption_timer(void)
4890 {
4891 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4892 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
4893 	u32 pin = saved_pin;
4894 	u32 exit = saved_exit;
4895 
4896 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
4897 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
4898 		printf("\"Save-VMX-preemption-timer\" control and/or \"Enable-VMX-preemption-timer\" control is not supported\n");
4899 		return;
4900 	}
4901 
4902 	pin |= PIN_PREEMPT;
4903 	vmcs_write(PIN_CONTROLS, pin);
4904 	exit &= ~EXI_SAVE_PREEMPT;
4905 	vmcs_write(EXI_CONTROLS, exit);
4906 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
4907 	test_vmx_valid_controls();
4908 	report_prefix_pop();
4909 
4910 	exit |= EXI_SAVE_PREEMPT;
4911 	vmcs_write(EXI_CONTROLS, exit);
4912 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
4913 	test_vmx_valid_controls();
4914 	report_prefix_pop();
4915 
4916 	pin &= ~PIN_PREEMPT;
4917 	vmcs_write(PIN_CONTROLS, pin);
4918 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
4919 	test_vmx_invalid_controls();
4920 	report_prefix_pop();
4921 
4922 	exit &= ~EXI_SAVE_PREEMPT;
4923 	vmcs_write(EXI_CONTROLS, exit);
4924 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
4925 	test_vmx_valid_controls();
4926 	report_prefix_pop();
4927 
4928 	vmcs_write(PIN_CONTROLS, saved_pin);
4929 	vmcs_write(EXI_CONTROLS, saved_exit);
4930 }
4931 
4932 /*
4933  * Tests for VM-execution control fields
4934  */
4935 static void test_vm_execution_ctls(void)
4936 {
4937 	test_pin_based_ctls();
4938 	test_primary_processor_based_ctls();
4939 	test_secondary_processor_based_ctls();
4940 	test_cr3_targets();
4941 	test_io_bitmaps();
4942 	test_msr_bitmap();
4943 	test_apic_ctls();
4944 	test_tpr_threshold();
4945 	test_nmi_ctrls();
4946 	test_pml();
4947 	test_vpid();
4948 	test_ept_eptp();
4949 	test_vmx_preemption_timer();
4950 }
4951 
4952  /*
4953   * The following checks are performed for the VM-entry MSR-load address if
4954   * the VM-entry MSR-load count field is non-zero:
4955   *
4956   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
4957   *      The address should not set any bits beyond the processor’s
4958   *      physical-address width.
4959   *
4960   *    - The address of the last byte in the VM-entry MSR-load area
4961   *      should not set any bits beyond the processor’s physical-address
4962   *      width. The address of this last byte is VM-entry MSR-load address
4963   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
4964   *      uses more bits than the processor’s physical-address width.)
4965   *
4966   *
4967   *  [Intel SDM]
4968   */
4969 static void test_entry_msr_load(void)
4970 {
4971 	entry_msr_load = alloc_page();
4972 	u64 tmp;
4973 	u32 entry_msr_ld_cnt = 1;
4974 	int i;
4975 	u32 addr_len = 64;
4976 
4977 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
4978 
4979 	/* Check first 4 bits of VM-entry MSR-load address */
4980 	for (i = 0; i < 4; i++) {
4981 		tmp = (u64)entry_msr_load | 1ull << i;
4982 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
4983 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
4984 				    tmp & 0xf);
4985 		test_vmx_invalid_controls();
4986 		report_prefix_pop();
4987 	}
4988 
4989 	if (basic.val & (1ul << 48))
4990 		addr_len = 32;
4991 
4992 	test_vmcs_addr_values("VM-entry-MSR-load address",
4993 				ENTER_MSR_LD_ADDR, 16, false, false,
4994 				4, addr_len - 1);
4995 
4996 	/*
4997 	 * Check last byte of VM-entry MSR-load address
4998 	 */
4999 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
5000 
5001 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5002 							i < 64; i++) {
5003 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
5004 			1ul << i;
5005 		vmcs_write(ENTER_MSR_LD_ADDR,
5006 			   tmp - (entry_msr_ld_cnt * 16 - 1));
5007 		test_vmx_invalid_controls();
5008 	}
5009 
5010 	vmcs_write(ENT_MSR_LD_CNT, 2);
5011 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5012 	test_vmx_invalid_controls();
5013 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5014 	test_vmx_valid_controls();
5015 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5016 	test_vmx_valid_controls();
5017 }
5018 
5019 static void guest_state_test_main(void)
5020 {
5021 	while (1) {
5022 		if (vmx_get_test_stage() != 2)
5023 			vmcall();
5024 		else
5025 			break;
5026 	}
5027 
5028 	asm volatile("fnop");
5029 }
5030 
5031 static void advance_guest_state_test(void)
5032 {
5033 	u32 reason = vmcs_read(EXI_REASON);
5034 	if (! (reason & 0x80000000)) {
5035 		u64 guest_rip = vmcs_read(GUEST_RIP);
5036 		u32 insn_len = vmcs_read(EXI_INST_LEN);
5037 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
5038 	}
5039 }
5040 
5041 static void report_guest_state_test(const char *test, u32 xreason,
5042 				    u64 field, const char * field_name)
5043 {
5044 	u32 reason = vmcs_read(EXI_REASON);
5045 
5046 	report("%s, %s %lx", reason == xreason, test, field_name, field);
5047 	advance_guest_state_test();
5048 }
5049 
5050 /*
5051  * Tests for VM-entry control fields
5052  */
5053 static void test_vm_entry_ctls(void)
5054 {
5055 	test_invalid_event_injection();
5056 	test_entry_msr_load();
5057 }
5058 
5059 /*
5060  * The following checks are performed for the VM-exit MSR-store address if
5061  * the VM-exit MSR-store count field is non-zero:
5062  *
5063  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
5064  *      The address should not set any bits beyond the processor’s
5065  *      physical-address width.
5066  *
5067  *    - The address of the last byte in the VM-exit MSR-store area
5068  *      should not set any bits beyond the processor’s physical-address
5069  *      width. The address of this last byte is VM-exit MSR-store address
5070  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5071  *      uses more bits than the processor’s physical-address width.)
5072  *
5073  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
5074  * in the range 63:32.
5075  *
5076  *  [Intel SDM]
5077  */
5078 static void test_exit_msr_store(void)
5079 {
5080 	exit_msr_store = alloc_page();
5081 	u64 tmp;
5082 	u32 exit_msr_st_cnt = 1;
5083 	int i;
5084 	u32 addr_len = 64;
5085 
5086 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
5087 
5088 	/* Check first 4 bits of VM-exit MSR-store address */
5089 	for (i = 0; i < 4; i++) {
5090 		tmp = (u64)exit_msr_store | 1ull << i;
5091 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
5092 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
5093 				    tmp & 0xf);
5094 		test_vmx_invalid_controls();
5095 		report_prefix_pop();
5096 	}
5097 
5098 	if (basic.val & (1ul << 48))
5099 		addr_len = 32;
5100 
5101 	test_vmcs_addr_values("VM-exit-MSR-store address",
5102 				EXIT_MSR_ST_ADDR, 16, false, false,
5103 				4, addr_len - 1);
5104 
5105 	/*
5106 	 * Check last byte of VM-exit MSR-store address
5107 	 */
5108 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
5109 
5110 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5111 							i < 64; i++) {
5112 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
5113 			1ul << i;
5114 		vmcs_write(EXIT_MSR_ST_ADDR,
5115 			   tmp - (exit_msr_st_cnt * 16 - 1));
5116 		test_vmx_invalid_controls();
5117 	}
5118 
5119 	vmcs_write(EXI_MSR_ST_CNT, 2);
5120 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5121 	test_vmx_invalid_controls();
5122 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5123 	test_vmx_valid_controls();
5124 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5125 	test_vmx_valid_controls();
5126 }
5127 
5128 /*
5129  * Tests for VM-exit controls
5130  */
5131 static void test_vm_exit_ctls(void)
5132 {
5133 	test_exit_msr_store();
5134 }
5135 
5136 /*
5137  * Check that the virtual CPU checks all of the VMX controls as
5138  * documented in the Intel SDM.
5139  */
5140 static void vmx_controls_test(void)
5141 {
5142 	/*
5143 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
5144 	 * fail due to invalid guest state, should we make it that
5145 	 * far.
5146 	 */
5147 	vmcs_write(GUEST_RFLAGS, 0);
5148 
5149 	test_vm_execution_ctls();
5150 	test_vm_exit_ctls();
5151 	test_vm_entry_ctls();
5152 }
5153 
5154 struct apic_reg_virt_config {
5155 	bool apic_register_virtualization;
5156 	bool use_tpr_shadow;
5157 	bool virtualize_apic_accesses;
5158 	bool virtualize_x2apic_mode;
5159 	bool activate_secondary_controls;
5160 };
5161 
5162 struct apic_reg_test {
5163 	const char *name;
5164 	struct apic_reg_virt_config apic_reg_virt_config;
5165 };
5166 
5167 struct apic_reg_virt_expectation {
5168 	enum Reason rd_exit_reason;
5169 	enum Reason wr_exit_reason;
5170 	u32 val;
5171 	u32 (*virt_fn)(u32);
5172 
5173 	/*
5174 	 * If false, accessing the APIC access address from L2 is treated as a
5175 	 * normal memory operation, rather than triggering virtualization.
5176 	 */
5177 	bool virtualize_apic_accesses;
5178 };
5179 
5180 static u32 apic_virt_identity(u32 val)
5181 {
5182 	return val;
5183 }
5184 
5185 static u32 apic_virt_nibble1(u32 val)
5186 {
5187 	return val & 0xf0;
5188 }
5189 
5190 static u32 apic_virt_byte3(u32 val)
5191 {
5192 	return val & (0xff << 24);
5193 }
5194 
5195 static bool apic_reg_virt_exit_expectation(
5196 	u32 reg, struct apic_reg_virt_config *config,
5197 	struct apic_reg_virt_expectation *expectation)
5198 {
5199 	/* Good configs, where some L2 APIC accesses are virtualized. */
5200 	bool virtualize_apic_accesses_only =
5201 		config->virtualize_apic_accesses &&
5202 		!config->use_tpr_shadow &&
5203 		!config->apic_register_virtualization &&
5204 		!config->virtualize_x2apic_mode &&
5205 		config->activate_secondary_controls;
5206 	bool virtualize_apic_accesses_and_use_tpr_shadow =
5207 		config->virtualize_apic_accesses &&
5208 		config->use_tpr_shadow &&
5209 		!config->apic_register_virtualization &&
5210 		!config->virtualize_x2apic_mode &&
5211 		config->activate_secondary_controls;
5212 	bool apic_register_virtualization =
5213 		config->virtualize_apic_accesses &&
5214 		config->use_tpr_shadow &&
5215 		config->apic_register_virtualization &&
5216 		!config->virtualize_x2apic_mode &&
5217 		config->activate_secondary_controls;
5218 
5219 	expectation->val = MAGIC_VAL_1;
5220 	expectation->virt_fn = apic_virt_identity;
5221 	expectation->virtualize_apic_accesses =
5222 		config->virtualize_apic_accesses &&
5223 		config->activate_secondary_controls;
5224 	if (virtualize_apic_accesses_only) {
5225 		expectation->rd_exit_reason = VMX_APIC_ACCESS;
5226 		expectation->wr_exit_reason = VMX_APIC_ACCESS;
5227 	} else if (virtualize_apic_accesses_and_use_tpr_shadow) {
5228 		switch (reg) {
5229 		case APIC_TASKPRI:
5230 			expectation->rd_exit_reason = VMX_VMCALL;
5231 			expectation->wr_exit_reason = VMX_VMCALL;
5232 			expectation->virt_fn = apic_virt_nibble1;
5233 			break;
5234 		default:
5235 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5236 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5237 		}
5238 	} else if (apic_register_virtualization) {
5239 		expectation->rd_exit_reason = VMX_VMCALL;
5240 
5241 		switch (reg) {
5242 		case APIC_ID:
5243 		case APIC_EOI:
5244 		case APIC_LDR:
5245 		case APIC_DFR:
5246 		case APIC_SPIV:
5247 		case APIC_ESR:
5248 		case APIC_ICR:
5249 		case APIC_LVTT:
5250 		case APIC_LVTTHMR:
5251 		case APIC_LVTPC:
5252 		case APIC_LVT0:
5253 		case APIC_LVT1:
5254 		case APIC_LVTERR:
5255 		case APIC_TMICT:
5256 		case APIC_TDCR:
5257 			expectation->wr_exit_reason = VMX_APIC_WRITE;
5258 			break;
5259 		case APIC_LVR:
5260 		case APIC_ISR ... APIC_ISR + 0x70:
5261 		case APIC_TMR ... APIC_TMR + 0x70:
5262 		case APIC_IRR ... APIC_IRR + 0x70:
5263 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5264 			break;
5265 		case APIC_TASKPRI:
5266 			expectation->wr_exit_reason = VMX_VMCALL;
5267 			expectation->virt_fn = apic_virt_nibble1;
5268 			break;
5269 		case APIC_ICR2:
5270 			expectation->wr_exit_reason = VMX_VMCALL;
5271 			expectation->virt_fn = apic_virt_byte3;
5272 			break;
5273 		default:
5274 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5275 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5276 		}
5277 	} else if (!expectation->virtualize_apic_accesses) {
5278 		/*
5279 		 * No APIC registers are directly virtualized. This includes
5280 		 * VTPR, which can be virtualized through MOV to/from CR8 via
5281 		 * the use TPR shadow control, but not through directly
5282 		 * accessing VTPR.
5283 		 */
5284 		expectation->rd_exit_reason = VMX_VMCALL;
5285 		expectation->wr_exit_reason = VMX_VMCALL;
5286 	} else {
5287 		printf("Cannot parse APIC register virtualization config:\n"
5288 		       "\tvirtualize_apic_accesses: %d\n"
5289 		       "\tuse_tpr_shadow: %d\n"
5290 		       "\tapic_register_virtualization: %d\n"
5291 		       "\tvirtualize_x2apic_mode: %d\n"
5292 		       "\tactivate_secondary_controls: %d\n",
5293 		       config->virtualize_apic_accesses,
5294 		       config->use_tpr_shadow,
5295 		       config->apic_register_virtualization,
5296 		       config->virtualize_x2apic_mode,
5297 		       config->activate_secondary_controls);
5298 
5299 		return false;
5300 	}
5301 
5302 	return true;
5303 }
5304 
5305 struct apic_reg_test apic_reg_tests[] = {
5306 	/* Good configs, where some L2 APIC accesses are virtualized. */
5307 	{
5308 		.name = "Virtualize APIC accesses",
5309 		.apic_reg_virt_config = {
5310 			.virtualize_apic_accesses = true,
5311 			.use_tpr_shadow = false,
5312 			.apic_register_virtualization = false,
5313 			.virtualize_x2apic_mode = false,
5314 			.activate_secondary_controls = true,
5315 		},
5316 	},
5317 	{
5318 		.name = "Virtualize APIC accesses + Use TPR shadow",
5319 		.apic_reg_virt_config = {
5320 			.virtualize_apic_accesses = true,
5321 			.use_tpr_shadow = true,
5322 			.apic_register_virtualization = false,
5323 			.virtualize_x2apic_mode = false,
5324 			.activate_secondary_controls = true,
5325 		},
5326 	},
5327 	{
5328 		.name = "APIC-register virtualization",
5329 		.apic_reg_virt_config = {
5330 			.virtualize_apic_accesses = true,
5331 			.use_tpr_shadow = true,
5332 			.apic_register_virtualization = true,
5333 			.virtualize_x2apic_mode = false,
5334 			.activate_secondary_controls = true,
5335 		},
5336 	},
5337 
5338 	/*
5339 	 * Test that the secondary processor-based VM-execution controls are
5340 	 * correctly ignored when "activate secondary controls" is disabled.
5341 	 */
5342 	{
5343 		.name = "Activate secondary controls off",
5344 		.apic_reg_virt_config = {
5345 			.virtualize_apic_accesses = true,
5346 			.use_tpr_shadow = false,
5347 			.apic_register_virtualization = true,
5348 			.virtualize_x2apic_mode = true,
5349 			.activate_secondary_controls = false,
5350 		},
5351 	},
5352 	{
5353 		.name = "Activate secondary controls off + Use TPR shadow",
5354 		.apic_reg_virt_config = {
5355 			.virtualize_apic_accesses = true,
5356 			.use_tpr_shadow = true,
5357 			.apic_register_virtualization = true,
5358 			.virtualize_x2apic_mode = true,
5359 			.activate_secondary_controls = false,
5360 		},
5361 	},
5362 
5363 	/*
5364 	 * Test that the APIC access address is treated like an arbitrary memory
5365 	 * address when "virtualize APIC accesses" is disabled.
5366 	 */
5367 	{
5368 		.name = "Virtualize APIC accesses off + Use TPR shadow",
5369 		.apic_reg_virt_config = {
5370 			.virtualize_apic_accesses = false,
5371 			.use_tpr_shadow = true,
5372 			.apic_register_virtualization = true,
5373 			.virtualize_x2apic_mode = true,
5374 			.activate_secondary_controls = true,
5375 		},
5376 	},
5377 
5378 	/*
5379 	 * Test that VM entry fails due to invalid controls when
5380 	 * "APIC-register virtualization" is enabled while "use TPR shadow" is
5381 	 * disabled.
5382 	 */
5383 	{
5384 		.name = "APIC-register virtualization + Use TPR shadow off",
5385 		.apic_reg_virt_config = {
5386 			.virtualize_apic_accesses = true,
5387 			.use_tpr_shadow = false,
5388 			.apic_register_virtualization = true,
5389 			.virtualize_x2apic_mode = false,
5390 			.activate_secondary_controls = true,
5391 		},
5392 	},
5393 
5394 	/*
5395 	 * Test that VM entry fails due to invalid controls when
5396 	 * "Virtualize x2APIC mode" is enabled while "use TPR shadow" is
5397 	 * disabled.
5398 	 */
5399 	{
5400 		.name = "Virtualize x2APIC mode + Use TPR shadow off",
5401 		.apic_reg_virt_config = {
5402 			.virtualize_apic_accesses = false,
5403 			.use_tpr_shadow = false,
5404 			.apic_register_virtualization = false,
5405 			.virtualize_x2apic_mode = true,
5406 			.activate_secondary_controls = true,
5407 		},
5408 	},
5409 	{
5410 		.name = "Virtualize x2APIC mode + Use TPR shadow off v2",
5411 		.apic_reg_virt_config = {
5412 			.virtualize_apic_accesses = false,
5413 			.use_tpr_shadow = false,
5414 			.apic_register_virtualization = true,
5415 			.virtualize_x2apic_mode = true,
5416 			.activate_secondary_controls = true,
5417 		},
5418 	},
5419 
5420 	/*
5421 	 * Test that VM entry fails due to invalid controls when
5422 	 * "virtualize x2APIC mode" is enabled while "virtualize APIC accesses"
5423 	 * is enabled.
5424 	 */
5425 	{
5426 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses",
5427 		.apic_reg_virt_config = {
5428 			.virtualize_apic_accesses = true,
5429 			.use_tpr_shadow = true,
5430 			.apic_register_virtualization = false,
5431 			.virtualize_x2apic_mode = true,
5432 			.activate_secondary_controls = true,
5433 		},
5434 	},
5435 	{
5436 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses v2",
5437 		.apic_reg_virt_config = {
5438 			.virtualize_apic_accesses = true,
5439 			.use_tpr_shadow = true,
5440 			.apic_register_virtualization = true,
5441 			.virtualize_x2apic_mode = true,
5442 			.activate_secondary_controls = true,
5443 		},
5444 	},
5445 };
5446 
5447 enum Apic_op {
5448 	APIC_OP_XAPIC_RD,
5449 	APIC_OP_XAPIC_WR,
5450 	TERMINATE,
5451 };
5452 
5453 static u32 vmx_xapic_read(u32 *apic_access_address, u32 reg)
5454 {
5455 	return *(volatile u32 *)((uintptr_t)apic_access_address + reg);
5456 }
5457 
5458 static void vmx_xapic_write(u32 *apic_access_address, u32 reg, u32 val)
5459 {
5460 	*(volatile u32 *)((uintptr_t)apic_access_address + reg) = val;
5461 }
5462 
5463 struct apic_reg_virt_guest_args {
5464 	enum Apic_op op;
5465 	u32 *apic_access_address;
5466 	u32 reg;
5467 	u32 val;
5468 	bool check_rd;
5469 	u32 (*virt_fn)(u32);
5470 } apic_reg_virt_guest_args;
5471 
5472 static void apic_reg_virt_guest(void)
5473 {
5474 	volatile struct apic_reg_virt_guest_args *args =
5475 		&apic_reg_virt_guest_args;
5476 
5477 	for (;;) {
5478 		enum Apic_op op = args->op;
5479 		u32 *apic_access_address = args->apic_access_address;
5480 		u32 reg = args->reg;
5481 		u32 val = args->val;
5482 		bool check_rd = args->check_rd;
5483 		u32 (*virt_fn)(u32) = args->virt_fn;
5484 
5485 		if (op == TERMINATE)
5486 			break;
5487 
5488 		if (op == APIC_OP_XAPIC_RD) {
5489 			u32 ret = vmx_xapic_read(apic_access_address, reg);
5490 
5491 			if (check_rd) {
5492 				u32 want = virt_fn(val);
5493 				u32 got = virt_fn(ret);
5494 
5495 				report("read 0x%x, expected 0x%x.",
5496 				       got == want, got, want);
5497 			}
5498 		} else if (op == APIC_OP_XAPIC_WR) {
5499 			vmx_xapic_write(apic_access_address, reg, val);
5500 		}
5501 
5502 		/*
5503 		 * The L1 should always execute a vmcall after it's done testing
5504 		 * an individual APIC operation. This helps to validate that the
5505 		 * L1 and L2 are in sync with each other, as expected.
5506 		 */
5507 		vmcall();
5508 	}
5509 }
5510 
5511 static void test_xapic_rd(
5512 	u32 reg, struct apic_reg_virt_expectation *expectation,
5513 	u32 *apic_access_address, u32 *virtual_apic_page)
5514 {
5515 	u32 val = expectation->val;
5516 	u32 exit_reason_want = expectation->rd_exit_reason;
5517 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5518 
5519 	report_prefix_pushf("xapic - reading 0x%03x", reg);
5520 
5521 	/* Configure guest to do an xapic read */
5522 	args->op = APIC_OP_XAPIC_RD;
5523 	args->apic_access_address = apic_access_address;
5524 	args->reg = reg;
5525 	args->val = val;
5526 	args->check_rd = exit_reason_want == VMX_VMCALL;
5527 	args->virt_fn = expectation->virt_fn;
5528 
5529 	/* Setup virtual APIC page */
5530 	if (!expectation->virtualize_apic_accesses) {
5531 		apic_access_address[apic_reg_index(reg)] = val;
5532 		virtual_apic_page[apic_reg_index(reg)] = 0;
5533 	} else if (exit_reason_want == VMX_VMCALL) {
5534 		apic_access_address[apic_reg_index(reg)] = 0;
5535 		virtual_apic_page[apic_reg_index(reg)] = val;
5536 	}
5537 
5538 	/* Enter guest */
5539 	enter_guest();
5540 
5541 	/*
5542 	 * Validate the behavior and
5543 	 * pass a magic value back to the guest.
5544 	 */
5545 	if (exit_reason_want == VMX_APIC_ACCESS) {
5546 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5547 
5548 		assert_exit_reason(exit_reason_want);
5549 		report("got APIC access exit @ page offset 0x%03x, want 0x%03x",
5550 		       apic_page_offset == reg, apic_page_offset, reg);
5551 		skip_exit_insn();
5552 
5553 		/* Reenter guest so it can consume/check rcx and exit again. */
5554 		enter_guest();
5555 	} else if (exit_reason_want != VMX_VMCALL) {
5556 		report("Oops, bad exit expectation: %u.", false,
5557 		       exit_reason_want);
5558 	}
5559 
5560 	skip_exit_vmcall();
5561 	report_prefix_pop();
5562 }
5563 
5564 static void test_xapic_wr(
5565 	u32 reg, struct apic_reg_virt_expectation *expectation,
5566 	u32 *apic_access_address, u32 *virtual_apic_page)
5567 {
5568 	u32 val = expectation->val;
5569 	u32 exit_reason_want = expectation->wr_exit_reason;
5570 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5571 	bool virtualized =
5572 		expectation->virtualize_apic_accesses &&
5573 		(exit_reason_want == VMX_APIC_WRITE ||
5574 		 exit_reason_want == VMX_VMCALL);
5575 	bool checked = false;
5576 
5577 	report_prefix_pushf("xapic - writing 0x%x to 0x%03x", val, reg);
5578 
5579 	/* Configure guest to do an xapic read */
5580 	args->op = APIC_OP_XAPIC_WR;
5581 	args->apic_access_address = apic_access_address;
5582 	args->reg = reg;
5583 	args->val = val;
5584 
5585 	/* Setup virtual APIC page */
5586 	if (virtualized || !expectation->virtualize_apic_accesses) {
5587 		apic_access_address[apic_reg_index(reg)] = 0;
5588 		virtual_apic_page[apic_reg_index(reg)] = 0;
5589 	}
5590 
5591 	/* Enter guest */
5592 	enter_guest();
5593 
5594 	/*
5595 	 * Validate the behavior and
5596 	 * pass a magic value back to the guest.
5597 	 */
5598 	if (exit_reason_want == VMX_APIC_ACCESS) {
5599 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5600 
5601 		assert_exit_reason(exit_reason_want);
5602 		report("got APIC access exit @ page offset 0x%03x, want 0x%03x",
5603 		       apic_page_offset == reg, apic_page_offset, reg);
5604 		skip_exit_insn();
5605 
5606 		/* Reenter guest so it can consume/check rcx and exit again. */
5607 		enter_guest();
5608 	} else if (exit_reason_want == VMX_APIC_WRITE) {
5609 		assert_exit_reason(exit_reason_want);
5610 		report("got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%x",
5611 		       virtual_apic_page[apic_reg_index(reg)] == val,
5612 		       apic_reg_index(reg),
5613 		       virtual_apic_page[apic_reg_index(reg)], val);
5614 		checked = true;
5615 
5616 		/* Reenter guest so it can consume/check rcx and exit again. */
5617 		enter_guest();
5618 	} else if (exit_reason_want != VMX_VMCALL) {
5619 		report("Oops, bad exit expectation: %u.", false,
5620 		       exit_reason_want);
5621 	}
5622 
5623 	assert_exit_reason(VMX_VMCALL);
5624 	if (virtualized && !checked) {
5625 		u32 want = expectation->virt_fn(val);
5626 		u32 got = virtual_apic_page[apic_reg_index(reg)];
5627 		got = expectation->virt_fn(got);
5628 
5629 		report("exitless write; val is 0x%x, want 0x%x",
5630 		       got == want, got, want);
5631 	} else if (!expectation->virtualize_apic_accesses && !checked) {
5632 		u32 got = apic_access_address[apic_reg_index(reg)];
5633 
5634 		report("non-virtualized write; val is 0x%x, want 0x%x",
5635 		       got == val, got, val);
5636 	} else if (!expectation->virtualize_apic_accesses && checked) {
5637 		report("Non-virtualized write was prematurely checked!", false);
5638 	}
5639 
5640 	skip_exit_vmcall();
5641 	report_prefix_pop();
5642 }
5643 
5644 enum Config_type {
5645 	CONFIG_TYPE_GOOD,
5646 	CONFIG_TYPE_UNSUPPORTED,
5647 	CONFIG_TYPE_VMENTRY_FAILS_EARLY,
5648 };
5649 
5650 static enum Config_type configure_apic_reg_virt_test(
5651 	struct apic_reg_virt_config *apic_reg_virt_config)
5652 {
5653 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5654 	u32 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
5655 	/* Configs where L2 entry fails early, due to invalid controls. */
5656 	bool use_tpr_shadow_incorrectly_off =
5657 		!apic_reg_virt_config->use_tpr_shadow &&
5658 		(apic_reg_virt_config->apic_register_virtualization ||
5659 		 apic_reg_virt_config->virtualize_x2apic_mode) &&
5660 		apic_reg_virt_config->activate_secondary_controls;
5661 	bool virtualize_apic_accesses_incorrectly_on =
5662 		apic_reg_virt_config->virtualize_apic_accesses &&
5663 		apic_reg_virt_config->virtualize_x2apic_mode &&
5664 		apic_reg_virt_config->activate_secondary_controls;
5665 	bool vmentry_fails_early =
5666 		use_tpr_shadow_incorrectly_off ||
5667 		virtualize_apic_accesses_incorrectly_on;
5668 
5669 	if (apic_reg_virt_config->activate_secondary_controls) {
5670 		if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
5671 			printf("VM-execution control \"activate secondary controls\" NOT supported.\n");
5672 			return CONFIG_TYPE_UNSUPPORTED;
5673 		}
5674 		cpu_exec_ctrl0 |= CPU_SECONDARY;
5675 	} else {
5676 		cpu_exec_ctrl0 &= ~CPU_SECONDARY;
5677 	}
5678 
5679 	if (apic_reg_virt_config->virtualize_apic_accesses) {
5680 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES)) {
5681 			printf("VM-execution control \"virtualize APIC accesses\" NOT supported.\n");
5682 			return CONFIG_TYPE_UNSUPPORTED;
5683 		}
5684 		cpu_exec_ctrl1 |= CPU_VIRT_APIC_ACCESSES;
5685 	} else {
5686 		cpu_exec_ctrl1 &= ~CPU_VIRT_APIC_ACCESSES;
5687 	}
5688 
5689 	if (apic_reg_virt_config->use_tpr_shadow) {
5690 		if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
5691 			printf("VM-execution control \"use TPR shadow\" NOT supported.\n");
5692 			return CONFIG_TYPE_UNSUPPORTED;
5693 		}
5694 		cpu_exec_ctrl0 |= CPU_TPR_SHADOW;
5695 	} else {
5696 		cpu_exec_ctrl0 &= ~CPU_TPR_SHADOW;
5697 	}
5698 
5699 	if (apic_reg_virt_config->apic_register_virtualization) {
5700 		if (!(ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT)) {
5701 			printf("VM-execution control \"APIC-register virtualization\" NOT supported.\n");
5702 			return CONFIG_TYPE_UNSUPPORTED;
5703 		}
5704 		cpu_exec_ctrl1 |= CPU_APIC_REG_VIRT;
5705 	} else {
5706 		cpu_exec_ctrl1 &= ~CPU_APIC_REG_VIRT;
5707 	}
5708 
5709 	if (apic_reg_virt_config->virtualize_x2apic_mode) {
5710 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_X2APIC)) {
5711 			printf("VM-execution control \"virtualize x2APIC mode\" NOT supported.\n");
5712 			return CONFIG_TYPE_UNSUPPORTED;
5713 		}
5714 		cpu_exec_ctrl1 |= CPU_VIRT_X2APIC;
5715 	} else {
5716 		cpu_exec_ctrl1 &= ~CPU_VIRT_X2APIC;
5717 	}
5718 
5719 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
5720 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
5721 
5722 	if (vmentry_fails_early)
5723 		return CONFIG_TYPE_VMENTRY_FAILS_EARLY;
5724 
5725 	return CONFIG_TYPE_GOOD;
5726 }
5727 
5728 static bool cpu_has_apicv(void)
5729 {
5730 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
5731 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
5732 		(ctrl_pin_rev.clr & PIN_POST_INTR));
5733 }
5734 
5735 /* Validates APIC register access across valid virtualization configurations. */
5736 static void apic_reg_virt_test(void)
5737 {
5738 	u32 *apic_access_address;
5739 	u32 *virtual_apic_page;
5740 	u64 control;
5741 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5742 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
5743 	int i;
5744 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5745 
5746 	if (!cpu_has_apicv()) {
5747 		report_skip(__func__);
5748 		return;
5749 	}
5750 
5751 	control = cpu_exec_ctrl1;
5752 	control &= ~CPU_VINTD;
5753 	vmcs_write(CPU_EXEC_CTRL1, control);
5754 
5755 	test_set_guest(apic_reg_virt_guest);
5756 
5757 	/*
5758 	 * From the SDM: The 1-setting of the "virtualize APIC accesses"
5759 	 * VM-execution is guaranteed to apply only if translations to the
5760 	 * APIC-access address use a 4-KByte page.
5761 	 */
5762 	apic_access_address = alloc_page();
5763 	force_4k_page(apic_access_address);
5764 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_address));
5765 
5766 	virtual_apic_page = alloc_page();
5767 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
5768 
5769 	for (i = 0; i < ARRAY_SIZE(apic_reg_tests); i++) {
5770 		struct apic_reg_test *apic_reg_test = &apic_reg_tests[i];
5771 		struct apic_reg_virt_config *apic_reg_virt_config =
5772 				&apic_reg_test->apic_reg_virt_config;
5773 		enum Config_type config_type;
5774 		u32 reg;
5775 
5776 		printf("--- %s test ---\n", apic_reg_test->name);
5777 		config_type =
5778 			configure_apic_reg_virt_test(apic_reg_virt_config);
5779 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
5780 			printf("Skip because of missing features.\n");
5781 			continue;
5782 		}
5783 
5784 		if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
5785 			enter_guest_with_bad_controls();
5786 			continue;
5787 		}
5788 
5789 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
5790 			struct apic_reg_virt_expectation expectation = {};
5791 			bool ok;
5792 
5793 			ok = apic_reg_virt_exit_expectation(
5794 				reg, apic_reg_virt_config, &expectation);
5795 			if (!ok) {
5796 				report("Malformed test.", false);
5797 				break;
5798 			}
5799 
5800 			test_xapic_rd(reg, &expectation, apic_access_address,
5801 				      virtual_apic_page);
5802 			test_xapic_wr(reg, &expectation, apic_access_address,
5803 				      virtual_apic_page);
5804 		}
5805 	}
5806 
5807 	/* Terminate the guest */
5808 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
5809 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
5810 	args->op = TERMINATE;
5811 	enter_guest();
5812 	assert_exit_reason(VMX_VMCALL);
5813 }
5814 
5815 struct virt_x2apic_mode_config {
5816 	struct apic_reg_virt_config apic_reg_virt_config;
5817 	bool virtual_interrupt_delivery;
5818 	bool use_msr_bitmaps;
5819 	bool disable_x2apic_msr_intercepts;
5820 	bool disable_x2apic;
5821 };
5822 
5823 struct virt_x2apic_mode_test_case {
5824 	const char *name;
5825 	struct virt_x2apic_mode_config virt_x2apic_mode_config;
5826 };
5827 
5828 enum Virt_x2apic_mode_behavior_type {
5829 	X2APIC_ACCESS_VIRTUALIZED,
5830 	X2APIC_ACCESS_PASSED_THROUGH,
5831 	X2APIC_ACCESS_TRIGGERS_GP,
5832 };
5833 
5834 struct virt_x2apic_mode_expectation {
5835 	enum Reason rd_exit_reason;
5836 	enum Reason wr_exit_reason;
5837 
5838 	/*
5839 	 * RDMSR and WRMSR handle 64-bit values. However, except for ICR, all of
5840 	 * the x2APIC registers are 32 bits. Notice:
5841 	 *   1. vmx_x2apic_read() clears the upper 32 bits for 32-bit registers.
5842 	 *   2. vmx_x2apic_write() expects the val arg to be well-formed.
5843 	 */
5844 	u64 rd_val;
5845 	u64 wr_val;
5846 
5847 	/*
5848 	 * Compares input to virtualized output;
5849 	 * 1st arg is pointer to return expected virtualization output.
5850 	 */
5851 	u64 (*virt_fn)(u64);
5852 
5853 	enum Virt_x2apic_mode_behavior_type rd_behavior;
5854 	enum Virt_x2apic_mode_behavior_type wr_behavior;
5855 	bool wr_only;
5856 };
5857 
5858 static u64 virt_x2apic_mode_identity(u64 val)
5859 {
5860 	return val;
5861 }
5862 
5863 static u64 virt_x2apic_mode_nibble1(u64 val)
5864 {
5865 	return val & 0xf0;
5866 }
5867 
5868 static bool is_cmci_enabled(void)
5869 {
5870 	return rdmsr(MSR_IA32_MCG_CAP) & BIT_ULL(10);
5871 }
5872 
5873 static void virt_x2apic_mode_rd_expectation(
5874 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
5875 	bool apic_register_virtualization, bool virtual_interrupt_delivery,
5876 	struct virt_x2apic_mode_expectation *expectation)
5877 {
5878 	bool readable =
5879 		!x2apic_reg_reserved(reg) &&
5880 		reg != APIC_EOI;
5881 
5882 	if (reg == APIC_CMCI && !is_cmci_enabled())
5883 		readable = false;
5884 
5885 	expectation->rd_exit_reason = VMX_VMCALL;
5886 	expectation->virt_fn = virt_x2apic_mode_identity;
5887 	if (virt_x2apic_mode_on && apic_register_virtualization) {
5888 		expectation->rd_val = MAGIC_VAL_1;
5889 		if (reg == APIC_PROCPRI && virtual_interrupt_delivery)
5890 			expectation->virt_fn = virt_x2apic_mode_nibble1;
5891 		else if (reg == APIC_TASKPRI)
5892 			expectation->virt_fn = virt_x2apic_mode_nibble1;
5893 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
5894 	} else if (virt_x2apic_mode_on && !apic_register_virtualization &&
5895 		   reg == APIC_TASKPRI) {
5896 		expectation->rd_val = MAGIC_VAL_1;
5897 		expectation->virt_fn = virt_x2apic_mode_nibble1;
5898 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
5899 	} else if (!disable_x2apic && readable) {
5900 		expectation->rd_val = apic_read(reg);
5901 		expectation->rd_behavior = X2APIC_ACCESS_PASSED_THROUGH;
5902 	} else {
5903 		expectation->rd_behavior = X2APIC_ACCESS_TRIGGERS_GP;
5904 	}
5905 }
5906 
5907 /*
5908  * get_x2apic_wr_val() creates an innocuous write value for an x2APIC register.
5909  *
5910  * For writable registers, get_x2apic_wr_val() deposits the write value into the
5911  * val pointer arg and returns true. For non-writable registers, val is not
5912  * modified and get_x2apic_wr_val() returns false.
5913  */
5914 static bool get_x2apic_wr_val(u32 reg, u64 *val)
5915 {
5916 	switch (reg) {
5917 	case APIC_TASKPRI:
5918 		/* Bits 31:8 are reserved. */
5919 		*val &= 0xff;
5920 		break;
5921 	case APIC_EOI:
5922 	case APIC_ESR:
5923 	case APIC_TMICT:
5924 		/*
5925 		 * EOI, ESR: WRMSR of a non-zero value causes #GP(0).
5926 		 * TMICT: A write of 0 to the initial-count register effectively
5927 		 *        stops the local APIC timer, in both one-shot and
5928 		 *        periodic mode.
5929 		 */
5930 		*val = 0;
5931 		break;
5932 	case APIC_SPIV:
5933 	case APIC_LVTT:
5934 	case APIC_LVTTHMR:
5935 	case APIC_LVTPC:
5936 	case APIC_LVT0:
5937 	case APIC_LVT1:
5938 	case APIC_LVTERR:
5939 	case APIC_TDCR:
5940 		/*
5941 		 * To avoid writing a 1 to a reserved bit or causing some other
5942 		 * unintended side effect, read the current value and use it as
5943 		 * the write value.
5944 		 */
5945 		*val = apic_read(reg);
5946 		break;
5947 	case APIC_CMCI:
5948 		if (!is_cmci_enabled())
5949 			return false;
5950 		*val = apic_read(reg);
5951 		break;
5952 	case APIC_ICR:
5953 		*val = 0x40000 | 0xf1;
5954 		break;
5955 	case APIC_SELF_IPI:
5956 		/*
5957 		 * With special processing (i.e., virtualize x2APIC mode +
5958 		 * virtual interrupt delivery), writing zero causes an
5959 		 * APIC-write VM exit. We plan to add a test for enabling
5960 		 * "virtual-interrupt delivery" in VMCS12, and that's where we
5961 		 * will test a self IPI with special processing.
5962 		 */
5963 		*val = 0x0;
5964 		break;
5965 	default:
5966 		return false;
5967 	}
5968 
5969 	return true;
5970 }
5971 
5972 static bool special_processing_applies(u32 reg, u64 *val,
5973 				       bool virt_int_delivery)
5974 {
5975 	bool special_processing =
5976 		(reg == APIC_TASKPRI) ||
5977 		(virt_int_delivery &&
5978 		 (reg == APIC_EOI || reg == APIC_SELF_IPI));
5979 
5980 	if (special_processing) {
5981 		TEST_ASSERT(get_x2apic_wr_val(reg, val));
5982 		return true;
5983 	}
5984 
5985 	return false;
5986 }
5987 
5988 static void virt_x2apic_mode_wr_expectation(
5989 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
5990 	bool virt_int_delivery,
5991 	struct virt_x2apic_mode_expectation *expectation)
5992 {
5993 	expectation->wr_exit_reason = VMX_VMCALL;
5994 	expectation->wr_val = MAGIC_VAL_1;
5995 	expectation->wr_only = false;
5996 
5997 	if (virt_x2apic_mode_on &&
5998 	    special_processing_applies(reg, &expectation->wr_val,
5999 				       virt_int_delivery)) {
6000 		expectation->wr_behavior = X2APIC_ACCESS_VIRTUALIZED;
6001 		if (reg == APIC_SELF_IPI)
6002 			expectation->wr_exit_reason = VMX_APIC_WRITE;
6003 	} else if (!disable_x2apic &&
6004 		   get_x2apic_wr_val(reg, &expectation->wr_val)) {
6005 		expectation->wr_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6006 		if (reg == APIC_EOI || reg == APIC_SELF_IPI)
6007 			expectation->wr_only = true;
6008 		if (reg == APIC_ICR)
6009 			expectation->wr_exit_reason = VMX_EXTINT;
6010 	} else {
6011 		expectation->wr_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6012 		/*
6013 		 * Writing 1 to a reserved bit triggers a #GP.
6014 		 * Thus, set the write value to 0, which seems
6015 		 * the most likely to detect a missed #GP.
6016 		 */
6017 		expectation->wr_val = 0;
6018 	}
6019 }
6020 
6021 static void virt_x2apic_mode_exit_expectation(
6022 	u32 reg, struct virt_x2apic_mode_config *config,
6023 	struct virt_x2apic_mode_expectation *expectation)
6024 {
6025 	struct apic_reg_virt_config *base_config =
6026 		&config->apic_reg_virt_config;
6027 	bool virt_x2apic_mode_on =
6028 		base_config->virtualize_x2apic_mode &&
6029 		config->use_msr_bitmaps &&
6030 		config->disable_x2apic_msr_intercepts &&
6031 		base_config->activate_secondary_controls;
6032 
6033 	virt_x2apic_mode_wr_expectation(
6034 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6035 		config->virtual_interrupt_delivery, expectation);
6036 	virt_x2apic_mode_rd_expectation(
6037 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6038 		base_config->apic_register_virtualization,
6039 		config->virtual_interrupt_delivery, expectation);
6040 }
6041 
6042 struct virt_x2apic_mode_test_case virt_x2apic_mode_tests[] = {
6043 	/*
6044 	 * Baseline "virtualize x2APIC mode" configuration:
6045 	 *   - virtualize x2APIC mode
6046 	 *   - virtual-interrupt delivery
6047 	 *   - APIC-register virtualization
6048 	 *   - x2APIC MSR intercepts disabled
6049 	 *
6050 	 * Reads come from virtual APIC page, special processing applies to
6051 	 * VTPR, EOI, and SELF IPI, and all other writes pass through to L1
6052 	 * APIC.
6053 	 */
6054 	{
6055 		.name = "Baseline",
6056 		.virt_x2apic_mode_config = {
6057 			.virtual_interrupt_delivery = true,
6058 			.use_msr_bitmaps = true,
6059 			.disable_x2apic_msr_intercepts = true,
6060 			.disable_x2apic = false,
6061 			.apic_reg_virt_config = {
6062 				.apic_register_virtualization = true,
6063 				.use_tpr_shadow = true,
6064 				.virtualize_apic_accesses = false,
6065 				.virtualize_x2apic_mode = true,
6066 				.activate_secondary_controls = true,
6067 			},
6068 		},
6069 	},
6070 	{
6071 		.name = "Baseline w/ x2apic disabled",
6072 		.virt_x2apic_mode_config = {
6073 			.virtual_interrupt_delivery = true,
6074 			.use_msr_bitmaps = true,
6075 			.disable_x2apic_msr_intercepts = true,
6076 			.disable_x2apic = true,
6077 			.apic_reg_virt_config = {
6078 				.apic_register_virtualization = true,
6079 				.use_tpr_shadow = true,
6080 				.virtualize_apic_accesses = false,
6081 				.virtualize_x2apic_mode = true,
6082 				.activate_secondary_controls = true,
6083 			},
6084 		},
6085 	},
6086 
6087 	/*
6088 	 * Baseline, minus virtual-interrupt delivery. Reads come from virtual
6089 	 * APIC page, special processing applies to VTPR, and all other writes
6090 	 * pass through to L1 APIC.
6091 	 */
6092 	{
6093 		.name = "Baseline - virtual interrupt delivery",
6094 		.virt_x2apic_mode_config = {
6095 			.virtual_interrupt_delivery = false,
6096 			.use_msr_bitmaps = true,
6097 			.disable_x2apic_msr_intercepts = true,
6098 			.disable_x2apic = false,
6099 			.apic_reg_virt_config = {
6100 				.apic_register_virtualization = true,
6101 				.use_tpr_shadow = true,
6102 				.virtualize_apic_accesses = false,
6103 				.virtualize_x2apic_mode = true,
6104 				.activate_secondary_controls = true,
6105 			},
6106 		},
6107 	},
6108 
6109 	/*
6110 	 * Baseline, minus APIC-register virtualization. x2APIC reads pass
6111 	 * through to L1's APIC, unless reading VTPR
6112 	 */
6113 	{
6114 		.name = "Virtualize x2APIC mode, no APIC reg virt",
6115 		.virt_x2apic_mode_config = {
6116 			.virtual_interrupt_delivery = true,
6117 			.use_msr_bitmaps = true,
6118 			.disable_x2apic_msr_intercepts = true,
6119 			.disable_x2apic = false,
6120 			.apic_reg_virt_config = {
6121 				.apic_register_virtualization = false,
6122 				.use_tpr_shadow = true,
6123 				.virtualize_apic_accesses = false,
6124 				.virtualize_x2apic_mode = true,
6125 				.activate_secondary_controls = true,
6126 			},
6127 		},
6128 	},
6129 	{
6130 		.name = "Virtualize x2APIC mode, no APIC reg virt, x2APIC off",
6131 		.virt_x2apic_mode_config = {
6132 			.virtual_interrupt_delivery = true,
6133 			.use_msr_bitmaps = true,
6134 			.disable_x2apic_msr_intercepts = true,
6135 			.disable_x2apic = true,
6136 			.apic_reg_virt_config = {
6137 				.apic_register_virtualization = false,
6138 				.use_tpr_shadow = true,
6139 				.virtualize_apic_accesses = false,
6140 				.virtualize_x2apic_mode = true,
6141 				.activate_secondary_controls = true,
6142 			},
6143 		},
6144 	},
6145 
6146 	/*
6147 	 * Enable "virtualize x2APIC mode" and "APIC-register virtualization",
6148 	 * and disable intercepts for the x2APIC MSRs, but fail to enable
6149 	 * "activate secondary controls" (i.e. L2 gets access to L1's x2APIC
6150 	 * MSRs).
6151 	 */
6152 	{
6153 		.name = "Fail to enable activate secondary controls",
6154 		.virt_x2apic_mode_config = {
6155 			.virtual_interrupt_delivery = true,
6156 			.use_msr_bitmaps = true,
6157 			.disable_x2apic_msr_intercepts = true,
6158 			.disable_x2apic = false,
6159 			.apic_reg_virt_config = {
6160 				.apic_register_virtualization = true,
6161 				.use_tpr_shadow = true,
6162 				.virtualize_apic_accesses = false,
6163 				.virtualize_x2apic_mode = true,
6164 				.activate_secondary_controls = false,
6165 			},
6166 		},
6167 	},
6168 
6169 	/*
6170 	 * Enable "APIC-register virtualization" and enable "activate secondary
6171 	 * controls" and disable intercepts for the x2APIC MSRs, but do not
6172 	 * enable the "virtualize x2APIC mode" VM-execution control (i.e. L2
6173 	 * gets access to L1's x2APIC MSRs).
6174 	 */
6175 	{
6176 		.name = "Fail to enable virtualize x2APIC mode",
6177 		.virt_x2apic_mode_config = {
6178 			.virtual_interrupt_delivery = true,
6179 			.use_msr_bitmaps = true,
6180 			.disable_x2apic_msr_intercepts = true,
6181 			.disable_x2apic = false,
6182 			.apic_reg_virt_config = {
6183 				.apic_register_virtualization = true,
6184 				.use_tpr_shadow = true,
6185 				.virtualize_apic_accesses = false,
6186 				.virtualize_x2apic_mode = false,
6187 				.activate_secondary_controls = true,
6188 			},
6189 		},
6190 	},
6191 
6192 	/*
6193 	 * Disable "Virtualize x2APIC mode", disable x2APIC MSR intercepts, and
6194 	 * enable "APIC-register virtualization" --> L2 gets L1's x2APIC MSRs.
6195 	 */
6196 	{
6197 		.name = "Baseline",
6198 		.virt_x2apic_mode_config = {
6199 			.virtual_interrupt_delivery = true,
6200 			.use_msr_bitmaps = true,
6201 			.disable_x2apic_msr_intercepts = true,
6202 			.disable_x2apic = false,
6203 			.apic_reg_virt_config = {
6204 				.apic_register_virtualization = true,
6205 				.use_tpr_shadow = true,
6206 				.virtualize_apic_accesses = false,
6207 				.virtualize_x2apic_mode = false,
6208 				.activate_secondary_controls = true,
6209 			},
6210 		},
6211 	},
6212 };
6213 
6214 enum X2apic_op {
6215 	X2APIC_OP_RD,
6216 	X2APIC_OP_WR,
6217 	X2APIC_TERMINATE,
6218 };
6219 
6220 static u64 vmx_x2apic_read(u32 reg)
6221 {
6222 	u32 msr_addr = x2apic_msr(reg);
6223 	u64 val;
6224 
6225 	val = rdmsr(msr_addr);
6226 
6227 	return val;
6228 }
6229 
6230 static void vmx_x2apic_write(u32 reg, u64 val)
6231 {
6232 	u32 msr_addr = x2apic_msr(reg);
6233 
6234 	wrmsr(msr_addr, val);
6235 }
6236 
6237 struct virt_x2apic_mode_guest_args {
6238 	enum X2apic_op op;
6239 	u32 reg;
6240 	u64 val;
6241 	bool should_gp;
6242 	u64 (*virt_fn)(u64);
6243 } virt_x2apic_mode_guest_args;
6244 
6245 static volatile bool handle_x2apic_gp_ran;
6246 static volatile u32 handle_x2apic_gp_insn_len;
6247 static void handle_x2apic_gp(struct ex_regs *regs)
6248 {
6249 	handle_x2apic_gp_ran = true;
6250 	regs->rip += handle_x2apic_gp_insn_len;
6251 }
6252 
6253 static handler setup_x2apic_gp_handler(void)
6254 {
6255 	handler old_handler;
6256 
6257 	old_handler = handle_exception(GP_VECTOR, handle_x2apic_gp);
6258 	/* RDMSR and WRMSR are both 2 bytes, assuming no prefixes. */
6259 	handle_x2apic_gp_insn_len = 2;
6260 
6261 	return old_handler;
6262 }
6263 
6264 static void teardown_x2apic_gp_handler(handler old_handler)
6265 {
6266 	handle_exception(GP_VECTOR, old_handler);
6267 
6268 	/*
6269 	 * Defensively reset instruction length, so that if the handler is
6270 	 * incorrectly used, it will loop infinitely, rather than run off into
6271 	 * la la land.
6272 	 */
6273 	handle_x2apic_gp_insn_len = 0;
6274 	handle_x2apic_gp_ran = false;
6275 }
6276 
6277 static void virt_x2apic_mode_guest(void)
6278 {
6279 	volatile struct virt_x2apic_mode_guest_args *args =
6280 		&virt_x2apic_mode_guest_args;
6281 
6282 	for (;;) {
6283 		enum X2apic_op op = args->op;
6284 		u32 reg = args->reg;
6285 		u64 val = args->val;
6286 		bool should_gp = args->should_gp;
6287 		u64 (*virt_fn)(u64) = args->virt_fn;
6288 		handler old_handler;
6289 
6290 		if (op == X2APIC_TERMINATE)
6291 			break;
6292 
6293 		if (should_gp) {
6294 			TEST_ASSERT(!handle_x2apic_gp_ran);
6295 			old_handler = setup_x2apic_gp_handler();
6296 		}
6297 
6298 		if (op == X2APIC_OP_RD) {
6299 			u64 ret = vmx_x2apic_read(reg);
6300 
6301 			if (!should_gp) {
6302 				u64 want = virt_fn(val);
6303 				u64 got = virt_fn(ret);
6304 
6305 				report("APIC read; got 0x%lx, want 0x%lx.",
6306 				       got == want, got, want);
6307 			}
6308 		} else if (op == X2APIC_OP_WR) {
6309 			vmx_x2apic_write(reg, val);
6310 		}
6311 
6312 		if (should_gp) {
6313 			report("x2APIC op triggered GP.",
6314 			       handle_x2apic_gp_ran);
6315 			teardown_x2apic_gp_handler(old_handler);
6316 		}
6317 
6318 		/*
6319 		 * The L1 should always execute a vmcall after it's done testing
6320 		 * an individual APIC operation. This helps to validate that the
6321 		 * L1 and L2 are in sync with each other, as expected.
6322 		 */
6323 		vmcall();
6324 	}
6325 }
6326 
6327 static void test_x2apic_rd(
6328 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6329 	u32 *virtual_apic_page)
6330 {
6331 	u64 val = expectation->rd_val;
6332 	u32 exit_reason_want = expectation->rd_exit_reason;
6333 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6334 
6335 	report_prefix_pushf("x2apic - reading 0x%03x", reg);
6336 
6337 	/* Configure guest to do an x2apic read */
6338 	args->op = X2APIC_OP_RD;
6339 	args->reg = reg;
6340 	args->val = val;
6341 	args->should_gp = expectation->rd_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6342 	args->virt_fn = expectation->virt_fn;
6343 
6344 	/* Setup virtual APIC page */
6345 	if (expectation->rd_behavior == X2APIC_ACCESS_VIRTUALIZED)
6346 		virtual_apic_page[apic_reg_index(reg)] = (u32)val;
6347 
6348 	/* Enter guest */
6349 	enter_guest();
6350 
6351 	if (exit_reason_want != VMX_VMCALL) {
6352 		report("Oops, bad exit expectation: %u.", false,
6353 		       exit_reason_want);
6354 	}
6355 
6356 	skip_exit_vmcall();
6357 	report_prefix_pop();
6358 }
6359 
6360 static volatile bool handle_x2apic_ipi_ran;
6361 static void handle_x2apic_ipi(isr_regs_t *regs)
6362 {
6363 	handle_x2apic_ipi_ran = true;
6364 	eoi();
6365 }
6366 
6367 static void test_x2apic_wr(
6368 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6369 	u32 *virtual_apic_page)
6370 {
6371 	u64 val = expectation->wr_val;
6372 	u32 exit_reason_want = expectation->wr_exit_reason;
6373 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6374 	int ipi_vector = 0xf1;
6375 	u32 restore_val = 0;
6376 
6377 	report_prefix_pushf("x2apic - writing 0x%lx to 0x%03x", val, reg);
6378 
6379 	/* Configure guest to do an x2apic read */
6380 	args->op = X2APIC_OP_WR;
6381 	args->reg = reg;
6382 	args->val = val;
6383 	args->should_gp = expectation->wr_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6384 
6385 	/* Setup virtual APIC page */
6386 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED)
6387 		virtual_apic_page[apic_reg_index(reg)] = 0;
6388 	if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH && !expectation->wr_only)
6389 		restore_val = apic_read(reg);
6390 
6391 	/* Setup IPI handler */
6392 	handle_x2apic_ipi_ran = false;
6393 	handle_irq(ipi_vector, handle_x2apic_ipi);
6394 
6395 	/* Enter guest */
6396 	enter_guest();
6397 
6398 	/*
6399 	 * Validate the behavior and
6400 	 * pass a magic value back to the guest.
6401 	 */
6402 	if (exit_reason_want == VMX_EXTINT) {
6403 		assert_exit_reason(exit_reason_want);
6404 
6405 		/* Clear the external interrupt. */
6406 		irq_enable();
6407 		asm volatile ("nop");
6408 		irq_disable();
6409 		report("Got pending interrupt after IRQ enabled.",
6410 		       handle_x2apic_ipi_ran);
6411 
6412 		enter_guest();
6413 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6414 		assert_exit_reason(exit_reason_want);
6415 		report("got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%lx",
6416 		       virtual_apic_page[apic_reg_index(reg)] == val,
6417 		       apic_reg_index(reg),
6418 		       virtual_apic_page[apic_reg_index(reg)], val);
6419 
6420 		/* Reenter guest so it can consume/check rcx and exit again. */
6421 		enter_guest();
6422 	} else if (exit_reason_want != VMX_VMCALL) {
6423 		report("Oops, bad exit expectation: %u.", false,
6424 		       exit_reason_want);
6425 	}
6426 
6427 	assert_exit_reason(VMX_VMCALL);
6428 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) {
6429 		u64 want = val;
6430 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6431 
6432 		report("x2APIC write; got 0x%x, want 0x%lx",
6433 		       got == want, got, want);
6434 	} else if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH) {
6435 		if (!expectation->wr_only) {
6436 			u32 got = apic_read(reg);
6437 			bool ok;
6438 
6439 			/*
6440 			 * When L1's TPR is passed through to L2, the lower
6441 			 * nibble can be lost. For example, if L2 executes
6442 			 * WRMSR(0x808, 0x78), then, L1 might read 0x70.
6443 			 *
6444 			 * Here's how the lower nibble can get lost:
6445 			 *   1. L2 executes WRMSR(0x808, 0x78).
6446 			 *   2. L2 exits to L0 with a WRMSR exit.
6447 			 *   3. L0 emulates WRMSR, by writing L1's TPR.
6448 			 *   4. L0 re-enters L2.
6449 			 *   5. L2 exits to L0 (reason doesn't matter).
6450 			 *   6. L0 reflects L2's exit to L1.
6451 			 *   7. Before entering L1, L0 exits to user-space
6452 			 *      (e.g., to satisfy TPR access reporting).
6453 			 *   8. User-space executes KVM_SET_REGS ioctl, which
6454 			 *      clears the lower nibble of L1's TPR.
6455 			 */
6456 			if (reg == APIC_TASKPRI) {
6457 				got = apic_virt_nibble1(got);
6458 				val = apic_virt_nibble1(val);
6459 			}
6460 
6461 			ok = got == val;
6462 			report("non-virtualized write; val is 0x%x, want 0x%lx",
6463 			       ok, got, val);
6464 			apic_write(reg, restore_val);
6465 		} else {
6466 			report("non-virtualized and write-only OK", true);
6467 		}
6468 	}
6469 	skip_exit_insn();
6470 
6471 	report_prefix_pop();
6472 }
6473 
6474 static enum Config_type configure_virt_x2apic_mode_test(
6475 	struct virt_x2apic_mode_config *virt_x2apic_mode_config,
6476 	u8 *msr_bitmap_page)
6477 {
6478 	int msr;
6479 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6480 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6481 
6482 	/* x2apic-specific VMCS config */
6483 	if (virt_x2apic_mode_config->use_msr_bitmaps) {
6484 		/* virt_x2apic_mode_test() checks for MSR bitmaps support */
6485 		cpu_exec_ctrl0 |= CPU_MSR_BITMAP;
6486 	} else {
6487 		cpu_exec_ctrl0 &= ~CPU_MSR_BITMAP;
6488 	}
6489 
6490 	if (virt_x2apic_mode_config->virtual_interrupt_delivery) {
6491 		if (!(ctrl_cpu_rev[1].clr & CPU_VINTD)) {
6492 			report_skip("VM-execution control \"virtual-interrupt delivery\" NOT supported.\n");
6493 			return CONFIG_TYPE_UNSUPPORTED;
6494 		}
6495 		cpu_exec_ctrl1 |= CPU_VINTD;
6496 	} else {
6497 		cpu_exec_ctrl1 &= ~CPU_VINTD;
6498 	}
6499 
6500 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6501 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6502 
6503 	/* x2APIC MSR intercepts are usually off for "Virtualize x2APIC mode" */
6504 	for (msr = 0x800; msr <= 0x8ff; msr++) {
6505 		if (virt_x2apic_mode_config->disable_x2apic_msr_intercepts) {
6506 			clear_bit(msr, msr_bitmap_page + 0x000);
6507 			clear_bit(msr, msr_bitmap_page + 0x800);
6508 		} else {
6509 			set_bit(msr, msr_bitmap_page + 0x000);
6510 			set_bit(msr, msr_bitmap_page + 0x800);
6511 		}
6512 	}
6513 
6514 	/* x2APIC mode can impact virtualization */
6515 	reset_apic();
6516 	if (!virt_x2apic_mode_config->disable_x2apic)
6517 		enable_x2apic();
6518 
6519 	return configure_apic_reg_virt_test(
6520 		&virt_x2apic_mode_config->apic_reg_virt_config);
6521 }
6522 
6523 static void virt_x2apic_mode_test(void)
6524 {
6525 	u32 *virtual_apic_page;
6526 	u8 *msr_bitmap_page;
6527 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6528 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6529 	int i;
6530 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6531 
6532 	if (!cpu_has_apicv()) {
6533 		report_skip(__func__);
6534 		return;
6535 	}
6536 
6537 	/*
6538 	 * This is to exercise an issue in KVM's logic to merge L0's and L1's
6539 	 * MSR bitmaps. Previously, an L1 could get at L0's x2APIC MSRs by
6540 	 * writing the IA32_SPEC_CTRL MSR or the IA32_PRED_CMD MSRs. KVM would
6541 	 * then proceed to manipulate the MSR bitmaps, as if VMCS12 had the
6542 	 * "Virtualize x2APIC mod" control set, even when it didn't.
6543 	 */
6544 	if (has_spec_ctrl())
6545 		wrmsr(MSR_IA32_SPEC_CTRL, 1);
6546 
6547 	/*
6548 	 * Check that VMCS12 supports:
6549 	 *   - "Virtual-APIC address", indicated by "use TPR shadow"
6550 	 *   - "MSR-bitmap address", indicated by "use MSR bitmaps"
6551 	 */
6552 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6553 		report_skip("VM-execution control \"use TPR shadow\" NOT supported.\n");
6554 		return;
6555 	} else if (!(ctrl_cpu_rev[0].clr & CPU_MSR_BITMAP)) {
6556 		report_skip("VM-execution control \"use MSR bitmaps\" NOT supported.\n");
6557 		return;
6558 	}
6559 
6560 	test_set_guest(virt_x2apic_mode_guest);
6561 
6562 	virtual_apic_page = alloc_page();
6563 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6564 
6565 	msr_bitmap_page = alloc_page();
6566 	memset(msr_bitmap_page, 0xff, PAGE_SIZE);
6567 	vmcs_write(MSR_BITMAP, virt_to_phys(msr_bitmap_page));
6568 
6569 	for (i = 0; i < ARRAY_SIZE(virt_x2apic_mode_tests); i++) {
6570 		struct virt_x2apic_mode_test_case *virt_x2apic_mode_test_case =
6571 			&virt_x2apic_mode_tests[i];
6572 		struct virt_x2apic_mode_config *virt_x2apic_mode_config =
6573 			&virt_x2apic_mode_test_case->virt_x2apic_mode_config;
6574 		enum Config_type config_type;
6575 		u32 reg;
6576 
6577 		printf("--- %s test ---\n", virt_x2apic_mode_test_case->name);
6578 		config_type =
6579 			configure_virt_x2apic_mode_test(virt_x2apic_mode_config,
6580 							msr_bitmap_page);
6581 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6582 			report_skip("Skip because of missing features.\n");
6583 			continue;
6584 		} else if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6585 			enter_guest_with_bad_controls();
6586 			continue;
6587 		}
6588 
6589 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6590 			struct virt_x2apic_mode_expectation expectation;
6591 
6592 			virt_x2apic_mode_exit_expectation(
6593 				reg, virt_x2apic_mode_config, &expectation);
6594 
6595 			test_x2apic_rd(reg, &expectation, virtual_apic_page);
6596 			test_x2apic_wr(reg, &expectation, virtual_apic_page);
6597 		}
6598 	}
6599 
6600 
6601 	/* Terminate the guest */
6602 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6603 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6604 	args->op = X2APIC_TERMINATE;
6605 	enter_guest();
6606 	assert_exit_reason(VMX_VMCALL);
6607 }
6608 
6609 /*
6610  * On processors that support Intel 64 architecture, the IA32_SYSENTER_ESP
6611  * field and the IA32_SYSENTER_EIP field must each contain a canonical
6612  * address.
6613  *
6614  *  [Intel SDM]
6615  */
6616 static void test_sysenter_field(u32 field, const char *name)
6617 {
6618 	u64 addr_saved = vmcs_read(field);
6619 
6620 	vmcs_write(field, NONCANONICAL);
6621 	report_prefix_pushf("%s non-canonical", name);
6622 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6623 	report_prefix_pop();
6624 
6625 	vmcs_write(field, 0xffffffff);
6626 	report_prefix_pushf("%s canonical", name);
6627 	test_vmx_vmlaunch(0);
6628 	report_prefix_pop();
6629 
6630 	vmcs_write(field, addr_saved);
6631 }
6632 
6633 static void test_ctl_reg(const char *cr_name, u64 cr, u64 fixed0, u64 fixed1)
6634 {
6635 	u64 val;
6636 	u64 cr_saved = vmcs_read(cr);
6637 	int i;
6638 
6639 	val = fixed0 & fixed1;
6640 	if (cr == HOST_CR4)
6641 		vmcs_write(cr, val | X86_CR4_PAE);
6642 	else
6643 		vmcs_write(cr, val);
6644 	report_prefix_pushf("%s %lx", cr_name, val);
6645 	if (val == fixed0)
6646 		test_vmx_vmlaunch(0);
6647 	else
6648 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6649 	report_prefix_pop();
6650 
6651 	for (i = 0; i < 64; i++) {
6652 
6653 		/* Set a bit when the corresponding bit in fixed1 is 0 */
6654 		if ((fixed1 & (1ull << i)) == 0) {
6655 			if (cr == HOST_CR4 && ((1ull << i) & X86_CR4_SMEP ||
6656 					       (1ull << i) & X86_CR4_SMAP))
6657 				continue;
6658 
6659 			vmcs_write(cr, cr_saved | (1ull << i));
6660 			report_prefix_pushf("%s %llx", cr_name,
6661 						cr_saved | (1ull << i));
6662 			test_vmx_vmlaunch(
6663 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6664 			report_prefix_pop();
6665 		}
6666 
6667 		/* Unset a bit when the corresponding bit in fixed0 is 1 */
6668 		if (fixed0 & (1ull << i)) {
6669 			vmcs_write(cr, cr_saved & ~(1ull << i));
6670 			report_prefix_pushf("%s %llx", cr_name,
6671 						cr_saved & ~(1ull << i));
6672 			test_vmx_vmlaunch(
6673 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6674 			report_prefix_pop();
6675 		}
6676 	}
6677 
6678 	vmcs_write(cr, cr_saved);
6679 }
6680 
6681 /*
6682  * 1. The CR0 field must not set any bit to a value not supported in VMX
6683  *    operation.
6684  * 2. The CR4 field must not set any bit to a value not supported in VMX
6685  *    operation.
6686  * 3. On processors that support Intel 64 architecture, the CR3 field must
6687  *    be such that bits 63:52 and bits in the range 51:32 beyond the
6688  *    processor’s physical-address width must be 0.
6689  *
6690  *  [Intel SDM]
6691  */
6692 static void test_host_ctl_regs(void)
6693 {
6694 	u64 fixed0, fixed1, cr3, cr3_saved;
6695 	int i;
6696 
6697 	/* Test CR0 */
6698 	fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0);
6699 	fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1);
6700 	test_ctl_reg("HOST_CR0", HOST_CR0, fixed0, fixed1);
6701 
6702 	/* Test CR4 */
6703 	fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0);
6704 	fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1) &
6705 		 ~(X86_CR4_SMEP | X86_CR4_SMAP);
6706 	test_ctl_reg("HOST_CR4", HOST_CR4, fixed0, fixed1);
6707 
6708 	/* Test CR3 */
6709 	cr3_saved = vmcs_read(HOST_CR3);
6710 	for (i = cpuid_maxphyaddr(); i < 64; i++) {
6711 		cr3 = cr3_saved | (1ul << i);
6712 		vmcs_write(HOST_CR3, cr3);
6713 		report_prefix_pushf("HOST_CR3 %lx", cr3);
6714 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6715 		report_prefix_pop();
6716 	}
6717 
6718 	vmcs_write(HOST_CR3, cr3_saved);
6719 }
6720 
6721 static void test_efer_vmlaunch(u32 fld, bool ok)
6722 {
6723 	if (fld == HOST_EFER) {
6724 		if (ok)
6725 			test_vmx_vmlaunch(0);
6726 		else
6727 			test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6728 	} else {
6729 		if (ok) {
6730 			enter_guest();
6731 			report("vmlaunch succeeds", vmcs_read(EXI_REASON) == VMX_VMCALL);
6732 		} else {
6733 			enter_guest_with_invalid_guest_state();
6734 			report("vmlaunch fails", vmcs_read(EXI_REASON) == (VMX_ENTRY_FAILURE | VMX_FAIL_STATE));
6735 		}
6736 		advance_guest_state_test();
6737 	}
6738 }
6739 
6740 static void test_efer_one(u32 fld, const char * fld_name, u64 efer,
6741 			  u32 ctrl_fld, u64 ctrl,
6742 			  int i, const char *efer_bit_name)
6743 {
6744 	bool ok;
6745 
6746 	ok = true;
6747 	if (ctrl_fld == EXI_CONTROLS && (ctrl & EXI_LOAD_EFER)) {
6748 		if (!!(efer & EFER_LMA) != !!(ctrl & EXI_HOST_64))
6749 			ok = false;
6750 		if (!!(efer & EFER_LME) != !!(ctrl & EXI_HOST_64))
6751 			ok = false;
6752 	}
6753 	if (ctrl_fld == ENT_CONTROLS && (ctrl & ENT_LOAD_EFER)) {
6754 		/* Check LMA too since CR0.PG is set.  */
6755 		if (!!(efer & EFER_LMA) != !!(ctrl & ENT_GUEST_64))
6756 			ok = false;
6757 		if (!!(efer & EFER_LME) != !!(ctrl & ENT_GUEST_64))
6758 			ok = false;
6759 	}
6760 
6761 	/*
6762 	 * Skip the test if it would enter the guest in 32-bit mode.
6763 	 * Perhaps write the test in assembly and make sure it
6764 	 * can be run in either mode?
6765 	 */
6766 	if (fld == GUEST_EFER && ok && !(ctrl & ENT_GUEST_64))
6767 		return;
6768 
6769 	vmcs_write(ctrl_fld, ctrl);
6770 	vmcs_write(fld, efer);
6771 	report_prefix_pushf("%s %s bit turned %s, controls %s",
6772 			    fld_name, efer_bit_name,
6773 			    (i & 1) ? "on" : "off",
6774 			    (i & 2) ? "on" : "off");
6775 
6776 	test_efer_vmlaunch(fld, ok);
6777 	report_prefix_pop();
6778 }
6779 
6780 static void test_efer_bit(u32 fld, const char * fld_name,
6781 			  u32 ctrl_fld, u64 ctrl_bit, u64 efer_bit,
6782 			  const char *efer_bit_name)
6783 {
6784 	u64 efer_saved = vmcs_read(fld);
6785 	u32 ctrl_saved = vmcs_read(ctrl_fld);
6786 	int i;
6787 
6788 	for (i = 0; i < 4; i++) {
6789 		u64 efer = efer_saved & ~efer_bit;
6790 		u64 ctrl = ctrl_saved & ~ctrl_bit;
6791 
6792 		if (i & 1)
6793 			efer |= efer_bit;
6794 		if (i & 2)
6795 			ctrl |= ctrl_bit;
6796 
6797 		test_efer_one(fld, fld_name, efer, ctrl_fld, ctrl,
6798 			      i, efer_bit_name);
6799 	}
6800 
6801 	vmcs_write(ctrl_fld, ctrl_saved);
6802 	vmcs_write(fld, efer_saved);
6803 }
6804 
6805 static void test_efer(u32 fld, const char * fld_name, u32 ctrl_fld,
6806 		      u64 ctrl_bit1, u64 ctrl_bit2)
6807 {
6808 	u64 efer_saved = vmcs_read(fld);
6809 	u32 ctrl_saved = vmcs_read(ctrl_fld);
6810 	u64 efer_reserved_bits =  ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
6811 	u64 i;
6812 	u64 efer;
6813 
6814 	if (cpu_has_efer_nx())
6815 		efer_reserved_bits &= ~EFER_NX;
6816 
6817 	if (!ctrl_bit1) {
6818 		printf("\"Load-IA32-EFER\" exit control not supported\n");
6819 		goto test_entry_exit_mode;
6820 	}
6821 
6822 	report_prefix_pushf("%s %lx", fld_name, efer_saved);
6823 	test_efer_vmlaunch(fld, true);
6824 	report_prefix_pop();
6825 
6826 	/*
6827 	 * Check reserved bits
6828 	 */
6829 	vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit1);
6830 	for (i = 0; i < 64; i++) {
6831 		if ((1ull << i) & efer_reserved_bits) {
6832 			efer = efer_saved | (1ull << i);
6833 			vmcs_write(fld, efer);
6834 			report_prefix_pushf("%s %lx", fld_name, efer);
6835 			test_efer_vmlaunch(fld, true);
6836 			report_prefix_pop();
6837 		}
6838 	}
6839 
6840 	vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit1);
6841 	for (i = 0; i < 64; i++) {
6842 		if ((1ull << i) & efer_reserved_bits) {
6843 			efer = efer_saved | (1ull << i);
6844 			vmcs_write(fld, efer);
6845 			report_prefix_pushf("%s %lx", fld_name, efer);
6846 			test_efer_vmlaunch(fld, false);
6847 			report_prefix_pop();
6848 		}
6849 	}
6850 
6851 	vmcs_write(ctrl_fld, ctrl_saved);
6852 	vmcs_write(fld, efer_saved);
6853 
6854 	/*
6855 	 * Check LMA and LME bits
6856 	 */
6857 	test_efer_bit(fld, fld_name,
6858 		      ctrl_fld, ctrl_bit1,
6859 		      EFER_LMA,
6860 		      "EFER_LMA");
6861 	test_efer_bit(fld, fld_name,
6862 		      ctrl_fld, ctrl_bit1,
6863 		      EFER_LME,
6864 		      "EFER_LME");
6865 
6866 test_entry_exit_mode:
6867 	test_efer_bit(fld, fld_name,
6868 		      ctrl_fld, ctrl_bit2,
6869 		      EFER_LMA,
6870 		      "EFER_LMA");
6871 	test_efer_bit(fld, fld_name,
6872 		      ctrl_fld, ctrl_bit2,
6873 		      EFER_LME,
6874 		      "EFER_LME");
6875 }
6876 
6877 /*
6878  * If the 'load IA32_EFER' VM-exit control is 1, bits reserved in the
6879  * IA32_EFER MSR must be 0 in the field for that register. In addition,
6880  * the values of the LMA and LME bits in the field must each be that of
6881  * the 'host address-space size' VM-exit control.
6882  *
6883  *  [Intel SDM]
6884  */
6885 static void test_host_efer(void)
6886 {
6887 	test_efer(HOST_EFER, "HOST_EFER", EXI_CONTROLS,
6888 		  ctrl_exit_rev.clr & EXI_LOAD_EFER,
6889 		  EXI_HOST_64);
6890 }
6891 
6892 /*
6893  * If the 'load IA32_EFER' VM-enter control is 1, bits reserved in the
6894  * IA32_EFER MSR must be 0 in the field for that register. In addition,
6895  * the values of the LMA and LME bits in the field must each be that of
6896  * the 'IA32e-mode guest' VM-exit control.
6897  */
6898 static void test_guest_efer(void)
6899 {
6900 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER)) {
6901 		printf("\"Load-IA32-EFER\" entry control not supported\n");
6902 		return;
6903 	}
6904 
6905 	vmcs_write(GUEST_EFER, rdmsr(MSR_EFER));
6906 	test_efer(GUEST_EFER, "GUEST_EFER", ENT_CONTROLS,
6907 		  ctrl_enter_rev.clr & ENT_LOAD_EFER,
6908 		  ENT_GUEST_64);
6909 }
6910 
6911 /*
6912  * PAT values higher than 8 are uninteresting since they're likely lumped
6913  * in with "8". We only test values above 8 one bit at a time,
6914  * in order to reduce the number of VM-Entries and keep the runtime reasonable.
6915  */
6916 #define	PAT_VAL_LIMIT	8
6917 
6918 static void test_pat(u32 field, const char * field_name, u32 ctrl_field,
6919 		     u64 ctrl_bit)
6920 {
6921 	u32 ctrl_saved = vmcs_read(ctrl_field);
6922 	u64 pat_saved = vmcs_read(field);
6923 	u64 i, val;
6924 	u32 j;
6925 	int error;
6926 
6927 	vmcs_clear_bits(ctrl_field, ctrl_bit);
6928 
6929 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
6930 		/* Test PAT0..PAT7 fields */
6931 		for (j = 0; j < (i ? 8 : 1); j++) {
6932 			val = i << j * 8;
6933 			vmcs_write(field, val);
6934 			if (field == HOST_PAT) {
6935 				report_prefix_pushf("%s %lx", field_name, val);
6936 				test_vmx_vmlaunch(0);
6937 				report_prefix_pop();
6938 
6939 			} else {	// GUEST_PAT
6940 				enter_guest();
6941 				report_guest_state_test("ENT_LOAD_PAT enabled",
6942 							VMX_VMCALL, val,
6943 							"GUEST_PAT");
6944 			}
6945 		}
6946 	}
6947 
6948 	vmcs_set_bits(ctrl_field, ctrl_bit);
6949 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
6950 		/* Test PAT0..PAT7 fields */
6951 		for (j = 0; j < (i ? 8 : 1); j++) {
6952 			val = i << j * 8;
6953 			vmcs_write(field, val);
6954 
6955 			if (field == HOST_PAT) {
6956 				report_prefix_pushf("%s %lx", field_name, val);
6957 				if (i == 0x2 || i == 0x3 || i >= 0x8)
6958 					error =
6959 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
6960 				else
6961 					error = 0;
6962 
6963 				test_vmx_vmlaunch(error);
6964 				report_prefix_pop();
6965 
6966 			} else {	// GUEST_PAT
6967 				if (i == 0x2 || i == 0x3 || i >= 0x8) {
6968 					enter_guest_with_invalid_guest_state();
6969 					report_guest_state_test("ENT_LOAD_PAT "
6970 							        "enabled",
6971 							        VMX_FAIL_STATE | VMX_ENTRY_FAILURE,
6972 							        val,
6973 							        "GUEST_PAT");
6974 				} else {
6975 					enter_guest();
6976 					report_guest_state_test("ENT_LOAD_PAT "
6977 							        "enabled",
6978 							        VMX_VMCALL,
6979 							        val,
6980 							        "GUEST_PAT");
6981 				}
6982 			}
6983 
6984 		}
6985 	}
6986 
6987 	vmcs_write(ctrl_field, ctrl_saved);
6988 	vmcs_write(field, pat_saved);
6989 }
6990 
6991 /*
6992  *  If the "load IA32_PAT" VM-exit control is 1, the value of the field
6993  *  for the IA32_PAT MSR must be one that could be written by WRMSR
6994  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
6995  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
6996  *  6 (WB), or 7 (UC-).
6997  *
6998  *  [Intel SDM]
6999  */
7000 static void test_load_host_pat(void)
7001 {
7002 	/*
7003 	 * "load IA32_PAT" VM-exit control
7004 	 */
7005 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) {
7006 		printf("\"Load-IA32-PAT\" exit control not supported\n");
7007 		return;
7008 	}
7009 
7010 	test_pat(HOST_PAT, "HOST_PAT", EXI_CONTROLS, EXI_LOAD_PAT);
7011 }
7012 
7013 /*
7014  * test_vmcs_field - test a value for the given VMCS field
7015  * @field: VMCS field
7016  * @field_name: string name of VMCS field
7017  * @bit_start: starting bit
7018  * @bit_end: ending bit
7019  * @val: value that the bit range must or must not contain
7020  * @valid_val: whether value given in 'val' must be valid or not
7021  * @error: expected VMCS error when vmentry fails for an invalid value
7022  */
7023 static void test_vmcs_field(u64 field, const char *field_name, u32 bit_start,
7024 			    u32 bit_end, u64 val, bool valid_val, u32 error)
7025 {
7026 	u64 field_saved = vmcs_read(field);
7027 	u32 i;
7028 	u64 tmp;
7029 	u32 bit_on;
7030 	u64 mask = ~0ull;
7031 
7032 	mask = (mask >> bit_end) << bit_end;
7033 	mask = mask | ((1 << bit_start) - 1);
7034 	tmp = (field_saved & mask) | (val << bit_start);
7035 
7036 	vmcs_write(field, tmp);
7037 	report_prefix_pushf("%s %lx", field_name, tmp);
7038 	if (valid_val)
7039 		test_vmx_vmlaunch(0);
7040 	else
7041 		test_vmx_vmlaunch(error);
7042 	report_prefix_pop();
7043 
7044 	for (i = bit_start; i <= bit_end; i = i + 2) {
7045 		bit_on = ((1ull < i) & (val << bit_start)) ? 0 : 1;
7046 		if (bit_on)
7047 			tmp = field_saved | (1ull << i);
7048 		else
7049 			tmp = field_saved & ~(1ull << i);
7050 		vmcs_write(field, tmp);
7051 		report_prefix_pushf("%s %lx", field_name, tmp);
7052 		if (valid_val)
7053 			test_vmx_vmlaunch(error);
7054 		else
7055 			test_vmx_vmlaunch(0);
7056 		report_prefix_pop();
7057 	}
7058 
7059 	vmcs_write(field, field_saved);
7060 }
7061 
7062 static void test_canonical(u64 field, const char * field_name)
7063 {
7064 	u64 addr_saved = vmcs_read(field);
7065 
7066 	report_prefix_pushf("%s %lx", field_name, addr_saved);
7067 	if (is_canonical(addr_saved)) {
7068 		test_vmx_vmlaunch(0);
7069 		report_prefix_pop();
7070 
7071 		vmcs_write(field, NONCANONICAL);
7072 		report_prefix_pushf("%s %llx", field_name, NONCANONICAL);
7073 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7074 
7075 		vmcs_write(field, addr_saved);
7076 	} else {
7077 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7078 	}
7079 	report_prefix_pop();
7080 }
7081 
7082 #define TEST_RPL_TI_FLAGS(reg, name)				\
7083 	test_vmcs_field(reg, name, 0, 2, 0x0, true,		\
7084 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7085 
7086 #define TEST_CS_TR_FLAGS(reg, name)				\
7087 	test_vmcs_field(reg, name, 3, 15, 0x0000, false,	\
7088 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7089 
7090 /*
7091  * 1. In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the
7092  *    RPL (bits 1:0) and the TI flag (bit 2) must be 0.
7093  * 2. The selector fields for CS and TR cannot be 0000H.
7094  * 3. The selector field for SS cannot be 0000H if the "host address-space
7095  *    size" VM-exit control is 0.
7096  * 4. On processors that support Intel 64 architecture, the base-address
7097  *    fields for FS, GS and TR must contain canonical addresses.
7098  */
7099 static void test_host_segment_regs(void)
7100 {
7101 	u16 selector_saved;
7102 
7103 	/*
7104 	 * Test RPL and TI flags
7105 	 */
7106 	TEST_RPL_TI_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7107 	TEST_RPL_TI_FLAGS(HOST_SEL_SS, "HOST_SEL_SS");
7108 	TEST_RPL_TI_FLAGS(HOST_SEL_DS, "HOST_SEL_DS");
7109 	TEST_RPL_TI_FLAGS(HOST_SEL_ES, "HOST_SEL_ES");
7110 	TEST_RPL_TI_FLAGS(HOST_SEL_FS, "HOST_SEL_FS");
7111 	TEST_RPL_TI_FLAGS(HOST_SEL_GS, "HOST_SEL_GS");
7112 	TEST_RPL_TI_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7113 
7114 	/*
7115 	 * Test that CS and TR fields can not be 0x0000
7116 	 */
7117 	TEST_CS_TR_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7118 	TEST_CS_TR_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7119 
7120 	/*
7121 	 * SS field can not be 0x0000 if "host address-space size" VM-exit
7122 	 * control is 0
7123 	 */
7124 	selector_saved = vmcs_read(HOST_SEL_SS);
7125 	vmcs_write(HOST_SEL_SS, 0);
7126 	report_prefix_pushf("HOST_SEL_SS 0");
7127 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7128 		test_vmx_vmlaunch(0);
7129 	} else {
7130 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7131 	}
7132 	report_prefix_pop();
7133 
7134 	vmcs_write(HOST_SEL_SS, selector_saved);
7135 
7136 #ifdef __x86_64__
7137 	/*
7138 	 * Base address for FS, GS and TR must be canonical
7139 	 */
7140 	test_canonical(HOST_BASE_FS, "HOST_BASE_FS");
7141 	test_canonical(HOST_BASE_GS, "HOST_BASE_GS");
7142 	test_canonical(HOST_BASE_TR, "HOST_BASE_TR");
7143 #endif
7144 }
7145 
7146 /*
7147  *  On processors that support Intel 64 architecture, the base-address
7148  *  fields for GDTR and IDTR must contain canonical addresses.
7149  */
7150 static void test_host_desc_tables(void)
7151 {
7152 #ifdef __x86_64__
7153 	test_canonical(HOST_BASE_GDTR, "HOST_BASE_GDTR");
7154 	test_canonical(HOST_BASE_IDTR, "HOST_BASE_IDTR");
7155 #endif
7156 }
7157 
7158 /*
7159  * If the "host address-space size" VM-exit control is 0, the following must
7160  * hold:
7161  *    - The "IA-32e mode guest" VM-entry control is 0.
7162  *    - Bit 17 of the CR4 field (corresponding to CR4.PCIDE) is 0.
7163  *    - Bits 63:32 in the RIP field are 0.
7164  *
7165  * If the "host address-space size" VM-exit control is 1, the following must
7166  * hold:
7167  *    - Bit 5 of the CR4 field (corresponding to CR4.PAE) is 1.
7168  *    - The RIP field contains a canonical address.
7169  *
7170  */
7171 static void test_host_addr_size(void)
7172 {
7173 	u64 cr4_saved = vmcs_read(HOST_CR4);
7174 	u64 rip_saved = vmcs_read(HOST_RIP);
7175 	u64 entry_ctrl_saved = vmcs_read(ENT_CONTROLS);
7176 	int i;
7177 	u64 tmp;
7178 
7179 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7180 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7181 		report_prefix_pushf("\"IA-32e mode guest\" enabled");
7182 		test_vmx_vmlaunch(0);
7183 		report_prefix_pop();
7184 
7185 		vmcs_write(HOST_CR4, cr4_saved | X86_CR4_PCIDE);
7186 		report_prefix_pushf("\"CR4.PCIDE\" set");
7187 		test_vmx_vmlaunch(0);
7188 		report_prefix_pop();
7189 
7190 		for (i = 32; i <= 63; i = i + 4) {
7191 			tmp = rip_saved | 1ull << i;
7192 			vmcs_write(HOST_RIP, tmp);
7193 			report_prefix_pushf("HOST_RIP %lx", tmp);
7194 			test_vmx_vmlaunch(0);
7195 			report_prefix_pop();
7196 		}
7197 
7198 		if (cr4_saved & X86_CR4_PAE) {
7199 			vmcs_write(HOST_CR4, cr4_saved  & ~X86_CR4_PAE);
7200 			report_prefix_pushf("\"CR4.PAE\" unset");
7201 			test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7202 		} else {
7203 			report_prefix_pushf("\"CR4.PAE\" set");
7204 			test_vmx_vmlaunch(0);
7205 		}
7206 		report_prefix_pop();
7207 
7208 		vmcs_write(HOST_RIP, NONCANONICAL);
7209 		report_prefix_pushf("HOST_RIP %llx", NONCANONICAL);
7210 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7211 		report_prefix_pop();
7212 
7213 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7214 		vmcs_write(HOST_RIP, rip_saved);
7215 		vmcs_write(HOST_CR4, cr4_saved);
7216 	}
7217 }
7218 
7219 /*
7220  * Check that the virtual CPU checks the VMX Host State Area as
7221  * documented in the Intel SDM.
7222  */
7223 static void vmx_host_state_area_test(void)
7224 {
7225 	/*
7226 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
7227 	 * fail due to invalid guest state, should we make it that
7228 	 * far.
7229 	 */
7230 	vmcs_write(GUEST_RFLAGS, 0);
7231 
7232 	test_host_ctl_regs();
7233 
7234 	test_sysenter_field(HOST_SYSENTER_ESP, "HOST_SYSENTER_ESP");
7235 	test_sysenter_field(HOST_SYSENTER_EIP, "HOST_SYSENTER_EIP");
7236 
7237 	test_host_efer();
7238 	test_load_host_pat();
7239 	test_host_segment_regs();
7240 	test_host_desc_tables();
7241 	test_host_addr_size();
7242 }
7243 
7244 /*
7245  *  If the "load IA32_PAT" VM-entry control is 1, the value of the field
7246  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7247  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7248  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7249  *  6 (WB), or 7 (UC-).
7250  *
7251  *  [Intel SDM]
7252  */
7253 static void test_load_guest_pat(void)
7254 {
7255 	/*
7256 	 * "load IA32_PAT" VM-entry control
7257 	 */
7258 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
7259 		printf("\"Load-IA32-PAT\" entry control not supported\n");
7260 		return;
7261 	}
7262 
7263 	test_pat(GUEST_PAT, "GUEST_PAT", ENT_CONTROLS, ENT_LOAD_PAT);
7264 }
7265 
7266 /*
7267  * Check that the virtual CPU checks the VMX Guest State Area as
7268  * documented in the Intel SDM.
7269  */
7270 static void vmx_guest_state_area_test(void)
7271 {
7272 	vmx_set_test_stage(1);
7273 	test_set_guest(guest_state_test_main);
7274 
7275 	test_load_guest_pat();
7276 	test_guest_efer();
7277 
7278 	/*
7279 	 * Let the guest finish execution
7280 	 */
7281 	vmx_set_test_stage(2);
7282 	enter_guest();
7283 }
7284 
7285 static bool valid_vmcs_for_vmentry(void)
7286 {
7287 	struct vmcs *current_vmcs = NULL;
7288 
7289 	if (vmcs_save(&current_vmcs))
7290 		return false;
7291 
7292 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
7293 }
7294 
7295 static void try_vmentry_in_movss_shadow(void)
7296 {
7297 	u32 vm_inst_err;
7298 	u32 flags;
7299 	bool early_failure = false;
7300 	u32 expected_flags = X86_EFLAGS_FIXED;
7301 	bool valid_vmcs = valid_vmcs_for_vmentry();
7302 
7303 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
7304 
7305 	/*
7306 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
7307 	 * unsupported VMCS component").
7308 	 */
7309 	vmcs_write(~0u, 0);
7310 
7311 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
7312 			      "vmwrite %%rsp, %%rdx;"
7313 			      "mov 0f, %%rax;"
7314 			      "mov %[host_rip], %%edx;"
7315 			      "vmwrite %%rax, %%rdx;"
7316 			      "mov $-1, %%ah;"
7317 			      "sahf;"
7318 			      "mov %%ss, %%ax;"
7319 			      "mov %%ax, %%ss;"
7320 			      "vmlaunch;"
7321 			      "mov $1, %[early_failure];"
7322 			      "0: lahf;"
7323 			      "movzbl %%ah, %[flags]"
7324 			      : [early_failure] "+r" (early_failure),
7325 				[flags] "=&a" (flags)
7326 			      : [host_rsp] "i" (HOST_RSP),
7327 				[host_rip] "i" (HOST_RIP)
7328 			      : "rdx", "cc", "memory");
7329 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
7330 
7331 	report("Early VM-entry failure", early_failure);
7332 	report("RFLAGS[8:0] is %x (actual %x)", flags == expected_flags,
7333 	       expected_flags, flags);
7334 	if (valid_vmcs)
7335 		report("VM-instruction error is %d (actual %d)",
7336 		       vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
7337 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
7338 }
7339 
7340 static void vmentry_movss_shadow_test(void)
7341 {
7342 	struct vmcs *orig_vmcs;
7343 
7344 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
7345 
7346 	/*
7347 	 * Set the launched flag on the current VMCS to verify the correct
7348 	 * error priority, below.
7349 	 */
7350 	test_set_guest(v2_null_test_guest);
7351 	enter_guest();
7352 
7353 	/*
7354 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
7355 	 * fail due to invalid guest state (if we make it that far).
7356 	 */
7357 	vmcs_write(GUEST_RFLAGS, 0);
7358 
7359 	/*
7360 	 * "VM entry with events blocked by MOV SS" takes precedence over
7361 	 * "VMLAUNCH with non-clear VMCS."
7362 	 */
7363 	report_prefix_push("valid current-VMCS");
7364 	try_vmentry_in_movss_shadow();
7365 	report_prefix_pop();
7366 
7367 	/*
7368 	 * VMfailInvalid takes precedence over "VM entry with events
7369 	 * blocked by MOV SS."
7370 	 */
7371 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
7372 	report_prefix_push("no current-VMCS");
7373 	try_vmentry_in_movss_shadow();
7374 	report_prefix_pop();
7375 
7376 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
7377 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
7378 }
7379 
7380 static void vmx_cr_load_test(void)
7381 {
7382 	unsigned long cr3, cr4, orig_cr3, orig_cr4;
7383 
7384 	orig_cr4 = read_cr4();
7385 	orig_cr3 = read_cr3();
7386 
7387 	if (!this_cpu_has(X86_FEATURE_PCID)) {
7388 		report_skip("PCID not detected");
7389 		return;
7390 	}
7391 	if (!this_cpu_has(X86_FEATURE_MCE)) {
7392 		report_skip("MCE not detected");
7393 		return;
7394 	}
7395 
7396 	TEST_ASSERT(!(orig_cr3 & X86_CR3_PCID_MASK));
7397 
7398 	/* Enable PCID for L1. */
7399 	cr4 = orig_cr4 | X86_CR4_PCIDE;
7400 	cr3 = orig_cr3 | 0x1;
7401 	TEST_ASSERT(!write_cr4_checking(cr4));
7402 	write_cr3(cr3);
7403 
7404 	test_set_guest(v2_null_test_guest);
7405 	vmcs_write(HOST_CR4, cr4);
7406 	vmcs_write(HOST_CR3, cr3);
7407 	enter_guest();
7408 
7409 	/*
7410 	 * No exception is expected.
7411 	 *
7412 	 * NB. KVM loads the last guest write to CR4 into CR4 read
7413 	 *     shadow. In order to trigger an exit to KVM, we can toggle a
7414 	 *     bit that is owned by KVM. We use CR4.MCE, which shall
7415 	 *     have no side effect because normally no guest MCE (e.g., as the
7416 	 *     result of bad memory) would happen during this test.
7417 	 */
7418 	TEST_ASSERT(!write_cr4_checking(cr4 ^ X86_CR4_MCE));
7419 
7420 	/* Cleanup L1 state. */
7421 	write_cr3(orig_cr3);
7422 	TEST_ASSERT(!write_cr4_checking(orig_cr4));
7423 }
7424 
7425 static void vmx_nm_test_guest(void)
7426 {
7427 	write_cr0(read_cr0() | X86_CR0_TS);
7428 	asm volatile("fnop");
7429 }
7430 
7431 static void check_nm_exit(const char *test)
7432 {
7433 	u32 reason = vmcs_read(EXI_REASON);
7434 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
7435 	const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
7436 		NM_VECTOR;
7437 
7438 	report("%s", reason == VMX_EXC_NMI && intr_info == expected, test);
7439 }
7440 
7441 /*
7442  * This test checks that:
7443  *
7444  * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then
7445  *     a subsequent #NM VM-exit is reflected to L1.
7446  *
7447  * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a
7448  *     subsequent #NM VM-exit is reflected to L1.
7449  */
7450 static void vmx_nm_test(void)
7451 {
7452 	unsigned long cr0 = read_cr0();
7453 
7454 	test_set_guest(vmx_nm_test_guest);
7455 
7456 	/*
7457 	 * L1 wants to intercept #NM exceptions encountered in L2.
7458 	 */
7459 	vmcs_write(EXC_BITMAP, 1 << NM_VECTOR);
7460 
7461 	/*
7462 	 * Launch L2 with CR0.TS clear, but don't claim host ownership of
7463 	 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop,
7464 	 * which will raise #NM. L0 should reflect the #NM VM-exit to L1.
7465 	 */
7466 	vmcs_write(CR0_MASK, 0);
7467 	vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS);
7468 	enter_guest();
7469 	check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1");
7470 
7471 	/*
7472 	 * Re-enter L2 at the fnop instruction, with CR0.TS clear but
7473 	 * CR0.EM set. The fnop will still raise #NM, and L0 should
7474 	 * reflect the #NM VM-exit to L1.
7475 	 */
7476 	vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM);
7477 	enter_guest();
7478 	check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1");
7479 
7480 	/*
7481 	 * Re-enter L2 at the fnop instruction, with both CR0.TS and
7482 	 * CR0.EM clear. There will be no #NM, and the L2 guest should
7483 	 * exit normally.
7484 	 */
7485 	vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM));
7486 	enter_guest();
7487 }
7488 
7489 bool vmx_pending_event_ipi_fired;
7490 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
7491 {
7492 	vmx_pending_event_ipi_fired = true;
7493 	eoi();
7494 }
7495 
7496 bool vmx_pending_event_guest_run;
7497 static void vmx_pending_event_guest(void)
7498 {
7499 	vmcall();
7500 	vmx_pending_event_guest_run = true;
7501 }
7502 
7503 static void vmx_pending_event_test_core(bool guest_hlt)
7504 {
7505 	int ipi_vector = 0xf1;
7506 
7507 	vmx_pending_event_ipi_fired = false;
7508 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
7509 
7510 	vmx_pending_event_guest_run = false;
7511 	test_set_guest(vmx_pending_event_guest);
7512 
7513 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
7514 
7515 	enter_guest();
7516 	skip_exit_vmcall();
7517 
7518 	if (guest_hlt)
7519 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7520 
7521 	irq_disable();
7522 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
7523 				   APIC_DM_FIXED | ipi_vector,
7524 				   0);
7525 
7526 	enter_guest();
7527 
7528 	assert_exit_reason(VMX_EXTINT);
7529 	report("Guest did not run before host received IPI",
7530 		   !vmx_pending_event_guest_run);
7531 
7532 	irq_enable();
7533 	asm volatile ("nop");
7534 	irq_disable();
7535 	report("Got pending interrupt after IRQ enabled",
7536 		   vmx_pending_event_ipi_fired);
7537 
7538 	if (guest_hlt)
7539 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
7540 
7541 	enter_guest();
7542 	report("Guest finished running when no interrupt",
7543 		   vmx_pending_event_guest_run);
7544 }
7545 
7546 static void vmx_pending_event_test(void)
7547 {
7548 	vmx_pending_event_test_core(false);
7549 }
7550 
7551 static void vmx_pending_event_hlt_test(void)
7552 {
7553 	vmx_pending_event_test_core(true);
7554 }
7555 
7556 static int vmx_window_test_db_count;
7557 
7558 static void vmx_window_test_db_handler(struct ex_regs *regs)
7559 {
7560 	vmx_window_test_db_count++;
7561 }
7562 
7563 static void vmx_nmi_window_test_guest(void)
7564 {
7565 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
7566 
7567 	asm volatile("vmcall\n\t"
7568 		     "nop\n\t");
7569 
7570 	handle_exception(DB_VECTOR, NULL);
7571 }
7572 
7573 static void verify_nmi_window_exit(u64 rip)
7574 {
7575 	u32 exit_reason = vmcs_read(EXI_REASON);
7576 
7577 	report("Exit reason (%d) is 'NMI window'",
7578 	       exit_reason == VMX_NMI_WINDOW, exit_reason);
7579 	report("RIP (%#lx) is %#lx", vmcs_read(GUEST_RIP) == rip,
7580 	       vmcs_read(GUEST_RIP), rip);
7581 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
7582 }
7583 
7584 static void vmx_nmi_window_test(void)
7585 {
7586 	u64 nop_addr;
7587 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
7588 
7589 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
7590 		report_skip("CPU does not support the \"Virtual NMIs\" VM-execution control.");
7591 		return;
7592 	}
7593 
7594 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
7595 		report_skip("CPU does not support the \"NMI-window exiting\" VM-execution control.");
7596 		return;
7597 	}
7598 
7599 	vmx_window_test_db_count = 0;
7600 
7601 	report_prefix_push("NMI-window");
7602 	test_set_guest(vmx_nmi_window_test_guest);
7603 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
7604 	enter_guest();
7605 	skip_exit_vmcall();
7606 	nop_addr = vmcs_read(GUEST_RIP);
7607 
7608 	/*
7609 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
7610 	 * RIP will not advance.
7611 	 */
7612 	report_prefix_push("active, no blocking");
7613 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
7614 	enter_guest();
7615 	verify_nmi_window_exit(nop_addr);
7616 	report_prefix_pop();
7617 
7618 	/*
7619 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
7620 	 * a VM-exit on the next instruction after the nop. (The nop
7621 	 * is one byte.)
7622 	 */
7623 	report_prefix_push("active, blocking by MOV-SS");
7624 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
7625 	enter_guest();
7626 	verify_nmi_window_exit(nop_addr + 1);
7627 	report_prefix_pop();
7628 
7629 	/*
7630 	 * Ask for "NMI-window exiting" (with event injection), and
7631 	 * expect a VM-exit after the event is injected. (RIP should
7632 	 * be at the address specified in the IDT entry for #DB.)
7633 	 */
7634 	report_prefix_push("active, no blocking, injecting #DB");
7635 	vmcs_write(ENT_INTR_INFO,
7636 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
7637 	enter_guest();
7638 	verify_nmi_window_exit((u64)db_fault_addr);
7639 	report_prefix_pop();
7640 
7641 	/*
7642 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
7643 	 * a VM-exit after the next IRET (i.e. after the #DB handler
7644 	 * returns). So, RIP should be back at one byte past the nop.
7645 	 */
7646 	report_prefix_push("active, blocking by NMI");
7647 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
7648 	enter_guest();
7649 	verify_nmi_window_exit(nop_addr + 1);
7650 	report("#DB handler executed once (actual %d times)",
7651 	       vmx_window_test_db_count == 1,
7652 	       vmx_window_test_db_count);
7653 	report_prefix_pop();
7654 
7655 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
7656 		report_skip("CPU does not support activity state HLT.");
7657 	} else {
7658 		/*
7659 		 * Ask for "NMI-window exiting" when entering activity
7660 		 * state HLT, and expect an immediate VM-exit. RIP is
7661 		 * still one byte past the nop.
7662 		 */
7663 		report_prefix_push("halted, no blocking");
7664 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7665 		enter_guest();
7666 		verify_nmi_window_exit(nop_addr + 1);
7667 		report_prefix_pop();
7668 
7669 		/*
7670 		 * Ask for "NMI-window exiting" when entering activity
7671 		 * state HLT (with event injection), and expect a
7672 		 * VM-exit after the event is injected. (RIP should be
7673 		 * at the address specified in the IDT entry for #DB.)
7674 		 */
7675 		report_prefix_push("halted, no blocking, injecting #DB");
7676 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7677 		vmcs_write(ENT_INTR_INFO,
7678 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
7679 			   DB_VECTOR);
7680 		enter_guest();
7681 		verify_nmi_window_exit((u64)db_fault_addr);
7682 		report_prefix_pop();
7683 	}
7684 
7685 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
7686 	enter_guest();
7687 	report_prefix_pop();
7688 }
7689 
7690 static void vmx_intr_window_test_guest(void)
7691 {
7692 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
7693 
7694 	/*
7695 	 * The two consecutive STIs are to ensure that only the first
7696 	 * one has a shadow. Note that NOP and STI are one byte
7697 	 * instructions.
7698 	 */
7699 	asm volatile("vmcall\n\t"
7700 		     "nop\n\t"
7701 		     "sti\n\t"
7702 		     "sti\n\t");
7703 
7704 	handle_exception(DB_VECTOR, NULL);
7705 }
7706 
7707 static void verify_intr_window_exit(u64 rip)
7708 {
7709 	u32 exit_reason = vmcs_read(EXI_REASON);
7710 
7711 	report("Exit reason (%d) is 'interrupt window'",
7712 	       exit_reason == VMX_INTR_WINDOW, exit_reason);
7713 	report("RIP (%#lx) is %#lx", vmcs_read(GUEST_RIP) == rip,
7714 	       vmcs_read(GUEST_RIP), rip);
7715 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
7716 }
7717 
7718 static void vmx_intr_window_test(void)
7719 {
7720 	u64 vmcall_addr;
7721 	u64 nop_addr;
7722 	unsigned int orig_db_gate_type;
7723 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
7724 
7725 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
7726 		report_skip("CPU does not support the \"interrupt-window exiting\" VM-execution control.");
7727 		return;
7728 	}
7729 
7730 	/*
7731 	 * Change the IDT entry for #DB from interrupt gate to trap gate,
7732 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
7733 	 * be disabled after vectoring a #DB.
7734 	 */
7735 	orig_db_gate_type = boot_idt[DB_VECTOR].type;
7736 	boot_idt[DB_VECTOR].type = 15;
7737 
7738 	report_prefix_push("interrupt-window");
7739 	test_set_guest(vmx_intr_window_test_guest);
7740 	enter_guest();
7741 	assert_exit_reason(VMX_VMCALL);
7742 	vmcall_addr = vmcs_read(GUEST_RIP);
7743 
7744 	/*
7745 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
7746 	 * no blocking; expect an immediate VM-exit. Note that we have
7747 	 * not advanced past the vmcall instruction yet, so RIP should
7748 	 * point to the vmcall instruction.
7749 	 */
7750 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
7751 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
7752 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
7753 	enter_guest();
7754 	verify_intr_window_exit(vmcall_addr);
7755 	report_prefix_pop();
7756 
7757 	/*
7758 	 * Ask for "interrupt-window exiting" (with event injection)
7759 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
7760 	 * the event is injected. That is, RIP should should be at the
7761 	 * address specified in the IDT entry for #DB.
7762 	 */
7763 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #DB");
7764 	vmcs_write(ENT_INTR_INFO,
7765 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
7766 	vmcall_addr = vmcs_read(GUEST_RIP);
7767 	enter_guest();
7768 	verify_intr_window_exit((u64)db_fault_addr);
7769 	report_prefix_pop();
7770 
7771 	/*
7772 	 * Let the L2 guest run through the IRET, back to the VMCALL.
7773 	 * We have to clear the "interrupt-window exiting"
7774 	 * VM-execution control, or it would just keep causing
7775 	 * VM-exits. Then, advance past the VMCALL and set the
7776 	 * "interrupt-window exiting" VM-execution control again.
7777 	 */
7778 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
7779 	enter_guest();
7780 	skip_exit_vmcall();
7781 	nop_addr = vmcs_read(GUEST_RIP);
7782 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
7783 
7784 	/*
7785 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
7786 	 * RFLAGS.IF set, and expect a VM-exit on the next
7787 	 * instruction. (NOP is one byte.)
7788 	 */
7789 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
7790 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
7791 	enter_guest();
7792 	verify_intr_window_exit(nop_addr + 1);
7793 	report_prefix_pop();
7794 
7795 	/*
7796 	 * Back up to the NOP and ask for "interrupt-window exiting"
7797 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
7798 	 * on the next instruction. (NOP is one byte.)
7799 	 */
7800 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
7801 	vmcs_write(GUEST_RIP, nop_addr);
7802 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
7803 	enter_guest();
7804 	verify_intr_window_exit(nop_addr + 1);
7805 	report_prefix_pop();
7806 
7807 	/*
7808 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
7809 	 * and expect a VM-exit on the instruction following the STI
7810 	 * shadow. Only the first STI (which is one byte past the NOP)
7811 	 * should have a shadow. The second STI (which is two bytes
7812 	 * past the NOP) has no shadow. Therefore, the interrupt
7813 	 * window opens at three bytes past the NOP.
7814 	 */
7815 	report_prefix_push("active, RFLAGS.IF = 0");
7816 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
7817 	enter_guest();
7818 	verify_intr_window_exit(nop_addr + 3);
7819 	report_prefix_pop();
7820 
7821 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
7822 		report_skip("CPU does not support activity state HLT.");
7823 	} else {
7824 		/*
7825 		 * Ask for "interrupt-window exiting" when entering
7826 		 * activity state HLT, and expect an immediate
7827 		 * VM-exit. RIP is still three bytes past the nop.
7828 		 */
7829 		report_prefix_push("halted, no blocking");
7830 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7831 		enter_guest();
7832 		verify_intr_window_exit(nop_addr + 3);
7833 		report_prefix_pop();
7834 
7835 		/*
7836 		 * Ask for "interrupt-window exiting" when entering
7837 		 * activity state HLT (with event injection), and
7838 		 * expect a VM-exit after the event is injected. That
7839 		 * is, RIP should should be at the address specified
7840 		 * in the IDT entry for #DB.
7841 		 */
7842 		report_prefix_push("halted, no blocking, injecting #DB");
7843 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7844 		vmcs_write(ENT_INTR_INFO,
7845 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
7846 			   DB_VECTOR);
7847 		enter_guest();
7848 		verify_intr_window_exit((u64)db_fault_addr);
7849 		report_prefix_pop();
7850 	}
7851 
7852 	boot_idt[DB_VECTOR].type = orig_db_gate_type;
7853 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
7854 	enter_guest();
7855 	report_prefix_pop();
7856 }
7857 
7858 #define GUEST_TSC_OFFSET (1u << 30)
7859 
7860 static u64 guest_tsc;
7861 
7862 static void vmx_store_tsc_test_guest(void)
7863 {
7864 	guest_tsc = rdtsc();
7865 }
7866 
7867 /*
7868  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
7869  * list, the value saved is not subject to the TSC offset that is
7870  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
7871  */
7872 static void vmx_store_tsc_test(void)
7873 {
7874 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
7875 	u64 low, high;
7876 
7877 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
7878 		report_skip("'Use TSC offsetting' not supported");
7879 		return;
7880 	}
7881 
7882 	test_set_guest(vmx_store_tsc_test_guest);
7883 
7884 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
7885 	vmcs_write(EXI_MSR_ST_CNT, 1);
7886 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
7887 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
7888 
7889 	low = rdtsc();
7890 	enter_guest();
7891 	high = rdtsc();
7892 
7893 	report("RDTSC value in the guest (%lu) is in range [%lu, %lu]",
7894 	       low + GUEST_TSC_OFFSET <= guest_tsc &&
7895 	       guest_tsc <= high + GUEST_TSC_OFFSET,
7896 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
7897 	report("IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
7898 	       low <= msr_entry.value && msr_entry.value <= high,
7899 	       msr_entry.value, low, high);
7900 }
7901 
7902 static void vmx_db_test_guest(void)
7903 {
7904 	/*
7905 	 * For a hardware generated single-step #DB.
7906 	 */
7907 	asm volatile("vmcall;"
7908 		     "nop;"
7909 		     ".Lpost_nop:");
7910 	/*
7911 	 * ...in a MOVSS shadow, with pending debug exceptions.
7912 	 */
7913 	asm volatile("vmcall;"
7914 		     "nop;"
7915 		     ".Lpost_movss_nop:");
7916 	/*
7917 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
7918 	 * emulates it in software.)
7919 	 */
7920 	asm volatile("vmcall;"
7921 		     "wbinvd;"
7922 		     ".Lpost_wbinvd:");
7923 	/*
7924 	 * ...in a MOVSS shadow, with pending debug exceptions.
7925 	 */
7926 	asm volatile("vmcall;"
7927 		     "wbinvd;"
7928 		     ".Lpost_movss_wbinvd:");
7929 	/*
7930 	 * For a hardware generated single-step #DB in a transactional region.
7931 	 */
7932 	asm volatile("vmcall;"
7933 		     ".Lxbegin: xbegin .Lskip_rtm;"
7934 		     "xend;"
7935 		     ".Lskip_rtm:");
7936 }
7937 
7938 /*
7939  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
7940  * L2. No #DB is delivered and L2 continues to the next point of
7941  * interest.
7942  */
7943 static void dismiss_db(void)
7944 {
7945 	vmcs_write(GUEST_PENDING_DEBUG, 0);
7946 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
7947 	enter_guest();
7948 }
7949 
7950 /*
7951  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
7952  * Then throw away the #DB exception and resume L2.
7953  */
7954 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
7955 			  void *expected_rip, u64 expected_exit_qual,
7956 			  u64 expected_dr6)
7957 {
7958 	u32 reason = vmcs_read(EXI_REASON);
7959 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
7960 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
7961 	u64 guest_rip = vmcs_read(GUEST_RIP);
7962 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
7963 	u64 dr6 = read_dr6();
7964 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
7965 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
7966 
7967 	report("Expected #DB VM-exit",
7968 	       reason == VMX_EXC_NMI && intr_info == expected_intr_info);
7969 	report("Expected RIP %p (actual %lx)", (u64)expected_rip == guest_rip,
7970 	       expected_rip, guest_rip);
7971 	report_xfail("Expected pending debug exceptions 0 (actual %lx)",
7972 		     xfail_pdbg, 0 == guest_pending_dbg, guest_pending_dbg);
7973 	report_xfail("Expected exit qualification %lx (actual %lx)", xfail_qual,
7974 		     expected_exit_qual == exit_qual,
7975 		     expected_exit_qual, exit_qual);
7976 	report_xfail("Expected DR6 %lx (actual %lx)", xfail_dr6,
7977 		     expected_dr6 == dr6, expected_dr6, dr6);
7978 	dismiss_db();
7979 }
7980 
7981 /*
7982  * Assuming the guest has just exited on a VMCALL instruction, skip
7983  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
7984  * pending debug exceptions are non-zero, set the VMCS up as if the
7985  * previous instruction was a MOVSS that generated the indicated
7986  * pending debug exceptions. Then enter L2.
7987  */
7988 static void single_step_guest(const char *test_name, u64 starting_dr6,
7989 			      u64 pending_debug_exceptions)
7990 {
7991 	printf("\n%s\n", test_name);
7992 	skip_exit_vmcall();
7993 	write_dr6(starting_dr6);
7994 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
7995 	if (pending_debug_exceptions) {
7996 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
7997 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
7998 	}
7999 	enter_guest();
8000 }
8001 
8002 /*
8003  * When L1 intercepts #DB, verify that a single-step trap clears
8004  * pending debug exceptions, populates the exit qualification field
8005  * properly, and that DR6 is not prematurely clobbered. In a
8006  * (simulated) MOVSS shadow, make sure that the pending debug
8007  * exception bits are properly accumulated into the exit qualification
8008  * field.
8009  */
8010 static void vmx_db_test(void)
8011 {
8012 	/*
8013 	 * We are going to set a few arbitrary bits in DR6 to verify that
8014 	 * (a) DR6 is not modified by an intercepted #DB, and
8015 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
8016          *     the exit qualification field for a subsequent #DB exception.
8017 	 */
8018 	const u64 starting_dr6 = DR6_RESERVED | BIT(13) | DR_TRAP3 | DR_TRAP1;
8019 	extern char post_nop asm(".Lpost_nop");
8020 	extern char post_movss_nop asm(".Lpost_movss_nop");
8021 	extern char post_wbinvd asm(".Lpost_wbinvd");
8022 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
8023 	extern char xbegin asm(".Lxbegin");
8024 	extern char skip_rtm asm(".Lskip_rtm");
8025 
8026 	/*
8027 	 * L1 wants to intercept #DB exceptions encountered in L2.
8028 	 */
8029 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
8030 
8031 	/*
8032 	 * Start L2 and run it up to the first point of interest.
8033 	 */
8034 	test_set_guest(vmx_db_test_guest);
8035 	enter_guest();
8036 
8037 	/*
8038 	 * Hardware-delivered #DB trap for single-step sets the
8039 	 * standard that L0 has to follow for emulated instructions.
8040 	 */
8041 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
8042 	check_db_exit(false, false, false, &post_nop, DR_STEP, starting_dr6);
8043 
8044 	/*
8045 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
8046 	 * also sets the standard that L0 has to follow for emulated
8047 	 * instructions. Here, we establish the VMCS pending debug
8048 	 * exceptions to indicate that the simulated MOVSS triggered a
8049 	 * data breakpoint as well as the single-step trap.
8050 	 */
8051 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
8052 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0 );
8053 	check_db_exit(false, false, false, &post_movss_nop, DR_STEP | DR_TRAP0,
8054 		      starting_dr6);
8055 
8056 	/*
8057 	 * L0 synthesized #DB trap for single-step is buggy, because
8058 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
8059 	 * reconstitute the exit qualification from the prematurely
8060 	 * modified DR6, but fails miserably.
8061 	 */
8062 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
8063 	check_db_exit(true, true, false, &post_wbinvd, DR_STEP, starting_dr6);
8064 
8065 	/*
8066 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
8067 	 * even worse, because L0 also leaves the pending debug
8068 	 * exceptions in the VMCS instead of accumulating them into
8069 	 * the exit qualification field for the #DB exception.
8070 	 */
8071 	single_step_guest("Software synthesized single-step in MOVSS shadow",
8072 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0);
8073 	check_db_exit(true, true, true, &post_movss_wbinvd, DR_STEP | DR_TRAP0,
8074 		      starting_dr6);
8075 
8076 	/*
8077 	 * Optional RTM test for hardware that supports RTM, to
8078 	 * demonstrate that the current volume 3 of the SDM
8079 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
8080 	 * qualification for debug exceptions is not reserved. It is
8081 	 * set to 1 if a debug exception (#DB) or a breakpoint
8082 	 * exception (#BP) occurs inside an RTM region while advanced
8083 	 * debugging of RTM transactional regions is enabled.
8084 	 */
8085 	if (this_cpu_has(X86_FEATURE_RTM)) {
8086 		vmcs_write(ENT_CONTROLS,
8087 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
8088 		/*
8089 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
8090 		 * in the guest to enable advanced debugging of RTM
8091 		 * transactional regions.
8092 		 */
8093 		vmcs_write(GUEST_DR7, BIT(11));
8094 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
8095 		single_step_guest("Hardware delivered single-step in "
8096 				  "transactional region", starting_dr6, 0);
8097 		check_db_exit(false, false, false, &xbegin, BIT(16),
8098 			      starting_dr6);
8099 	} else {
8100 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
8101 		enter_guest();
8102 	}
8103 }
8104 
8105 static void enable_vid(void)
8106 {
8107 	void *virtual_apic_page;
8108 
8109 	assert(cpu_has_apicv());
8110 
8111 	disable_intercept_for_x2apic_msrs();
8112 
8113 	virtual_apic_page = alloc_page();
8114 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
8115 
8116 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
8117 
8118 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
8119 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
8120 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
8121 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
8122 
8123 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
8124 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
8125 }
8126 
8127 static void trigger_ioapic_scan_thread(void *data)
8128 {
8129 	/* Wait until other CPU entered L2 */
8130 	while (vmx_get_test_stage() != 1)
8131 		;
8132 
8133 	/* Trigger ioapic scan */
8134 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
8135 	vmx_set_test_stage(2);
8136 }
8137 
8138 static void irq_79_handler_guest(isr_regs_t *regs)
8139 {
8140 	eoi();
8141 
8142 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
8143 	vmcall();
8144 }
8145 
8146 /*
8147  * Constant for num of busy-loop iterations after which
8148  * a timer interrupt should have happened in host
8149  */
8150 #define TIMER_INTERRUPT_DELAY 100000000
8151 
8152 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
8153 {
8154 	handle_irq(0x79, irq_79_handler_guest);
8155 	irq_enable();
8156 
8157 	/* Signal to L1 CPU to trigger ioapic scan */
8158 	vmx_set_test_stage(1);
8159 	/* Wait until L1 CPU to trigger ioapic scan */
8160 	while (vmx_get_test_stage() != 2)
8161 		;
8162 
8163 	/*
8164 	 * Wait for L0 timer interrupt to be raised while we run in L2
8165 	 * such that L0 will process the IOAPIC scan request before
8166 	 * resuming L2
8167 	 */
8168 	delay(TIMER_INTERRUPT_DELAY);
8169 
8170 	asm volatile ("int $0x79");
8171 }
8172 
8173 static void vmx_eoi_bitmap_ioapic_scan_test(void)
8174 {
8175 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
8176 		report_skip(__func__);
8177 		return;
8178 	}
8179 
8180 	enable_vid();
8181 
8182 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
8183 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
8184 
8185 	/*
8186 	 * Launch L2.
8187 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
8188 	 * In case the reason isn't VMX_VMCALL, the asserion inside
8189 	 * skip_exit_vmcall() will fail.
8190 	 */
8191 	enter_guest();
8192 	skip_exit_vmcall();
8193 
8194 	/* Let L2 finish */
8195 	enter_guest();
8196 	report(__func__, 1);
8197 }
8198 
8199 #define HLT_WITH_RVI_VECTOR		(0xf1)
8200 
8201 bool vmx_hlt_with_rvi_guest_isr_fired;
8202 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
8203 {
8204 	vmx_hlt_with_rvi_guest_isr_fired = true;
8205 	eoi();
8206 }
8207 
8208 static void vmx_hlt_with_rvi_guest(void)
8209 {
8210 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
8211 
8212 	irq_enable();
8213 	asm volatile ("nop");
8214 
8215 	vmcall();
8216 }
8217 
8218 static void vmx_hlt_with_rvi_test(void)
8219 {
8220 	if (!cpu_has_apicv()) {
8221 		report_skip(__func__);
8222 		return;
8223 	}
8224 
8225 	enable_vid();
8226 
8227 	vmx_hlt_with_rvi_guest_isr_fired = false;
8228 	test_set_guest(vmx_hlt_with_rvi_guest);
8229 
8230 	enter_guest();
8231 	skip_exit_vmcall();
8232 
8233 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8234 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
8235 	enter_guest();
8236 
8237 	report("Interrupt raised in guest", vmx_hlt_with_rvi_guest_isr_fired);
8238 }
8239 
8240 static void set_irq_line_thread(void *data)
8241 {
8242 	/* Wait until other CPU entered L2 */
8243 	while (vmx_get_test_stage() != 1)
8244 		;
8245 
8246 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
8247 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
8248 	vmx_set_test_stage(2);
8249 }
8250 
8251 static bool irq_78_handler_vmcall_before_eoi;
8252 static void irq_78_handler_guest(isr_regs_t *regs)
8253 {
8254 	set_irq_line(0xf, 0);
8255 	if (irq_78_handler_vmcall_before_eoi)
8256 		vmcall();
8257 	eoi();
8258 	vmcall();
8259 }
8260 
8261 static void vmx_apic_passthrough_guest(void)
8262 {
8263 	handle_irq(0x78, irq_78_handler_guest);
8264 	irq_enable();
8265 
8266 	/* If requested, wait for other CPU to trigger ioapic scan */
8267 	if (vmx_get_test_stage() < 1) {
8268 		vmx_set_test_stage(1);
8269 		while (vmx_get_test_stage() != 2)
8270 			;
8271 	}
8272 
8273 	set_irq_line(0xf, 1);
8274 }
8275 
8276 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
8277 {
8278 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
8279 		report_skip(__func__);
8280 		return;
8281 	}
8282 
8283 	/* Test device is required for generating IRQs */
8284 	if (!test_device_enabled()) {
8285 		report_skip(__func__);
8286 		return;
8287 	}
8288 	u64 cpu_ctrl_0 = CPU_SECONDARY;
8289 	u64 cpu_ctrl_1 = 0;
8290 
8291 	disable_intercept_for_x2apic_msrs();
8292 
8293 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
8294 
8295 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
8296 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
8297 
8298 	if (set_irq_line_from_thread) {
8299 		irq_78_handler_vmcall_before_eoi = false;
8300 		on_cpu_async(1, set_irq_line_thread, NULL);
8301 	} else {
8302 		irq_78_handler_vmcall_before_eoi = true;
8303 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
8304 		vmx_set_test_stage(2);
8305 	}
8306 	test_set_guest(vmx_apic_passthrough_guest);
8307 
8308 	if (irq_78_handler_vmcall_before_eoi) {
8309 		/* Before EOI remote_irr should still be set */
8310 		enter_guest();
8311 		skip_exit_vmcall();
8312 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
8313 			"IOAPIC pass-through: remote_irr=1 before EOI");
8314 	}
8315 
8316 	/* After EOI remote_irr should be cleared */
8317 	enter_guest();
8318 	skip_exit_vmcall();
8319 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
8320 		"IOAPIC pass-through: remote_irr=0 after EOI");
8321 
8322 	/* Let L2 finish */
8323 	enter_guest();
8324 	report(__func__, 1);
8325 }
8326 
8327 static void vmx_apic_passthrough_test(void)
8328 {
8329 	vmx_apic_passthrough(false);
8330 }
8331 
8332 static void vmx_apic_passthrough_thread_test(void)
8333 {
8334 	vmx_apic_passthrough(true);
8335 }
8336 
8337 static u64 init_signal_test_exit_reason;
8338 static bool init_signal_test_thread_continued;
8339 
8340 static void init_signal_test_thread(void *data)
8341 {
8342 	struct vmcs *test_vmcs = data;
8343 
8344 	/* Enter VMX operation (i.e. exec VMXON) */
8345 	u64 *ap_vmxon_region = alloc_page();
8346 	enable_vmx();
8347 	init_vmx(ap_vmxon_region);
8348 	_vmx_on(ap_vmxon_region);
8349 
8350 	/* Signal CPU have entered VMX operation */
8351 	vmx_set_test_stage(1);
8352 
8353 	/* Wait for BSP CPU to send INIT signal */
8354 	while (vmx_get_test_stage() != 2)
8355 		;
8356 
8357 	/*
8358 	 * Signal that we continue as usual as INIT signal
8359 	 * should be blocked while CPU is in VMX operation
8360 	 */
8361 	vmx_set_test_stage(3);
8362 
8363 	/* Wait for signal to enter VMX non-root mode */
8364 	while (vmx_get_test_stage() != 4)
8365 		;
8366 
8367 	/* Enter VMX non-root mode */
8368 	test_set_guest(v2_null_test_guest);
8369 	make_vmcs_current(test_vmcs);
8370 	enter_guest();
8371 	/* Save exit reason for BSP CPU to compare to expected result */
8372 	init_signal_test_exit_reason = vmcs_read(EXI_REASON);
8373 	/* VMCLEAR test-vmcs so it could be loaded by BSP CPU */
8374 	vmcs_clear(test_vmcs);
8375 	launched = false;
8376 	/* Signal that CPU exited to VMX root mode */
8377 	vmx_set_test_stage(5);
8378 
8379 	/* Wait for signal to exit VMX operation */
8380 	while (vmx_get_test_stage() != 6)
8381 		;
8382 
8383 	/* Exit VMX operation (i.e. exec VMXOFF) */
8384 	vmx_off();
8385 
8386 	/*
8387 	 * Exiting VMX operation should result in latched
8388 	 * INIT signal being processed. Therefore, we should
8389 	 * never reach the below code. Thus, signal to BSP
8390 	 * CPU if we have reached here so it is able to
8391 	 * report an issue if it happens.
8392 	 */
8393 	init_signal_test_thread_continued = true;
8394 }
8395 
8396 #define INIT_SIGNAL_TEST_DELAY	100000000ULL
8397 
8398 static void vmx_init_signal_test(void)
8399 {
8400 	struct vmcs *test_vmcs;
8401 
8402 	if (cpu_count() < 2) {
8403 		report_skip(__func__);
8404 		return;
8405 	}
8406 
8407 	/* VMCLEAR test-vmcs so it could be loaded by other CPU */
8408 	vmcs_save(&test_vmcs);
8409 	vmcs_clear(test_vmcs);
8410 
8411 	vmx_set_test_stage(0);
8412 	on_cpu_async(1, init_signal_test_thread, test_vmcs);
8413 
8414 	/* Wait for other CPU to enter VMX operation */
8415 	while (vmx_get_test_stage() != 1)
8416 		;
8417 
8418 	/* Send INIT signal to other CPU */
8419 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
8420 				   id_map[1]);
8421 	/* Signal other CPU we have sent INIT signal */
8422 	vmx_set_test_stage(2);
8423 
8424 	/*
8425 	 * Wait reasonable amount of time for INIT signal to
8426 	 * be received on other CPU and verify that other CPU
8427 	 * have proceed as usual to next test stage as INIT
8428 	 * signal should be blocked while other CPU in
8429 	 * VMX operation
8430 	 */
8431 	delay(INIT_SIGNAL_TEST_DELAY);
8432 	report("INIT signal blocked when CPU in VMX operation",
8433 		   vmx_get_test_stage() == 3);
8434 	/* No point to continue if we failed at this point */
8435 	if (vmx_get_test_stage() != 3)
8436 		return;
8437 
8438 	/* Signal other CPU to enter VMX non-root mode */
8439 	init_signal_test_exit_reason = -1ull;
8440 	vmx_set_test_stage(4);
8441 	/*
8442 	 * Wait reasonable amont of time for other CPU
8443 	 * to exit to VMX root mode
8444 	 */
8445 	delay(INIT_SIGNAL_TEST_DELAY);
8446 	if (vmx_get_test_stage() != 5) {
8447 		report("Pending INIT signal didn't result in VMX exit", false);
8448 		return;
8449 	}
8450 	report("INIT signal during VMX non-root mode result in exit-reason %s (%lu)",
8451 			init_signal_test_exit_reason == VMX_INIT,
8452 			exit_reason_description(init_signal_test_exit_reason),
8453 			init_signal_test_exit_reason);
8454 
8455 	/* Run guest to completion */
8456 	make_vmcs_current(test_vmcs);
8457 	enter_guest();
8458 
8459 	/* Signal other CPU to exit VMX operation */
8460 	init_signal_test_thread_continued = false;
8461 	vmx_set_test_stage(6);
8462 
8463 	/*
8464 	 * Wait reasonable amount of time for other CPU
8465 	 * to run after INIT signal was processed
8466 	 */
8467 	delay(INIT_SIGNAL_TEST_DELAY);
8468 	report("INIT signal processed after exit VMX operation",
8469 		   !init_signal_test_thread_continued);
8470 
8471 	/*
8472 	 * TODO: Send SIPI to other CPU to sipi_entry (See x86/cstart64.S)
8473 	 * to re-init it to kvm-unit-tests standard environment.
8474 	 * Somehow (?) verify that SIPI was indeed received.
8475 	 */
8476 }
8477 
8478 enum vmcs_access {
8479 	ACCESS_VMREAD,
8480 	ACCESS_VMWRITE,
8481 	ACCESS_NONE,
8482 };
8483 
8484 struct vmcs_shadow_test_common {
8485 	enum vmcs_access op;
8486 	enum Reason reason;
8487 	u64 field;
8488 	u64 value;
8489 	u64 flags;
8490 	u64 time;
8491 } l1_l2_common;
8492 
8493 static inline u64 vmread_flags(u64 field, u64 *val)
8494 {
8495 	u64 flags;
8496 
8497 	asm volatile ("vmread %2, %1; pushf; pop %0"
8498 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
8499 	return flags & X86_EFLAGS_ALU;
8500 }
8501 
8502 static inline u64 vmwrite_flags(u64 field, u64 val)
8503 {
8504 	u64 flags;
8505 
8506 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
8507 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
8508 	return flags & X86_EFLAGS_ALU;
8509 }
8510 
8511 static void vmx_vmcs_shadow_test_guest(void)
8512 {
8513 	struct vmcs_shadow_test_common *c = &l1_l2_common;
8514 	u64 start;
8515 
8516 	while (c->op != ACCESS_NONE) {
8517 		start = rdtsc();
8518 		switch (c->op) {
8519 		default:
8520 			c->flags = -1ull;
8521 			break;
8522 		case ACCESS_VMREAD:
8523 			c->flags = vmread_flags(c->field, &c->value);
8524 			break;
8525 		case ACCESS_VMWRITE:
8526 			c->flags = vmwrite_flags(c->field, 0);
8527 			break;
8528 		}
8529 		c->time = rdtsc() - start;
8530 		vmcall();
8531 	}
8532 }
8533 
8534 static u64 vmread_from_shadow(u64 field)
8535 {
8536 	struct vmcs *primary;
8537 	struct vmcs *shadow;
8538 	u64 value;
8539 
8540 	TEST_ASSERT(!vmcs_save(&primary));
8541 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
8542 	TEST_ASSERT(!make_vmcs_current(shadow));
8543 	value = vmcs_read(field);
8544 	TEST_ASSERT(!make_vmcs_current(primary));
8545 	return value;
8546 }
8547 
8548 static u64 vmwrite_to_shadow(u64 field, u64 value)
8549 {
8550 	struct vmcs *primary;
8551 	struct vmcs *shadow;
8552 
8553 	TEST_ASSERT(!vmcs_save(&primary));
8554 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
8555 	TEST_ASSERT(!make_vmcs_current(shadow));
8556 	vmcs_write(field, value);
8557 	value = vmcs_read(field);
8558 	TEST_ASSERT(!make_vmcs_current(primary));
8559 	return value;
8560 }
8561 
8562 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
8563 {
8564 	struct vmcs_shadow_test_common *c = &l1_l2_common;
8565 
8566 	c->op = access;
8567 	vmcs_write(VMX_INST_ERROR, 0);
8568 	enter_guest();
8569 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
8570 	if (c->reason != VMX_VMCALL) {
8571 		skip_exit_insn();
8572 		enter_guest();
8573 	}
8574 	skip_exit_vmcall();
8575 }
8576 
8577 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
8578 {
8579 	struct vmcs_shadow_test_common *c = &l1_l2_common;
8580 	struct vmcs *shadow;
8581 	u64 value;
8582 	uintptr_t flags[2];
8583 	bool good_shadow;
8584 	u32 vmx_inst_error;
8585 
8586 	report_prefix_pushf("field %lx", field);
8587 	c->field = field;
8588 
8589 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
8590 	if (shadow != (struct vmcs *)-1ull) {
8591 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
8592 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
8593 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
8594 	} else {
8595 		/*
8596 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
8597 		 * shadowed-fields should fail with setting RFLAGS.CF.
8598 		 */
8599 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
8600 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
8601 		good_shadow = false;
8602 	}
8603 
8604 	/* Intercept both VMREAD and VMWRITE. */
8605 	report_prefix_push("no VMREAD/VMWRITE permission");
8606 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
8607 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
8608 		set_bit(field, bitmap[ACCESS_VMREAD]);
8609 		set_bit(field, bitmap[ACCESS_VMWRITE]);
8610 	}
8611 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
8612 	report("not shadowed for VMWRITE", c->reason == VMX_VMWRITE);
8613 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
8614 	report("not shadowed for VMREAD", c->reason == VMX_VMREAD);
8615 	report_prefix_pop();
8616 
8617 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
8618 		goto out;
8619 
8620 	/* Permit shadowed VMREAD. */
8621 	report_prefix_push("VMREAD permission only");
8622 	clear_bit(field, bitmap[ACCESS_VMREAD]);
8623 	set_bit(field, bitmap[ACCESS_VMWRITE]);
8624 	if (good_shadow)
8625 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
8626 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
8627 	report("not shadowed for VMWRITE", c->reason == VMX_VMWRITE);
8628 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
8629 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
8630 	report("shadowed for VMREAD (in %ld cycles)", c->reason == VMX_VMCALL,
8631 	       c->time);
8632 	report("ALU flags after VMREAD (%lx) are as expected (%lx)",
8633 	       c->flags == flags[ACCESS_VMREAD],
8634 	       c->flags, flags[ACCESS_VMREAD]);
8635 	if (good_shadow)
8636 		report("value read from shadow (%lx) is as expected (%lx)",
8637 		       c->value == value, c->value, value);
8638 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
8639 		report("VMX_INST_ERROR (%d) is as expected (%d)",
8640 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
8641 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8642 	report_prefix_pop();
8643 
8644 	/* Permit shadowed VMWRITE. */
8645 	report_prefix_push("VMWRITE permission only");
8646 	set_bit(field, bitmap[ACCESS_VMREAD]);
8647 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
8648 	if (good_shadow)
8649 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
8650 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
8651 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
8652 	report("shadowed for VMWRITE (in %ld cycles)", c->reason == VMX_VMCALL,
8653 		c->time);
8654 	report("ALU flags after VMWRITE (%lx) are as expected (%lx)",
8655 	       c->flags == flags[ACCESS_VMREAD],
8656 	       c->flags, flags[ACCESS_VMREAD]);
8657 	if (good_shadow) {
8658 		value = vmread_from_shadow(field);
8659 		report("shadow VMCS value (%lx) is as expected (%lx)",
8660 		       value == 0, value, 0ul);
8661 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
8662 		report("VMX_INST_ERROR (%d) is as expected (%d)",
8663 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
8664 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8665 	}
8666 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
8667 	report("not shadowed for VMREAD", c->reason == VMX_VMREAD);
8668 	report_prefix_pop();
8669 
8670 	/* Permit shadowed VMREAD and VMWRITE. */
8671 	report_prefix_push("VMREAD and VMWRITE permission");
8672 	clear_bit(field, bitmap[ACCESS_VMREAD]);
8673 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
8674 	if (good_shadow)
8675 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
8676 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
8677 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
8678 	report("shadowed for VMWRITE (in %ld cycles)", c->reason == VMX_VMCALL,
8679 		c->time);
8680 	report("ALU flags after VMWRITE (%lx) are as expected (%lx)",
8681 	       c->flags == flags[ACCESS_VMREAD],
8682 	       c->flags, flags[ACCESS_VMREAD]);
8683 	if (good_shadow) {
8684 		value = vmread_from_shadow(field);
8685 		report("shadow VMCS value (%lx) is as expected (%lx)",
8686 		       value == 0, value, 0ul);
8687 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
8688 		report("VMX_INST_ERROR (%d) is as expected (%d)",
8689 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
8690 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8691 	}
8692 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
8693 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
8694 	report("shadowed for VMREAD (in %ld cycles)", c->reason == VMX_VMCALL,
8695 	       c->time);
8696 	report("ALU flags after VMREAD (%lx) are as expected (%lx)",
8697 	       c->flags == flags[ACCESS_VMREAD],
8698 	       c->flags, flags[ACCESS_VMREAD]);
8699 	if (good_shadow)
8700 		report("value read from shadow (%lx) is as expected (%lx)",
8701 		       c->value == 0, c->value, 0ul);
8702 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
8703 		report("VMX_INST_ERROR (%d) is as expected (%d)",
8704 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
8705 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8706 	report_prefix_pop();
8707 
8708 out:
8709 	report_prefix_pop();
8710 }
8711 
8712 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
8713 {
8714 	unsigned base;
8715 	unsigned index;
8716 	unsigned bit;
8717 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
8718 
8719 	/* Run test on all possible valid VMCS fields */
8720 	for (base = 0;
8721 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
8722 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
8723 		for (index = 0; index <= highest_index; index++)
8724 			vmcs_shadow_test_field(bitmap, base + index);
8725 
8726 	/*
8727 	 * Run tests on some invalid VMCS fields
8728 	 * (Have reserved bit set).
8729 	 */
8730 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
8731 		vmcs_shadow_test_field(bitmap, (1ull << bit));
8732 }
8733 
8734 static void vmx_vmcs_shadow_test(void)
8735 {
8736 	u8 *bitmap[2];
8737 	struct vmcs *shadow;
8738 
8739 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
8740 		printf("\t'Activate secondary controls' not supported.\n");
8741 		return;
8742 	}
8743 
8744 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
8745 		printf("\t'VMCS shadowing' not supported.\n");
8746 		return;
8747 	}
8748 
8749 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
8750 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
8751 		printf("\tVMWRITE can't modify VM-exit information fields.\n");
8752 		return;
8753 	}
8754 
8755 	test_set_guest(vmx_vmcs_shadow_test_guest);
8756 
8757 	bitmap[ACCESS_VMREAD] = alloc_page();
8758 	bitmap[ACCESS_VMWRITE] = alloc_page();
8759 
8760 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
8761 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
8762 
8763 	shadow = alloc_page();
8764 	shadow->hdr.revision_id = basic.revision;
8765 	shadow->hdr.shadow_vmcs = 1;
8766 	TEST_ASSERT(!vmcs_clear(shadow));
8767 
8768 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
8769 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
8770 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
8771 
8772 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
8773 	report_prefix_push("valid link pointer");
8774 	vmx_vmcs_shadow_test_body(bitmap);
8775 	report_prefix_pop();
8776 
8777 	vmcs_write(VMCS_LINK_PTR, -1ull);
8778 	report_prefix_push("invalid link pointer");
8779 	vmx_vmcs_shadow_test_body(bitmap);
8780 	report_prefix_pop();
8781 
8782 	l1_l2_common.op = ACCESS_NONE;
8783 	enter_guest();
8784 }
8785 
8786 
8787 
8788 static int invalid_msr_init(struct vmcs *vmcs)
8789 {
8790 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
8791 		printf("\tPreemption timer is not supported\n");
8792 		return VMX_TEST_EXIT;
8793 	}
8794 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
8795 	preempt_val = 10000000;
8796 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
8797 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
8798 
8799 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
8800 		printf("\tSave preemption value is not supported\n");
8801 
8802 	vmcs_write(ENT_MSR_LD_CNT, 1);
8803 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
8804 
8805 	return VMX_TEST_START;
8806 }
8807 
8808 
8809 static void invalid_msr_main(void)
8810 {
8811 	report("Invalid MSR load", 0);
8812 }
8813 
8814 static int invalid_msr_exit_handler(void)
8815 {
8816 	report("Invalid MSR load", 0);
8817 	print_vmexit_info();
8818 	return VMX_TEST_EXIT;
8819 }
8820 
8821 static int invalid_msr_entry_failure(struct vmentry_failure *failure)
8822 {
8823 	ulong reason;
8824 
8825 	reason = vmcs_read(EXI_REASON);
8826 	report("Invalid MSR load", reason == (0x80000000u | VMX_FAIL_MSR));
8827 	return VMX_TEST_VMEXIT;
8828 }
8829 
8830 /*
8831  * The max number of MSRs in an atomic switch MSR list is:
8832  * (111B + 1) * 512 = 4096
8833  *
8834  * Each list entry consumes:
8835  * 4-byte MSR index + 4 bytes reserved + 8-byte data = 16 bytes
8836  *
8837  * Allocate 128 kB to cover max_msr_list_size (i.e., 64 kB) and then some.
8838  */
8839 static const u32 msr_list_page_order = 5;
8840 
8841 static void atomic_switch_msr_limit_test_guest(void)
8842 {
8843 	vmcall();
8844 }
8845 
8846 static void populate_msr_list(struct vmx_msr_entry *msr_list,
8847 			      size_t byte_capacity, int count)
8848 {
8849 	int i;
8850 
8851 	for (i = 0; i < count; i++) {
8852 		msr_list[i].index = MSR_IA32_TSC;
8853 		msr_list[i].reserved = 0;
8854 		msr_list[i].value = 0x1234567890abcdef;
8855 	}
8856 
8857 	memset(msr_list + count, 0xff,
8858 	       byte_capacity - count * sizeof(*msr_list));
8859 }
8860 
8861 static int max_msr_list_size(void)
8862 {
8863 	u32 vmx_misc = rdmsr(MSR_IA32_VMX_MISC);
8864 	u32 factor = ((vmx_misc & GENMASK(27, 25)) >> 25) + 1;
8865 
8866 	return factor * 512;
8867 }
8868 
8869 static void atomic_switch_msrs_test(int count)
8870 {
8871 	struct vmx_msr_entry *vm_enter_load;
8872         struct vmx_msr_entry *vm_exit_load;
8873         struct vmx_msr_entry *vm_exit_store;
8874 	int max_allowed = max_msr_list_size();
8875 	int byte_capacity = 1ul << (msr_list_page_order + PAGE_SHIFT);
8876 	/* Exceeding the max MSR list size at exit trigers KVM to abort. */
8877 	int exit_count = count > max_allowed ? max_allowed : count;
8878 	int cleanup_count = count > max_allowed ? 2 : 1;
8879 	int i;
8880 
8881 	/*
8882 	 * Check for the IA32_TSC MSR,
8883 	 * available with the "TSC flag" and used to populate the MSR lists.
8884 	 */
8885 	if (!(cpuid(1).d & (1 << 4))) {
8886 		report_skip(__func__);
8887 		return;
8888 	}
8889 
8890 	/* Set L2 guest. */
8891 	test_set_guest(atomic_switch_msr_limit_test_guest);
8892 
8893 	/* Setup atomic MSR switch lists. */
8894 	vm_enter_load = alloc_pages(msr_list_page_order);
8895 	vm_exit_load = alloc_pages(msr_list_page_order);
8896 	vm_exit_store = alloc_pages(msr_list_page_order);
8897 
8898 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)vm_enter_load);
8899 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)vm_exit_load);
8900 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)vm_exit_store);
8901 
8902 	/*
8903 	 * VM-Enter should succeed up to the max number of MSRs per list, and
8904 	 * should not consume junk beyond the last entry.
8905 	 */
8906 	populate_msr_list(vm_enter_load, byte_capacity, count);
8907 	populate_msr_list(vm_exit_load, byte_capacity, exit_count);
8908 	populate_msr_list(vm_exit_store, byte_capacity, exit_count);
8909 
8910 	vmcs_write(ENT_MSR_LD_CNT, count);
8911 	vmcs_write(EXI_MSR_LD_CNT, exit_count);
8912 	vmcs_write(EXI_MSR_ST_CNT, exit_count);
8913 
8914 	if (count <= max_allowed) {
8915 		enter_guest();
8916 		assert_exit_reason(VMX_VMCALL);
8917 		skip_exit_vmcall();
8918 	} else {
8919 		u32 exit_reason;
8920 		u32 exit_reason_want;
8921 		u32 exit_qual;
8922 
8923 		enter_guest_with_invalid_guest_state();
8924 
8925 		exit_reason = vmcs_read(EXI_REASON);
8926 		exit_reason_want = VMX_FAIL_MSR | VMX_ENTRY_FAILURE;
8927 		report("exit_reason, %u, is %u.",
8928 		       exit_reason == exit_reason_want, exit_reason,
8929 		       exit_reason_want);
8930 
8931 		exit_qual = vmcs_read(EXI_QUALIFICATION);
8932 		report("exit_qual, %u, is %u.", exit_qual == max_allowed + 1,
8933 		       exit_qual, max_allowed + 1);
8934 	}
8935 
8936 	/* Cleanup. */
8937 	vmcs_write(ENT_MSR_LD_CNT, 0);
8938 	vmcs_write(EXI_MSR_LD_CNT, 0);
8939 	vmcs_write(EXI_MSR_ST_CNT, 0);
8940 	for (i = 0; i < cleanup_count; i++) {
8941 		enter_guest();
8942 		skip_exit_vmcall();
8943 	}
8944 	free_pages_by_order(vm_enter_load, msr_list_page_order);
8945 	free_pages_by_order(vm_exit_load, msr_list_page_order);
8946 	free_pages_by_order(vm_exit_store, msr_list_page_order);
8947 }
8948 
8949 static void atomic_switch_max_msrs_test(void)
8950 {
8951 	atomic_switch_msrs_test(max_msr_list_size());
8952 }
8953 
8954 static void atomic_switch_overflow_msrs_test(void)
8955 {
8956 	atomic_switch_msrs_test(max_msr_list_size() + 1);
8957 }
8958 
8959 #define TEST(name) { #name, .v2 = name }
8960 
8961 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
8962 struct vmx_test vmx_tests[] = {
8963 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
8964 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
8965 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
8966 		preemption_timer_exit_handler, NULL, {0} },
8967 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
8968 		test_ctrl_pat_exit_handler, NULL, {0} },
8969 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
8970 		test_ctrl_efer_exit_handler, NULL, {0} },
8971 	{ "CR shadowing", NULL, cr_shadowing_main,
8972 		cr_shadowing_exit_handler, NULL, {0} },
8973 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
8974 		NULL, {0} },
8975 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
8976 		insn_intercept_exit_handler, NULL, {0} },
8977 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
8978 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
8979 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
8980 	{ "VPID", vpid_init, vpid_main, vpid_exit_handler, NULL, {0} },
8981 	{ "interrupt", interrupt_init, interrupt_main,
8982 		interrupt_exit_handler, NULL, {0} },
8983 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
8984 		NULL, {0} },
8985 	{ "MSR switch", msr_switch_init, msr_switch_main,
8986 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
8987 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
8988 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
8989 		disable_rdtscp_exit_handler, NULL, {0} },
8990 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
8991 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
8992 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
8993 		exit_monitor_from_l2_handler, NULL, {0} },
8994 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
8995 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
8996 	/* Basic V2 tests. */
8997 	TEST(v2_null_test),
8998 	TEST(v2_multiple_entries_test),
8999 	TEST(fixture_test_case1),
9000 	TEST(fixture_test_case2),
9001 	/* Opcode tests. */
9002 	TEST(invvpid_test_v2),
9003 	/* VM-entry tests */
9004 	TEST(vmx_controls_test),
9005 	TEST(vmx_host_state_area_test),
9006 	TEST(vmx_guest_state_area_test),
9007 	TEST(vmentry_movss_shadow_test),
9008 	/* APICv tests */
9009 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
9010 	TEST(vmx_hlt_with_rvi_test),
9011 	TEST(apic_reg_virt_test),
9012 	TEST(virt_x2apic_mode_test),
9013 	/* APIC pass-through tests */
9014 	TEST(vmx_apic_passthrough_test),
9015 	TEST(vmx_apic_passthrough_thread_test),
9016 	TEST(vmx_init_signal_test),
9017 	/* VMCS Shadowing tests */
9018 	TEST(vmx_vmcs_shadow_test),
9019 	/* Regression tests */
9020 	TEST(vmx_cr_load_test),
9021 	TEST(vmx_nm_test),
9022 	TEST(vmx_db_test),
9023 	TEST(vmx_nmi_window_test),
9024 	TEST(vmx_intr_window_test),
9025 	TEST(vmx_pending_event_test),
9026 	TEST(vmx_pending_event_hlt_test),
9027 	TEST(vmx_store_tsc_test),
9028 	/* EPT access tests. */
9029 	TEST(ept_access_test_not_present),
9030 	TEST(ept_access_test_read_only),
9031 	TEST(ept_access_test_write_only),
9032 	TEST(ept_access_test_read_write),
9033 	TEST(ept_access_test_execute_only),
9034 	TEST(ept_access_test_read_execute),
9035 	TEST(ept_access_test_write_execute),
9036 	TEST(ept_access_test_read_write_execute),
9037 	TEST(ept_access_test_reserved_bits),
9038 	TEST(ept_access_test_ignored_bits),
9039 	TEST(ept_access_test_paddr_not_present_ad_disabled),
9040 	TEST(ept_access_test_paddr_not_present_ad_enabled),
9041 	TEST(ept_access_test_paddr_read_only_ad_disabled),
9042 	TEST(ept_access_test_paddr_read_only_ad_enabled),
9043 	TEST(ept_access_test_paddr_read_write),
9044 	TEST(ept_access_test_paddr_read_write_execute),
9045 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
9046 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
9047 	TEST(ept_access_test_paddr_not_present_page_fault),
9048 	TEST(ept_access_test_force_2m_page),
9049 	/* Atomic MSR switch tests. */
9050 	TEST(atomic_switch_max_msrs_test),
9051 	TEST(atomic_switch_overflow_msrs_test),
9052 	{ NULL, NULL, NULL, NULL, NULL, {0} },
9053 };
9054