xref: /kvm-unit-tests/x86/vmx_tests.c (revision d4ae0a719d12f3927aa831b0c2c534a854a44845)
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 "pmu.h"
13 #include "vm.h"
14 #include "pci.h"
15 #include "fwcfg.h"
16 #include "isr.h"
17 #include "desc.h"
18 #include "apic.h"
19 #include "vmalloc.h"
20 #include "alloc_page.h"
21 #include "smp.h"
22 #include "delay.h"
23 #include "access.h"
24 #include "x86/usermode.h"
25 
26 /*
27  * vmcs.GUEST_PENDING_DEBUG has the same format as DR6, although some bits that
28  * are legal in DR6 are reserved in vmcs.GUEST_PENDING_DEBUG.  And if any data
29  * or I/O breakpoint matches *and* was enabled, bit 12 is also set.
30  */
31 #define PENDING_DBG_TRAP	BIT(12)
32 
33 #define VPID_CAP_INVVPID_TYPES_SHIFT 40
34 
35 u64 ia32_pat;
36 u64 ia32_efer;
37 void *io_bitmap_a, *io_bitmap_b;
38 u16 ioport;
39 
40 unsigned long *pml4;
41 u64 eptp;
42 void *data_page1, *data_page2;
43 
44 phys_addr_t pci_physaddr;
45 
46 void *pml_log;
47 #define PML_INDEX 512
48 
49 static inline unsigned ffs(unsigned x)
50 {
51 	int pos = -1;
52 
53 	__asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0"
54 			     : "+r"(pos) : "rm"(x) : "eax");
55 	return pos + 1;
56 }
57 
58 static inline void vmcall(void)
59 {
60 	asm volatile("vmcall");
61 }
62 
63 static void basic_guest_main(void)
64 {
65 	report_pass("Basic VMX test");
66 }
67 
68 static int basic_exit_handler(union exit_reason exit_reason)
69 {
70 	report_fail("Basic VMX test");
71 	print_vmexit_info(exit_reason);
72 	return VMX_TEST_EXIT;
73 }
74 
75 static void vmenter_main(void)
76 {
77 	u64 rax;
78 	u64 rsp, resume_rsp;
79 
80 	report_pass("test vmlaunch");
81 
82 	asm volatile(
83 		"mov %%rsp, %0\n\t"
84 		"mov %3, %%rax\n\t"
85 		"vmcall\n\t"
86 		"mov %%rax, %1\n\t"
87 		"mov %%rsp, %2\n\t"
88 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
89 		: "g"(0xABCD));
90 	report((rax == 0xFFFF) && (rsp == resume_rsp), "test vmresume");
91 }
92 
93 static int vmenter_exit_handler(union exit_reason exit_reason)
94 {
95 	u64 guest_rip = vmcs_read(GUEST_RIP);
96 
97 	switch (exit_reason.basic) {
98 	case VMX_VMCALL:
99 		if (regs.rax != 0xABCD) {
100 			report_fail("test vmresume");
101 			return VMX_TEST_VMEXIT;
102 		}
103 		regs.rax = 0xFFFF;
104 		vmcs_write(GUEST_RIP, guest_rip + 3);
105 		return VMX_TEST_RESUME;
106 	default:
107 		report_fail("test vmresume");
108 		print_vmexit_info(exit_reason);
109 	}
110 	return VMX_TEST_VMEXIT;
111 }
112 
113 u32 preempt_scale;
114 volatile unsigned long long tsc_val;
115 volatile u32 preempt_val;
116 u64 saved_rip;
117 
118 static int preemption_timer_init(struct vmcs *vmcs)
119 {
120 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
121 		printf("\tPreemption timer is not supported\n");
122 		return VMX_TEST_EXIT;
123 	}
124 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
125 	preempt_val = 10000000;
126 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
127 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
128 
129 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
130 		printf("\tSave preemption value is not supported\n");
131 
132 	return VMX_TEST_START;
133 }
134 
135 static void preemption_timer_main(void)
136 {
137 	tsc_val = rdtsc();
138 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
139 		vmx_set_test_stage(0);
140 		vmcall();
141 		if (vmx_get_test_stage() == 1)
142 			vmcall();
143 	}
144 	vmx_set_test_stage(1);
145 	while (vmx_get_test_stage() == 1) {
146 		if (((rdtsc() - tsc_val) >> preempt_scale)
147 				> 10 * preempt_val) {
148 			vmx_set_test_stage(2);
149 			vmcall();
150 		}
151 	}
152 	tsc_val = rdtsc();
153 	asm volatile ("hlt");
154 	vmcall();
155 	vmx_set_test_stage(5);
156 	vmcall();
157 }
158 
159 static int preemption_timer_exit_handler(union exit_reason exit_reason)
160 {
161 	bool guest_halted;
162 	u64 guest_rip;
163 	u32 insn_len;
164 	u32 ctrl_exit;
165 
166 	guest_rip = vmcs_read(GUEST_RIP);
167 	insn_len = vmcs_read(EXI_INST_LEN);
168 	switch (exit_reason.basic) {
169 	case VMX_PREEMPT:
170 		switch (vmx_get_test_stage()) {
171 		case 1:
172 		case 2:
173 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val,
174 			       "busy-wait for preemption timer");
175 			vmx_set_test_stage(3);
176 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
177 			return VMX_TEST_RESUME;
178 		case 3:
179 			guest_halted =
180 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
181 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val
182 			        && guest_halted,
183 			       "preemption timer during hlt");
184 			vmx_set_test_stage(4);
185 			vmcs_write(PIN_CONTROLS,
186 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
187 			vmcs_write(EXI_CONTROLS,
188 				   vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT);
189 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
190 			return VMX_TEST_RESUME;
191 		case 4:
192 			report(saved_rip == guest_rip,
193 			       "preemption timer with 0 value");
194 			break;
195 		default:
196 			report_fail("Invalid stage.");
197 			print_vmexit_info(exit_reason);
198 			break;
199 		}
200 		break;
201 	case VMX_VMCALL:
202 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
203 		switch (vmx_get_test_stage()) {
204 		case 0:
205 			report(vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val,
206 			       "Keep preemption value");
207 			vmx_set_test_stage(1);
208 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
209 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
210 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
211 			vmcs_write(EXI_CONTROLS, ctrl_exit);
212 			return VMX_TEST_RESUME;
213 		case 1:
214 			report(vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val,
215 			       "Save preemption value");
216 			return VMX_TEST_RESUME;
217 		case 2:
218 			report_fail("busy-wait for preemption timer");
219 			vmx_set_test_stage(3);
220 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
221 			return VMX_TEST_RESUME;
222 		case 3:
223 			report_fail("preemption timer during hlt");
224 			vmx_set_test_stage(4);
225 			/* fall through */
226 		case 4:
227 			vmcs_write(PIN_CONTROLS,
228 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
229 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
230 			saved_rip = guest_rip + insn_len;
231 			return VMX_TEST_RESUME;
232 		case 5:
233 			report_fail("preemption timer with 0 value (vmcall stage 5)");
234 			break;
235 		default:
236 			// Should not reach here
237 			report_fail("unexpected stage, %d",
238 				    vmx_get_test_stage());
239 			print_vmexit_info(exit_reason);
240 			return VMX_TEST_VMEXIT;
241 		}
242 		break;
243 	default:
244 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
245 		print_vmexit_info(exit_reason);
246 	}
247 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
248 	return VMX_TEST_VMEXIT;
249 }
250 
251 static void msr_bmp_init(void)
252 {
253 	void *msr_bitmap;
254 	u32 ctrl_cpu0;
255 
256 	msr_bitmap = alloc_page();
257 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
258 	ctrl_cpu0 |= CPU_MSR_BITMAP;
259 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
260 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
261 }
262 
263 static void *get_msr_bitmap(void)
264 {
265 	void *msr_bitmap;
266 
267 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
268 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
269 	} else {
270 		msr_bitmap = alloc_page();
271 		memset(msr_bitmap, 0xff, PAGE_SIZE);
272 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
273 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
274 	}
275 
276 	return msr_bitmap;
277 }
278 
279 static void disable_intercept_for_x2apic_msrs(void)
280 {
281 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
282 	u32 msr;
283 
284 	for (msr = APIC_BASE_MSR;
285 		 msr < (APIC_BASE_MSR+0xff);
286 		 msr += BITS_PER_LONG) {
287 		unsigned int word = msr / BITS_PER_LONG;
288 
289 		msr_bitmap[word] = 0;
290 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
291 	}
292 }
293 
294 static int test_ctrl_pat_init(struct vmcs *vmcs)
295 {
296 	u64 ctrl_ent;
297 	u64 ctrl_exi;
298 
299 	msr_bmp_init();
300 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
301 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
302 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
303 		printf("\tSave/load PAT is not supported\n");
304 		return 1;
305 	}
306 
307 	ctrl_ent = vmcs_read(ENT_CONTROLS);
308 	ctrl_exi = vmcs_read(EXI_CONTROLS);
309 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
310 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
311 	vmcs_write(ENT_CONTROLS, ctrl_ent);
312 	vmcs_write(EXI_CONTROLS, ctrl_exi);
313 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
314 	vmcs_write(GUEST_PAT, 0x0);
315 	vmcs_write(HOST_PAT, ia32_pat);
316 	return VMX_TEST_START;
317 }
318 
319 static void test_ctrl_pat_main(void)
320 {
321 	u64 guest_ia32_pat;
322 
323 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
324 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
325 		printf("\tENT_LOAD_PAT is not supported.\n");
326 	else {
327 		if (guest_ia32_pat != 0) {
328 			report_fail("Entry load PAT");
329 			return;
330 		}
331 	}
332 	wrmsr(MSR_IA32_CR_PAT, 0x6);
333 	vmcall();
334 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
335 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
336 		report(guest_ia32_pat == ia32_pat, "Entry load PAT");
337 }
338 
339 static int test_ctrl_pat_exit_handler(union exit_reason exit_reason)
340 {
341 	u64 guest_rip;
342 	u64 guest_pat;
343 
344 	guest_rip = vmcs_read(GUEST_RIP);
345 	switch (exit_reason.basic) {
346 	case VMX_VMCALL:
347 		guest_pat = vmcs_read(GUEST_PAT);
348 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
349 			printf("\tEXI_SAVE_PAT is not supported\n");
350 			vmcs_write(GUEST_PAT, 0x6);
351 		} else {
352 			report(guest_pat == 0x6, "Exit save PAT");
353 		}
354 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
355 			printf("\tEXI_LOAD_PAT is not supported\n");
356 		else
357 			report(rdmsr(MSR_IA32_CR_PAT) == ia32_pat,
358 			       "Exit load 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 : Unknown exit reason, 0x%x.\n", exit_reason.full);
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_fail("Entry load EFER");
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(guest_ia32_efer == ia32_efer, "Entry load EFER");
403 }
404 
405 static int test_ctrl_efer_exit_handler(union exit_reason exit_reason)
406 {
407 	u64 guest_rip;
408 	u64 guest_efer;
409 
410 	guest_rip = vmcs_read(GUEST_RIP);
411 	switch (exit_reason.basic) {
412 	case VMX_VMCALL:
413 		guest_efer = vmcs_read(GUEST_EFER);
414 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
415 			printf("\tEXI_SAVE_EFER is not supported\n");
416 			vmcs_write(GUEST_EFER, ia32_efer);
417 		} else {
418 			report(guest_efer == ia32_efer, "Exit save EFER");
419 		}
420 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
421 			printf("\tEXI_LOAD_EFER is not supported\n");
422 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
423 		} else {
424 			report(rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX),
425 			       "Exit load EFER");
426 		}
427 		vmcs_write(GUEST_PAT, ia32_efer);
428 		vmcs_write(GUEST_RIP, guest_rip + 3);
429 		return VMX_TEST_RESUME;
430 	default:
431 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
432 		break;
433 	}
434 	return VMX_TEST_VMEXIT;
435 }
436 
437 u32 guest_cr0, guest_cr4;
438 
439 static void cr_shadowing_main(void)
440 {
441 	u32 cr0, cr4, tmp;
442 
443 	// Test read through
444 	vmx_set_test_stage(0);
445 	guest_cr0 = read_cr0();
446 	if (vmx_get_test_stage() == 1)
447 		report_fail("Read through CR0");
448 	else
449 		vmcall();
450 	vmx_set_test_stage(1);
451 	guest_cr4 = read_cr4();
452 	if (vmx_get_test_stage() == 2)
453 		report_fail("Read through CR4");
454 	else
455 		vmcall();
456 	// Test write through
457 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
458 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
459 	vmx_set_test_stage(2);
460 	write_cr0(guest_cr0);
461 	if (vmx_get_test_stage() == 3)
462 		report_fail("Write through CR0");
463 	else
464 		vmcall();
465 	vmx_set_test_stage(3);
466 	write_cr4(guest_cr4);
467 	if (vmx_get_test_stage() == 4)
468 		report_fail("Write through CR4");
469 	else
470 		vmcall();
471 	// Test read shadow
472 	vmx_set_test_stage(4);
473 	vmcall();
474 	cr0 = read_cr0();
475 	if (vmx_get_test_stage() != 5)
476 		report(cr0 == guest_cr0, "Read shadowing CR0");
477 	vmx_set_test_stage(5);
478 	cr4 = read_cr4();
479 	if (vmx_get_test_stage() != 6)
480 		report(cr4 == guest_cr4, "Read shadowing CR4");
481 	// Test write shadow (same value with shadow)
482 	vmx_set_test_stage(6);
483 	write_cr0(guest_cr0);
484 	if (vmx_get_test_stage() == 7)
485 		report_fail("Write shadowing CR0 (same value with shadow)");
486 	else
487 		vmcall();
488 	vmx_set_test_stage(7);
489 	write_cr4(guest_cr4);
490 	if (vmx_get_test_stage() == 8)
491 		report_fail("Write shadowing CR4 (same value with shadow)");
492 	else
493 		vmcall();
494 	// Test write shadow (different value)
495 	vmx_set_test_stage(8);
496 	tmp = guest_cr0 ^ X86_CR0_TS;
497 	asm volatile("mov %0, %%rsi\n\t"
498 		"mov %%rsi, %%cr0\n\t"
499 		::"m"(tmp)
500 		:"rsi", "memory", "cc");
501 	report(vmx_get_test_stage() == 9,
502 	       "Write shadowing different X86_CR0_TS");
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(vmx_get_test_stage() == 10,
510 	       "Write shadowing different X86_CR0_MP");
511 	vmx_set_test_stage(10);
512 	tmp = guest_cr4 ^ X86_CR4_TSD;
513 	asm volatile("mov %0, %%rsi\n\t"
514 		"mov %%rsi, %%cr4\n\t"
515 		::"m"(tmp)
516 		:"rsi", "memory", "cc");
517 	report(vmx_get_test_stage() == 11,
518 	       "Write shadowing different X86_CR4_TSD");
519 	vmx_set_test_stage(11);
520 	tmp = guest_cr4 ^ X86_CR4_DE;
521 	asm volatile("mov %0, %%rsi\n\t"
522 		"mov %%rsi, %%cr4\n\t"
523 		::"m"(tmp)
524 		:"rsi", "memory", "cc");
525 	report(vmx_get_test_stage() == 12,
526 	       "Write shadowing different X86_CR4_DE");
527 }
528 
529 static int cr_shadowing_exit_handler(union exit_reason exit_reason)
530 {
531 	u64 guest_rip;
532 	u32 insn_len;
533 	u32 exit_qual;
534 
535 	guest_rip = vmcs_read(GUEST_RIP);
536 	insn_len = vmcs_read(EXI_INST_LEN);
537 	exit_qual = vmcs_read(EXI_QUALIFICATION);
538 	switch (exit_reason.basic) {
539 	case VMX_VMCALL:
540 		switch (vmx_get_test_stage()) {
541 		case 0:
542 			report(guest_cr0 == vmcs_read(GUEST_CR0),
543 			       "Read through CR0");
544 			break;
545 		case 1:
546 			report(guest_cr4 == vmcs_read(GUEST_CR4),
547 			       "Read through CR4");
548 			break;
549 		case 2:
550 			report(guest_cr0 == vmcs_read(GUEST_CR0),
551 			       "Write through CR0");
552 			break;
553 		case 3:
554 			report(guest_cr4 == vmcs_read(GUEST_CR4),
555 			       "Write through CR4");
556 			break;
557 		case 4:
558 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
559 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
560 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
561 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
562 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
563 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
564 			break;
565 		case 6:
566 			report(guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)),
567 			       "Write shadowing CR0 (same value)");
568 			break;
569 		case 7:
570 			report(guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)),
571 			       "Write shadowing CR4 (same value)");
572 			break;
573 		default:
574 			// Should not reach here
575 			report_fail("unexpected stage, %d",
576 				    vmx_get_test_stage());
577 			print_vmexit_info(exit_reason);
578 			return VMX_TEST_VMEXIT;
579 		}
580 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
581 		return VMX_TEST_RESUME;
582 	case VMX_CR:
583 		switch (vmx_get_test_stage()) {
584 		case 4:
585 			report_fail("Read shadowing CR0");
586 			vmx_inc_test_stage();
587 			break;
588 		case 5:
589 			report_fail("Read shadowing CR4");
590 			vmx_inc_test_stage();
591 			break;
592 		case 6:
593 			report_fail("Write shadowing CR0 (same value)");
594 			vmx_inc_test_stage();
595 			break;
596 		case 7:
597 			report_fail("Write shadowing CR4 (same value)");
598 			vmx_inc_test_stage();
599 			break;
600 		case 8:
601 		case 9:
602 			// 0x600 encodes "mov %esi, %cr0"
603 			if (exit_qual == 0x600)
604 				vmx_inc_test_stage();
605 			break;
606 		case 10:
607 		case 11:
608 			// 0x604 encodes "mov %esi, %cr4"
609 			if (exit_qual == 0x604)
610 				vmx_inc_test_stage();
611 			break;
612 		default:
613 			// Should not reach here
614 			report_fail("unexpected stage, %d",
615 				    vmx_get_test_stage());
616 			print_vmexit_info(exit_reason);
617 			return VMX_TEST_VMEXIT;
618 		}
619 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
620 		return VMX_TEST_RESUME;
621 	default:
622 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
623 		print_vmexit_info(exit_reason);
624 	}
625 	return VMX_TEST_VMEXIT;
626 }
627 
628 static int iobmp_init(struct vmcs *vmcs)
629 {
630 	u32 ctrl_cpu0;
631 
632 	io_bitmap_a = alloc_page();
633 	io_bitmap_b = alloc_page();
634 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
635 	ctrl_cpu0 |= CPU_IO_BITMAP;
636 	ctrl_cpu0 &= (~CPU_IO);
637 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
638 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
639 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
640 	return VMX_TEST_START;
641 }
642 
643 static void iobmp_main(void)
644 {
645 	// stage 0, test IO pass
646 	vmx_set_test_stage(0);
647 	inb(0x5000);
648 	outb(0x0, 0x5000);
649 	report(vmx_get_test_stage() == 0, "I/O bitmap - I/O pass");
650 	// test IO width, in/out
651 	((u8 *)io_bitmap_a)[0] = 0xFF;
652 	vmx_set_test_stage(2);
653 	inb(0x0);
654 	report(vmx_get_test_stage() == 3, "I/O bitmap - trap in");
655 	vmx_set_test_stage(3);
656 	outw(0x0, 0x0);
657 	report(vmx_get_test_stage() == 4, "I/O bitmap - trap out");
658 	vmx_set_test_stage(4);
659 	inl(0x0);
660 	report(vmx_get_test_stage() == 5, "I/O bitmap - I/O width, long");
661 	// test low/high IO port
662 	vmx_set_test_stage(5);
663 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
664 	inb(0x5000);
665 	report(vmx_get_test_stage() == 6, "I/O bitmap - I/O port, low part");
666 	vmx_set_test_stage(6);
667 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
668 	inb(0x9000);
669 	report(vmx_get_test_stage() == 7, "I/O bitmap - I/O port, high part");
670 	// test partial pass
671 	vmx_set_test_stage(7);
672 	inl(0x4FFF);
673 	report(vmx_get_test_stage() == 8, "I/O bitmap - partial pass");
674 	// test overrun
675 	vmx_set_test_stage(8);
676 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
677 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
678 	inl(0xFFFF);
679 	report(vmx_get_test_stage() == 9, "I/O bitmap - overrun");
680 	vmx_set_test_stage(9);
681 	vmcall();
682 	outb(0x0, 0x0);
683 	report(vmx_get_test_stage() == 9,
684 	       "I/O bitmap - ignore unconditional exiting");
685 	vmx_set_test_stage(10);
686 	vmcall();
687 	outb(0x0, 0x0);
688 	report(vmx_get_test_stage() == 11,
689 	       "I/O bitmap - unconditional exiting");
690 }
691 
692 static int iobmp_exit_handler(union exit_reason exit_reason)
693 {
694 	u64 guest_rip;
695 	ulong exit_qual;
696 	u32 insn_len, ctrl_cpu0;
697 
698 	guest_rip = vmcs_read(GUEST_RIP);
699 	exit_qual = vmcs_read(EXI_QUALIFICATION);
700 	insn_len = vmcs_read(EXI_INST_LEN);
701 	switch (exit_reason.basic) {
702 	case VMX_IO:
703 		switch (vmx_get_test_stage()) {
704 		case 0:
705 		case 1:
706 			vmx_inc_test_stage();
707 			break;
708 		case 2:
709 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE,
710 			       "I/O bitmap - I/O width, byte");
711 			report(exit_qual & VMX_IO_IN,
712 			       "I/O bitmap - I/O direction, in");
713 			vmx_inc_test_stage();
714 			break;
715 		case 3:
716 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD,
717 			       "I/O bitmap - I/O width, word");
718 			report(!(exit_qual & VMX_IO_IN),
719 			       "I/O bitmap - I/O direction, out");
720 			vmx_inc_test_stage();
721 			break;
722 		case 4:
723 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG,
724 			       "I/O bitmap - I/O width, long");
725 			vmx_inc_test_stage();
726 			break;
727 		case 5:
728 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
729 				vmx_inc_test_stage();
730 			break;
731 		case 6:
732 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
733 				vmx_inc_test_stage();
734 			break;
735 		case 7:
736 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
737 				vmx_inc_test_stage();
738 			break;
739 		case 8:
740 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
741 				vmx_inc_test_stage();
742 			break;
743 		case 9:
744 		case 10:
745 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
746 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
747 			vmx_inc_test_stage();
748 			break;
749 		default:
750 			// Should not reach here
751 			report_fail("unexpected stage, %d",
752 				    vmx_get_test_stage());
753 			print_vmexit_info(exit_reason);
754 			return VMX_TEST_VMEXIT;
755 		}
756 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
757 		return VMX_TEST_RESUME;
758 	case VMX_VMCALL:
759 		switch (vmx_get_test_stage()) {
760 		case 9:
761 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
762 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
763 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
764 			break;
765 		case 10:
766 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
767 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
768 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
769 			break;
770 		default:
771 			// Should not reach here
772 			report_fail("unexpected stage, %d",
773 				    vmx_get_test_stage());
774 			print_vmexit_info(exit_reason);
775 			return VMX_TEST_VMEXIT;
776 		}
777 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
778 		return VMX_TEST_RESUME;
779 	default:
780 		printf("guest_rip = %#lx\n", guest_rip);
781 		printf("\tERROR : Unknown exit reason, 0x%x\n", exit_reason.full);
782 		break;
783 	}
784 	return VMX_TEST_VMEXIT;
785 }
786 
787 #define INSN_CPU0		0
788 #define INSN_CPU1		1
789 #define INSN_ALWAYS_TRAP	2
790 
791 #define FIELD_EXIT_QUAL		(1 << 0)
792 #define FIELD_INSN_INFO		(1 << 1)
793 
794 asm(
795 	"insn_hlt: hlt;ret\n\t"
796 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
797 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
798 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
799 	"insn_rdtsc: rdtsc;ret\n\t"
800 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
801 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
802 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
803 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
804 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
805 	"insn_pause: pause;ret\n\t"
806 	"insn_wbinvd: wbinvd;ret\n\t"
807 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
808 	"insn_invd: invd;ret\n\t"
809 	"insn_sgdt: sgdt gdt_descr;ret\n\t"
810 	"insn_lgdt: lgdt gdt_descr;ret\n\t"
811 	"insn_sidt: sidt idt_descr;ret\n\t"
812 	"insn_lidt: lidt idt_descr;ret\n\t"
813 	"insn_sldt: sldt %ax;ret\n\t"
814 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
815 	"insn_str: str %ax;ret\n\t"
816 	"insn_rdrand: rdrand %rax;ret\n\t"
817 	"insn_rdseed: rdseed %rax;ret\n\t"
818 );
819 extern void insn_hlt(void);
820 extern void insn_invlpg(void);
821 extern void insn_mwait(void);
822 extern void insn_rdpmc(void);
823 extern void insn_rdtsc(void);
824 extern void insn_cr3_load(void);
825 extern void insn_cr3_store(void);
826 extern void insn_cr8_load(void);
827 extern void insn_cr8_store(void);
828 extern void insn_monitor(void);
829 extern void insn_pause(void);
830 extern void insn_wbinvd(void);
831 extern void insn_sgdt(void);
832 extern void insn_lgdt(void);
833 extern void insn_sidt(void);
834 extern void insn_lidt(void);
835 extern void insn_sldt(void);
836 extern void insn_lldt(void);
837 extern void insn_str(void);
838 extern void insn_cpuid(void);
839 extern void insn_invd(void);
840 extern void insn_rdrand(void);
841 extern void insn_rdseed(void);
842 
843 u32 cur_insn;
844 u64 cr3;
845 
846 typedef bool (*supported_fn)(void);
847 
848 static bool this_cpu_has_mwait(void)
849 {
850 	return this_cpu_has(X86_FEATURE_MWAIT);
851 }
852 
853 struct insn_table {
854 	const char *name;
855 	u32 flag;
856 	void (*insn_func)(void);
857 	u32 type;
858 	u32 reason;
859 	ulong exit_qual;
860 	u32 insn_info;
861 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
862 	// which field need to be tested, reason is always tested
863 	u32 test_field;
864 	const supported_fn supported_fn;
865 	u8 disabled;
866 };
867 
868 /*
869  * Add more test cases of instruction intercept here. Elements in this
870  * table is:
871  *	name/control flag/insn function/type/exit reason/exit qulification/
872  *	instruction info/field to test
873  * The last field defines which fields (exit_qual and insn_info) need to be
874  * tested in exit handler. If set to 0, only "reason" is checked.
875  */
876 static struct insn_table insn_table[] = {
877 	// Flags for Primary Processor-Based VM-Execution Controls
878 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
879 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
880 		0x12345678, 0, FIELD_EXIT_QUAL},
881 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0, this_cpu_has_mwait},
882 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0, this_cpu_has_pmu},
883 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
884 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
885 		FIELD_EXIT_QUAL},
886 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
887 		FIELD_EXIT_QUAL},
888 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
889 		FIELD_EXIT_QUAL},
890 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
891 		FIELD_EXIT_QUAL},
892 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0, this_cpu_has_mwait},
893 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
894 	// Flags for Secondary Processor-Based VM-Execution Controls
895 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
896 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
897 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
898 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
899 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
900 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
901 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
902 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
903 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
904 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
905 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
906 	// Instructions always trap
907 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
908 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
909 	// Instructions never trap
910 	{NULL},
911 };
912 
913 static int insn_intercept_init(struct vmcs *vmcs)
914 {
915 	u32 ctrl_cpu, cur_insn;
916 
917 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
918 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
919 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
920 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
921 	cr3 = read_cr3();
922 
923 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
924 		if (insn_table[cur_insn].supported_fn == NULL)
925 			continue;
926 		insn_table[cur_insn].disabled = !insn_table[cur_insn].supported_fn();
927 	}
928 	return VMX_TEST_START;
929 }
930 
931 static void insn_intercept_main(void)
932 {
933 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
934 		vmx_set_test_stage(cur_insn * 2);
935 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
936 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
937 		    (insn_table[cur_insn].type == INSN_CPU1 &&
938 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
939 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
940 			       insn_table[cur_insn].type - INSN_CPU0,
941 			       insn_table[cur_insn].name);
942 			continue;
943 		}
944 
945 		if (insn_table[cur_insn].disabled) {
946 			printf("\tFeature required for %s is not supported.\n",
947 			       insn_table[cur_insn].name);
948 			continue;
949 		}
950 
951 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
952 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
953 		    (insn_table[cur_insn].type == INSN_CPU1 &&
954 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
955 			/* skip hlt, it stalls the guest and is tested below */
956 			if (insn_table[cur_insn].insn_func != insn_hlt)
957 				insn_table[cur_insn].insn_func();
958 			report(vmx_get_test_stage() == cur_insn * 2,
959 					"execute %s",
960 					insn_table[cur_insn].name);
961 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
962 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
963 			       insn_table[cur_insn].type - INSN_CPU0,
964 			       insn_table[cur_insn].name);
965 
966 		vmcall();
967 
968 		insn_table[cur_insn].insn_func();
969 		report(vmx_get_test_stage() == cur_insn * 2 + 1,
970 				"intercept %s",
971 				insn_table[cur_insn].name);
972 
973 		vmx_set_test_stage(cur_insn * 2 + 1);
974 		vmcall();
975 	}
976 }
977 
978 static int insn_intercept_exit_handler(union exit_reason exit_reason)
979 {
980 	u64 guest_rip;
981 	ulong exit_qual;
982 	u32 insn_len;
983 	u32 insn_info;
984 	bool pass;
985 
986 	guest_rip = vmcs_read(GUEST_RIP);
987 	exit_qual = vmcs_read(EXI_QUALIFICATION);
988 	insn_len = vmcs_read(EXI_INST_LEN);
989 	insn_info = vmcs_read(EXI_INST_INFO);
990 
991 	if (exit_reason.basic == VMX_VMCALL) {
992 		u32 val = 0;
993 
994 		if (insn_table[cur_insn].type == INSN_CPU0)
995 			val = vmcs_read(CPU_EXEC_CTRL0);
996 		else if (insn_table[cur_insn].type == INSN_CPU1)
997 			val = vmcs_read(CPU_EXEC_CTRL1);
998 
999 		if (vmx_get_test_stage() & 1)
1000 			val &= ~insn_table[cur_insn].flag;
1001 		else
1002 			val |= insn_table[cur_insn].flag;
1003 
1004 		if (insn_table[cur_insn].type == INSN_CPU0)
1005 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
1006 		else if (insn_table[cur_insn].type == INSN_CPU1)
1007 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
1008 	} else {
1009 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
1010 			insn_table[cur_insn].reason == exit_reason.full;
1011 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
1012 		    insn_table[cur_insn].exit_qual != exit_qual)
1013 			pass = false;
1014 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
1015 		    insn_table[cur_insn].insn_info != insn_info)
1016 			pass = false;
1017 		if (pass)
1018 			vmx_inc_test_stage();
1019 	}
1020 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1021 	return VMX_TEST_RESUME;
1022 }
1023 
1024 /**
1025  * __setup_ept - Setup the VMCS fields to enable Extended Page Tables (EPT)
1026  * @hpa:	Host physical address of the top-level, a.k.a. root, EPT table
1027  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1028  *
1029  * Returns 0 on success, 1 on failure.
1030  *
1031  * Note that @hpa doesn't need to point at actual memory if VM-Launch is
1032  * expected to fail, e.g. setup_dummy_ept() arbitrarily passes '0' to satisfy
1033  * the various EPTP consistency checks, but doesn't ensure backing for HPA '0'.
1034  */
1035 static int __setup_ept(u64 hpa, bool enable_ad)
1036 {
1037 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1038 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1039 		printf("\tEPT is not supported\n");
1040 		return 1;
1041 	}
1042 	if (!(ept_vpid.val & EPT_CAP_WB)) {
1043 		printf("\tWB memtype for EPT walks not supported\n");
1044 		return 1;
1045 	}
1046 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
1047 		printf("\tPWL4 is not supported\n");
1048 		return 1;
1049 	}
1050 
1051 	eptp = EPT_MEM_TYPE_WB;
1052 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1053 	eptp |= hpa;
1054 	if (enable_ad)
1055 		eptp |= EPTP_AD_FLAG;
1056 
1057 	vmcs_write(EPTP, eptp);
1058 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY);
1059 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT);
1060 
1061 	return 0;
1062 }
1063 
1064 /**
1065  * setup_ept - Enable Extended Page Tables (EPT) and setup an identity map
1066  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1067  *
1068  * Returns 0 on success, 1 on failure.
1069  *
1070  * This is the "real" function for setting up EPT tables, i.e. use this for
1071  * tests that need to run code in the guest with EPT enabled.
1072  */
1073 static int setup_ept(bool enable_ad)
1074 {
1075 	unsigned long end_of_memory;
1076 
1077 	pml4 = alloc_page();
1078 
1079 	if (__setup_ept(virt_to_phys(pml4), enable_ad))
1080 		return 1;
1081 
1082 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1083 	if (end_of_memory < (1ul << 32))
1084 		end_of_memory = (1ul << 32);
1085 	/* Cannot use large EPT pages if we need to track EPT
1086 	 * accessed/dirty bits at 4K granularity.
1087 	 */
1088 	setup_ept_range(pml4, 0, end_of_memory, 0,
1089 			!enable_ad && ept_2m_supported(),
1090 			EPT_WA | EPT_RA | EPT_EA);
1091 	return 0;
1092 }
1093 
1094 /**
1095  * setup_dummy_ept - Enable Extended Page Tables (EPT) with a dummy root HPA
1096  *
1097  * Setup EPT using a semi-arbitrary dummy root HPA.  This function is intended
1098  * for use by tests that need EPT enabled to verify dependent VMCS controls
1099  * but never expect to fully enter the guest, i.e. don't need setup the actual
1100  * EPT tables.
1101  */
1102 static void setup_dummy_ept(void)
1103 {
1104 	if (__setup_ept(0, false))
1105 		report_abort("EPT setup unexpectedly failed");
1106 }
1107 
1108 static int enable_unrestricted_guest(bool need_valid_ept)
1109 {
1110 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1111 	    !(ctrl_cpu_rev[1].clr & CPU_URG) ||
1112 	    !(ctrl_cpu_rev[1].clr & CPU_EPT))
1113 		return 1;
1114 
1115 	if (need_valid_ept)
1116 		setup_ept(false);
1117 	else
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 int ept_ad_enabled(void)
1139 {
1140 	return eptp & EPTP_AD_FLAG;
1141 }
1142 
1143 static void ept_enable_ad_bits_or_skip_test(void)
1144 {
1145 	if (!ept_ad_bits_supported())
1146 		test_skip("EPT AD bits not supported.");
1147 	ept_enable_ad_bits();
1148 }
1149 
1150 static int apic_version;
1151 
1152 static int ept_init_common(bool have_ad)
1153 {
1154 	int ret;
1155 	struct pci_dev pcidev;
1156 
1157 	/* INVEPT is required by the EPT violation handler. */
1158 	if (!is_invept_type_supported(INVEPT_SINGLE))
1159 		return VMX_TEST_EXIT;
1160 
1161 	if (setup_ept(have_ad))
1162 		return VMX_TEST_EXIT;
1163 
1164 	data_page1 = alloc_page();
1165 	data_page2 = alloc_page();
1166 	*((u32 *)data_page1) = MAGIC_VAL_1;
1167 	*((u32 *)data_page2) = MAGIC_VAL_2;
1168 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1169 			EPT_RA | EPT_WA | EPT_EA);
1170 
1171 	apic_version = apic_read(APIC_LVR);
1172 
1173 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1174 	if (ret != PCIDEVADDR_INVALID) {
1175 		pci_dev_init(&pcidev, ret);
1176 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1177 	}
1178 
1179 	return VMX_TEST_START;
1180 }
1181 
1182 static int ept_init(struct vmcs *vmcs)
1183 {
1184 	return ept_init_common(false);
1185 }
1186 
1187 static void ept_common(void)
1188 {
1189 	vmx_set_test_stage(0);
1190 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1191 			*((u32 *)data_page1) != MAGIC_VAL_1)
1192 		report_fail("EPT basic framework - read");
1193 	else {
1194 		*((u32 *)data_page2) = MAGIC_VAL_3;
1195 		vmcall();
1196 		if (vmx_get_test_stage() == 1) {
1197 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1198 					*((u32 *)data_page2) == MAGIC_VAL_2)
1199 				report_pass("EPT basic framework");
1200 			else
1201 				report_pass("EPT basic framework - remap");
1202 		}
1203 	}
1204 	// Test EPT Misconfigurations
1205 	vmx_set_test_stage(1);
1206 	vmcall();
1207 	*((u32 *)data_page1) = MAGIC_VAL_1;
1208 	if (vmx_get_test_stage() != 2) {
1209 		report_fail("EPT misconfigurations");
1210 		goto t1;
1211 	}
1212 	vmx_set_test_stage(2);
1213 	vmcall();
1214 	*((u32 *)data_page1) = MAGIC_VAL_1;
1215 	report(vmx_get_test_stage() == 3, "EPT misconfigurations");
1216 t1:
1217 	// Test EPT violation
1218 	vmx_set_test_stage(3);
1219 	vmcall();
1220 	*((u32 *)data_page1) = MAGIC_VAL_1;
1221 	report(vmx_get_test_stage() == 4, "EPT violation - page permission");
1222 	// Violation caused by EPT paging structure
1223 	vmx_set_test_stage(4);
1224 	vmcall();
1225 	*((u32 *)data_page1) = MAGIC_VAL_2;
1226 	report(vmx_get_test_stage() == 5, "EPT violation - paging structure");
1227 
1228 	// MMIO Read/Write
1229 	vmx_set_test_stage(5);
1230 	vmcall();
1231 
1232 	*(u32 volatile *)pci_physaddr;
1233 	report(vmx_get_test_stage() == 6, "MMIO EPT violation - read");
1234 
1235 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1236 	report(vmx_get_test_stage() == 7, "MMIO EPT violation - write");
1237 }
1238 
1239 static void ept_main(void)
1240 {
1241 	ept_common();
1242 
1243 	// Test EPT access to L1 MMIO
1244 	vmx_set_test_stage(7);
1245 	report(*((u32 *)0xfee00030UL) == apic_version, "EPT - MMIO access");
1246 
1247 	// Test invalid operand for INVEPT
1248 	vmcall();
1249 	report(vmx_get_test_stage() == 8, "EPT - unsupported INVEPT");
1250 }
1251 
1252 static bool invept_test(int type, u64 eptp)
1253 {
1254 	bool ret, supported;
1255 
1256 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1257 	ret = __invept(type, eptp);
1258 
1259 	if (ret == !supported)
1260 		return false;
1261 
1262 	if (!supported)
1263 		printf("WARNING: unsupported invept passed!\n");
1264 	else
1265 		printf("WARNING: invept failed!\n");
1266 
1267 	return true;
1268 }
1269 
1270 static int pml_exit_handler(union exit_reason exit_reason)
1271 {
1272 	u16 index, count;
1273 	u64 *pmlbuf = pml_log;
1274 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1275 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1276 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1277 
1278 	switch (exit_reason.basic) {
1279 	case VMX_VMCALL:
1280 		switch (vmx_get_test_stage()) {
1281 		case 0:
1282 			index = vmcs_read(GUEST_PML_INDEX);
1283 			for (count = index + 1; count < PML_INDEX; count++) {
1284 				if (pmlbuf[count] == (u64)data_page2) {
1285 					vmx_inc_test_stage();
1286 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1287 					break;
1288 				}
1289 			}
1290 			break;
1291 		case 1:
1292 			index = vmcs_read(GUEST_PML_INDEX);
1293 			/* Keep clearing the dirty bit till a overflow */
1294 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1295 			break;
1296 		default:
1297 			report_fail("unexpected stage, %d.",
1298 			       vmx_get_test_stage());
1299 			print_vmexit_info(exit_reason);
1300 			return VMX_TEST_VMEXIT;
1301 		}
1302 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1303 		return VMX_TEST_RESUME;
1304 	case VMX_PML_FULL:
1305 		vmx_inc_test_stage();
1306 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1307 		return VMX_TEST_RESUME;
1308 	default:
1309 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1310 		print_vmexit_info(exit_reason);
1311 	}
1312 	return VMX_TEST_VMEXIT;
1313 }
1314 
1315 static int ept_exit_handler_common(union exit_reason exit_reason, bool have_ad)
1316 {
1317 	u64 guest_rip;
1318 	u64 guest_cr3;
1319 	u32 insn_len;
1320 	u32 exit_qual;
1321 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte,
1322 			     guest_pte_addr;
1323 
1324 	guest_rip = vmcs_read(GUEST_RIP);
1325 	guest_cr3 = vmcs_read(GUEST_CR3);
1326 	insn_len = vmcs_read(EXI_INST_LEN);
1327 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1328 	pteval_t *ptep;
1329 	switch (exit_reason.basic) {
1330 	case VMX_VMCALL:
1331 		switch (vmx_get_test_stage()) {
1332 		case 0:
1333 			check_ept_ad(pml4, guest_cr3,
1334 				     (unsigned long)data_page1,
1335 				     have_ad ? EPT_ACCESS_FLAG : 0,
1336 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1337 			check_ept_ad(pml4, guest_cr3,
1338 				     (unsigned long)data_page2,
1339 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1340 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1341 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1342 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1343 			if (have_ad)
1344 				invept(INVEPT_SINGLE, eptp);
1345 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1346 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1347 				vmx_inc_test_stage();
1348 				install_ept(pml4, (unsigned long)data_page2,
1349 						(unsigned long)data_page2,
1350 						EPT_RA | EPT_WA | EPT_EA);
1351 			} else
1352 				report_fail("EPT basic framework - write");
1353 			break;
1354 		case 1:
1355 			install_ept(pml4, (unsigned long)data_page1,
1356  				(unsigned long)data_page1, EPT_WA);
1357 			invept(INVEPT_SINGLE, eptp);
1358 			break;
1359 		case 2:
1360 			install_ept(pml4, (unsigned long)data_page1,
1361  				(unsigned long)data_page1,
1362  				EPT_RA | EPT_WA | EPT_EA |
1363  				(2 << EPT_MEM_TYPE_SHIFT));
1364 			invept(INVEPT_SINGLE, eptp);
1365 			break;
1366 		case 3:
1367 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1368 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1369 						1, &data_page1_pte));
1370 			set_ept_pte(pml4, (unsigned long)data_page1,
1371 				1, data_page1_pte & ~EPT_PRESENT);
1372 			invept(INVEPT_SINGLE, eptp);
1373 			break;
1374 		case 4:
1375 			ptep = get_pte_level((pgd_t *)guest_cr3, data_page1, /*level=*/2);
1376 			guest_pte_addr = virt_to_phys(ptep) & PAGE_MASK;
1377 
1378 			TEST_ASSERT(get_ept_pte(pml4, guest_pte_addr, 2, &data_page1_pte_pte));
1379 			set_ept_pte(pml4, guest_pte_addr, 2,
1380 				data_page1_pte_pte & ~EPT_PRESENT);
1381 			invept(INVEPT_SINGLE, eptp);
1382 			break;
1383 		case 5:
1384 			install_ept(pml4, (unsigned long)pci_physaddr,
1385 				(unsigned long)pci_physaddr, 0);
1386 			invept(INVEPT_SINGLE, eptp);
1387 			break;
1388 		case 7:
1389 			if (!invept_test(0, eptp))
1390 				vmx_inc_test_stage();
1391 			break;
1392 		// Should not reach here
1393 		default:
1394 			report_fail("ERROR - unexpected stage, %d.",
1395 			       vmx_get_test_stage());
1396 			print_vmexit_info(exit_reason);
1397 			return VMX_TEST_VMEXIT;
1398 		}
1399 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1400 		return VMX_TEST_RESUME;
1401 	case VMX_EPT_MISCONFIG:
1402 		switch (vmx_get_test_stage()) {
1403 		case 1:
1404 		case 2:
1405 			vmx_inc_test_stage();
1406 			install_ept(pml4, (unsigned long)data_page1,
1407  				(unsigned long)data_page1,
1408  				EPT_RA | EPT_WA | EPT_EA);
1409 			invept(INVEPT_SINGLE, eptp);
1410 			break;
1411 		// Should not reach here
1412 		default:
1413 			report_fail("ERROR - unexpected stage, %d.",
1414 			       vmx_get_test_stage());
1415 			print_vmexit_info(exit_reason);
1416 			return VMX_TEST_VMEXIT;
1417 		}
1418 		return VMX_TEST_RESUME;
1419 	case VMX_EPT_VIOLATION:
1420 		/*
1421 		 * Exit-qualifications are masked not to account for advanced
1422 		 * VM-exit information. Once KVM supports this feature, this
1423 		 * masking should be removed.
1424 		 */
1425 		exit_qual &= ~EPT_VLT_GUEST_MASK;
1426 
1427 		switch(vmx_get_test_stage()) {
1428 		case 3:
1429 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1430 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1431 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1432 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1433 					EPT_VLT_PADDR))
1434 				vmx_inc_test_stage();
1435 			set_ept_pte(pml4, (unsigned long)data_page1,
1436 				1, data_page1_pte | (EPT_PRESENT));
1437 			invept(INVEPT_SINGLE, eptp);
1438 			break;
1439 		case 4:
1440 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1441 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1442 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1443 			if (exit_qual == (EPT_VLT_RD |
1444 					  (have_ad ? EPT_VLT_WR : 0) |
1445 					  EPT_VLT_LADDR_VLD))
1446 				vmx_inc_test_stage();
1447 			set_ept_pte(pml4, guest_pte_addr, 2,
1448 				data_page1_pte_pte | (EPT_PRESENT));
1449 			invept(INVEPT_SINGLE, eptp);
1450 			break;
1451 		case 5:
1452 			if (exit_qual & EPT_VLT_RD)
1453 				vmx_inc_test_stage();
1454 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1455 						1, &memaddr_pte));
1456 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1457 			invept(INVEPT_SINGLE, eptp);
1458 			break;
1459 		case 6:
1460 			if (exit_qual & EPT_VLT_WR)
1461 				vmx_inc_test_stage();
1462 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1463 						1, &memaddr_pte));
1464 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1465 			invept(INVEPT_SINGLE, eptp);
1466 			break;
1467 		default:
1468 			// Should not reach here
1469 			report_fail("ERROR : unexpected stage, %d",
1470 			       vmx_get_test_stage());
1471 			print_vmexit_info(exit_reason);
1472 			return VMX_TEST_VMEXIT;
1473 		}
1474 		return VMX_TEST_RESUME;
1475 	default:
1476 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1477 		print_vmexit_info(exit_reason);
1478 	}
1479 	return VMX_TEST_VMEXIT;
1480 }
1481 
1482 static int ept_exit_handler(union exit_reason exit_reason)
1483 {
1484 	return ept_exit_handler_common(exit_reason, false);
1485 }
1486 
1487 static int eptad_init(struct vmcs *vmcs)
1488 {
1489 	int r = ept_init_common(true);
1490 
1491 	if (r == VMX_TEST_EXIT)
1492 		return r;
1493 
1494 	if (!ept_ad_bits_supported()) {
1495 		printf("\tEPT A/D bits are not supported");
1496 		return VMX_TEST_EXIT;
1497 	}
1498 
1499 	return r;
1500 }
1501 
1502 static int pml_init(struct vmcs *vmcs)
1503 {
1504 	u32 ctrl_cpu;
1505 	int r = eptad_init(vmcs);
1506 
1507 	if (r == VMX_TEST_EXIT)
1508 		return r;
1509 
1510 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1511 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1512 		printf("\tPML is not supported");
1513 		return VMX_TEST_EXIT;
1514 	}
1515 
1516 	pml_log = alloc_page();
1517 	vmcs_write(PMLADDR, (u64)pml_log);
1518 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1519 
1520 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1521 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1522 
1523 	return VMX_TEST_START;
1524 }
1525 
1526 static void pml_main(void)
1527 {
1528 	int count = 0;
1529 
1530 	vmx_set_test_stage(0);
1531 	*((u32 *)data_page2) = 0x1;
1532 	vmcall();
1533 	report(vmx_get_test_stage() == 1, "PML - Dirty GPA Logging");
1534 
1535 	while (vmx_get_test_stage() == 1) {
1536 		vmcall();
1537 		*((u32 *)data_page2) = 0x1;
1538 		if (count++ > PML_INDEX)
1539 			break;
1540 	}
1541 	report(vmx_get_test_stage() == 2, "PML Full Event");
1542 }
1543 
1544 static void eptad_main(void)
1545 {
1546 	ept_common();
1547 }
1548 
1549 static int eptad_exit_handler(union exit_reason exit_reason)
1550 {
1551 	return ept_exit_handler_common(exit_reason, true);
1552 }
1553 
1554 #define TIMER_VECTOR	222
1555 
1556 static volatile bool timer_fired;
1557 
1558 static void timer_isr(isr_regs_t *regs)
1559 {
1560 	timer_fired = true;
1561 	apic_write(APIC_EOI, 0);
1562 }
1563 
1564 static int interrupt_init(struct vmcs *vmcs)
1565 {
1566 	msr_bmp_init();
1567 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1568 	handle_irq(TIMER_VECTOR, timer_isr);
1569 	return VMX_TEST_START;
1570 }
1571 
1572 static void interrupt_main(void)
1573 {
1574 	long long start, loops;
1575 
1576 	vmx_set_test_stage(0);
1577 
1578 	apic_write(APIC_LVTT, TIMER_VECTOR);
1579 	sti();
1580 
1581 	apic_write(APIC_TMICT, 1);
1582 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1583 		asm volatile ("nop");
1584 	report(timer_fired, "direct interrupt while running guest");
1585 
1586 	apic_write(APIC_TMICT, 0);
1587 	cli();
1588 	vmcall();
1589 	timer_fired = false;
1590 	apic_write(APIC_TMICT, 1);
1591 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1592 		asm volatile ("nop");
1593 	report(timer_fired, "intercepted interrupt while running guest");
1594 
1595 	sti();
1596 	apic_write(APIC_TMICT, 0);
1597 	cli();
1598 	vmcall();
1599 	timer_fired = false;
1600 	start = rdtsc();
1601 	apic_write(APIC_TMICT, 1000000);
1602 
1603 	safe_halt();
1604 
1605 	report(rdtsc() - start > 1000000 && timer_fired,
1606 	       "direct interrupt + hlt");
1607 
1608 	apic_write(APIC_TMICT, 0);
1609 	cli();
1610 	vmcall();
1611 	timer_fired = false;
1612 	start = rdtsc();
1613 	apic_write(APIC_TMICT, 1000000);
1614 
1615 	safe_halt();
1616 
1617 	report(rdtsc() - start > 10000 && timer_fired,
1618 	       "intercepted interrupt + hlt");
1619 
1620 	apic_write(APIC_TMICT, 0);
1621 	cli();
1622 	vmcall();
1623 	timer_fired = false;
1624 	start = rdtsc();
1625 	apic_write(APIC_TMICT, 1000000);
1626 
1627 	sti_nop();
1628 	vmcall();
1629 
1630 	report(rdtsc() - start > 10000 && timer_fired,
1631 	       "direct interrupt + activity state hlt");
1632 
1633 	apic_write(APIC_TMICT, 0);
1634 	cli();
1635 	vmcall();
1636 	timer_fired = false;
1637 	start = rdtsc();
1638 	apic_write(APIC_TMICT, 1000000);
1639 
1640 	sti_nop();
1641 	vmcall();
1642 
1643 	report(rdtsc() - start > 10000 && timer_fired,
1644 	       "intercepted interrupt + activity state hlt");
1645 
1646 	apic_write(APIC_TMICT, 0);
1647 	cli();
1648 	vmx_set_test_stage(7);
1649 	vmcall();
1650 	timer_fired = false;
1651 	apic_write(APIC_TMICT, 1);
1652 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1653 		asm volatile ("nop");
1654 	report(timer_fired,
1655 	       "running a guest with interrupt acknowledgement set");
1656 
1657 	apic_write(APIC_TMICT, 0);
1658 	sti();
1659 	timer_fired = false;
1660 	vmcall();
1661 	report(timer_fired, "Inject an event to a halted guest");
1662 }
1663 
1664 static int interrupt_exit_handler(union exit_reason exit_reason)
1665 {
1666 	u64 guest_rip = vmcs_read(GUEST_RIP);
1667 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1668 
1669 	switch (exit_reason.basic) {
1670 	case VMX_VMCALL:
1671 		switch (vmx_get_test_stage()) {
1672 		case 0:
1673 		case 2:
1674 		case 5:
1675 			vmcs_write(PIN_CONTROLS,
1676 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1677 			break;
1678 		case 7:
1679 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1680 			vmcs_write(PIN_CONTROLS,
1681 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1682 			break;
1683 		case 1:
1684 		case 3:
1685 			vmcs_write(PIN_CONTROLS,
1686 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1687 			break;
1688 		case 4:
1689 		case 6:
1690 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1691 			break;
1692 
1693 		case 8:
1694 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1695 			vmcs_write(ENT_INTR_INFO,
1696 				   TIMER_VECTOR |
1697 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1698 				   INTR_INFO_VALID_MASK);
1699 			break;
1700 		}
1701 		vmx_inc_test_stage();
1702 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1703 		return VMX_TEST_RESUME;
1704 	case VMX_EXTINT:
1705 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1706 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1707 			handle_external_interrupt(vector);
1708 		} else {
1709 			sti_nop_cli();
1710 		}
1711 		if (vmx_get_test_stage() >= 2)
1712 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1713 		return VMX_TEST_RESUME;
1714 	default:
1715 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1716 		print_vmexit_info(exit_reason);
1717 	}
1718 
1719 	return VMX_TEST_VMEXIT;
1720 }
1721 
1722 
1723 static volatile int nmi_fired;
1724 
1725 #define NMI_DELAY 100000000ULL
1726 
1727 static void nmi_isr(isr_regs_t *regs)
1728 {
1729 	nmi_fired = true;
1730 }
1731 
1732 static int nmi_hlt_init(struct vmcs *vmcs)
1733 {
1734 	msr_bmp_init();
1735 	handle_irq(NMI_VECTOR, nmi_isr);
1736 	vmcs_write(PIN_CONTROLS,
1737 		   vmcs_read(PIN_CONTROLS) & ~PIN_NMI);
1738 	vmcs_write(PIN_CONTROLS,
1739 		   vmcs_read(PIN_CONTROLS) & ~PIN_VIRT_NMI);
1740 	return VMX_TEST_START;
1741 }
1742 
1743 static void nmi_message_thread(void *data)
1744 {
1745     while (vmx_get_test_stage() != 1)
1746         pause();
1747 
1748     delay(NMI_DELAY);
1749     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1750 
1751     while (vmx_get_test_stage() != 2)
1752         pause();
1753 
1754     delay(NMI_DELAY);
1755     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1756 }
1757 
1758 static void nmi_hlt_main(void)
1759 {
1760     long long start;
1761 
1762     if (cpu_count() < 2) {
1763         report_skip("%s : CPU count < 2", __func__);
1764         vmx_set_test_stage(-1);
1765         return;
1766     }
1767 
1768     vmx_set_test_stage(0);
1769     on_cpu_async(1, nmi_message_thread, NULL);
1770     start = rdtsc();
1771     vmx_set_test_stage(1);
1772     asm volatile ("hlt");
1773     report((rdtsc() - start > NMI_DELAY) && nmi_fired,
1774             "direct NMI + hlt");
1775     if (!nmi_fired)
1776         vmx_set_test_stage(-1);
1777     nmi_fired = false;
1778 
1779     vmcall();
1780 
1781     start = rdtsc();
1782     vmx_set_test_stage(2);
1783     asm volatile ("hlt");
1784     report((rdtsc() - start > NMI_DELAY) && !nmi_fired,
1785             "intercepted NMI + hlt");
1786     if (nmi_fired) {
1787         report(!nmi_fired, "intercepted NMI was dispatched");
1788         vmx_set_test_stage(-1);
1789         return;
1790     }
1791     vmx_set_test_stage(3);
1792 }
1793 
1794 static int nmi_hlt_exit_handler(union exit_reason exit_reason)
1795 {
1796     u64 guest_rip = vmcs_read(GUEST_RIP);
1797     u32 insn_len = vmcs_read(EXI_INST_LEN);
1798 
1799     switch (vmx_get_test_stage()) {
1800     case 1:
1801         if (exit_reason.basic != VMX_VMCALL) {
1802             report_fail("VMEXIT not due to vmcall. Exit reason 0x%x",
1803                         exit_reason.full);
1804             print_vmexit_info(exit_reason);
1805             return VMX_TEST_VMEXIT;
1806         }
1807 
1808         vmcs_write(PIN_CONTROLS,
1809                vmcs_read(PIN_CONTROLS) | PIN_NMI);
1810         vmcs_write(PIN_CONTROLS,
1811                vmcs_read(PIN_CONTROLS) | PIN_VIRT_NMI);
1812         vmcs_write(GUEST_RIP, guest_rip + insn_len);
1813         break;
1814 
1815     case 2:
1816         if (exit_reason.basic != VMX_EXC_NMI) {
1817             report_fail("VMEXIT not due to NMI intercept. Exit reason 0x%x",
1818                         exit_reason.full);
1819             print_vmexit_info(exit_reason);
1820             return VMX_TEST_VMEXIT;
1821         }
1822         report_pass("NMI intercept while running guest");
1823         vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1824         break;
1825 
1826     case 3:
1827         break;
1828 
1829     default:
1830         return VMX_TEST_VMEXIT;
1831     }
1832 
1833     if (vmx_get_test_stage() == 3)
1834         return VMX_TEST_VMEXIT;
1835 
1836     return VMX_TEST_RESUME;
1837 }
1838 
1839 
1840 static int dbgctls_init(struct vmcs *vmcs)
1841 {
1842 	u64 dr7 = 0x402;
1843 	u64 zero = 0;
1844 
1845 	msr_bmp_init();
1846 	asm volatile(
1847 		"mov %0,%%dr0\n\t"
1848 		"mov %0,%%dr1\n\t"
1849 		"mov %0,%%dr2\n\t"
1850 		"mov %1,%%dr7\n\t"
1851 		: : "r" (zero), "r" (dr7));
1852 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1853 	vmcs_write(GUEST_DR7, 0x404);
1854 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1855 
1856 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1857 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1858 
1859 	return VMX_TEST_START;
1860 }
1861 
1862 static void dbgctls_main(void)
1863 {
1864 	u64 dr7, debugctl;
1865 
1866 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1867 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1868 	/* Commented out: KVM does not support DEBUGCTL so far */
1869 	(void)debugctl;
1870 	report(dr7 == 0x404, "Load debug controls" /* && debugctl == 0x2 */);
1871 
1872 	dr7 = 0x408;
1873 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1874 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1875 
1876 	vmx_set_test_stage(0);
1877 	vmcall();
1878 	report(vmx_get_test_stage() == 1, "Save debug controls");
1879 
1880 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1881 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1882 		printf("\tDebug controls are always loaded/saved\n");
1883 		return;
1884 	}
1885 	vmx_set_test_stage(2);
1886 	vmcall();
1887 
1888 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1889 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1890 	/* Commented out: KVM does not support DEBUGCTL so far */
1891 	(void)debugctl;
1892 	report(dr7 == 0x402,
1893 	       "Guest=host debug controls" /* && debugctl == 0x1 */);
1894 
1895 	dr7 = 0x408;
1896 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1897 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1898 
1899 	vmx_set_test_stage(3);
1900 	vmcall();
1901 	report(vmx_get_test_stage() == 4, "Don't save debug controls");
1902 }
1903 
1904 static int dbgctls_exit_handler(union exit_reason exit_reason)
1905 {
1906 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1907 	u64 guest_rip = vmcs_read(GUEST_RIP);
1908 	u64 dr7, debugctl;
1909 
1910 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1911 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1912 
1913 	switch (exit_reason.basic) {
1914 	case VMX_VMCALL:
1915 		switch (vmx_get_test_stage()) {
1916 		case 0:
1917 			if (dr7 == 0x400 && debugctl == 0 &&
1918 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1919 			    Commented out: KVM does not support DEBUGCTL so far
1920 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1921 				vmx_inc_test_stage();
1922 			break;
1923 		case 2:
1924 			dr7 = 0x402;
1925 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1926 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1927 			vmcs_write(GUEST_DR7, 0x404);
1928 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1929 
1930 			vmcs_write(ENT_CONTROLS,
1931 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1932 			vmcs_write(EXI_CONTROLS,
1933 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1934 			break;
1935 		case 3:
1936 			if (dr7 == 0x400 && debugctl == 0 &&
1937 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1938 			    Commented out: KVM does not support DEBUGCTL so far
1939 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1940 				vmx_inc_test_stage();
1941 			break;
1942 		}
1943 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1944 		return VMX_TEST_RESUME;
1945 	default:
1946 		report_fail("Unknown exit reason, %d", exit_reason.full);
1947 		print_vmexit_info(exit_reason);
1948 	}
1949 	return VMX_TEST_VMEXIT;
1950 }
1951 
1952 struct vmx_msr_entry {
1953 	u32 index;
1954 	u32 reserved;
1955 	u64 value;
1956 } __attribute__((packed));
1957 
1958 #define MSR_MAGIC 0x31415926
1959 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1960 
1961 static int msr_switch_init(struct vmcs *vmcs)
1962 {
1963 	msr_bmp_init();
1964 	exit_msr_store = alloc_page();
1965 	exit_msr_load = alloc_page();
1966 	entry_msr_load = alloc_page();
1967 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1968 	entry_msr_load[0].value = MSR_MAGIC;
1969 
1970 	vmx_set_test_stage(1);
1971 	vmcs_write(ENT_MSR_LD_CNT, 1);
1972 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1973 	vmcs_write(EXI_MSR_ST_CNT, 1);
1974 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1975 	vmcs_write(EXI_MSR_LD_CNT, 1);
1976 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1977 	return VMX_TEST_START;
1978 }
1979 
1980 static void msr_switch_main(void)
1981 {
1982 	if (vmx_get_test_stage() == 1) {
1983 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC,
1984 		       "VM entry MSR load");
1985 		vmx_set_test_stage(2);
1986 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
1987 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
1988 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
1989 		exit_msr_load[0].value = MSR_MAGIC + 2;
1990 	}
1991 	vmcall();
1992 }
1993 
1994 static int msr_switch_exit_handler(union exit_reason exit_reason)
1995 {
1996 	if (exit_reason.basic == VMX_VMCALL && vmx_get_test_stage() == 2) {
1997 		report(exit_msr_store[0].value == MSR_MAGIC + 1,
1998 		       "VM exit MSR store");
1999 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2,
2000 		       "VM exit MSR load");
2001 		vmx_set_test_stage(3);
2002 		entry_msr_load[0].index = MSR_FS_BASE;
2003 		return VMX_TEST_RESUME;
2004 	}
2005 	printf("ERROR %s: unexpected stage=%u or reason=0x%x\n",
2006 		__func__, vmx_get_test_stage(), exit_reason.full);
2007 	return VMX_TEST_EXIT;
2008 }
2009 
2010 static int msr_switch_entry_failure(struct vmentry_result *result)
2011 {
2012 	if (result->vm_fail) {
2013 		printf("ERROR %s: VM-Fail on %s\n", __func__, result->instr);
2014 		return VMX_TEST_EXIT;
2015 	}
2016 
2017 	if (result->exit_reason.failed_vmentry &&
2018 	    result->exit_reason.basic == VMX_FAIL_MSR &&
2019 	    vmx_get_test_stage() == 3) {
2020 		report(vmcs_read(EXI_QUALIFICATION) == 1,
2021 		       "VM entry MSR load: try to load FS_BASE");
2022 		return VMX_TEST_VMEXIT;
2023 	}
2024 	printf("ERROR %s: unexpected stage=%u or reason=%x\n",
2025 		__func__, vmx_get_test_stage(), result->exit_reason.full);
2026 	return VMX_TEST_EXIT;
2027 }
2028 
2029 static int vmmcall_init(struct vmcs *vmcs)
2030 {
2031 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
2032 	return VMX_TEST_START;
2033 }
2034 
2035 static void vmmcall_main(void)
2036 {
2037 	asm volatile(
2038 		"mov $0xABCD, %%rax\n\t"
2039 		"vmmcall\n\t"
2040 		::: "rax");
2041 
2042 	report_fail("VMMCALL");
2043 }
2044 
2045 static int vmmcall_exit_handler(union exit_reason exit_reason)
2046 {
2047 	switch (exit_reason.basic) {
2048 	case VMX_VMCALL:
2049 		printf("here\n");
2050 		report_fail("VMMCALL triggers #UD");
2051 		break;
2052 	case VMX_EXC_NMI:
2053 		report((vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR,
2054 		       "VMMCALL triggers #UD");
2055 		break;
2056 	default:
2057 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2058 		print_vmexit_info(exit_reason);
2059 	}
2060 
2061 	return VMX_TEST_VMEXIT;
2062 }
2063 
2064 static int disable_rdtscp_init(struct vmcs *vmcs)
2065 {
2066 	u32 ctrl_cpu1;
2067 
2068 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
2069 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
2070 		ctrl_cpu1 &= ~CPU_RDTSCP;
2071 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
2072 	}
2073 
2074 	return VMX_TEST_START;
2075 }
2076 
2077 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
2078 {
2079 	switch (vmx_get_test_stage()) {
2080 	case 0:
2081 		report_pass("RDTSCP triggers #UD");
2082 		vmx_inc_test_stage();
2083 		regs->rip += 3;
2084 		break;
2085 	case 2:
2086 		report_pass("RDPID triggers #UD");
2087 		vmx_inc_test_stage();
2088 		regs->rip += 4;
2089 		break;
2090 	}
2091 	return;
2092 
2093 }
2094 
2095 static void disable_rdtscp_main(void)
2096 {
2097 	/* Test that #UD is properly injected in L2.  */
2098 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2099 
2100 	vmx_set_test_stage(0);
2101 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2102 	vmcall();
2103 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2104 
2105 	handle_exception(UD_VECTOR, 0);
2106 	vmcall();
2107 }
2108 
2109 static int disable_rdtscp_exit_handler(union exit_reason exit_reason)
2110 {
2111 	switch (exit_reason.basic) {
2112 	case VMX_VMCALL:
2113 		switch (vmx_get_test_stage()) {
2114 		case 0:
2115 			report_fail("RDTSCP triggers #UD");
2116 			vmx_inc_test_stage();
2117 			/* fallthrough */
2118 		case 1:
2119 			vmx_inc_test_stage();
2120 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2121 			return VMX_TEST_RESUME;
2122 		case 2:
2123 			report_fail("RDPID triggers #UD");
2124 			break;
2125 		}
2126 		break;
2127 
2128 	default:
2129 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2130 		print_vmexit_info(exit_reason);
2131 	}
2132 	return VMX_TEST_VMEXIT;
2133 }
2134 
2135 static void exit_monitor_from_l2_main(void)
2136 {
2137 	printf("Calling exit(0) from l2...\n");
2138 	exit(0);
2139 }
2140 
2141 static int exit_monitor_from_l2_handler(union exit_reason exit_reason)
2142 {
2143 	report_fail("The guest should have killed the VMM");
2144 	return VMX_TEST_EXIT;
2145 }
2146 
2147 static void assert_exit_reason(u64 expected)
2148 {
2149 	u64 actual = vmcs_read(EXI_REASON);
2150 
2151 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2152 			   exit_reason_description(expected),
2153 			   exit_reason_description(actual));
2154 }
2155 
2156 static void skip_exit_insn(void)
2157 {
2158 	u64 guest_rip = vmcs_read(GUEST_RIP);
2159 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2160 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2161 }
2162 
2163 static void skip_exit_vmcall(void)
2164 {
2165 	assert_exit_reason(VMX_VMCALL);
2166 	skip_exit_insn();
2167 }
2168 
2169 static void v2_null_test_guest(void)
2170 {
2171 }
2172 
2173 static void v2_null_test(void)
2174 {
2175 	test_set_guest(v2_null_test_guest);
2176 	enter_guest();
2177 	report_pass(__func__);
2178 }
2179 
2180 static void v2_multiple_entries_test_guest(void)
2181 {
2182 	vmx_set_test_stage(1);
2183 	vmcall();
2184 	vmx_set_test_stage(2);
2185 }
2186 
2187 static void v2_multiple_entries_test(void)
2188 {
2189 	test_set_guest(v2_multiple_entries_test_guest);
2190 	enter_guest();
2191 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2192 	skip_exit_vmcall();
2193 	enter_guest();
2194 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2195 	report_pass(__func__);
2196 }
2197 
2198 static int fixture_test_data = 1;
2199 
2200 static void fixture_test_teardown(void *data)
2201 {
2202 	*((int *) data) = 1;
2203 }
2204 
2205 static void fixture_test_guest(void)
2206 {
2207 	fixture_test_data++;
2208 }
2209 
2210 
2211 static void fixture_test_setup(void)
2212 {
2213 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2214 			   "fixture_test_teardown didn't run?!");
2215 	fixture_test_data = 2;
2216 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2217 	test_set_guest(fixture_test_guest);
2218 }
2219 
2220 static void fixture_test_case1(void)
2221 {
2222 	fixture_test_setup();
2223 	TEST_ASSERT_EQ(2, fixture_test_data);
2224 	enter_guest();
2225 	TEST_ASSERT_EQ(3, fixture_test_data);
2226 	report_pass(__func__);
2227 }
2228 
2229 static void fixture_test_case2(void)
2230 {
2231 	fixture_test_setup();
2232 	TEST_ASSERT_EQ(2, fixture_test_data);
2233 	enter_guest();
2234 	TEST_ASSERT_EQ(3, fixture_test_data);
2235 	report_pass(__func__);
2236 }
2237 
2238 enum ept_access_op {
2239 	OP_READ,
2240 	OP_WRITE,
2241 	OP_EXEC,
2242 	OP_FLUSH_TLB,
2243 	OP_EXIT,
2244 };
2245 
2246 static struct ept_access_test_data {
2247 	unsigned long gpa;
2248 	unsigned long *gva;
2249 	unsigned long hpa;
2250 	unsigned long *hva;
2251 	enum ept_access_op op;
2252 } ept_access_test_data;
2253 
2254 extern unsigned char ret42_start;
2255 extern unsigned char ret42_end;
2256 
2257 /* Returns 42. */
2258 asm(
2259 	".align 64\n"
2260 	"ret42_start:\n"
2261 	"mov $42, %eax\n"
2262 	"ret\n"
2263 	"ret42_end:\n"
2264 );
2265 
2266 static void
2267 diagnose_ept_violation_qual(u64 expected, u64 actual)
2268 {
2269 
2270 #define DIAGNOSE(flag)							\
2271 do {									\
2272 	if ((expected & flag) != (actual & flag))			\
2273 		printf(#flag " %sexpected\n",				\
2274 		       (expected & flag) ? "" : "un");			\
2275 } while (0)
2276 
2277 	DIAGNOSE(EPT_VLT_RD);
2278 	DIAGNOSE(EPT_VLT_WR);
2279 	DIAGNOSE(EPT_VLT_FETCH);
2280 	DIAGNOSE(EPT_VLT_PERM_RD);
2281 	DIAGNOSE(EPT_VLT_PERM_WR);
2282 	DIAGNOSE(EPT_VLT_PERM_EX);
2283 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2284 	DIAGNOSE(EPT_VLT_PADDR);
2285 
2286 #undef DIAGNOSE
2287 }
2288 
2289 static void do_ept_access_op(enum ept_access_op op)
2290 {
2291 	ept_access_test_data.op = op;
2292 	enter_guest();
2293 }
2294 
2295 /*
2296  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2297  * needed by tests that modify guest PTEs.
2298  */
2299 static void ept_access_test_guest_flush_tlb(void)
2300 {
2301 	do_ept_access_op(OP_FLUSH_TLB);
2302 	skip_exit_vmcall();
2303 }
2304 
2305 /*
2306  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2307  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2308  * a huge page.
2309  */
2310 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2311 				 unsigned long clear, unsigned long set)
2312 {
2313 	struct ept_access_test_data *data = &ept_access_test_data;
2314 	unsigned long orig_pte;
2315 	unsigned long pte;
2316 
2317 	/* Screw with the mapping at the requested level. */
2318 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2319 	pte = orig_pte;
2320 	if (mkhuge)
2321 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2322 	else
2323 		pte = orig_pte;
2324 	pte = (pte & ~clear) | set;
2325 	set_ept_pte(pml4, gpa, level, pte);
2326 	invept(INVEPT_SINGLE, eptp);
2327 
2328 	return orig_pte;
2329 }
2330 
2331 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2332 {
2333 	set_ept_pte(pml4, gpa, level, orig_pte);
2334 	invept(INVEPT_SINGLE, eptp);
2335 }
2336 
2337 static void do_ept_violation(bool leaf, enum ept_access_op op,
2338 			     u64 expected_qual, u64 expected_paddr)
2339 {
2340 	u64 qual;
2341 
2342 	/* Try the access and observe the violation. */
2343 	do_ept_access_op(op);
2344 
2345 	assert_exit_reason(VMX_EPT_VIOLATION);
2346 
2347 	qual = vmcs_read(EXI_QUALIFICATION);
2348 
2349 	/* Mask undefined bits (which may later be defined in certain cases). */
2350 	qual &= ~(EPT_VLT_GUEST_USER | EPT_VLT_GUEST_RW | EPT_VLT_GUEST_EX |
2351 		 EPT_VLT_PERM_USER_EX);
2352 
2353 	diagnose_ept_violation_qual(expected_qual, qual);
2354 	TEST_EXPECT_EQ(expected_qual, qual);
2355 
2356 	#if 0
2357 	/* Disable for now otherwise every test will fail */
2358 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2359 		       (unsigned long) (
2360 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2361 	#endif
2362 	/*
2363 	 * TODO: tests that probe expected_paddr in pages other than the one at
2364 	 * the beginning of the 1g region.
2365 	 */
2366 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2367 }
2368 
2369 static void
2370 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2371 			      unsigned long set, enum ept_access_op op,
2372 			      u64 expected_qual)
2373 {
2374 	struct ept_access_test_data *data = &ept_access_test_data;
2375 	unsigned long orig_pte;
2376 
2377 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2378 
2379 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2380 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2381 					 data->gpa);
2382 
2383 	/* Fix the violation and resume the op loop. */
2384 	ept_untwiddle(data->gpa, level, orig_pte);
2385 	enter_guest();
2386 	skip_exit_vmcall();
2387 }
2388 
2389 static void
2390 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2391 		       enum ept_access_op op, u64 expected_qual)
2392 {
2393 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2394 				      expected_qual);
2395 	if (ept_huge_pages_supported(level))
2396 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2397 					      expected_qual);
2398 }
2399 
2400 static void ept_violation(unsigned long clear, unsigned long set,
2401 			  enum ept_access_op op, u64 expected_qual)
2402 {
2403 	ept_violation_at_level(1, clear, set, op, expected_qual);
2404 	ept_violation_at_level(2, clear, set, op, expected_qual);
2405 	ept_violation_at_level(3, clear, set, op, expected_qual);
2406 	ept_violation_at_level(4, clear, set, op, expected_qual);
2407 }
2408 
2409 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2410 				       u64 expected_qual)
2411 {
2412 	ept_violation(EPT_PRESENT, access, op,
2413 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2414 }
2415 
2416 /*
2417  * For translations that don't involve a GVA, that is physical address (paddr)
2418  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2419  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2420  * translations don't involve GVAs, such as when the hardware does the guest
2421  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2422  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2423  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2424  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2425  *
2426  * Note that paddr violations can also be triggered by loading PAE page tables
2427  * with wonky addresses. We don't test that yet.
2428  *
2429  * This function modifies the EPT entry that maps the GPA that the guest page
2430  * table entry mapping ept_access_test_data.gva resides on.
2431  *
2432  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2433  *
2434  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2435  *
2436  *	@op		Guest operation to perform with
2437  *			ept_access_test_data.gva.
2438  *
2439  *	@expect_violation
2440  *			Is a violation expected during the paddr access?
2441  *
2442  *	@expected_qual	Expected qualification for the EPT violation.
2443  *			EPT_VLT_PADDR should be clear.
2444  */
2445 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2446 			     enum ept_access_op op, bool expect_violation,
2447 			     u64 expected_qual)
2448 {
2449 	struct ept_access_test_data *data = &ept_access_test_data;
2450 	unsigned long *ptep;
2451 	unsigned long gpa;
2452 	unsigned long orig_epte;
2453 	unsigned long epte;
2454 	int i;
2455 
2456 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2457 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2458 	TEST_ASSERT(ptep);
2459 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2460 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2461 	ept_access_test_guest_flush_tlb();
2462 
2463 	/*
2464 	 * Now modify the access bits on the EPT entry for the GPA that the
2465 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2466 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2467 	 * constructed our test such that those other 511 PTEs aren't used by
2468 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2469 	 * PTE we're modifying is at the beginning of a 4K page and the
2470 	 * following 511 entries are also under our control (and not touched by
2471 	 * the guest).
2472 	 */
2473 	gpa = virt_to_phys(ptep);
2474 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2475 	/*
2476 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2477 	 * otherwise our level=1 twiddling below will fail. We use the
2478 	 * identity map (gpa = gpa) since page tables are shared with the host.
2479 	 */
2480 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2481 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2482 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2483 
2484 	if (expect_violation) {
2485 		do_ept_violation(/*leaf=*/true, op,
2486 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2487 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2488 		do_ept_access_op(op);
2489 	} else {
2490 		do_ept_access_op(op);
2491 		if (ept_ad_enabled()) {
2492 			for (i = EPT_PAGE_LEVEL; i > 0; i--) {
2493 				TEST_ASSERT(get_ept_pte(pml4, gpa, i, &epte));
2494 				TEST_ASSERT(epte & EPT_ACCESS_FLAG);
2495 				if (i == 1)
2496 					TEST_ASSERT(epte & EPT_DIRTY_FLAG);
2497 				else
2498 					TEST_ASSERT_EQ(epte & EPT_DIRTY_FLAG, 0);
2499 			}
2500 		}
2501 
2502 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2503 	}
2504 
2505 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2506 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2507 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2508 
2509 	skip_exit_vmcall();
2510 }
2511 
2512 static void ept_access_allowed_paddr(unsigned long ept_access,
2513 				     unsigned long pte_ad,
2514 				     enum ept_access_op op)
2515 {
2516 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2517 			 /*expected_qual=*/-1);
2518 }
2519 
2520 static void ept_access_violation_paddr(unsigned long ept_access,
2521 				       unsigned long pte_ad,
2522 				       enum ept_access_op op,
2523 				       u64 expected_qual)
2524 {
2525 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2526 			 expected_qual);
2527 }
2528 
2529 
2530 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2531 					unsigned long clear,
2532 					unsigned long set,
2533 					enum ept_access_op op)
2534 {
2535 	struct ept_access_test_data *data = &ept_access_test_data;
2536 	unsigned long orig_pte;
2537 
2538 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2539 
2540 	/* No violation. Should proceed to vmcall. */
2541 	do_ept_access_op(op);
2542 	skip_exit_vmcall();
2543 
2544 	ept_untwiddle(data->gpa, level, orig_pte);
2545 }
2546 
2547 static void ept_allowed_at_level(int level, unsigned long clear,
2548 				 unsigned long set, enum ept_access_op op)
2549 {
2550 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2551 	if (ept_huge_pages_supported(level))
2552 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2553 }
2554 
2555 static void ept_allowed(unsigned long clear, unsigned long set,
2556 			enum ept_access_op op)
2557 {
2558 	ept_allowed_at_level(1, clear, set, op);
2559 	ept_allowed_at_level(2, clear, set, op);
2560 	ept_allowed_at_level(3, clear, set, op);
2561 	ept_allowed_at_level(4, clear, set, op);
2562 }
2563 
2564 static void ept_ignored_bit(int bit)
2565 {
2566 	/* Set the bit. */
2567 	ept_allowed(0, 1ul << bit, OP_READ);
2568 	ept_allowed(0, 1ul << bit, OP_WRITE);
2569 	ept_allowed(0, 1ul << bit, OP_EXEC);
2570 
2571 	/* Clear the bit. */
2572 	ept_allowed(1ul << bit, 0, OP_READ);
2573 	ept_allowed(1ul << bit, 0, OP_WRITE);
2574 	ept_allowed(1ul << bit, 0, OP_EXEC);
2575 }
2576 
2577 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2578 {
2579 	ept_allowed(EPT_PRESENT, access, op);
2580 }
2581 
2582 
2583 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2584 					     unsigned long clear,
2585 					     unsigned long set,
2586 					     enum ept_access_op op)
2587 {
2588 	struct ept_access_test_data *data = &ept_access_test_data;
2589 	unsigned long orig_pte;
2590 
2591 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2592 
2593 	do_ept_access_op(op);
2594 	assert_exit_reason(VMX_EPT_MISCONFIG);
2595 
2596 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2597 	#if 0
2598 	/* broken: */
2599 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2600 	#endif
2601 	#if 0
2602 	/*
2603 	 * broken:
2604 	 * According to description of exit qual for EPT violation,
2605 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2606 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2607 	 * to be set for msiconfig.
2608 	 */
2609 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2610 		       (unsigned long) (
2611 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2612 	#endif
2613 
2614 	/* Fix the violation and resume the op loop. */
2615 	ept_untwiddle(data->gpa, level, orig_pte);
2616 	enter_guest();
2617 	skip_exit_vmcall();
2618 }
2619 
2620 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2621 					  unsigned long clear,
2622 					  unsigned long set)
2623 {
2624 	/* The op shouldn't matter (read, write, exec), so try them all! */
2625 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2626 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2627 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2628 }
2629 
2630 static void ept_misconfig_at_level(int level, unsigned long clear,
2631 				   unsigned long set)
2632 {
2633 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2634 	if (ept_huge_pages_supported(level))
2635 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2636 }
2637 
2638 static void ept_misconfig(unsigned long clear, unsigned long set)
2639 {
2640 	ept_misconfig_at_level(1, clear, set);
2641 	ept_misconfig_at_level(2, clear, set);
2642 	ept_misconfig_at_level(3, clear, set);
2643 	ept_misconfig_at_level(4, clear, set);
2644 }
2645 
2646 static void ept_access_misconfig(unsigned long access)
2647 {
2648 	ept_misconfig(EPT_PRESENT, access);
2649 }
2650 
2651 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2652 {
2653 	/* Setting the bit causes a misconfig. */
2654 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2655 
2656 	/* Making the entry non-present turns reserved bits into ignored. */
2657 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2658 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2659 }
2660 
2661 static void ept_reserved_bit_at_level_huge(int level, int bit)
2662 {
2663 	/* Setting the bit causes a misconfig. */
2664 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2665 
2666 	/* Making the entry non-present turns reserved bits into ignored. */
2667 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2668 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2669 }
2670 
2671 static void ept_reserved_bit_at_level(int level, int bit)
2672 {
2673 	/* Setting the bit causes a misconfig. */
2674 	ept_misconfig_at_level(level, 0, 1ul << bit);
2675 
2676 	/* Making the entry non-present turns reserved bits into ignored. */
2677 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2678 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2679 }
2680 
2681 static void ept_reserved_bit(int bit)
2682 {
2683 	ept_reserved_bit_at_level(1, bit);
2684 	ept_reserved_bit_at_level(2, bit);
2685 	ept_reserved_bit_at_level(3, bit);
2686 	ept_reserved_bit_at_level(4, bit);
2687 }
2688 
2689 #define PAGE_2M_ORDER 9
2690 #define PAGE_1G_ORDER 18
2691 
2692 static void *get_1g_page(void)
2693 {
2694 	static void *alloc;
2695 
2696 	if (!alloc)
2697 		alloc = alloc_pages(PAGE_1G_ORDER);
2698 	return alloc;
2699 }
2700 
2701 static void ept_access_test_teardown(void *unused)
2702 {
2703 	/* Exit the guest cleanly. */
2704 	do_ept_access_op(OP_EXIT);
2705 }
2706 
2707 static void ept_access_test_guest(void)
2708 {
2709 	struct ept_access_test_data *data = &ept_access_test_data;
2710 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2711 
2712 	while (true) {
2713 		switch (data->op) {
2714 		case OP_READ:
2715 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2716 			break;
2717 		case OP_WRITE:
2718 			*data->gva = MAGIC_VAL_2;
2719 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2720 			*data->gva = MAGIC_VAL_1;
2721 			break;
2722 		case OP_EXEC:
2723 			TEST_ASSERT_EQ(42, code());
2724 			break;
2725 		case OP_FLUSH_TLB:
2726 			write_cr3(read_cr3());
2727 			break;
2728 		case OP_EXIT:
2729 			return;
2730 		default:
2731 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2732 		}
2733 		vmcall();
2734 	}
2735 }
2736 
2737 static void ept_access_test_setup(void)
2738 {
2739 	struct ept_access_test_data *data = &ept_access_test_data;
2740 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2741 	unsigned long size = npages * PAGE_SIZE;
2742 	unsigned long *page_table = current_page_table();
2743 	unsigned long pte;
2744 
2745 	if (setup_ept(false))
2746 		test_skip("EPT not supported");
2747 
2748 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2749 	if (cpuid_maxphyaddr() < 40)
2750 		test_skip("Test needs MAXPHYADDR >= 40");
2751 
2752 	test_set_guest(ept_access_test_guest);
2753 	test_add_teardown(ept_access_test_teardown, NULL);
2754 
2755 	data->hva = get_1g_page();
2756 	TEST_ASSERT(data->hva);
2757 	data->hpa = virt_to_phys(data->hva);
2758 
2759 	data->gpa = 1ul << 39;
2760 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2761 				   size);
2762 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2763 	install_pages(page_table, data->gpa, size, data->gva);
2764 
2765 	/*
2766 	 * Make sure nothing's mapped here so the tests that screw with the
2767 	 * pml4 entry don't inadvertently break something.
2768 	 */
2769 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2770 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2771 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2772 
2773 	data->hva[0] = MAGIC_VAL_1;
2774 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2775 }
2776 
2777 static void ept_access_test_not_present(void)
2778 {
2779 	ept_access_test_setup();
2780 	/* --- */
2781 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2782 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2783 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2784 }
2785 
2786 static void ept_access_test_read_only(void)
2787 {
2788 	ept_access_test_setup();
2789 
2790 	/* r-- */
2791 	ept_access_allowed(EPT_RA, OP_READ);
2792 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2793 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2794 }
2795 
2796 static void ept_access_test_write_only(void)
2797 {
2798 	ept_access_test_setup();
2799 	/* -w- */
2800 	ept_access_misconfig(EPT_WA);
2801 }
2802 
2803 static void ept_access_test_read_write(void)
2804 {
2805 	ept_access_test_setup();
2806 	/* rw- */
2807 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2808 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2809 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2810 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2811 }
2812 
2813 
2814 static void ept_access_test_execute_only(void)
2815 {
2816 	ept_access_test_setup();
2817 	/* --x */
2818 	if (ept_execute_only_supported()) {
2819 		ept_access_violation(EPT_EA, OP_READ,
2820 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2821 		ept_access_violation(EPT_EA, OP_WRITE,
2822 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2823 		ept_access_allowed(EPT_EA, OP_EXEC);
2824 	} else {
2825 		ept_access_misconfig(EPT_EA);
2826 	}
2827 }
2828 
2829 static void ept_access_test_read_execute(void)
2830 {
2831 	ept_access_test_setup();
2832 	/* r-x */
2833 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2834 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2835 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2836 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2837 }
2838 
2839 static void ept_access_test_write_execute(void)
2840 {
2841 	ept_access_test_setup();
2842 	/* -wx */
2843 	ept_access_misconfig(EPT_WA | EPT_EA);
2844 }
2845 
2846 static void ept_access_test_read_write_execute(void)
2847 {
2848 	ept_access_test_setup();
2849 	/* rwx */
2850 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2851 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2852 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2853 }
2854 
2855 static void ept_access_test_reserved_bits(void)
2856 {
2857 	int i;
2858 	int maxphyaddr;
2859 
2860 	ept_access_test_setup();
2861 
2862 	/* Reserved bits above maxphyaddr. */
2863 	maxphyaddr = cpuid_maxphyaddr();
2864 	for (i = maxphyaddr; i <= 51; i++) {
2865 		report_prefix_pushf("reserved_bit=%d", i);
2866 		ept_reserved_bit(i);
2867 		report_prefix_pop();
2868 	}
2869 
2870 	/* Level-specific reserved bits. */
2871 	ept_reserved_bit_at_level_nohuge(2, 3);
2872 	ept_reserved_bit_at_level_nohuge(2, 4);
2873 	ept_reserved_bit_at_level_nohuge(2, 5);
2874 	ept_reserved_bit_at_level_nohuge(2, 6);
2875 	/* 2M alignment. */
2876 	for (i = 12; i < 20; i++) {
2877 		report_prefix_pushf("reserved_bit=%d", i);
2878 		ept_reserved_bit_at_level_huge(2, i);
2879 		report_prefix_pop();
2880 	}
2881 	ept_reserved_bit_at_level_nohuge(3, 3);
2882 	ept_reserved_bit_at_level_nohuge(3, 4);
2883 	ept_reserved_bit_at_level_nohuge(3, 5);
2884 	ept_reserved_bit_at_level_nohuge(3, 6);
2885 	/* 1G alignment. */
2886 	for (i = 12; i < 29; i++) {
2887 		report_prefix_pushf("reserved_bit=%d", i);
2888 		ept_reserved_bit_at_level_huge(3, i);
2889 		report_prefix_pop();
2890 	}
2891 	ept_reserved_bit_at_level(4, 3);
2892 	ept_reserved_bit_at_level(4, 4);
2893 	ept_reserved_bit_at_level(4, 5);
2894 	ept_reserved_bit_at_level(4, 6);
2895 	ept_reserved_bit_at_level(4, 7);
2896 }
2897 
2898 static void ept_access_test_ignored_bits(void)
2899 {
2900 	ept_access_test_setup();
2901 	/*
2902 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2903 	 * far as translation is concerned even if AD bits are enabled in the
2904 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2905 	 * control is 0.
2906 	 */
2907 	ept_ignored_bit(8);
2908 	ept_ignored_bit(9);
2909 	ept_ignored_bit(10);
2910 	ept_ignored_bit(11);
2911 	ept_ignored_bit(52);
2912 	ept_ignored_bit(53);
2913 	ept_ignored_bit(54);
2914 	ept_ignored_bit(55);
2915 	ept_ignored_bit(56);
2916 	ept_ignored_bit(57);
2917 	ept_ignored_bit(58);
2918 	ept_ignored_bit(59);
2919 	ept_ignored_bit(60);
2920 	ept_ignored_bit(61);
2921 	ept_ignored_bit(62);
2922 	ept_ignored_bit(63);
2923 }
2924 
2925 static void ept_access_test_paddr_not_present_ad_disabled(void)
2926 {
2927 	ept_access_test_setup();
2928 	ept_disable_ad_bits();
2929 
2930 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
2931 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
2932 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
2933 }
2934 
2935 static void ept_access_test_paddr_not_present_ad_enabled(void)
2936 {
2937 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
2938 
2939 	ept_access_test_setup();
2940 	ept_enable_ad_bits_or_skip_test();
2941 
2942 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
2943 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
2944 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
2945 }
2946 
2947 static void ept_access_test_paddr_read_only_ad_disabled(void)
2948 {
2949 	/*
2950 	 * When EPT AD bits are disabled, all accesses to guest paging
2951 	 * structures are reported separately as a read and (after
2952 	 * translation of the GPA to host physical address) a read+write
2953 	 * if the A/D bits have to be set.
2954 	 */
2955 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2956 
2957 	ept_access_test_setup();
2958 	ept_disable_ad_bits();
2959 
2960 	/* Can't update A bit, so all accesses fail. */
2961 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2962 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
2963 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
2964 	/* AD bits disabled, so only writes try to update the D bit. */
2965 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
2966 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
2967 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
2968 	/* Both A and D already set, so read-only is OK. */
2969 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
2970 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
2971 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
2972 }
2973 
2974 static void ept_access_test_paddr_read_only_ad_enabled(void)
2975 {
2976 	/*
2977 	 * When EPT AD bits are enabled, all accesses to guest paging
2978 	 * structures are considered writes as far as EPT translation
2979 	 * is concerned.
2980 	 */
2981 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2982 
2983 	ept_access_test_setup();
2984 	ept_enable_ad_bits_or_skip_test();
2985 
2986 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2987 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
2988 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
2989 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
2990 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
2991 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
2992 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
2993 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
2994 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
2995 }
2996 
2997 static void ept_access_test_paddr_read_write(void)
2998 {
2999 	ept_access_test_setup();
3000 	/* Read-write access to paging structure. */
3001 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
3002 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
3003 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
3004 }
3005 
3006 static void ept_access_test_paddr_read_write_execute(void)
3007 {
3008 	ept_access_test_setup();
3009 	/* RWX access to paging structure. */
3010 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
3011 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
3012 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
3013 }
3014 
3015 static void ept_access_test_paddr_read_execute_ad_disabled(void)
3016 {
3017   	/*
3018 	 * When EPT AD bits are disabled, all accesses to guest paging
3019 	 * structures are reported separately as a read and (after
3020 	 * translation of the GPA to host physical address) a read+write
3021 	 * if the A/D bits have to be set.
3022 	 */
3023 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3024 
3025 	ept_access_test_setup();
3026 	ept_disable_ad_bits();
3027 
3028 	/* Can't update A bit, so all accesses fail. */
3029 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3030 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3031 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3032 	/* AD bits disabled, so only writes try to update the D bit. */
3033 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
3034 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3035 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3036 	/* Both A and D already set, so read-only is OK. */
3037 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3038 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3039 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3040 }
3041 
3042 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3043 {
3044 	/*
3045 	 * When EPT AD bits are enabled, all accesses to guest paging
3046 	 * structures are considered writes as far as EPT translation
3047 	 * is concerned.
3048 	 */
3049 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3050 
3051 	ept_access_test_setup();
3052 	ept_enable_ad_bits_or_skip_test();
3053 
3054 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3055 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3056 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3057 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3058 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3059 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3060 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3061 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3062 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3063 }
3064 
3065 static void ept_access_test_paddr_not_present_page_fault(void)
3066 {
3067 	ept_access_test_setup();
3068 	/*
3069 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3070 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3071 	 * Thus guest page fault before host takes EPT violation for trying to
3072 	 * update A bit.
3073 	 */
3074 }
3075 
3076 static void ept_access_test_force_2m_page(void)
3077 {
3078 	ept_access_test_setup();
3079 
3080 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3081 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3082 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3083 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3084 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3085 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3086 }
3087 
3088 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3089 {
3090 	if (!is_invvpid_type_supported(type))
3091 		return false;
3092 
3093 	if (vpid >> 16)
3094 		return false;
3095 
3096 	if (type != INVVPID_ALL && !vpid)
3097 		return false;
3098 
3099 	if (type == INVVPID_ADDR && !is_canonical(gla))
3100 		return false;
3101 
3102 	return true;
3103 }
3104 
3105 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3106 {
3107 	int rc;
3108 	bool valid = invvpid_valid(type, vpid, gla);
3109 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3110 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3111 	/*
3112 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3113 	 * that we can tell if it is updated by INVVPID.
3114 	 */
3115 	vmcs_read(~0);
3116 	rc = __invvpid(type, vpid, gla);
3117 	report(!rc == valid, "INVVPID type %ld VPID %lx GLA %lx %s", type,
3118 	       vpid, gla,
3119 	       valid ? "passes" : "fails");
3120 	report(vmcs_read(VMX_INST_ERROR) == expected,
3121 	       "After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3122 	       rc ? "failed" : "successful",
3123 	       expected, vmcs_read(VMX_INST_ERROR));
3124 }
3125 
3126 static inline unsigned long get_first_supported_invvpid_type(void)
3127 {
3128 	u64 type = ffs(ept_vpid.val >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3129 
3130 	__TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3131 	return type;
3132 }
3133 
3134 static void ds_invvpid(void *data)
3135 {
3136 	asm volatile("invvpid %0, %1"
3137 		     :
3138 		     : "m"(*(struct invvpid_operand *)data),
3139 		       "r"(get_first_supported_invvpid_type()));
3140 }
3141 
3142 /*
3143  * The SS override is ignored in 64-bit mode, so we use an addressing
3144  * mode with %rsp as the base register to generate an implicit SS
3145  * reference.
3146  */
3147 static void ss_invvpid(void *data)
3148 {
3149 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3150 		     : "+r"(data)
3151 		     : "r"(get_first_supported_invvpid_type()));
3152 }
3153 
3154 static void invvpid_test_gp(void)
3155 {
3156 	bool fault;
3157 
3158 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3159 				   (void *)NONCANONICAL);
3160 	report(fault, "INVVPID with non-canonical DS operand raises #GP");
3161 }
3162 
3163 static void invvpid_test_ss(void)
3164 {
3165 	bool fault;
3166 
3167 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3168 				   (void *)NONCANONICAL);
3169 	report(fault, "INVVPID with non-canonical SS operand raises #SS");
3170 }
3171 
3172 static void invvpid_test_pf(void)
3173 {
3174 	void *vpage = alloc_vpage();
3175 	bool fault;
3176 
3177 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3178 	report(fault, "INVVPID with unmapped operand raises #PF");
3179 }
3180 
3181 static void try_compat_invvpid(void *unused)
3182 {
3183 	struct far_pointer32 fp = {
3184 		.offset = (uintptr_t)&&invvpid,
3185 		.selector = KERNEL_CS32,
3186 	};
3187 	uintptr_t rsp;
3188 
3189 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
3190 
3191 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3192 			"Code address too high.");
3193 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3194 
3195 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3196 	return;
3197 invvpid:
3198 	asm volatile (".code32;"
3199 		      "invvpid (%eax), %eax;"
3200 		      "lret;"
3201 		      ".code64");
3202 	__builtin_unreachable();
3203 }
3204 
3205 static void invvpid_test_compatibility_mode(void)
3206 {
3207 	bool fault;
3208 
3209 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3210 	report(fault, "Compatibility mode INVVPID raises #UD");
3211 }
3212 
3213 static void invvpid_test_not_in_vmx_operation(void)
3214 {
3215 	bool fault;
3216 
3217 	TEST_ASSERT(!vmx_off());
3218 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3219 	report(fault, "INVVPID outside of VMX operation raises #UD");
3220 	TEST_ASSERT(!vmx_on());
3221 }
3222 
3223 /*
3224  * This does not test real-address mode, virtual-8086 mode, protected mode,
3225  * or CPL > 0.
3226  */
3227 static void invvpid_test(void)
3228 {
3229 	int i;
3230 	unsigned types = 0;
3231 	unsigned type;
3232 
3233 	if (!is_vpid_supported())
3234 		test_skip("VPID not supported");
3235 
3236 	if (!is_invvpid_supported())
3237 		test_skip("INVVPID not supported.\n");
3238 
3239 	if (is_invvpid_type_supported(INVVPID_ADDR))
3240 		types |= 1u << INVVPID_ADDR;
3241 	if (is_invvpid_type_supported(INVVPID_CONTEXT_GLOBAL))
3242 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3243 	if (is_invvpid_type_supported(INVVPID_ALL))
3244 		types |= 1u << INVVPID_ALL;
3245 	if (is_invvpid_type_supported(INVVPID_CONTEXT_LOCAL))
3246 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3247 
3248 	if (!types)
3249 		test_skip("No INVVPID types supported.\n");
3250 
3251 	for (i = -127; i < 128; i++)
3252 		try_invvpid(i, 0xffff, 0);
3253 
3254 	/*
3255 	 * VPID must not be more than 16 bits.
3256 	 */
3257 	for (i = 0; i < 64; i++)
3258 		for (type = 0; type < 4; type++)
3259 			if (types & (1u << type))
3260 				try_invvpid(type, 1ul << i, 0);
3261 
3262 	/*
3263 	 * VPID must not be zero, except for "all contexts."
3264 	 */
3265 	for (type = 0; type < 4; type++)
3266 		if (types & (1u << type))
3267 			try_invvpid(type, 0, 0);
3268 
3269 	/*
3270 	 * The gla operand is only validated for single-address INVVPID.
3271 	 */
3272 	if (types & (1u << INVVPID_ADDR))
3273 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3274 
3275 	invvpid_test_gp();
3276 	invvpid_test_ss();
3277 	invvpid_test_pf();
3278 	invvpid_test_compatibility_mode();
3279 	invvpid_test_not_in_vmx_operation();
3280 }
3281 
3282 static void test_assert_vmlaunch_inst_error(u32 expected_error)
3283 {
3284 	u32 vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3285 
3286 	report(vmx_inst_err == expected_error,
3287 	       "VMX inst error is %d (actual %d)", expected_error, vmx_inst_err);
3288 }
3289 
3290 /*
3291  * This version is wildly unsafe and should _only_ be used to test VM-Fail
3292  * scenarios involving HOST_RIP.
3293  */
3294 static void test_vmx_vmlaunch_must_fail(u32 expected_error)
3295 {
3296 	/* Read the function name. */
3297 	TEST_ASSERT(expected_error);
3298 
3299 	/*
3300 	 * Don't bother with any prep work, if VMLAUNCH passes the VM-Fail
3301 	 * consistency checks and generates a VM-Exit, then the test is doomed
3302 	 * no matter what as it will jump to a garbage RIP.
3303 	 */
3304 	__asm__ __volatile__ ("vmlaunch");
3305 	test_assert_vmlaunch_inst_error(expected_error);
3306 }
3307 
3308 /*
3309  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3310  * at least as far as the guest-state checks. Returns false if the
3311  * VMLAUNCH fails early and execution falls through to the next
3312  * instruction.
3313  */
3314 static bool vmlaunch(void)
3315 {
3316 	u32 exit_reason;
3317 
3318 	/*
3319 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3320 	 * unsupported VMCS component"). The caller can then check
3321 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3322 	 */
3323 	vmcs_write(~0u, 0);
3324 
3325 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3326 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3327 				   :
3328 				   : "r" ((u64)HOST_RSP)
3329 				   : "cc", "memory"
3330 				   : success);
3331 	return false;
3332 success:
3333 	exit_reason = vmcs_read(EXI_REASON);
3334 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3335 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3336 	return true;
3337 }
3338 
3339 /*
3340  * Try to launch the current VMCS.
3341  */
3342 static void test_vmx_vmlaunch(u32 xerror)
3343 {
3344 	bool success = vmlaunch();
3345 
3346 	report(success == !xerror, "vmlaunch %s",
3347 	       !xerror ? "succeeds" : "fails");
3348 	if (!success && xerror)
3349 		test_assert_vmlaunch_inst_error(xerror);
3350 }
3351 
3352 /*
3353  * Try to launch the current VMCS, and expect one of two possible
3354  * errors (or success) codes.
3355  */
3356 static void test_vmx_vmlaunch2(u32 xerror1, u32 xerror2)
3357 {
3358 	bool success = vmlaunch();
3359 	u32 vmx_inst_err;
3360 
3361 	if (!xerror1 == !xerror2)
3362 		report(success == !xerror1, "vmlaunch %s",
3363 		       !xerror1 ? "succeeds" : "fails");
3364 
3365 	if (!success && (xerror1 || xerror2)) {
3366 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3367 		report(vmx_inst_err == xerror1 || vmx_inst_err == xerror2,
3368 		       "VMX inst error is %d or %d (actual %d)", xerror1,
3369 		       xerror2, vmx_inst_err);
3370 	}
3371 }
3372 
3373 static void test_vmx_invalid_controls(void)
3374 {
3375 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3376 }
3377 
3378 static void test_vmx_valid_controls(void)
3379 {
3380 	test_vmx_vmlaunch(0);
3381 }
3382 
3383 /*
3384  * Test a particular value of a VM-execution control bit, if the value
3385  * is required or if the value is zero.
3386  */
3387 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3388 				    enum Encoding encoding, unsigned bit,
3389 				    unsigned val)
3390 {
3391 	u32 mask = 1u << bit;
3392 	bool expected;
3393 	u32 controls;
3394 
3395 	if (msr.set & mask)
3396 		TEST_ASSERT(msr.clr & mask);
3397 
3398 	/*
3399 	 * We can't arbitrarily turn on a control bit, because it may
3400 	 * introduce dependencies on other VMCS fields. So, we only
3401 	 * test turning on bits that have a required setting.
3402 	 */
3403 	if (val && (msr.clr & mask) && !(msr.set & mask))
3404 		return;
3405 
3406 	report_prefix_pushf("%s %s bit %d",
3407 			    val ? "Set" : "Clear", name, bit);
3408 
3409 	controls = vmcs_read(encoding);
3410 	if (val) {
3411 		vmcs_write(encoding, msr.set | mask);
3412 		expected = (msr.clr & mask);
3413 	} else {
3414 		vmcs_write(encoding, msr.set & ~mask);
3415 		expected = !(msr.set & mask);
3416 	}
3417 	if (expected)
3418 		test_vmx_valid_controls();
3419 	else
3420 		test_vmx_invalid_controls();
3421 	vmcs_write(encoding, controls);
3422 	report_prefix_pop();
3423 }
3424 
3425 /*
3426  * Test reserved values of a VM-execution control bit, based on the
3427  * allowed bit settings from the corresponding VMX capability MSR.
3428  */
3429 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3430 			      enum Encoding encoding, unsigned bit)
3431 {
3432 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3433 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3434 }
3435 
3436 /*
3437  * Reserved bits in the pin-based VM-execution controls must be set
3438  * properly. Software may consult the VMX capability MSRs to determine
3439  * the proper settings.
3440  * [Intel SDM]
3441  */
3442 static void test_pin_based_ctls(void)
3443 {
3444 	unsigned bit;
3445 
3446 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3447 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3448 	for (bit = 0; bit < 32; bit++)
3449 		test_rsvd_ctl_bit("pin-based controls",
3450 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3451 }
3452 
3453 /*
3454  * Reserved bits in the primary processor-based VM-execution controls
3455  * must be set properly. Software may consult the VMX capability MSRs
3456  * to determine the proper settings.
3457  * [Intel SDM]
3458  */
3459 static void test_primary_processor_based_ctls(void)
3460 {
3461 	unsigned bit;
3462 
3463 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3464 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3465 	for (bit = 0; bit < 32; bit++)
3466 		test_rsvd_ctl_bit("primary processor-based controls",
3467 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3468 }
3469 
3470 /*
3471  * If the "activate secondary controls" primary processor-based
3472  * VM-execution control is 1, reserved bits in the secondary
3473  * processor-based VM-execution controls must be cleared. Software may
3474  * consult the VMX capability MSRs to determine which bits are
3475  * reserved.
3476  * If the "activate secondary controls" primary processor-based
3477  * VM-execution control is 0 (or if the processor does not support the
3478  * 1-setting of that control), no checks are performed on the
3479  * secondary processor-based VM-execution controls.
3480  * [Intel SDM]
3481  */
3482 static void test_secondary_processor_based_ctls(void)
3483 {
3484 	u32 primary;
3485 	u32 secondary;
3486 	unsigned bit;
3487 
3488 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3489 		return;
3490 
3491 	primary = vmcs_read(CPU_EXEC_CTRL0);
3492 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3493 
3494 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3495 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3496 	for (bit = 0; bit < 32; bit++)
3497 		test_rsvd_ctl_bit("secondary processor-based controls",
3498 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3499 
3500 	/*
3501 	 * When the "activate secondary controls" VM-execution control
3502 	 * is clear, there are no checks on the secondary controls.
3503 	 */
3504 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3505 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3506 	report(vmlaunch(),
3507 	       "Secondary processor-based controls ignored");
3508 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3509 	vmcs_write(CPU_EXEC_CTRL0, primary);
3510 }
3511 
3512 static void try_cr3_target_count(unsigned i, unsigned max)
3513 {
3514 	report_prefix_pushf("CR3 target count 0x%x", i);
3515 	vmcs_write(CR3_TARGET_COUNT, i);
3516 	if (i <= max)
3517 		test_vmx_valid_controls();
3518 	else
3519 		test_vmx_invalid_controls();
3520 	report_prefix_pop();
3521 }
3522 
3523 /*
3524  * The CR3-target count must not be greater than 4. Future processors
3525  * may support a different number of CR3-target values. Software
3526  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3527  * number of values supported.
3528  * [Intel SDM]
3529  */
3530 static void test_cr3_targets(void)
3531 {
3532 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3533 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3534 	unsigned i;
3535 
3536 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3537 	TEST_ASSERT(supported_targets <= 256);
3538 
3539 	try_cr3_target_count(-1u, supported_targets);
3540 	try_cr3_target_count(0x80000000, supported_targets);
3541 	try_cr3_target_count(0x7fffffff, supported_targets);
3542 	for (i = 0; i <= supported_targets + 1; i++)
3543 		try_cr3_target_count(i, supported_targets);
3544 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3545 
3546 	/* VMWRITE to nonexistent target fields should fail. */
3547 	for (i = supported_targets; i < 256; i++)
3548 		TEST_ASSERT(vmcs_write(CR3_TARGET_0 + i*2, 0));
3549 }
3550 
3551 /*
3552  * Test a particular address setting in the VMCS
3553  */
3554 static void test_vmcs_addr(const char *name,
3555 			   enum Encoding encoding,
3556 			   u64 align,
3557 			   bool ignored,
3558 			   bool skip_beyond_mapped_ram,
3559 			   u64 addr)
3560 {
3561 	report_prefix_pushf("%s = %lx", name, addr);
3562 	vmcs_write(encoding, addr);
3563 	if (skip_beyond_mapped_ram &&
3564 	    addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3565 	    addr < (1ul << cpuid_maxphyaddr()))
3566 		printf("Skipping physical address beyond mapped RAM\n");
3567 	else if (ignored || (IS_ALIGNED(addr, align) &&
3568 	    addr < (1ul << cpuid_maxphyaddr())))
3569 		test_vmx_valid_controls();
3570 	else
3571 		test_vmx_invalid_controls();
3572 	report_prefix_pop();
3573 }
3574 
3575 /*
3576  * Test interesting values for a VMCS address
3577  */
3578 static void test_vmcs_addr_values(const char *name,
3579 				  enum Encoding encoding,
3580 				  u64 align,
3581 				  bool ignored,
3582 				  bool skip_beyond_mapped_ram,
3583 				  u32 bit_start, u32 bit_end)
3584 {
3585 	unsigned i;
3586 	u64 orig_val = vmcs_read(encoding);
3587 
3588 	for (i = bit_start; i <= bit_end; i++)
3589 		test_vmcs_addr(name, encoding, align, ignored,
3590 			       skip_beyond_mapped_ram, 1ul << i);
3591 
3592 	test_vmcs_addr(name, encoding, align, ignored,
3593 		       skip_beyond_mapped_ram, PAGE_SIZE - 1);
3594 	test_vmcs_addr(name, encoding, align, ignored,
3595 		       skip_beyond_mapped_ram, PAGE_SIZE);
3596 	test_vmcs_addr(name, encoding, align, ignored,
3597 		       skip_beyond_mapped_ram,
3598 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3599 	test_vmcs_addr(name, encoding, align, ignored,
3600 		       skip_beyond_mapped_ram, -1ul);
3601 
3602 	vmcs_write(encoding, orig_val);
3603 }
3604 
3605 /*
3606  * Test a physical address reference in the VMCS, when the corresponding
3607  * feature is enabled and when the corresponding feature is disabled.
3608  */
3609 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3610 				     const char *field_name,
3611 				     const char *control_name, u64 align,
3612 				     bool skip_beyond_mapped_ram,
3613 				     bool control_primary)
3614 {
3615 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3616 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3617 	u64 page_addr;
3618 
3619 	if (control_primary) {
3620 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3621 			return;
3622 	} else {
3623 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3624 			return;
3625 	}
3626 
3627 	page_addr = vmcs_read(field);
3628 
3629 	report_prefix_pushf("%s enabled", control_name);
3630 	if (control_primary) {
3631 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3632 	} else {
3633 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3634 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3635 	}
3636 
3637 	test_vmcs_addr_values(field_name, field, align, false,
3638 			      skip_beyond_mapped_ram, 0, 63);
3639 	report_prefix_pop();
3640 
3641 	report_prefix_pushf("%s disabled", control_name);
3642 	if (control_primary) {
3643 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3644 	} else {
3645 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3646 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3647 	}
3648 
3649 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3650 	report_prefix_pop();
3651 
3652 	vmcs_write(field, page_addr);
3653 	vmcs_write(CPU_EXEC_CTRL0, primary);
3654 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3655 }
3656 
3657 /*
3658  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3659  * each I/O-bitmap address must be 0. Neither address should set any
3660  * bits beyond the processor's physical-address width.
3661  * [Intel SDM]
3662  */
3663 static void test_io_bitmaps(void)
3664 {
3665 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3666 				 "I/O bitmap A", "Use I/O bitmaps",
3667 				 PAGE_SIZE, false, true);
3668 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3669 				 "I/O bitmap B", "Use I/O bitmaps",
3670 				 PAGE_SIZE, false, true);
3671 }
3672 
3673 /*
3674  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3675  * the MSR-bitmap address must be 0. The address should not set any
3676  * bits beyond the processor's physical-address width.
3677  * [Intel SDM]
3678  */
3679 static void test_msr_bitmap(void)
3680 {
3681 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3682 				 "MSR bitmap", "Use MSR bitmaps",
3683 				 PAGE_SIZE, false, true);
3684 }
3685 
3686 /*
3687  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3688  * address must satisfy the following checks:
3689  * - Bits 11:0 of the address must be 0.
3690  * - The address should not set any bits beyond the processor's
3691  *   physical-address width.
3692  * [Intel SDM]
3693  */
3694 static void test_apic_virt_addr(void)
3695 {
3696 	/*
3697 	 * Ensure the processor will never use the virtual-APIC page, since
3698 	 * we will point it to invalid RAM.  Otherwise KVM is puzzled about
3699 	 * what we're trying to achieve and fails vmentry.
3700 	 */
3701 	u32 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
3702 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0 | CPU_CR8_LOAD | CPU_CR8_STORE);
3703 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3704 				 "virtual-APIC address", "Use TPR shadow",
3705 				 PAGE_SIZE, false, true);
3706 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
3707 }
3708 
3709 /*
3710  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3711  * APIC-access address must satisfy the following checks:
3712  *  - Bits 11:0 of the address must be 0.
3713  *  - The address should not set any bits beyond the processor's
3714  *    physical-address width.
3715  * [Intel SDM]
3716  */
3717 static void test_apic_access_addr(void)
3718 {
3719 	void *apic_access_page = alloc_page();
3720 
3721 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3722 
3723 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3724 				 "APIC-access address",
3725 				 "virtualize APIC-accesses", PAGE_SIZE,
3726 				 true, false);
3727 }
3728 
3729 static bool set_bit_pattern(u8 mask, u32 *secondary)
3730 {
3731 	u8 i;
3732 	bool flag = false;
3733 	u32 test_bits[3] = {
3734 		CPU_VIRT_X2APIC,
3735 		CPU_APIC_REG_VIRT,
3736 		CPU_VINTD
3737 	};
3738 
3739         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3740 		if ((mask & (1u << i)) &&
3741 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3742 			*secondary |= test_bits[i];
3743 			flag = true;
3744 		}
3745 	}
3746 
3747 	return (flag);
3748 }
3749 
3750 /*
3751  * If the "use TPR shadow" VM-execution control is 0, the following
3752  * VM-execution controls must also be 0:
3753  * 	- virtualize x2APIC mode
3754  *	- APIC-register virtualization
3755  *	- virtual-interrupt delivery
3756  *    [Intel SDM]
3757  *
3758  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3759  *    "virtualize APIC accesses" VM-execution control must be 0.
3760  *    [Intel SDM]
3761  */
3762 static void test_apic_virtual_ctls(void)
3763 {
3764 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3765 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3766 	u32 primary = saved_primary;
3767 	u32 secondary = saved_secondary;
3768 	bool is_ctrl_valid = false;
3769 	char str[10] = "disabled";
3770 	u8 i = 0, j;
3771 
3772 	/*
3773 	 * First test
3774 	 */
3775 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3776 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3777 		return;
3778 
3779 	primary |= CPU_SECONDARY;
3780 	primary &= ~CPU_TPR_SHADOW;
3781 	vmcs_write(CPU_EXEC_CTRL0, primary);
3782 
3783 	while (1) {
3784 		for (j = 1; j < 8; j++) {
3785 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3786 			if (primary & CPU_TPR_SHADOW) {
3787 				is_ctrl_valid = true;
3788 			} else {
3789 				if (! set_bit_pattern(j, &secondary))
3790 					is_ctrl_valid = true;
3791 				else
3792 					is_ctrl_valid = false;
3793 			}
3794 
3795 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3796 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3797 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3798 			if (is_ctrl_valid)
3799 				test_vmx_valid_controls();
3800 			else
3801 				test_vmx_invalid_controls();
3802 			report_prefix_pop();
3803 		}
3804 
3805 		if (i == 1)
3806 			break;
3807 		i++;
3808 
3809 		primary |= CPU_TPR_SHADOW;
3810 		vmcs_write(CPU_EXEC_CTRL0, primary);
3811 		strcpy(str, "enabled");
3812 	}
3813 
3814 	/*
3815 	 * Second test
3816 	 */
3817 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3818 
3819 	primary = saved_primary;
3820 	secondary = saved_secondary;
3821 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3822 		return;
3823 
3824 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3825 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3826 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3827 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3828 	test_vmx_valid_controls();
3829 	report_prefix_pop();
3830 
3831 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3832 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3833 	test_vmx_valid_controls();
3834 	report_prefix_pop();
3835 
3836 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3837 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3838 	test_vmx_invalid_controls();
3839 	report_prefix_pop();
3840 
3841 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3842 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3843 	test_vmx_valid_controls();
3844 	report_prefix_pop();
3845 
3846 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3847 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3848 }
3849 
3850 /*
3851  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3852  * "external-interrupt exiting" VM-execution control must be 1.
3853  * [Intel SDM]
3854  */
3855 static void test_virtual_intr_ctls(void)
3856 {
3857 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3858 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3859 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3860 	u32 primary = saved_primary;
3861 	u32 secondary = saved_secondary;
3862 	u32 pin = saved_pin;
3863 
3864 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3865 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
3866 		return;
3867 
3868 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3869 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
3870 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3871 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
3872 	test_vmx_valid_controls();
3873 	report_prefix_pop();
3874 
3875 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
3876 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3877 	test_vmx_invalid_controls();
3878 	report_prefix_pop();
3879 
3880 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
3881 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
3882 	test_vmx_valid_controls();
3883 	report_prefix_pop();
3884 
3885 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3886 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3887 	test_vmx_invalid_controls();
3888 	report_prefix_pop();
3889 
3890 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3891 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3892 	vmcs_write(PIN_CONTROLS, saved_pin);
3893 }
3894 
3895 static void test_pi_desc_addr(u64 addr, bool is_ctrl_valid)
3896 {
3897 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
3898 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
3899 	if (is_ctrl_valid)
3900 		test_vmx_valid_controls();
3901 	else
3902 		test_vmx_invalid_controls();
3903 	report_prefix_pop();
3904 }
3905 
3906 /*
3907  * If the "process posted interrupts" VM-execution control is 1, the
3908  * following must be true:
3909  *
3910  *	- The "virtual-interrupt delivery" VM-execution control is 1.
3911  *	- The "acknowledge interrupt on exit" VM-exit control is 1.
3912  *	- The posted-interrupt notification vector has a value in the
3913  *	- range 0 - 255 (bits 15:8 are all 0).
3914  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
3915  *	- The posted-interrupt descriptor address does not set any bits
3916  *	  beyond the processor's physical-address width.
3917  * [Intel SDM]
3918  */
3919 static void test_posted_intr(void)
3920 {
3921 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3922 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3923 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3924 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
3925 	u32 primary = saved_primary;
3926 	u32 secondary = saved_secondary;
3927 	u32 pin = saved_pin;
3928 	u32 exit_ctl = exit_ctl_saved;
3929 	u16 vec;
3930 	int i;
3931 
3932 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
3933 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3934 	    (ctrl_exit_rev.clr & EXI_INTA)))
3935 		return;
3936 
3937 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3938 
3939 	/*
3940 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
3941 	 */
3942 	pin |= PIN_POST_INTR;
3943 	vmcs_write(PIN_CONTROLS, pin);
3944 	secondary &= ~CPU_VINTD;
3945 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3946 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
3947 	test_vmx_invalid_controls();
3948 	report_prefix_pop();
3949 
3950 	secondary |= CPU_VINTD;
3951 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3952 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
3953 	test_vmx_invalid_controls();
3954 	report_prefix_pop();
3955 
3956 	exit_ctl &= ~EXI_INTA;
3957 	vmcs_write(EXI_CONTROLS, exit_ctl);
3958 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
3959 	test_vmx_invalid_controls();
3960 	report_prefix_pop();
3961 
3962 	exit_ctl |= EXI_INTA;
3963 	vmcs_write(EXI_CONTROLS, exit_ctl);
3964 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3965 	test_vmx_valid_controls();
3966 	report_prefix_pop();
3967 
3968 	secondary &= ~CPU_VINTD;
3969 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3970 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
3971 	test_vmx_invalid_controls();
3972 	report_prefix_pop();
3973 
3974 	secondary |= CPU_VINTD;
3975 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3976 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3977 	test_vmx_valid_controls();
3978 	report_prefix_pop();
3979 
3980 	/*
3981 	 * Test posted-interrupt notification vector
3982 	 */
3983 	for (i = 0; i < 8; i++) {
3984 		vec = (1ul << i);
3985 		vmcs_write(PINV, vec);
3986 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3987 		test_vmx_valid_controls();
3988 		report_prefix_pop();
3989 	}
3990 	for (i = 8; i < 16; i++) {
3991 		vec = (1ul << i);
3992 		vmcs_write(PINV, vec);
3993 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3994 		test_vmx_invalid_controls();
3995 		report_prefix_pop();
3996 	}
3997 
3998 	vec &= ~(0xff << 8);
3999 	vmcs_write(PINV, vec);
4000 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4001 	test_vmx_valid_controls();
4002 	report_prefix_pop();
4003 
4004 	/*
4005 	 * Test posted-interrupt descriptor address
4006 	 */
4007 	for (i = 0; i < 6; i++) {
4008 		test_pi_desc_addr(1ul << i, false);
4009 	}
4010 
4011 	test_pi_desc_addr(0xf0, false);
4012 	test_pi_desc_addr(0xff, false);
4013 	test_pi_desc_addr(0x0f, false);
4014 	test_pi_desc_addr(0x8000, true);
4015 	test_pi_desc_addr(0x00, true);
4016 	test_pi_desc_addr(0xc000, true);
4017 
4018 	test_vmcs_addr_values("process-posted interrupts",
4019 			       POSTED_INTR_DESC_ADDR, 64,
4020 			       false, false, 0, 63);
4021 
4022 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4023 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4024 	vmcs_write(PIN_CONTROLS, saved_pin);
4025 }
4026 
4027 static void test_apic_ctls(void)
4028 {
4029 	test_apic_virt_addr();
4030 	test_apic_access_addr();
4031 	test_apic_virtual_ctls();
4032 	test_virtual_intr_ctls();
4033 	test_posted_intr();
4034 }
4035 
4036 /*
4037  * If the "enable VPID" VM-execution control is 1, the value of the
4038  * of the VPID VM-execution control field must not be 0000H.
4039  * [Intel SDM]
4040  */
4041 static void test_vpid(void)
4042 {
4043 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4044 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4045 	u16 vpid = 0x0000;
4046 	int i;
4047 
4048 	if (!is_vpid_supported()) {
4049 		report_skip("%s : Secondary controls and/or VPID not supported", __func__);
4050 		return;
4051 	}
4052 
4053 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
4054 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
4055 	vmcs_write(VPID, vpid);
4056 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
4057 	test_vmx_valid_controls();
4058 	report_prefix_pop();
4059 
4060 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
4061 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4062 	test_vmx_invalid_controls();
4063 	report_prefix_pop();
4064 
4065 	for (i = 0; i < 16; i++) {
4066 		vpid = (short)1 << i;;
4067 		vmcs_write(VPID, vpid);
4068 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4069 		test_vmx_valid_controls();
4070 		report_prefix_pop();
4071 	}
4072 
4073 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4074 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4075 }
4076 
4077 static void set_vtpr(unsigned vtpr)
4078 {
4079 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
4080 }
4081 
4082 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
4083 {
4084 	bool valid = true;
4085 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4086 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4087 
4088 	if ((primary & CPU_TPR_SHADOW) &&
4089 	    (!(primary & CPU_SECONDARY) ||
4090 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
4091 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
4092 
4093 	set_vtpr(vtpr);
4094 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
4095 	    threshold, (vtpr >> 4) & 0xf);
4096 	if (valid)
4097 		test_vmx_valid_controls();
4098 	else
4099 		test_vmx_invalid_controls();
4100 	report_prefix_pop();
4101 }
4102 
4103 static void test_invalid_event_injection(void)
4104 {
4105 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
4106 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
4107 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
4108 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
4109 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
4110 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4111 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4112 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4113 	u32 cnt;
4114 
4115 	/* Setup */
4116 	report_prefix_push("invalid event injection");
4117 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4118 	vmcs_write(ENT_INST_LEN, 0x00000001);
4119 
4120 	/* The field's interruption type is not set to a reserved value. */
4121 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4122 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4123 			    "RESERVED interruption type invalid [-]",
4124 			    ent_intr_info);
4125 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4126 	test_vmx_invalid_controls();
4127 	report_prefix_pop();
4128 
4129 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4130 			DE_VECTOR;
4131 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4132 			    "RESERVED interruption type invalid [+]",
4133 			    ent_intr_info);
4134 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4135 	test_vmx_valid_controls();
4136 	report_prefix_pop();
4137 
4138 	/* If the interruption type is other event, the vector is 0. */
4139 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4140 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4141 			    "(OTHER EVENT && vector != 0) invalid [-]",
4142 			    ent_intr_info);
4143 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4144 	test_vmx_invalid_controls();
4145 	report_prefix_pop();
4146 
4147 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4148 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4149 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4150 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4151 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4152 	test_vmx_invalid_controls();
4153 	report_prefix_pop();
4154 
4155 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4156 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4157 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4158 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4159 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4160 	test_vmx_valid_controls();
4161 	report_prefix_pop();
4162 
4163 	/*
4164 	 * If the interruption type
4165 	 * is HW exception, the vector is at most 31.
4166 	 */
4167 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4168 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4169 			    "(HW exception && vector > 31) invalid [-]",
4170 			    ent_intr_info);
4171 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4172 	test_vmx_invalid_controls();
4173 	report_prefix_pop();
4174 
4175 	/*
4176 	 * deliver-error-code is 1 iff either
4177 	 * (a) the "unrestricted guest" VM-execution control is 0
4178 	 * (b) CR0.PE is set.
4179 	 */
4180 
4181 	/* Assert that unrestricted guest is disabled or unsupported */
4182 	assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
4183 	       !(secondary_save & CPU_URG));
4184 
4185 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4186 			GP_VECTOR;
4187 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4188 			    "error code <-> (!URG || prot_mode) [-]",
4189 			    ent_intr_info);
4190 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4191 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4192 	test_vmx_invalid_controls();
4193 	report_prefix_pop();
4194 
4195 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4196 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4197 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4198 			    "error code <-> (!URG || prot_mode) [+]",
4199 			    ent_intr_info);
4200 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4201 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4202 	test_vmx_valid_controls();
4203 	report_prefix_pop();
4204 
4205 	if (enable_unrestricted_guest(false))
4206 		goto skip_unrestricted_guest;
4207 
4208 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4209 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4210 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4211 			    "error code <-> (!URG || prot_mode) [-]",
4212 			    ent_intr_info);
4213 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4214 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4215 	test_vmx_invalid_controls();
4216 	report_prefix_pop();
4217 
4218 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4219 			GP_VECTOR;
4220 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4221 			    "error code <-> (!URG || prot_mode) [-]",
4222 			    ent_intr_info);
4223 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4224 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4225 	test_vmx_invalid_controls();
4226 	report_prefix_pop();
4227 
4228 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4229 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4230 
4231 skip_unrestricted_guest:
4232 	vmcs_write(GUEST_CR0, guest_cr0_save);
4233 
4234 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4235 	report_prefix_push("error code <-> HW exception");
4236 	for (cnt = 0; cnt < 8; cnt++) {
4237 		u32 exception_type_mask = cnt << 8;
4238 		u32 deliver_error_code_mask =
4239 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4240 			INTR_INFO_DELIVER_CODE_MASK : 0;
4241 
4242 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4243 				exception_type_mask | GP_VECTOR;
4244 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4245 				    ent_intr_info);
4246 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4247 		test_vmx_invalid_controls();
4248 		report_prefix_pop();
4249 	}
4250 	report_prefix_pop();
4251 
4252 	/*
4253 	 * deliver-error-code is 1 iff the the vector
4254 	 * indicates an exception that would normally deliver an error code
4255 	 */
4256 	report_prefix_push("error code <-> vector delivers error code");
4257 	for (cnt = 0; cnt < 32; cnt++) {
4258 		bool has_error_code = false;
4259 		u32 deliver_error_code_mask;
4260 
4261 		switch (cnt) {
4262 		case DF_VECTOR:
4263 		case TS_VECTOR:
4264 		case NP_VECTOR:
4265 		case SS_VECTOR:
4266 		case GP_VECTOR:
4267 		case PF_VECTOR:
4268 		case AC_VECTOR:
4269 			has_error_code = true;
4270 		case CP_VECTOR:
4271 			/* Some CPUs have error code and some do not, skip */
4272 			continue;
4273 		}
4274 
4275 		/* Negative case */
4276 		deliver_error_code_mask = has_error_code ?
4277 						0 :
4278 						INTR_INFO_DELIVER_CODE_MASK;
4279 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4280 				INTR_TYPE_HARD_EXCEPTION | cnt;
4281 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4282 				    ent_intr_info);
4283 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4284 		test_vmx_invalid_controls();
4285 		report_prefix_pop();
4286 
4287 		/* Positive case */
4288 		deliver_error_code_mask = has_error_code ?
4289 						INTR_INFO_DELIVER_CODE_MASK :
4290 						0;
4291 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4292 				INTR_TYPE_HARD_EXCEPTION | cnt;
4293 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4294 				    ent_intr_info);
4295 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4296 		test_vmx_valid_controls();
4297 		report_prefix_pop();
4298 	}
4299 	report_prefix_pop();
4300 
4301 	/* Reserved bits in the field (30:12) are 0. */
4302 	report_prefix_push("reserved bits clear");
4303 	for (cnt = 12; cnt <= 30; cnt++) {
4304 		ent_intr_info = ent_intr_info_base |
4305 				INTR_INFO_DELIVER_CODE_MASK |
4306 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4307 				(1U << cnt);
4308 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4309 				    ent_intr_info);
4310 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4311 		test_vmx_invalid_controls();
4312 		report_prefix_pop();
4313 	}
4314 	report_prefix_pop();
4315 
4316 	/*
4317 	 * If deliver-error-code is 1
4318 	 * bits 31:16 of the VM-entry exception error-code field are 0.
4319 	 */
4320 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4321 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4322 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4323 			    "VM-entry exception error code[31:16] clear",
4324 			    ent_intr_info);
4325 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4326 	for (cnt = 16; cnt <= 31; cnt++) {
4327 		ent_intr_err = 1U << cnt;
4328 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4329 				    ent_intr_err);
4330 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4331 		test_vmx_invalid_controls();
4332 		report_prefix_pop();
4333 	}
4334 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4335 	report_prefix_pop();
4336 
4337 	/*
4338 	 * If the interruption type is software interrupt, software exception,
4339 	 * or privileged software exception, the VM-entry instruction-length
4340 	 * field is in the range 0 - 15.
4341 	 */
4342 
4343 	for (cnt = 0; cnt < 3; cnt++) {
4344 		switch (cnt) {
4345 		case 0:
4346 			ent_intr_info = ent_intr_info_base |
4347 					INTR_TYPE_SOFT_INTR;
4348 			break;
4349 		case 1:
4350 			ent_intr_info = ent_intr_info_base |
4351 					INTR_TYPE_SOFT_EXCEPTION;
4352 			break;
4353 		case 2:
4354 			ent_intr_info = ent_intr_info_base |
4355 					INTR_TYPE_PRIV_SW_EXCEPTION;
4356 			break;
4357 		}
4358 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4359 				    "VM-entry instruction-length check",
4360 				    ent_intr_info);
4361 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4362 
4363 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4364 		ent_intr_len = -1;
4365 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4366 				    ent_intr_len);
4367 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4368 		test_vmx_invalid_controls();
4369 		report_prefix_pop();
4370 
4371 		/* Instruction length set to 16 should fail */
4372 		ent_intr_len = 0x00000010;
4373 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4374 				    ent_intr_len);
4375 		vmcs_write(ENT_INST_LEN, 0x00000010);
4376 		test_vmx_invalid_controls();
4377 		report_prefix_pop();
4378 
4379 		report_prefix_pop();
4380 	}
4381 
4382 	/* Cleanup */
4383 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4384 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4385 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4386 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4387 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4388 	vmcs_write(GUEST_CR0, guest_cr0_save);
4389 	report_prefix_pop();
4390 }
4391 
4392 /*
4393  * Test interesting vTPR values for a given TPR threshold.
4394  */
4395 static void test_vtpr_values(unsigned threshold)
4396 {
4397 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4398 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4399 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4400 }
4401 
4402 static void try_tpr_threshold(unsigned threshold)
4403 {
4404 	bool valid = true;
4405 
4406 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4407 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4408 
4409 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4410 	    (secondary & CPU_VINTD)))
4411 		valid = !(threshold >> 4);
4412 
4413 	set_vtpr(-1);
4414 	vmcs_write(TPR_THRESHOLD, threshold);
4415 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4416 	if (valid)
4417 		test_vmx_valid_controls();
4418 	else
4419 		test_vmx_invalid_controls();
4420 	report_prefix_pop();
4421 
4422 	if (valid)
4423 		test_vtpr_values(threshold);
4424 }
4425 
4426 /*
4427  * Test interesting TPR threshold values.
4428  */
4429 static void test_tpr_threshold_values(void)
4430 {
4431 	unsigned i;
4432 
4433 	for (i = 0; i < 0x10; i++)
4434 		try_tpr_threshold(i);
4435 	for (i = 4; i < 32; i++)
4436 		try_tpr_threshold(1u << i);
4437 	try_tpr_threshold(-1u);
4438 	try_tpr_threshold(0x7fffffff);
4439 }
4440 
4441 /*
4442  * This test covers the following two VM entry checks:
4443  *
4444  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4445  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4446  *         31:4 of the TPR threshold VM-execution control field must
4447 	   be 0.
4448  *         [Intel SDM]
4449  *
4450  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4451  *          "virtual-interrupt delivery" VM-execution control is 0
4452  *          and the "virtualize APIC accesses" VM-execution control
4453  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4454  *          control field must not be greater than the value of bits
4455  *          7:4 of VTPR.
4456  *          [Intel SDM]
4457  */
4458 static void test_tpr_threshold(void)
4459 {
4460 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4461 	u64 apic_virt_addr = vmcs_read(APIC_VIRT_ADDR);
4462 	u64 threshold = vmcs_read(TPR_THRESHOLD);
4463 	void *virtual_apic_page;
4464 
4465 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4466 		return;
4467 
4468 	virtual_apic_page = alloc_page();
4469 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4470 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4471 
4472 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4473 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4474 	test_tpr_threshold_values();
4475 	report_prefix_pop();
4476 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4477 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4478 	test_tpr_threshold_values();
4479 	report_prefix_pop();
4480 
4481 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4482 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES))))
4483 		goto out;
4484 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4485 
4486 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4487 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4488 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4489 		test_tpr_threshold_values();
4490 		report_prefix_pop();
4491 
4492 		vmcs_write(CPU_EXEC_CTRL0,
4493 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4494 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4495 		test_tpr_threshold_values();
4496 		report_prefix_pop();
4497 	}
4498 
4499 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4500 		vmcs_write(CPU_EXEC_CTRL0,
4501 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4502 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4503 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4504 		test_tpr_threshold_values();
4505 		report_prefix_pop();
4506 
4507 		vmcs_write(CPU_EXEC_CTRL0,
4508 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4509 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4510 		test_tpr_threshold_values();
4511 		report_prefix_pop();
4512 	}
4513 
4514 	if ((ctrl_cpu_rev[1].clr &
4515 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4516 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4517 		vmcs_write(CPU_EXEC_CTRL0,
4518 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4519 		vmcs_write(CPU_EXEC_CTRL1,
4520 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4521 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4522 		test_tpr_threshold_values();
4523 		report_prefix_pop();
4524 
4525 		vmcs_write(CPU_EXEC_CTRL0,
4526 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4527 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4528 		test_tpr_threshold_values();
4529 		report_prefix_pop();
4530 	}
4531 
4532 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4533 out:
4534 	vmcs_write(TPR_THRESHOLD, threshold);
4535 	vmcs_write(APIC_VIRT_ADDR, apic_virt_addr);
4536 	vmcs_write(CPU_EXEC_CTRL0, primary);
4537 }
4538 
4539 /*
4540  * This test verifies the following two vmentry checks:
4541  *
4542  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4543  *  VM-execution control must be 0.
4544  *  [Intel SDM]
4545  *
4546  *  If the "virtual NMIs" VM-execution control is 0, the "NMI-window
4547  *  exiting" VM-execution control must be 0.
4548  *  [Intel SDM]
4549  */
4550 static void test_nmi_ctrls(void)
4551 {
4552 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4553 
4554 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4555 	    (PIN_NMI | PIN_VIRT_NMI)) {
4556 		report_skip("%s : NMI exiting and/or Virtual NMIs not supported", __func__);
4557 		return;
4558 	}
4559 
4560 	/* Save the controls so that we can restore them after our tests */
4561 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4562 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4563 
4564 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4565 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4566 
4567 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4568 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4569 	test_vmx_valid_controls();
4570 	report_prefix_pop();
4571 
4572 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4573 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4574 	test_vmx_invalid_controls();
4575 	report_prefix_pop();
4576 
4577 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4578 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4579 	test_vmx_valid_controls();
4580 	report_prefix_pop();
4581 
4582 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4583 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4584 	test_vmx_valid_controls();
4585 	report_prefix_pop();
4586 
4587 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4588 		report_info("NMI-window exiting is not supported, skipping...");
4589 		goto done;
4590 	}
4591 
4592 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4593 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4594 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4595 	test_vmx_invalid_controls();
4596 	report_prefix_pop();
4597 
4598 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4599 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4600 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4601 	test_vmx_valid_controls();
4602 	report_prefix_pop();
4603 
4604 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4605 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4606 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4607 	test_vmx_valid_controls();
4608 	report_prefix_pop();
4609 
4610 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4611 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4612 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4613 	test_vmx_valid_controls();
4614 	report_prefix_pop();
4615 
4616 	/* Restore the controls to their original values */
4617 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4618 done:
4619 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4620 }
4621 
4622 static void test_eptp_ad_bit(u64 eptp, bool is_ctrl_valid)
4623 {
4624 	vmcs_write(EPTP, eptp);
4625 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4626 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4627 	if (is_ctrl_valid)
4628 		test_vmx_valid_controls();
4629 	else
4630 		test_vmx_invalid_controls();
4631 	report_prefix_pop();
4632 
4633 }
4634 
4635 /*
4636  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4637  *    control field must satisfy the following checks:
4638  *
4639  *     - The EPT memory type (bits 2:0) must be a value supported by the
4640  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4641  *     - Bits 5:3 (1 less than the EPT page-walk length) must indicate a
4642  *	 supported EPT page-walk length.
4643  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4644  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4645  *	 indicating that the processor does not support accessed and dirty
4646  *	 dirty flags for EPT.
4647  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4648  *	 physical-address width) must all be 0.
4649  *
4650  * 2. If the "unrestricted guest" VM-execution control is 1, the
4651  *    "enable EPT" VM-execution control must also be 1.
4652  */
4653 static void test_ept_eptp(void)
4654 {
4655 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4656 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4657 	u64 eptp_saved = vmcs_read(EPTP);
4658 	u32 primary = primary_saved;
4659 	u32 secondary = secondary_saved;
4660 	u64 eptp = eptp_saved;
4661 	u32 i, maxphysaddr;
4662 	u64 j, resv_bits_mask = 0;
4663 
4664 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4665 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4666 		report_skip("%s : \"CPU secondary\" and/or \"enable EPT\" exec control not supported", __func__);
4667 		return;
4668 	}
4669 
4670 	/* Support for 4-level EPT is mandatory. */
4671 	report(is_4_level_ept_supported(), "4-level EPT support check");
4672 
4673 	primary |= CPU_SECONDARY;
4674 	vmcs_write(CPU_EXEC_CTRL0, primary);
4675 	secondary |= CPU_EPT;
4676 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4677 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4678 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4679 	vmcs_write(EPTP, eptp);
4680 
4681 	for (i = 0; i < 8; i++) {
4682 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4683 		vmcs_write(EPTP, eptp);
4684 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4685 		    eptp & EPT_MEM_TYPE_MASK);
4686 		if (is_ept_memtype_supported(i))
4687 			test_vmx_valid_controls();
4688 		else
4689 			test_vmx_invalid_controls();
4690 		report_prefix_pop();
4691 	}
4692 
4693 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4694 
4695 	/*
4696 	 * Page walk length (bits 5:3).  Note, the value in VMCS.EPTP "is 1
4697 	 * less than the EPT page-walk length".
4698 	 */
4699 	for (i = 0; i < 8; i++) {
4700 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4701 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4702 
4703 		vmcs_write(EPTP, eptp);
4704 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4705 		    eptp & EPTP_PG_WALK_LEN_MASK);
4706 		if (i == 3 || (i == 4 && is_5_level_ept_supported()))
4707 			test_vmx_valid_controls();
4708 		else
4709 			test_vmx_invalid_controls();
4710 		report_prefix_pop();
4711 	}
4712 
4713 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4714 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4715 
4716 	/*
4717 	 * Accessed and dirty flag (bit 6)
4718 	 */
4719 	if (ept_ad_bits_supported()) {
4720 		report_info("Processor supports accessed and dirty flag");
4721 		eptp &= ~EPTP_AD_FLAG;
4722 		test_eptp_ad_bit(eptp, true);
4723 
4724 		eptp |= EPTP_AD_FLAG;
4725 		test_eptp_ad_bit(eptp, true);
4726 	} else {
4727 		report_info("Processor does not supports accessed and dirty flag");
4728 		eptp &= ~EPTP_AD_FLAG;
4729 		test_eptp_ad_bit(eptp, true);
4730 
4731 		eptp |= EPTP_AD_FLAG;
4732 		test_eptp_ad_bit(eptp, false);
4733 	}
4734 
4735 	/*
4736 	 * Reserved bits [11:7] and [63:N]
4737 	 */
4738 	for (i = 0; i < 32; i++) {
4739 		eptp = (eptp &
4740 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4741 		    (i << EPTP_RESERV_BITS_SHIFT);
4742 		vmcs_write(EPTP, eptp);
4743 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4744 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4745 		    EPTP_RESERV_BITS_MASK);
4746 		if (i == 0)
4747 			test_vmx_valid_controls();
4748 		else
4749 			test_vmx_invalid_controls();
4750 		report_prefix_pop();
4751 	}
4752 
4753 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4754 
4755 	maxphysaddr = cpuid_maxphyaddr();
4756 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4757 		resv_bits_mask |= 1ul << i;
4758 	}
4759 
4760 	for (j = maxphysaddr - 1; j <= 63; j++) {
4761 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4762 		    (j < maxphysaddr ? 0 : 1ul << j);
4763 		vmcs_write(EPTP, eptp);
4764 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4765 		    (eptp >> maxphysaddr) & resv_bits_mask);
4766 		if (j < maxphysaddr)
4767 			test_vmx_valid_controls();
4768 		else
4769 			test_vmx_invalid_controls();
4770 		report_prefix_pop();
4771 	}
4772 
4773 	secondary &= ~(CPU_EPT | CPU_URG);
4774 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4775 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4776 	test_vmx_valid_controls();
4777 	report_prefix_pop();
4778 
4779 	if (!(ctrl_cpu_rev[1].clr & CPU_URG))
4780 		goto skip_unrestricted_guest;
4781 
4782 	secondary |= CPU_URG;
4783 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4784 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4785 	test_vmx_invalid_controls();
4786 	report_prefix_pop();
4787 
4788 	secondary |= CPU_EPT;
4789 	setup_dummy_ept();
4790 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4791 	test_vmx_valid_controls();
4792 	report_prefix_pop();
4793 
4794 skip_unrestricted_guest:
4795 	secondary &= ~CPU_URG;
4796 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4797 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4798 	test_vmx_valid_controls();
4799 	report_prefix_pop();
4800 
4801 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4802 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4803 	vmcs_write(EPTP, eptp_saved);
4804 }
4805 
4806 /*
4807  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4808  * VM-execution control must also be 1. In addition, the PML address
4809  * must satisfy the following checks:
4810  *
4811  *    * Bits 11:0 of the address must be 0.
4812  *    * The address should not set any bits beyond the processor's
4813  *	physical-address width.
4814  *
4815  *  [Intel SDM]
4816  */
4817 static void test_pml(void)
4818 {
4819 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4820 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4821 	u32 primary = primary_saved;
4822 	u32 secondary = secondary_saved;
4823 
4824 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4825 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
4826 		report_skip("%s : \"Secondary execution\" or \"enable EPT\" or \"enable PML\" control not supported", __func__);
4827 		return;
4828 	}
4829 
4830 	primary |= CPU_SECONDARY;
4831 	vmcs_write(CPU_EXEC_CTRL0, primary);
4832 	secondary &= ~(CPU_PML | CPU_EPT);
4833 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4834 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
4835 	test_vmx_valid_controls();
4836 	report_prefix_pop();
4837 
4838 	secondary |= CPU_PML;
4839 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4840 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
4841 	test_vmx_invalid_controls();
4842 	report_prefix_pop();
4843 
4844 	secondary |= CPU_EPT;
4845 	setup_dummy_ept();
4846 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
4847 	test_vmx_valid_controls();
4848 	report_prefix_pop();
4849 
4850 	secondary &= ~CPU_PML;
4851 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4852 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
4853 	test_vmx_valid_controls();
4854 	report_prefix_pop();
4855 
4856 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
4857 				 PAGE_SIZE, false, false);
4858 
4859 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4860 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4861 }
4862 
4863  /*
4864  * If the "activate VMX-preemption timer" VM-execution control is 0, the
4865  * the "save VMX-preemption timer value" VM-exit control must also be 0.
4866  *
4867  *  [Intel SDM]
4868  */
4869 static void test_vmx_preemption_timer(void)
4870 {
4871 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4872 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
4873 	u32 pin = saved_pin;
4874 	u32 exit = saved_exit;
4875 
4876 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
4877 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
4878 		report_skip("%s : \"Save-VMX-preemption-timer\" and/or \"Enable-VMX-preemption-timer\" control not supported", __func__);
4879 		return;
4880 	}
4881 
4882 	pin |= PIN_PREEMPT;
4883 	vmcs_write(PIN_CONTROLS, pin);
4884 	exit &= ~EXI_SAVE_PREEMPT;
4885 	vmcs_write(EXI_CONTROLS, exit);
4886 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
4887 	test_vmx_valid_controls();
4888 	report_prefix_pop();
4889 
4890 	exit |= EXI_SAVE_PREEMPT;
4891 	vmcs_write(EXI_CONTROLS, exit);
4892 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
4893 	test_vmx_valid_controls();
4894 	report_prefix_pop();
4895 
4896 	pin &= ~PIN_PREEMPT;
4897 	vmcs_write(PIN_CONTROLS, pin);
4898 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
4899 	test_vmx_invalid_controls();
4900 	report_prefix_pop();
4901 
4902 	exit &= ~EXI_SAVE_PREEMPT;
4903 	vmcs_write(EXI_CONTROLS, exit);
4904 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
4905 	test_vmx_valid_controls();
4906 	report_prefix_pop();
4907 
4908 	vmcs_write(PIN_CONTROLS, saved_pin);
4909 	vmcs_write(EXI_CONTROLS, saved_exit);
4910 }
4911 
4912 extern unsigned char test_mtf1;
4913 extern unsigned char test_mtf2;
4914 extern unsigned char test_mtf3;
4915 extern unsigned char test_mtf4;
4916 
4917 static void test_mtf_guest(void)
4918 {
4919 	asm ("vmcall;\n\t"
4920 	     "out %al, $0x80;\n\t"
4921 	     "test_mtf1:\n\t"
4922 	     "vmcall;\n\t"
4923 	     "out %al, $0x80;\n\t"
4924 	     "test_mtf2:\n\t"
4925 	     /*
4926 	      * Prepare for the 'MOV CR3' test. Attempt to induce a
4927 	      * general-protection fault by moving a non-canonical address into
4928 	      * CR3. The 'MOV CR3' instruction does not take an imm64 operand,
4929 	      * so we must MOV the desired value into a register first.
4930 	      *
4931 	      * MOV RAX is done before the VMCALL such that MTF is only enabled
4932 	      * for the instruction under test.
4933 	      */
4934 	     "mov $0xaaaaaaaaaaaaaaaa, %rax;\n\t"
4935 	     "vmcall;\n\t"
4936 	     "mov %rax, %cr3;\n\t"
4937 	     "test_mtf3:\n\t"
4938 	     "vmcall;\n\t"
4939 	     /*
4940 	      * ICEBP/INT1 instruction. Though the instruction is now
4941 	      * documented, don't rely on assemblers enumerating the
4942 	      * instruction. Resort to hand assembly.
4943 	      */
4944 	     ".byte 0xf1;\n\t"
4945 	     "vmcall;\n\t"
4946 	     "test_mtf4:\n\t"
4947 	     "mov $0, %eax;\n\t");
4948 }
4949 
4950 static void test_mtf_gp_handler(struct ex_regs *regs)
4951 {
4952 	regs->rip = (unsigned long) &test_mtf3;
4953 }
4954 
4955 static void test_mtf_db_handler(struct ex_regs *regs)
4956 {
4957 }
4958 
4959 static void enable_mtf(void)
4960 {
4961 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
4962 
4963 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 | CPU_MTF);
4964 }
4965 
4966 static void disable_mtf(void)
4967 {
4968 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
4969 
4970 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 & ~CPU_MTF);
4971 }
4972 
4973 static void enable_tf(void)
4974 {
4975 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
4976 
4977 	vmcs_write(GUEST_RFLAGS, rflags | X86_EFLAGS_TF);
4978 }
4979 
4980 static void disable_tf(void)
4981 {
4982 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
4983 
4984 	vmcs_write(GUEST_RFLAGS, rflags & ~X86_EFLAGS_TF);
4985 }
4986 
4987 static void report_mtf(const char *insn_name, unsigned long exp_rip)
4988 {
4989 	unsigned long rip = vmcs_read(GUEST_RIP);
4990 
4991 	assert_exit_reason(VMX_MTF);
4992 	report(rip == exp_rip, "MTF VM-exit after %s. RIP: 0x%lx (expected 0x%lx)",
4993 	       insn_name, rip, exp_rip);
4994 }
4995 
4996 static void vmx_mtf_test(void)
4997 {
4998 	unsigned long pending_dbg;
4999 	handler old_gp, old_db;
5000 
5001 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5002 		report_skip("%s : \"Monitor trap flag\" exec control not supported", __func__);
5003 		return;
5004 	}
5005 
5006 	test_set_guest(test_mtf_guest);
5007 
5008 	/* Expect an MTF VM-exit after OUT instruction */
5009 	enter_guest();
5010 	skip_exit_vmcall();
5011 
5012 	enable_mtf();
5013 	enter_guest();
5014 	report_mtf("OUT", (unsigned long) &test_mtf1);
5015 	disable_mtf();
5016 
5017 	/*
5018 	 * Concurrent #DB trap and MTF on instruction boundary. Expect MTF
5019 	 * VM-exit with populated 'pending debug exceptions' VMCS field.
5020 	 */
5021 	enter_guest();
5022 	skip_exit_vmcall();
5023 
5024 	enable_mtf();
5025 	enable_tf();
5026 
5027 	enter_guest();
5028 	report_mtf("OUT", (unsigned long) &test_mtf2);
5029 	pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
5030 	report(pending_dbg & DR6_BS,
5031 	       "'pending debug exceptions' field after MTF VM-exit: 0x%lx (expected 0x%lx)",
5032 	       pending_dbg, (unsigned long) DR6_BS);
5033 
5034 	disable_mtf();
5035 	disable_tf();
5036 	vmcs_write(GUEST_PENDING_DEBUG, 0);
5037 
5038 	/*
5039 	 * #GP exception takes priority over MTF. Expect MTF VM-exit with RIP
5040 	 * advanced to first instruction of #GP handler.
5041 	 */
5042 	enter_guest();
5043 	skip_exit_vmcall();
5044 
5045 	old_gp = handle_exception(GP_VECTOR, test_mtf_gp_handler);
5046 
5047 	enable_mtf();
5048 	enter_guest();
5049 	report_mtf("MOV CR3", (unsigned long) get_idt_addr(&boot_idt[GP_VECTOR]));
5050 	disable_mtf();
5051 
5052 	/*
5053 	 * Concurrent MTF and privileged software exception (i.e. ICEBP/INT1).
5054 	 * MTF should follow the delivery of #DB trap, though the SDM doesn't
5055 	 * provide clear indication of the relative priority.
5056 	 */
5057 	enter_guest();
5058 	skip_exit_vmcall();
5059 
5060 	handle_exception(GP_VECTOR, old_gp);
5061 	old_db = handle_exception(DB_VECTOR, test_mtf_db_handler);
5062 
5063 	enable_mtf();
5064 	enter_guest();
5065 	report_mtf("INT1", (unsigned long) get_idt_addr(&boot_idt[DB_VECTOR]));
5066 	disable_mtf();
5067 
5068 	enter_guest();
5069 	skip_exit_vmcall();
5070 	handle_exception(DB_VECTOR, old_db);
5071 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK | INTR_TYPE_OTHER_EVENT);
5072 	enter_guest();
5073 	report_mtf("injected MTF", (unsigned long) &test_mtf4);
5074 	enter_guest();
5075 }
5076 
5077 extern char vmx_mtf_pdpte_guest_begin;
5078 extern char vmx_mtf_pdpte_guest_end;
5079 
5080 asm("vmx_mtf_pdpte_guest_begin:\n\t"
5081     "mov %cr0, %rax\n\t"    /* save CR0 with PG=1                 */
5082     "vmcall\n\t"            /* on return from this CR0.PG=0       */
5083     "mov %rax, %cr0\n\t"    /* restore CR0.PG=1 to enter PAE mode */
5084     "vmcall\n\t"
5085     "retq\n\t"
5086     "vmx_mtf_pdpte_guest_end:");
5087 
5088 static void vmx_mtf_pdpte_test(void)
5089 {
5090 	void *test_mtf_pdpte_guest;
5091 	pteval_t *pdpt;
5092 	u32 guest_ar_cs;
5093 	u64 guest_efer;
5094 	pteval_t *pte;
5095 	u64 guest_cr0;
5096 	u64 guest_cr3;
5097 	u64 guest_cr4;
5098 	u64 ent_ctls;
5099 	int i;
5100 
5101 	if (setup_ept(false))
5102 		return;
5103 
5104 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5105 		report_skip("%s : \"Monitor trap flag\" exec control not supported", __func__);
5106 		return;
5107 	}
5108 
5109 	if (!(ctrl_cpu_rev[1].clr & CPU_URG)) {
5110 		report_skip("%s : \"Unrestricted guest\" exec control not supported", __func__);
5111 		return;
5112 	}
5113 
5114 	vmcs_write(EXC_BITMAP, ~0);
5115 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
5116 
5117 	/*
5118 	 * Copy the guest code to an identity-mapped page.
5119 	 */
5120 	test_mtf_pdpte_guest = alloc_page();
5121 	memcpy(test_mtf_pdpte_guest, &vmx_mtf_pdpte_guest_begin,
5122 	       &vmx_mtf_pdpte_guest_end - &vmx_mtf_pdpte_guest_begin);
5123 
5124 	test_set_guest(test_mtf_pdpte_guest);
5125 
5126 	enter_guest();
5127 	skip_exit_vmcall();
5128 
5129 	/*
5130 	 * Put the guest in non-paged 32-bit protected mode, ready to enter
5131 	 * PAE mode when CR0.PG is set. CR4.PAE will already have been set
5132 	 * when the guest started out in long mode.
5133 	 */
5134 	ent_ctls = vmcs_read(ENT_CONTROLS);
5135 	vmcs_write(ENT_CONTROLS, ent_ctls & ~ENT_GUEST_64);
5136 
5137 	guest_efer = vmcs_read(GUEST_EFER);
5138 	vmcs_write(GUEST_EFER, guest_efer & ~(EFER_LMA | EFER_LME));
5139 
5140 	/*
5141 	 * Set CS access rights bits for 32-bit protected mode:
5142 	 * 3:0    B execute/read/accessed
5143 	 * 4      1 code or data
5144 	 * 6:5    0 descriptor privilege level
5145 	 * 7      1 present
5146 	 * 11:8   0 reserved
5147 	 * 12     0 available for use by system software
5148 	 * 13     0 64 bit mode not active
5149 	 * 14     1 default operation size 32-bit segment
5150 	 * 15     1 page granularity: segment limit in 4K units
5151 	 * 16     0 segment usable
5152 	 * 31:17  0 reserved
5153 	 */
5154 	guest_ar_cs = vmcs_read(GUEST_AR_CS);
5155 	vmcs_write(GUEST_AR_CS, 0xc09b);
5156 
5157 	guest_cr0 = vmcs_read(GUEST_CR0);
5158 	vmcs_write(GUEST_CR0, guest_cr0 & ~X86_CR0_PG);
5159 
5160 	guest_cr4 = vmcs_read(GUEST_CR4);
5161 	vmcs_write(GUEST_CR4, guest_cr4 & ~X86_CR4_PCIDE);
5162 
5163 	guest_cr3 = vmcs_read(GUEST_CR3);
5164 
5165 	/*
5166 	 * Turn the 4-level page table into a PAE page table by following the 0th
5167 	 * PML4 entry to a PDPT page, and grab the first four PDPTEs from that
5168 	 * page.
5169 	 *
5170 	 * Why does this work?
5171 	 *
5172 	 * PAE uses 32-bit addressing which implies:
5173 	 * Bits 11:0   page offset
5174 	 * Bits 20:12  entry into 512-entry page table
5175 	 * Bits 29:21  entry into a 512-entry directory table
5176 	 * Bits 31:30  entry into the page directory pointer table.
5177 	 * Bits 63:32  zero
5178 	 *
5179 	 * As only 2 bits are needed to select the PDPTEs for the entire
5180 	 * 32-bit address space, take the first 4 PDPTEs in the level 3 page
5181 	 * directory pointer table. It doesn't matter which of these PDPTEs
5182 	 * are present because they must cover the guest code given that it
5183 	 * has already run successfully.
5184 	 *
5185 	 * Get a pointer to PTE for GVA=0 in the page directory pointer table
5186 	 */
5187 	pte = get_pte_level(
5188             (pgd_t *)phys_to_virt(guest_cr3 & ~X86_CR3_PCID_MASK), 0,
5189             PDPT_LEVEL);
5190 
5191 	/*
5192 	 * Need some memory for the 4-entry PAE page directory pointer
5193 	 * table. Use the end of the identity-mapped page where the guest code
5194 	 * is stored. There is definitely space as the guest code is only a
5195 	 * few bytes.
5196 	 */
5197 	pdpt = test_mtf_pdpte_guest + PAGE_SIZE - 4 * sizeof(pteval_t);
5198 
5199 	/*
5200 	 * Copy the first four PDPTEs into the PAE page table with reserved
5201 	 * bits cleared. Note that permission bits from the PML4E and PDPTE
5202 	 * are not propagated.
5203 	 */
5204 	for (i = 0; i < 4; i++) {
5205 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_RSVD_MASK),
5206 				   "PDPTE has invalid reserved bits");
5207 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_PAGE_SIZE_MASK),
5208 				   "Cannot use 1GB super pages for PAE");
5209 		pdpt[i] = pte[i] & ~(PAE_PDPTE_RSVD_MASK);
5210 	}
5211 	vmcs_write(GUEST_CR3, virt_to_phys(pdpt));
5212 
5213 	enable_mtf();
5214 	enter_guest();
5215 	assert_exit_reason(VMX_MTF);
5216 	disable_mtf();
5217 
5218 	/*
5219 	 * The four PDPTEs should have been loaded into the VMCS when
5220 	 * the guest set CR0.PG to enter PAE mode.
5221 	 */
5222 	for (i = 0; i < 4; i++) {
5223 		u64 pdpte = vmcs_read(GUEST_PDPTE + 2 * i);
5224 
5225 		report(pdpte == pdpt[i], "PDPTE%d is 0x%lx (expected 0x%lx)",
5226 		       i, pdpte, pdpt[i]);
5227 	}
5228 
5229 	/*
5230 	 * Now, try to enter the guest in PAE mode. If the PDPTEs in the
5231 	 * vmcs are wrong, this will fail.
5232 	 */
5233 	enter_guest();
5234 	skip_exit_vmcall();
5235 
5236 	/*
5237 	 * Return guest to 64-bit mode and wrap up.
5238 	 */
5239 	vmcs_write(ENT_CONTROLS, ent_ctls);
5240 	vmcs_write(GUEST_EFER, guest_efer);
5241 	vmcs_write(GUEST_AR_CS, guest_ar_cs);
5242 	vmcs_write(GUEST_CR0, guest_cr0);
5243 	vmcs_write(GUEST_CR4, guest_cr4);
5244 	vmcs_write(GUEST_CR3, guest_cr3);
5245 
5246 	enter_guest();
5247 }
5248 
5249 /*
5250  * Tests for VM-execution control fields
5251  */
5252 static void test_vm_execution_ctls(void)
5253 {
5254 	test_pin_based_ctls();
5255 	test_primary_processor_based_ctls();
5256 	test_secondary_processor_based_ctls();
5257 	test_cr3_targets();
5258 	test_io_bitmaps();
5259 	test_msr_bitmap();
5260 	test_apic_ctls();
5261 	test_tpr_threshold();
5262 	test_nmi_ctrls();
5263 	test_pml();
5264 	test_vpid();
5265 	test_ept_eptp();
5266 	test_vmx_preemption_timer();
5267 }
5268 
5269  /*
5270   * The following checks are performed for the VM-entry MSR-load address if
5271   * the VM-entry MSR-load count field is non-zero:
5272   *
5273   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
5274   *      The address should not set any bits beyond the processor's
5275   *      physical-address width.
5276   *
5277   *    - The address of the last byte in the VM-entry MSR-load area
5278   *      should not set any bits beyond the processor's physical-address
5279   *      width. The address of this last byte is VM-entry MSR-load address
5280   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5281   *      uses more bits than the processor's physical-address width.)
5282   *
5283   *
5284   *  [Intel SDM]
5285   */
5286 static void test_entry_msr_load(void)
5287 {
5288 	entry_msr_load = alloc_page();
5289 	u64 tmp;
5290 	u32 entry_msr_ld_cnt = 1;
5291 	int i;
5292 	u32 addr_len = 64;
5293 
5294 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
5295 
5296 	/* Check first 4 bits of VM-entry MSR-load address */
5297 	for (i = 0; i < 4; i++) {
5298 		tmp = (u64)entry_msr_load | 1ull << i;
5299 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
5300 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
5301 				    tmp & 0xf);
5302 		test_vmx_invalid_controls();
5303 		report_prefix_pop();
5304 	}
5305 
5306 	if (basic.val & (1ul << 48))
5307 		addr_len = 32;
5308 
5309 	test_vmcs_addr_values("VM-entry-MSR-load address",
5310 				ENTER_MSR_LD_ADDR, 16, false, false,
5311 				4, addr_len - 1);
5312 
5313 	/*
5314 	 * Check last byte of VM-entry MSR-load address
5315 	 */
5316 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
5317 
5318 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5319 							i < 64; i++) {
5320 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
5321 			1ul << i;
5322 		vmcs_write(ENTER_MSR_LD_ADDR,
5323 			   tmp - (entry_msr_ld_cnt * 16 - 1));
5324 		test_vmx_invalid_controls();
5325 	}
5326 
5327 	vmcs_write(ENT_MSR_LD_CNT, 2);
5328 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5329 	test_vmx_invalid_controls();
5330 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5331 	test_vmx_valid_controls();
5332 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5333 	test_vmx_valid_controls();
5334 }
5335 
5336 static struct vmx_state_area_test_data {
5337 	u32 msr;
5338 	u64 exp;
5339 	bool enabled;
5340 } vmx_state_area_test_data;
5341 
5342 static void guest_state_test_main(void)
5343 {
5344 	u64 obs;
5345 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
5346 
5347 	while (1) {
5348 		if (vmx_get_test_stage() == 2)
5349 			break;
5350 
5351 		if (data->enabled) {
5352 			obs = rdmsr(data->msr);
5353 			report(data->exp == obs,
5354 			       "Guest state is 0x%lx (expected 0x%lx)",
5355 			       obs, data->exp);
5356 		}
5357 
5358 		vmcall();
5359 	}
5360 
5361 	asm volatile("fnop");
5362 }
5363 
5364 static void test_guest_state(const char *test, bool xfail, u64 field,
5365 			     const char * field_name)
5366 {
5367 	struct vmentry_result result;
5368 	u8 abort_flags;
5369 
5370 	abort_flags = ABORT_ON_EARLY_VMENTRY_FAIL;
5371 	if (!xfail)
5372 		abort_flags = ABORT_ON_INVALID_GUEST_STATE;
5373 
5374 	__enter_guest(abort_flags, &result);
5375 
5376 	report(result.exit_reason.failed_vmentry == xfail &&
5377 	       ((xfail && result.exit_reason.basic == VMX_FAIL_STATE) ||
5378 	        (!xfail && result.exit_reason.basic == VMX_VMCALL)) &&
5379 		(!xfail || vmcs_read(EXI_QUALIFICATION) == ENTRY_FAIL_DEFAULT),
5380 	        "%s, %s = %lx", test, field_name, field);
5381 
5382 	if (!result.exit_reason.failed_vmentry)
5383 		skip_exit_insn();
5384 }
5385 
5386 /*
5387  * Tests for VM-entry control fields
5388  */
5389 static void test_vm_entry_ctls(void)
5390 {
5391 	test_invalid_event_injection();
5392 	test_entry_msr_load();
5393 }
5394 
5395 /*
5396  * The following checks are performed for the VM-exit MSR-store address if
5397  * the VM-exit MSR-store count field is non-zero:
5398  *
5399  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
5400  *      The address should not set any bits beyond the processor's
5401  *      physical-address width.
5402  *
5403  *    - The address of the last byte in the VM-exit MSR-store area
5404  *      should not set any bits beyond the processor's physical-address
5405  *      width. The address of this last byte is VM-exit MSR-store address
5406  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5407  *      uses more bits than the processor's physical-address width.)
5408  *
5409  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
5410  * in the range 63:32.
5411  *
5412  *  [Intel SDM]
5413  */
5414 static void test_exit_msr_store(void)
5415 {
5416 	exit_msr_store = alloc_page();
5417 	u64 tmp;
5418 	u32 exit_msr_st_cnt = 1;
5419 	int i;
5420 	u32 addr_len = 64;
5421 
5422 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
5423 
5424 	/* Check first 4 bits of VM-exit MSR-store address */
5425 	for (i = 0; i < 4; i++) {
5426 		tmp = (u64)exit_msr_store | 1ull << i;
5427 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
5428 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
5429 				    tmp & 0xf);
5430 		test_vmx_invalid_controls();
5431 		report_prefix_pop();
5432 	}
5433 
5434 	if (basic.val & (1ul << 48))
5435 		addr_len = 32;
5436 
5437 	test_vmcs_addr_values("VM-exit-MSR-store address",
5438 				EXIT_MSR_ST_ADDR, 16, false, false,
5439 				4, addr_len - 1);
5440 
5441 	/*
5442 	 * Check last byte of VM-exit MSR-store address
5443 	 */
5444 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
5445 
5446 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5447 							i < 64; i++) {
5448 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
5449 			1ul << i;
5450 		vmcs_write(EXIT_MSR_ST_ADDR,
5451 			   tmp - (exit_msr_st_cnt * 16 - 1));
5452 		test_vmx_invalid_controls();
5453 	}
5454 
5455 	vmcs_write(EXI_MSR_ST_CNT, 2);
5456 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5457 	test_vmx_invalid_controls();
5458 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5459 	test_vmx_valid_controls();
5460 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5461 	test_vmx_valid_controls();
5462 }
5463 
5464 /*
5465  * Tests for VM-exit controls
5466  */
5467 static void test_vm_exit_ctls(void)
5468 {
5469 	test_exit_msr_store();
5470 }
5471 
5472 /*
5473  * Check that the virtual CPU checks all of the VMX controls as
5474  * documented in the Intel SDM.
5475  */
5476 static void vmx_controls_test(void)
5477 {
5478 	/*
5479 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
5480 	 * fail due to invalid guest state, should we make it that
5481 	 * far.
5482 	 */
5483 	vmcs_write(GUEST_RFLAGS, 0);
5484 
5485 	test_vm_execution_ctls();
5486 	test_vm_exit_ctls();
5487 	test_vm_entry_ctls();
5488 }
5489 
5490 struct apic_reg_virt_config {
5491 	bool apic_register_virtualization;
5492 	bool use_tpr_shadow;
5493 	bool virtualize_apic_accesses;
5494 	bool virtualize_x2apic_mode;
5495 	bool activate_secondary_controls;
5496 };
5497 
5498 struct apic_reg_test {
5499 	const char *name;
5500 	struct apic_reg_virt_config apic_reg_virt_config;
5501 };
5502 
5503 struct apic_reg_virt_expectation {
5504 	enum Reason rd_exit_reason;
5505 	enum Reason wr_exit_reason;
5506 	u32 val;
5507 	u32 (*virt_fn)(u32);
5508 
5509 	/*
5510 	 * If false, accessing the APIC access address from L2 is treated as a
5511 	 * normal memory operation, rather than triggering virtualization.
5512 	 */
5513 	bool virtualize_apic_accesses;
5514 };
5515 
5516 static u32 apic_virt_identity(u32 val)
5517 {
5518 	return val;
5519 }
5520 
5521 static u32 apic_virt_nibble1(u32 val)
5522 {
5523 	return val & 0xf0;
5524 }
5525 
5526 static u32 apic_virt_byte3(u32 val)
5527 {
5528 	return val & (0xff << 24);
5529 }
5530 
5531 static bool apic_reg_virt_exit_expectation(
5532 	u32 reg, struct apic_reg_virt_config *config,
5533 	struct apic_reg_virt_expectation *expectation)
5534 {
5535 	/* Good configs, where some L2 APIC accesses are virtualized. */
5536 	bool virtualize_apic_accesses_only =
5537 		config->virtualize_apic_accesses &&
5538 		!config->use_tpr_shadow &&
5539 		!config->apic_register_virtualization &&
5540 		!config->virtualize_x2apic_mode &&
5541 		config->activate_secondary_controls;
5542 	bool virtualize_apic_accesses_and_use_tpr_shadow =
5543 		config->virtualize_apic_accesses &&
5544 		config->use_tpr_shadow &&
5545 		!config->apic_register_virtualization &&
5546 		!config->virtualize_x2apic_mode &&
5547 		config->activate_secondary_controls;
5548 	bool apic_register_virtualization =
5549 		config->virtualize_apic_accesses &&
5550 		config->use_tpr_shadow &&
5551 		config->apic_register_virtualization &&
5552 		!config->virtualize_x2apic_mode &&
5553 		config->activate_secondary_controls;
5554 
5555 	expectation->val = MAGIC_VAL_1;
5556 	expectation->virt_fn = apic_virt_identity;
5557 	expectation->virtualize_apic_accesses =
5558 		config->virtualize_apic_accesses &&
5559 		config->activate_secondary_controls;
5560 	if (virtualize_apic_accesses_only) {
5561 		expectation->rd_exit_reason = VMX_APIC_ACCESS;
5562 		expectation->wr_exit_reason = VMX_APIC_ACCESS;
5563 	} else if (virtualize_apic_accesses_and_use_tpr_shadow) {
5564 		switch (reg) {
5565 		case APIC_TASKPRI:
5566 			expectation->rd_exit_reason = VMX_VMCALL;
5567 			expectation->wr_exit_reason = VMX_VMCALL;
5568 			expectation->virt_fn = apic_virt_nibble1;
5569 			break;
5570 		default:
5571 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5572 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5573 		}
5574 	} else if (apic_register_virtualization) {
5575 		expectation->rd_exit_reason = VMX_VMCALL;
5576 
5577 		switch (reg) {
5578 		case APIC_ID:
5579 		case APIC_EOI:
5580 		case APIC_LDR:
5581 		case APIC_DFR:
5582 		case APIC_SPIV:
5583 		case APIC_ESR:
5584 		case APIC_ICR:
5585 		case APIC_LVTT:
5586 		case APIC_LVTTHMR:
5587 		case APIC_LVTPC:
5588 		case APIC_LVT0:
5589 		case APIC_LVT1:
5590 		case APIC_LVTERR:
5591 		case APIC_TMICT:
5592 		case APIC_TDCR:
5593 			expectation->wr_exit_reason = VMX_APIC_WRITE;
5594 			break;
5595 		case APIC_LVR:
5596 		case APIC_ISR ... APIC_ISR + 0x70:
5597 		case APIC_TMR ... APIC_TMR + 0x70:
5598 		case APIC_IRR ... APIC_IRR + 0x70:
5599 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5600 			break;
5601 		case APIC_TASKPRI:
5602 			expectation->wr_exit_reason = VMX_VMCALL;
5603 			expectation->virt_fn = apic_virt_nibble1;
5604 			break;
5605 		case APIC_ICR2:
5606 			expectation->wr_exit_reason = VMX_VMCALL;
5607 			expectation->virt_fn = apic_virt_byte3;
5608 			break;
5609 		default:
5610 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5611 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5612 		}
5613 	} else if (!expectation->virtualize_apic_accesses) {
5614 		/*
5615 		 * No APIC registers are directly virtualized. This includes
5616 		 * VTPR, which can be virtualized through MOV to/from CR8 via
5617 		 * the use TPR shadow control, but not through directly
5618 		 * accessing VTPR.
5619 		 */
5620 		expectation->rd_exit_reason = VMX_VMCALL;
5621 		expectation->wr_exit_reason = VMX_VMCALL;
5622 	} else {
5623 		printf("Cannot parse APIC register virtualization config:\n"
5624 		       "\tvirtualize_apic_accesses: %d\n"
5625 		       "\tuse_tpr_shadow: %d\n"
5626 		       "\tapic_register_virtualization: %d\n"
5627 		       "\tvirtualize_x2apic_mode: %d\n"
5628 		       "\tactivate_secondary_controls: %d\n",
5629 		       config->virtualize_apic_accesses,
5630 		       config->use_tpr_shadow,
5631 		       config->apic_register_virtualization,
5632 		       config->virtualize_x2apic_mode,
5633 		       config->activate_secondary_controls);
5634 
5635 		return false;
5636 	}
5637 
5638 	return true;
5639 }
5640 
5641 struct apic_reg_test apic_reg_tests[] = {
5642 	/* Good configs, where some L2 APIC accesses are virtualized. */
5643 	{
5644 		.name = "Virtualize APIC accesses",
5645 		.apic_reg_virt_config = {
5646 			.virtualize_apic_accesses = true,
5647 			.use_tpr_shadow = false,
5648 			.apic_register_virtualization = false,
5649 			.virtualize_x2apic_mode = false,
5650 			.activate_secondary_controls = true,
5651 		},
5652 	},
5653 	{
5654 		.name = "Virtualize APIC accesses + Use TPR shadow",
5655 		.apic_reg_virt_config = {
5656 			.virtualize_apic_accesses = true,
5657 			.use_tpr_shadow = true,
5658 			.apic_register_virtualization = false,
5659 			.virtualize_x2apic_mode = false,
5660 			.activate_secondary_controls = true,
5661 		},
5662 	},
5663 	{
5664 		.name = "APIC-register virtualization",
5665 		.apic_reg_virt_config = {
5666 			.virtualize_apic_accesses = true,
5667 			.use_tpr_shadow = true,
5668 			.apic_register_virtualization = true,
5669 			.virtualize_x2apic_mode = false,
5670 			.activate_secondary_controls = true,
5671 		},
5672 	},
5673 
5674 	/*
5675 	 * Test that the secondary processor-based VM-execution controls are
5676 	 * correctly ignored when "activate secondary controls" is disabled.
5677 	 */
5678 	{
5679 		.name = "Activate secondary controls off",
5680 		.apic_reg_virt_config = {
5681 			.virtualize_apic_accesses = true,
5682 			.use_tpr_shadow = false,
5683 			.apic_register_virtualization = true,
5684 			.virtualize_x2apic_mode = true,
5685 			.activate_secondary_controls = false,
5686 		},
5687 	},
5688 	{
5689 		.name = "Activate secondary controls off + Use TPR shadow",
5690 		.apic_reg_virt_config = {
5691 			.virtualize_apic_accesses = true,
5692 			.use_tpr_shadow = true,
5693 			.apic_register_virtualization = true,
5694 			.virtualize_x2apic_mode = true,
5695 			.activate_secondary_controls = false,
5696 		},
5697 	},
5698 
5699 	/*
5700 	 * Test that the APIC access address is treated like an arbitrary memory
5701 	 * address when "virtualize APIC accesses" is disabled.
5702 	 */
5703 	{
5704 		.name = "Virtualize APIC accesses off + Use TPR shadow",
5705 		.apic_reg_virt_config = {
5706 			.virtualize_apic_accesses = false,
5707 			.use_tpr_shadow = true,
5708 			.apic_register_virtualization = true,
5709 			.virtualize_x2apic_mode = true,
5710 			.activate_secondary_controls = true,
5711 		},
5712 	},
5713 
5714 	/*
5715 	 * Test that VM entry fails due to invalid controls when
5716 	 * "APIC-register virtualization" is enabled while "use TPR shadow" is
5717 	 * disabled.
5718 	 */
5719 	{
5720 		.name = "APIC-register virtualization + Use TPR shadow off",
5721 		.apic_reg_virt_config = {
5722 			.virtualize_apic_accesses = true,
5723 			.use_tpr_shadow = false,
5724 			.apic_register_virtualization = true,
5725 			.virtualize_x2apic_mode = false,
5726 			.activate_secondary_controls = true,
5727 		},
5728 	},
5729 
5730 	/*
5731 	 * Test that VM entry fails due to invalid controls when
5732 	 * "Virtualize x2APIC mode" is enabled while "use TPR shadow" is
5733 	 * disabled.
5734 	 */
5735 	{
5736 		.name = "Virtualize x2APIC mode + Use TPR shadow off",
5737 		.apic_reg_virt_config = {
5738 			.virtualize_apic_accesses = false,
5739 			.use_tpr_shadow = false,
5740 			.apic_register_virtualization = false,
5741 			.virtualize_x2apic_mode = true,
5742 			.activate_secondary_controls = true,
5743 		},
5744 	},
5745 	{
5746 		.name = "Virtualize x2APIC mode + Use TPR shadow off v2",
5747 		.apic_reg_virt_config = {
5748 			.virtualize_apic_accesses = false,
5749 			.use_tpr_shadow = false,
5750 			.apic_register_virtualization = true,
5751 			.virtualize_x2apic_mode = true,
5752 			.activate_secondary_controls = true,
5753 		},
5754 	},
5755 
5756 	/*
5757 	 * Test that VM entry fails due to invalid controls when
5758 	 * "virtualize x2APIC mode" is enabled while "virtualize APIC accesses"
5759 	 * is enabled.
5760 	 */
5761 	{
5762 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses",
5763 		.apic_reg_virt_config = {
5764 			.virtualize_apic_accesses = true,
5765 			.use_tpr_shadow = true,
5766 			.apic_register_virtualization = false,
5767 			.virtualize_x2apic_mode = true,
5768 			.activate_secondary_controls = true,
5769 		},
5770 	},
5771 	{
5772 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses v2",
5773 		.apic_reg_virt_config = {
5774 			.virtualize_apic_accesses = true,
5775 			.use_tpr_shadow = true,
5776 			.apic_register_virtualization = true,
5777 			.virtualize_x2apic_mode = true,
5778 			.activate_secondary_controls = true,
5779 		},
5780 	},
5781 };
5782 
5783 enum Apic_op {
5784 	APIC_OP_XAPIC_RD,
5785 	APIC_OP_XAPIC_WR,
5786 	TERMINATE,
5787 };
5788 
5789 static u32 vmx_xapic_read(u32 *apic_access_address, u32 reg)
5790 {
5791 	return *(volatile u32 *)((uintptr_t)apic_access_address + reg);
5792 }
5793 
5794 static void vmx_xapic_write(u32 *apic_access_address, u32 reg, u32 val)
5795 {
5796 	*(volatile u32 *)((uintptr_t)apic_access_address + reg) = val;
5797 }
5798 
5799 struct apic_reg_virt_guest_args {
5800 	enum Apic_op op;
5801 	u32 *apic_access_address;
5802 	u32 reg;
5803 	u32 val;
5804 	bool check_rd;
5805 	u32 (*virt_fn)(u32);
5806 } apic_reg_virt_guest_args;
5807 
5808 static void apic_reg_virt_guest(void)
5809 {
5810 	volatile struct apic_reg_virt_guest_args *args =
5811 		&apic_reg_virt_guest_args;
5812 
5813 	for (;;) {
5814 		enum Apic_op op = args->op;
5815 		u32 *apic_access_address = args->apic_access_address;
5816 		u32 reg = args->reg;
5817 		u32 val = args->val;
5818 		bool check_rd = args->check_rd;
5819 		u32 (*virt_fn)(u32) = args->virt_fn;
5820 
5821 		if (op == TERMINATE)
5822 			break;
5823 
5824 		if (op == APIC_OP_XAPIC_RD) {
5825 			u32 ret = vmx_xapic_read(apic_access_address, reg);
5826 
5827 			if (check_rd) {
5828 				u32 want = virt_fn(val);
5829 				u32 got = virt_fn(ret);
5830 
5831 				report(got == want,
5832 				       "read 0x%x, expected 0x%x.", got, want);
5833 			}
5834 		} else if (op == APIC_OP_XAPIC_WR) {
5835 			vmx_xapic_write(apic_access_address, reg, val);
5836 		}
5837 
5838 		/*
5839 		 * The L1 should always execute a vmcall after it's done testing
5840 		 * an individual APIC operation. This helps to validate that the
5841 		 * L1 and L2 are in sync with each other, as expected.
5842 		 */
5843 		vmcall();
5844 	}
5845 }
5846 
5847 static void test_xapic_rd(
5848 	u32 reg, struct apic_reg_virt_expectation *expectation,
5849 	u32 *apic_access_address, u32 *virtual_apic_page)
5850 {
5851 	u32 val = expectation->val;
5852 	u32 exit_reason_want = expectation->rd_exit_reason;
5853 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5854 
5855 	report_prefix_pushf("xapic - reading 0x%03x", reg);
5856 
5857 	/* Configure guest to do an xapic read */
5858 	args->op = APIC_OP_XAPIC_RD;
5859 	args->apic_access_address = apic_access_address;
5860 	args->reg = reg;
5861 	args->val = val;
5862 	args->check_rd = exit_reason_want == VMX_VMCALL;
5863 	args->virt_fn = expectation->virt_fn;
5864 
5865 	/* Setup virtual APIC page */
5866 	if (!expectation->virtualize_apic_accesses) {
5867 		apic_access_address[apic_reg_index(reg)] = val;
5868 		virtual_apic_page[apic_reg_index(reg)] = 0;
5869 	} else if (exit_reason_want == VMX_VMCALL) {
5870 		apic_access_address[apic_reg_index(reg)] = 0;
5871 		virtual_apic_page[apic_reg_index(reg)] = val;
5872 	}
5873 
5874 	/* Enter guest */
5875 	enter_guest();
5876 
5877 	/*
5878 	 * Validate the behavior and
5879 	 * pass a magic value back to the guest.
5880 	 */
5881 	if (exit_reason_want == VMX_APIC_ACCESS) {
5882 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5883 
5884 		assert_exit_reason(exit_reason_want);
5885 		report(apic_page_offset == reg,
5886 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
5887 		       apic_page_offset, reg);
5888 		skip_exit_insn();
5889 
5890 		/* Reenter guest so it can consume/check rcx and exit again. */
5891 		enter_guest();
5892 	} else if (exit_reason_want != VMX_VMCALL) {
5893 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
5894 	}
5895 
5896 	skip_exit_vmcall();
5897 	report_prefix_pop();
5898 }
5899 
5900 static void test_xapic_wr(
5901 	u32 reg, struct apic_reg_virt_expectation *expectation,
5902 	u32 *apic_access_address, u32 *virtual_apic_page)
5903 {
5904 	u32 val = expectation->val;
5905 	u32 exit_reason_want = expectation->wr_exit_reason;
5906 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5907 	bool virtualized =
5908 		expectation->virtualize_apic_accesses &&
5909 		(exit_reason_want == VMX_APIC_WRITE ||
5910 		 exit_reason_want == VMX_VMCALL);
5911 	bool checked = false;
5912 
5913 	report_prefix_pushf("xapic - writing 0x%x to 0x%03x", val, reg);
5914 
5915 	/* Configure guest to do an xapic read */
5916 	args->op = APIC_OP_XAPIC_WR;
5917 	args->apic_access_address = apic_access_address;
5918 	args->reg = reg;
5919 	args->val = val;
5920 
5921 	/* Setup virtual APIC page */
5922 	if (virtualized || !expectation->virtualize_apic_accesses) {
5923 		apic_access_address[apic_reg_index(reg)] = 0;
5924 		virtual_apic_page[apic_reg_index(reg)] = 0;
5925 	}
5926 
5927 	/* Enter guest */
5928 	enter_guest();
5929 
5930 	/*
5931 	 * Validate the behavior and
5932 	 * pass a magic value back to the guest.
5933 	 */
5934 	if (exit_reason_want == VMX_APIC_ACCESS) {
5935 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5936 
5937 		assert_exit_reason(exit_reason_want);
5938 		report(apic_page_offset == reg,
5939 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
5940 		       apic_page_offset, reg);
5941 		skip_exit_insn();
5942 
5943 		/* Reenter guest so it can consume/check rcx and exit again. */
5944 		enter_guest();
5945 	} else if (exit_reason_want == VMX_APIC_WRITE) {
5946 		assert_exit_reason(exit_reason_want);
5947 		report(virtual_apic_page[apic_reg_index(reg)] == val,
5948 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%x",
5949 		       apic_reg_index(reg),
5950 		       virtual_apic_page[apic_reg_index(reg)], val);
5951 		checked = true;
5952 
5953 		/* Reenter guest so it can consume/check rcx and exit again. */
5954 		enter_guest();
5955 	} else if (exit_reason_want != VMX_VMCALL) {
5956 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
5957 	}
5958 
5959 	assert_exit_reason(VMX_VMCALL);
5960 	if (virtualized && !checked) {
5961 		u32 want = expectation->virt_fn(val);
5962 		u32 got = virtual_apic_page[apic_reg_index(reg)];
5963 		got = expectation->virt_fn(got);
5964 
5965 		report(got == want, "exitless write; val is 0x%x, want 0x%x",
5966 		       got, want);
5967 	} else if (!expectation->virtualize_apic_accesses && !checked) {
5968 		u32 got = apic_access_address[apic_reg_index(reg)];
5969 
5970 		report(got == val,
5971 		       "non-virtualized write; val is 0x%x, want 0x%x", got,
5972 		       val);
5973 	} else if (!expectation->virtualize_apic_accesses && checked) {
5974 		report_fail("Non-virtualized write was prematurely checked!");
5975 	}
5976 
5977 	skip_exit_vmcall();
5978 	report_prefix_pop();
5979 }
5980 
5981 enum Config_type {
5982 	CONFIG_TYPE_GOOD,
5983 	CONFIG_TYPE_UNSUPPORTED,
5984 	CONFIG_TYPE_VMENTRY_FAILS_EARLY,
5985 };
5986 
5987 static enum Config_type configure_apic_reg_virt_test(
5988 	struct apic_reg_virt_config *apic_reg_virt_config)
5989 {
5990 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5991 	u32 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
5992 	/* Configs where L2 entry fails early, due to invalid controls. */
5993 	bool use_tpr_shadow_incorrectly_off =
5994 		!apic_reg_virt_config->use_tpr_shadow &&
5995 		(apic_reg_virt_config->apic_register_virtualization ||
5996 		 apic_reg_virt_config->virtualize_x2apic_mode) &&
5997 		apic_reg_virt_config->activate_secondary_controls;
5998 	bool virtualize_apic_accesses_incorrectly_on =
5999 		apic_reg_virt_config->virtualize_apic_accesses &&
6000 		apic_reg_virt_config->virtualize_x2apic_mode &&
6001 		apic_reg_virt_config->activate_secondary_controls;
6002 	bool vmentry_fails_early =
6003 		use_tpr_shadow_incorrectly_off ||
6004 		virtualize_apic_accesses_incorrectly_on;
6005 
6006 	if (apic_reg_virt_config->activate_secondary_controls) {
6007 		if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
6008 			printf("VM-execution control \"activate secondary controls\" NOT supported.\n");
6009 			return CONFIG_TYPE_UNSUPPORTED;
6010 		}
6011 		cpu_exec_ctrl0 |= CPU_SECONDARY;
6012 	} else {
6013 		cpu_exec_ctrl0 &= ~CPU_SECONDARY;
6014 	}
6015 
6016 	if (apic_reg_virt_config->virtualize_apic_accesses) {
6017 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES)) {
6018 			printf("VM-execution control \"virtualize APIC accesses\" NOT supported.\n");
6019 			return CONFIG_TYPE_UNSUPPORTED;
6020 		}
6021 		cpu_exec_ctrl1 |= CPU_VIRT_APIC_ACCESSES;
6022 	} else {
6023 		cpu_exec_ctrl1 &= ~CPU_VIRT_APIC_ACCESSES;
6024 	}
6025 
6026 	if (apic_reg_virt_config->use_tpr_shadow) {
6027 		if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6028 			printf("VM-execution control \"use TPR shadow\" NOT supported.\n");
6029 			return CONFIG_TYPE_UNSUPPORTED;
6030 		}
6031 		cpu_exec_ctrl0 |= CPU_TPR_SHADOW;
6032 	} else {
6033 		cpu_exec_ctrl0 &= ~CPU_TPR_SHADOW;
6034 	}
6035 
6036 	if (apic_reg_virt_config->apic_register_virtualization) {
6037 		if (!(ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT)) {
6038 			printf("VM-execution control \"APIC-register virtualization\" NOT supported.\n");
6039 			return CONFIG_TYPE_UNSUPPORTED;
6040 		}
6041 		cpu_exec_ctrl1 |= CPU_APIC_REG_VIRT;
6042 	} else {
6043 		cpu_exec_ctrl1 &= ~CPU_APIC_REG_VIRT;
6044 	}
6045 
6046 	if (apic_reg_virt_config->virtualize_x2apic_mode) {
6047 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_X2APIC)) {
6048 			printf("VM-execution control \"virtualize x2APIC mode\" NOT supported.\n");
6049 			return CONFIG_TYPE_UNSUPPORTED;
6050 		}
6051 		cpu_exec_ctrl1 |= CPU_VIRT_X2APIC;
6052 	} else {
6053 		cpu_exec_ctrl1 &= ~CPU_VIRT_X2APIC;
6054 	}
6055 
6056 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6057 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6058 
6059 	if (vmentry_fails_early)
6060 		return CONFIG_TYPE_VMENTRY_FAILS_EARLY;
6061 
6062 	return CONFIG_TYPE_GOOD;
6063 }
6064 
6065 static bool cpu_has_apicv(void)
6066 {
6067 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
6068 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
6069 		(ctrl_pin_rev.clr & PIN_POST_INTR));
6070 }
6071 
6072 /* Validates APIC register access across valid virtualization configurations. */
6073 static void apic_reg_virt_test(void)
6074 {
6075 	u32 *apic_access_address;
6076 	u32 *virtual_apic_page;
6077 	u64 control;
6078 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6079 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6080 	int i;
6081 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
6082 
6083 	if (!cpu_has_apicv()) {
6084 		report_skip("%s : Not all required APICv bits supported", __func__);
6085 		return;
6086 	}
6087 
6088 	control = cpu_exec_ctrl1;
6089 	control &= ~CPU_VINTD;
6090 	vmcs_write(CPU_EXEC_CTRL1, control);
6091 
6092 	test_set_guest(apic_reg_virt_guest);
6093 
6094 	/*
6095 	 * From the SDM: The 1-setting of the "virtualize APIC accesses"
6096 	 * VM-execution is guaranteed to apply only if translations to the
6097 	 * APIC-access address use a 4-KByte page.
6098 	 */
6099 	apic_access_address = alloc_page();
6100 	force_4k_page(apic_access_address);
6101 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_address));
6102 
6103 	virtual_apic_page = alloc_page();
6104 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6105 
6106 	for (i = 0; i < ARRAY_SIZE(apic_reg_tests); i++) {
6107 		struct apic_reg_test *apic_reg_test = &apic_reg_tests[i];
6108 		struct apic_reg_virt_config *apic_reg_virt_config =
6109 				&apic_reg_test->apic_reg_virt_config;
6110 		enum Config_type config_type;
6111 		u32 reg;
6112 
6113 		printf("--- %s test ---\n", apic_reg_test->name);
6114 		config_type =
6115 			configure_apic_reg_virt_test(apic_reg_virt_config);
6116 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6117 			printf("Skip because of missing features.\n");
6118 			continue;
6119 		}
6120 
6121 		if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6122 			enter_guest_with_bad_controls();
6123 			continue;
6124 		}
6125 
6126 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6127 			struct apic_reg_virt_expectation expectation = {};
6128 			bool ok;
6129 
6130 			ok = apic_reg_virt_exit_expectation(
6131 				reg, apic_reg_virt_config, &expectation);
6132 			if (!ok) {
6133 				report_fail("Malformed test.");
6134 				break;
6135 			}
6136 
6137 			test_xapic_rd(reg, &expectation, apic_access_address,
6138 				      virtual_apic_page);
6139 			test_xapic_wr(reg, &expectation, apic_access_address,
6140 				      virtual_apic_page);
6141 		}
6142 	}
6143 
6144 	/* Terminate the guest */
6145 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6146 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6147 	args->op = TERMINATE;
6148 	enter_guest();
6149 	assert_exit_reason(VMX_VMCALL);
6150 }
6151 
6152 struct virt_x2apic_mode_config {
6153 	struct apic_reg_virt_config apic_reg_virt_config;
6154 	bool virtual_interrupt_delivery;
6155 	bool use_msr_bitmaps;
6156 	bool disable_x2apic_msr_intercepts;
6157 	bool disable_x2apic;
6158 };
6159 
6160 struct virt_x2apic_mode_test_case {
6161 	const char *name;
6162 	struct virt_x2apic_mode_config virt_x2apic_mode_config;
6163 };
6164 
6165 enum Virt_x2apic_mode_behavior_type {
6166 	X2APIC_ACCESS_VIRTUALIZED,
6167 	X2APIC_ACCESS_PASSED_THROUGH,
6168 	X2APIC_ACCESS_TRIGGERS_GP,
6169 };
6170 
6171 struct virt_x2apic_mode_expectation {
6172 	enum Reason rd_exit_reason;
6173 	enum Reason wr_exit_reason;
6174 
6175 	/*
6176 	 * RDMSR and WRMSR handle 64-bit values. However, except for ICR, all of
6177 	 * the x2APIC registers are 32 bits. Notice:
6178 	 *   1. vmx_x2apic_read() clears the upper 32 bits for 32-bit registers.
6179 	 *   2. vmx_x2apic_write() expects the val arg to be well-formed.
6180 	 */
6181 	u64 rd_val;
6182 	u64 wr_val;
6183 
6184 	/*
6185 	 * Compares input to virtualized output;
6186 	 * 1st arg is pointer to return expected virtualization output.
6187 	 */
6188 	u64 (*virt_fn)(u64);
6189 
6190 	enum Virt_x2apic_mode_behavior_type rd_behavior;
6191 	enum Virt_x2apic_mode_behavior_type wr_behavior;
6192 	bool wr_only;
6193 };
6194 
6195 static u64 virt_x2apic_mode_identity(u64 val)
6196 {
6197 	return val;
6198 }
6199 
6200 static u64 virt_x2apic_mode_nibble1(u64 val)
6201 {
6202 	return val & 0xf0;
6203 }
6204 
6205 static void virt_x2apic_mode_rd_expectation(
6206 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6207 	bool apic_register_virtualization, bool virtual_interrupt_delivery,
6208 	struct virt_x2apic_mode_expectation *expectation)
6209 {
6210 	enum x2apic_reg_semantics semantics = get_x2apic_reg_semantics(reg);
6211 
6212 	expectation->rd_exit_reason = VMX_VMCALL;
6213 	expectation->virt_fn = virt_x2apic_mode_identity;
6214 	if (virt_x2apic_mode_on && apic_register_virtualization) {
6215 		expectation->rd_val = MAGIC_VAL_1;
6216 		if (reg == APIC_PROCPRI && virtual_interrupt_delivery)
6217 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6218 		else if (reg == APIC_TASKPRI)
6219 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6220 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6221 	} else if (virt_x2apic_mode_on && !apic_register_virtualization &&
6222 		   reg == APIC_TASKPRI) {
6223 		expectation->rd_val = MAGIC_VAL_1;
6224 		expectation->virt_fn = virt_x2apic_mode_nibble1;
6225 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6226 	} else if (!disable_x2apic && (semantics & X2APIC_READABLE)) {
6227 		expectation->rd_val = apic_read(reg);
6228 		expectation->rd_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6229 	} else {
6230 		expectation->rd_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6231 	}
6232 }
6233 
6234 /*
6235  * get_x2apic_wr_val() creates an innocuous write value for an x2APIC register.
6236  *
6237  * For writable registers, get_x2apic_wr_val() deposits the write value into the
6238  * val pointer arg and returns true. For non-writable registers, val is not
6239  * modified and get_x2apic_wr_val() returns false.
6240  */
6241 static bool get_x2apic_wr_val(u32 reg, u64 *val)
6242 {
6243 	switch (reg) {
6244 	case APIC_TASKPRI:
6245 		/* Bits 31:8 are reserved. */
6246 		*val &= 0xff;
6247 		break;
6248 	case APIC_EOI:
6249 	case APIC_ESR:
6250 	case APIC_TMICT:
6251 		/*
6252 		 * EOI, ESR: WRMSR of a non-zero value causes #GP(0).
6253 		 * TMICT: A write of 0 to the initial-count register effectively
6254 		 *        stops the local APIC timer, in both one-shot and
6255 		 *        periodic mode.
6256 		 */
6257 		*val = 0;
6258 		break;
6259 	case APIC_SPIV:
6260 	case APIC_LVTT:
6261 	case APIC_LVTTHMR:
6262 	case APIC_LVTPC:
6263 	case APIC_LVT0:
6264 	case APIC_LVT1:
6265 	case APIC_LVTERR:
6266 	case APIC_TDCR:
6267 		/*
6268 		 * To avoid writing a 1 to a reserved bit or causing some other
6269 		 * unintended side effect, read the current value and use it as
6270 		 * the write value.
6271 		 */
6272 		*val = apic_read(reg);
6273 		break;
6274 	case APIC_CMCI:
6275 		if (!apic_lvt_entry_supported(6))
6276 			return false;
6277 		*val = apic_read(reg);
6278 		break;
6279 	case APIC_ICR:
6280 		*val = 0x40000 | 0xf1;
6281 		break;
6282 	case APIC_SELF_IPI:
6283 		/*
6284 		 * With special processing (i.e., virtualize x2APIC mode +
6285 		 * virtual interrupt delivery), writing zero causes an
6286 		 * APIC-write VM exit. We plan to add a test for enabling
6287 		 * "virtual-interrupt delivery" in VMCS12, and that's where we
6288 		 * will test a self IPI with special processing.
6289 		 */
6290 		*val = 0x0;
6291 		break;
6292 	default:
6293 		return false;
6294 	}
6295 
6296 	return true;
6297 }
6298 
6299 static bool special_processing_applies(u32 reg, u64 *val,
6300 				       bool virt_int_delivery)
6301 {
6302 	bool special_processing =
6303 		(reg == APIC_TASKPRI) ||
6304 		(virt_int_delivery &&
6305 		 (reg == APIC_EOI || reg == APIC_SELF_IPI));
6306 
6307 	if (special_processing) {
6308 		TEST_ASSERT(get_x2apic_wr_val(reg, val));
6309 		return true;
6310 	}
6311 
6312 	return false;
6313 }
6314 
6315 static void virt_x2apic_mode_wr_expectation(
6316 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6317 	bool virt_int_delivery,
6318 	struct virt_x2apic_mode_expectation *expectation)
6319 {
6320 	expectation->wr_exit_reason = VMX_VMCALL;
6321 	expectation->wr_val = MAGIC_VAL_1;
6322 	expectation->wr_only = false;
6323 
6324 	if (virt_x2apic_mode_on &&
6325 	    special_processing_applies(reg, &expectation->wr_val,
6326 				       virt_int_delivery)) {
6327 		expectation->wr_behavior = X2APIC_ACCESS_VIRTUALIZED;
6328 		if (reg == APIC_SELF_IPI)
6329 			expectation->wr_exit_reason = VMX_APIC_WRITE;
6330 	} else if (!disable_x2apic &&
6331 		   get_x2apic_wr_val(reg, &expectation->wr_val)) {
6332 		expectation->wr_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6333 		if (reg == APIC_EOI || reg == APIC_SELF_IPI)
6334 			expectation->wr_only = true;
6335 		if (reg == APIC_ICR)
6336 			expectation->wr_exit_reason = VMX_EXTINT;
6337 	} else {
6338 		expectation->wr_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6339 		/*
6340 		 * Writing 1 to a reserved bit triggers a #GP.
6341 		 * Thus, set the write value to 0, which seems
6342 		 * the most likely to detect a missed #GP.
6343 		 */
6344 		expectation->wr_val = 0;
6345 	}
6346 }
6347 
6348 static void virt_x2apic_mode_exit_expectation(
6349 	u32 reg, struct virt_x2apic_mode_config *config,
6350 	struct virt_x2apic_mode_expectation *expectation)
6351 {
6352 	struct apic_reg_virt_config *base_config =
6353 		&config->apic_reg_virt_config;
6354 	bool virt_x2apic_mode_on =
6355 		base_config->virtualize_x2apic_mode &&
6356 		config->use_msr_bitmaps &&
6357 		config->disable_x2apic_msr_intercepts &&
6358 		base_config->activate_secondary_controls;
6359 
6360 	virt_x2apic_mode_wr_expectation(
6361 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6362 		config->virtual_interrupt_delivery, expectation);
6363 	virt_x2apic_mode_rd_expectation(
6364 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6365 		base_config->apic_register_virtualization,
6366 		config->virtual_interrupt_delivery, expectation);
6367 }
6368 
6369 struct virt_x2apic_mode_test_case virt_x2apic_mode_tests[] = {
6370 	/*
6371 	 * Baseline "virtualize x2APIC mode" configuration:
6372 	 *   - virtualize x2APIC mode
6373 	 *   - virtual-interrupt delivery
6374 	 *   - APIC-register virtualization
6375 	 *   - x2APIC MSR intercepts disabled
6376 	 *
6377 	 * Reads come from virtual APIC page, special processing applies to
6378 	 * VTPR, EOI, and SELF IPI, and all other writes pass through to L1
6379 	 * APIC.
6380 	 */
6381 	{
6382 		.name = "Baseline",
6383 		.virt_x2apic_mode_config = {
6384 			.virtual_interrupt_delivery = true,
6385 			.use_msr_bitmaps = true,
6386 			.disable_x2apic_msr_intercepts = true,
6387 			.disable_x2apic = false,
6388 			.apic_reg_virt_config = {
6389 				.apic_register_virtualization = true,
6390 				.use_tpr_shadow = true,
6391 				.virtualize_apic_accesses = false,
6392 				.virtualize_x2apic_mode = true,
6393 				.activate_secondary_controls = true,
6394 			},
6395 		},
6396 	},
6397 	{
6398 		.name = "Baseline w/ x2apic disabled",
6399 		.virt_x2apic_mode_config = {
6400 			.virtual_interrupt_delivery = true,
6401 			.use_msr_bitmaps = true,
6402 			.disable_x2apic_msr_intercepts = true,
6403 			.disable_x2apic = true,
6404 			.apic_reg_virt_config = {
6405 				.apic_register_virtualization = true,
6406 				.use_tpr_shadow = true,
6407 				.virtualize_apic_accesses = false,
6408 				.virtualize_x2apic_mode = true,
6409 				.activate_secondary_controls = true,
6410 			},
6411 		},
6412 	},
6413 
6414 	/*
6415 	 * Baseline, minus virtual-interrupt delivery. Reads come from virtual
6416 	 * APIC page, special processing applies to VTPR, and all other writes
6417 	 * pass through to L1 APIC.
6418 	 */
6419 	{
6420 		.name = "Baseline - virtual interrupt delivery",
6421 		.virt_x2apic_mode_config = {
6422 			.virtual_interrupt_delivery = false,
6423 			.use_msr_bitmaps = true,
6424 			.disable_x2apic_msr_intercepts = true,
6425 			.disable_x2apic = false,
6426 			.apic_reg_virt_config = {
6427 				.apic_register_virtualization = true,
6428 				.use_tpr_shadow = true,
6429 				.virtualize_apic_accesses = false,
6430 				.virtualize_x2apic_mode = true,
6431 				.activate_secondary_controls = true,
6432 			},
6433 		},
6434 	},
6435 
6436 	/*
6437 	 * Baseline, minus APIC-register virtualization. x2APIC reads pass
6438 	 * through to L1's APIC, unless reading VTPR
6439 	 */
6440 	{
6441 		.name = "Virtualize x2APIC mode, no APIC reg virt",
6442 		.virt_x2apic_mode_config = {
6443 			.virtual_interrupt_delivery = true,
6444 			.use_msr_bitmaps = true,
6445 			.disable_x2apic_msr_intercepts = true,
6446 			.disable_x2apic = false,
6447 			.apic_reg_virt_config = {
6448 				.apic_register_virtualization = false,
6449 				.use_tpr_shadow = true,
6450 				.virtualize_apic_accesses = false,
6451 				.virtualize_x2apic_mode = true,
6452 				.activate_secondary_controls = true,
6453 			},
6454 		},
6455 	},
6456 	{
6457 		.name = "Virtualize x2APIC mode, no APIC reg virt, x2APIC off",
6458 		.virt_x2apic_mode_config = {
6459 			.virtual_interrupt_delivery = true,
6460 			.use_msr_bitmaps = true,
6461 			.disable_x2apic_msr_intercepts = true,
6462 			.disable_x2apic = true,
6463 			.apic_reg_virt_config = {
6464 				.apic_register_virtualization = false,
6465 				.use_tpr_shadow = true,
6466 				.virtualize_apic_accesses = false,
6467 				.virtualize_x2apic_mode = true,
6468 				.activate_secondary_controls = true,
6469 			},
6470 		},
6471 	},
6472 
6473 	/*
6474 	 * Enable "virtualize x2APIC mode" and "APIC-register virtualization",
6475 	 * and disable intercepts for the x2APIC MSRs, but fail to enable
6476 	 * "activate secondary controls" (i.e. L2 gets access to L1's x2APIC
6477 	 * MSRs).
6478 	 */
6479 	{
6480 		.name = "Fail to enable activate secondary controls",
6481 		.virt_x2apic_mode_config = {
6482 			.virtual_interrupt_delivery = true,
6483 			.use_msr_bitmaps = true,
6484 			.disable_x2apic_msr_intercepts = true,
6485 			.disable_x2apic = false,
6486 			.apic_reg_virt_config = {
6487 				.apic_register_virtualization = true,
6488 				.use_tpr_shadow = true,
6489 				.virtualize_apic_accesses = false,
6490 				.virtualize_x2apic_mode = true,
6491 				.activate_secondary_controls = false,
6492 			},
6493 		},
6494 	},
6495 
6496 	/*
6497 	 * Enable "APIC-register virtualization" and enable "activate secondary
6498 	 * controls" and disable intercepts for the x2APIC MSRs, but do not
6499 	 * enable the "virtualize x2APIC mode" VM-execution control (i.e. L2
6500 	 * gets access to L1's x2APIC MSRs).
6501 	 */
6502 	{
6503 		.name = "Fail to enable virtualize x2APIC mode",
6504 		.virt_x2apic_mode_config = {
6505 			.virtual_interrupt_delivery = true,
6506 			.use_msr_bitmaps = true,
6507 			.disable_x2apic_msr_intercepts = true,
6508 			.disable_x2apic = false,
6509 			.apic_reg_virt_config = {
6510 				.apic_register_virtualization = true,
6511 				.use_tpr_shadow = true,
6512 				.virtualize_apic_accesses = false,
6513 				.virtualize_x2apic_mode = false,
6514 				.activate_secondary_controls = true,
6515 			},
6516 		},
6517 	},
6518 
6519 	/*
6520 	 * Disable "Virtualize x2APIC mode", disable x2APIC MSR intercepts, and
6521 	 * enable "APIC-register virtualization" --> L2 gets L1's x2APIC MSRs.
6522 	 */
6523 	{
6524 		.name = "Baseline",
6525 		.virt_x2apic_mode_config = {
6526 			.virtual_interrupt_delivery = true,
6527 			.use_msr_bitmaps = true,
6528 			.disable_x2apic_msr_intercepts = true,
6529 			.disable_x2apic = false,
6530 			.apic_reg_virt_config = {
6531 				.apic_register_virtualization = true,
6532 				.use_tpr_shadow = true,
6533 				.virtualize_apic_accesses = false,
6534 				.virtualize_x2apic_mode = false,
6535 				.activate_secondary_controls = true,
6536 			},
6537 		},
6538 	},
6539 };
6540 
6541 enum X2apic_op {
6542 	X2APIC_OP_RD,
6543 	X2APIC_OP_WR,
6544 	X2APIC_TERMINATE,
6545 };
6546 
6547 static u64 vmx_x2apic_read(u32 reg)
6548 {
6549 	u32 msr_addr = x2apic_msr(reg);
6550 	u64 val;
6551 
6552 	val = rdmsr(msr_addr);
6553 
6554 	return val;
6555 }
6556 
6557 static void vmx_x2apic_write(u32 reg, u64 val)
6558 {
6559 	u32 msr_addr = x2apic_msr(reg);
6560 
6561 	wrmsr(msr_addr, val);
6562 }
6563 
6564 struct virt_x2apic_mode_guest_args {
6565 	enum X2apic_op op;
6566 	u32 reg;
6567 	u64 val;
6568 	bool should_gp;
6569 	u64 (*virt_fn)(u64);
6570 } virt_x2apic_mode_guest_args;
6571 
6572 static volatile bool handle_x2apic_gp_ran;
6573 static volatile u32 handle_x2apic_gp_insn_len;
6574 static void handle_x2apic_gp(struct ex_regs *regs)
6575 {
6576 	handle_x2apic_gp_ran = true;
6577 	regs->rip += handle_x2apic_gp_insn_len;
6578 }
6579 
6580 static handler setup_x2apic_gp_handler(void)
6581 {
6582 	handler old_handler;
6583 
6584 	old_handler = handle_exception(GP_VECTOR, handle_x2apic_gp);
6585 	/* RDMSR and WRMSR are both 2 bytes, assuming no prefixes. */
6586 	handle_x2apic_gp_insn_len = 2;
6587 
6588 	return old_handler;
6589 }
6590 
6591 static void teardown_x2apic_gp_handler(handler old_handler)
6592 {
6593 	handle_exception(GP_VECTOR, old_handler);
6594 
6595 	/*
6596 	 * Defensively reset instruction length, so that if the handler is
6597 	 * incorrectly used, it will loop infinitely, rather than run off into
6598 	 * la la land.
6599 	 */
6600 	handle_x2apic_gp_insn_len = 0;
6601 	handle_x2apic_gp_ran = false;
6602 }
6603 
6604 static void virt_x2apic_mode_guest(void)
6605 {
6606 	volatile struct virt_x2apic_mode_guest_args *args =
6607 		&virt_x2apic_mode_guest_args;
6608 
6609 	for (;;) {
6610 		enum X2apic_op op = args->op;
6611 		u32 reg = args->reg;
6612 		u64 val = args->val;
6613 		bool should_gp = args->should_gp;
6614 		u64 (*virt_fn)(u64) = args->virt_fn;
6615 		handler old_handler;
6616 
6617 		if (op == X2APIC_TERMINATE)
6618 			break;
6619 
6620 		if (should_gp) {
6621 			TEST_ASSERT(!handle_x2apic_gp_ran);
6622 			old_handler = setup_x2apic_gp_handler();
6623 		}
6624 
6625 		if (op == X2APIC_OP_RD) {
6626 			u64 ret = vmx_x2apic_read(reg);
6627 
6628 			if (!should_gp) {
6629 				u64 want = virt_fn(val);
6630 				u64 got = virt_fn(ret);
6631 
6632 				report(got == want,
6633 				       "APIC read; got 0x%lx, want 0x%lx.",
6634 				       got, want);
6635 			}
6636 		} else if (op == X2APIC_OP_WR) {
6637 			vmx_x2apic_write(reg, val);
6638 		}
6639 
6640 		if (should_gp) {
6641 			report(handle_x2apic_gp_ran,
6642 			       "x2APIC op triggered GP.");
6643 			teardown_x2apic_gp_handler(old_handler);
6644 		}
6645 
6646 		/*
6647 		 * The L1 should always execute a vmcall after it's done testing
6648 		 * an individual APIC operation. This helps to validate that the
6649 		 * L1 and L2 are in sync with each other, as expected.
6650 		 */
6651 		vmcall();
6652 	}
6653 }
6654 
6655 static void test_x2apic_rd(
6656 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6657 	u32 *virtual_apic_page)
6658 {
6659 	u64 val = expectation->rd_val;
6660 	u32 exit_reason_want = expectation->rd_exit_reason;
6661 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6662 
6663 	report_prefix_pushf("x2apic - reading 0x%03x", reg);
6664 
6665 	/* Configure guest to do an x2apic read */
6666 	args->op = X2APIC_OP_RD;
6667 	args->reg = reg;
6668 	args->val = val;
6669 	args->should_gp = expectation->rd_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6670 	args->virt_fn = expectation->virt_fn;
6671 
6672 	/* Setup virtual APIC page */
6673 	if (expectation->rd_behavior == X2APIC_ACCESS_VIRTUALIZED)
6674 		virtual_apic_page[apic_reg_index(reg)] = (u32)val;
6675 
6676 	/* Enter guest */
6677 	enter_guest();
6678 
6679 	if (exit_reason_want != VMX_VMCALL) {
6680 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6681 	}
6682 
6683 	skip_exit_vmcall();
6684 	report_prefix_pop();
6685 }
6686 
6687 static volatile bool handle_x2apic_ipi_ran;
6688 static void handle_x2apic_ipi(isr_regs_t *regs)
6689 {
6690 	handle_x2apic_ipi_ran = true;
6691 	eoi();
6692 }
6693 
6694 static void test_x2apic_wr(
6695 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6696 	u32 *virtual_apic_page)
6697 {
6698 	u64 val = expectation->wr_val;
6699 	u32 exit_reason_want = expectation->wr_exit_reason;
6700 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6701 	int ipi_vector = 0xf1;
6702 	u32 restore_val = 0;
6703 
6704 	report_prefix_pushf("x2apic - writing 0x%lx to 0x%03x", val, reg);
6705 
6706 	/* Configure guest to do an x2apic read */
6707 	args->op = X2APIC_OP_WR;
6708 	args->reg = reg;
6709 	args->val = val;
6710 	args->should_gp = expectation->wr_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6711 
6712 	/* Setup virtual APIC page */
6713 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED)
6714 		virtual_apic_page[apic_reg_index(reg)] = 0;
6715 	if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH && !expectation->wr_only)
6716 		restore_val = apic_read(reg);
6717 
6718 	/* Setup IPI handler */
6719 	handle_x2apic_ipi_ran = false;
6720 	handle_irq(ipi_vector, handle_x2apic_ipi);
6721 
6722 	/* Enter guest */
6723 	enter_guest();
6724 
6725 	/*
6726 	 * Validate the behavior and
6727 	 * pass a magic value back to the guest.
6728 	 */
6729 	if (exit_reason_want == VMX_EXTINT) {
6730 		assert_exit_reason(exit_reason_want);
6731 
6732 		/* Clear the external interrupt. */
6733 		sti_nop_cli();
6734 		report(handle_x2apic_ipi_ran,
6735 		       "Got pending interrupt after IRQ enabled.");
6736 
6737 		enter_guest();
6738 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6739 		assert_exit_reason(exit_reason_want);
6740 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6741 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%lx",
6742 		       apic_reg_index(reg),
6743 		       virtual_apic_page[apic_reg_index(reg)], val);
6744 
6745 		/* Reenter guest so it can consume/check rcx and exit again. */
6746 		enter_guest();
6747 	} else if (exit_reason_want != VMX_VMCALL) {
6748 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6749 	}
6750 
6751 	assert_exit_reason(VMX_VMCALL);
6752 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) {
6753 		u64 want = val;
6754 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6755 
6756 		report(got == want, "x2APIC write; got 0x%x, want 0x%lx", got,
6757 		       want);
6758 	} else if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH) {
6759 		if (!expectation->wr_only) {
6760 			u32 got = apic_read(reg);
6761 			bool ok;
6762 
6763 			/*
6764 			 * When L1's TPR is passed through to L2, the lower
6765 			 * nibble can be lost. For example, if L2 executes
6766 			 * WRMSR(0x808, 0x78), then, L1 might read 0x70.
6767 			 *
6768 			 * Here's how the lower nibble can get lost:
6769 			 *   1. L2 executes WRMSR(0x808, 0x78).
6770 			 *   2. L2 exits to L0 with a WRMSR exit.
6771 			 *   3. L0 emulates WRMSR, by writing L1's TPR.
6772 			 *   4. L0 re-enters L2.
6773 			 *   5. L2 exits to L0 (reason doesn't matter).
6774 			 *   6. L0 reflects L2's exit to L1.
6775 			 *   7. Before entering L1, L0 exits to user-space
6776 			 *      (e.g., to satisfy TPR access reporting).
6777 			 *   8. User-space executes KVM_SET_REGS ioctl, which
6778 			 *      clears the lower nibble of L1's TPR.
6779 			 */
6780 			if (reg == APIC_TASKPRI) {
6781 				got = apic_virt_nibble1(got);
6782 				val = apic_virt_nibble1(val);
6783 			}
6784 
6785 			ok = got == val;
6786 			report(ok,
6787 			       "non-virtualized write; val is 0x%x, want 0x%lx",
6788 			       got, val);
6789 			apic_write(reg, restore_val);
6790 		} else {
6791 			report_pass("non-virtualized and write-only OK");
6792 		}
6793 	}
6794 	skip_exit_insn();
6795 
6796 	report_prefix_pop();
6797 }
6798 
6799 static enum Config_type configure_virt_x2apic_mode_test(
6800 	struct virt_x2apic_mode_config *virt_x2apic_mode_config,
6801 	u8 *msr_bitmap_page)
6802 {
6803 	int msr;
6804 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6805 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6806 
6807 	/* x2apic-specific VMCS config */
6808 	if (virt_x2apic_mode_config->use_msr_bitmaps) {
6809 		/* virt_x2apic_mode_test() checks for MSR bitmaps support */
6810 		cpu_exec_ctrl0 |= CPU_MSR_BITMAP;
6811 	} else {
6812 		cpu_exec_ctrl0 &= ~CPU_MSR_BITMAP;
6813 	}
6814 
6815 	if (virt_x2apic_mode_config->virtual_interrupt_delivery) {
6816 		if (!(ctrl_cpu_rev[1].clr & CPU_VINTD)) {
6817 			report_skip("%s : \"virtual-interrupt delivery\" exec control not supported", __func__);
6818 			return CONFIG_TYPE_UNSUPPORTED;
6819 		}
6820 		cpu_exec_ctrl1 |= CPU_VINTD;
6821 	} else {
6822 		cpu_exec_ctrl1 &= ~CPU_VINTD;
6823 	}
6824 
6825 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6826 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6827 
6828 	/* x2APIC MSR intercepts are usually off for "Virtualize x2APIC mode" */
6829 	for (msr = 0x800; msr <= 0x8ff; msr++) {
6830 		if (virt_x2apic_mode_config->disable_x2apic_msr_intercepts) {
6831 			clear_bit(msr, msr_bitmap_page + 0x000);
6832 			clear_bit(msr, msr_bitmap_page + 0x800);
6833 		} else {
6834 			set_bit(msr, msr_bitmap_page + 0x000);
6835 			set_bit(msr, msr_bitmap_page + 0x800);
6836 		}
6837 	}
6838 
6839 	/* x2APIC mode can impact virtualization */
6840 	reset_apic();
6841 	if (!virt_x2apic_mode_config->disable_x2apic)
6842 		enable_x2apic();
6843 
6844 	return configure_apic_reg_virt_test(
6845 		&virt_x2apic_mode_config->apic_reg_virt_config);
6846 }
6847 
6848 static void virt_x2apic_mode_test(void)
6849 {
6850 	u32 *virtual_apic_page;
6851 	u8 *msr_bitmap_page;
6852 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6853 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6854 	int i;
6855 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6856 
6857 	if (!cpu_has_apicv()) {
6858 		report_skip("%s : Not all required APICv bits supported", __func__);
6859 		return;
6860 	}
6861 
6862 	/*
6863 	 * This is to exercise an issue in KVM's logic to merge L0's and L1's
6864 	 * MSR bitmaps. Previously, an L1 could get at L0's x2APIC MSRs by
6865 	 * writing the IA32_SPEC_CTRL MSR or the IA32_PRED_CMD MSRs. KVM would
6866 	 * then proceed to manipulate the MSR bitmaps, as if VMCS12 had the
6867 	 * "Virtualize x2APIC mod" control set, even when it didn't.
6868 	 */
6869 	if (this_cpu_has(X86_FEATURE_SPEC_CTRL))
6870 		wrmsr(MSR_IA32_SPEC_CTRL, 1);
6871 
6872 	/*
6873 	 * Check that VMCS12 supports:
6874 	 *   - "Virtual-APIC address", indicated by "use TPR shadow"
6875 	 *   - "MSR-bitmap address", indicated by "use MSR bitmaps"
6876 	 */
6877 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6878 		report_skip("%s : \"Use TPR shadow\" exec control not supported", __func__);
6879 		return;
6880 	} else if (!(ctrl_cpu_rev[0].clr & CPU_MSR_BITMAP)) {
6881 		report_skip("%s : \"Use MSR bitmaps\" exec control not supported", __func__);
6882 		return;
6883 	}
6884 
6885 	test_set_guest(virt_x2apic_mode_guest);
6886 
6887 	virtual_apic_page = alloc_page();
6888 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6889 
6890 	msr_bitmap_page = alloc_page();
6891 	memset(msr_bitmap_page, 0xff, PAGE_SIZE);
6892 	vmcs_write(MSR_BITMAP, virt_to_phys(msr_bitmap_page));
6893 
6894 	for (i = 0; i < ARRAY_SIZE(virt_x2apic_mode_tests); i++) {
6895 		struct virt_x2apic_mode_test_case *virt_x2apic_mode_test_case =
6896 			&virt_x2apic_mode_tests[i];
6897 		struct virt_x2apic_mode_config *virt_x2apic_mode_config =
6898 			&virt_x2apic_mode_test_case->virt_x2apic_mode_config;
6899 		enum Config_type config_type;
6900 		u32 reg;
6901 
6902 		printf("--- %s test ---\n", virt_x2apic_mode_test_case->name);
6903 		config_type =
6904 			configure_virt_x2apic_mode_test(virt_x2apic_mode_config,
6905 							msr_bitmap_page);
6906 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6907 			report_skip("Skip because of missing features.");
6908 			continue;
6909 		} else if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6910 			enter_guest_with_bad_controls();
6911 			continue;
6912 		}
6913 
6914 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6915 			struct virt_x2apic_mode_expectation expectation;
6916 
6917 			virt_x2apic_mode_exit_expectation(
6918 				reg, virt_x2apic_mode_config, &expectation);
6919 
6920 			test_x2apic_rd(reg, &expectation, virtual_apic_page);
6921 			test_x2apic_wr(reg, &expectation, virtual_apic_page);
6922 		}
6923 	}
6924 
6925 
6926 	/* Terminate the guest */
6927 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6928 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6929 	args->op = X2APIC_TERMINATE;
6930 	enter_guest();
6931 	assert_exit_reason(VMX_VMCALL);
6932 }
6933 
6934 static void test_ctl_reg(const char *cr_name, u64 cr, u64 fixed0, u64 fixed1)
6935 {
6936 	u64 val;
6937 	u64 cr_saved = vmcs_read(cr);
6938 	int i;
6939 
6940 	val = fixed0 & fixed1;
6941 	if (cr == HOST_CR4)
6942 		vmcs_write(cr, val | X86_CR4_PAE);
6943 	else
6944 		vmcs_write(cr, val);
6945 	report_prefix_pushf("%s %lx", cr_name, val);
6946 	if (val == fixed0)
6947 		test_vmx_vmlaunch(0);
6948 	else
6949 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6950 	report_prefix_pop();
6951 
6952 	for (i = 0; i < 64; i++) {
6953 
6954 		/* Set a bit when the corresponding bit in fixed1 is 0 */
6955 		if ((fixed1 & (1ull << i)) == 0) {
6956 			if (cr == HOST_CR4 && ((1ull << i) & X86_CR4_SMEP ||
6957 					       (1ull << i) & X86_CR4_SMAP))
6958 				continue;
6959 
6960 			vmcs_write(cr, cr_saved | (1ull << i));
6961 			report_prefix_pushf("%s %llx", cr_name,
6962 						cr_saved | (1ull << i));
6963 			test_vmx_vmlaunch(
6964 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6965 			report_prefix_pop();
6966 		}
6967 
6968 		/* Unset a bit when the corresponding bit in fixed0 is 1 */
6969 		if (fixed0 & (1ull << i)) {
6970 			vmcs_write(cr, cr_saved & ~(1ull << i));
6971 			report_prefix_pushf("%s %llx", cr_name,
6972 						cr_saved & ~(1ull << i));
6973 			test_vmx_vmlaunch(
6974 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6975 			report_prefix_pop();
6976 		}
6977 	}
6978 
6979 	vmcs_write(cr, cr_saved);
6980 }
6981 
6982 /*
6983  * 1. The CR0 field must not set any bit to a value not supported in VMX
6984  *    operation.
6985  * 2. The CR4 field must not set any bit to a value not supported in VMX
6986  *    operation.
6987  * 3. On processors that support Intel 64 architecture, the CR3 field must
6988  *    be such that bits 63:52 and bits in the range 51:32 beyond the
6989  *    processor's physical-address width must be 0.
6990  *
6991  *  [Intel SDM]
6992  */
6993 static void test_host_ctl_regs(void)
6994 {
6995 	u64 fixed0, fixed1, cr3, cr3_saved;
6996 	int i;
6997 
6998 	/* Test CR0 */
6999 	fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0);
7000 	fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1);
7001 	test_ctl_reg("HOST_CR0", HOST_CR0, fixed0, fixed1);
7002 
7003 	/* Test CR4 */
7004 	fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0);
7005 	fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1) &
7006 		 ~(X86_CR4_SMEP | X86_CR4_SMAP);
7007 	test_ctl_reg("HOST_CR4", HOST_CR4, fixed0, fixed1);
7008 
7009 	/* Test CR3 */
7010 	cr3_saved = vmcs_read(HOST_CR3);
7011 	for (i = cpuid_maxphyaddr(); i < 64; i++) {
7012 		cr3 = cr3_saved | (1ul << i);
7013 		vmcs_write(HOST_CR3, cr3);
7014 		report_prefix_pushf("HOST_CR3 %lx", cr3);
7015 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7016 		report_prefix_pop();
7017 	}
7018 
7019 	vmcs_write(HOST_CR3, cr3_saved);
7020 }
7021 
7022 static void test_efer_vmlaunch(u32 fld, bool ok)
7023 {
7024 	if (fld == HOST_EFER) {
7025 		if (ok)
7026 			test_vmx_vmlaunch(0);
7027 		else
7028 			test_vmx_vmlaunch2(VMXERR_ENTRY_INVALID_CONTROL_FIELD,
7029 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7030 	} else {
7031 		test_guest_state("EFER test", !ok, GUEST_EFER, "GUEST_EFER");
7032 	}
7033 }
7034 
7035 static void test_efer_one(u32 fld, const char * fld_name, u64 efer,
7036 			  u32 ctrl_fld, u64 ctrl,
7037 			  int i, const char *efer_bit_name)
7038 {
7039 	bool ok;
7040 
7041 	ok = true;
7042 	if (ctrl_fld == EXI_CONTROLS && (ctrl & EXI_LOAD_EFER)) {
7043 		if (!!(efer & EFER_LMA) != !!(ctrl & EXI_HOST_64))
7044 			ok = false;
7045 		if (!!(efer & EFER_LME) != !!(ctrl & EXI_HOST_64))
7046 			ok = false;
7047 	}
7048 	if (ctrl_fld == ENT_CONTROLS && (ctrl & ENT_LOAD_EFER)) {
7049 		/* Check LMA too since CR0.PG is set.  */
7050 		if (!!(efer & EFER_LMA) != !!(ctrl & ENT_GUEST_64))
7051 			ok = false;
7052 		if (!!(efer & EFER_LME) != !!(ctrl & ENT_GUEST_64))
7053 			ok = false;
7054 	}
7055 
7056 	/*
7057 	 * Skip the test if it would enter the guest in 32-bit mode.
7058 	 * Perhaps write the test in assembly and make sure it
7059 	 * can be run in either mode?
7060 	 */
7061 	if (fld == GUEST_EFER && ok && !(ctrl & ENT_GUEST_64))
7062 		return;
7063 
7064 	vmcs_write(ctrl_fld, ctrl);
7065 	vmcs_write(fld, efer);
7066 	report_prefix_pushf("%s %s bit turned %s, controls %s",
7067 			    fld_name, efer_bit_name,
7068 			    (i & 1) ? "on" : "off",
7069 			    (i & 2) ? "on" : "off");
7070 
7071 	test_efer_vmlaunch(fld, ok);
7072 	report_prefix_pop();
7073 }
7074 
7075 static void test_efer_bit(u32 fld, const char * fld_name,
7076 			  u32 ctrl_fld, u64 ctrl_bit, u64 efer_bit,
7077 			  const char *efer_bit_name)
7078 {
7079 	u64 efer_saved = vmcs_read(fld);
7080 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7081 	int i;
7082 
7083 	for (i = 0; i < 4; i++) {
7084 		u64 efer = efer_saved & ~efer_bit;
7085 		u64 ctrl = ctrl_saved & ~ctrl_bit;
7086 
7087 		if (i & 1)
7088 			efer |= efer_bit;
7089 		if (i & 2)
7090 			ctrl |= ctrl_bit;
7091 
7092 		test_efer_one(fld, fld_name, efer, ctrl_fld, ctrl,
7093 			      i, efer_bit_name);
7094 	}
7095 
7096 	vmcs_write(ctrl_fld, ctrl_saved);
7097 	vmcs_write(fld, efer_saved);
7098 }
7099 
7100 static void test_efer(u32 fld, const char * fld_name, u32 ctrl_fld,
7101 		      u64 ctrl_bit1, u64 ctrl_bit2)
7102 {
7103 	u64 efer_saved = vmcs_read(fld);
7104 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7105 	u64 efer_reserved_bits =  ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
7106 	u64 i;
7107 	u64 efer;
7108 
7109 	if (this_cpu_has(X86_FEATURE_NX))
7110 		efer_reserved_bits &= ~EFER_NX;
7111 
7112 	if (!ctrl_bit1) {
7113 		report_skip("%s : \"Load-IA32-EFER\" exit control not supported", __func__);
7114 		goto test_entry_exit_mode;
7115 	}
7116 
7117 	report_prefix_pushf("%s %lx", fld_name, efer_saved);
7118 	test_efer_vmlaunch(fld, true);
7119 	report_prefix_pop();
7120 
7121 	/*
7122 	 * Check reserved bits
7123 	 */
7124 	vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit1);
7125 	for (i = 0; i < 64; i++) {
7126 		if ((1ull << i) & efer_reserved_bits) {
7127 			efer = efer_saved | (1ull << i);
7128 			vmcs_write(fld, efer);
7129 			report_prefix_pushf("%s %lx", fld_name, efer);
7130 			test_efer_vmlaunch(fld, true);
7131 			report_prefix_pop();
7132 		}
7133 	}
7134 
7135 	vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit1);
7136 	for (i = 0; i < 64; i++) {
7137 		if ((1ull << i) & efer_reserved_bits) {
7138 			efer = efer_saved | (1ull << i);
7139 			vmcs_write(fld, efer);
7140 			report_prefix_pushf("%s %lx", fld_name, efer);
7141 			test_efer_vmlaunch(fld, false);
7142 			report_prefix_pop();
7143 		}
7144 	}
7145 
7146 	vmcs_write(ctrl_fld, ctrl_saved);
7147 	vmcs_write(fld, efer_saved);
7148 
7149 	/*
7150 	 * Check LMA and LME bits
7151 	 */
7152 	test_efer_bit(fld, fld_name,
7153 		      ctrl_fld, ctrl_bit1,
7154 		      EFER_LMA,
7155 		      "EFER_LMA");
7156 	test_efer_bit(fld, fld_name,
7157 		      ctrl_fld, ctrl_bit1,
7158 		      EFER_LME,
7159 		      "EFER_LME");
7160 
7161 test_entry_exit_mode:
7162 	test_efer_bit(fld, fld_name,
7163 		      ctrl_fld, ctrl_bit2,
7164 		      EFER_LMA,
7165 		      "EFER_LMA");
7166 	test_efer_bit(fld, fld_name,
7167 		      ctrl_fld, ctrl_bit2,
7168 		      EFER_LME,
7169 		      "EFER_LME");
7170 }
7171 
7172 /*
7173  * If the 'load IA32_EFER' VM-exit control is 1, bits reserved in the
7174  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7175  * the values of the LMA and LME bits in the field must each be that of
7176  * the 'host address-space size' VM-exit control.
7177  *
7178  *  [Intel SDM]
7179  */
7180 static void test_host_efer(void)
7181 {
7182 	test_efer(HOST_EFER, "HOST_EFER", EXI_CONTROLS,
7183 		  ctrl_exit_rev.clr & EXI_LOAD_EFER,
7184 		  EXI_HOST_64);
7185 }
7186 
7187 /*
7188  * If the 'load IA32_EFER' VM-enter control is 1, bits reserved in the
7189  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7190  * the values of the LMA and LME bits in the field must each be that of
7191  * the 'IA32e-mode guest' VM-exit control.
7192  */
7193 static void test_guest_efer(void)
7194 {
7195 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER)) {
7196 		report_skip("%s : \"Load-IA32-EFER\" entry control not supported", __func__);
7197 		return;
7198 	}
7199 
7200 	vmcs_write(GUEST_EFER, rdmsr(MSR_EFER));
7201 	test_efer(GUEST_EFER, "GUEST_EFER", ENT_CONTROLS,
7202 		  ctrl_enter_rev.clr & ENT_LOAD_EFER,
7203 		  ENT_GUEST_64);
7204 }
7205 
7206 /*
7207  * PAT values higher than 8 are uninteresting since they're likely lumped
7208  * in with "8". We only test values above 8 one bit at a time,
7209  * in order to reduce the number of VM-Entries and keep the runtime reasonable.
7210  */
7211 #define	PAT_VAL_LIMIT	8
7212 
7213 static void test_pat(u32 field, const char * field_name, u32 ctrl_field,
7214 		     u64 ctrl_bit)
7215 {
7216 	u32 ctrl_saved = vmcs_read(ctrl_field);
7217 	u64 pat_saved = vmcs_read(field);
7218 	u64 i, val;
7219 	u32 j;
7220 	int error;
7221 
7222 	vmcs_clear_bits(ctrl_field, ctrl_bit);
7223 
7224 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7225 		/* Test PAT0..PAT7 fields */
7226 		for (j = 0; j < (i ? 8 : 1); j++) {
7227 			val = i << j * 8;
7228 			vmcs_write(field, val);
7229 			if (field == HOST_PAT) {
7230 				report_prefix_pushf("%s %lx", field_name, val);
7231 				test_vmx_vmlaunch(0);
7232 				report_prefix_pop();
7233 
7234 			} else {	// GUEST_PAT
7235 				test_guest_state("ENT_LOAD_PAT enabled", false,
7236 						 val, "GUEST_PAT");
7237 			}
7238 		}
7239 	}
7240 
7241 	vmcs_set_bits(ctrl_field, ctrl_bit);
7242 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7243 		/* Test PAT0..PAT7 fields */
7244 		for (j = 0; j < (i ? 8 : 1); j++) {
7245 			val = i << j * 8;
7246 			vmcs_write(field, val);
7247 
7248 			if (field == HOST_PAT) {
7249 				report_prefix_pushf("%s %lx", field_name, val);
7250 				if (i == 0x2 || i == 0x3 || i >= 0x8)
7251 					error =
7252 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
7253 				else
7254 					error = 0;
7255 
7256 				test_vmx_vmlaunch(error);
7257 				report_prefix_pop();
7258 
7259 			} else {	// GUEST_PAT
7260 				error = (i == 0x2 || i == 0x3 || i >= 0x8);
7261 				test_guest_state("ENT_LOAD_PAT enabled", !!error,
7262 						 val, "GUEST_PAT");
7263 			}
7264 
7265 		}
7266 	}
7267 
7268 	vmcs_write(ctrl_field, ctrl_saved);
7269 	vmcs_write(field, pat_saved);
7270 }
7271 
7272 /*
7273  *  If the "load IA32_PAT" VM-exit control is 1, the value of the field
7274  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7275  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7276  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7277  *  6 (WB), or 7 (UC-).
7278  *
7279  *  [Intel SDM]
7280  */
7281 static void test_load_host_pat(void)
7282 {
7283 	/*
7284 	 * "load IA32_PAT" VM-exit control
7285 	 */
7286 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) {
7287 		report_skip("%s : \"Load-IA32-PAT\" exit control not supported", __func__);
7288 		return;
7289 	}
7290 
7291 	test_pat(HOST_PAT, "HOST_PAT", EXI_CONTROLS, EXI_LOAD_PAT);
7292 }
7293 
7294 union cpuidA_eax {
7295 	struct {
7296 		unsigned int version_id:8;
7297 		unsigned int num_counters_gp:8;
7298 		unsigned int bit_width:8;
7299 		unsigned int mask_length:8;
7300 	} split;
7301 	unsigned int full;
7302 };
7303 
7304 union cpuidA_edx {
7305 	struct {
7306 		unsigned int num_counters_fixed:5;
7307 		unsigned int bit_width_fixed:8;
7308 		unsigned int reserved:9;
7309 	} split;
7310 	unsigned int full;
7311 };
7312 
7313 static bool valid_pgc(u64 val)
7314 {
7315 	struct cpuid id;
7316 	union cpuidA_eax eax;
7317 	union cpuidA_edx edx;
7318 	u64 mask;
7319 
7320 	id = cpuid(0xA);
7321 	eax.full = id.a;
7322 	edx.full = id.d;
7323 	mask = ~(((1ull << eax.split.num_counters_gp) - 1) |
7324 		 (((1ull << edx.split.num_counters_fixed) - 1) << 32));
7325 
7326 	return !(val & mask);
7327 }
7328 
7329 static void test_pgc_vmlaunch(u32 xerror, u32 xreason, bool xfail, bool host)
7330 {
7331 	u32 inst_err;
7332 	u64 obs;
7333 	bool success;
7334 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7335 
7336 	if (host) {
7337 		success = vmlaunch();
7338 		obs = rdmsr(data->msr);
7339 		if (!success) {
7340 			inst_err = vmcs_read(VMX_INST_ERROR);
7341 			report(xerror == inst_err, "vmlaunch failed, "
7342 			       "VMX Inst Error is %d (expected %d)",
7343 			       inst_err, xerror);
7344 		} else {
7345 			report(!data->enabled || data->exp == obs,
7346 			       "Host state is 0x%lx (expected 0x%lx)",
7347 			       obs, data->exp);
7348 			report(success != xfail, "vmlaunch succeeded");
7349 		}
7350 	} else {
7351 		test_guest_state("load GUEST_PERF_GLOBAL_CTRL", xfail,
7352 				 GUEST_PERF_GLOBAL_CTRL,
7353 				 "GUEST_PERF_GLOBAL_CTRL");
7354 	}
7355 }
7356 
7357 /*
7358  * test_load_perf_global_ctrl is a generic function for testing the
7359  * "load IA32_PERF_GLOBAL_CTRL" VM-{Entry,Exit} controls. This test function
7360  * tests the provided ctrl_val when disabled and enabled.
7361  *
7362  * @nr: VMCS field number corresponding to the host/guest state field
7363  * @name: Name of the above VMCS field for printing in test report
7364  * @ctrl_nr: VMCS field number corresponding to the VM-{Entry,Exit} control
7365  * @ctrl_val: Bit to set on the ctrl_field
7366  */
7367 static void test_perf_global_ctrl(u32 nr, const char *name, u32 ctrl_nr,
7368 				  const char *ctrl_name, u64 ctrl_val)
7369 {
7370 	u64 ctrl_saved = vmcs_read(ctrl_nr);
7371 	u64 pgc_saved = vmcs_read(nr);
7372 	u64 i, val;
7373 	bool host = nr == HOST_PERF_GLOBAL_CTRL;
7374 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7375 
7376 	data->msr = MSR_CORE_PERF_GLOBAL_CTRL;
7377 	msr_bmp_init();
7378 	vmcs_write(ctrl_nr, ctrl_saved & ~ctrl_val);
7379 	data->enabled = false;
7380 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=0 on %s",
7381 			    ctrl_name);
7382 
7383 	for (i = 0; i < 64; i++) {
7384 		val = 1ull << i;
7385 		vmcs_write(nr, val);
7386 		report_prefix_pushf("%s = 0x%lx", name, val);
7387 		test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7388 		report_prefix_pop();
7389 	}
7390 	report_prefix_pop();
7391 
7392 	vmcs_write(ctrl_nr, ctrl_saved | ctrl_val);
7393 	data->enabled = true;
7394 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=1 on %s",
7395 			    ctrl_name);
7396 	for (i = 0; i < 64; i++) {
7397 		val = 1ull << i;
7398 		data->exp = val;
7399 		vmcs_write(nr, val);
7400 		report_prefix_pushf("%s = 0x%lx", name, val);
7401 		if (valid_pgc(val)) {
7402 			test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7403 		} else {
7404 			if (host)
7405 				test_pgc_vmlaunch(
7406 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
7407 					0,
7408 					true,
7409 					host);
7410 			else
7411 				test_pgc_vmlaunch(
7412 					0,
7413 					VMX_ENTRY_FAILURE | VMX_FAIL_STATE,
7414 					true,
7415 					host);
7416 		}
7417 		report_prefix_pop();
7418 	}
7419 
7420 	data->enabled = false;
7421 	report_prefix_pop();
7422 	vmcs_write(ctrl_nr, ctrl_saved);
7423 	vmcs_write(nr, pgc_saved);
7424 }
7425 
7426 static void test_load_host_perf_global_ctrl(void)
7427 {
7428 	if (!this_cpu_has_perf_global_ctrl()) {
7429 		report_skip("%s : \"IA32_PERF_GLOBAL_CTRL\" MSR not supported", __func__);
7430 		return;
7431 	}
7432 
7433 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PERF)) {
7434 		report_skip("%s : \"Load IA32_PERF_GLOBAL_CTRL\" exit control not supported", __func__);
7435 		return;
7436 	}
7437 
7438 	test_perf_global_ctrl(HOST_PERF_GLOBAL_CTRL, "HOST_PERF_GLOBAL_CTRL",
7439 				   EXI_CONTROLS, "EXI_CONTROLS", EXI_LOAD_PERF);
7440 }
7441 
7442 
7443 static void test_load_guest_perf_global_ctrl(void)
7444 {
7445 	if (!this_cpu_has_perf_global_ctrl()) {
7446 		report_skip("%s : \"IA32_PERF_GLOBAL_CTRL\" MSR not supported", __func__);
7447 		return;
7448 	}
7449 
7450 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PERF)) {
7451 		report_skip("%s : \"Load IA32_PERF_GLOBAL_CTRL\" entry control not supported", __func__);
7452 		return;
7453 	}
7454 
7455 	test_perf_global_ctrl(GUEST_PERF_GLOBAL_CTRL, "GUEST_PERF_GLOBAL_CTRL",
7456 				   ENT_CONTROLS, "ENT_CONTROLS", ENT_LOAD_PERF);
7457 }
7458 
7459 
7460 /*
7461  * test_vmcs_field - test a value for the given VMCS field
7462  * @field: VMCS field
7463  * @field_name: string name of VMCS field
7464  * @bit_start: starting bit
7465  * @bit_end: ending bit
7466  * @val: value that the bit range must or must not contain
7467  * @valid_val: whether value given in 'val' must be valid or not
7468  * @error: expected VMCS error when vmentry fails for an invalid value
7469  */
7470 static void test_vmcs_field(u64 field, const char *field_name, u32 bit_start,
7471 			    u32 bit_end, u64 val, bool valid_val, u32 error)
7472 {
7473 	u64 field_saved = vmcs_read(field);
7474 	u32 i;
7475 	u64 tmp;
7476 	u32 bit_on;
7477 	u64 mask = ~0ull;
7478 
7479 	mask = (mask >> bit_end) << bit_end;
7480 	mask = mask | ((1 << bit_start) - 1);
7481 	tmp = (field_saved & mask) | (val << bit_start);
7482 
7483 	vmcs_write(field, tmp);
7484 	report_prefix_pushf("%s %lx", field_name, tmp);
7485 	if (valid_val)
7486 		test_vmx_vmlaunch(0);
7487 	else
7488 		test_vmx_vmlaunch(error);
7489 	report_prefix_pop();
7490 
7491 	for (i = bit_start; i <= bit_end; i = i + 2) {
7492 		bit_on = ((1ull < i) & (val << bit_start)) ? 0 : 1;
7493 		if (bit_on)
7494 			tmp = field_saved | (1ull << i);
7495 		else
7496 			tmp = field_saved & ~(1ull << i);
7497 		vmcs_write(field, tmp);
7498 		report_prefix_pushf("%s %lx", field_name, tmp);
7499 		if (valid_val)
7500 			test_vmx_vmlaunch(error);
7501 		else
7502 			test_vmx_vmlaunch(0);
7503 		report_prefix_pop();
7504 	}
7505 
7506 	vmcs_write(field, field_saved);
7507 }
7508 
7509 static void test_canonical(u64 field, const char * field_name, bool host)
7510 {
7511 	u64 addr_saved = vmcs_read(field);
7512 
7513 	/*
7514 	 * Use the existing value if possible.  Writing a random canonical
7515 	 * value is not an option as doing so would corrupt the field being
7516 	 * tested and likely hose the test.
7517 	 */
7518 	if (is_canonical(addr_saved)) {
7519 		if (host) {
7520 			report_prefix_pushf("%s %lx", field_name, addr_saved);
7521 			test_vmx_vmlaunch(0);
7522 			report_prefix_pop();
7523 		} else {
7524 			test_guest_state("Test canonical address", false,
7525 					 addr_saved, field_name);
7526 		}
7527 	}
7528 
7529 	vmcs_write(field, NONCANONICAL);
7530 
7531 	if (host) {
7532 		report_prefix_pushf("%s %llx", field_name, NONCANONICAL);
7533 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7534 		report_prefix_pop();
7535 	} else {
7536 		test_guest_state("Test non-canonical address", true,
7537 				 NONCANONICAL, field_name);
7538 	}
7539 
7540 	vmcs_write(field, addr_saved);
7541 }
7542 
7543 #define TEST_RPL_TI_FLAGS(reg, name)				\
7544 	test_vmcs_field(reg, name, 0, 2, 0x0, true,		\
7545 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7546 
7547 #define TEST_CS_TR_FLAGS(reg, name)				\
7548 	test_vmcs_field(reg, name, 3, 15, 0x0000, false,	\
7549 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7550 
7551 /*
7552  * 1. In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the
7553  *    RPL (bits 1:0) and the TI flag (bit 2) must be 0.
7554  * 2. The selector fields for CS and TR cannot be 0000H.
7555  * 3. The selector field for SS cannot be 0000H if the "host address-space
7556  *    size" VM-exit control is 0.
7557  * 4. On processors that support Intel 64 architecture, the base-address
7558  *    fields for FS, GS and TR must contain canonical addresses.
7559  */
7560 static void test_host_segment_regs(void)
7561 {
7562 	u16 selector_saved;
7563 
7564 	/*
7565 	 * Test RPL and TI flags
7566 	 */
7567 	TEST_RPL_TI_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7568 	TEST_RPL_TI_FLAGS(HOST_SEL_SS, "HOST_SEL_SS");
7569 	TEST_RPL_TI_FLAGS(HOST_SEL_DS, "HOST_SEL_DS");
7570 	TEST_RPL_TI_FLAGS(HOST_SEL_ES, "HOST_SEL_ES");
7571 	TEST_RPL_TI_FLAGS(HOST_SEL_FS, "HOST_SEL_FS");
7572 	TEST_RPL_TI_FLAGS(HOST_SEL_GS, "HOST_SEL_GS");
7573 	TEST_RPL_TI_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7574 
7575 	/*
7576 	 * Test that CS and TR fields can not be 0x0000
7577 	 */
7578 	TEST_CS_TR_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7579 	TEST_CS_TR_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7580 
7581 	/*
7582 	 * SS field can not be 0x0000 if "host address-space size" VM-exit
7583 	 * control is 0
7584 	 */
7585 	selector_saved = vmcs_read(HOST_SEL_SS);
7586 	vmcs_write(HOST_SEL_SS, 0);
7587 	report_prefix_pushf("HOST_SEL_SS 0");
7588 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7589 		test_vmx_vmlaunch(0);
7590 	} else {
7591 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7592 	}
7593 	report_prefix_pop();
7594 
7595 	vmcs_write(HOST_SEL_SS, selector_saved);
7596 
7597 	/*
7598 	 * Base address for FS, GS and TR must be canonical
7599 	 */
7600 	test_canonical(HOST_BASE_FS, "HOST_BASE_FS", true);
7601 	test_canonical(HOST_BASE_GS, "HOST_BASE_GS", true);
7602 	test_canonical(HOST_BASE_TR, "HOST_BASE_TR", true);
7603 }
7604 
7605 /*
7606  *  On processors that support Intel 64 architecture, the base-address
7607  *  fields for GDTR and IDTR must contain canonical addresses.
7608  */
7609 static void test_host_desc_tables(void)
7610 {
7611 	test_canonical(HOST_BASE_GDTR, "HOST_BASE_GDTR", true);
7612 	test_canonical(HOST_BASE_IDTR, "HOST_BASE_IDTR", true);
7613 }
7614 
7615 /*
7616  * If the "host address-space size" VM-exit control is 0, the following must
7617  * hold:
7618  *    - The "IA-32e mode guest" VM-entry control is 0.
7619  *    - Bit 17 of the CR4 field (corresponding to CR4.PCIDE) is 0.
7620  *    - Bits 63:32 in the RIP field are 0.
7621  *
7622  * If the "host address-space size" VM-exit control is 1, the following must
7623  * hold:
7624  *    - Bit 5 of the CR4 field (corresponding to CR4.PAE) is 1.
7625  *    - The RIP field contains a canonical address.
7626  *
7627  */
7628 static void test_host_addr_size(void)
7629 {
7630 	u64 cr4_saved = vmcs_read(HOST_CR4);
7631 	u64 rip_saved = vmcs_read(HOST_RIP);
7632 	u64 entry_ctrl_saved = vmcs_read(ENT_CONTROLS);
7633 
7634 	assert(vmcs_read(EXI_CONTROLS) & EXI_HOST_64);
7635 	assert(cr4_saved & X86_CR4_PAE);
7636 
7637 	vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7638 	report_prefix_pushf("\"IA-32e mode guest\" enabled");
7639 	test_vmx_vmlaunch(0);
7640 	report_prefix_pop();
7641 
7642 	if (this_cpu_has(X86_FEATURE_PCID)) {
7643 		vmcs_write(HOST_CR4, cr4_saved | X86_CR4_PCIDE);
7644 		report_prefix_pushf("\"CR4.PCIDE\" set");
7645 		test_vmx_vmlaunch(0);
7646 		report_prefix_pop();
7647 	}
7648 
7649 	vmcs_write(HOST_CR4, cr4_saved  & ~X86_CR4_PAE);
7650 	report_prefix_pushf("\"CR4.PAE\" unset");
7651 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7652 	vmcs_write(HOST_CR4, cr4_saved);
7653 	report_prefix_pop();
7654 
7655 	vmcs_write(HOST_RIP, NONCANONICAL);
7656 	report_prefix_pushf("HOST_RIP %llx", NONCANONICAL);
7657 	test_vmx_vmlaunch_must_fail(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7658 	report_prefix_pop();
7659 
7660 	vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7661 	vmcs_write(HOST_RIP, rip_saved);
7662 	vmcs_write(HOST_CR4, cr4_saved);
7663 
7664 	/*
7665 	 * Restore host's active CR4 and RIP values by triggering a VM-Exit.
7666 	 * The original CR4 and RIP values in the VMCS are restored between
7667 	 * testcases as needed, but don't guarantee a VM-Exit and so the active
7668 	 * CR4 and RIP may still hold a test value.  Running with the test CR4
7669 	 * and RIP values at some point is unavoidable, and the active values
7670 	 * are unlikely to affect VM-Enter, so the above doesn't force a VM-exit
7671 	 * between testcases.  Note, if VM-Enter is surrounded by CALL+RET then
7672 	 * the active RIP will already be restored, but that's also not
7673 	 * guaranteed, and CR4 needs to be restored regardless.
7674 	 */
7675 	report_prefix_pushf("restore host state");
7676 	test_vmx_vmlaunch(0);
7677 	report_prefix_pop();
7678 }
7679 
7680 /*
7681  * Check that the virtual CPU checks the VMX Host State Area as
7682  * documented in the Intel SDM.
7683  */
7684 static void vmx_host_state_area_test(void)
7685 {
7686 	/*
7687 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
7688 	 * fail due to invalid guest state, should we make it that
7689 	 * far.
7690 	 */
7691 	vmcs_write(GUEST_RFLAGS, 0);
7692 
7693 	test_host_ctl_regs();
7694 
7695 	test_canonical(HOST_SYSENTER_ESP, "HOST_SYSENTER_ESP", true);
7696 	test_canonical(HOST_SYSENTER_EIP, "HOST_SYSENTER_EIP", true);
7697 
7698 	test_host_efer();
7699 	test_load_host_pat();
7700 	test_host_segment_regs();
7701 	test_host_desc_tables();
7702 	test_host_addr_size();
7703 	test_load_host_perf_global_ctrl();
7704 }
7705 
7706 /*
7707  * If the "load debug controls" VM-entry control is 1, bits 63:32 in
7708  * the DR7 field must be 0.
7709  *
7710  * [Intel SDM]
7711  */
7712 static void test_guest_dr7(void)
7713 {
7714 	u32 ent_saved = vmcs_read(ENT_CONTROLS);
7715 	u64 dr7_saved = vmcs_read(GUEST_DR7);
7716 	u64 val;
7717 	int i;
7718 
7719 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS) {
7720 		vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7721 		for (i = 0; i < 64; i++) {
7722 			val = 1ull << i;
7723 			vmcs_write(GUEST_DR7, val);
7724 			test_guest_state("ENT_LOAD_DBGCTLS disabled", false,
7725 					 val, "GUEST_DR7");
7726 		}
7727 	}
7728 	if (ctrl_enter_rev.clr & ENT_LOAD_DBGCTLS) {
7729 		vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7730 		for (i = 0; i < 64; i++) {
7731 			val = 1ull << i;
7732 			vmcs_write(GUEST_DR7, val);
7733 			test_guest_state("ENT_LOAD_DBGCTLS enabled", i >= 32,
7734 					 val, "GUEST_DR7");
7735 		}
7736 	}
7737 	vmcs_write(GUEST_DR7, dr7_saved);
7738 	vmcs_write(ENT_CONTROLS, ent_saved);
7739 }
7740 
7741 /*
7742  *  If the "load IA32_PAT" VM-entry control is 1, the value of the field
7743  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7744  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7745  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7746  *  6 (WB), or 7 (UC-).
7747  *
7748  *  [Intel SDM]
7749  */
7750 static void test_load_guest_pat(void)
7751 {
7752 	/*
7753 	 * "load IA32_PAT" VM-entry control
7754 	 */
7755 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
7756 		report_skip("%s : \"Load-IA32-PAT\" entry control not supported", __func__);
7757 		return;
7758 	}
7759 
7760 	test_pat(GUEST_PAT, "GUEST_PAT", ENT_CONTROLS, ENT_LOAD_PAT);
7761 }
7762 
7763 #define MSR_IA32_BNDCFGS_RSVD_MASK	0x00000ffc
7764 
7765 /*
7766  * If the "load IA32_BNDCFGS" VM-entry control is 1, the following
7767  * checks are performed on the field for the IA32_BNDCFGS MSR:
7768  *
7769  *   - Bits reserved in the IA32_BNDCFGS MSR must be 0.
7770  *   - The linear address in bits 63:12 must be canonical.
7771  *
7772  *  [Intel SDM]
7773  */
7774 static void test_load_guest_bndcfgs(void)
7775 {
7776 	u64 bndcfgs_saved = vmcs_read(GUEST_BNDCFGS);
7777 	u64 bndcfgs;
7778 
7779 	if (!(ctrl_enter_rev.clr & ENT_LOAD_BNDCFGS)) {
7780 		report_skip("%s : \"Load-IA32-BNDCFGS\" entry control not supported", __func__);
7781 		return;
7782 	}
7783 
7784 	vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7785 
7786 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7787 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7788 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7789 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7790 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7791 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7792 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7793 
7794 	vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7795 
7796 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7797 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7798 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7799 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7800 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7801 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7802 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7803 
7804 	vmcs_write(GUEST_BNDCFGS, bndcfgs_saved);
7805 }
7806 
7807 #define	GUEST_SEG_UNUSABLE_MASK	(1u << 16)
7808 #define	GUEST_SEG_SEL_TI_MASK	(1u << 2)
7809 
7810 
7811 #define	TEST_SEGMENT_SEL(test, xfail, sel, val)				\
7812 do {									\
7813 	vmcs_write(sel, val);						\
7814 	test_guest_state(test " segment", xfail, val, xstr(sel));	\
7815 } while (0)
7816 
7817 #define	TEST_INVALID_SEG_SEL(sel, val) \
7818 	TEST_SEGMENT_SEL("Invalid: " xstr(val), true, sel, val);
7819 
7820 #define	TEST_VALID_SEG_SEL(sel, val) \
7821 	TEST_SEGMENT_SEL("Valid: " xstr(val), false, sel, val);
7822 
7823 /*
7824  * The following checks are done on the Selector field of the Guest Segment
7825  * Registers:
7826  *    - TR. The TI flag (bit 2) must be 0.
7827  *    - LDTR. If LDTR is usable, the TI flag (bit 2) must be 0.
7828  *    - SS. If the guest will not be virtual-8086 and the "unrestricted
7829  *	guest" VM-execution control is 0, the RPL (bits 1:0) must equal
7830  *	the RPL of the selector field for CS.
7831  *
7832  *  [Intel SDM]
7833  */
7834 static void test_guest_segment_sel_fields(void)
7835 {
7836 	u16 sel_saved;
7837 	u32 ar_saved;
7838 	u32 cpu_ctrl0_saved;
7839 	u32 cpu_ctrl1_saved;
7840 	u16 cs_rpl_bits;
7841 
7842 	/*
7843 	 * Test for GUEST_SEL_TR
7844 	 */
7845 	sel_saved = vmcs_read(GUEST_SEL_TR);
7846 	TEST_INVALID_SEG_SEL(GUEST_SEL_TR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7847 	vmcs_write(GUEST_SEL_TR, sel_saved);
7848 
7849 	/*
7850 	 * Test for GUEST_SEL_LDTR
7851 	 */
7852 	sel_saved = vmcs_read(GUEST_SEL_LDTR);
7853 	ar_saved = vmcs_read(GUEST_AR_LDTR);
7854 	/* LDTR is set unusable */
7855 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7856 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7857 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7858 	/* LDTR is set usable */
7859 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7860 	TEST_INVALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7861 
7862 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7863 
7864 	vmcs_write(GUEST_AR_LDTR, ar_saved);
7865 	vmcs_write(GUEST_SEL_LDTR, sel_saved);
7866 
7867 	/*
7868 	 * Test for GUEST_SEL_SS
7869 	 */
7870 	cpu_ctrl0_saved = vmcs_read(CPU_EXEC_CTRL0);
7871 	cpu_ctrl1_saved = vmcs_read(CPU_EXEC_CTRL1);
7872 	ar_saved = vmcs_read(GUEST_AR_SS);
7873 	/* Turn off "unrestricted guest" vm-execution control */
7874 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved & ~CPU_URG);
7875 	cs_rpl_bits = vmcs_read(GUEST_SEL_CS) & 0x3;
7876 	sel_saved = vmcs_read(GUEST_SEL_SS);
7877 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7878 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7879 	/* Make SS usable if it's unusable or vice-versa */
7880 	if (ar_saved & GUEST_SEG_UNUSABLE_MASK)
7881 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7882 	else
7883 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7884 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7885 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7886 
7887 	/* Need a valid EPTP as the passing case fully enters the guest. */
7888 	if (enable_unrestricted_guest(true))
7889 		goto skip_ss_tests;
7890 
7891 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7892 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7893 
7894 	/* Make SS usable if it's unusable or vice-versa */
7895 	if (vmcs_read(GUEST_AR_SS) & GUEST_SEG_UNUSABLE_MASK)
7896 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7897 	else
7898 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7899 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7900 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7901 skip_ss_tests:
7902 
7903 	vmcs_write(GUEST_AR_SS, ar_saved);
7904 	vmcs_write(GUEST_SEL_SS, sel_saved);
7905 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrl0_saved);
7906 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved);
7907 }
7908 
7909 #define	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(xfail, seg_base)			\
7910 do {										\
7911 	addr_saved = vmcs_read(seg_base);					\
7912 	for (i = 32; i < 63; i = i + 4) {					\
7913 		addr = addr_saved | 1ull << i;					\
7914 		vmcs_write(seg_base, addr);					\
7915 		test_guest_state("seg.BASE[63:32] != 0, usable = " xstr(xfail),	\
7916 				 xfail, addr, xstr(seg_base));			\
7917 	}									\
7918 	vmcs_write(seg_base, addr_saved);					\
7919 } while (0)
7920 
7921 #define	TEST_SEGMENT_BASE_ADDR_CANONICAL(xfail, seg_base)		  \
7922 do {									  \
7923 	addr_saved = vmcs_read(seg_base);				  \
7924 	vmcs_write(seg_base, NONCANONICAL);				  \
7925 	test_guest_state("seg.BASE non-canonical, usable = " xstr(xfail), \
7926 			 xfail, NONCANONICAL, xstr(seg_base));		  \
7927 	vmcs_write(seg_base, addr_saved);				  \
7928 } while (0)
7929 
7930 /*
7931  * The following checks are done on the Base Address field of the Guest
7932  * Segment Registers on processors that support Intel 64 architecture:
7933  *    - TR, FS, GS : The address must be canonical.
7934  *    - LDTR : If LDTR is usable, the address must be canonical.
7935  *    - CS : Bits 63:32 of the address must be zero.
7936  *    - SS, DS, ES : If the register is usable, bits 63:32 of the address
7937  *	must be zero.
7938  *
7939  *  [Intel SDM]
7940  */
7941 static void test_guest_segment_base_addr_fields(void)
7942 {
7943 	u64 addr_saved;
7944 	u64 addr;
7945 	u32 ar_saved;
7946 	int i;
7947 
7948 	/*
7949 	 * The address of TR, FS, GS and LDTR must be canonical.
7950 	 */
7951 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_TR);
7952 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_FS);
7953 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_GS);
7954 	ar_saved = vmcs_read(GUEST_AR_LDTR);
7955 	/* Make LDTR unusable */
7956 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7957 	TEST_SEGMENT_BASE_ADDR_CANONICAL(false, GUEST_BASE_LDTR);
7958 	/* Make LDTR usable */
7959 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7960 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_LDTR);
7961 
7962 	vmcs_write(GUEST_AR_LDTR, ar_saved);
7963 
7964 	/*
7965 	 * Bits 63:32 in CS, SS, DS and ES base address must be zero
7966 	 */
7967 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_CS);
7968 	ar_saved = vmcs_read(GUEST_AR_SS);
7969 	/* Make SS unusable */
7970 	vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7971 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_SS);
7972 	/* Make SS usable */
7973 	vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7974 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_SS);
7975 	vmcs_write(GUEST_AR_SS, ar_saved);
7976 
7977 	ar_saved = vmcs_read(GUEST_AR_DS);
7978 	/* Make DS unusable */
7979 	vmcs_write(GUEST_AR_DS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7980 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_DS);
7981 	/* Make DS usable */
7982 	vmcs_write(GUEST_AR_DS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7983 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_DS);
7984 	vmcs_write(GUEST_AR_DS, ar_saved);
7985 
7986 	ar_saved = vmcs_read(GUEST_AR_ES);
7987 	/* Make ES unusable */
7988 	vmcs_write(GUEST_AR_ES, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7989 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_ES);
7990 	/* Make ES usable */
7991 	vmcs_write(GUEST_AR_ES, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7992 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_ES);
7993 	vmcs_write(GUEST_AR_ES, ar_saved);
7994 }
7995 
7996 /*
7997  * Check that the virtual CPU checks the VMX Guest State Area as
7998  * documented in the Intel SDM.
7999  */
8000 static void vmx_guest_state_area_test(void)
8001 {
8002 	vmx_set_test_stage(1);
8003 	test_set_guest(guest_state_test_main);
8004 
8005 	/*
8006 	 * The IA32_SYSENTER_ESP field and the IA32_SYSENTER_EIP field
8007 	 * must each contain a canonical address.
8008 	 */
8009 	test_canonical(GUEST_SYSENTER_ESP, "GUEST_SYSENTER_ESP", false);
8010 	test_canonical(GUEST_SYSENTER_EIP, "GUEST_SYSENTER_EIP", false);
8011 
8012 	test_guest_dr7();
8013 	test_load_guest_pat();
8014 	test_guest_efer();
8015 	test_load_guest_perf_global_ctrl();
8016 	test_load_guest_bndcfgs();
8017 
8018 	test_guest_segment_sel_fields();
8019 	test_guest_segment_base_addr_fields();
8020 
8021 	test_canonical(GUEST_BASE_GDTR, "GUEST_BASE_GDTR", false);
8022 	test_canonical(GUEST_BASE_IDTR, "GUEST_BASE_IDTR", false);
8023 
8024 	u32 guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_GDTR);
8025 	int i;
8026 	for (i = 16; i <= 31; i++) {
8027 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8028 		vmcs_write(GUEST_LIMIT_GDTR, tmp);
8029 		test_guest_state("GDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_GDTR");
8030 	}
8031 	vmcs_write(GUEST_LIMIT_GDTR, guest_desc_limit_saved);
8032 
8033 	guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_IDTR);
8034 	for (i = 16; i <= 31; i++) {
8035 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8036 		vmcs_write(GUEST_LIMIT_IDTR, tmp);
8037 		test_guest_state("IDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_IDTR");
8038 	}
8039 	vmcs_write(GUEST_LIMIT_IDTR, guest_desc_limit_saved);
8040 
8041 	/*
8042 	 * Let the guest finish execution
8043 	 */
8044 	vmx_set_test_stage(2);
8045 	enter_guest();
8046 }
8047 
8048 extern void unrestricted_guest_main(void);
8049 asm (".code32\n"
8050 	"unrestricted_guest_main:\n"
8051 	"vmcall\n"
8052 	"nop\n"
8053 	"mov $1, %edi\n"
8054 	"call hypercall\n"
8055 	".code64\n");
8056 
8057 static void setup_unrestricted_guest(void)
8058 {
8059 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) & ~(X86_CR0_PG));
8060 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) & ~ENT_GUEST_64);
8061 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) & ~EFER_LMA);
8062 	vmcs_write(GUEST_RIP, virt_to_phys(unrestricted_guest_main));
8063 }
8064 
8065 static void unsetup_unrestricted_guest(void)
8066 {
8067 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) | X86_CR0_PG);
8068 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_GUEST_64);
8069 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) | EFER_LMA);
8070 	vmcs_write(GUEST_RIP, (u64) phys_to_virt(vmcs_read(GUEST_RIP)));
8071 	vmcs_write(GUEST_RSP, (u64) phys_to_virt(vmcs_read(GUEST_RSP)));
8072 }
8073 
8074 /*
8075  * If "unrestricted guest" secondary VM-execution control is set, guests
8076  * can run in unpaged protected mode.
8077  */
8078 static void vmentry_unrestricted_guest_test(void)
8079 {
8080 	if (enable_unrestricted_guest(true)) {
8081 		report_skip("%s: \"Unrestricted guest\" exec control not supported", __func__);
8082 		return;
8083 	}
8084 
8085 	test_set_guest(unrestricted_guest_main);
8086 	setup_unrestricted_guest();
8087 	test_guest_state("Unrestricted guest test", false, CPU_URG, "CPU_URG");
8088 
8089 	/*
8090 	 * Let the guest finish execution as a regular guest
8091 	 */
8092 	unsetup_unrestricted_guest();
8093 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) & ~CPU_URG);
8094 	enter_guest();
8095 }
8096 
8097 static bool valid_vmcs_for_vmentry(void)
8098 {
8099 	struct vmcs *current_vmcs = NULL;
8100 
8101 	if (vmcs_save(&current_vmcs))
8102 		return false;
8103 
8104 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
8105 }
8106 
8107 static void try_vmentry_in_movss_shadow(void)
8108 {
8109 	u32 vm_inst_err;
8110 	u32 flags;
8111 	bool early_failure = false;
8112 	u32 expected_flags = X86_EFLAGS_FIXED;
8113 	bool valid_vmcs = valid_vmcs_for_vmentry();
8114 
8115 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
8116 
8117 	/*
8118 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
8119 	 * unsupported VMCS component").
8120 	 */
8121 	vmcs_write(~0u, 0);
8122 
8123 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
8124 			      "vmwrite %%rsp, %%rdx;"
8125 			      "mov 0f, %%rax;"
8126 			      "mov %[host_rip], %%edx;"
8127 			      "vmwrite %%rax, %%rdx;"
8128 			      "mov $-1, %%ah;"
8129 			      "sahf;"
8130 			      "mov %%ss, %%ax;"
8131 			      "mov %%ax, %%ss;"
8132 			      "vmlaunch;"
8133 			      "mov $1, %[early_failure];"
8134 			      "0: lahf;"
8135 			      "movzbl %%ah, %[flags]"
8136 			      : [early_failure] "+r" (early_failure),
8137 				[flags] "=&a" (flags)
8138 			      : [host_rsp] "i" (HOST_RSP),
8139 				[host_rip] "i" (HOST_RIP)
8140 			      : "rdx", "cc", "memory");
8141 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
8142 
8143 	report(early_failure, "Early VM-entry failure");
8144 	report(flags == expected_flags, "RFLAGS[8:0] is %x (actual %x)",
8145 	       expected_flags, flags);
8146 	if (valid_vmcs)
8147 		report(vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
8148 		       "VM-instruction error is %d (actual %d)",
8149 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
8150 }
8151 
8152 static void vmentry_movss_shadow_test(void)
8153 {
8154 	struct vmcs *orig_vmcs;
8155 
8156 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
8157 
8158 	/*
8159 	 * Set the launched flag on the current VMCS to verify the correct
8160 	 * error priority, below.
8161 	 */
8162 	test_set_guest(v2_null_test_guest);
8163 	enter_guest();
8164 
8165 	/*
8166 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
8167 	 * fail due to invalid guest state (if we make it that far).
8168 	 */
8169 	vmcs_write(GUEST_RFLAGS, 0);
8170 
8171 	/*
8172 	 * "VM entry with events blocked by MOV SS" takes precedence over
8173 	 * "VMLAUNCH with non-clear VMCS."
8174 	 */
8175 	report_prefix_push("valid current-VMCS");
8176 	try_vmentry_in_movss_shadow();
8177 	report_prefix_pop();
8178 
8179 	/*
8180 	 * VMfailInvalid takes precedence over "VM entry with events
8181 	 * blocked by MOV SS."
8182 	 */
8183 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
8184 	report_prefix_push("no current-VMCS");
8185 	try_vmentry_in_movss_shadow();
8186 	report_prefix_pop();
8187 
8188 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
8189 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8190 }
8191 
8192 static void vmx_ldtr_test_guest(void)
8193 {
8194 	u16 ldtr = sldt();
8195 
8196 	report(ldtr == NP_SEL, "Expected %x for L2 LDTR selector (got %x)",
8197 	       NP_SEL, ldtr);
8198 }
8199 
8200 /*
8201  * Ensure that the L1 LDTR is set to 0 on VM-exit.
8202  */
8203 static void vmx_ldtr_test(void)
8204 {
8205 	const u8 ldt_ar = 0x82; /* Present LDT */
8206 	u16 sel = FIRST_SPARE_SEL;
8207 
8208 	/* Set up a non-zero L1 LDTR prior to VM-entry. */
8209 	set_gdt_entry(sel, 0, 0, ldt_ar, 0);
8210 	lldt(sel);
8211 
8212 	test_set_guest(vmx_ldtr_test_guest);
8213 	/*
8214 	 * Set up a different LDTR for L2. The actual GDT contents are
8215 	 * irrelevant, since we stuff the hidden descriptor state
8216 	 * straight into the VMCS rather than reading it from the GDT.
8217 	 */
8218 	vmcs_write(GUEST_SEL_LDTR, NP_SEL);
8219 	vmcs_write(GUEST_AR_LDTR, ldt_ar);
8220 	enter_guest();
8221 
8222 	/*
8223 	 * VM-exit should clear LDTR (and make it unusable, but we
8224 	 * won't verify that here).
8225 	 */
8226 	sel = sldt();
8227 	report(!sel, "Expected 0 for L1 LDTR selector (got %x)", sel);
8228 }
8229 
8230 static void vmx_single_vmcall_guest(void)
8231 {
8232 	vmcall();
8233 }
8234 
8235 static void vmx_cr_load_test(void)
8236 {
8237 	unsigned long cr3, cr4, orig_cr3, orig_cr4;
8238 	u32 ctrls[2] = {0};
8239 	pgd_t *pml5;
8240 
8241 	orig_cr4 = read_cr4();
8242 	orig_cr3 = read_cr3();
8243 
8244 	if (!this_cpu_has(X86_FEATURE_PCID)) {
8245 		report_skip("%s : PCID not detected", __func__);
8246 		return;
8247 	}
8248 	if (!this_cpu_has(X86_FEATURE_MCE)) {
8249 		report_skip("%s : MCE not detected", __func__);
8250 		return;
8251 	}
8252 
8253 	TEST_ASSERT(!(orig_cr3 & X86_CR3_PCID_MASK));
8254 
8255 	/* Enable PCID for L1. */
8256 	cr4 = orig_cr4 | X86_CR4_PCIDE;
8257 	cr3 = orig_cr3 | 0x1;
8258 	TEST_ASSERT(!write_cr4_safe(cr4));
8259 	write_cr3(cr3);
8260 
8261 	test_set_guest(vmx_single_vmcall_guest);
8262 	vmcs_write(HOST_CR4, cr4);
8263 	vmcs_write(HOST_CR3, cr3);
8264 	enter_guest();
8265 
8266 	/*
8267 	 * No exception is expected.
8268 	 *
8269 	 * NB. KVM loads the last guest write to CR4 into CR4 read
8270 	 *     shadow. In order to trigger an exit to KVM, we can toggle a
8271 	 *     bit that is owned by KVM. We use CR4.MCE, which shall
8272 	 *     have no side effect because normally no guest MCE (e.g., as the
8273 	 *     result of bad memory) would happen during this test.
8274 	 */
8275 	TEST_ASSERT(!write_cr4_safe(cr4 ^ X86_CR4_MCE));
8276 
8277 	/* Cleanup L1 state. */
8278 	write_cr3(orig_cr3);
8279 	TEST_ASSERT(!write_cr4_safe(orig_cr4));
8280 
8281 	if (!this_cpu_has(X86_FEATURE_LA57))
8282 		goto done;
8283 
8284 	/*
8285 	 * Allocate a full page for PML5 to guarantee alignment, though only
8286 	 * the first entry needs to be filled (the test's virtual addresses
8287 	 * most definitely do not have any of bits 56:48 set).
8288 	 */
8289 	pml5 = alloc_page();
8290 	*pml5 = orig_cr3 | PT_PRESENT_MASK | PT_WRITABLE_MASK;
8291 
8292 	/*
8293 	 * Transition to/from 5-level paging in the host via VM-Exit.  CR4.LA57
8294 	 * can't be toggled while long is active via MOV CR4, but there are no
8295 	 * such restrictions on VM-Exit.
8296 	 */
8297 lol_5level:
8298 	vmcs_write(HOST_CR4, orig_cr4 | X86_CR4_LA57);
8299 	vmcs_write(HOST_CR3, virt_to_phys(pml5));
8300 	enter_guest();
8301 
8302 	/*
8303 	 * VMREAD with a memory operand to verify KVM detects the LA57 change,
8304 	 * e.g. uses the correct guest root level in gva_to_gpa().
8305 	 */
8306 	TEST_ASSERT(vmcs_readm(HOST_CR3) == virt_to_phys(pml5));
8307 	TEST_ASSERT(vmcs_readm(HOST_CR4) == (orig_cr4 | X86_CR4_LA57));
8308 
8309 	vmcs_write(HOST_CR4, orig_cr4);
8310 	vmcs_write(HOST_CR3, orig_cr3);
8311 	enter_guest();
8312 
8313 	TEST_ASSERT(vmcs_readm(HOST_CR3) == orig_cr3);
8314 	TEST_ASSERT(vmcs_readm(HOST_CR4) == orig_cr4);
8315 
8316 	/*
8317 	 * And now do the same LA57 shenanigans with EPT enabled.  KVM uses
8318 	 * two separate MMUs when L1 uses TDP, whereas the above shadow paging
8319 	 * version shares an MMU between L1 and L2.
8320 	 *
8321 	 * If the saved execution controls are non-zero then the EPT version
8322 	 * has already run.  In that case, restore the old controls.  If EPT
8323 	 * setup fails, e.g. EPT isn't supported, fall through and finish up.
8324 	 */
8325 	if (ctrls[0]) {
8326 		vmcs_write(CPU_EXEC_CTRL0, ctrls[0]);
8327 		vmcs_write(CPU_EXEC_CTRL1, ctrls[1]);
8328 	} else if (!setup_ept(false)) {
8329 		ctrls[0] = vmcs_read(CPU_EXEC_CTRL0);
8330 		ctrls[1]  = vmcs_read(CPU_EXEC_CTRL1);
8331 		goto lol_5level;
8332 	}
8333 
8334 	free_page(pml5);
8335 
8336 done:
8337 	skip_exit_vmcall();
8338 	enter_guest();
8339 }
8340 
8341 static void vmx_cr4_osxsave_test_guest(void)
8342 {
8343 	write_cr4(read_cr4() & ~X86_CR4_OSXSAVE);
8344 }
8345 
8346 /*
8347  * Ensure that kvm recalculates the L1 guest's CPUID.01H:ECX.OSXSAVE
8348  * after VM-exit from an L2 guest that sets CR4.OSXSAVE to a different
8349  * value than in L1.
8350  */
8351 static void vmx_cr4_osxsave_test(void)
8352 {
8353 	if (!this_cpu_has(X86_FEATURE_XSAVE)) {
8354 		report_skip("%s : XSAVE not detected", __func__);
8355 		return;
8356 	}
8357 
8358 	if (!(read_cr4() & X86_CR4_OSXSAVE)) {
8359 		unsigned long cr4 = read_cr4() | X86_CR4_OSXSAVE;
8360 
8361 		write_cr4(cr4);
8362 		vmcs_write(GUEST_CR4, cr4);
8363 		vmcs_write(HOST_CR4, cr4);
8364 	}
8365 
8366 	TEST_ASSERT(this_cpu_has(X86_FEATURE_OSXSAVE));
8367 
8368 	test_set_guest(vmx_cr4_osxsave_test_guest);
8369 	enter_guest();
8370 
8371 	TEST_ASSERT(this_cpu_has(X86_FEATURE_OSXSAVE));
8372 }
8373 
8374 /*
8375  * FNOP with both CR0.TS and CR0.EM clear should not generate #NM, and the L2
8376  * guest should exit normally.
8377  */
8378 static void vmx_no_nm_test(void)
8379 {
8380 	test_set_guest(fnop);
8381 	vmcs_write(GUEST_CR0, read_cr0() & ~(X86_CR0_TS | X86_CR0_EM));
8382 	enter_guest();
8383 }
8384 
8385 bool vmx_pending_event_ipi_fired;
8386 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
8387 {
8388 	vmx_pending_event_ipi_fired = true;
8389 	eoi();
8390 }
8391 
8392 bool vmx_pending_event_guest_run;
8393 static void vmx_pending_event_guest(void)
8394 {
8395 	vmcall();
8396 	vmx_pending_event_guest_run = true;
8397 }
8398 
8399 static void vmx_pending_event_test_core(bool guest_hlt)
8400 {
8401 	int ipi_vector = 0xf1;
8402 
8403 	vmx_pending_event_ipi_fired = false;
8404 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
8405 
8406 	vmx_pending_event_guest_run = false;
8407 	test_set_guest(vmx_pending_event_guest);
8408 
8409 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
8410 
8411 	enter_guest();
8412 	skip_exit_vmcall();
8413 
8414 	if (guest_hlt)
8415 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8416 
8417 	cli();
8418 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
8419 				   APIC_DM_FIXED | ipi_vector,
8420 				   0);
8421 
8422 	enter_guest();
8423 
8424 	assert_exit_reason(VMX_EXTINT);
8425 	report(!vmx_pending_event_guest_run,
8426 	       "Guest did not run before host received IPI");
8427 
8428 	sti_nop_cli();
8429 	report(vmx_pending_event_ipi_fired,
8430 	       "Got pending interrupt after IRQ enabled");
8431 
8432 	if (guest_hlt)
8433 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8434 
8435 	enter_guest();
8436 	report(vmx_pending_event_guest_run,
8437 	       "Guest finished running when no interrupt");
8438 }
8439 
8440 static void vmx_pending_event_test(void)
8441 {
8442 	vmx_pending_event_test_core(false);
8443 }
8444 
8445 static void vmx_pending_event_hlt_test(void)
8446 {
8447 	vmx_pending_event_test_core(true);
8448 }
8449 
8450 static int vmx_window_test_db_count;
8451 
8452 static void vmx_window_test_db_handler(struct ex_regs *regs)
8453 {
8454 	vmx_window_test_db_count++;
8455 }
8456 
8457 static void vmx_nmi_window_test_guest(void)
8458 {
8459 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8460 
8461 	asm volatile("vmcall\n\t"
8462 		     "nop\n\t");
8463 
8464 	handle_exception(DB_VECTOR, NULL);
8465 }
8466 
8467 static void verify_nmi_window_exit(u64 rip)
8468 {
8469 	u32 exit_reason = vmcs_read(EXI_REASON);
8470 
8471 	report(exit_reason == VMX_NMI_WINDOW,
8472 	       "Exit reason (%d) is 'NMI window'", exit_reason);
8473 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8474 	       vmcs_read(GUEST_RIP), rip);
8475 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8476 }
8477 
8478 static void vmx_nmi_window_test(void)
8479 {
8480 	u64 nop_addr;
8481 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8482 
8483 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
8484 		report_skip("%s : \"Virtual NMIs\" exec control not supported", __func__);
8485 		return;
8486 	}
8487 
8488 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
8489 		report_skip("%s : \"NMI-window exiting\" exec control not supported", __func__);
8490 		return;
8491 	}
8492 
8493 	vmx_window_test_db_count = 0;
8494 
8495 	report_prefix_push("NMI-window");
8496 	test_set_guest(vmx_nmi_window_test_guest);
8497 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
8498 	enter_guest();
8499 	skip_exit_vmcall();
8500 	nop_addr = vmcs_read(GUEST_RIP);
8501 
8502 	/*
8503 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
8504 	 * RIP will not advance.
8505 	 */
8506 	report_prefix_push("active, no blocking");
8507 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8508 	enter_guest();
8509 	verify_nmi_window_exit(nop_addr);
8510 	report_prefix_pop();
8511 
8512 	/*
8513 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
8514 	 * a VM-exit on the next instruction after the nop. (The nop
8515 	 * is one byte.)
8516 	 */
8517 	report_prefix_push("active, blocking by MOV-SS");
8518 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8519 	enter_guest();
8520 	verify_nmi_window_exit(nop_addr + 1);
8521 	report_prefix_pop();
8522 
8523 	/*
8524 	 * Ask for "NMI-window exiting" (with event injection), and
8525 	 * expect a VM-exit after the event is injected. (RIP should
8526 	 * be at the address specified in the IDT entry for #DB.)
8527 	 */
8528 	report_prefix_push("active, no blocking, injecting #DB");
8529 	vmcs_write(ENT_INTR_INFO,
8530 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8531 	enter_guest();
8532 	verify_nmi_window_exit((u64)db_fault_addr);
8533 	report_prefix_pop();
8534 
8535 	/*
8536 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
8537 	 * a VM-exit after the next IRET (i.e. after the #DB handler
8538 	 * returns). So, RIP should be back at one byte past the nop.
8539 	 */
8540 	report_prefix_push("active, blocking by NMI");
8541 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
8542 	enter_guest();
8543 	verify_nmi_window_exit(nop_addr + 1);
8544 	report(vmx_window_test_db_count == 1,
8545 	       "#DB handler executed once (actual %d times)",
8546 	       vmx_window_test_db_count);
8547 	report_prefix_pop();
8548 
8549 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8550 		report_skip("CPU does not support activity state HLT.");
8551 	} else {
8552 		/*
8553 		 * Ask for "NMI-window exiting" when entering activity
8554 		 * state HLT, and expect an immediate VM-exit. RIP is
8555 		 * still one byte past the nop.
8556 		 */
8557 		report_prefix_push("halted, no blocking");
8558 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8559 		enter_guest();
8560 		verify_nmi_window_exit(nop_addr + 1);
8561 		report_prefix_pop();
8562 
8563 		/*
8564 		 * Ask for "NMI-window exiting" when entering activity
8565 		 * state HLT (with event injection), and expect a
8566 		 * VM-exit after the event is injected. (RIP should be
8567 		 * at the address specified in the IDT entry for #DB.)
8568 		 */
8569 		report_prefix_push("halted, no blocking, injecting #DB");
8570 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8571 		vmcs_write(ENT_INTR_INFO,
8572 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8573 			   DB_VECTOR);
8574 		enter_guest();
8575 		verify_nmi_window_exit((u64)db_fault_addr);
8576 		report_prefix_pop();
8577 	}
8578 
8579 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8580 	enter_guest();
8581 	report_prefix_pop();
8582 }
8583 
8584 static void vmx_intr_window_test_guest(void)
8585 {
8586 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8587 
8588 	/*
8589 	 * The two consecutive STIs are to ensure that only the first
8590 	 * one has a shadow. Note that NOP and STI are one byte
8591 	 * instructions.
8592 	 */
8593 	asm volatile("vmcall\n\t"
8594 		     "nop\n\t"
8595 		     "sti\n\t"
8596 		     "sti\n\t");
8597 
8598 	handle_exception(DB_VECTOR, NULL);
8599 }
8600 
8601 static void verify_intr_window_exit(u64 rip)
8602 {
8603 	u32 exit_reason = vmcs_read(EXI_REASON);
8604 
8605 	report(exit_reason == VMX_INTR_WINDOW,
8606 	       "Exit reason (%d) is 'interrupt window'", exit_reason);
8607 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8608 	       vmcs_read(GUEST_RIP), rip);
8609 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8610 }
8611 
8612 static void vmx_intr_window_test(void)
8613 {
8614 	u64 vmcall_addr;
8615 	u64 nop_addr;
8616 	unsigned int orig_db_gate_type;
8617 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8618 
8619 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
8620 		report_skip("%s : \"Interrupt-window exiting\" exec control not supported", __func__);
8621 		return;
8622 	}
8623 
8624 	/*
8625 	 * Change the IDT entry for #DB from interrupt gate to trap gate,
8626 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
8627 	 * be disabled after vectoring a #DB.
8628 	 */
8629 	orig_db_gate_type = boot_idt[DB_VECTOR].type;
8630 	boot_idt[DB_VECTOR].type = 15;
8631 
8632 	report_prefix_push("interrupt-window");
8633 	test_set_guest(vmx_intr_window_test_guest);
8634 	enter_guest();
8635 	assert_exit_reason(VMX_VMCALL);
8636 	vmcall_addr = vmcs_read(GUEST_RIP);
8637 
8638 	/*
8639 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
8640 	 * no blocking; expect an immediate VM-exit. Note that we have
8641 	 * not advanced past the vmcall instruction yet, so RIP should
8642 	 * point to the vmcall instruction.
8643 	 */
8644 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
8645 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8646 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
8647 	enter_guest();
8648 	verify_intr_window_exit(vmcall_addr);
8649 	report_prefix_pop();
8650 
8651 	/*
8652 	 * Ask for "interrupt-window exiting" (with event injection)
8653 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
8654 	 * the event is injected. That is, RIP should should be at the
8655 	 * address specified in the IDT entry for #DB.
8656 	 */
8657 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #DB");
8658 	vmcs_write(ENT_INTR_INFO,
8659 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8660 	vmcall_addr = vmcs_read(GUEST_RIP);
8661 	enter_guest();
8662 	verify_intr_window_exit((u64)db_fault_addr);
8663 	report_prefix_pop();
8664 
8665 	/*
8666 	 * Let the L2 guest run through the IRET, back to the VMCALL.
8667 	 * We have to clear the "interrupt-window exiting"
8668 	 * VM-execution control, or it would just keep causing
8669 	 * VM-exits. Then, advance past the VMCALL and set the
8670 	 * "interrupt-window exiting" VM-execution control again.
8671 	 */
8672 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8673 	enter_guest();
8674 	skip_exit_vmcall();
8675 	nop_addr = vmcs_read(GUEST_RIP);
8676 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8677 
8678 	/*
8679 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
8680 	 * RFLAGS.IF set, and expect a VM-exit on the next
8681 	 * instruction. (NOP is one byte.)
8682 	 */
8683 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
8684 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8685 	enter_guest();
8686 	verify_intr_window_exit(nop_addr + 1);
8687 	report_prefix_pop();
8688 
8689 	/*
8690 	 * Back up to the NOP and ask for "interrupt-window exiting"
8691 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
8692 	 * on the next instruction. (NOP is one byte.)
8693 	 */
8694 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
8695 	vmcs_write(GUEST_RIP, nop_addr);
8696 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
8697 	enter_guest();
8698 	verify_intr_window_exit(nop_addr + 1);
8699 	report_prefix_pop();
8700 
8701 	/*
8702 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
8703 	 * and expect a VM-exit on the instruction following the STI
8704 	 * shadow. Only the first STI (which is one byte past the NOP)
8705 	 * should have a shadow. The second STI (which is two bytes
8706 	 * past the NOP) has no shadow. Therefore, the interrupt
8707 	 * window opens at three bytes past the NOP.
8708 	 */
8709 	report_prefix_push("active, RFLAGS.IF = 0");
8710 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8711 	enter_guest();
8712 	verify_intr_window_exit(nop_addr + 3);
8713 	report_prefix_pop();
8714 
8715 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8716 		report_skip("CPU does not support activity state HLT.");
8717 	} else {
8718 		/*
8719 		 * Ask for "interrupt-window exiting" when entering
8720 		 * activity state HLT, and expect an immediate
8721 		 * VM-exit. RIP is still three bytes past the nop.
8722 		 */
8723 		report_prefix_push("halted, no blocking");
8724 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8725 		enter_guest();
8726 		verify_intr_window_exit(nop_addr + 3);
8727 		report_prefix_pop();
8728 
8729 		/*
8730 		 * Ask for "interrupt-window exiting" when entering
8731 		 * activity state HLT (with event injection), and
8732 		 * expect a VM-exit after the event is injected. That
8733 		 * is, RIP should should be at the address specified
8734 		 * in the IDT entry for #DB.
8735 		 */
8736 		report_prefix_push("halted, no blocking, injecting #DB");
8737 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8738 		vmcs_write(ENT_INTR_INFO,
8739 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8740 			   DB_VECTOR);
8741 		enter_guest();
8742 		verify_intr_window_exit((u64)db_fault_addr);
8743 		report_prefix_pop();
8744 	}
8745 
8746 	boot_idt[DB_VECTOR].type = orig_db_gate_type;
8747 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8748 	enter_guest();
8749 	report_prefix_pop();
8750 }
8751 
8752 #define GUEST_TSC_OFFSET (1u << 30)
8753 
8754 static u64 guest_tsc;
8755 
8756 static void vmx_store_tsc_test_guest(void)
8757 {
8758 	guest_tsc = rdtsc();
8759 }
8760 
8761 /*
8762  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
8763  * list, the value saved is not subject to the TSC offset that is
8764  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
8765  */
8766 static void vmx_store_tsc_test(void)
8767 {
8768 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
8769 	u64 low, high;
8770 
8771 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
8772 		report_skip("%s : \"Use TSC offsetting\" exec control not supported", __func__);
8773 		return;
8774 	}
8775 
8776 	test_set_guest(vmx_store_tsc_test_guest);
8777 
8778 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
8779 	vmcs_write(EXI_MSR_ST_CNT, 1);
8780 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
8781 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
8782 
8783 	low = rdtsc();
8784 	enter_guest();
8785 	high = rdtsc();
8786 
8787 	report(low + GUEST_TSC_OFFSET <= guest_tsc &&
8788 	       guest_tsc <= high + GUEST_TSC_OFFSET,
8789 	       "RDTSC value in the guest (%lu) is in range [%lu, %lu]",
8790 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
8791 	report(low <= msr_entry.value && msr_entry.value <= high,
8792 	       "IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
8793 	       msr_entry.value, low, high);
8794 }
8795 
8796 static void vmx_preemption_timer_zero_test_db_handler(struct ex_regs *regs)
8797 {
8798 }
8799 
8800 static void vmx_preemption_timer_zero_test_guest(void)
8801 {
8802 	while (vmx_get_test_stage() < 3)
8803 		vmcall();
8804 }
8805 
8806 static void vmx_preemption_timer_zero_activate_preemption_timer(void)
8807 {
8808 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
8809 	vmcs_write(PREEMPT_TIMER_VALUE, 0);
8810 }
8811 
8812 static void vmx_preemption_timer_zero_advance_past_vmcall(void)
8813 {
8814 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
8815 	enter_guest();
8816 	skip_exit_vmcall();
8817 }
8818 
8819 static void vmx_preemption_timer_zero_inject_db(bool intercept_db)
8820 {
8821 	vmx_preemption_timer_zero_activate_preemption_timer();
8822 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK |
8823 		   INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8824 	vmcs_write(EXC_BITMAP, intercept_db ? 1 << DB_VECTOR : 0);
8825 	enter_guest();
8826 }
8827 
8828 static void vmx_preemption_timer_zero_set_pending_dbg(u32 exception_bitmap)
8829 {
8830 	vmx_preemption_timer_zero_activate_preemption_timer();
8831 	vmcs_write(GUEST_PENDING_DEBUG, PENDING_DBG_TRAP | DR6_TRAP1);
8832 	vmcs_write(EXC_BITMAP, exception_bitmap);
8833 	enter_guest();
8834 }
8835 
8836 static void vmx_preemption_timer_zero_expect_preempt_at_rip(u64 expected_rip)
8837 {
8838 	u32 reason = (u32)vmcs_read(EXI_REASON);
8839 	u64 guest_rip = vmcs_read(GUEST_RIP);
8840 
8841 	report(reason == VMX_PREEMPT && guest_rip == expected_rip,
8842 	       "Exit reason is 0x%x (expected 0x%x) and guest RIP is %lx (0x%lx expected).",
8843 	       reason, VMX_PREEMPT, guest_rip, expected_rip);
8844 }
8845 
8846 /*
8847  * This test ensures that when the VMX preemption timer is zero at
8848  * VM-entry, a VM-exit occurs after any event injection and after any
8849  * pending debug exceptions are raised, but before execution of any
8850  * guest instructions.
8851  */
8852 static void vmx_preemption_timer_zero_test(void)
8853 {
8854 	u64 db_fault_address = (u64)get_idt_addr(&boot_idt[DB_VECTOR]);
8855 	handler old_db;
8856 	u32 reason;
8857 
8858 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
8859 		report_skip("%s : \"Activate VMX-preemption timer\" pin control not supported", __func__);
8860 		return;
8861 	}
8862 
8863 	/*
8864 	 * Install a custom #DB handler that doesn't abort.
8865 	 */
8866 	old_db = handle_exception(DB_VECTOR,
8867 				  vmx_preemption_timer_zero_test_db_handler);
8868 
8869 	test_set_guest(vmx_preemption_timer_zero_test_guest);
8870 
8871 	/*
8872 	 * VMX-preemption timer should fire after event injection.
8873 	 */
8874 	vmx_set_test_stage(0);
8875 	vmx_preemption_timer_zero_inject_db(0);
8876 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8877 	vmx_preemption_timer_zero_advance_past_vmcall();
8878 
8879 	/*
8880 	 * VMX-preemption timer should fire after event injection.
8881 	 * Exception bitmap is irrelevant, since you can't intercept
8882 	 * an event that you injected.
8883 	 */
8884 	vmx_set_test_stage(1);
8885 	vmx_preemption_timer_zero_inject_db(true);
8886 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8887 	vmx_preemption_timer_zero_advance_past_vmcall();
8888 
8889 	/*
8890 	 * VMX-preemption timer should fire after pending debug exceptions
8891 	 * have delivered a #DB trap.
8892 	 */
8893 	vmx_set_test_stage(2);
8894 	vmx_preemption_timer_zero_set_pending_dbg(0);
8895 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8896 	vmx_preemption_timer_zero_advance_past_vmcall();
8897 
8898 	/*
8899 	 * VMX-preemption timer would fire after pending debug exceptions
8900 	 * have delivered a #DB trap, but in this case, the #DB trap is
8901 	 * intercepted.
8902 	 */
8903 	vmx_set_test_stage(3);
8904 	vmx_preemption_timer_zero_set_pending_dbg(1 << DB_VECTOR);
8905 	reason = (u32)vmcs_read(EXI_REASON);
8906 	report(reason == VMX_EXC_NMI, "Exit reason is 0x%x (expected 0x%x)",
8907 	       reason, VMX_EXC_NMI);
8908 
8909 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
8910 	enter_guest();
8911 
8912 	handle_exception(DB_VECTOR, old_db);
8913 }
8914 
8915 static u64 vmx_preemption_timer_tf_test_prev_rip;
8916 
8917 static void vmx_preemption_timer_tf_test_db_handler(struct ex_regs *regs)
8918 {
8919 	extern char vmx_preemption_timer_tf_test_endloop;
8920 
8921 	if (vmx_get_test_stage() == 2) {
8922 		/*
8923 		 * Stage 2 means that we're done, one way or another.
8924 		 * Arrange for the iret to drop us out of the wbinvd
8925 		 * loop and stop single-stepping.
8926 		 */
8927 		regs->rip = (u64)&vmx_preemption_timer_tf_test_endloop;
8928 		regs->rflags &= ~X86_EFLAGS_TF;
8929 	} else if (regs->rip == vmx_preemption_timer_tf_test_prev_rip) {
8930 		/*
8931 		 * The RIP should alternate between the wbinvd and the
8932 		 * jmp instruction in the code below. If we ever see
8933 		 * the same instruction twice in a row, that means a
8934 		 * single-step trap has been dropped. Let the
8935 		 * hypervisor know about the failure by executing a
8936 		 * VMCALL.
8937 		 */
8938 		vmcall();
8939 	}
8940 	vmx_preemption_timer_tf_test_prev_rip = regs->rip;
8941 }
8942 
8943 static void vmx_preemption_timer_tf_test_guest(void)
8944 {
8945 	/*
8946 	 * The hypervisor doesn't intercept WBINVD, so the loop below
8947 	 * shouldn't be a problem--it's just two instructions
8948 	 * executing in VMX non-root mode. However, when the
8949 	 * hypervisor is running in a virtual environment, the parent
8950 	 * hypervisor might intercept WBINVD and emulate it. If the
8951 	 * parent hypervisor is broken, the single-step trap after the
8952 	 * WBINVD might be lost.
8953 	 */
8954 	asm volatile("vmcall\n\t"
8955 		     "0: wbinvd\n\t"
8956 		     "1: jmp 0b\n\t"
8957 		     "vmx_preemption_timer_tf_test_endloop:");
8958 }
8959 
8960 /*
8961  * Ensure that the delivery of a "VMX-preemption timer expired"
8962  * VM-exit doesn't disrupt single-stepping in the guest. Note that
8963  * passing this test doesn't ensure correctness, because the test will
8964  * only fail if the VMX-preemtion timer fires at the right time (or
8965  * the wrong time, as it were).
8966  */
8967 static void vmx_preemption_timer_tf_test(void)
8968 {
8969 	handler old_db;
8970 	u32 reason;
8971 	int i;
8972 
8973 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
8974 		report_skip("%s : \"Activate VMX-preemption timer\" pin control not supported", __func__);
8975 		return;
8976 	}
8977 
8978 	old_db = handle_exception(DB_VECTOR,
8979 				  vmx_preemption_timer_tf_test_db_handler);
8980 
8981 	test_set_guest(vmx_preemption_timer_tf_test_guest);
8982 
8983 	enter_guest();
8984 	skip_exit_vmcall();
8985 
8986 	vmx_set_test_stage(1);
8987 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
8988 	vmcs_write(PREEMPT_TIMER_VALUE, 50000);
8989 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
8990 
8991 	/*
8992 	 * The only exit we should see is "VMX-preemption timer
8993 	 * expired."  If we get a VMCALL exit, that means the #DB
8994 	 * handler has detected a missing single-step trap. It doesn't
8995 	 * matter where the guest RIP is when the VMX-preemption timer
8996 	 * expires (whether it's in the WBINVD loop or in the #DB
8997 	 * handler)--a single-step trap should never be discarded.
8998 	 */
8999 	for (i = 0; i < 10000; i++) {
9000 		enter_guest();
9001 		reason = (u32)vmcs_read(EXI_REASON);
9002 		if (reason == VMX_PREEMPT)
9003 			continue;
9004 		TEST_ASSERT(reason == VMX_VMCALL);
9005 		skip_exit_insn();
9006 		break;
9007 	}
9008 
9009 	report(reason == VMX_PREEMPT, "No single-step traps skipped");
9010 
9011 	vmx_set_test_stage(2);
9012 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9013 	enter_guest();
9014 
9015 	handle_exception(DB_VECTOR, old_db);
9016 }
9017 
9018 #define VMX_PREEMPTION_TIMER_EXPIRY_CYCLES 1000000
9019 
9020 static u64 vmx_preemption_timer_expiry_start;
9021 static u64 vmx_preemption_timer_expiry_finish;
9022 
9023 static void vmx_preemption_timer_expiry_test_guest(void)
9024 {
9025 	vmcall();
9026 	vmx_preemption_timer_expiry_start = fenced_rdtsc();
9027 
9028 	while (vmx_get_test_stage() == 0)
9029 		vmx_preemption_timer_expiry_finish = fenced_rdtsc();
9030 }
9031 
9032 /*
9033  * Test that the VMX-preemption timer is not excessively delayed.
9034  *
9035  * Per the SDM, volume 3, VM-entry starts the VMX-preemption timer
9036  * with the unsigned value in the VMX-preemption timer-value field,
9037  * and the VMX-preemption timer counts down by 1 every time bit X in
9038  * the TSC changes due to a TSC increment (where X is
9039  * IA32_VMX_MISC[4:0]). If the timer counts down to zero in any state
9040  * other than the wait-for-SIPI state, the logical processor
9041  * transitions to the C0 C-state and causes a VM-exit.
9042  *
9043  * The guest code above reads the starting TSC after VM-entry. At this
9044  * point, the VMX-preemption timer has already been activated. Next,
9045  * the guest code reads the current TSC in a loop, storing the value
9046  * read to memory.
9047  *
9048  * If the RDTSC in the loop reads a value past the VMX-preemption
9049  * timer deadline, then the VMX-preemption timer VM-exit must be
9050  * delivered before the next instruction retires. Even if a higher
9051  * priority SMI is delivered first, the VMX-preemption timer VM-exit
9052  * must be delivered before the next instruction retires. Hence, a TSC
9053  * value past the VMX-preemption timer deadline might be read, but it
9054  * cannot be stored. If a TSC value past the deadline *is* stored,
9055  * then the architectural specification has been violated.
9056  */
9057 static void vmx_preemption_timer_expiry_test(void)
9058 {
9059 	u32 preemption_timer_value;
9060 	union vmx_misc misc;
9061 	u64 tsc_deadline;
9062 	u32 reason;
9063 
9064 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9065 		report_skip("%s : \"Activate VMX-preemption timer\" pin control not supported", __func__);
9066 		return;
9067 	}
9068 
9069 	test_set_guest(vmx_preemption_timer_expiry_test_guest);
9070 
9071 	enter_guest();
9072 	skip_exit_vmcall();
9073 
9074 	misc.val = rdmsr(MSR_IA32_VMX_MISC);
9075 	preemption_timer_value =
9076 		VMX_PREEMPTION_TIMER_EXPIRY_CYCLES >> misc.pt_bit;
9077 
9078 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9079 	vmcs_write(PREEMPT_TIMER_VALUE, preemption_timer_value);
9080 	vmx_set_test_stage(0);
9081 
9082 	enter_guest();
9083 	reason = (u32)vmcs_read(EXI_REASON);
9084 	TEST_ASSERT(reason == VMX_PREEMPT);
9085 
9086 	tsc_deadline = ((vmx_preemption_timer_expiry_start >> misc.pt_bit) <<
9087 			misc.pt_bit) + (preemption_timer_value << misc.pt_bit);
9088 
9089 	report(vmx_preemption_timer_expiry_finish < tsc_deadline,
9090 	       "Last stored guest TSC (%lu) < TSC deadline (%lu)",
9091 	       vmx_preemption_timer_expiry_finish, tsc_deadline);
9092 
9093 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9094 	vmx_set_test_stage(1);
9095 	enter_guest();
9096 }
9097 
9098 static void vmx_db_test_guest(void)
9099 {
9100 	/*
9101 	 * For a hardware generated single-step #DB.
9102 	 */
9103 	asm volatile("vmcall;"
9104 		     "nop;"
9105 		     ".Lpost_nop:");
9106 	/*
9107 	 * ...in a MOVSS shadow, with pending debug exceptions.
9108 	 */
9109 	asm volatile("vmcall;"
9110 		     "nop;"
9111 		     ".Lpost_movss_nop:");
9112 	/*
9113 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
9114 	 * emulates it in software.)
9115 	 */
9116 	asm volatile("vmcall;"
9117 		     "wbinvd;"
9118 		     ".Lpost_wbinvd:");
9119 	/*
9120 	 * ...in a MOVSS shadow, with pending debug exceptions.
9121 	 */
9122 	asm volatile("vmcall;"
9123 		     "wbinvd;"
9124 		     ".Lpost_movss_wbinvd:");
9125 	/*
9126 	 * For a hardware generated single-step #DB in a transactional region.
9127 	 */
9128 	asm volatile("vmcall;"
9129 		     ".Lxbegin: xbegin .Lskip_rtm;"
9130 		     "xend;"
9131 		     ".Lskip_rtm:");
9132 }
9133 
9134 /*
9135  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
9136  * L2. No #DB is delivered and L2 continues to the next point of
9137  * interest.
9138  */
9139 static void dismiss_db(void)
9140 {
9141 	vmcs_write(GUEST_PENDING_DEBUG, 0);
9142 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
9143 	enter_guest();
9144 }
9145 
9146 /*
9147  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
9148  * Then throw away the #DB exception and resume L2.
9149  */
9150 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
9151 			  void *expected_rip, u64 expected_exit_qual,
9152 			  u64 expected_dr6)
9153 {
9154 	u32 reason = vmcs_read(EXI_REASON);
9155 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
9156 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
9157 	u64 guest_rip = vmcs_read(GUEST_RIP);
9158 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
9159 	u64 dr6 = read_dr6();
9160 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
9161 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
9162 
9163 	report(reason == VMX_EXC_NMI && intr_info == expected_intr_info,
9164 	       "Expected #DB VM-exit");
9165 	report((u64)expected_rip == guest_rip, "Expected RIP %p (actual %lx)",
9166 	       expected_rip, guest_rip);
9167 	report_xfail(xfail_pdbg, 0 == guest_pending_dbg,
9168 		     "Expected pending debug exceptions 0 (actual %lx)",
9169 		     guest_pending_dbg);
9170 	report_xfail(xfail_qual, expected_exit_qual == exit_qual,
9171 		     "Expected exit qualification %lx (actual %lx)",
9172 		     expected_exit_qual, exit_qual);
9173 	report_xfail(xfail_dr6, expected_dr6 == dr6,
9174 		     "Expected DR6 %lx (actual %lx)", expected_dr6, dr6);
9175 	dismiss_db();
9176 }
9177 
9178 /*
9179  * Assuming the guest has just exited on a VMCALL instruction, skip
9180  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
9181  * pending debug exceptions are non-zero, set the VMCS up as if the
9182  * previous instruction was a MOVSS that generated the indicated
9183  * pending debug exceptions. Then enter L2.
9184  */
9185 static void single_step_guest(const char *test_name, u64 starting_dr6,
9186 			      u64 pending_debug_exceptions)
9187 {
9188 	printf("\n%s\n", test_name);
9189 	skip_exit_vmcall();
9190 	write_dr6(starting_dr6);
9191 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9192 	if (pending_debug_exceptions) {
9193 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
9194 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
9195 	}
9196 	enter_guest();
9197 }
9198 
9199 /*
9200  * When L1 intercepts #DB, verify that a single-step trap clears
9201  * pending debug exceptions, populates the exit qualification field
9202  * properly, and that DR6 is not prematurely clobbered. In a
9203  * (simulated) MOVSS shadow, make sure that the pending debug
9204  * exception bits are properly accumulated into the exit qualification
9205  * field.
9206  */
9207 static void vmx_db_test(void)
9208 {
9209 	/*
9210 	 * We are going to set a few arbitrary bits in DR6 to verify that
9211 	 * (a) DR6 is not modified by an intercepted #DB, and
9212 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
9213          *     the exit qualification field for a subsequent #DB exception.
9214 	 */
9215 	const u64 starting_dr6 = DR6_ACTIVE_LOW | DR6_BS | DR6_TRAP3 | DR6_TRAP1;
9216 	extern char post_nop asm(".Lpost_nop");
9217 	extern char post_movss_nop asm(".Lpost_movss_nop");
9218 	extern char post_wbinvd asm(".Lpost_wbinvd");
9219 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
9220 	extern char xbegin asm(".Lxbegin");
9221 	extern char skip_rtm asm(".Lskip_rtm");
9222 
9223 	/*
9224 	 * L1 wants to intercept #DB exceptions encountered in L2.
9225 	 */
9226 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
9227 
9228 	/*
9229 	 * Start L2 and run it up to the first point of interest.
9230 	 */
9231 	test_set_guest(vmx_db_test_guest);
9232 	enter_guest();
9233 
9234 	/*
9235 	 * Hardware-delivered #DB trap for single-step sets the
9236 	 * standard that L0 has to follow for emulated instructions.
9237 	 */
9238 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
9239 	check_db_exit(false, false, false, &post_nop, DR6_BS, starting_dr6);
9240 
9241 	/*
9242 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
9243 	 * also sets the standard that L0 has to follow for emulated
9244 	 * instructions. Here, we establish the VMCS pending debug
9245 	 * exceptions to indicate that the simulated MOVSS triggered a
9246 	 * data breakpoint as well as the single-step trap.
9247 	 */
9248 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
9249 			  starting_dr6, DR6_BS | PENDING_DBG_TRAP | DR6_TRAP0);
9250 	check_db_exit(false, false, false, &post_movss_nop, DR6_BS | DR6_TRAP0,
9251 		      starting_dr6);
9252 
9253 	/*
9254 	 * L0 synthesized #DB trap for single-step is buggy, because
9255 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
9256 	 * reconstitute the exit qualification from the prematurely
9257 	 * modified DR6, but fails miserably.
9258 	 */
9259 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
9260 	check_db_exit(false, false, false, &post_wbinvd, DR6_BS, starting_dr6);
9261 
9262 	/*
9263 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
9264 	 * even worse, because L0 also leaves the pending debug
9265 	 * exceptions in the VMCS instead of accumulating them into
9266 	 * the exit qualification field for the #DB exception.
9267 	 */
9268 	single_step_guest("Software synthesized single-step in MOVSS shadow",
9269 			  starting_dr6, DR6_BS | PENDING_DBG_TRAP | DR6_TRAP0);
9270 	check_db_exit(true, false, true, &post_movss_wbinvd, DR6_BS | DR6_TRAP0,
9271 		      starting_dr6);
9272 
9273 	/*
9274 	 * Optional RTM test for hardware that supports RTM, to
9275 	 * demonstrate that the current volume 3 of the SDM
9276 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
9277 	 * qualification for debug exceptions is not reserved. It is
9278 	 * set to 1 if a debug exception (#DB) or a breakpoint
9279 	 * exception (#BP) occurs inside an RTM region while advanced
9280 	 * debugging of RTM transactional regions is enabled.
9281 	 */
9282 	if (this_cpu_has(X86_FEATURE_RTM)) {
9283 		vmcs_write(ENT_CONTROLS,
9284 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
9285 		/*
9286 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
9287 		 * in the guest to enable advanced debugging of RTM
9288 		 * transactional regions.
9289 		 */
9290 		vmcs_write(GUEST_DR7, BIT(11));
9291 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
9292 		single_step_guest("Hardware delivered single-step in "
9293 				  "transactional region", starting_dr6, 0);
9294 		check_db_exit(false, false, false, &xbegin, BIT(16),
9295 			      starting_dr6);
9296 	} else {
9297 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
9298 		enter_guest();
9299 	}
9300 }
9301 
9302 static void enable_vid(void)
9303 {
9304 	void *virtual_apic_page;
9305 
9306 	assert(cpu_has_apicv());
9307 
9308 	disable_intercept_for_x2apic_msrs();
9309 
9310 	virtual_apic_page = alloc_page();
9311 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
9312 
9313 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
9314 
9315 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
9316 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
9317 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
9318 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
9319 
9320 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
9321 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
9322 }
9323 
9324 static void trigger_ioapic_scan_thread(void *data)
9325 {
9326 	/* Wait until other CPU entered L2 */
9327 	while (vmx_get_test_stage() != 1)
9328 		;
9329 
9330 	/* Trigger ioapic scan */
9331 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
9332 	vmx_set_test_stage(2);
9333 }
9334 
9335 static void irq_79_handler_guest(isr_regs_t *regs)
9336 {
9337 	eoi();
9338 
9339 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
9340 	vmcall();
9341 }
9342 
9343 /*
9344  * Constant for num of busy-loop iterations after which
9345  * a timer interrupt should have happened in host
9346  */
9347 #define TIMER_INTERRUPT_DELAY 100000000
9348 
9349 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
9350 {
9351 	handle_irq(0x79, irq_79_handler_guest);
9352 	sti();
9353 
9354 	/* Signal to L1 CPU to trigger ioapic scan */
9355 	vmx_set_test_stage(1);
9356 	/* Wait until L1 CPU to trigger ioapic scan */
9357 	while (vmx_get_test_stage() != 2)
9358 		;
9359 
9360 	/*
9361 	 * Wait for L0 timer interrupt to be raised while we run in L2
9362 	 * such that L0 will process the IOAPIC scan request before
9363 	 * resuming L2
9364 	 */
9365 	delay(TIMER_INTERRUPT_DELAY);
9366 
9367 	asm volatile ("int $0x79");
9368 }
9369 
9370 static void vmx_eoi_bitmap_ioapic_scan_test(void)
9371 {
9372 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
9373 		report_skip("%s : Not all required APICv bits supported or CPU count < 2", __func__);
9374 		return;
9375 	}
9376 
9377 	enable_vid();
9378 
9379 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
9380 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
9381 
9382 	/*
9383 	 * Launch L2.
9384 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
9385 	 * In case the reason isn't VMX_VMCALL, the assertion inside
9386 	 * skip_exit_vmcall() will fail.
9387 	 */
9388 	enter_guest();
9389 	skip_exit_vmcall();
9390 
9391 	/* Let L2 finish */
9392 	enter_guest();
9393 	report_pass(__func__);
9394 }
9395 
9396 #define HLT_WITH_RVI_VECTOR		(0xf1)
9397 
9398 bool vmx_hlt_with_rvi_guest_isr_fired;
9399 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
9400 {
9401 	vmx_hlt_with_rvi_guest_isr_fired = true;
9402 	eoi();
9403 }
9404 
9405 static void vmx_hlt_with_rvi_guest(void)
9406 {
9407 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
9408 
9409 	sti_nop();
9410 	asm volatile ("nop");
9411 
9412 	vmcall();
9413 }
9414 
9415 static void vmx_hlt_with_rvi_test(void)
9416 {
9417 	if (!cpu_has_apicv()) {
9418 		report_skip("%s : Not all required APICv bits supported", __func__);
9419 		return;
9420 	}
9421 
9422 	enable_vid();
9423 
9424 	vmx_hlt_with_rvi_guest_isr_fired = false;
9425 	test_set_guest(vmx_hlt_with_rvi_guest);
9426 
9427 	enter_guest();
9428 	skip_exit_vmcall();
9429 
9430 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
9431 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
9432 	enter_guest();
9433 
9434 	report(vmx_hlt_with_rvi_guest_isr_fired, "Interrupt raised in guest");
9435 }
9436 
9437 static void set_irq_line_thread(void *data)
9438 {
9439 	/* Wait until other CPU entered L2 */
9440 	while (vmx_get_test_stage() != 1)
9441 		;
9442 
9443 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
9444 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9445 	vmx_set_test_stage(2);
9446 }
9447 
9448 static bool irq_78_handler_vmcall_before_eoi;
9449 static void irq_78_handler_guest(isr_regs_t *regs)
9450 {
9451 	set_irq_line(0xf, 0);
9452 	if (irq_78_handler_vmcall_before_eoi)
9453 		vmcall();
9454 	eoi();
9455 	vmcall();
9456 }
9457 
9458 static void vmx_apic_passthrough_guest(void)
9459 {
9460 	handle_irq(0x78, irq_78_handler_guest);
9461 	sti();
9462 
9463 	/* If requested, wait for other CPU to trigger ioapic scan */
9464 	if (vmx_get_test_stage() < 1) {
9465 		vmx_set_test_stage(1);
9466 		while (vmx_get_test_stage() != 2)
9467 			;
9468 	}
9469 
9470 	set_irq_line(0xf, 1);
9471 }
9472 
9473 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
9474 {
9475 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
9476 		report_skip("%s : CPU count < 2", __func__);
9477 		return;
9478 	}
9479 
9480 	/* Test device is required for generating IRQs */
9481 	if (!test_device_enabled()) {
9482 		report_skip("%s : No test device enabled", __func__);
9483 		return;
9484 	}
9485 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9486 	u64 cpu_ctrl_1 = 0;
9487 
9488 	disable_intercept_for_x2apic_msrs();
9489 
9490 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9491 
9492 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9493 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9494 
9495 	if (set_irq_line_from_thread) {
9496 		irq_78_handler_vmcall_before_eoi = false;
9497 		on_cpu_async(1, set_irq_line_thread, NULL);
9498 	} else {
9499 		irq_78_handler_vmcall_before_eoi = true;
9500 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9501 		vmx_set_test_stage(2);
9502 	}
9503 	test_set_guest(vmx_apic_passthrough_guest);
9504 
9505 	if (irq_78_handler_vmcall_before_eoi) {
9506 		/* Before EOI remote_irr should still be set */
9507 		enter_guest();
9508 		skip_exit_vmcall();
9509 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
9510 			"IOAPIC pass-through: remote_irr=1 before EOI");
9511 	}
9512 
9513 	/* After EOI remote_irr should be cleared */
9514 	enter_guest();
9515 	skip_exit_vmcall();
9516 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
9517 		"IOAPIC pass-through: remote_irr=0 after EOI");
9518 
9519 	/* Let L2 finish */
9520 	enter_guest();
9521 	report_pass(__func__);
9522 }
9523 
9524 static void vmx_apic_passthrough_test(void)
9525 {
9526 	vmx_apic_passthrough(false);
9527 }
9528 
9529 static void vmx_apic_passthrough_thread_test(void)
9530 {
9531 	vmx_apic_passthrough(true);
9532 }
9533 
9534 static void vmx_apic_passthrough_tpr_threshold_guest(void)
9535 {
9536 	cli();
9537 	apic_set_tpr(0);
9538 }
9539 
9540 static bool vmx_apic_passthrough_tpr_threshold_ipi_isr_fired;
9541 static void vmx_apic_passthrough_tpr_threshold_ipi_isr(isr_regs_t *regs)
9542 {
9543 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = true;
9544 	eoi();
9545 }
9546 
9547 static void vmx_apic_passthrough_tpr_threshold_test(void)
9548 {
9549 	int ipi_vector = 0xe1;
9550 
9551 	disable_intercept_for_x2apic_msrs();
9552 	vmcs_clear_bits(PIN_CONTROLS, PIN_EXTINT);
9553 
9554 	/* Raise L0 TPR-threshold by queueing vector in LAPIC IRR */
9555 	cli();
9556 	apic_set_tpr((ipi_vector >> 4) + 1);
9557 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
9558 			APIC_DM_FIXED | ipi_vector,
9559 			0);
9560 
9561 	test_set_guest(vmx_apic_passthrough_tpr_threshold_guest);
9562 	enter_guest();
9563 
9564 	report(apic_get_tpr() == 0, "TPR was zero by guest");
9565 
9566 	/* Clean pending self-IPI */
9567 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = false;
9568 	handle_irq(ipi_vector, vmx_apic_passthrough_tpr_threshold_ipi_isr);
9569 	sti_nop();
9570 	report(vmx_apic_passthrough_tpr_threshold_ipi_isr_fired, "self-IPI fired");
9571 
9572 	report_pass(__func__);
9573 }
9574 
9575 static u64 init_signal_test_exit_reason;
9576 static bool init_signal_test_thread_continued;
9577 
9578 static void init_signal_test_thread(void *data)
9579 {
9580 	struct vmcs *test_vmcs = data;
9581 
9582 	/* Enter VMX operation (i.e. exec VMXON) */
9583 	u64 *ap_vmxon_region = alloc_page();
9584 	enable_vmx();
9585 	init_vmx(ap_vmxon_region);
9586 	TEST_ASSERT(!__vmxon_safe(ap_vmxon_region));
9587 
9588 	/* Signal CPU have entered VMX operation */
9589 	vmx_set_test_stage(1);
9590 
9591 	/* Wait for BSP CPU to send INIT signal */
9592 	while (vmx_get_test_stage() != 2)
9593 		;
9594 
9595 	/*
9596 	 * Signal that we continue as usual as INIT signal
9597 	 * should be blocked while CPU is in VMX operation
9598 	 */
9599 	vmx_set_test_stage(3);
9600 
9601 	/* Wait for signal to enter VMX non-root mode */
9602 	while (vmx_get_test_stage() != 4)
9603 		;
9604 
9605 	/* Enter VMX non-root mode */
9606 	test_set_guest(v2_null_test_guest);
9607 	make_vmcs_current(test_vmcs);
9608 	enter_guest();
9609 	/* Save exit reason for BSP CPU to compare to expected result */
9610 	init_signal_test_exit_reason = vmcs_read(EXI_REASON);
9611 	/* VMCLEAR test-vmcs so it could be loaded by BSP CPU */
9612 	vmcs_clear(test_vmcs);
9613 	launched = false;
9614 	/* Signal that CPU exited to VMX root mode */
9615 	vmx_set_test_stage(5);
9616 
9617 	/* Wait for BSP CPU to signal to exit VMX operation */
9618 	while (vmx_get_test_stage() != 6)
9619 		;
9620 
9621 	/* Exit VMX operation (i.e. exec VMXOFF) */
9622 	vmx_off();
9623 
9624 	/*
9625 	 * Signal to BSP CPU that we continue as usual as INIT signal
9626 	 * should have been consumed by VMX_INIT exit from guest
9627 	 */
9628 	vmx_set_test_stage(7);
9629 
9630 	/* Wait for BSP CPU to signal to enter VMX operation */
9631 	while (vmx_get_test_stage() != 8)
9632 		;
9633 	/* Enter VMX operation (i.e. exec VMXON) */
9634 	TEST_ASSERT(!__vmxon_safe(ap_vmxon_region));
9635 	/* Signal to BSP we are in VMX operation */
9636 	vmx_set_test_stage(9);
9637 
9638 	/* Wait for BSP CPU to send INIT signal */
9639 	while (vmx_get_test_stage() != 10)
9640 		;
9641 
9642 	/* Exit VMX operation (i.e. exec VMXOFF) */
9643 	vmx_off();
9644 
9645 	/*
9646 	 * Exiting VMX operation should result in latched
9647 	 * INIT signal being processed. Therefore, we should
9648 	 * never reach the below code. Thus, signal to BSP
9649 	 * CPU if we have reached here so it is able to
9650 	 * report an issue if it happens.
9651 	 */
9652 	init_signal_test_thread_continued = true;
9653 }
9654 
9655 #define INIT_SIGNAL_TEST_DELAY	100000000ULL
9656 
9657 static void vmx_init_signal_test(void)
9658 {
9659 	struct vmcs *test_vmcs;
9660 
9661 	if (cpu_count() < 2) {
9662 		report_skip("%s : CPU count < 2", __func__);
9663 		return;
9664 	}
9665 
9666 	/* VMCLEAR test-vmcs so it could be loaded by other CPU */
9667 	vmcs_save(&test_vmcs);
9668 	vmcs_clear(test_vmcs);
9669 
9670 	vmx_set_test_stage(0);
9671 	on_cpu_async(1, init_signal_test_thread, test_vmcs);
9672 
9673 	/* Wait for other CPU to enter VMX operation */
9674 	while (vmx_get_test_stage() != 1)
9675 		;
9676 
9677 	/* Send INIT signal to other CPU */
9678 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9679 				   id_map[1]);
9680 	/* Signal other CPU we have sent INIT signal */
9681 	vmx_set_test_stage(2);
9682 
9683 	/*
9684 	 * Wait reasonable amount of time for INIT signal to
9685 	 * be received on other CPU and verify that other CPU
9686 	 * have proceed as usual to next test stage as INIT
9687 	 * signal should be blocked while other CPU in
9688 	 * VMX operation
9689 	 */
9690 	delay(INIT_SIGNAL_TEST_DELAY);
9691 	report(vmx_get_test_stage() == 3,
9692 	       "INIT signal blocked when CPU in VMX operation");
9693 	/* No point to continue if we failed at this point */
9694 	if (vmx_get_test_stage() != 3)
9695 		return;
9696 
9697 	/* Signal other CPU to enter VMX non-root mode */
9698 	init_signal_test_exit_reason = -1ull;
9699 	vmx_set_test_stage(4);
9700 	/*
9701 	 * Wait reasonable amount of time for other CPU
9702 	 * to exit to VMX root mode
9703 	 */
9704 	delay(INIT_SIGNAL_TEST_DELAY);
9705 	if (vmx_get_test_stage() != 5) {
9706 		report_fail("Pending INIT signal didn't result in VMX exit");
9707 		return;
9708 	}
9709 	report(init_signal_test_exit_reason == VMX_INIT,
9710 			"INIT signal during VMX non-root mode result in exit-reason %s (%lu)",
9711 			exit_reason_description(init_signal_test_exit_reason),
9712 			init_signal_test_exit_reason);
9713 
9714 	/* Run guest to completion */
9715 	make_vmcs_current(test_vmcs);
9716 	enter_guest();
9717 
9718 	/* Signal other CPU to exit VMX operation */
9719 	init_signal_test_thread_continued = false;
9720 	vmx_set_test_stage(6);
9721 
9722 	/* Wait reasonable amount of time for other CPU to exit VMX operation */
9723 	delay(INIT_SIGNAL_TEST_DELAY);
9724 	report(vmx_get_test_stage() == 7,
9725 	       "INIT signal consumed on VMX_INIT exit");
9726 	/* No point to continue if we failed at this point */
9727 	if (vmx_get_test_stage() != 7)
9728 		return;
9729 
9730 	/* Signal other CPU to enter VMX operation */
9731 	vmx_set_test_stage(8);
9732 	/* Wait for other CPU to enter VMX operation */
9733 	while (vmx_get_test_stage() != 9)
9734 		;
9735 
9736 	/* Send INIT signal to other CPU */
9737 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9738 				   id_map[1]);
9739 	/* Signal other CPU we have sent INIT signal */
9740 	vmx_set_test_stage(10);
9741 
9742 	/*
9743 	 * Wait reasonable amount of time for other CPU
9744 	 * to exit VMX operation and process INIT signal
9745 	 */
9746 	delay(INIT_SIGNAL_TEST_DELAY);
9747 	report(!init_signal_test_thread_continued,
9748 	       "INIT signal processed after exit VMX operation");
9749 
9750 	/*
9751 	 * TODO: Send SIPI to other CPU to sipi_entry (See x86/cstart64.S)
9752 	 * to re-init it to kvm-unit-tests standard environment.
9753 	 * Somehow (?) verify that SIPI was indeed received.
9754 	 */
9755 }
9756 
9757 #define SIPI_SIGNAL_TEST_DELAY	100000000ULL
9758 
9759 static void vmx_sipi_test_guest(void)
9760 {
9761 	if (apic_id() == 0) {
9762 		/* wait AP enter guest with activity=WAIT_SIPI */
9763 		while (vmx_get_test_stage() != 1)
9764 			;
9765 		delay(SIPI_SIGNAL_TEST_DELAY);
9766 
9767 		/* First SIPI signal */
9768 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9769 		report_pass("BSP(L2): Send first SIPI to cpu[%d]", id_map[1]);
9770 
9771 		/* wait AP enter guest */
9772 		while (vmx_get_test_stage() != 2)
9773 			;
9774 		delay(SIPI_SIGNAL_TEST_DELAY);
9775 
9776 		/* Second SIPI signal should be ignored since AP is not in WAIT_SIPI state */
9777 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9778 		report_pass("BSP(L2): Send second SIPI to cpu[%d]", id_map[1]);
9779 
9780 		/* Delay a while to check whether second SIPI would cause VMExit */
9781 		delay(SIPI_SIGNAL_TEST_DELAY);
9782 
9783 		/* Test is done, notify AP to exit test */
9784 		vmx_set_test_stage(3);
9785 
9786 		/* wait AP exit non-root mode */
9787 		while (vmx_get_test_stage() != 5)
9788 			;
9789 	} else {
9790 		/* wait BSP notify test is done */
9791 		while (vmx_get_test_stage() != 3)
9792 			;
9793 
9794 		/* AP exit guest */
9795 		vmx_set_test_stage(4);
9796 	}
9797 }
9798 
9799 static void sipi_test_ap_thread(void *data)
9800 {
9801 	struct vmcs *ap_vmcs;
9802 	u64 *ap_vmxon_region;
9803 	void *ap_stack, *ap_syscall_stack;
9804 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9805 	u64 cpu_ctrl_1 = 0;
9806 
9807 	/* Enter VMX operation (i.e. exec VMXON) */
9808 	ap_vmxon_region = alloc_page();
9809 	enable_vmx();
9810 	init_vmx(ap_vmxon_region);
9811 	TEST_ASSERT(!__vmxon_safe(ap_vmxon_region));
9812 	init_vmcs(&ap_vmcs);
9813 	make_vmcs_current(ap_vmcs);
9814 
9815 	/* Set stack for AP */
9816 	ap_stack = alloc_page();
9817 	ap_syscall_stack = alloc_page();
9818 	vmcs_write(GUEST_RSP, (u64)(ap_stack + PAGE_SIZE - 1));
9819 	vmcs_write(GUEST_SYSENTER_ESP, (u64)(ap_syscall_stack + PAGE_SIZE - 1));
9820 
9821 	/* passthrough lapic to L2 */
9822 	disable_intercept_for_x2apic_msrs();
9823 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9824 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9825 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9826 
9827 	/* Set guest activity state to wait-for-SIPI state */
9828 	vmcs_write(GUEST_ACTV_STATE, ACTV_WAIT_SIPI);
9829 
9830 	vmx_set_test_stage(1);
9831 
9832 	/* AP enter guest */
9833 	enter_guest();
9834 
9835 	if (vmcs_read(EXI_REASON) == VMX_SIPI) {
9836 		report_pass("AP: Handle SIPI VMExit");
9837 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
9838 		vmx_set_test_stage(2);
9839 	} else {
9840 		report_fail("AP: Unexpected VMExit, reason=%ld", vmcs_read(EXI_REASON));
9841 		vmx_off();
9842 		return;
9843 	}
9844 
9845 	/* AP enter guest */
9846 	enter_guest();
9847 
9848 	report(vmcs_read(EXI_REASON) != VMX_SIPI,
9849 		"AP: should no SIPI VMExit since activity is not in WAIT_SIPI state");
9850 
9851 	/* notify BSP that AP is already exit from non-root mode */
9852 	vmx_set_test_stage(5);
9853 
9854 	/* Leave VMX operation */
9855 	vmx_off();
9856 }
9857 
9858 static void vmx_sipi_signal_test(void)
9859 {
9860 	if (!(rdmsr(MSR_IA32_VMX_MISC) & MSR_IA32_VMX_MISC_ACTIVITY_WAIT_SIPI)) {
9861 		report_skip("%s : \"ACTIVITY_WAIT_SIPI state\" not supported", __func__);
9862 		return;
9863 	}
9864 
9865 	if (cpu_count() < 2) {
9866 		report_skip("%s : CPU count < 2", __func__);
9867 		return;
9868 	}
9869 
9870 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9871 	u64 cpu_ctrl_1 = 0;
9872 
9873 	/* passthrough lapic to L2 */
9874 	disable_intercept_for_x2apic_msrs();
9875 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9876 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9877 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9878 
9879 	test_set_guest(vmx_sipi_test_guest);
9880 
9881 	/* update CR3 on AP */
9882 	on_cpu(1, update_cr3, (void *)read_cr3());
9883 
9884 	/* start AP */
9885 	on_cpu_async(1, sipi_test_ap_thread, NULL);
9886 
9887 	vmx_set_test_stage(0);
9888 
9889 	/* BSP enter guest */
9890 	enter_guest();
9891 }
9892 
9893 
9894 enum vmcs_access {
9895 	ACCESS_VMREAD,
9896 	ACCESS_VMWRITE,
9897 	ACCESS_NONE,
9898 };
9899 
9900 struct vmcs_shadow_test_common {
9901 	enum vmcs_access op;
9902 	enum Reason reason;
9903 	u64 field;
9904 	u64 value;
9905 	u64 flags;
9906 	u64 time;
9907 } l1_l2_common;
9908 
9909 static inline u64 vmread_flags(u64 field, u64 *val)
9910 {
9911 	u64 flags;
9912 
9913 	asm volatile ("vmread %2, %1; pushf; pop %0"
9914 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
9915 	return flags & X86_EFLAGS_ALU;
9916 }
9917 
9918 static inline u64 vmwrite_flags(u64 field, u64 val)
9919 {
9920 	u64 flags;
9921 
9922 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
9923 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
9924 	return flags & X86_EFLAGS_ALU;
9925 }
9926 
9927 static void vmx_vmcs_shadow_test_guest(void)
9928 {
9929 	struct vmcs_shadow_test_common *c = &l1_l2_common;
9930 	u64 start;
9931 
9932 	while (c->op != ACCESS_NONE) {
9933 		start = rdtsc();
9934 		switch (c->op) {
9935 		default:
9936 			c->flags = -1ull;
9937 			break;
9938 		case ACCESS_VMREAD:
9939 			c->flags = vmread_flags(c->field, &c->value);
9940 			break;
9941 		case ACCESS_VMWRITE:
9942 			c->flags = vmwrite_flags(c->field, 0);
9943 			break;
9944 		}
9945 		c->time = rdtsc() - start;
9946 		vmcall();
9947 	}
9948 }
9949 
9950 static u64 vmread_from_shadow(u64 field)
9951 {
9952 	struct vmcs *primary;
9953 	struct vmcs *shadow;
9954 	u64 value;
9955 
9956 	TEST_ASSERT(!vmcs_save(&primary));
9957 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
9958 	TEST_ASSERT(!make_vmcs_current(shadow));
9959 	value = vmcs_read(field);
9960 	TEST_ASSERT(!make_vmcs_current(primary));
9961 	return value;
9962 }
9963 
9964 static u64 vmwrite_to_shadow(u64 field, u64 value)
9965 {
9966 	struct vmcs *primary;
9967 	struct vmcs *shadow;
9968 
9969 	TEST_ASSERT(!vmcs_save(&primary));
9970 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
9971 	TEST_ASSERT(!make_vmcs_current(shadow));
9972 	vmcs_write(field, value);
9973 	value = vmcs_read(field);
9974 	TEST_ASSERT(!make_vmcs_current(primary));
9975 	return value;
9976 }
9977 
9978 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
9979 {
9980 	struct vmcs_shadow_test_common *c = &l1_l2_common;
9981 
9982 	c->op = access;
9983 	vmcs_write(VMX_INST_ERROR, 0);
9984 	enter_guest();
9985 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
9986 	if (c->reason != VMX_VMCALL) {
9987 		skip_exit_insn();
9988 		enter_guest();
9989 	}
9990 	skip_exit_vmcall();
9991 }
9992 
9993 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
9994 {
9995 	struct vmcs_shadow_test_common *c = &l1_l2_common;
9996 	struct vmcs *shadow;
9997 	u64 value;
9998 	uintptr_t flags[2];
9999 	bool good_shadow;
10000 	u32 vmx_inst_error;
10001 
10002 	report_prefix_pushf("field %lx", field);
10003 	c->field = field;
10004 
10005 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10006 	if (shadow != (struct vmcs *)-1ull) {
10007 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
10008 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
10009 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
10010 	} else {
10011 		/*
10012 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
10013 		 * shadowed-fields should fail with setting RFLAGS.CF.
10014 		 */
10015 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
10016 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
10017 		good_shadow = false;
10018 	}
10019 
10020 	/* Intercept both VMREAD and VMWRITE. */
10021 	report_prefix_push("no VMREAD/VMWRITE permission");
10022 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
10023 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
10024 		set_bit(field, bitmap[ACCESS_VMREAD]);
10025 		set_bit(field, bitmap[ACCESS_VMWRITE]);
10026 	}
10027 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10028 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10029 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10030 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10031 	report_prefix_pop();
10032 
10033 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
10034 		goto out;
10035 
10036 	/* Permit shadowed VMREAD. */
10037 	report_prefix_push("VMREAD permission only");
10038 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10039 	set_bit(field, bitmap[ACCESS_VMWRITE]);
10040 	if (good_shadow)
10041 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10042 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10043 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10044 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10045 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10046 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10047 	       c->time);
10048 	report(c->flags == flags[ACCESS_VMREAD],
10049 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10050 	       c->flags, flags[ACCESS_VMREAD]);
10051 	if (good_shadow)
10052 		report(c->value == value,
10053 		       "value read from shadow (%lx) is as expected (%lx)",
10054 		       c->value, value);
10055 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10056 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10057 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10058 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10059 	report_prefix_pop();
10060 
10061 	/* Permit shadowed VMWRITE. */
10062 	report_prefix_push("VMWRITE permission only");
10063 	set_bit(field, bitmap[ACCESS_VMREAD]);
10064 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10065 	if (good_shadow)
10066 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10067 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10068 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10069 	report(c->reason == VMX_VMCALL,
10070 		"shadowed for VMWRITE (in %ld cycles)",
10071 		c->time);
10072 	report(c->flags == flags[ACCESS_VMREAD],
10073 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10074 	       c->flags, flags[ACCESS_VMREAD]);
10075 	if (good_shadow) {
10076 		value = vmread_from_shadow(field);
10077 		report(value == 0,
10078 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10079 		       0ul);
10080 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10081 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10082 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10083 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10084 	}
10085 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10086 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10087 	report_prefix_pop();
10088 
10089 	/* Permit shadowed VMREAD and VMWRITE. */
10090 	report_prefix_push("VMREAD and VMWRITE permission");
10091 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10092 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10093 	if (good_shadow)
10094 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10095 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10096 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10097 	report(c->reason == VMX_VMCALL,
10098 		"shadowed for VMWRITE (in %ld cycles)",
10099 		c->time);
10100 	report(c->flags == flags[ACCESS_VMREAD],
10101 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10102 	       c->flags, flags[ACCESS_VMREAD]);
10103 	if (good_shadow) {
10104 		value = vmread_from_shadow(field);
10105 		report(value == 0,
10106 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10107 		       0ul);
10108 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10109 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10110 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10111 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10112 	}
10113 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10114 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10115 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10116 	       c->time);
10117 	report(c->flags == flags[ACCESS_VMREAD],
10118 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10119 	       c->flags, flags[ACCESS_VMREAD]);
10120 	if (good_shadow)
10121 		report(c->value == 0,
10122 		       "value read from shadow (%lx) is as expected (%lx)",
10123 		       c->value, 0ul);
10124 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10125 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10126 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10127 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10128 	report_prefix_pop();
10129 
10130 out:
10131 	report_prefix_pop();
10132 }
10133 
10134 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
10135 {
10136 	unsigned base;
10137 	unsigned index;
10138 	unsigned bit;
10139 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
10140 
10141 	/* Run test on all possible valid VMCS fields */
10142 	for (base = 0;
10143 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
10144 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
10145 		for (index = 0; index <= highest_index; index++)
10146 			vmcs_shadow_test_field(bitmap, base + index);
10147 
10148 	/*
10149 	 * Run tests on some invalid VMCS fields
10150 	 * (Have reserved bit set).
10151 	 */
10152 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
10153 		vmcs_shadow_test_field(bitmap, (1ull << bit));
10154 }
10155 
10156 static void vmx_vmcs_shadow_test(void)
10157 {
10158 	u8 *bitmap[2];
10159 	struct vmcs *shadow;
10160 
10161 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
10162 		report_skip("%s : \"Activate secondary controls\" not supported", __func__);
10163 		return;
10164 	}
10165 
10166 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
10167 		report_skip("%s : \"VMCS shadowing\" not supported", __func__);
10168 		return;
10169 	}
10170 
10171 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
10172 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
10173 		report_skip("%s : VMWRITE can't modify VM-exit information fields.", __func__);
10174 		return;
10175 	}
10176 
10177 	test_set_guest(vmx_vmcs_shadow_test_guest);
10178 
10179 	bitmap[ACCESS_VMREAD] = alloc_page();
10180 	bitmap[ACCESS_VMWRITE] = alloc_page();
10181 
10182 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
10183 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
10184 
10185 	shadow = alloc_page();
10186 	shadow->hdr.revision_id = basic.revision;
10187 	shadow->hdr.shadow_vmcs = 1;
10188 	TEST_ASSERT(!vmcs_clear(shadow));
10189 
10190 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
10191 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
10192 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
10193 
10194 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
10195 	report_prefix_push("valid link pointer");
10196 	vmx_vmcs_shadow_test_body(bitmap);
10197 	report_prefix_pop();
10198 
10199 	vmcs_write(VMCS_LINK_PTR, -1ull);
10200 	report_prefix_push("invalid link pointer");
10201 	vmx_vmcs_shadow_test_body(bitmap);
10202 	report_prefix_pop();
10203 
10204 	l1_l2_common.op = ACCESS_NONE;
10205 	enter_guest();
10206 }
10207 
10208 /*
10209  * This test monitors the difference between a guest RDTSC instruction
10210  * and the IA32_TIME_STAMP_COUNTER MSR value stored in the VMCS12
10211  * VM-exit MSR-store list when taking a VM-exit on the instruction
10212  * following RDTSC.
10213  */
10214 #define RDTSC_DIFF_ITERS 100000
10215 #define RDTSC_DIFF_FAILS 100
10216 #define HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD 750
10217 
10218 /*
10219  * Set 'use TSC offsetting' and set the guest offset to the
10220  * inverse of the host's current TSC value, so that the guest starts running
10221  * with an effective TSC value of 0.
10222  */
10223 static void reset_guest_tsc_to_zero(void)
10224 {
10225 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
10226 	vmcs_write(TSC_OFFSET, -rdtsc());
10227 }
10228 
10229 static void rdtsc_vmexit_diff_test_guest(void)
10230 {
10231 	int i;
10232 
10233 	for (i = 0; i < RDTSC_DIFF_ITERS; i++)
10234 		/* Ensure rdtsc is the last instruction before the vmcall. */
10235 		asm volatile("rdtsc; vmcall" : : : "eax", "edx");
10236 }
10237 
10238 /*
10239  * This function only considers the "use TSC offsetting" VM-execution
10240  * control.  It does not handle "use TSC scaling" (because the latter
10241  * isn't available to the host today.)
10242  */
10243 static unsigned long long host_time_to_guest_time(unsigned long long t)
10244 {
10245 	TEST_ASSERT(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
10246 		    !(vmcs_read(CPU_EXEC_CTRL1) & CPU_USE_TSC_SCALING));
10247 
10248 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_USE_TSC_OFFSET)
10249 		t += vmcs_read(TSC_OFFSET);
10250 
10251 	return t;
10252 }
10253 
10254 static unsigned long long rdtsc_vmexit_diff_test_iteration(void)
10255 {
10256 	unsigned long long guest_tsc, host_to_guest_tsc;
10257 
10258 	enter_guest();
10259 	skip_exit_vmcall();
10260 	guest_tsc = (u32) regs.rax + (regs.rdx << 32);
10261 	host_to_guest_tsc = host_time_to_guest_time(exit_msr_store[0].value);
10262 
10263 	return host_to_guest_tsc - guest_tsc;
10264 }
10265 
10266 static void rdtsc_vmexit_diff_test(void)
10267 {
10268 	unsigned long long delta;
10269 	int fail = 0;
10270 	int i;
10271 
10272 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET))
10273 		test_skip("CPU doesn't support the 'use TSC offsetting' processor-based VM-execution control.\n");
10274 
10275 	test_set_guest(rdtsc_vmexit_diff_test_guest);
10276 
10277 	reset_guest_tsc_to_zero();
10278 
10279 	/*
10280 	 * Set up the VMCS12 VM-exit MSR-store list to store just one
10281 	 * MSR: IA32_TIME_STAMP_COUNTER. Note that the value stored is
10282 	 * in the host time domain (i.e., it is not adjusted according
10283 	 * to the TSC multiplier and TSC offset fields in the VMCS12,
10284 	 * as a guest RDTSC would be.)
10285 	 */
10286 	exit_msr_store = alloc_page();
10287 	exit_msr_store[0].index = MSR_IA32_TSC;
10288 	vmcs_write(EXI_MSR_ST_CNT, 1);
10289 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(exit_msr_store));
10290 
10291 	for (i = 0; i < RDTSC_DIFF_ITERS && fail < RDTSC_DIFF_FAILS; i++) {
10292 		delta = rdtsc_vmexit_diff_test_iteration();
10293 		if (delta >= HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD)
10294 			fail++;
10295 	}
10296 
10297 	enter_guest();
10298 
10299 	report(fail < RDTSC_DIFF_FAILS,
10300 	       "RDTSC to VM-exit delta too high in %d of %d iterations, last = %llu",
10301 	       fail, i, delta);
10302 }
10303 
10304 static int invalid_msr_init(struct vmcs *vmcs)
10305 {
10306 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
10307 		printf("\tPreemption timer is not supported\n");
10308 		return VMX_TEST_EXIT;
10309 	}
10310 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
10311 	preempt_val = 10000000;
10312 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
10313 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
10314 
10315 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
10316 		printf("\tSave preemption value is not supported\n");
10317 
10318 	vmcs_write(ENT_MSR_LD_CNT, 1);
10319 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
10320 
10321 	return VMX_TEST_START;
10322 }
10323 
10324 
10325 static void invalid_msr_main(void)
10326 {
10327 	report_fail("Invalid MSR load");
10328 }
10329 
10330 static int invalid_msr_exit_handler(union exit_reason exit_reason)
10331 {
10332 	report_fail("Invalid MSR load");
10333 	print_vmexit_info(exit_reason);
10334 	return VMX_TEST_EXIT;
10335 }
10336 
10337 static int invalid_msr_entry_failure(struct vmentry_result *result)
10338 {
10339 	report(result->exit_reason.failed_vmentry &&
10340 	       result->exit_reason.basic == VMX_FAIL_MSR, "Invalid MSR load");
10341 	return VMX_TEST_VMEXIT;
10342 }
10343 
10344 /*
10345  * The max number of MSRs in an atomic switch MSR list is:
10346  * (111B + 1) * 512 = 4096
10347  *
10348  * Each list entry consumes:
10349  * 4-byte MSR index + 4 bytes reserved + 8-byte data = 16 bytes
10350  *
10351  * Allocate 128 kB to cover max_msr_list_size (i.e., 64 kB) and then some.
10352  */
10353 static const u32 msr_list_page_order = 5;
10354 
10355 static void atomic_switch_msr_limit_test_guest(void)
10356 {
10357 	vmcall();
10358 }
10359 
10360 static void populate_msr_list(struct vmx_msr_entry *msr_list,
10361 			      size_t byte_capacity, int count)
10362 {
10363 	int i;
10364 
10365 	for (i = 0; i < count; i++) {
10366 		msr_list[i].index = MSR_IA32_TSC;
10367 		msr_list[i].reserved = 0;
10368 		msr_list[i].value = 0x1234567890abcdef;
10369 	}
10370 
10371 	memset(msr_list + count, 0xff,
10372 	       byte_capacity - count * sizeof(*msr_list));
10373 }
10374 
10375 static int max_msr_list_size(void)
10376 {
10377 	u32 vmx_misc = rdmsr(MSR_IA32_VMX_MISC);
10378 	u32 factor = ((vmx_misc & GENMASK(27, 25)) >> 25) + 1;
10379 
10380 	return factor * 512;
10381 }
10382 
10383 static void atomic_switch_msrs_test(int count)
10384 {
10385 	struct vmx_msr_entry *vm_enter_load;
10386         struct vmx_msr_entry *vm_exit_load;
10387         struct vmx_msr_entry *vm_exit_store;
10388 	int max_allowed = max_msr_list_size();
10389 	int byte_capacity = 1ul << (msr_list_page_order + PAGE_SHIFT);
10390 	/* Exceeding the max MSR list size at exit triggers KVM to abort. */
10391 	int exit_count = count > max_allowed ? max_allowed : count;
10392 	int cleanup_count = count > max_allowed ? 2 : 1;
10393 	int i;
10394 
10395 	/*
10396 	 * Check for the IA32_TSC MSR,
10397 	 * available with the "TSC flag" and used to populate the MSR lists.
10398 	 */
10399 	if (!(cpuid(1).d & (1 << 4))) {
10400 		report_skip("%s : \"Time Stamp Counter\" not supported", __func__);
10401 		return;
10402 	}
10403 
10404 	/* Set L2 guest. */
10405 	test_set_guest(atomic_switch_msr_limit_test_guest);
10406 
10407 	/* Setup atomic MSR switch lists. */
10408 	vm_enter_load = alloc_pages(msr_list_page_order);
10409 	vm_exit_load = alloc_pages(msr_list_page_order);
10410 	vm_exit_store = alloc_pages(msr_list_page_order);
10411 
10412 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)vm_enter_load);
10413 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)vm_exit_load);
10414 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)vm_exit_store);
10415 
10416 	/*
10417 	 * VM-Enter should succeed up to the max number of MSRs per list, and
10418 	 * should not consume junk beyond the last entry.
10419 	 */
10420 	populate_msr_list(vm_enter_load, byte_capacity, count);
10421 	populate_msr_list(vm_exit_load, byte_capacity, exit_count);
10422 	populate_msr_list(vm_exit_store, byte_capacity, exit_count);
10423 
10424 	vmcs_write(ENT_MSR_LD_CNT, count);
10425 	vmcs_write(EXI_MSR_LD_CNT, exit_count);
10426 	vmcs_write(EXI_MSR_ST_CNT, exit_count);
10427 
10428 	if (count <= max_allowed) {
10429 		enter_guest();
10430 		assert_exit_reason(VMX_VMCALL);
10431 		skip_exit_vmcall();
10432 	} else {
10433 		u32 exit_qual;
10434 
10435 		test_guest_state("Invalid MSR Load Count", true, count,
10436 				 "ENT_MSR_LD_CNT");
10437 
10438 		exit_qual = vmcs_read(EXI_QUALIFICATION);
10439 		report(exit_qual == max_allowed + 1, "exit_qual, %u, is %u.",
10440 		       exit_qual, max_allowed + 1);
10441 	}
10442 
10443 	/* Cleanup. */
10444 	vmcs_write(ENT_MSR_LD_CNT, 0);
10445 	vmcs_write(EXI_MSR_LD_CNT, 0);
10446 	vmcs_write(EXI_MSR_ST_CNT, 0);
10447 	for (i = 0; i < cleanup_count; i++) {
10448 		enter_guest();
10449 		skip_exit_vmcall();
10450 	}
10451 	free_pages_by_order(vm_enter_load, msr_list_page_order);
10452 	free_pages_by_order(vm_exit_load, msr_list_page_order);
10453 	free_pages_by_order(vm_exit_store, msr_list_page_order);
10454 }
10455 
10456 static void atomic_switch_max_msrs_test(void)
10457 {
10458 	atomic_switch_msrs_test(max_msr_list_size());
10459 }
10460 
10461 static void atomic_switch_overflow_msrs_test(void)
10462 {
10463 	if (test_device_enabled())
10464 		atomic_switch_msrs_test(max_msr_list_size() + 1);
10465 	else
10466 		test_skip("Test is only supported on KVM");
10467 }
10468 
10469 static void vmx_pf_exception_test_guest(void)
10470 {
10471 	ac_test_run(PT_LEVEL_PML4, false);
10472 }
10473 
10474 static void vmx_pf_exception_forced_emulation_test_guest(void)
10475 {
10476 	ac_test_run(PT_LEVEL_PML4, true);
10477 }
10478 
10479 typedef void (*invalidate_tlb_t)(void *data);
10480 typedef void (*pf_exception_test_guest_t)(void);
10481 
10482 
10483 static void __vmx_pf_exception_test(invalidate_tlb_t inv_fn, void *data,
10484 				    pf_exception_test_guest_t guest_fn)
10485 {
10486 	u64 efer;
10487 	struct cpuid cpuid;
10488 
10489 	test_set_guest(guest_fn);
10490 
10491 	/* Intercept INVLPG when to perform TLB invalidation from L1 (this). */
10492 	if (inv_fn)
10493 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INVLPG);
10494 	else
10495 		vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INVLPG);
10496 
10497 	enter_guest();
10498 
10499 	while (vmcs_read(EXI_REASON) != VMX_VMCALL) {
10500 		switch (vmcs_read(EXI_REASON)) {
10501 		case VMX_RDMSR:
10502 			assert(regs.rcx == MSR_EFER);
10503 			efer = vmcs_read(GUEST_EFER);
10504 			regs.rdx = efer >> 32;
10505 			regs.rax = efer & 0xffffffff;
10506 			break;
10507 		case VMX_WRMSR:
10508 			assert(regs.rcx == MSR_EFER);
10509 			efer = regs.rdx << 32 | (regs.rax & 0xffffffff);
10510 			vmcs_write(GUEST_EFER, efer);
10511 			break;
10512 		case VMX_CPUID:
10513 			cpuid = (struct cpuid) {0, 0, 0, 0};
10514 			cpuid = raw_cpuid(regs.rax, regs.rcx);
10515 			regs.rax = cpuid.a;
10516 			regs.rbx = cpuid.b;
10517 			regs.rcx = cpuid.c;
10518 			regs.rdx = cpuid.d;
10519 			break;
10520 		case VMX_INVLPG:
10521 			inv_fn(data);
10522 			break;
10523 		default:
10524 			assert_msg(false,
10525 				"Unexpected exit to L1, exit_reason: %s (0x%lx)",
10526 				exit_reason_description(vmcs_read(EXI_REASON)),
10527 				vmcs_read(EXI_REASON));
10528 		}
10529 		skip_exit_insn();
10530 		enter_guest();
10531 	}
10532 
10533 	assert_exit_reason(VMX_VMCALL);
10534 }
10535 
10536 static void vmx_pf_exception_test(void)
10537 {
10538 	__vmx_pf_exception_test(NULL, NULL, vmx_pf_exception_test_guest);
10539 }
10540 
10541 static void vmx_pf_exception_forced_emulation_test(void)
10542 {
10543 	__vmx_pf_exception_test(NULL, NULL, vmx_pf_exception_forced_emulation_test_guest);
10544 }
10545 
10546 static void invalidate_tlb_no_vpid(void *data)
10547 {
10548 	/* If VPID is disabled, the TLB is flushed on VM-Enter and VM-Exit. */
10549 }
10550 
10551 static void vmx_pf_no_vpid_test(void)
10552 {
10553 	if (is_vpid_supported())
10554 		vmcs_clear_bits(CPU_EXEC_CTRL1, CPU_VPID);
10555 
10556 	__vmx_pf_exception_test(invalidate_tlb_no_vpid, NULL,
10557 				vmx_pf_exception_test_guest);
10558 }
10559 
10560 static void invalidate_tlb_invvpid_addr(void *data)
10561 {
10562 	invvpid(INVVPID_ALL, *(u16 *)data, vmcs_read(EXI_QUALIFICATION));
10563 }
10564 
10565 static void invalidate_tlb_new_vpid(void *data)
10566 {
10567 	u16 *vpid = data;
10568 
10569 	/*
10570 	 * Bump VPID to effectively flush L2's TLB from L0's perspective.
10571 	 * Invalidate all VPIDs when the VPID wraps to zero as hardware/KVM is
10572 	 * architecturally allowed to keep TLB entries indefinitely.
10573 	 */
10574 	++(*vpid);
10575 	if (*vpid == 0) {
10576 		++(*vpid);
10577 		invvpid(INVVPID_ALL, 0, 0);
10578 	}
10579 	vmcs_write(VPID, *vpid);
10580 }
10581 
10582 static void __vmx_pf_vpid_test(invalidate_tlb_t inv_fn, u16 vpid)
10583 {
10584 	if (!is_vpid_supported())
10585 		test_skip("VPID unsupported");
10586 
10587 	if (!is_invvpid_supported())
10588 		test_skip("INVVPID unsupported");
10589 
10590 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
10591 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VPID);
10592 	vmcs_write(VPID, vpid);
10593 
10594 	__vmx_pf_exception_test(inv_fn, &vpid, vmx_pf_exception_test_guest);
10595 }
10596 
10597 static void vmx_pf_invvpid_test(void)
10598 {
10599 	if (!is_invvpid_type_supported(INVVPID_ADDR))
10600 		test_skip("INVVPID ADDR unsupported");
10601 
10602 	__vmx_pf_vpid_test(invalidate_tlb_invvpid_addr, 0xaaaa);
10603 }
10604 
10605 static void vmx_pf_vpid_test(void)
10606 {
10607 	/* Need INVVPID(ALL) to flush VPIDs upon wrap/reuse. */
10608 	if (!is_invvpid_type_supported(INVVPID_ALL))
10609 		test_skip("INVVPID ALL unsupported");
10610 
10611 	__vmx_pf_vpid_test(invalidate_tlb_new_vpid, 1);
10612 }
10613 
10614 static void vmx_l2_ac_test(void)
10615 {
10616 	bool hit_ac = false;
10617 
10618 	write_cr0(read_cr0() | X86_CR0_AM);
10619 	write_rflags(read_rflags() | X86_EFLAGS_AC);
10620 
10621 	run_in_user(generate_usermode_ac, AC_VECTOR, 0, 0, 0, 0, &hit_ac);
10622 	report(hit_ac, "Usermode #AC handled in L2");
10623 	vmcall();
10624 }
10625 
10626 struct vmx_exception_test {
10627 	u8 vector;
10628 	void (*guest_code)(void);
10629 };
10630 
10631 struct vmx_exception_test vmx_exception_tests[] = {
10632 	{ GP_VECTOR, generate_non_canonical_gp },
10633 	{ UD_VECTOR, generate_ud },
10634 	{ DE_VECTOR, generate_de },
10635 	{ DB_VECTOR, generate_single_step_db },
10636 	{ BP_VECTOR, generate_bp },
10637 	{ AC_VECTOR, vmx_l2_ac_test },
10638 	{ OF_VECTOR, generate_of },
10639 	{ NM_VECTOR, generate_cr0_ts_nm },
10640 	{ NM_VECTOR, generate_cr0_em_nm },
10641 };
10642 
10643 static u8 vmx_exception_test_vector;
10644 
10645 static void vmx_exception_handler(struct ex_regs *regs)
10646 {
10647 	report(regs->vector == vmx_exception_test_vector,
10648 	       "Handling %s in L2's exception handler",
10649 	       exception_mnemonic(vmx_exception_test_vector));
10650 	vmcall();
10651 }
10652 
10653 static void handle_exception_in_l2(u8 vector)
10654 {
10655 	handler old_handler = handle_exception(vector, vmx_exception_handler);
10656 
10657 	vmx_exception_test_vector = vector;
10658 
10659 	enter_guest();
10660 	report(vmcs_read(EXI_REASON) == VMX_VMCALL,
10661 	       "%s handled by L2", exception_mnemonic(vector));
10662 
10663 	handle_exception(vector, old_handler);
10664 }
10665 
10666 static void handle_exception_in_l1(u32 vector)
10667 {
10668 	u32 old_eb = vmcs_read(EXC_BITMAP);
10669 	u32 intr_type;
10670 	u32 intr_info;
10671 
10672 	vmcs_write(EXC_BITMAP, old_eb | (1u << vector));
10673 
10674 	enter_guest();
10675 
10676 	if (vector == BP_VECTOR || vector == OF_VECTOR)
10677 		intr_type = VMX_INTR_TYPE_SOFT_EXCEPTION;
10678 	else
10679 		intr_type = VMX_INTR_TYPE_HARD_EXCEPTION;
10680 
10681 	intr_info = vmcs_read(EXI_INTR_INFO);
10682 	report((vmcs_read(EXI_REASON) == VMX_EXC_NMI) &&
10683 	       (intr_info & INTR_INFO_VALID_MASK) &&
10684 	       (intr_info & INTR_INFO_VECTOR_MASK) == vector &&
10685 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >> INTR_INFO_INTR_TYPE_SHIFT) == intr_type,
10686 	       "%s correctly routed to L1", exception_mnemonic(vector));
10687 
10688 	vmcs_write(EXC_BITMAP, old_eb);
10689 }
10690 
10691 static void vmx_exception_test(void)
10692 {
10693 	struct vmx_exception_test *t;
10694 	int i;
10695 
10696 	for (i = 0; i < ARRAY_SIZE(vmx_exception_tests); i++) {
10697 		t = &vmx_exception_tests[i];
10698 
10699 		/*
10700 		 * Override the guest code before each run even though it's the
10701 		 * same code, the VMCS guest state needs to be reinitialized.
10702 		 */
10703 		test_override_guest(t->guest_code);
10704 		handle_exception_in_l2(t->vector);
10705 
10706 		test_override_guest(t->guest_code);
10707 		handle_exception_in_l1(t->vector);
10708 	}
10709 
10710 	test_set_guest_finished();
10711 }
10712 
10713 #define TEST(name) { #name, .v2 = name }
10714 
10715 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
10716 struct vmx_test vmx_tests[] = {
10717 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
10718 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
10719 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
10720 		preemption_timer_exit_handler, NULL, {0} },
10721 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
10722 		test_ctrl_pat_exit_handler, NULL, {0} },
10723 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
10724 		test_ctrl_efer_exit_handler, NULL, {0} },
10725 	{ "CR shadowing", NULL, cr_shadowing_main,
10726 		cr_shadowing_exit_handler, NULL, {0} },
10727 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
10728 		NULL, {0} },
10729 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
10730 		insn_intercept_exit_handler, NULL, {0} },
10731 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
10732 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
10733 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
10734 	{ "interrupt", interrupt_init, interrupt_main,
10735 		interrupt_exit_handler, NULL, {0} },
10736 	{ "nmi_hlt", nmi_hlt_init, nmi_hlt_main,
10737 		nmi_hlt_exit_handler, NULL, {0} },
10738 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
10739 		NULL, {0} },
10740 	{ "MSR switch", msr_switch_init, msr_switch_main,
10741 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
10742 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
10743 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
10744 		disable_rdtscp_exit_handler, NULL, {0} },
10745 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
10746 		exit_monitor_from_l2_handler, NULL, {0} },
10747 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
10748 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
10749 	/* Basic V2 tests. */
10750 	TEST(v2_null_test),
10751 	TEST(v2_multiple_entries_test),
10752 	TEST(fixture_test_case1),
10753 	TEST(fixture_test_case2),
10754 	/* Opcode tests. */
10755 	TEST(invvpid_test),
10756 	/* VM-entry tests */
10757 	TEST(vmx_controls_test),
10758 	TEST(vmx_host_state_area_test),
10759 	TEST(vmx_guest_state_area_test),
10760 	TEST(vmentry_movss_shadow_test),
10761 	TEST(vmentry_unrestricted_guest_test),
10762 	/* APICv tests */
10763 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
10764 	TEST(vmx_hlt_with_rvi_test),
10765 	TEST(apic_reg_virt_test),
10766 	TEST(virt_x2apic_mode_test),
10767 	/* APIC pass-through tests */
10768 	TEST(vmx_apic_passthrough_test),
10769 	TEST(vmx_apic_passthrough_thread_test),
10770 	TEST(vmx_apic_passthrough_tpr_threshold_test),
10771 	TEST(vmx_init_signal_test),
10772 	TEST(vmx_sipi_signal_test),
10773 	/* VMCS Shadowing tests */
10774 	TEST(vmx_vmcs_shadow_test),
10775 	/* Regression tests */
10776 	TEST(vmx_ldtr_test),
10777 	TEST(vmx_cr_load_test),
10778 	TEST(vmx_cr4_osxsave_test),
10779 	TEST(vmx_no_nm_test),
10780 	TEST(vmx_db_test),
10781 	TEST(vmx_nmi_window_test),
10782 	TEST(vmx_intr_window_test),
10783 	TEST(vmx_pending_event_test),
10784 	TEST(vmx_pending_event_hlt_test),
10785 	TEST(vmx_store_tsc_test),
10786 	TEST(vmx_preemption_timer_zero_test),
10787 	TEST(vmx_preemption_timer_tf_test),
10788 	TEST(vmx_preemption_timer_expiry_test),
10789 	/* EPT access tests. */
10790 	TEST(ept_access_test_not_present),
10791 	TEST(ept_access_test_read_only),
10792 	TEST(ept_access_test_write_only),
10793 	TEST(ept_access_test_read_write),
10794 	TEST(ept_access_test_execute_only),
10795 	TEST(ept_access_test_read_execute),
10796 	TEST(ept_access_test_write_execute),
10797 	TEST(ept_access_test_read_write_execute),
10798 	TEST(ept_access_test_reserved_bits),
10799 	TEST(ept_access_test_ignored_bits),
10800 	TEST(ept_access_test_paddr_not_present_ad_disabled),
10801 	TEST(ept_access_test_paddr_not_present_ad_enabled),
10802 	TEST(ept_access_test_paddr_read_only_ad_disabled),
10803 	TEST(ept_access_test_paddr_read_only_ad_enabled),
10804 	TEST(ept_access_test_paddr_read_write),
10805 	TEST(ept_access_test_paddr_read_write_execute),
10806 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
10807 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
10808 	TEST(ept_access_test_paddr_not_present_page_fault),
10809 	TEST(ept_access_test_force_2m_page),
10810 	/* Atomic MSR switch tests. */
10811 	TEST(atomic_switch_max_msrs_test),
10812 	TEST(atomic_switch_overflow_msrs_test),
10813 	TEST(rdtsc_vmexit_diff_test),
10814 	TEST(vmx_mtf_test),
10815 	TEST(vmx_mtf_pdpte_test),
10816 	TEST(vmx_pf_exception_test),
10817 	TEST(vmx_pf_exception_forced_emulation_test),
10818 	TEST(vmx_pf_no_vpid_test),
10819 	TEST(vmx_pf_invvpid_test),
10820 	TEST(vmx_pf_vpid_test),
10821 	TEST(vmx_exception_test),
10822 	{ NULL, NULL, NULL, NULL, NULL, {0} },
10823 };
10824