xref: /kvm-unit-tests/x86/vmx_tests.c (revision f19da7cc04a6e94d6827f74cc54ee8b0f0830b40)
1 /*
2  * All test cases of nested virtualization should be in this file
3  *
4  * Author : Arthur Chunqi Li <yzt356@gmail.com>
5  */
6 
7 #include <asm/debugreg.h>
8 
9 #include "vmx.h"
10 #include "msr.h"
11 #include "processor.h"
12 #include "vm.h"
13 #include "pci.h"
14 #include "fwcfg.h"
15 #include "isr.h"
16 #include "desc.h"
17 #include "apic.h"
18 #include "types.h"
19 #include "vmalloc.h"
20 #include "alloc_page.h"
21 #include "smp.h"
22 #include "delay.h"
23 #include "access.h"
24 
25 #define VPID_CAP_INVVPID_TYPES_SHIFT 40
26 
27 u64 ia32_pat;
28 u64 ia32_efer;
29 void *io_bitmap_a, *io_bitmap_b;
30 u16 ioport;
31 
32 unsigned long *pml4;
33 u64 eptp;
34 void *data_page1, *data_page2;
35 
36 phys_addr_t pci_physaddr;
37 
38 void *pml_log;
39 #define PML_INDEX 512
40 
41 static inline unsigned ffs(unsigned x)
42 {
43 	int pos = -1;
44 
45 	__asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0"
46 			     : "+r"(pos) : "rm"(x) : "eax");
47 	return pos + 1;
48 }
49 
50 static inline void vmcall(void)
51 {
52 	asm volatile("vmcall");
53 }
54 
55 static void basic_guest_main(void)
56 {
57 	report_pass("Basic VMX test");
58 }
59 
60 static int basic_exit_handler(union exit_reason exit_reason)
61 {
62 	report_fail("Basic VMX test");
63 	print_vmexit_info(exit_reason);
64 	return VMX_TEST_EXIT;
65 }
66 
67 static void vmenter_main(void)
68 {
69 	u64 rax;
70 	u64 rsp, resume_rsp;
71 
72 	report_pass("test vmlaunch");
73 
74 	asm volatile(
75 		"mov %%rsp, %0\n\t"
76 		"mov %3, %%rax\n\t"
77 		"vmcall\n\t"
78 		"mov %%rax, %1\n\t"
79 		"mov %%rsp, %2\n\t"
80 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
81 		: "g"(0xABCD));
82 	report((rax == 0xFFFF) && (rsp == resume_rsp), "test vmresume");
83 }
84 
85 static int vmenter_exit_handler(union exit_reason exit_reason)
86 {
87 	u64 guest_rip = vmcs_read(GUEST_RIP);
88 
89 	switch (exit_reason.basic) {
90 	case VMX_VMCALL:
91 		if (regs.rax != 0xABCD) {
92 			report_fail("test vmresume");
93 			return VMX_TEST_VMEXIT;
94 		}
95 		regs.rax = 0xFFFF;
96 		vmcs_write(GUEST_RIP, guest_rip + 3);
97 		return VMX_TEST_RESUME;
98 	default:
99 		report_fail("test vmresume");
100 		print_vmexit_info(exit_reason);
101 	}
102 	return VMX_TEST_VMEXIT;
103 }
104 
105 u32 preempt_scale;
106 volatile unsigned long long tsc_val;
107 volatile u32 preempt_val;
108 u64 saved_rip;
109 
110 static int preemption_timer_init(struct vmcs *vmcs)
111 {
112 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
113 		printf("\tPreemption timer is not supported\n");
114 		return VMX_TEST_EXIT;
115 	}
116 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
117 	preempt_val = 10000000;
118 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
119 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
120 
121 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
122 		printf("\tSave preemption value is not supported\n");
123 
124 	return VMX_TEST_START;
125 }
126 
127 static void preemption_timer_main(void)
128 {
129 	tsc_val = rdtsc();
130 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
131 		vmx_set_test_stage(0);
132 		vmcall();
133 		if (vmx_get_test_stage() == 1)
134 			vmcall();
135 	}
136 	vmx_set_test_stage(1);
137 	while (vmx_get_test_stage() == 1) {
138 		if (((rdtsc() - tsc_val) >> preempt_scale)
139 				> 10 * preempt_val) {
140 			vmx_set_test_stage(2);
141 			vmcall();
142 		}
143 	}
144 	tsc_val = rdtsc();
145 	asm volatile ("hlt");
146 	vmcall();
147 	vmx_set_test_stage(5);
148 	vmcall();
149 }
150 
151 static int preemption_timer_exit_handler(union exit_reason exit_reason)
152 {
153 	bool guest_halted;
154 	u64 guest_rip;
155 	u32 insn_len;
156 	u32 ctrl_exit;
157 
158 	guest_rip = vmcs_read(GUEST_RIP);
159 	insn_len = vmcs_read(EXI_INST_LEN);
160 	switch (exit_reason.basic) {
161 	case VMX_PREEMPT:
162 		switch (vmx_get_test_stage()) {
163 		case 1:
164 		case 2:
165 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val,
166 			       "busy-wait for preemption timer");
167 			vmx_set_test_stage(3);
168 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
169 			return VMX_TEST_RESUME;
170 		case 3:
171 			guest_halted =
172 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
173 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val
174 			        && guest_halted,
175 			       "preemption timer during hlt");
176 			vmx_set_test_stage(4);
177 			vmcs_write(PIN_CONTROLS,
178 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
179 			vmcs_write(EXI_CONTROLS,
180 				   vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT);
181 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
182 			return VMX_TEST_RESUME;
183 		case 4:
184 			report(saved_rip == guest_rip,
185 			       "preemption timer with 0 value");
186 			break;
187 		default:
188 			report_fail("Invalid stage.");
189 			print_vmexit_info(exit_reason);
190 			break;
191 		}
192 		break;
193 	case VMX_VMCALL:
194 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
195 		switch (vmx_get_test_stage()) {
196 		case 0:
197 			report(vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val,
198 			       "Keep preemption value");
199 			vmx_set_test_stage(1);
200 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
201 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
202 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
203 			vmcs_write(EXI_CONTROLS, ctrl_exit);
204 			return VMX_TEST_RESUME;
205 		case 1:
206 			report(vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val,
207 			       "Save preemption value");
208 			return VMX_TEST_RESUME;
209 		case 2:
210 			report_fail("busy-wait for preemption timer");
211 			vmx_set_test_stage(3);
212 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
213 			return VMX_TEST_RESUME;
214 		case 3:
215 			report_fail("preemption timer during hlt");
216 			vmx_set_test_stage(4);
217 			/* fall through */
218 		case 4:
219 			vmcs_write(PIN_CONTROLS,
220 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
221 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
222 			saved_rip = guest_rip + insn_len;
223 			return VMX_TEST_RESUME;
224 		case 5:
225 			report_fail("preemption timer with 0 value (vmcall stage 5)");
226 			break;
227 		default:
228 			// Should not reach here
229 			report_fail("unexpected stage, %d",
230 				    vmx_get_test_stage());
231 			print_vmexit_info(exit_reason);
232 			return VMX_TEST_VMEXIT;
233 		}
234 		break;
235 	default:
236 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
237 		print_vmexit_info(exit_reason);
238 	}
239 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
240 	return VMX_TEST_VMEXIT;
241 }
242 
243 static void msr_bmp_init(void)
244 {
245 	void *msr_bitmap;
246 	u32 ctrl_cpu0;
247 
248 	msr_bitmap = alloc_page();
249 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
250 	ctrl_cpu0 |= CPU_MSR_BITMAP;
251 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
252 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
253 }
254 
255 static void *get_msr_bitmap(void)
256 {
257 	void *msr_bitmap;
258 
259 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
260 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
261 	} else {
262 		msr_bitmap = alloc_page();
263 		memset(msr_bitmap, 0xff, PAGE_SIZE);
264 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
265 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
266 	}
267 
268 	return msr_bitmap;
269 }
270 
271 static void disable_intercept_for_x2apic_msrs(void)
272 {
273 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
274 	u32 msr;
275 
276 	for (msr = APIC_BASE_MSR;
277 		 msr < (APIC_BASE_MSR+0xff);
278 		 msr += BITS_PER_LONG) {
279 		unsigned int word = msr / BITS_PER_LONG;
280 
281 		msr_bitmap[word] = 0;
282 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
283 	}
284 }
285 
286 static int test_ctrl_pat_init(struct vmcs *vmcs)
287 {
288 	u64 ctrl_ent;
289 	u64 ctrl_exi;
290 
291 	msr_bmp_init();
292 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
293 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
294 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
295 		printf("\tSave/load PAT is not supported\n");
296 		return 1;
297 	}
298 
299 	ctrl_ent = vmcs_read(ENT_CONTROLS);
300 	ctrl_exi = vmcs_read(EXI_CONTROLS);
301 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
302 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
303 	vmcs_write(ENT_CONTROLS, ctrl_ent);
304 	vmcs_write(EXI_CONTROLS, ctrl_exi);
305 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
306 	vmcs_write(GUEST_PAT, 0x0);
307 	vmcs_write(HOST_PAT, ia32_pat);
308 	return VMX_TEST_START;
309 }
310 
311 static void test_ctrl_pat_main(void)
312 {
313 	u64 guest_ia32_pat;
314 
315 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
316 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
317 		printf("\tENT_LOAD_PAT is not supported.\n");
318 	else {
319 		if (guest_ia32_pat != 0) {
320 			report_fail("Entry load PAT");
321 			return;
322 		}
323 	}
324 	wrmsr(MSR_IA32_CR_PAT, 0x6);
325 	vmcall();
326 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
327 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
328 		report(guest_ia32_pat == ia32_pat, "Entry load PAT");
329 }
330 
331 static int test_ctrl_pat_exit_handler(union exit_reason exit_reason)
332 {
333 	u64 guest_rip;
334 	u64 guest_pat;
335 
336 	guest_rip = vmcs_read(GUEST_RIP);
337 	switch (exit_reason.basic) {
338 	case VMX_VMCALL:
339 		guest_pat = vmcs_read(GUEST_PAT);
340 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
341 			printf("\tEXI_SAVE_PAT is not supported\n");
342 			vmcs_write(GUEST_PAT, 0x6);
343 		} else {
344 			report(guest_pat == 0x6, "Exit save PAT");
345 		}
346 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
347 			printf("\tEXI_LOAD_PAT is not supported\n");
348 		else
349 			report(rdmsr(MSR_IA32_CR_PAT) == ia32_pat,
350 			       "Exit load PAT");
351 		vmcs_write(GUEST_PAT, ia32_pat);
352 		vmcs_write(GUEST_RIP, guest_rip + 3);
353 		return VMX_TEST_RESUME;
354 	default:
355 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
356 		break;
357 	}
358 	return VMX_TEST_VMEXIT;
359 }
360 
361 static int test_ctrl_efer_init(struct vmcs *vmcs)
362 {
363 	u64 ctrl_ent;
364 	u64 ctrl_exi;
365 
366 	msr_bmp_init();
367 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
368 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
369 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
370 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
371 	ia32_efer = rdmsr(MSR_EFER);
372 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
373 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
374 	return VMX_TEST_START;
375 }
376 
377 static void test_ctrl_efer_main(void)
378 {
379 	u64 guest_ia32_efer;
380 
381 	guest_ia32_efer = rdmsr(MSR_EFER);
382 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
383 		printf("\tENT_LOAD_EFER is not supported.\n");
384 	else {
385 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
386 			report_fail("Entry load EFER");
387 			return;
388 		}
389 	}
390 	wrmsr(MSR_EFER, ia32_efer);
391 	vmcall();
392 	guest_ia32_efer = rdmsr(MSR_EFER);
393 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
394 		report(guest_ia32_efer == ia32_efer, "Entry load EFER");
395 }
396 
397 static int test_ctrl_efer_exit_handler(union exit_reason exit_reason)
398 {
399 	u64 guest_rip;
400 	u64 guest_efer;
401 
402 	guest_rip = vmcs_read(GUEST_RIP);
403 	switch (exit_reason.basic) {
404 	case VMX_VMCALL:
405 		guest_efer = vmcs_read(GUEST_EFER);
406 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
407 			printf("\tEXI_SAVE_EFER is not supported\n");
408 			vmcs_write(GUEST_EFER, ia32_efer);
409 		} else {
410 			report(guest_efer == ia32_efer, "Exit save EFER");
411 		}
412 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
413 			printf("\tEXI_LOAD_EFER is not supported\n");
414 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
415 		} else {
416 			report(rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX),
417 			       "Exit load EFER");
418 		}
419 		vmcs_write(GUEST_PAT, ia32_efer);
420 		vmcs_write(GUEST_RIP, guest_rip + 3);
421 		return VMX_TEST_RESUME;
422 	default:
423 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
424 		break;
425 	}
426 	return VMX_TEST_VMEXIT;
427 }
428 
429 u32 guest_cr0, guest_cr4;
430 
431 static void cr_shadowing_main(void)
432 {
433 	u32 cr0, cr4, tmp;
434 
435 	// Test read through
436 	vmx_set_test_stage(0);
437 	guest_cr0 = read_cr0();
438 	if (vmx_get_test_stage() == 1)
439 		report_fail("Read through CR0");
440 	else
441 		vmcall();
442 	vmx_set_test_stage(1);
443 	guest_cr4 = read_cr4();
444 	if (vmx_get_test_stage() == 2)
445 		report_fail("Read through CR4");
446 	else
447 		vmcall();
448 	// Test write through
449 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
450 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
451 	vmx_set_test_stage(2);
452 	write_cr0(guest_cr0);
453 	if (vmx_get_test_stage() == 3)
454 		report_fail("Write throuth CR0");
455 	else
456 		vmcall();
457 	vmx_set_test_stage(3);
458 	write_cr4(guest_cr4);
459 	if (vmx_get_test_stage() == 4)
460 		report_fail("Write through CR4");
461 	else
462 		vmcall();
463 	// Test read shadow
464 	vmx_set_test_stage(4);
465 	vmcall();
466 	cr0 = read_cr0();
467 	if (vmx_get_test_stage() != 5)
468 		report(cr0 == guest_cr0, "Read shadowing CR0");
469 	vmx_set_test_stage(5);
470 	cr4 = read_cr4();
471 	if (vmx_get_test_stage() != 6)
472 		report(cr4 == guest_cr4, "Read shadowing CR4");
473 	// Test write shadow (same value with shadow)
474 	vmx_set_test_stage(6);
475 	write_cr0(guest_cr0);
476 	if (vmx_get_test_stage() == 7)
477 		report_fail("Write shadowing CR0 (same value with shadow)");
478 	else
479 		vmcall();
480 	vmx_set_test_stage(7);
481 	write_cr4(guest_cr4);
482 	if (vmx_get_test_stage() == 8)
483 		report_fail("Write shadowing CR4 (same value with shadow)");
484 	else
485 		vmcall();
486 	// Test write shadow (different value)
487 	vmx_set_test_stage(8);
488 	tmp = guest_cr0 ^ X86_CR0_TS;
489 	asm volatile("mov %0, %%rsi\n\t"
490 		"mov %%rsi, %%cr0\n\t"
491 		::"m"(tmp)
492 		:"rsi", "memory", "cc");
493 	report(vmx_get_test_stage() == 9,
494 	       "Write shadowing different X86_CR0_TS");
495 	vmx_set_test_stage(9);
496 	tmp = guest_cr0 ^ X86_CR0_MP;
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() == 10,
502 	       "Write shadowing different X86_CR0_MP");
503 	vmx_set_test_stage(10);
504 	tmp = guest_cr4 ^ X86_CR4_TSD;
505 	asm volatile("mov %0, %%rsi\n\t"
506 		"mov %%rsi, %%cr4\n\t"
507 		::"m"(tmp)
508 		:"rsi", "memory", "cc");
509 	report(vmx_get_test_stage() == 11,
510 	       "Write shadowing different X86_CR4_TSD");
511 	vmx_set_test_stage(11);
512 	tmp = guest_cr4 ^ X86_CR4_DE;
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() == 12,
518 	       "Write shadowing different X86_CR4_DE");
519 }
520 
521 static int cr_shadowing_exit_handler(union exit_reason exit_reason)
522 {
523 	u64 guest_rip;
524 	u32 insn_len;
525 	u32 exit_qual;
526 
527 	guest_rip = vmcs_read(GUEST_RIP);
528 	insn_len = vmcs_read(EXI_INST_LEN);
529 	exit_qual = vmcs_read(EXI_QUALIFICATION);
530 	switch (exit_reason.basic) {
531 	case VMX_VMCALL:
532 		switch (vmx_get_test_stage()) {
533 		case 0:
534 			report(guest_cr0 == vmcs_read(GUEST_CR0),
535 			       "Read through CR0");
536 			break;
537 		case 1:
538 			report(guest_cr4 == vmcs_read(GUEST_CR4),
539 			       "Read through CR4");
540 			break;
541 		case 2:
542 			report(guest_cr0 == vmcs_read(GUEST_CR0),
543 			       "Write through CR0");
544 			break;
545 		case 3:
546 			report(guest_cr4 == vmcs_read(GUEST_CR4),
547 			       "Write through CR4");
548 			break;
549 		case 4:
550 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
551 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
552 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
553 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
554 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
555 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
556 			break;
557 		case 6:
558 			report(guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)),
559 			       "Write shadowing CR0 (same value)");
560 			break;
561 		case 7:
562 			report(guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)),
563 			       "Write shadowing CR4 (same value)");
564 			break;
565 		default:
566 			// Should not reach here
567 			report_fail("unexpected stage, %d",
568 				    vmx_get_test_stage());
569 			print_vmexit_info(exit_reason);
570 			return VMX_TEST_VMEXIT;
571 		}
572 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
573 		return VMX_TEST_RESUME;
574 	case VMX_CR:
575 		switch (vmx_get_test_stage()) {
576 		case 4:
577 			report_fail("Read shadowing CR0");
578 			vmx_inc_test_stage();
579 			break;
580 		case 5:
581 			report_fail("Read shadowing CR4");
582 			vmx_inc_test_stage();
583 			break;
584 		case 6:
585 			report_fail("Write shadowing CR0 (same value)");
586 			vmx_inc_test_stage();
587 			break;
588 		case 7:
589 			report_fail("Write shadowing CR4 (same value)");
590 			vmx_inc_test_stage();
591 			break;
592 		case 8:
593 		case 9:
594 			// 0x600 encodes "mov %esi, %cr0"
595 			if (exit_qual == 0x600)
596 				vmx_inc_test_stage();
597 			break;
598 		case 10:
599 		case 11:
600 			// 0x604 encodes "mov %esi, %cr4"
601 			if (exit_qual == 0x604)
602 				vmx_inc_test_stage();
603 			break;
604 		default:
605 			// Should not reach here
606 			report_fail("unexpected stage, %d",
607 				    vmx_get_test_stage());
608 			print_vmexit_info(exit_reason);
609 			return VMX_TEST_VMEXIT;
610 		}
611 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
612 		return VMX_TEST_RESUME;
613 	default:
614 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
615 		print_vmexit_info(exit_reason);
616 	}
617 	return VMX_TEST_VMEXIT;
618 }
619 
620 static int iobmp_init(struct vmcs *vmcs)
621 {
622 	u32 ctrl_cpu0;
623 
624 	io_bitmap_a = alloc_page();
625 	io_bitmap_b = alloc_page();
626 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
627 	ctrl_cpu0 |= CPU_IO_BITMAP;
628 	ctrl_cpu0 &= (~CPU_IO);
629 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
630 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
631 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
632 	return VMX_TEST_START;
633 }
634 
635 static void iobmp_main(void)
636 {
637 	// stage 0, test IO pass
638 	vmx_set_test_stage(0);
639 	inb(0x5000);
640 	outb(0x0, 0x5000);
641 	report(vmx_get_test_stage() == 0, "I/O bitmap - I/O pass");
642 	// test IO width, in/out
643 	((u8 *)io_bitmap_a)[0] = 0xFF;
644 	vmx_set_test_stage(2);
645 	inb(0x0);
646 	report(vmx_get_test_stage() == 3, "I/O bitmap - trap in");
647 	vmx_set_test_stage(3);
648 	outw(0x0, 0x0);
649 	report(vmx_get_test_stage() == 4, "I/O bitmap - trap out");
650 	vmx_set_test_stage(4);
651 	inl(0x0);
652 	report(vmx_get_test_stage() == 5, "I/O bitmap - I/O width, long");
653 	// test low/high IO port
654 	vmx_set_test_stage(5);
655 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
656 	inb(0x5000);
657 	report(vmx_get_test_stage() == 6, "I/O bitmap - I/O port, low part");
658 	vmx_set_test_stage(6);
659 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
660 	inb(0x9000);
661 	report(vmx_get_test_stage() == 7, "I/O bitmap - I/O port, high part");
662 	// test partial pass
663 	vmx_set_test_stage(7);
664 	inl(0x4FFF);
665 	report(vmx_get_test_stage() == 8, "I/O bitmap - partial pass");
666 	// test overrun
667 	vmx_set_test_stage(8);
668 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
669 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
670 	inl(0xFFFF);
671 	report(vmx_get_test_stage() == 9, "I/O bitmap - overrun");
672 	vmx_set_test_stage(9);
673 	vmcall();
674 	outb(0x0, 0x0);
675 	report(vmx_get_test_stage() == 9,
676 	       "I/O bitmap - ignore unconditional exiting");
677 	vmx_set_test_stage(10);
678 	vmcall();
679 	outb(0x0, 0x0);
680 	report(vmx_get_test_stage() == 11,
681 	       "I/O bitmap - unconditional exiting");
682 }
683 
684 static int iobmp_exit_handler(union exit_reason exit_reason)
685 {
686 	u64 guest_rip;
687 	ulong exit_qual;
688 	u32 insn_len, ctrl_cpu0;
689 
690 	guest_rip = vmcs_read(GUEST_RIP);
691 	exit_qual = vmcs_read(EXI_QUALIFICATION);
692 	insn_len = vmcs_read(EXI_INST_LEN);
693 	switch (exit_reason.basic) {
694 	case VMX_IO:
695 		switch (vmx_get_test_stage()) {
696 		case 0:
697 		case 1:
698 			vmx_inc_test_stage();
699 			break;
700 		case 2:
701 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE,
702 			       "I/O bitmap - I/O width, byte");
703 			report(exit_qual & VMX_IO_IN,
704 			       "I/O bitmap - I/O direction, in");
705 			vmx_inc_test_stage();
706 			break;
707 		case 3:
708 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD,
709 			       "I/O bitmap - I/O width, word");
710 			report(!(exit_qual & VMX_IO_IN),
711 			       "I/O bitmap - I/O direction, out");
712 			vmx_inc_test_stage();
713 			break;
714 		case 4:
715 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG,
716 			       "I/O bitmap - I/O width, long");
717 			vmx_inc_test_stage();
718 			break;
719 		case 5:
720 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
721 				vmx_inc_test_stage();
722 			break;
723 		case 6:
724 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
725 				vmx_inc_test_stage();
726 			break;
727 		case 7:
728 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
729 				vmx_inc_test_stage();
730 			break;
731 		case 8:
732 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
733 				vmx_inc_test_stage();
734 			break;
735 		case 9:
736 		case 10:
737 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
738 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
739 			vmx_inc_test_stage();
740 			break;
741 		default:
742 			// Should not reach here
743 			report_fail("unexpected stage, %d",
744 				    vmx_get_test_stage());
745 			print_vmexit_info(exit_reason);
746 			return VMX_TEST_VMEXIT;
747 		}
748 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
749 		return VMX_TEST_RESUME;
750 	case VMX_VMCALL:
751 		switch (vmx_get_test_stage()) {
752 		case 9:
753 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
754 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
755 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
756 			break;
757 		case 10:
758 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
759 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
760 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
761 			break;
762 		default:
763 			// Should not reach here
764 			report_fail("unexpected stage, %d",
765 				    vmx_get_test_stage());
766 			print_vmexit_info(exit_reason);
767 			return VMX_TEST_VMEXIT;
768 		}
769 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
770 		return VMX_TEST_RESUME;
771 	default:
772 		printf("guest_rip = %#lx\n", guest_rip);
773 		printf("\tERROR : Unknown exit reason, 0x%x\n", exit_reason.full);
774 		break;
775 	}
776 	return VMX_TEST_VMEXIT;
777 }
778 
779 #define INSN_CPU0		0
780 #define INSN_CPU1		1
781 #define INSN_ALWAYS_TRAP	2
782 
783 #define FIELD_EXIT_QUAL		(1 << 0)
784 #define FIELD_INSN_INFO		(1 << 1)
785 
786 asm(
787 	"insn_hlt: hlt;ret\n\t"
788 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
789 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
790 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
791 	"insn_rdtsc: rdtsc;ret\n\t"
792 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
793 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
794 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
795 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
796 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
797 	"insn_pause: pause;ret\n\t"
798 	"insn_wbinvd: wbinvd;ret\n\t"
799 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
800 	"insn_invd: invd;ret\n\t"
801 	"insn_sgdt: sgdt gdt_descr;ret\n\t"
802 	"insn_lgdt: lgdt gdt_descr;ret\n\t"
803 	"insn_sidt: sidt idt_descr;ret\n\t"
804 	"insn_lidt: lidt idt_descr;ret\n\t"
805 	"insn_sldt: sldt %ax;ret\n\t"
806 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
807 	"insn_str: str %ax;ret\n\t"
808 	"insn_rdrand: rdrand %rax;ret\n\t"
809 	"insn_rdseed: rdseed %rax;ret\n\t"
810 );
811 extern void insn_hlt(void);
812 extern void insn_invlpg(void);
813 extern void insn_mwait(void);
814 extern void insn_rdpmc(void);
815 extern void insn_rdtsc(void);
816 extern void insn_cr3_load(void);
817 extern void insn_cr3_store(void);
818 extern void insn_cr8_load(void);
819 extern void insn_cr8_store(void);
820 extern void insn_monitor(void);
821 extern void insn_pause(void);
822 extern void insn_wbinvd(void);
823 extern void insn_sgdt(void);
824 extern void insn_lgdt(void);
825 extern void insn_sidt(void);
826 extern void insn_lidt(void);
827 extern void insn_sldt(void);
828 extern void insn_lldt(void);
829 extern void insn_str(void);
830 extern void insn_cpuid(void);
831 extern void insn_invd(void);
832 extern void insn_rdrand(void);
833 extern void insn_rdseed(void);
834 
835 u32 cur_insn;
836 u64 cr3;
837 
838 #define X86_FEATURE_MONITOR	(1 << 3)
839 
840 typedef bool (*supported_fn)(void);
841 
842 static bool monitor_supported(void)
843 {
844 	return this_cpu_has(X86_FEATURE_MWAIT);
845 }
846 
847 struct insn_table {
848 	const char *name;
849 	u32 flag;
850 	void (*insn_func)(void);
851 	u32 type;
852 	u32 reason;
853 	ulong exit_qual;
854 	u32 insn_info;
855 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
856 	// which field need to be tested, reason is always tested
857 	u32 test_field;
858 	const supported_fn supported_fn;
859 	u8 disabled;
860 };
861 
862 /*
863  * Add more test cases of instruction intercept here. Elements in this
864  * table is:
865  *	name/control flag/insn function/type/exit reason/exit qulification/
866  *	instruction info/field to test
867  * The last field defines which fields (exit_qual and insn_info) need to be
868  * tested in exit handler. If set to 0, only "reason" is checked.
869  */
870 static struct insn_table insn_table[] = {
871 	// Flags for Primary Processor-Based VM-Execution Controls
872 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
873 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
874 		0x12345678, 0, FIELD_EXIT_QUAL},
875 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0, &monitor_supported},
876 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0},
877 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
878 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
879 		FIELD_EXIT_QUAL},
880 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
881 		FIELD_EXIT_QUAL},
882 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
883 		FIELD_EXIT_QUAL},
884 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
885 		FIELD_EXIT_QUAL},
886 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0, &monitor_supported},
887 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
888 	// Flags for Secondary Processor-Based VM-Execution Controls
889 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
890 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
891 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
892 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
893 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
894 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
895 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
896 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
897 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
898 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
899 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
900 	// Instructions always trap
901 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
902 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
903 	// Instructions never trap
904 	{NULL},
905 };
906 
907 static int insn_intercept_init(struct vmcs *vmcs)
908 {
909 	u32 ctrl_cpu, cur_insn;
910 
911 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
912 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
913 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
914 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
915 	cr3 = read_cr3();
916 
917 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
918 		if (insn_table[cur_insn].supported_fn == NULL)
919 			continue;
920 		insn_table[cur_insn].disabled = !insn_table[cur_insn].supported_fn();
921 	}
922 	return VMX_TEST_START;
923 }
924 
925 static void insn_intercept_main(void)
926 {
927 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
928 		vmx_set_test_stage(cur_insn * 2);
929 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
930 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
931 		    (insn_table[cur_insn].type == INSN_CPU1 &&
932 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
933 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
934 			       insn_table[cur_insn].type - INSN_CPU0,
935 			       insn_table[cur_insn].name);
936 			continue;
937 		}
938 
939 		if (insn_table[cur_insn].disabled) {
940 			printf("\tFeature required for %s is not supported.\n",
941 			       insn_table[cur_insn].name);
942 			continue;
943 		}
944 
945 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
946 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
947 		    (insn_table[cur_insn].type == INSN_CPU1 &&
948 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
949 			/* skip hlt, it stalls the guest and is tested below */
950 			if (insn_table[cur_insn].insn_func != insn_hlt)
951 				insn_table[cur_insn].insn_func();
952 			report(vmx_get_test_stage() == cur_insn * 2,
953 					"execute %s",
954 					insn_table[cur_insn].name);
955 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
956 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
957 			       insn_table[cur_insn].type - INSN_CPU0,
958 			       insn_table[cur_insn].name);
959 
960 		vmcall();
961 
962 		insn_table[cur_insn].insn_func();
963 		report(vmx_get_test_stage() == cur_insn * 2 + 1,
964 				"intercept %s",
965 				insn_table[cur_insn].name);
966 
967 		vmx_set_test_stage(cur_insn * 2 + 1);
968 		vmcall();
969 	}
970 }
971 
972 static int insn_intercept_exit_handler(union exit_reason exit_reason)
973 {
974 	u64 guest_rip;
975 	ulong exit_qual;
976 	u32 insn_len;
977 	u32 insn_info;
978 	bool pass;
979 
980 	guest_rip = vmcs_read(GUEST_RIP);
981 	exit_qual = vmcs_read(EXI_QUALIFICATION);
982 	insn_len = vmcs_read(EXI_INST_LEN);
983 	insn_info = vmcs_read(EXI_INST_INFO);
984 
985 	if (exit_reason.basic == VMX_VMCALL) {
986 		u32 val = 0;
987 
988 		if (insn_table[cur_insn].type == INSN_CPU0)
989 			val = vmcs_read(CPU_EXEC_CTRL0);
990 		else if (insn_table[cur_insn].type == INSN_CPU1)
991 			val = vmcs_read(CPU_EXEC_CTRL1);
992 
993 		if (vmx_get_test_stage() & 1)
994 			val &= ~insn_table[cur_insn].flag;
995 		else
996 			val |= insn_table[cur_insn].flag;
997 
998 		if (insn_table[cur_insn].type == INSN_CPU0)
999 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
1000 		else if (insn_table[cur_insn].type == INSN_CPU1)
1001 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
1002 	} else {
1003 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
1004 			insn_table[cur_insn].reason == exit_reason.full;
1005 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
1006 		    insn_table[cur_insn].exit_qual != exit_qual)
1007 			pass = false;
1008 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
1009 		    insn_table[cur_insn].insn_info != insn_info)
1010 			pass = false;
1011 		if (pass)
1012 			vmx_inc_test_stage();
1013 	}
1014 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1015 	return VMX_TEST_RESUME;
1016 }
1017 
1018 /**
1019  * __setup_ept - Setup the VMCS fields to enable Extended Page Tables (EPT)
1020  * @hpa:	Host physical address of the top-level, a.k.a. root, EPT table
1021  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1022  *
1023  * Returns 0 on success, 1 on failure.
1024  *
1025  * Note that @hpa doesn't need to point at actual memory if VM-Launch is
1026  * expected to fail, e.g. setup_dummy_ept() arbitrarily passes '0' to satisfy
1027  * the various EPTP consistency checks, but doesn't ensure backing for HPA '0'.
1028  */
1029 static int __setup_ept(u64 hpa, bool enable_ad)
1030 {
1031 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1032 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1033 		printf("\tEPT is not supported\n");
1034 		return 1;
1035 	}
1036 	if (!(ept_vpid.val & EPT_CAP_WB)) {
1037 		printf("\tWB memtype for EPT walks not supported\n");
1038 		return 1;
1039 	}
1040 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
1041 		printf("\tPWL4 is not supported\n");
1042 		return 1;
1043 	}
1044 
1045 	eptp = EPT_MEM_TYPE_WB;
1046 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1047 	eptp |= hpa;
1048 	if (enable_ad)
1049 		eptp |= EPTP_AD_FLAG;
1050 
1051 	vmcs_write(EPTP, eptp);
1052 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY);
1053 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT);
1054 
1055 	return 0;
1056 }
1057 
1058 /**
1059  * setup_ept - Enable Extended Page Tables (EPT) and setup an identity map
1060  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1061  *
1062  * Returns 0 on success, 1 on failure.
1063  *
1064  * This is the "real" function for setting up EPT tables, i.e. use this for
1065  * tests that need to run code in the guest with EPT enabled.
1066  */
1067 static int setup_ept(bool enable_ad)
1068 {
1069 	unsigned long end_of_memory;
1070 
1071 	pml4 = alloc_page();
1072 
1073 	if (__setup_ept(virt_to_phys(pml4), enable_ad))
1074 		return 1;
1075 
1076 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1077 	if (end_of_memory < (1ul << 32))
1078 		end_of_memory = (1ul << 32);
1079 	/* Cannot use large EPT pages if we need to track EPT
1080 	 * accessed/dirty bits at 4K granularity.
1081 	 */
1082 	setup_ept_range(pml4, 0, end_of_memory, 0,
1083 			!enable_ad && ept_2m_supported(),
1084 			EPT_WA | EPT_RA | EPT_EA);
1085 	return 0;
1086 }
1087 
1088 /**
1089  * setup_dummy_ept - Enable Extended Page Tables (EPT) with a dummy root HPA
1090  *
1091  * Setup EPT using a semi-arbitrary dummy root HPA.  This function is intended
1092  * for use by tests that need EPT enabled to verify dependent VMCS controls
1093  * but never expect to fully enter the guest, i.e. don't need setup the actual
1094  * EPT tables.
1095  */
1096 static void setup_dummy_ept(void)
1097 {
1098 	if (__setup_ept(0, false))
1099 		report_abort("EPT setup unexpectedly failed");
1100 }
1101 
1102 static int enable_unrestricted_guest(bool need_valid_ept)
1103 {
1104 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1105 	    !(ctrl_cpu_rev[1].clr & CPU_URG) ||
1106 	    !(ctrl_cpu_rev[1].clr & CPU_EPT))
1107 		return 1;
1108 
1109 	if (need_valid_ept)
1110 		setup_ept(false);
1111 	else
1112 		setup_dummy_ept();
1113 
1114 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
1115 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
1116 
1117 	return 0;
1118 }
1119 
1120 static void ept_enable_ad_bits(void)
1121 {
1122 	eptp |= EPTP_AD_FLAG;
1123 	vmcs_write(EPTP, eptp);
1124 }
1125 
1126 static void ept_disable_ad_bits(void)
1127 {
1128 	eptp &= ~EPTP_AD_FLAG;
1129 	vmcs_write(EPTP, eptp);
1130 }
1131 
1132 static int ept_ad_enabled(void)
1133 {
1134 	return eptp & EPTP_AD_FLAG;
1135 }
1136 
1137 static void ept_enable_ad_bits_or_skip_test(void)
1138 {
1139 	if (!ept_ad_bits_supported())
1140 		test_skip("EPT AD bits not supported.");
1141 	ept_enable_ad_bits();
1142 }
1143 
1144 static int apic_version;
1145 
1146 static int ept_init_common(bool have_ad)
1147 {
1148 	int ret;
1149 	struct pci_dev pcidev;
1150 
1151 	/* INVEPT is required by the EPT violation handler. */
1152 	if (!is_invept_type_supported(INVEPT_SINGLE))
1153 		return VMX_TEST_EXIT;
1154 
1155 	if (setup_ept(have_ad))
1156 		return VMX_TEST_EXIT;
1157 
1158 	data_page1 = alloc_page();
1159 	data_page2 = alloc_page();
1160 	*((u32 *)data_page1) = MAGIC_VAL_1;
1161 	*((u32 *)data_page2) = MAGIC_VAL_2;
1162 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1163 			EPT_RA | EPT_WA | EPT_EA);
1164 
1165 	apic_version = apic_read(APIC_LVR);
1166 
1167 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1168 	if (ret != PCIDEVADDR_INVALID) {
1169 		pci_dev_init(&pcidev, ret);
1170 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1171 	}
1172 
1173 	return VMX_TEST_START;
1174 }
1175 
1176 static int ept_init(struct vmcs *vmcs)
1177 {
1178 	return ept_init_common(false);
1179 }
1180 
1181 static void ept_common(void)
1182 {
1183 	vmx_set_test_stage(0);
1184 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1185 			*((u32 *)data_page1) != MAGIC_VAL_1)
1186 		report_fail("EPT basic framework - read");
1187 	else {
1188 		*((u32 *)data_page2) = MAGIC_VAL_3;
1189 		vmcall();
1190 		if (vmx_get_test_stage() == 1) {
1191 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1192 					*((u32 *)data_page2) == MAGIC_VAL_2)
1193 				report_pass("EPT basic framework");
1194 			else
1195 				report_pass("EPT basic framework - remap");
1196 		}
1197 	}
1198 	// Test EPT Misconfigurations
1199 	vmx_set_test_stage(1);
1200 	vmcall();
1201 	*((u32 *)data_page1) = MAGIC_VAL_1;
1202 	if (vmx_get_test_stage() != 2) {
1203 		report_fail("EPT misconfigurations");
1204 		goto t1;
1205 	}
1206 	vmx_set_test_stage(2);
1207 	vmcall();
1208 	*((u32 *)data_page1) = MAGIC_VAL_1;
1209 	report(vmx_get_test_stage() == 3, "EPT misconfigurations");
1210 t1:
1211 	// Test EPT violation
1212 	vmx_set_test_stage(3);
1213 	vmcall();
1214 	*((u32 *)data_page1) = MAGIC_VAL_1;
1215 	report(vmx_get_test_stage() == 4, "EPT violation - page permission");
1216 	// Violation caused by EPT paging structure
1217 	vmx_set_test_stage(4);
1218 	vmcall();
1219 	*((u32 *)data_page1) = MAGIC_VAL_2;
1220 	report(vmx_get_test_stage() == 5, "EPT violation - paging structure");
1221 
1222 	// MMIO Read/Write
1223 	vmx_set_test_stage(5);
1224 	vmcall();
1225 
1226 	*(u32 volatile *)pci_physaddr;
1227 	report(vmx_get_test_stage() == 6, "MMIO EPT violation - read");
1228 
1229 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1230 	report(vmx_get_test_stage() == 7, "MMIO EPT violation - write");
1231 }
1232 
1233 static void ept_main(void)
1234 {
1235 	ept_common();
1236 
1237 	// Test EPT access to L1 MMIO
1238 	vmx_set_test_stage(7);
1239 	report(*((u32 *)0xfee00030UL) == apic_version, "EPT - MMIO access");
1240 
1241 	// Test invalid operand for INVEPT
1242 	vmcall();
1243 	report(vmx_get_test_stage() == 8, "EPT - unsupported INVEPT");
1244 }
1245 
1246 static bool invept_test(int type, u64 eptp)
1247 {
1248 	bool ret, supported;
1249 
1250 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1251 	ret = __invept(type, eptp);
1252 
1253 	if (ret == !supported)
1254 		return false;
1255 
1256 	if (!supported)
1257 		printf("WARNING: unsupported invept passed!\n");
1258 	else
1259 		printf("WARNING: invept failed!\n");
1260 
1261 	return true;
1262 }
1263 
1264 static int pml_exit_handler(union exit_reason exit_reason)
1265 {
1266 	u16 index, count;
1267 	u64 *pmlbuf = pml_log;
1268 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1269 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1270 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1271 
1272 	switch (exit_reason.basic) {
1273 	case VMX_VMCALL:
1274 		switch (vmx_get_test_stage()) {
1275 		case 0:
1276 			index = vmcs_read(GUEST_PML_INDEX);
1277 			for (count = index + 1; count < PML_INDEX; count++) {
1278 				if (pmlbuf[count] == (u64)data_page2) {
1279 					vmx_inc_test_stage();
1280 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1281 					break;
1282 				}
1283 			}
1284 			break;
1285 		case 1:
1286 			index = vmcs_read(GUEST_PML_INDEX);
1287 			/* Keep clearing the dirty bit till a overflow */
1288 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1289 			break;
1290 		default:
1291 			report_fail("unexpected stage, %d.",
1292 			       vmx_get_test_stage());
1293 			print_vmexit_info(exit_reason);
1294 			return VMX_TEST_VMEXIT;
1295 		}
1296 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1297 		return VMX_TEST_RESUME;
1298 	case VMX_PML_FULL:
1299 		vmx_inc_test_stage();
1300 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1301 		return VMX_TEST_RESUME;
1302 	default:
1303 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1304 		print_vmexit_info(exit_reason);
1305 	}
1306 	return VMX_TEST_VMEXIT;
1307 }
1308 
1309 static int ept_exit_handler_common(union exit_reason exit_reason, bool have_ad)
1310 {
1311 	u64 guest_rip;
1312 	u64 guest_cr3;
1313 	u32 insn_len;
1314 	u32 exit_qual;
1315 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte,
1316 			     guest_pte_addr;
1317 
1318 	guest_rip = vmcs_read(GUEST_RIP);
1319 	guest_cr3 = vmcs_read(GUEST_CR3);
1320 	insn_len = vmcs_read(EXI_INST_LEN);
1321 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1322 	pteval_t *ptep;
1323 	switch (exit_reason.basic) {
1324 	case VMX_VMCALL:
1325 		switch (vmx_get_test_stage()) {
1326 		case 0:
1327 			check_ept_ad(pml4, guest_cr3,
1328 				     (unsigned long)data_page1,
1329 				     have_ad ? EPT_ACCESS_FLAG : 0,
1330 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1331 			check_ept_ad(pml4, guest_cr3,
1332 				     (unsigned long)data_page2,
1333 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1334 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1335 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1336 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1337 			if (have_ad)
1338 				invept(INVEPT_SINGLE, eptp);
1339 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1340 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1341 				vmx_inc_test_stage();
1342 				install_ept(pml4, (unsigned long)data_page2,
1343 						(unsigned long)data_page2,
1344 						EPT_RA | EPT_WA | EPT_EA);
1345 			} else
1346 				report_fail("EPT basic framework - write");
1347 			break;
1348 		case 1:
1349 			install_ept(pml4, (unsigned long)data_page1,
1350  				(unsigned long)data_page1, EPT_WA);
1351 			invept(INVEPT_SINGLE, eptp);
1352 			break;
1353 		case 2:
1354 			install_ept(pml4, (unsigned long)data_page1,
1355  				(unsigned long)data_page1,
1356  				EPT_RA | EPT_WA | EPT_EA |
1357  				(2 << EPT_MEM_TYPE_SHIFT));
1358 			invept(INVEPT_SINGLE, eptp);
1359 			break;
1360 		case 3:
1361 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1362 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1363 						1, &data_page1_pte));
1364 			set_ept_pte(pml4, (unsigned long)data_page1,
1365 				1, data_page1_pte & ~EPT_PRESENT);
1366 			invept(INVEPT_SINGLE, eptp);
1367 			break;
1368 		case 4:
1369 			ptep = get_pte_level((pgd_t *)guest_cr3, data_page1, /*level=*/2);
1370 			guest_pte_addr = virt_to_phys(ptep) & PAGE_MASK;
1371 
1372 			TEST_ASSERT(get_ept_pte(pml4, guest_pte_addr, 2, &data_page1_pte_pte));
1373 			set_ept_pte(pml4, guest_pte_addr, 2,
1374 				data_page1_pte_pte & ~EPT_PRESENT);
1375 			invept(INVEPT_SINGLE, eptp);
1376 			break;
1377 		case 5:
1378 			install_ept(pml4, (unsigned long)pci_physaddr,
1379 				(unsigned long)pci_physaddr, 0);
1380 			invept(INVEPT_SINGLE, eptp);
1381 			break;
1382 		case 7:
1383 			if (!invept_test(0, eptp))
1384 				vmx_inc_test_stage();
1385 			break;
1386 		// Should not reach here
1387 		default:
1388 			report_fail("ERROR - unexpected stage, %d.",
1389 			       vmx_get_test_stage());
1390 			print_vmexit_info(exit_reason);
1391 			return VMX_TEST_VMEXIT;
1392 		}
1393 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1394 		return VMX_TEST_RESUME;
1395 	case VMX_EPT_MISCONFIG:
1396 		switch (vmx_get_test_stage()) {
1397 		case 1:
1398 		case 2:
1399 			vmx_inc_test_stage();
1400 			install_ept(pml4, (unsigned long)data_page1,
1401  				(unsigned long)data_page1,
1402  				EPT_RA | EPT_WA | EPT_EA);
1403 			invept(INVEPT_SINGLE, eptp);
1404 			break;
1405 		// Should not reach here
1406 		default:
1407 			report_fail("ERROR - unexpected stage, %d.",
1408 			       vmx_get_test_stage());
1409 			print_vmexit_info(exit_reason);
1410 			return VMX_TEST_VMEXIT;
1411 		}
1412 		return VMX_TEST_RESUME;
1413 	case VMX_EPT_VIOLATION:
1414 		/*
1415 		 * Exit-qualifications are masked not to account for advanced
1416 		 * VM-exit information. Once KVM supports this feature, this
1417 		 * masking should be removed.
1418 		 */
1419 		exit_qual &= ~EPT_VLT_GUEST_MASK;
1420 
1421 		switch(vmx_get_test_stage()) {
1422 		case 3:
1423 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1424 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1425 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1426 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1427 					EPT_VLT_PADDR))
1428 				vmx_inc_test_stage();
1429 			set_ept_pte(pml4, (unsigned long)data_page1,
1430 				1, data_page1_pte | (EPT_PRESENT));
1431 			invept(INVEPT_SINGLE, eptp);
1432 			break;
1433 		case 4:
1434 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1435 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1436 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1437 			if (exit_qual == (EPT_VLT_RD |
1438 					  (have_ad ? EPT_VLT_WR : 0) |
1439 					  EPT_VLT_LADDR_VLD))
1440 				vmx_inc_test_stage();
1441 			set_ept_pte(pml4, guest_pte_addr, 2,
1442 				data_page1_pte_pte | (EPT_PRESENT));
1443 			invept(INVEPT_SINGLE, eptp);
1444 			break;
1445 		case 5:
1446 			if (exit_qual & EPT_VLT_RD)
1447 				vmx_inc_test_stage();
1448 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1449 						1, &memaddr_pte));
1450 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1451 			invept(INVEPT_SINGLE, eptp);
1452 			break;
1453 		case 6:
1454 			if (exit_qual & EPT_VLT_WR)
1455 				vmx_inc_test_stage();
1456 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1457 						1, &memaddr_pte));
1458 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1459 			invept(INVEPT_SINGLE, eptp);
1460 			break;
1461 		default:
1462 			// Should not reach here
1463 			report_fail("ERROR : unexpected stage, %d",
1464 			       vmx_get_test_stage());
1465 			print_vmexit_info(exit_reason);
1466 			return VMX_TEST_VMEXIT;
1467 		}
1468 		return VMX_TEST_RESUME;
1469 	default:
1470 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1471 		print_vmexit_info(exit_reason);
1472 	}
1473 	return VMX_TEST_VMEXIT;
1474 }
1475 
1476 static int ept_exit_handler(union exit_reason exit_reason)
1477 {
1478 	return ept_exit_handler_common(exit_reason, false);
1479 }
1480 
1481 static int eptad_init(struct vmcs *vmcs)
1482 {
1483 	int r = ept_init_common(true);
1484 
1485 	if (r == VMX_TEST_EXIT)
1486 		return r;
1487 
1488 	if ((rdmsr(MSR_IA32_VMX_EPT_VPID_CAP) & EPT_CAP_AD_FLAG) == 0) {
1489 		printf("\tEPT A/D bits are not supported");
1490 		return VMX_TEST_EXIT;
1491 	}
1492 
1493 	return r;
1494 }
1495 
1496 static int pml_init(struct vmcs *vmcs)
1497 {
1498 	u32 ctrl_cpu;
1499 	int r = eptad_init(vmcs);
1500 
1501 	if (r == VMX_TEST_EXIT)
1502 		return r;
1503 
1504 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1505 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1506 		printf("\tPML is not supported");
1507 		return VMX_TEST_EXIT;
1508 	}
1509 
1510 	pml_log = alloc_page();
1511 	vmcs_write(PMLADDR, (u64)pml_log);
1512 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1513 
1514 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1515 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1516 
1517 	return VMX_TEST_START;
1518 }
1519 
1520 static void pml_main(void)
1521 {
1522 	int count = 0;
1523 
1524 	vmx_set_test_stage(0);
1525 	*((u32 *)data_page2) = 0x1;
1526 	vmcall();
1527 	report(vmx_get_test_stage() == 1, "PML - Dirty GPA Logging");
1528 
1529 	while (vmx_get_test_stage() == 1) {
1530 		vmcall();
1531 		*((u32 *)data_page2) = 0x1;
1532 		if (count++ > PML_INDEX)
1533 			break;
1534 	}
1535 	report(vmx_get_test_stage() == 2, "PML Full Event");
1536 }
1537 
1538 static void eptad_main(void)
1539 {
1540 	ept_common();
1541 }
1542 
1543 static int eptad_exit_handler(union exit_reason exit_reason)
1544 {
1545 	return ept_exit_handler_common(exit_reason, true);
1546 }
1547 
1548 #define TIMER_VECTOR	222
1549 
1550 static volatile bool timer_fired;
1551 
1552 static void timer_isr(isr_regs_t *regs)
1553 {
1554 	timer_fired = true;
1555 	apic_write(APIC_EOI, 0);
1556 }
1557 
1558 static int interrupt_init(struct vmcs *vmcs)
1559 {
1560 	msr_bmp_init();
1561 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1562 	handle_irq(TIMER_VECTOR, timer_isr);
1563 	return VMX_TEST_START;
1564 }
1565 
1566 static void interrupt_main(void)
1567 {
1568 	long long start, loops;
1569 
1570 	vmx_set_test_stage(0);
1571 
1572 	apic_write(APIC_LVTT, TIMER_VECTOR);
1573 	irq_enable();
1574 
1575 	apic_write(APIC_TMICT, 1);
1576 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1577 		asm volatile ("nop");
1578 	report(timer_fired, "direct interrupt while running guest");
1579 
1580 	apic_write(APIC_TMICT, 0);
1581 	irq_disable();
1582 	vmcall();
1583 	timer_fired = false;
1584 	apic_write(APIC_TMICT, 1);
1585 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1586 		asm volatile ("nop");
1587 	report(timer_fired, "intercepted interrupt while running guest");
1588 
1589 	irq_enable();
1590 	apic_write(APIC_TMICT, 0);
1591 	irq_disable();
1592 	vmcall();
1593 	timer_fired = false;
1594 	start = rdtsc();
1595 	apic_write(APIC_TMICT, 1000000);
1596 
1597 	safe_halt();
1598 
1599 	report(rdtsc() - start > 1000000 && timer_fired,
1600 	       "direct interrupt + hlt");
1601 
1602 	apic_write(APIC_TMICT, 0);
1603 	irq_disable();
1604 	vmcall();
1605 	timer_fired = false;
1606 	start = rdtsc();
1607 	apic_write(APIC_TMICT, 1000000);
1608 
1609 	safe_halt();
1610 
1611 	report(rdtsc() - start > 10000 && timer_fired,
1612 	       "intercepted interrupt + hlt");
1613 
1614 	apic_write(APIC_TMICT, 0);
1615 	irq_disable();
1616 	vmcall();
1617 	timer_fired = false;
1618 	start = rdtsc();
1619 	apic_write(APIC_TMICT, 1000000);
1620 
1621 	irq_enable();
1622 	asm volatile ("nop");
1623 	vmcall();
1624 
1625 	report(rdtsc() - start > 10000 && timer_fired,
1626 	       "direct interrupt + activity state hlt");
1627 
1628 	apic_write(APIC_TMICT, 0);
1629 	irq_disable();
1630 	vmcall();
1631 	timer_fired = false;
1632 	start = rdtsc();
1633 	apic_write(APIC_TMICT, 1000000);
1634 
1635 	irq_enable();
1636 	asm volatile ("nop");
1637 	vmcall();
1638 
1639 	report(rdtsc() - start > 10000 && timer_fired,
1640 	       "intercepted interrupt + activity state hlt");
1641 
1642 	apic_write(APIC_TMICT, 0);
1643 	irq_disable();
1644 	vmx_set_test_stage(7);
1645 	vmcall();
1646 	timer_fired = false;
1647 	apic_write(APIC_TMICT, 1);
1648 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1649 		asm volatile ("nop");
1650 	report(timer_fired,
1651 	       "running a guest with interrupt acknowledgement set");
1652 
1653 	apic_write(APIC_TMICT, 0);
1654 	irq_enable();
1655 	timer_fired = false;
1656 	vmcall();
1657 	report(timer_fired, "Inject an event to a halted guest");
1658 }
1659 
1660 static int interrupt_exit_handler(union exit_reason exit_reason)
1661 {
1662 	u64 guest_rip = vmcs_read(GUEST_RIP);
1663 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1664 
1665 	switch (exit_reason.basic) {
1666 	case VMX_VMCALL:
1667 		switch (vmx_get_test_stage()) {
1668 		case 0:
1669 		case 2:
1670 		case 5:
1671 			vmcs_write(PIN_CONTROLS,
1672 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1673 			break;
1674 		case 7:
1675 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1676 			vmcs_write(PIN_CONTROLS,
1677 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1678 			break;
1679 		case 1:
1680 		case 3:
1681 			vmcs_write(PIN_CONTROLS,
1682 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1683 			break;
1684 		case 4:
1685 		case 6:
1686 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1687 			break;
1688 
1689 		case 8:
1690 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1691 			vmcs_write(ENT_INTR_INFO,
1692 				   TIMER_VECTOR |
1693 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1694 				   INTR_INFO_VALID_MASK);
1695 			break;
1696 		}
1697 		vmx_inc_test_stage();
1698 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1699 		return VMX_TEST_RESUME;
1700 	case VMX_EXTINT:
1701 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1702 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1703 			handle_external_interrupt(vector);
1704 		} else {
1705 			irq_enable();
1706 			asm volatile ("nop");
1707 			irq_disable();
1708 		}
1709 		if (vmx_get_test_stage() >= 2)
1710 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1711 		return VMX_TEST_RESUME;
1712 	default:
1713 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1714 		print_vmexit_info(exit_reason);
1715 	}
1716 
1717 	return VMX_TEST_VMEXIT;
1718 }
1719 
1720 
1721 static volatile int nmi_fired;
1722 
1723 #define NMI_DELAY 100000000ULL
1724 
1725 static void nmi_isr(isr_regs_t *regs)
1726 {
1727 	nmi_fired = true;
1728 }
1729 
1730 static int nmi_hlt_init(struct vmcs *vmcs)
1731 {
1732 	msr_bmp_init();
1733 	handle_irq(NMI_VECTOR, nmi_isr);
1734 	vmcs_write(PIN_CONTROLS,
1735 		   vmcs_read(PIN_CONTROLS) & ~PIN_NMI);
1736 	vmcs_write(PIN_CONTROLS,
1737 		   vmcs_read(PIN_CONTROLS) & ~PIN_VIRT_NMI);
1738 	return VMX_TEST_START;
1739 }
1740 
1741 static void nmi_message_thread(void *data)
1742 {
1743     while (vmx_get_test_stage() != 1)
1744         pause();
1745 
1746     delay(NMI_DELAY);
1747     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1748 
1749     while (vmx_get_test_stage() != 2)
1750         pause();
1751 
1752     delay(NMI_DELAY);
1753     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1754 }
1755 
1756 static void nmi_hlt_main(void)
1757 {
1758     long long start;
1759 
1760     if (cpu_count() < 2) {
1761         report_skip(__func__);
1762         vmx_set_test_stage(-1);
1763         return;
1764     }
1765 
1766     vmx_set_test_stage(0);
1767     on_cpu_async(1, nmi_message_thread, NULL);
1768     start = rdtsc();
1769     vmx_set_test_stage(1);
1770     asm volatile ("hlt");
1771     report((rdtsc() - start > NMI_DELAY) && nmi_fired,
1772             "direct NMI + hlt");
1773     if (!nmi_fired)
1774         vmx_set_test_stage(-1);
1775     nmi_fired = false;
1776 
1777     vmcall();
1778 
1779     start = rdtsc();
1780     vmx_set_test_stage(2);
1781     asm volatile ("hlt");
1782     report((rdtsc() - start > NMI_DELAY) && !nmi_fired,
1783             "intercepted NMI + hlt");
1784     if (nmi_fired) {
1785         report(!nmi_fired, "intercepted NMI was dispatched");
1786         vmx_set_test_stage(-1);
1787         return;
1788     }
1789     vmx_set_test_stage(3);
1790 }
1791 
1792 static int nmi_hlt_exit_handler(union exit_reason exit_reason)
1793 {
1794     u64 guest_rip = vmcs_read(GUEST_RIP);
1795     u32 insn_len = vmcs_read(EXI_INST_LEN);
1796 
1797     switch (vmx_get_test_stage()) {
1798     case 1:
1799         if (exit_reason.basic != VMX_VMCALL) {
1800             report_fail("VMEXIT not due to vmcall. Exit reason 0x%x",
1801                         exit_reason.full);
1802             print_vmexit_info(exit_reason);
1803             return VMX_TEST_VMEXIT;
1804         }
1805 
1806         vmcs_write(PIN_CONTROLS,
1807                vmcs_read(PIN_CONTROLS) | PIN_NMI);
1808         vmcs_write(PIN_CONTROLS,
1809                vmcs_read(PIN_CONTROLS) | PIN_VIRT_NMI);
1810         vmcs_write(GUEST_RIP, guest_rip + insn_len);
1811         break;
1812 
1813     case 2:
1814         if (exit_reason.basic != VMX_EXC_NMI) {
1815             report_fail("VMEXIT not due to NMI intercept. Exit reason 0x%x",
1816                         exit_reason.full);
1817             print_vmexit_info(exit_reason);
1818             return VMX_TEST_VMEXIT;
1819         }
1820         report_pass("NMI intercept while running guest");
1821         vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1822         break;
1823 
1824     case 3:
1825         break;
1826 
1827     default:
1828         return VMX_TEST_VMEXIT;
1829     }
1830 
1831     if (vmx_get_test_stage() == 3)
1832         return VMX_TEST_VMEXIT;
1833 
1834     return VMX_TEST_RESUME;
1835 }
1836 
1837 
1838 static int dbgctls_init(struct vmcs *vmcs)
1839 {
1840 	u64 dr7 = 0x402;
1841 	u64 zero = 0;
1842 
1843 	msr_bmp_init();
1844 	asm volatile(
1845 		"mov %0,%%dr0\n\t"
1846 		"mov %0,%%dr1\n\t"
1847 		"mov %0,%%dr2\n\t"
1848 		"mov %1,%%dr7\n\t"
1849 		: : "r" (zero), "r" (dr7));
1850 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1851 	vmcs_write(GUEST_DR7, 0x404);
1852 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1853 
1854 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1855 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1856 
1857 	return VMX_TEST_START;
1858 }
1859 
1860 static void dbgctls_main(void)
1861 {
1862 	u64 dr7, debugctl;
1863 
1864 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1865 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1866 	/* Commented out: KVM does not support DEBUGCTL so far */
1867 	(void)debugctl;
1868 	report(dr7 == 0x404, "Load debug controls" /* && debugctl == 0x2 */);
1869 
1870 	dr7 = 0x408;
1871 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1872 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1873 
1874 	vmx_set_test_stage(0);
1875 	vmcall();
1876 	report(vmx_get_test_stage() == 1, "Save debug controls");
1877 
1878 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1879 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1880 		printf("\tDebug controls are always loaded/saved\n");
1881 		return;
1882 	}
1883 	vmx_set_test_stage(2);
1884 	vmcall();
1885 
1886 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1887 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1888 	/* Commented out: KVM does not support DEBUGCTL so far */
1889 	(void)debugctl;
1890 	report(dr7 == 0x402,
1891 	       "Guest=host debug controls" /* && debugctl == 0x1 */);
1892 
1893 	dr7 = 0x408;
1894 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1895 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1896 
1897 	vmx_set_test_stage(3);
1898 	vmcall();
1899 	report(vmx_get_test_stage() == 4, "Don't save debug controls");
1900 }
1901 
1902 static int dbgctls_exit_handler(union exit_reason exit_reason)
1903 {
1904 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1905 	u64 guest_rip = vmcs_read(GUEST_RIP);
1906 	u64 dr7, debugctl;
1907 
1908 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1909 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1910 
1911 	switch (exit_reason.basic) {
1912 	case VMX_VMCALL:
1913 		switch (vmx_get_test_stage()) {
1914 		case 0:
1915 			if (dr7 == 0x400 && debugctl == 0 &&
1916 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1917 			    Commented out: KVM does not support DEBUGCTL so far
1918 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1919 				vmx_inc_test_stage();
1920 			break;
1921 		case 2:
1922 			dr7 = 0x402;
1923 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1924 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1925 			vmcs_write(GUEST_DR7, 0x404);
1926 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1927 
1928 			vmcs_write(ENT_CONTROLS,
1929 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1930 			vmcs_write(EXI_CONTROLS,
1931 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1932 			break;
1933 		case 3:
1934 			if (dr7 == 0x400 && debugctl == 0 &&
1935 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1936 			    Commented out: KVM does not support DEBUGCTL so far
1937 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1938 				vmx_inc_test_stage();
1939 			break;
1940 		}
1941 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1942 		return VMX_TEST_RESUME;
1943 	default:
1944 		report_fail("Unknown exit reason, %d", exit_reason.full);
1945 		print_vmexit_info(exit_reason);
1946 	}
1947 	return VMX_TEST_VMEXIT;
1948 }
1949 
1950 struct vmx_msr_entry {
1951 	u32 index;
1952 	u32 reserved;
1953 	u64 value;
1954 } __attribute__((packed));
1955 
1956 #define MSR_MAGIC 0x31415926
1957 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1958 
1959 static int msr_switch_init(struct vmcs *vmcs)
1960 {
1961 	msr_bmp_init();
1962 	exit_msr_store = alloc_page();
1963 	exit_msr_load = alloc_page();
1964 	entry_msr_load = alloc_page();
1965 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1966 	entry_msr_load[0].value = MSR_MAGIC;
1967 
1968 	vmx_set_test_stage(1);
1969 	vmcs_write(ENT_MSR_LD_CNT, 1);
1970 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1971 	vmcs_write(EXI_MSR_ST_CNT, 1);
1972 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1973 	vmcs_write(EXI_MSR_LD_CNT, 1);
1974 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1975 	return VMX_TEST_START;
1976 }
1977 
1978 static void msr_switch_main(void)
1979 {
1980 	if (vmx_get_test_stage() == 1) {
1981 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC,
1982 		       "VM entry MSR load");
1983 		vmx_set_test_stage(2);
1984 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
1985 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
1986 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
1987 		exit_msr_load[0].value = MSR_MAGIC + 2;
1988 	}
1989 	vmcall();
1990 }
1991 
1992 static int msr_switch_exit_handler(union exit_reason exit_reason)
1993 {
1994 	if (exit_reason.basic == VMX_VMCALL && vmx_get_test_stage() == 2) {
1995 		report(exit_msr_store[0].value == MSR_MAGIC + 1,
1996 		       "VM exit MSR store");
1997 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2,
1998 		       "VM exit MSR load");
1999 		vmx_set_test_stage(3);
2000 		entry_msr_load[0].index = MSR_FS_BASE;
2001 		return VMX_TEST_RESUME;
2002 	}
2003 	printf("ERROR %s: unexpected stage=%u or reason=0x%x\n",
2004 		__func__, vmx_get_test_stage(), exit_reason.full);
2005 	return VMX_TEST_EXIT;
2006 }
2007 
2008 static int msr_switch_entry_failure(struct vmentry_result *result)
2009 {
2010 	if (result->vm_fail) {
2011 		printf("ERROR %s: VM-Fail on %s\n", __func__, result->instr);
2012 		return VMX_TEST_EXIT;
2013 	}
2014 
2015 	if (result->exit_reason.failed_vmentry &&
2016 	    result->exit_reason.basic == VMX_FAIL_MSR &&
2017 	    vmx_get_test_stage() == 3) {
2018 		report(vmcs_read(EXI_QUALIFICATION) == 1,
2019 		       "VM entry MSR load: try to load FS_BASE");
2020 		return VMX_TEST_VMEXIT;
2021 	}
2022 	printf("ERROR %s: unexpected stage=%u or reason=%x\n",
2023 		__func__, vmx_get_test_stage(), result->exit_reason.full);
2024 	return VMX_TEST_EXIT;
2025 }
2026 
2027 static int vmmcall_init(struct vmcs *vmcs)
2028 {
2029 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
2030 	return VMX_TEST_START;
2031 }
2032 
2033 static void vmmcall_main(void)
2034 {
2035 	asm volatile(
2036 		"mov $0xABCD, %%rax\n\t"
2037 		"vmmcall\n\t"
2038 		::: "rax");
2039 
2040 	report_fail("VMMCALL");
2041 }
2042 
2043 static int vmmcall_exit_handler(union exit_reason exit_reason)
2044 {
2045 	switch (exit_reason.basic) {
2046 	case VMX_VMCALL:
2047 		printf("here\n");
2048 		report_fail("VMMCALL triggers #UD");
2049 		break;
2050 	case VMX_EXC_NMI:
2051 		report((vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR,
2052 		       "VMMCALL triggers #UD");
2053 		break;
2054 	default:
2055 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2056 		print_vmexit_info(exit_reason);
2057 	}
2058 
2059 	return VMX_TEST_VMEXIT;
2060 }
2061 
2062 static int disable_rdtscp_init(struct vmcs *vmcs)
2063 {
2064 	u32 ctrl_cpu1;
2065 
2066 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
2067 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
2068 		ctrl_cpu1 &= ~CPU_RDTSCP;
2069 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
2070 	}
2071 
2072 	return VMX_TEST_START;
2073 }
2074 
2075 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
2076 {
2077 	switch (vmx_get_test_stage()) {
2078 	case 0:
2079 		report_pass("RDTSCP triggers #UD");
2080 		vmx_inc_test_stage();
2081 		regs->rip += 3;
2082 		break;
2083 	case 2:
2084 		report_pass("RDPID triggers #UD");
2085 		vmx_inc_test_stage();
2086 		regs->rip += 4;
2087 		break;
2088 	}
2089 	return;
2090 
2091 }
2092 
2093 static void disable_rdtscp_main(void)
2094 {
2095 	/* Test that #UD is properly injected in L2.  */
2096 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2097 
2098 	vmx_set_test_stage(0);
2099 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2100 	vmcall();
2101 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2102 
2103 	handle_exception(UD_VECTOR, 0);
2104 	vmcall();
2105 }
2106 
2107 static int disable_rdtscp_exit_handler(union exit_reason exit_reason)
2108 {
2109 	switch (exit_reason.basic) {
2110 	case VMX_VMCALL:
2111 		switch (vmx_get_test_stage()) {
2112 		case 0:
2113 			report_fail("RDTSCP triggers #UD");
2114 			vmx_inc_test_stage();
2115 			/* fallthrough */
2116 		case 1:
2117 			vmx_inc_test_stage();
2118 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2119 			return VMX_TEST_RESUME;
2120 		case 2:
2121 			report_fail("RDPID triggers #UD");
2122 			break;
2123 		}
2124 		break;
2125 
2126 	default:
2127 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2128 		print_vmexit_info(exit_reason);
2129 	}
2130 	return VMX_TEST_VMEXIT;
2131 }
2132 
2133 static int int3_init(struct vmcs *vmcs)
2134 {
2135 	vmcs_write(EXC_BITMAP, ~0u);
2136 	return VMX_TEST_START;
2137 }
2138 
2139 static void int3_guest_main(void)
2140 {
2141 	asm volatile ("int3");
2142 }
2143 
2144 static int int3_exit_handler(union exit_reason exit_reason)
2145 {
2146 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2147 
2148 	report(exit_reason.basic == VMX_EXC_NMI &&
2149 	       (intr_info & INTR_INFO_VALID_MASK) &&
2150 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2151 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2152 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2153 	       "L1 intercepts #BP");
2154 
2155 	return VMX_TEST_VMEXIT;
2156 }
2157 
2158 static int into_init(struct vmcs *vmcs)
2159 {
2160 	vmcs_write(EXC_BITMAP, ~0u);
2161 	return VMX_TEST_START;
2162 }
2163 
2164 static void into_guest_main(void)
2165 {
2166 	struct far_pointer32 fp = {
2167 		.offset = (uintptr_t)&&into,
2168 		.selector = KERNEL_CS32,
2169 	};
2170 	uintptr_t rsp;
2171 
2172 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
2173 
2174 	if (fp.offset != (uintptr_t)&&into) {
2175 		printf("Code address too high.\n");
2176 		return;
2177 	}
2178 	if ((u32)rsp != rsp) {
2179 		printf("Stack address too high.\n");
2180 		return;
2181 	}
2182 
2183 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2184 	return;
2185 into:
2186 	asm volatile (".code32;"
2187 		      "movl $0x7fffffff, %eax;"
2188 		      "addl %eax, %eax;"
2189 		      "into;"
2190 		      "lret;"
2191 		      ".code64");
2192 	__builtin_unreachable();
2193 }
2194 
2195 static int into_exit_handler(union exit_reason exit_reason)
2196 {
2197 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2198 
2199 	report(exit_reason.basic == VMX_EXC_NMI &&
2200 	       (intr_info & INTR_INFO_VALID_MASK) &&
2201 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2202 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2203 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2204 	       "L1 intercepts #OF");
2205 
2206 	return VMX_TEST_VMEXIT;
2207 }
2208 
2209 static void exit_monitor_from_l2_main(void)
2210 {
2211 	printf("Calling exit(0) from l2...\n");
2212 	exit(0);
2213 }
2214 
2215 static int exit_monitor_from_l2_handler(union exit_reason exit_reason)
2216 {
2217 	report_fail("The guest should have killed the VMM");
2218 	return VMX_TEST_EXIT;
2219 }
2220 
2221 static void assert_exit_reason(u64 expected)
2222 {
2223 	u64 actual = vmcs_read(EXI_REASON);
2224 
2225 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2226 			   exit_reason_description(expected),
2227 			   exit_reason_description(actual));
2228 }
2229 
2230 static void skip_exit_insn(void)
2231 {
2232 	u64 guest_rip = vmcs_read(GUEST_RIP);
2233 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2234 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2235 }
2236 
2237 static void skip_exit_vmcall(void)
2238 {
2239 	assert_exit_reason(VMX_VMCALL);
2240 	skip_exit_insn();
2241 }
2242 
2243 static void v2_null_test_guest(void)
2244 {
2245 }
2246 
2247 static void v2_null_test(void)
2248 {
2249 	test_set_guest(v2_null_test_guest);
2250 	enter_guest();
2251 	report_pass(__func__);
2252 }
2253 
2254 static void v2_multiple_entries_test_guest(void)
2255 {
2256 	vmx_set_test_stage(1);
2257 	vmcall();
2258 	vmx_set_test_stage(2);
2259 }
2260 
2261 static void v2_multiple_entries_test(void)
2262 {
2263 	test_set_guest(v2_multiple_entries_test_guest);
2264 	enter_guest();
2265 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2266 	skip_exit_vmcall();
2267 	enter_guest();
2268 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2269 	report_pass(__func__);
2270 }
2271 
2272 static int fixture_test_data = 1;
2273 
2274 static void fixture_test_teardown(void *data)
2275 {
2276 	*((int *) data) = 1;
2277 }
2278 
2279 static void fixture_test_guest(void)
2280 {
2281 	fixture_test_data++;
2282 }
2283 
2284 
2285 static void fixture_test_setup(void)
2286 {
2287 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2288 			   "fixture_test_teardown didn't run?!");
2289 	fixture_test_data = 2;
2290 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2291 	test_set_guest(fixture_test_guest);
2292 }
2293 
2294 static void fixture_test_case1(void)
2295 {
2296 	fixture_test_setup();
2297 	TEST_ASSERT_EQ(2, fixture_test_data);
2298 	enter_guest();
2299 	TEST_ASSERT_EQ(3, fixture_test_data);
2300 	report_pass(__func__);
2301 }
2302 
2303 static void fixture_test_case2(void)
2304 {
2305 	fixture_test_setup();
2306 	TEST_ASSERT_EQ(2, fixture_test_data);
2307 	enter_guest();
2308 	TEST_ASSERT_EQ(3, fixture_test_data);
2309 	report_pass(__func__);
2310 }
2311 
2312 enum ept_access_op {
2313 	OP_READ,
2314 	OP_WRITE,
2315 	OP_EXEC,
2316 	OP_FLUSH_TLB,
2317 	OP_EXIT,
2318 };
2319 
2320 static struct ept_access_test_data {
2321 	unsigned long gpa;
2322 	unsigned long *gva;
2323 	unsigned long hpa;
2324 	unsigned long *hva;
2325 	enum ept_access_op op;
2326 } ept_access_test_data;
2327 
2328 extern unsigned char ret42_start;
2329 extern unsigned char ret42_end;
2330 
2331 /* Returns 42. */
2332 asm(
2333 	".align 64\n"
2334 	"ret42_start:\n"
2335 	"mov $42, %eax\n"
2336 	"ret\n"
2337 	"ret42_end:\n"
2338 );
2339 
2340 static void
2341 diagnose_ept_violation_qual(u64 expected, u64 actual)
2342 {
2343 
2344 #define DIAGNOSE(flag)							\
2345 do {									\
2346 	if ((expected & flag) != (actual & flag))			\
2347 		printf(#flag " %sexpected\n",				\
2348 		       (expected & flag) ? "" : "un");			\
2349 } while (0)
2350 
2351 	DIAGNOSE(EPT_VLT_RD);
2352 	DIAGNOSE(EPT_VLT_WR);
2353 	DIAGNOSE(EPT_VLT_FETCH);
2354 	DIAGNOSE(EPT_VLT_PERM_RD);
2355 	DIAGNOSE(EPT_VLT_PERM_WR);
2356 	DIAGNOSE(EPT_VLT_PERM_EX);
2357 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2358 	DIAGNOSE(EPT_VLT_PADDR);
2359 
2360 #undef DIAGNOSE
2361 }
2362 
2363 static void do_ept_access_op(enum ept_access_op op)
2364 {
2365 	ept_access_test_data.op = op;
2366 	enter_guest();
2367 }
2368 
2369 /*
2370  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2371  * needed by tests that modify guest PTEs.
2372  */
2373 static void ept_access_test_guest_flush_tlb(void)
2374 {
2375 	do_ept_access_op(OP_FLUSH_TLB);
2376 	skip_exit_vmcall();
2377 }
2378 
2379 /*
2380  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2381  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2382  * a huge page.
2383  */
2384 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2385 				 unsigned long clear, unsigned long set)
2386 {
2387 	struct ept_access_test_data *data = &ept_access_test_data;
2388 	unsigned long orig_pte;
2389 	unsigned long pte;
2390 
2391 	/* Screw with the mapping at the requested level. */
2392 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2393 	pte = orig_pte;
2394 	if (mkhuge)
2395 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2396 	else
2397 		pte = orig_pte;
2398 	pte = (pte & ~clear) | set;
2399 	set_ept_pte(pml4, gpa, level, pte);
2400 	invept(INVEPT_SINGLE, eptp);
2401 
2402 	return orig_pte;
2403 }
2404 
2405 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2406 {
2407 	set_ept_pte(pml4, gpa, level, orig_pte);
2408 	invept(INVEPT_SINGLE, eptp);
2409 }
2410 
2411 static void do_ept_violation(bool leaf, enum ept_access_op op,
2412 			     u64 expected_qual, u64 expected_paddr)
2413 {
2414 	u64 qual;
2415 
2416 	/* Try the access and observe the violation. */
2417 	do_ept_access_op(op);
2418 
2419 	assert_exit_reason(VMX_EPT_VIOLATION);
2420 
2421 	qual = vmcs_read(EXI_QUALIFICATION);
2422 
2423 	/* Mask undefined bits (which may later be defined in certain cases). */
2424 	qual &= ~(EPT_VLT_GUEST_USER | EPT_VLT_GUEST_RW | EPT_VLT_GUEST_EX |
2425 		 EPT_VLT_PERM_USER_EX);
2426 
2427 	diagnose_ept_violation_qual(expected_qual, qual);
2428 	TEST_EXPECT_EQ(expected_qual, qual);
2429 
2430 	#if 0
2431 	/* Disable for now otherwise every test will fail */
2432 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2433 		       (unsigned long) (
2434 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2435 	#endif
2436 	/*
2437 	 * TODO: tests that probe expected_paddr in pages other than the one at
2438 	 * the beginning of the 1g region.
2439 	 */
2440 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2441 }
2442 
2443 static void
2444 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2445 			      unsigned long set, enum ept_access_op op,
2446 			      u64 expected_qual)
2447 {
2448 	struct ept_access_test_data *data = &ept_access_test_data;
2449 	unsigned long orig_pte;
2450 
2451 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2452 
2453 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2454 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2455 					 data->gpa);
2456 
2457 	/* Fix the violation and resume the op loop. */
2458 	ept_untwiddle(data->gpa, level, orig_pte);
2459 	enter_guest();
2460 	skip_exit_vmcall();
2461 }
2462 
2463 static void
2464 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2465 		       enum ept_access_op op, u64 expected_qual)
2466 {
2467 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2468 				      expected_qual);
2469 	if (ept_huge_pages_supported(level))
2470 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2471 					      expected_qual);
2472 }
2473 
2474 static void ept_violation(unsigned long clear, unsigned long set,
2475 			  enum ept_access_op op, u64 expected_qual)
2476 {
2477 	ept_violation_at_level(1, clear, set, op, expected_qual);
2478 	ept_violation_at_level(2, clear, set, op, expected_qual);
2479 	ept_violation_at_level(3, clear, set, op, expected_qual);
2480 	ept_violation_at_level(4, clear, set, op, expected_qual);
2481 }
2482 
2483 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2484 				       u64 expected_qual)
2485 {
2486 	ept_violation(EPT_PRESENT, access, op,
2487 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2488 }
2489 
2490 /*
2491  * For translations that don't involve a GVA, that is physical address (paddr)
2492  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2493  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2494  * translations don't involve GVAs, such as when the hardware does the guest
2495  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2496  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2497  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2498  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2499  *
2500  * Note that paddr violations can also be triggered by loading PAE page tables
2501  * with wonky addresses. We don't test that yet.
2502  *
2503  * This function modifies the EPT entry that maps the GPA that the guest page
2504  * table entry mapping ept_access_test_data.gva resides on.
2505  *
2506  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2507  *
2508  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2509  *
2510  *	@op		Guest operation to perform with
2511  *			ept_access_test_data.gva.
2512  *
2513  *	@expect_violation
2514  *			Is a violation expected during the paddr access?
2515  *
2516  *	@expected_qual	Expected qualification for the EPT violation.
2517  *			EPT_VLT_PADDR should be clear.
2518  */
2519 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2520 			     enum ept_access_op op, bool expect_violation,
2521 			     u64 expected_qual)
2522 {
2523 	struct ept_access_test_data *data = &ept_access_test_data;
2524 	unsigned long *ptep;
2525 	unsigned long gpa;
2526 	unsigned long orig_epte;
2527 	unsigned long epte;
2528 	int i;
2529 
2530 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2531 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2532 	TEST_ASSERT(ptep);
2533 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2534 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2535 	ept_access_test_guest_flush_tlb();
2536 
2537 	/*
2538 	 * Now modify the access bits on the EPT entry for the GPA that the
2539 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2540 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2541 	 * constructed our test such that those other 511 PTEs aren't used by
2542 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2543 	 * PTE we're modifying is at the beginning of a 4K page and the
2544 	 * following 511 entires are also under our control (and not touched by
2545 	 * the guest).
2546 	 */
2547 	gpa = virt_to_phys(ptep);
2548 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2549 	/*
2550 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2551 	 * otherwise our level=1 twiddling below will fail. We use the
2552 	 * identity map (gpa = gpa) since page tables are shared with the host.
2553 	 */
2554 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2555 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2556 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2557 
2558 	if (expect_violation) {
2559 		do_ept_violation(/*leaf=*/true, op,
2560 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2561 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2562 		do_ept_access_op(op);
2563 	} else {
2564 		do_ept_access_op(op);
2565 		if (ept_ad_enabled()) {
2566 			for (i = EPT_PAGE_LEVEL; i > 0; i--) {
2567 				TEST_ASSERT(get_ept_pte(pml4, gpa, i, &epte));
2568 				TEST_ASSERT(epte & EPT_ACCESS_FLAG);
2569 				if (i == 1)
2570 					TEST_ASSERT(epte & EPT_DIRTY_FLAG);
2571 				else
2572 					TEST_ASSERT_EQ(epte & EPT_DIRTY_FLAG, 0);
2573 			}
2574 		}
2575 
2576 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2577 	}
2578 
2579 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2580 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2581 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2582 
2583 	skip_exit_vmcall();
2584 }
2585 
2586 static void ept_access_allowed_paddr(unsigned long ept_access,
2587 				     unsigned long pte_ad,
2588 				     enum ept_access_op op)
2589 {
2590 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2591 			 /*expected_qual=*/-1);
2592 }
2593 
2594 static void ept_access_violation_paddr(unsigned long ept_access,
2595 				       unsigned long pte_ad,
2596 				       enum ept_access_op op,
2597 				       u64 expected_qual)
2598 {
2599 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2600 			 expected_qual);
2601 }
2602 
2603 
2604 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2605 					unsigned long clear,
2606 					unsigned long set,
2607 					enum ept_access_op op)
2608 {
2609 	struct ept_access_test_data *data = &ept_access_test_data;
2610 	unsigned long orig_pte;
2611 
2612 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2613 
2614 	/* No violation. Should proceed to vmcall. */
2615 	do_ept_access_op(op);
2616 	skip_exit_vmcall();
2617 
2618 	ept_untwiddle(data->gpa, level, orig_pte);
2619 }
2620 
2621 static void ept_allowed_at_level(int level, unsigned long clear,
2622 				 unsigned long set, enum ept_access_op op)
2623 {
2624 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2625 	if (ept_huge_pages_supported(level))
2626 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2627 }
2628 
2629 static void ept_allowed(unsigned long clear, unsigned long set,
2630 			enum ept_access_op op)
2631 {
2632 	ept_allowed_at_level(1, clear, set, op);
2633 	ept_allowed_at_level(2, clear, set, op);
2634 	ept_allowed_at_level(3, clear, set, op);
2635 	ept_allowed_at_level(4, clear, set, op);
2636 }
2637 
2638 static void ept_ignored_bit(int bit)
2639 {
2640 	/* Set the bit. */
2641 	ept_allowed(0, 1ul << bit, OP_READ);
2642 	ept_allowed(0, 1ul << bit, OP_WRITE);
2643 	ept_allowed(0, 1ul << bit, OP_EXEC);
2644 
2645 	/* Clear the bit. */
2646 	ept_allowed(1ul << bit, 0, OP_READ);
2647 	ept_allowed(1ul << bit, 0, OP_WRITE);
2648 	ept_allowed(1ul << bit, 0, OP_EXEC);
2649 }
2650 
2651 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2652 {
2653 	ept_allowed(EPT_PRESENT, access, op);
2654 }
2655 
2656 
2657 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2658 					     unsigned long clear,
2659 					     unsigned long set,
2660 					     enum ept_access_op op)
2661 {
2662 	struct ept_access_test_data *data = &ept_access_test_data;
2663 	unsigned long orig_pte;
2664 
2665 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2666 
2667 	do_ept_access_op(op);
2668 	assert_exit_reason(VMX_EPT_MISCONFIG);
2669 
2670 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2671 	#if 0
2672 	/* broken: */
2673 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2674 	#endif
2675 	#if 0
2676 	/*
2677 	 * broken:
2678 	 * According to description of exit qual for EPT violation,
2679 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2680 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2681 	 * to be set for msiconfig.
2682 	 */
2683 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2684 		       (unsigned long) (
2685 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2686 	#endif
2687 
2688 	/* Fix the violation and resume the op loop. */
2689 	ept_untwiddle(data->gpa, level, orig_pte);
2690 	enter_guest();
2691 	skip_exit_vmcall();
2692 }
2693 
2694 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2695 					  unsigned long clear,
2696 					  unsigned long set)
2697 {
2698 	/* The op shouldn't matter (read, write, exec), so try them all! */
2699 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2700 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2701 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2702 }
2703 
2704 static void ept_misconfig_at_level(int level, unsigned long clear,
2705 				   unsigned long set)
2706 {
2707 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2708 	if (ept_huge_pages_supported(level))
2709 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2710 }
2711 
2712 static void ept_misconfig(unsigned long clear, unsigned long set)
2713 {
2714 	ept_misconfig_at_level(1, clear, set);
2715 	ept_misconfig_at_level(2, clear, set);
2716 	ept_misconfig_at_level(3, clear, set);
2717 	ept_misconfig_at_level(4, clear, set);
2718 }
2719 
2720 static void ept_access_misconfig(unsigned long access)
2721 {
2722 	ept_misconfig(EPT_PRESENT, access);
2723 }
2724 
2725 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2726 {
2727 	/* Setting the bit causes a misconfig. */
2728 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2729 
2730 	/* Making the entry non-present turns reserved bits into ignored. */
2731 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2732 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2733 }
2734 
2735 static void ept_reserved_bit_at_level_huge(int level, int bit)
2736 {
2737 	/* Setting the bit causes a misconfig. */
2738 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2739 
2740 	/* Making the entry non-present turns reserved bits into ignored. */
2741 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2742 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2743 }
2744 
2745 static void ept_reserved_bit_at_level(int level, int bit)
2746 {
2747 	/* Setting the bit causes a misconfig. */
2748 	ept_misconfig_at_level(level, 0, 1ul << bit);
2749 
2750 	/* Making the entry non-present turns reserved bits into ignored. */
2751 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2752 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2753 }
2754 
2755 static void ept_reserved_bit(int bit)
2756 {
2757 	ept_reserved_bit_at_level(1, bit);
2758 	ept_reserved_bit_at_level(2, bit);
2759 	ept_reserved_bit_at_level(3, bit);
2760 	ept_reserved_bit_at_level(4, bit);
2761 }
2762 
2763 #define PAGE_2M_ORDER 9
2764 #define PAGE_1G_ORDER 18
2765 
2766 static void *get_1g_page(void)
2767 {
2768 	static void *alloc;
2769 
2770 	if (!alloc)
2771 		alloc = alloc_pages(PAGE_1G_ORDER);
2772 	return alloc;
2773 }
2774 
2775 static void ept_access_test_teardown(void *unused)
2776 {
2777 	/* Exit the guest cleanly. */
2778 	do_ept_access_op(OP_EXIT);
2779 }
2780 
2781 static void ept_access_test_guest(void)
2782 {
2783 	struct ept_access_test_data *data = &ept_access_test_data;
2784 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2785 
2786 	while (true) {
2787 		switch (data->op) {
2788 		case OP_READ:
2789 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2790 			break;
2791 		case OP_WRITE:
2792 			*data->gva = MAGIC_VAL_2;
2793 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2794 			*data->gva = MAGIC_VAL_1;
2795 			break;
2796 		case OP_EXEC:
2797 			TEST_ASSERT_EQ(42, code());
2798 			break;
2799 		case OP_FLUSH_TLB:
2800 			write_cr3(read_cr3());
2801 			break;
2802 		case OP_EXIT:
2803 			return;
2804 		default:
2805 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2806 		}
2807 		vmcall();
2808 	}
2809 }
2810 
2811 static void ept_access_test_setup(void)
2812 {
2813 	struct ept_access_test_data *data = &ept_access_test_data;
2814 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2815 	unsigned long size = npages * PAGE_SIZE;
2816 	unsigned long *page_table = current_page_table();
2817 	unsigned long pte;
2818 
2819 	if (setup_ept(false))
2820 		test_skip("EPT not supported");
2821 
2822 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2823 	if (cpuid_maxphyaddr() < 40)
2824 		test_skip("Test needs MAXPHYADDR >= 40");
2825 
2826 	test_set_guest(ept_access_test_guest);
2827 	test_add_teardown(ept_access_test_teardown, NULL);
2828 
2829 	data->hva = get_1g_page();
2830 	TEST_ASSERT(data->hva);
2831 	data->hpa = virt_to_phys(data->hva);
2832 
2833 	data->gpa = 1ul << 39;
2834 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2835 				   size);
2836 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2837 	install_pages(page_table, data->gpa, size, data->gva);
2838 
2839 	/*
2840 	 * Make sure nothing's mapped here so the tests that screw with the
2841 	 * pml4 entry don't inadvertently break something.
2842 	 */
2843 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2844 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2845 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2846 
2847 	data->hva[0] = MAGIC_VAL_1;
2848 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2849 }
2850 
2851 static void ept_access_test_not_present(void)
2852 {
2853 	ept_access_test_setup();
2854 	/* --- */
2855 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2856 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2857 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2858 }
2859 
2860 static void ept_access_test_read_only(void)
2861 {
2862 	ept_access_test_setup();
2863 
2864 	/* r-- */
2865 	ept_access_allowed(EPT_RA, OP_READ);
2866 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2867 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2868 }
2869 
2870 static void ept_access_test_write_only(void)
2871 {
2872 	ept_access_test_setup();
2873 	/* -w- */
2874 	ept_access_misconfig(EPT_WA);
2875 }
2876 
2877 static void ept_access_test_read_write(void)
2878 {
2879 	ept_access_test_setup();
2880 	/* rw- */
2881 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2882 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2883 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2884 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2885 }
2886 
2887 
2888 static void ept_access_test_execute_only(void)
2889 {
2890 	ept_access_test_setup();
2891 	/* --x */
2892 	if (ept_execute_only_supported()) {
2893 		ept_access_violation(EPT_EA, OP_READ,
2894 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2895 		ept_access_violation(EPT_EA, OP_WRITE,
2896 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2897 		ept_access_allowed(EPT_EA, OP_EXEC);
2898 	} else {
2899 		ept_access_misconfig(EPT_EA);
2900 	}
2901 }
2902 
2903 static void ept_access_test_read_execute(void)
2904 {
2905 	ept_access_test_setup();
2906 	/* r-x */
2907 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2908 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2909 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2910 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2911 }
2912 
2913 static void ept_access_test_write_execute(void)
2914 {
2915 	ept_access_test_setup();
2916 	/* -wx */
2917 	ept_access_misconfig(EPT_WA | EPT_EA);
2918 }
2919 
2920 static void ept_access_test_read_write_execute(void)
2921 {
2922 	ept_access_test_setup();
2923 	/* rwx */
2924 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2925 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2926 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2927 }
2928 
2929 static void ept_access_test_reserved_bits(void)
2930 {
2931 	int i;
2932 	int maxphyaddr;
2933 
2934 	ept_access_test_setup();
2935 
2936 	/* Reserved bits above maxphyaddr. */
2937 	maxphyaddr = cpuid_maxphyaddr();
2938 	for (i = maxphyaddr; i <= 51; i++) {
2939 		report_prefix_pushf("reserved_bit=%d", i);
2940 		ept_reserved_bit(i);
2941 		report_prefix_pop();
2942 	}
2943 
2944 	/* Level-specific reserved bits. */
2945 	ept_reserved_bit_at_level_nohuge(2, 3);
2946 	ept_reserved_bit_at_level_nohuge(2, 4);
2947 	ept_reserved_bit_at_level_nohuge(2, 5);
2948 	ept_reserved_bit_at_level_nohuge(2, 6);
2949 	/* 2M alignment. */
2950 	for (i = 12; i < 20; i++) {
2951 		report_prefix_pushf("reserved_bit=%d", i);
2952 		ept_reserved_bit_at_level_huge(2, i);
2953 		report_prefix_pop();
2954 	}
2955 	ept_reserved_bit_at_level_nohuge(3, 3);
2956 	ept_reserved_bit_at_level_nohuge(3, 4);
2957 	ept_reserved_bit_at_level_nohuge(3, 5);
2958 	ept_reserved_bit_at_level_nohuge(3, 6);
2959 	/* 1G alignment. */
2960 	for (i = 12; i < 29; i++) {
2961 		report_prefix_pushf("reserved_bit=%d", i);
2962 		ept_reserved_bit_at_level_huge(3, i);
2963 		report_prefix_pop();
2964 	}
2965 	ept_reserved_bit_at_level(4, 3);
2966 	ept_reserved_bit_at_level(4, 4);
2967 	ept_reserved_bit_at_level(4, 5);
2968 	ept_reserved_bit_at_level(4, 6);
2969 	ept_reserved_bit_at_level(4, 7);
2970 }
2971 
2972 static void ept_access_test_ignored_bits(void)
2973 {
2974 	ept_access_test_setup();
2975 	/*
2976 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2977 	 * far as translation is concerned even if AD bits are enabled in the
2978 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2979 	 * control is 0.
2980 	 */
2981 	ept_ignored_bit(8);
2982 	ept_ignored_bit(9);
2983 	ept_ignored_bit(10);
2984 	ept_ignored_bit(11);
2985 	ept_ignored_bit(52);
2986 	ept_ignored_bit(53);
2987 	ept_ignored_bit(54);
2988 	ept_ignored_bit(55);
2989 	ept_ignored_bit(56);
2990 	ept_ignored_bit(57);
2991 	ept_ignored_bit(58);
2992 	ept_ignored_bit(59);
2993 	ept_ignored_bit(60);
2994 	ept_ignored_bit(61);
2995 	ept_ignored_bit(62);
2996 	ept_ignored_bit(63);
2997 }
2998 
2999 static void ept_access_test_paddr_not_present_ad_disabled(void)
3000 {
3001 	ept_access_test_setup();
3002 	ept_disable_ad_bits();
3003 
3004 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
3005 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
3006 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
3007 }
3008 
3009 static void ept_access_test_paddr_not_present_ad_enabled(void)
3010 {
3011 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
3012 
3013 	ept_access_test_setup();
3014 	ept_enable_ad_bits_or_skip_test();
3015 
3016 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
3017 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
3018 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
3019 }
3020 
3021 static void ept_access_test_paddr_read_only_ad_disabled(void)
3022 {
3023 	/*
3024 	 * When EPT AD bits are disabled, all accesses to guest paging
3025 	 * structures are reported separately as a read and (after
3026 	 * translation of the GPA to host physical address) a read+write
3027 	 * if the A/D bits have to be set.
3028 	 */
3029 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3030 
3031 	ept_access_test_setup();
3032 	ept_disable_ad_bits();
3033 
3034 	/* Can't update A bit, so all accesses fail. */
3035 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3036 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3037 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3038 	/* AD bits disabled, so only writes try to update the D bit. */
3039 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
3040 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3041 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
3042 	/* Both A and D already set, so read-only is OK. */
3043 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
3044 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
3045 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
3046 }
3047 
3048 static void ept_access_test_paddr_read_only_ad_enabled(void)
3049 {
3050 	/*
3051 	 * When EPT AD bits are enabled, all accesses to guest paging
3052 	 * structures are considered writes as far as EPT translation
3053 	 * is concerned.
3054 	 */
3055 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3056 
3057 	ept_access_test_setup();
3058 	ept_enable_ad_bits_or_skip_test();
3059 
3060 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3061 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3062 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3063 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
3064 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3065 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
3066 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
3067 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
3068 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
3069 }
3070 
3071 static void ept_access_test_paddr_read_write(void)
3072 {
3073 	ept_access_test_setup();
3074 	/* Read-write access to paging structure. */
3075 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
3076 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
3077 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
3078 }
3079 
3080 static void ept_access_test_paddr_read_write_execute(void)
3081 {
3082 	ept_access_test_setup();
3083 	/* RWX access to paging structure. */
3084 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
3085 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
3086 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
3087 }
3088 
3089 static void ept_access_test_paddr_read_execute_ad_disabled(void)
3090 {
3091   	/*
3092 	 * When EPT AD bits are disabled, all accesses to guest paging
3093 	 * structures are reported separately as a read and (after
3094 	 * translation of the GPA to host physical address) a read+write
3095 	 * if the A/D bits have to be set.
3096 	 */
3097 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3098 
3099 	ept_access_test_setup();
3100 	ept_disable_ad_bits();
3101 
3102 	/* Can't update A bit, so all accesses fail. */
3103 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3104 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3105 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3106 	/* AD bits disabled, so only writes try to update the D bit. */
3107 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
3108 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3109 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3110 	/* Both A and D already set, so read-only is OK. */
3111 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3112 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3113 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3114 }
3115 
3116 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3117 {
3118 	/*
3119 	 * When EPT AD bits are enabled, all accesses to guest paging
3120 	 * structures are considered writes as far as EPT translation
3121 	 * is concerned.
3122 	 */
3123 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3124 
3125 	ept_access_test_setup();
3126 	ept_enable_ad_bits_or_skip_test();
3127 
3128 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3129 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3130 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3131 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3132 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3133 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3134 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3135 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3136 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3137 }
3138 
3139 static void ept_access_test_paddr_not_present_page_fault(void)
3140 {
3141 	ept_access_test_setup();
3142 	/*
3143 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3144 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3145 	 * Thus guest page fault before host takes EPT violation for trying to
3146 	 * update A bit.
3147 	 */
3148 }
3149 
3150 static void ept_access_test_force_2m_page(void)
3151 {
3152 	ept_access_test_setup();
3153 
3154 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3155 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3156 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3157 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3158 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3159 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3160 }
3161 
3162 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3163 {
3164 	if (!is_invvpid_type_supported(type))
3165 		return false;
3166 
3167 	if (vpid >> 16)
3168 		return false;
3169 
3170 	if (type != INVVPID_ALL && !vpid)
3171 		return false;
3172 
3173 	if (type == INVVPID_ADDR && !is_canonical(gla))
3174 		return false;
3175 
3176 	return true;
3177 }
3178 
3179 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3180 {
3181 	int rc;
3182 	bool valid = invvpid_valid(type, vpid, gla);
3183 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3184 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3185 	/*
3186 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3187 	 * that we can tell if it is updated by INVVPID.
3188 	 */
3189 	vmcs_read(~0);
3190 	rc = __invvpid(type, vpid, gla);
3191 	report(!rc == valid, "INVVPID type %ld VPID %lx GLA %lx %s", type,
3192 	       vpid, gla,
3193 	       valid ? "passes" : "fails");
3194 	report(vmcs_read(VMX_INST_ERROR) == expected,
3195 	       "After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3196 	       rc ? "failed" : "successful",
3197 	       expected, vmcs_read(VMX_INST_ERROR));
3198 }
3199 
3200 static void ds_invvpid(void *data)
3201 {
3202 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3203 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3204 
3205 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3206 	asm volatile("invvpid %0, %1"
3207 		     :
3208 		     : "m"(*(struct invvpid_operand *)data),
3209 		       "r"(type));
3210 }
3211 
3212 /*
3213  * The SS override is ignored in 64-bit mode, so we use an addressing
3214  * mode with %rsp as the base register to generate an implicit SS
3215  * reference.
3216  */
3217 static void ss_invvpid(void *data)
3218 {
3219 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3220 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3221 
3222 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3223 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3224 		     : "+r"(data)
3225 		     : "r"(type));
3226 }
3227 
3228 static void invvpid_test_gp(void)
3229 {
3230 	bool fault;
3231 
3232 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3233 				   (void *)NONCANONICAL);
3234 	report(fault, "INVVPID with non-canonical DS operand raises #GP");
3235 }
3236 
3237 static void invvpid_test_ss(void)
3238 {
3239 	bool fault;
3240 
3241 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3242 				   (void *)NONCANONICAL);
3243 	report(fault, "INVVPID with non-canonical SS operand raises #SS");
3244 }
3245 
3246 static void invvpid_test_pf(void)
3247 {
3248 	void *vpage = alloc_vpage();
3249 	bool fault;
3250 
3251 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3252 	report(fault, "INVVPID with unmapped operand raises #PF");
3253 }
3254 
3255 static void try_compat_invvpid(void *unused)
3256 {
3257 	struct far_pointer32 fp = {
3258 		.offset = (uintptr_t)&&invvpid,
3259 		.selector = KERNEL_CS32,
3260 	};
3261 	uintptr_t rsp;
3262 
3263 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
3264 
3265 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3266 			"Code address too high.");
3267 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3268 
3269 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3270 	return;
3271 invvpid:
3272 	asm volatile (".code32;"
3273 		      "invvpid (%eax), %eax;"
3274 		      "lret;"
3275 		      ".code64");
3276 	__builtin_unreachable();
3277 }
3278 
3279 static void invvpid_test_compatibility_mode(void)
3280 {
3281 	bool fault;
3282 
3283 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3284 	report(fault, "Compatibility mode INVVPID raises #UD");
3285 }
3286 
3287 static void invvpid_test_not_in_vmx_operation(void)
3288 {
3289 	bool fault;
3290 
3291 	TEST_ASSERT(!vmx_off());
3292 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3293 	report(fault, "INVVPID outside of VMX operation raises #UD");
3294 	TEST_ASSERT(!vmx_on());
3295 }
3296 
3297 /*
3298  * This does not test real-address mode, virtual-8086 mode, protected mode,
3299  * or CPL > 0.
3300  */
3301 static void invvpid_test(void)
3302 {
3303 	u64 msr;
3304 	int i;
3305 	unsigned types = 0;
3306 	unsigned type;
3307 
3308 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
3309 	    !(ctrl_cpu_rev[1].clr & CPU_VPID))
3310 		test_skip("VPID not supported");
3311 
3312 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3313 
3314 	if (!(msr & VPID_CAP_INVVPID))
3315 		test_skip("INVVPID not supported.\n");
3316 
3317 	if (is_invvpid_type_supported(INVVPID_ADDR))
3318 		types |= 1u << INVVPID_ADDR;
3319 	if (is_invvpid_type_supported(INVVPID_CONTEXT_GLOBAL))
3320 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3321 	if (is_invvpid_type_supported(INVVPID_ALL))
3322 		types |= 1u << INVVPID_ALL;
3323 	if (is_invvpid_type_supported(INVVPID_CONTEXT_LOCAL))
3324 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3325 
3326 	if (!types)
3327 		test_skip("No INVVPID types supported.\n");
3328 
3329 	for (i = -127; i < 128; i++)
3330 		try_invvpid(i, 0xffff, 0);
3331 
3332 	/*
3333 	 * VPID must not be more than 16 bits.
3334 	 */
3335 	for (i = 0; i < 64; i++)
3336 		for (type = 0; type < 4; type++)
3337 			if (types & (1u << type))
3338 				try_invvpid(type, 1ul << i, 0);
3339 
3340 	/*
3341 	 * VPID must not be zero, except for "all contexts."
3342 	 */
3343 	for (type = 0; type < 4; type++)
3344 		if (types & (1u << type))
3345 			try_invvpid(type, 0, 0);
3346 
3347 	/*
3348 	 * The gla operand is only validated for single-address INVVPID.
3349 	 */
3350 	if (types & (1u << INVVPID_ADDR))
3351 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3352 
3353 	invvpid_test_gp();
3354 	invvpid_test_ss();
3355 	invvpid_test_pf();
3356 	invvpid_test_compatibility_mode();
3357 	invvpid_test_not_in_vmx_operation();
3358 }
3359 
3360 /*
3361  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3362  * at least as far as the guest-state checks. Returns false if the
3363  * VMLAUNCH fails early and execution falls through to the next
3364  * instruction.
3365  */
3366 static bool vmlaunch_succeeds(void)
3367 {
3368 	u32 exit_reason;
3369 
3370 	/*
3371 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3372 	 * unsupported VMCS component"). The caller can then check
3373 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3374 	 */
3375 	vmcs_write(~0u, 0);
3376 
3377 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3378 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3379 				   :
3380 				   : "r" ((u64)HOST_RSP)
3381 				   : "cc", "memory"
3382 				   : success);
3383 	return false;
3384 success:
3385 	exit_reason = vmcs_read(EXI_REASON);
3386 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3387 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3388 	return true;
3389 }
3390 
3391 /*
3392  * Try to launch the current VMCS.
3393  */
3394 static void test_vmx_vmlaunch(u32 xerror)
3395 {
3396 	bool success = vmlaunch_succeeds();
3397 	u32 vmx_inst_err;
3398 
3399 	report(success == !xerror, "vmlaunch %s",
3400 	       !xerror ? "succeeds" : "fails");
3401 	if (!success && xerror) {
3402 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3403 		report(vmx_inst_err == xerror,
3404 		       "VMX inst error is %d (actual %d)", xerror,
3405 		       vmx_inst_err);
3406 	}
3407 }
3408 
3409 /*
3410  * Try to launch the current VMCS, and expect one of two possible
3411  * errors (or success) codes.
3412  */
3413 static void test_vmx_vmlaunch2(u32 xerror1, u32 xerror2)
3414 {
3415 	bool success = vmlaunch_succeeds();
3416 	u32 vmx_inst_err;
3417 
3418 	if (!xerror1 == !xerror2)
3419 		report(success == !xerror1, "vmlaunch %s",
3420 		       !xerror1 ? "succeeds" : "fails");
3421 
3422 	if (!success && (xerror1 || xerror2)) {
3423 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3424 		report(vmx_inst_err == xerror1 || vmx_inst_err == xerror2,
3425 		       "VMX inst error is %d or %d (actual %d)", xerror1,
3426 		       xerror2, vmx_inst_err);
3427 	}
3428 }
3429 
3430 static void test_vmx_invalid_controls(void)
3431 {
3432 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3433 }
3434 
3435 static void test_vmx_valid_controls(void)
3436 {
3437 	test_vmx_vmlaunch(0);
3438 }
3439 
3440 /*
3441  * Test a particular value of a VM-execution control bit, if the value
3442  * is required or if the value is zero.
3443  */
3444 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3445 				    enum Encoding encoding, unsigned bit,
3446 				    unsigned val)
3447 {
3448 	u32 mask = 1u << bit;
3449 	bool expected;
3450 	u32 controls;
3451 
3452 	if (msr.set & mask)
3453 		TEST_ASSERT(msr.clr & mask);
3454 
3455 	/*
3456 	 * We can't arbitrarily turn on a control bit, because it may
3457 	 * introduce dependencies on other VMCS fields. So, we only
3458 	 * test turning on bits that have a required setting.
3459 	 */
3460 	if (val && (msr.clr & mask) && !(msr.set & mask))
3461 		return;
3462 
3463 	report_prefix_pushf("%s %s bit %d",
3464 			    val ? "Set" : "Clear", name, bit);
3465 
3466 	controls = vmcs_read(encoding);
3467 	if (val) {
3468 		vmcs_write(encoding, msr.set | mask);
3469 		expected = (msr.clr & mask);
3470 	} else {
3471 		vmcs_write(encoding, msr.set & ~mask);
3472 		expected = !(msr.set & mask);
3473 	}
3474 	if (expected)
3475 		test_vmx_valid_controls();
3476 	else
3477 		test_vmx_invalid_controls();
3478 	vmcs_write(encoding, controls);
3479 	report_prefix_pop();
3480 }
3481 
3482 /*
3483  * Test reserved values of a VM-execution control bit, based on the
3484  * allowed bit settings from the corresponding VMX capability MSR.
3485  */
3486 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3487 			      enum Encoding encoding, unsigned bit)
3488 {
3489 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3490 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3491 }
3492 
3493 /*
3494  * Reserved bits in the pin-based VM-execution controls must be set
3495  * properly. Software may consult the VMX capability MSRs to determine
3496  * the proper settings.
3497  * [Intel SDM]
3498  */
3499 static void test_pin_based_ctls(void)
3500 {
3501 	unsigned bit;
3502 
3503 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3504 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3505 	for (bit = 0; bit < 32; bit++)
3506 		test_rsvd_ctl_bit("pin-based controls",
3507 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3508 }
3509 
3510 /*
3511  * Reserved bits in the primary processor-based VM-execution controls
3512  * must be set properly. Software may consult the VMX capability MSRs
3513  * to determine the proper settings.
3514  * [Intel SDM]
3515  */
3516 static void test_primary_processor_based_ctls(void)
3517 {
3518 	unsigned bit;
3519 
3520 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3521 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3522 	for (bit = 0; bit < 32; bit++)
3523 		test_rsvd_ctl_bit("primary processor-based controls",
3524 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3525 }
3526 
3527 /*
3528  * If the "activate secondary controls" primary processor-based
3529  * VM-execution control is 1, reserved bits in the secondary
3530  * processor-based VM-execution controls must be cleared. Software may
3531  * consult the VMX capability MSRs to determine which bits are
3532  * reserved.
3533  * If the "activate secondary controls" primary processor-based
3534  * VM-execution control is 0 (or if the processor does not support the
3535  * 1-setting of that control), no checks are performed on the
3536  * secondary processor-based VM-execution controls.
3537  * [Intel SDM]
3538  */
3539 static void test_secondary_processor_based_ctls(void)
3540 {
3541 	u32 primary;
3542 	u32 secondary;
3543 	unsigned bit;
3544 
3545 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3546 		return;
3547 
3548 	primary = vmcs_read(CPU_EXEC_CTRL0);
3549 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3550 
3551 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3552 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3553 	for (bit = 0; bit < 32; bit++)
3554 		test_rsvd_ctl_bit("secondary processor-based controls",
3555 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3556 
3557 	/*
3558 	 * When the "activate secondary controls" VM-execution control
3559 	 * is clear, there are no checks on the secondary controls.
3560 	 */
3561 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3562 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3563 	report(vmlaunch_succeeds(),
3564 	       "Secondary processor-based controls ignored");
3565 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3566 	vmcs_write(CPU_EXEC_CTRL0, primary);
3567 }
3568 
3569 static void try_cr3_target_count(unsigned i, unsigned max)
3570 {
3571 	report_prefix_pushf("CR3 target count 0x%x", i);
3572 	vmcs_write(CR3_TARGET_COUNT, i);
3573 	if (i <= max)
3574 		test_vmx_valid_controls();
3575 	else
3576 		test_vmx_invalid_controls();
3577 	report_prefix_pop();
3578 }
3579 
3580 /*
3581  * The CR3-target count must not be greater than 4. Future processors
3582  * may support a different number of CR3-target values. Software
3583  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3584  * number of values supported.
3585  * [Intel SDM]
3586  */
3587 static void test_cr3_targets(void)
3588 {
3589 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3590 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3591 	unsigned i;
3592 
3593 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3594 	TEST_ASSERT(supported_targets <= 256);
3595 
3596 	try_cr3_target_count(-1u, supported_targets);
3597 	try_cr3_target_count(0x80000000, supported_targets);
3598 	try_cr3_target_count(0x7fffffff, supported_targets);
3599 	for (i = 0; i <= supported_targets + 1; i++)
3600 		try_cr3_target_count(i, supported_targets);
3601 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3602 
3603 	/* VMWRITE to nonexistent target fields should fail. */
3604 	for (i = supported_targets; i < 256; i++)
3605 		TEST_ASSERT(vmcs_write(CR3_TARGET_0 + i*2, 0));
3606 }
3607 
3608 /*
3609  * Test a particular address setting in the VMCS
3610  */
3611 static void test_vmcs_addr(const char *name,
3612 			   enum Encoding encoding,
3613 			   u64 align,
3614 			   bool ignored,
3615 			   bool skip_beyond_mapped_ram,
3616 			   u64 addr)
3617 {
3618 	report_prefix_pushf("%s = %lx", name, addr);
3619 	vmcs_write(encoding, addr);
3620 	if (skip_beyond_mapped_ram &&
3621 	    addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3622 	    addr < (1ul << cpuid_maxphyaddr()))
3623 		printf("Skipping physical address beyond mapped RAM\n");
3624 	else if (ignored || (IS_ALIGNED(addr, align) &&
3625 	    addr < (1ul << cpuid_maxphyaddr())))
3626 		test_vmx_valid_controls();
3627 	else
3628 		test_vmx_invalid_controls();
3629 	report_prefix_pop();
3630 }
3631 
3632 /*
3633  * Test interesting values for a VMCS address
3634  */
3635 static void test_vmcs_addr_values(const char *name,
3636 				  enum Encoding encoding,
3637 				  u64 align,
3638 				  bool ignored,
3639 				  bool skip_beyond_mapped_ram,
3640 				  u32 bit_start, u32 bit_end)
3641 {
3642 	unsigned i;
3643 	u64 orig_val = vmcs_read(encoding);
3644 
3645 	for (i = bit_start; i <= bit_end; i++)
3646 		test_vmcs_addr(name, encoding, align, ignored,
3647 			       skip_beyond_mapped_ram, 1ul << i);
3648 
3649 	test_vmcs_addr(name, encoding, align, ignored,
3650 		       skip_beyond_mapped_ram, PAGE_SIZE - 1);
3651 	test_vmcs_addr(name, encoding, align, ignored,
3652 		       skip_beyond_mapped_ram, PAGE_SIZE);
3653 	test_vmcs_addr(name, encoding, align, ignored,
3654 		       skip_beyond_mapped_ram,
3655 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3656 	test_vmcs_addr(name, encoding, align, ignored,
3657 		       skip_beyond_mapped_ram, -1ul);
3658 
3659 	vmcs_write(encoding, orig_val);
3660 }
3661 
3662 /*
3663  * Test a physical address reference in the VMCS, when the corresponding
3664  * feature is enabled and when the corresponding feature is disabled.
3665  */
3666 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3667 				     const char *field_name,
3668 				     const char *control_name, u64 align,
3669 				     bool skip_beyond_mapped_ram,
3670 				     bool control_primary)
3671 {
3672 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3673 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3674 	u64 page_addr;
3675 
3676 	if (control_primary) {
3677 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3678 			return;
3679 	} else {
3680 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3681 			return;
3682 	}
3683 
3684 	page_addr = vmcs_read(field);
3685 
3686 	report_prefix_pushf("%s enabled", control_name);
3687 	if (control_primary) {
3688 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3689 	} else {
3690 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3691 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3692 	}
3693 
3694 	test_vmcs_addr_values(field_name, field, align, false,
3695 			      skip_beyond_mapped_ram, 0, 63);
3696 	report_prefix_pop();
3697 
3698 	report_prefix_pushf("%s disabled", control_name);
3699 	if (control_primary) {
3700 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3701 	} else {
3702 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3703 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3704 	}
3705 
3706 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3707 	report_prefix_pop();
3708 
3709 	vmcs_write(field, page_addr);
3710 	vmcs_write(CPU_EXEC_CTRL0, primary);
3711 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3712 }
3713 
3714 /*
3715  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3716  * each I/O-bitmap address must be 0. Neither address should set any
3717  * bits beyond the processor's physical-address width.
3718  * [Intel SDM]
3719  */
3720 static void test_io_bitmaps(void)
3721 {
3722 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3723 				 "I/O bitmap A", "Use I/O bitmaps",
3724 				 PAGE_SIZE, false, true);
3725 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3726 				 "I/O bitmap B", "Use I/O bitmaps",
3727 				 PAGE_SIZE, false, true);
3728 }
3729 
3730 /*
3731  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3732  * the MSR-bitmap address must be 0. The address should not set any
3733  * bits beyond the processor's physical-address width.
3734  * [Intel SDM]
3735  */
3736 static void test_msr_bitmap(void)
3737 {
3738 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3739 				 "MSR bitmap", "Use MSR bitmaps",
3740 				 PAGE_SIZE, false, true);
3741 }
3742 
3743 /*
3744  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3745  * address must satisfy the following checks:
3746  * - Bits 11:0 of the address must be 0.
3747  * - The address should not set any bits beyond the processor's
3748  *   physical-address width.
3749  * [Intel SDM]
3750  */
3751 static void test_apic_virt_addr(void)
3752 {
3753 	/*
3754 	 * Ensure the processor will never use the virtual-APIC page, since
3755 	 * we will point it to invalid RAM.  Otherwise KVM is puzzled about
3756 	 * what we're trying to achieve and fails vmentry.
3757 	 */
3758 	u32 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
3759 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0 | CPU_CR8_LOAD | CPU_CR8_STORE);
3760 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3761 				 "virtual-APIC address", "Use TPR shadow",
3762 				 PAGE_SIZE, false, true);
3763 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
3764 }
3765 
3766 /*
3767  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3768  * APIC-access address must satisfy the following checks:
3769  *  - Bits 11:0 of the address must be 0.
3770  *  - The address should not set any bits beyond the processor's
3771  *    physical-address width.
3772  * [Intel SDM]
3773  */
3774 static void test_apic_access_addr(void)
3775 {
3776 	void *apic_access_page = alloc_page();
3777 
3778 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3779 
3780 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3781 				 "APIC-access address",
3782 				 "virtualize APIC-accesses", PAGE_SIZE,
3783 				 true, false);
3784 }
3785 
3786 static bool set_bit_pattern(u8 mask, u32 *secondary)
3787 {
3788 	u8 i;
3789 	bool flag = false;
3790 	u32 test_bits[3] = {
3791 		CPU_VIRT_X2APIC,
3792 		CPU_APIC_REG_VIRT,
3793 		CPU_VINTD
3794 	};
3795 
3796         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3797 		if ((mask & (1u << i)) &&
3798 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3799 			*secondary |= test_bits[i];
3800 			flag = true;
3801 		}
3802 	}
3803 
3804 	return (flag);
3805 }
3806 
3807 /*
3808  * If the "use TPR shadow" VM-execution control is 0, the following
3809  * VM-execution controls must also be 0:
3810  * 	- virtualize x2APIC mode
3811  *	- APIC-register virtualization
3812  *	- virtual-interrupt delivery
3813  *    [Intel SDM]
3814  *
3815  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3816  *    "virtualize APIC accesses" VM-execution control must be 0.
3817  *    [Intel SDM]
3818  */
3819 static void test_apic_virtual_ctls(void)
3820 {
3821 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3822 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3823 	u32 primary = saved_primary;
3824 	u32 secondary = saved_secondary;
3825 	bool ctrl = false;
3826 	char str[10] = "disabled";
3827 	u8 i = 0, j;
3828 
3829 	/*
3830 	 * First test
3831 	 */
3832 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3833 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3834 		return;
3835 
3836 	primary |= CPU_SECONDARY;
3837 	primary &= ~CPU_TPR_SHADOW;
3838 	vmcs_write(CPU_EXEC_CTRL0, primary);
3839 
3840 	while (1) {
3841 		for (j = 1; j < 8; j++) {
3842 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3843 			if (primary & CPU_TPR_SHADOW) {
3844 				ctrl = true;
3845 			} else {
3846 				if (! set_bit_pattern(j, &secondary))
3847 					ctrl = true;
3848 				else
3849 					ctrl = false;
3850 			}
3851 
3852 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3853 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3854 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3855 			if (ctrl)
3856 				test_vmx_valid_controls();
3857 			else
3858 				test_vmx_invalid_controls();
3859 			report_prefix_pop();
3860 		}
3861 
3862 		if (i == 1)
3863 			break;
3864 		i++;
3865 
3866 		primary |= CPU_TPR_SHADOW;
3867 		vmcs_write(CPU_EXEC_CTRL0, primary);
3868 		strcpy(str, "enabled");
3869 	}
3870 
3871 	/*
3872 	 * Second test
3873 	 */
3874 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3875 
3876 	primary = saved_primary;
3877 	secondary = saved_secondary;
3878 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3879 		return;
3880 
3881 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3882 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3883 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3884 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3885 	test_vmx_valid_controls();
3886 	report_prefix_pop();
3887 
3888 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3889 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3890 	test_vmx_valid_controls();
3891 	report_prefix_pop();
3892 
3893 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3894 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3895 	test_vmx_invalid_controls();
3896 	report_prefix_pop();
3897 
3898 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3899 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3900 	test_vmx_valid_controls();
3901 	report_prefix_pop();
3902 
3903 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3904 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3905 }
3906 
3907 /*
3908  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3909  * "external-interrupt exiting" VM-execution control must be 1.
3910  * [Intel SDM]
3911  */
3912 static void test_virtual_intr_ctls(void)
3913 {
3914 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3915 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3916 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3917 	u32 primary = saved_primary;
3918 	u32 secondary = saved_secondary;
3919 	u32 pin = saved_pin;
3920 
3921 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3922 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
3923 		return;
3924 
3925 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3926 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
3927 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3928 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
3929 	test_vmx_valid_controls();
3930 	report_prefix_pop();
3931 
3932 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
3933 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3934 	test_vmx_invalid_controls();
3935 	report_prefix_pop();
3936 
3937 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
3938 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
3939 	test_vmx_valid_controls();
3940 	report_prefix_pop();
3941 
3942 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3943 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3944 	test_vmx_invalid_controls();
3945 	report_prefix_pop();
3946 
3947 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3948 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3949 	vmcs_write(PIN_CONTROLS, saved_pin);
3950 }
3951 
3952 static void test_pi_desc_addr(u64 addr, bool ctrl)
3953 {
3954 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
3955 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
3956 	if (ctrl)
3957 		test_vmx_valid_controls();
3958 	else
3959 		test_vmx_invalid_controls();
3960 	report_prefix_pop();
3961 }
3962 
3963 /*
3964  * If the "process posted interrupts" VM-execution control is 1, the
3965  * following must be true:
3966  *
3967  *	- The "virtual-interrupt delivery" VM-execution control is 1.
3968  *	- The "acknowledge interrupt on exit" VM-exit control is 1.
3969  *	- The posted-interrupt notification vector has a value in the
3970  *	- range 0 - 255 (bits 15:8 are all 0).
3971  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
3972  *	- The posted-interrupt descriptor address does not set any bits
3973  *	  beyond the processor's physical-address width.
3974  * [Intel SDM]
3975  */
3976 static void test_posted_intr(void)
3977 {
3978 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3979 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3980 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3981 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
3982 	u32 primary = saved_primary;
3983 	u32 secondary = saved_secondary;
3984 	u32 pin = saved_pin;
3985 	u32 exit_ctl = exit_ctl_saved;
3986 	u16 vec;
3987 	int i;
3988 
3989 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
3990 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3991 	    (ctrl_exit_rev.clr & EXI_INTA)))
3992 		return;
3993 
3994 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3995 
3996 	/*
3997 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
3998 	 */
3999 	pin |= PIN_POST_INTR;
4000 	vmcs_write(PIN_CONTROLS, pin);
4001 	secondary &= ~CPU_VINTD;
4002 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4003 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
4004 	test_vmx_invalid_controls();
4005 	report_prefix_pop();
4006 
4007 	secondary |= CPU_VINTD;
4008 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4009 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
4010 	test_vmx_invalid_controls();
4011 	report_prefix_pop();
4012 
4013 	exit_ctl &= ~EXI_INTA;
4014 	vmcs_write(EXI_CONTROLS, exit_ctl);
4015 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
4016 	test_vmx_invalid_controls();
4017 	report_prefix_pop();
4018 
4019 	exit_ctl |= EXI_INTA;
4020 	vmcs_write(EXI_CONTROLS, exit_ctl);
4021 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4022 	test_vmx_valid_controls();
4023 	report_prefix_pop();
4024 
4025 	secondary &= ~CPU_VINTD;
4026 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4027 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
4028 	test_vmx_invalid_controls();
4029 	report_prefix_pop();
4030 
4031 	secondary |= CPU_VINTD;
4032 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4033 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4034 	test_vmx_valid_controls();
4035 	report_prefix_pop();
4036 
4037 	/*
4038 	 * Test posted-interrupt notification vector
4039 	 */
4040 	for (i = 0; i < 8; i++) {
4041 		vec = (1ul << i);
4042 		vmcs_write(PINV, vec);
4043 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4044 		test_vmx_valid_controls();
4045 		report_prefix_pop();
4046 	}
4047 	for (i = 8; i < 16; i++) {
4048 		vec = (1ul << i);
4049 		vmcs_write(PINV, vec);
4050 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4051 		test_vmx_invalid_controls();
4052 		report_prefix_pop();
4053 	}
4054 
4055 	vec &= ~(0xff << 8);
4056 	vmcs_write(PINV, vec);
4057 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4058 	test_vmx_valid_controls();
4059 	report_prefix_pop();
4060 
4061 	/*
4062 	 * Test posted-interrupt descriptor addresss
4063 	 */
4064 	for (i = 0; i < 6; i++) {
4065 		test_pi_desc_addr(1ul << i, false);
4066 	}
4067 
4068 	test_pi_desc_addr(0xf0, false);
4069 	test_pi_desc_addr(0xff, false);
4070 	test_pi_desc_addr(0x0f, false);
4071 	test_pi_desc_addr(0x8000, true);
4072 	test_pi_desc_addr(0x00, true);
4073 	test_pi_desc_addr(0xc000, true);
4074 
4075 	test_vmcs_addr_values("process-posted interrupts",
4076 			       POSTED_INTR_DESC_ADDR, 64,
4077 			       false, false, 0, 63);
4078 
4079 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4080 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4081 	vmcs_write(PIN_CONTROLS, saved_pin);
4082 }
4083 
4084 static void test_apic_ctls(void)
4085 {
4086 	test_apic_virt_addr();
4087 	test_apic_access_addr();
4088 	test_apic_virtual_ctls();
4089 	test_virtual_intr_ctls();
4090 	test_posted_intr();
4091 }
4092 
4093 /*
4094  * If the "enable VPID" VM-execution control is 1, the value of the
4095  * of the VPID VM-execution control field must not be 0000H.
4096  * [Intel SDM]
4097  */
4098 static void test_vpid(void)
4099 {
4100 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4101 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4102 	u16 vpid = 0x0000;
4103 	int i;
4104 
4105 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4106 	    (ctrl_cpu_rev[1].clr & CPU_VPID))) {
4107 		printf("Secondary controls and/or VPID not supported\n");
4108 		return;
4109 	}
4110 
4111 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
4112 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
4113 	vmcs_write(VPID, vpid);
4114 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
4115 	test_vmx_valid_controls();
4116 	report_prefix_pop();
4117 
4118 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
4119 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4120 	test_vmx_invalid_controls();
4121 	report_prefix_pop();
4122 
4123 	for (i = 0; i < 16; i++) {
4124 		vpid = (short)1 << i;;
4125 		vmcs_write(VPID, vpid);
4126 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4127 		test_vmx_valid_controls();
4128 		report_prefix_pop();
4129 	}
4130 
4131 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4132 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4133 }
4134 
4135 static void set_vtpr(unsigned vtpr)
4136 {
4137 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
4138 }
4139 
4140 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
4141 {
4142 	bool valid = true;
4143 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4144 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4145 
4146 	if ((primary & CPU_TPR_SHADOW) &&
4147 	    (!(primary & CPU_SECONDARY) ||
4148 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
4149 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
4150 
4151 	set_vtpr(vtpr);
4152 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
4153 	    threshold, (vtpr >> 4) & 0xf);
4154 	if (valid)
4155 		test_vmx_valid_controls();
4156 	else
4157 		test_vmx_invalid_controls();
4158 	report_prefix_pop();
4159 }
4160 
4161 static void test_invalid_event_injection(void)
4162 {
4163 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
4164 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
4165 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
4166 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
4167 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
4168 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4169 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4170 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4171 	u32 cnt;
4172 
4173 	/* Setup */
4174 	report_prefix_push("invalid event injection");
4175 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4176 	vmcs_write(ENT_INST_LEN, 0x00000001);
4177 
4178 	/* The field's interruption type is not set to a reserved value. */
4179 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4180 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4181 			    "RESERVED interruption type invalid [-]",
4182 			    ent_intr_info);
4183 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4184 	test_vmx_invalid_controls();
4185 	report_prefix_pop();
4186 
4187 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4188 			DE_VECTOR;
4189 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4190 			    "RESERVED interruption type invalid [+]",
4191 			    ent_intr_info);
4192 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4193 	test_vmx_valid_controls();
4194 	report_prefix_pop();
4195 
4196 	/* If the interruption type is other event, the vector is 0. */
4197 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4198 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4199 			    "(OTHER EVENT && vector != 0) invalid [-]",
4200 			    ent_intr_info);
4201 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4202 	test_vmx_invalid_controls();
4203 	report_prefix_pop();
4204 
4205 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4206 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4207 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4208 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4209 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4210 	test_vmx_invalid_controls();
4211 	report_prefix_pop();
4212 
4213 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4214 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4215 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4216 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4217 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4218 	test_vmx_valid_controls();
4219 	report_prefix_pop();
4220 
4221 	/*
4222 	 * If the interruption type
4223 	 * is HW exception, the vector is at most 31.
4224 	 */
4225 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4226 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4227 			    "(HW exception && vector > 31) invalid [-]",
4228 			    ent_intr_info);
4229 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4230 	test_vmx_invalid_controls();
4231 	report_prefix_pop();
4232 
4233 	/*
4234 	 * deliver-error-code is 1 iff either
4235 	 * (a) the "unrestricted guest" VM-execution control is 0
4236 	 * (b) CR0.PE is set.
4237 	 */
4238 
4239 	/* Assert that unrestricted guest is disabled or unsupported */
4240 	assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
4241 	       !(secondary_save & CPU_URG));
4242 
4243 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4244 			GP_VECTOR;
4245 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4246 			    "error code <-> (!URG || prot_mode) [-]",
4247 			    ent_intr_info);
4248 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4249 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4250 	test_vmx_invalid_controls();
4251 	report_prefix_pop();
4252 
4253 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4254 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4255 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4256 			    "error code <-> (!URG || prot_mode) [+]",
4257 			    ent_intr_info);
4258 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4259 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4260 	test_vmx_valid_controls();
4261 	report_prefix_pop();
4262 
4263 	if (enable_unrestricted_guest(false))
4264 		goto skip_unrestricted_guest;
4265 
4266 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4267 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4268 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4269 			    "error code <-> (!URG || prot_mode) [-]",
4270 			    ent_intr_info);
4271 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4272 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4273 	test_vmx_invalid_controls();
4274 	report_prefix_pop();
4275 
4276 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4277 			GP_VECTOR;
4278 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4279 			    "error code <-> (!URG || prot_mode) [-]",
4280 			    ent_intr_info);
4281 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4282 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4283 	test_vmx_invalid_controls();
4284 	report_prefix_pop();
4285 
4286 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4287 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4288 
4289 skip_unrestricted_guest:
4290 	vmcs_write(GUEST_CR0, guest_cr0_save);
4291 
4292 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4293 	report_prefix_push("error code <-> HW exception");
4294 	for (cnt = 0; cnt < 8; cnt++) {
4295 		u32 exception_type_mask = cnt << 8;
4296 		u32 deliver_error_code_mask =
4297 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4298 			INTR_INFO_DELIVER_CODE_MASK : 0;
4299 
4300 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4301 				exception_type_mask | GP_VECTOR;
4302 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4303 				    ent_intr_info);
4304 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4305 		test_vmx_invalid_controls();
4306 		report_prefix_pop();
4307 	}
4308 	report_prefix_pop();
4309 
4310 	/*
4311 	 * deliver-error-code is 1 iff the the vector
4312 	 * indicates an exception that would normally deliver an error code
4313 	 */
4314 	report_prefix_push("error code <-> vector delivers error code");
4315 	for (cnt = 0; cnt < 32; cnt++) {
4316 		bool has_error_code = false;
4317 		u32 deliver_error_code_mask;
4318 
4319 		switch (cnt) {
4320 		case DF_VECTOR:
4321 		case TS_VECTOR:
4322 		case NP_VECTOR:
4323 		case SS_VECTOR:
4324 		case GP_VECTOR:
4325 		case PF_VECTOR:
4326 		case AC_VECTOR:
4327 			has_error_code = true;
4328 		case CP_VECTOR:
4329 			/* Some CPUs have error code and some do not, skip */
4330 			continue;
4331 		}
4332 
4333 		/* Negative case */
4334 		deliver_error_code_mask = has_error_code ?
4335 						0 :
4336 						INTR_INFO_DELIVER_CODE_MASK;
4337 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4338 				INTR_TYPE_HARD_EXCEPTION | cnt;
4339 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4340 				    ent_intr_info);
4341 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4342 		test_vmx_invalid_controls();
4343 		report_prefix_pop();
4344 
4345 		/* Positive case */
4346 		deliver_error_code_mask = has_error_code ?
4347 						INTR_INFO_DELIVER_CODE_MASK :
4348 						0;
4349 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4350 				INTR_TYPE_HARD_EXCEPTION | cnt;
4351 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4352 				    ent_intr_info);
4353 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4354 		test_vmx_valid_controls();
4355 		report_prefix_pop();
4356 	}
4357 	report_prefix_pop();
4358 
4359 	/* Reserved bits in the field (30:12) are 0. */
4360 	report_prefix_push("reserved bits clear");
4361 	for (cnt = 12; cnt <= 30; cnt++) {
4362 		ent_intr_info = ent_intr_info_base |
4363 				INTR_INFO_DELIVER_CODE_MASK |
4364 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4365 				(1U << cnt);
4366 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4367 				    ent_intr_info);
4368 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4369 		test_vmx_invalid_controls();
4370 		report_prefix_pop();
4371 	}
4372 	report_prefix_pop();
4373 
4374 	/*
4375 	 * If deliver-error-code is 1
4376 	 * bits 31:16 of the VM-entry exception error-code field are 0.
4377 	 */
4378 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4379 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4380 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4381 			    "VM-entry exception error code[31:16] clear",
4382 			    ent_intr_info);
4383 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4384 	for (cnt = 16; cnt <= 31; cnt++) {
4385 		ent_intr_err = 1U << cnt;
4386 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4387 				    ent_intr_err);
4388 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4389 		test_vmx_invalid_controls();
4390 		report_prefix_pop();
4391 	}
4392 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4393 	report_prefix_pop();
4394 
4395 	/*
4396 	 * If the interruption type is software interrupt, software exception,
4397 	 * or privileged software exception, the VM-entry instruction-length
4398 	 * field is in the range 0 - 15.
4399 	 */
4400 
4401 	for (cnt = 0; cnt < 3; cnt++) {
4402 		switch (cnt) {
4403 		case 0:
4404 			ent_intr_info = ent_intr_info_base |
4405 					INTR_TYPE_SOFT_INTR;
4406 			break;
4407 		case 1:
4408 			ent_intr_info = ent_intr_info_base |
4409 					INTR_TYPE_SOFT_EXCEPTION;
4410 			break;
4411 		case 2:
4412 			ent_intr_info = ent_intr_info_base |
4413 					INTR_TYPE_PRIV_SW_EXCEPTION;
4414 			break;
4415 		}
4416 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4417 				    "VM-entry instruction-length check",
4418 				    ent_intr_info);
4419 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4420 
4421 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4422 		ent_intr_len = -1;
4423 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4424 				    ent_intr_len);
4425 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4426 		test_vmx_invalid_controls();
4427 		report_prefix_pop();
4428 
4429 		/* Instruction length set to 16 should fail */
4430 		ent_intr_len = 0x00000010;
4431 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4432 				    ent_intr_len);
4433 		vmcs_write(ENT_INST_LEN, 0x00000010);
4434 		test_vmx_invalid_controls();
4435 		report_prefix_pop();
4436 
4437 		report_prefix_pop();
4438 	}
4439 
4440 	/* Cleanup */
4441 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4442 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4443 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4444 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4445 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4446 	vmcs_write(GUEST_CR0, guest_cr0_save);
4447 	report_prefix_pop();
4448 }
4449 
4450 /*
4451  * Test interesting vTPR values for a given TPR threshold.
4452  */
4453 static void test_vtpr_values(unsigned threshold)
4454 {
4455 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4456 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4457 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4458 }
4459 
4460 static void try_tpr_threshold(unsigned threshold)
4461 {
4462 	bool valid = true;
4463 
4464 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4465 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4466 
4467 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4468 	    (secondary & CPU_VINTD)))
4469 		valid = !(threshold >> 4);
4470 
4471 	set_vtpr(-1);
4472 	vmcs_write(TPR_THRESHOLD, threshold);
4473 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4474 	if (valid)
4475 		test_vmx_valid_controls();
4476 	else
4477 		test_vmx_invalid_controls();
4478 	report_prefix_pop();
4479 
4480 	if (valid)
4481 		test_vtpr_values(threshold);
4482 }
4483 
4484 /*
4485  * Test interesting TPR threshold values.
4486  */
4487 static void test_tpr_threshold_values(void)
4488 {
4489 	unsigned i;
4490 
4491 	for (i = 0; i < 0x10; i++)
4492 		try_tpr_threshold(i);
4493 	for (i = 4; i < 32; i++)
4494 		try_tpr_threshold(1u << i);
4495 	try_tpr_threshold(-1u);
4496 	try_tpr_threshold(0x7fffffff);
4497 }
4498 
4499 /*
4500  * This test covers the following two VM entry checks:
4501  *
4502  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4503  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4504  *         31:4 of the TPR threshold VM-execution control field must
4505 	   be 0.
4506  *         [Intel SDM]
4507  *
4508  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4509  *          "virtual-interrupt delivery" VM-execution control is 0
4510  *          and the "virtualize APIC accesses" VM-execution control
4511  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4512  *          control field must not be greater than the value of bits
4513  *          7:4 of VTPR.
4514  *          [Intel SDM]
4515  */
4516 static void test_tpr_threshold(void)
4517 {
4518 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4519 	u64 apic_virt_addr = vmcs_read(APIC_VIRT_ADDR);
4520 	u64 threshold = vmcs_read(TPR_THRESHOLD);
4521 	void *virtual_apic_page;
4522 
4523 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4524 		return;
4525 
4526 	virtual_apic_page = alloc_page();
4527 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4528 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4529 
4530 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4531 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4532 	test_tpr_threshold_values();
4533 	report_prefix_pop();
4534 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4535 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4536 	test_tpr_threshold_values();
4537 	report_prefix_pop();
4538 
4539 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4540 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES))))
4541 		goto out;
4542 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4543 
4544 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4545 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4546 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4547 		test_tpr_threshold_values();
4548 		report_prefix_pop();
4549 
4550 		vmcs_write(CPU_EXEC_CTRL0,
4551 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4552 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4553 		test_tpr_threshold_values();
4554 		report_prefix_pop();
4555 	}
4556 
4557 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4558 		vmcs_write(CPU_EXEC_CTRL0,
4559 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4560 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4561 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4562 		test_tpr_threshold_values();
4563 		report_prefix_pop();
4564 
4565 		vmcs_write(CPU_EXEC_CTRL0,
4566 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4567 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4568 		test_tpr_threshold_values();
4569 		report_prefix_pop();
4570 	}
4571 
4572 	if ((ctrl_cpu_rev[1].clr &
4573 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4574 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4575 		vmcs_write(CPU_EXEC_CTRL0,
4576 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4577 		vmcs_write(CPU_EXEC_CTRL1,
4578 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4579 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4580 		test_tpr_threshold_values();
4581 		report_prefix_pop();
4582 
4583 		vmcs_write(CPU_EXEC_CTRL0,
4584 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4585 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4586 		test_tpr_threshold_values();
4587 		report_prefix_pop();
4588 	}
4589 
4590 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4591 out:
4592 	vmcs_write(TPR_THRESHOLD, threshold);
4593 	vmcs_write(APIC_VIRT_ADDR, apic_virt_addr);
4594 	vmcs_write(CPU_EXEC_CTRL0, primary);
4595 }
4596 
4597 /*
4598  * This test verifies the following two vmentry checks:
4599  *
4600  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4601  *  VM-execution control must be 0.
4602  *  [Intel SDM]
4603  *
4604  *  If the "virtual NMIs" VM-execution control is 0, the "NMI-window
4605  *  exiting" VM-execution control must be 0.
4606  *  [Intel SDM]
4607  */
4608 static void test_nmi_ctrls(void)
4609 {
4610 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4611 
4612 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4613 	    (PIN_NMI | PIN_VIRT_NMI)) {
4614 		printf("NMI exiting and Virtual NMIs are not supported !\n");
4615 		return;
4616 	}
4617 
4618 	/* Save the controls so that we can restore them after our tests */
4619 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4620 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4621 
4622 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4623 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4624 
4625 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4626 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4627 	test_vmx_valid_controls();
4628 	report_prefix_pop();
4629 
4630 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4631 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4632 	test_vmx_invalid_controls();
4633 	report_prefix_pop();
4634 
4635 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4636 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4637 	test_vmx_valid_controls();
4638 	report_prefix_pop();
4639 
4640 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4641 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4642 	test_vmx_valid_controls();
4643 	report_prefix_pop();
4644 
4645 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4646 		report_info("NMI-window exiting is not supported, skipping...");
4647 		goto done;
4648 	}
4649 
4650 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4651 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4652 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4653 	test_vmx_invalid_controls();
4654 	report_prefix_pop();
4655 
4656 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4657 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4658 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4659 	test_vmx_valid_controls();
4660 	report_prefix_pop();
4661 
4662 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4663 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4664 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4665 	test_vmx_valid_controls();
4666 	report_prefix_pop();
4667 
4668 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4669 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4670 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4671 	test_vmx_valid_controls();
4672 	report_prefix_pop();
4673 
4674 	/* Restore the controls to their original values */
4675 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4676 done:
4677 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4678 }
4679 
4680 static void test_eptp_ad_bit(u64 eptp, bool ctrl)
4681 {
4682 	vmcs_write(EPTP, eptp);
4683 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4684 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4685 	if (ctrl)
4686 		test_vmx_valid_controls();
4687 	else
4688 		test_vmx_invalid_controls();
4689 	report_prefix_pop();
4690 
4691 }
4692 
4693 /*
4694  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4695  *    control field must satisfy the following checks:
4696  *
4697  *     - The EPT memory type (bits 2:0) must be a value supported by the
4698  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4699  *     - Bits 5:3 (1 less than the EPT page-walk length) must indicate a
4700  *	 supported EPT page-walk length.
4701  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4702  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4703  *	 indicating that the processor does not support accessed and dirty
4704  *	 dirty flags for EPT.
4705  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4706  *	 physical-address width) must all be 0.
4707  *
4708  * 2. If the "unrestricted guest" VM-execution control is 1, the
4709  *    "enable EPT" VM-execution control must also be 1.
4710  */
4711 static void test_ept_eptp(void)
4712 {
4713 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4714 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4715 	u64 eptp_saved = vmcs_read(EPTP);
4716 	u32 primary = primary_saved;
4717 	u32 secondary = secondary_saved;
4718 	u64 msr, eptp = eptp_saved;
4719 	bool un_cache = false;
4720 	bool wr_bk = false;
4721 	bool ctrl;
4722 	u32 i, maxphysaddr;
4723 	u64 j, resv_bits_mask = 0;
4724 
4725 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4726 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4727 		printf("\"CPU secondary\" and/or \"enable EPT\" execution controls are not supported !\n");
4728 		return;
4729 	}
4730 
4731 	/*
4732 	 * Memory type (bits 2:0)
4733 	 */
4734 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
4735 	if (msr & EPT_CAP_UC)
4736 		un_cache = true;
4737 	if (msr & EPT_CAP_WB)
4738 		wr_bk = true;
4739 
4740 	/* Support for 4-level EPT is mandatory. */
4741 	report(msr & EPT_CAP_PWL4, "4-level EPT support check");
4742 
4743 	primary |= CPU_SECONDARY;
4744 	vmcs_write(CPU_EXEC_CTRL0, primary);
4745 	secondary |= CPU_EPT;
4746 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4747 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4748 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4749 	vmcs_write(EPTP, eptp);
4750 
4751 	for (i = 0; i < 8; i++) {
4752 		if (i == 0) {
4753 			if (un_cache) {
4754 				report_info("EPT paging structure memory-type is Un-cacheable\n");
4755 				ctrl = true;
4756 			} else {
4757 				ctrl = false;
4758 			}
4759 		} else if (i == 6) {
4760 			if (wr_bk) {
4761 				report_info("EPT paging structure memory-type is Write-back\n");
4762 				ctrl = true;
4763 			} else {
4764 				ctrl = false;
4765 			}
4766 		} else {
4767 			ctrl = false;
4768 		}
4769 
4770 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4771 		vmcs_write(EPTP, eptp);
4772 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4773 		    eptp & EPT_MEM_TYPE_MASK);
4774 		if (ctrl)
4775 			test_vmx_valid_controls();
4776 		else
4777 			test_vmx_invalid_controls();
4778 		report_prefix_pop();
4779 	}
4780 
4781 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4782 
4783 	/*
4784 	 * Page walk length (bits 5:3).  Note, the value in VMCS.EPTP "is 1
4785 	 * less than the EPT page-walk length".
4786 	 */
4787 	for (i = 0; i < 8; i++) {
4788 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4789 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4790 		if (i == 3 || (i == 4 && (msr & EPT_CAP_PWL5)))
4791 			ctrl = true;
4792 		else
4793 			ctrl = false;
4794 
4795 		vmcs_write(EPTP, eptp);
4796 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4797 		    eptp & EPTP_PG_WALK_LEN_MASK);
4798 		if (ctrl)
4799 			test_vmx_valid_controls();
4800 		else
4801 			test_vmx_invalid_controls();
4802 		report_prefix_pop();
4803 	}
4804 
4805 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4806 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4807 
4808 	/*
4809 	 * Accessed and dirty flag (bit 6)
4810 	 */
4811 	if (msr & EPT_CAP_AD_FLAG) {
4812 		report_info("Processor supports accessed and dirty flag");
4813 		eptp &= ~EPTP_AD_FLAG;
4814 		test_eptp_ad_bit(eptp, true);
4815 
4816 		eptp |= EPTP_AD_FLAG;
4817 		test_eptp_ad_bit(eptp, true);
4818 	} else {
4819 		report_info("Processor does not supports accessed and dirty flag");
4820 		eptp &= ~EPTP_AD_FLAG;
4821 		test_eptp_ad_bit(eptp, true);
4822 
4823 		eptp |= EPTP_AD_FLAG;
4824 		test_eptp_ad_bit(eptp, false);
4825 	}
4826 
4827 	/*
4828 	 * Reserved bits [11:7] and [63:N]
4829 	 */
4830 	for (i = 0; i < 32; i++) {
4831 		eptp = (eptp &
4832 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4833 		    (i << EPTP_RESERV_BITS_SHIFT);
4834 		vmcs_write(EPTP, eptp);
4835 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4836 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4837 		    EPTP_RESERV_BITS_MASK);
4838 		if (i == 0)
4839 			test_vmx_valid_controls();
4840 		else
4841 			test_vmx_invalid_controls();
4842 		report_prefix_pop();
4843 	}
4844 
4845 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4846 
4847 	maxphysaddr = cpuid_maxphyaddr();
4848 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4849 		resv_bits_mask |= 1ul << i;
4850 	}
4851 
4852 	for (j = maxphysaddr - 1; j <= 63; j++) {
4853 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4854 		    (j < maxphysaddr ? 0 : 1ul << j);
4855 		vmcs_write(EPTP, eptp);
4856 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4857 		    (eptp >> maxphysaddr) & resv_bits_mask);
4858 		if (j < maxphysaddr)
4859 			test_vmx_valid_controls();
4860 		else
4861 			test_vmx_invalid_controls();
4862 		report_prefix_pop();
4863 	}
4864 
4865 	secondary &= ~(CPU_EPT | CPU_URG);
4866 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4867 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4868 	test_vmx_valid_controls();
4869 	report_prefix_pop();
4870 
4871 	if (!(ctrl_cpu_rev[1].clr & CPU_URG))
4872 		goto skip_unrestricted_guest;
4873 
4874 	secondary |= CPU_URG;
4875 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4876 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4877 	test_vmx_invalid_controls();
4878 	report_prefix_pop();
4879 
4880 	secondary |= CPU_EPT;
4881 	setup_dummy_ept();
4882 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4883 	test_vmx_valid_controls();
4884 	report_prefix_pop();
4885 
4886 skip_unrestricted_guest:
4887 	secondary &= ~CPU_URG;
4888 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4889 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4890 	test_vmx_valid_controls();
4891 	report_prefix_pop();
4892 
4893 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4894 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4895 	vmcs_write(EPTP, eptp_saved);
4896 }
4897 
4898 /*
4899  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4900  * VM-execution control must also be 1. In addition, the PML address
4901  * must satisfy the following checks:
4902  *
4903  *    * Bits 11:0 of the address must be 0.
4904  *    * The address should not set any bits beyond the processor's
4905  *	physical-address width.
4906  *
4907  *  [Intel SDM]
4908  */
4909 static void test_pml(void)
4910 {
4911 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4912 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4913 	u32 primary = primary_saved;
4914 	u32 secondary = secondary_saved;
4915 
4916 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4917 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
4918 		printf("\"Secondary execution\" control or \"enable EPT\" control or \"enable PML\" control is not supported !\n");
4919 		return;
4920 	}
4921 
4922 	primary |= CPU_SECONDARY;
4923 	vmcs_write(CPU_EXEC_CTRL0, primary);
4924 	secondary &= ~(CPU_PML | CPU_EPT);
4925 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4926 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
4927 	test_vmx_valid_controls();
4928 	report_prefix_pop();
4929 
4930 	secondary |= CPU_PML;
4931 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4932 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
4933 	test_vmx_invalid_controls();
4934 	report_prefix_pop();
4935 
4936 	secondary |= CPU_EPT;
4937 	setup_dummy_ept();
4938 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
4939 	test_vmx_valid_controls();
4940 	report_prefix_pop();
4941 
4942 	secondary &= ~CPU_PML;
4943 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4944 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
4945 	test_vmx_valid_controls();
4946 	report_prefix_pop();
4947 
4948 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
4949 				 PAGE_SIZE, false, false);
4950 
4951 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4952 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4953 }
4954 
4955  /*
4956  * If the "activate VMX-preemption timer" VM-execution control is 0, the
4957  * the "save VMX-preemption timer value" VM-exit control must also be 0.
4958  *
4959  *  [Intel SDM]
4960  */
4961 static void test_vmx_preemption_timer(void)
4962 {
4963 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4964 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
4965 	u32 pin = saved_pin;
4966 	u32 exit = saved_exit;
4967 
4968 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
4969 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
4970 		printf("\"Save-VMX-preemption-timer\" control and/or \"Enable-VMX-preemption-timer\" control is not supported\n");
4971 		return;
4972 	}
4973 
4974 	pin |= PIN_PREEMPT;
4975 	vmcs_write(PIN_CONTROLS, pin);
4976 	exit &= ~EXI_SAVE_PREEMPT;
4977 	vmcs_write(EXI_CONTROLS, exit);
4978 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
4979 	test_vmx_valid_controls();
4980 	report_prefix_pop();
4981 
4982 	exit |= EXI_SAVE_PREEMPT;
4983 	vmcs_write(EXI_CONTROLS, exit);
4984 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
4985 	test_vmx_valid_controls();
4986 	report_prefix_pop();
4987 
4988 	pin &= ~PIN_PREEMPT;
4989 	vmcs_write(PIN_CONTROLS, pin);
4990 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
4991 	test_vmx_invalid_controls();
4992 	report_prefix_pop();
4993 
4994 	exit &= ~EXI_SAVE_PREEMPT;
4995 	vmcs_write(EXI_CONTROLS, exit);
4996 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
4997 	test_vmx_valid_controls();
4998 	report_prefix_pop();
4999 
5000 	vmcs_write(PIN_CONTROLS, saved_pin);
5001 	vmcs_write(EXI_CONTROLS, saved_exit);
5002 }
5003 
5004 extern unsigned char test_mtf1;
5005 extern unsigned char test_mtf2;
5006 extern unsigned char test_mtf3;
5007 extern unsigned char test_mtf4;
5008 
5009 static void test_mtf_guest(void)
5010 {
5011 	asm ("vmcall;\n\t"
5012 	     "out %al, $0x80;\n\t"
5013 	     "test_mtf1:\n\t"
5014 	     "vmcall;\n\t"
5015 	     "out %al, $0x80;\n\t"
5016 	     "test_mtf2:\n\t"
5017 	     /*
5018 	      * Prepare for the 'MOV CR3' test. Attempt to induce a
5019 	      * general-protection fault by moving a non-canonical address into
5020 	      * CR3. The 'MOV CR3' instruction does not take an imm64 operand,
5021 	      * so we must MOV the desired value into a register first.
5022 	      *
5023 	      * MOV RAX is done before the VMCALL such that MTF is only enabled
5024 	      * for the instruction under test.
5025 	      */
5026 	     "mov $0xaaaaaaaaaaaaaaaa, %rax;\n\t"
5027 	     "vmcall;\n\t"
5028 	     "mov %rax, %cr3;\n\t"
5029 	     "test_mtf3:\n\t"
5030 	     "vmcall;\n\t"
5031 	     /*
5032 	      * ICEBP/INT1 instruction. Though the instruction is now
5033 	      * documented, don't rely on assemblers enumerating the
5034 	      * instruction. Resort to hand assembly.
5035 	      */
5036 	     ".byte 0xf1;\n\t"
5037 	     "vmcall;\n\t"
5038 	     "test_mtf4:\n\t"
5039 	     "mov $0, %eax;\n\t");
5040 }
5041 
5042 static void test_mtf_gp_handler(struct ex_regs *regs)
5043 {
5044 	regs->rip = (unsigned long) &test_mtf3;
5045 }
5046 
5047 static void test_mtf_db_handler(struct ex_regs *regs)
5048 {
5049 }
5050 
5051 static void enable_mtf(void)
5052 {
5053 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5054 
5055 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 | CPU_MTF);
5056 }
5057 
5058 static void disable_mtf(void)
5059 {
5060 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5061 
5062 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 & ~CPU_MTF);
5063 }
5064 
5065 static void enable_tf(void)
5066 {
5067 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5068 
5069 	vmcs_write(GUEST_RFLAGS, rflags | X86_EFLAGS_TF);
5070 }
5071 
5072 static void disable_tf(void)
5073 {
5074 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5075 
5076 	vmcs_write(GUEST_RFLAGS, rflags & ~X86_EFLAGS_TF);
5077 }
5078 
5079 static void report_mtf(const char *insn_name, unsigned long exp_rip)
5080 {
5081 	unsigned long rip = vmcs_read(GUEST_RIP);
5082 
5083 	assert_exit_reason(VMX_MTF);
5084 	report(rip == exp_rip, "MTF VM-exit after %s. RIP: 0x%lx (expected 0x%lx)",
5085 	       insn_name, rip, exp_rip);
5086 }
5087 
5088 static void vmx_mtf_test(void)
5089 {
5090 	unsigned long pending_dbg;
5091 	handler old_gp, old_db;
5092 
5093 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5094 		printf("CPU does not support the 'monitor trap flag' processor-based VM-execution control.\n");
5095 		return;
5096 	}
5097 
5098 	test_set_guest(test_mtf_guest);
5099 
5100 	/* Expect an MTF VM-exit after OUT instruction */
5101 	enter_guest();
5102 	skip_exit_vmcall();
5103 
5104 	enable_mtf();
5105 	enter_guest();
5106 	report_mtf("OUT", (unsigned long) &test_mtf1);
5107 	disable_mtf();
5108 
5109 	/*
5110 	 * Concurrent #DB trap and MTF on instruction boundary. Expect MTF
5111 	 * VM-exit with populated 'pending debug exceptions' VMCS field.
5112 	 */
5113 	enter_guest();
5114 	skip_exit_vmcall();
5115 
5116 	enable_mtf();
5117 	enable_tf();
5118 
5119 	enter_guest();
5120 	report_mtf("OUT", (unsigned long) &test_mtf2);
5121 	pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
5122 	report(pending_dbg & DR_STEP,
5123 	       "'pending debug exceptions' field after MTF VM-exit: 0x%lx (expected 0x%lx)",
5124 	       pending_dbg, (unsigned long) DR_STEP);
5125 
5126 	disable_mtf();
5127 	disable_tf();
5128 	vmcs_write(GUEST_PENDING_DEBUG, 0);
5129 
5130 	/*
5131 	 * #GP exception takes priority over MTF. Expect MTF VM-exit with RIP
5132 	 * advanced to first instruction of #GP handler.
5133 	 */
5134 	enter_guest();
5135 	skip_exit_vmcall();
5136 
5137 	old_gp = handle_exception(GP_VECTOR, test_mtf_gp_handler);
5138 
5139 	enable_mtf();
5140 	enter_guest();
5141 	report_mtf("MOV CR3", (unsigned long) get_idt_addr(&boot_idt[GP_VECTOR]));
5142 	disable_mtf();
5143 
5144 	/*
5145 	 * Concurrent MTF and privileged software exception (i.e. ICEBP/INT1).
5146 	 * MTF should follow the delivery of #DB trap, though the SDM doesn't
5147 	 * provide clear indication of the relative priority.
5148 	 */
5149 	enter_guest();
5150 	skip_exit_vmcall();
5151 
5152 	handle_exception(GP_VECTOR, old_gp);
5153 	old_db = handle_exception(DB_VECTOR, test_mtf_db_handler);
5154 
5155 	enable_mtf();
5156 	enter_guest();
5157 	report_mtf("INT1", (unsigned long) get_idt_addr(&boot_idt[DB_VECTOR]));
5158 	disable_mtf();
5159 
5160 	enter_guest();
5161 	skip_exit_vmcall();
5162 	handle_exception(DB_VECTOR, old_db);
5163 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK | INTR_TYPE_OTHER_EVENT);
5164 	enter_guest();
5165 	report_mtf("injected MTF", (unsigned long) &test_mtf4);
5166 	enter_guest();
5167 }
5168 
5169 extern char vmx_mtf_pdpte_guest_begin;
5170 extern char vmx_mtf_pdpte_guest_end;
5171 
5172 asm("vmx_mtf_pdpte_guest_begin:\n\t"
5173     "mov %cr0, %rax\n\t"    /* save CR0 with PG=1                 */
5174     "vmcall\n\t"            /* on return from this CR0.PG=0       */
5175     "mov %rax, %cr0\n\t"    /* restore CR0.PG=1 to enter PAE mode */
5176     "vmcall\n\t"
5177     "retq\n\t"
5178     "vmx_mtf_pdpte_guest_end:");
5179 
5180 static void vmx_mtf_pdpte_test(void)
5181 {
5182 	void *test_mtf_pdpte_guest;
5183 	pteval_t *pdpt;
5184 	u32 guest_ar_cs;
5185 	u64 guest_efer;
5186 	pteval_t *pte;
5187 	u64 guest_cr0;
5188 	u64 guest_cr3;
5189 	u64 guest_cr4;
5190 	u64 ent_ctls;
5191 	int i;
5192 
5193 	if (setup_ept(false))
5194 		return;
5195 
5196 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5197 		printf("CPU does not support 'monitor trap flag.'\n");
5198 		return;
5199 	}
5200 
5201 	if (!(ctrl_cpu_rev[1].clr & CPU_URG)) {
5202 		printf("CPU does not support 'unrestricted guest.'\n");
5203 		return;
5204 	}
5205 
5206 	vmcs_write(EXC_BITMAP, ~0);
5207 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
5208 
5209 	/*
5210 	 * Copy the guest code to an identity-mapped page.
5211 	 */
5212 	test_mtf_pdpte_guest = alloc_page();
5213 	memcpy(test_mtf_pdpte_guest, &vmx_mtf_pdpte_guest_begin,
5214 	       &vmx_mtf_pdpte_guest_end - &vmx_mtf_pdpte_guest_begin);
5215 
5216 	test_set_guest(test_mtf_pdpte_guest);
5217 
5218 	enter_guest();
5219 	skip_exit_vmcall();
5220 
5221 	/*
5222 	 * Put the guest in non-paged 32-bit protected mode, ready to enter
5223 	 * PAE mode when CR0.PG is set. CR4.PAE will already have been set
5224 	 * when the guest started out in long mode.
5225 	 */
5226 	ent_ctls = vmcs_read(ENT_CONTROLS);
5227 	vmcs_write(ENT_CONTROLS, ent_ctls & ~ENT_GUEST_64);
5228 
5229 	guest_efer = vmcs_read(GUEST_EFER);
5230 	vmcs_write(GUEST_EFER, guest_efer & ~(EFER_LMA | EFER_LME));
5231 
5232 	/*
5233 	 * Set CS access rights bits for 32-bit protected mode:
5234 	 * 3:0    B execute/read/accessed
5235 	 * 4      1 code or data
5236 	 * 6:5    0 descriptor privilege level
5237 	 * 7      1 present
5238 	 * 11:8   0 reserved
5239 	 * 12     0 available for use by system software
5240 	 * 13     0 64 bit mode not active
5241 	 * 14     1 default operation size 32-bit segment
5242 	 * 15     1 page granularity: segment limit in 4K units
5243 	 * 16     0 segment usable
5244 	 * 31:17  0 reserved
5245 	 */
5246 	guest_ar_cs = vmcs_read(GUEST_AR_CS);
5247 	vmcs_write(GUEST_AR_CS, 0xc09b);
5248 
5249 	guest_cr0 = vmcs_read(GUEST_CR0);
5250 	vmcs_write(GUEST_CR0, guest_cr0 & ~X86_CR0_PG);
5251 
5252 	guest_cr4 = vmcs_read(GUEST_CR4);
5253 	vmcs_write(GUEST_CR4, guest_cr4 & ~X86_CR4_PCIDE);
5254 
5255 	guest_cr3 = vmcs_read(GUEST_CR3);
5256 
5257 	/*
5258 	 * Turn the 4-level page table into a PAE page table by following the 0th
5259 	 * PML4 entry to a PDPT page, and grab the first four PDPTEs from that
5260 	 * page.
5261 	 *
5262 	 * Why does this work?
5263 	 *
5264 	 * PAE uses 32-bit addressing which implies:
5265 	 * Bits 11:0   page offset
5266 	 * Bits 20:12  entry into 512-entry page table
5267 	 * Bits 29:21  entry into a 512-entry directory table
5268 	 * Bits 31:30  entry into the page directory pointer table.
5269 	 * Bits 63:32  zero
5270 	 *
5271 	 * As only 2 bits are needed to select the PDPTEs for the entire
5272 	 * 32-bit address space, take the first 4 PDPTEs in the level 3 page
5273 	 * directory pointer table. It doesn't matter which of these PDPTEs
5274 	 * are present because they must cover the guest code given that it
5275 	 * has already run successfully.
5276 	 *
5277 	 * Get a pointer to PTE for GVA=0 in the page directory pointer table
5278 	 */
5279 	pte = get_pte_level(
5280             (pgd_t *)phys_to_virt(guest_cr3 & ~X86_CR3_PCID_MASK), 0,
5281             PDPT_LEVEL);
5282 
5283 	/*
5284 	 * Need some memory for the 4-entry PAE page directory pointer
5285 	 * table. Use the end of the identity-mapped page where the guest code
5286 	 * is stored. There is definitely space as the guest code is only a
5287 	 * few bytes.
5288 	 */
5289 	pdpt = test_mtf_pdpte_guest + PAGE_SIZE - 4 * sizeof(pteval_t);
5290 
5291 	/*
5292 	 * Copy the first four PDPTEs into the PAE page table with reserved
5293 	 * bits cleared. Note that permission bits from the PML4E and PDPTE
5294 	 * are not propagated.
5295 	 */
5296 	for (i = 0; i < 4; i++) {
5297 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_RSVD_MASK),
5298 				   "PDPTE has invalid reserved bits");
5299 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_PAGE_SIZE_MASK),
5300 				   "Cannot use 1GB super pages for PAE");
5301 		pdpt[i] = pte[i] & ~(PAE_PDPTE_RSVD_MASK);
5302 	}
5303 	vmcs_write(GUEST_CR3, virt_to_phys(pdpt));
5304 
5305 	enable_mtf();
5306 	enter_guest();
5307 	assert_exit_reason(VMX_MTF);
5308 	disable_mtf();
5309 
5310 	/*
5311 	 * The four PDPTEs should have been loaded into the VMCS when
5312 	 * the guest set CR0.PG to enter PAE mode.
5313 	 */
5314 	for (i = 0; i < 4; i++) {
5315 		u64 pdpte = vmcs_read(GUEST_PDPTE + 2 * i);
5316 
5317 		report(pdpte == pdpt[i], "PDPTE%d is 0x%lx (expected 0x%lx)",
5318 		       i, pdpte, pdpt[i]);
5319 	}
5320 
5321 	/*
5322 	 * Now, try to enter the guest in PAE mode. If the PDPTEs in the
5323 	 * vmcs are wrong, this will fail.
5324 	 */
5325 	enter_guest();
5326 	skip_exit_vmcall();
5327 
5328 	/*
5329 	 * Return guest to 64-bit mode and wrap up.
5330 	 */
5331 	vmcs_write(ENT_CONTROLS, ent_ctls);
5332 	vmcs_write(GUEST_EFER, guest_efer);
5333 	vmcs_write(GUEST_AR_CS, guest_ar_cs);
5334 	vmcs_write(GUEST_CR0, guest_cr0);
5335 	vmcs_write(GUEST_CR4, guest_cr4);
5336 	vmcs_write(GUEST_CR3, guest_cr3);
5337 
5338 	enter_guest();
5339 }
5340 
5341 /*
5342  * Tests for VM-execution control fields
5343  */
5344 static void test_vm_execution_ctls(void)
5345 {
5346 	test_pin_based_ctls();
5347 	test_primary_processor_based_ctls();
5348 	test_secondary_processor_based_ctls();
5349 	test_cr3_targets();
5350 	test_io_bitmaps();
5351 	test_msr_bitmap();
5352 	test_apic_ctls();
5353 	test_tpr_threshold();
5354 	test_nmi_ctrls();
5355 	test_pml();
5356 	test_vpid();
5357 	test_ept_eptp();
5358 	test_vmx_preemption_timer();
5359 }
5360 
5361  /*
5362   * The following checks are performed for the VM-entry MSR-load address if
5363   * the VM-entry MSR-load count field is non-zero:
5364   *
5365   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
5366   *      The address should not set any bits beyond the processor's
5367   *      physical-address width.
5368   *
5369   *    - The address of the last byte in the VM-entry MSR-load area
5370   *      should not set any bits beyond the processor's physical-address
5371   *      width. The address of this last byte is VM-entry MSR-load address
5372   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5373   *      uses more bits than the processor's physical-address width.)
5374   *
5375   *
5376   *  [Intel SDM]
5377   */
5378 static void test_entry_msr_load(void)
5379 {
5380 	entry_msr_load = alloc_page();
5381 	u64 tmp;
5382 	u32 entry_msr_ld_cnt = 1;
5383 	int i;
5384 	u32 addr_len = 64;
5385 
5386 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
5387 
5388 	/* Check first 4 bits of VM-entry MSR-load address */
5389 	for (i = 0; i < 4; i++) {
5390 		tmp = (u64)entry_msr_load | 1ull << i;
5391 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
5392 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
5393 				    tmp & 0xf);
5394 		test_vmx_invalid_controls();
5395 		report_prefix_pop();
5396 	}
5397 
5398 	if (basic.val & (1ul << 48))
5399 		addr_len = 32;
5400 
5401 	test_vmcs_addr_values("VM-entry-MSR-load address",
5402 				ENTER_MSR_LD_ADDR, 16, false, false,
5403 				4, addr_len - 1);
5404 
5405 	/*
5406 	 * Check last byte of VM-entry MSR-load address
5407 	 */
5408 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
5409 
5410 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5411 							i < 64; i++) {
5412 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
5413 			1ul << i;
5414 		vmcs_write(ENTER_MSR_LD_ADDR,
5415 			   tmp - (entry_msr_ld_cnt * 16 - 1));
5416 		test_vmx_invalid_controls();
5417 	}
5418 
5419 	vmcs_write(ENT_MSR_LD_CNT, 2);
5420 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5421 	test_vmx_invalid_controls();
5422 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5423 	test_vmx_valid_controls();
5424 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5425 	test_vmx_valid_controls();
5426 }
5427 
5428 static struct vmx_state_area_test_data {
5429 	u32 msr;
5430 	u64 exp;
5431 	bool enabled;
5432 } vmx_state_area_test_data;
5433 
5434 static void guest_state_test_main(void)
5435 {
5436 	u64 obs;
5437 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
5438 
5439 	while (1) {
5440 		if (vmx_get_test_stage() == 2)
5441 			break;
5442 
5443 		if (data->enabled) {
5444 			obs = rdmsr(data->msr);
5445 			report(data->exp == obs,
5446 			       "Guest state is 0x%lx (expected 0x%lx)",
5447 			       obs, data->exp);
5448 		}
5449 
5450 		vmcall();
5451 	}
5452 
5453 	asm volatile("fnop");
5454 }
5455 
5456 static void test_guest_state(const char *test, bool xfail, u64 field,
5457 			     const char * field_name)
5458 {
5459 	struct vmentry_result result;
5460 	u8 abort_flags;
5461 
5462 	abort_flags = ABORT_ON_EARLY_VMENTRY_FAIL;
5463 	if (!xfail)
5464 		abort_flags = ABORT_ON_INVALID_GUEST_STATE;
5465 
5466 	__enter_guest(abort_flags, &result);
5467 
5468 	report(result.exit_reason.failed_vmentry == xfail &&
5469 	       ((xfail && result.exit_reason.basic == VMX_FAIL_STATE) ||
5470 	        (!xfail && result.exit_reason.basic == VMX_VMCALL)) &&
5471 		(!xfail || vmcs_read(EXI_QUALIFICATION) == ENTRY_FAIL_DEFAULT),
5472 	        "%s, %s = %lx", test, field_name, field);
5473 
5474 	if (!result.exit_reason.failed_vmentry)
5475 		skip_exit_insn();
5476 }
5477 
5478 /*
5479  * Tests for VM-entry control fields
5480  */
5481 static void test_vm_entry_ctls(void)
5482 {
5483 	test_invalid_event_injection();
5484 	test_entry_msr_load();
5485 }
5486 
5487 /*
5488  * The following checks are performed for the VM-exit MSR-store address if
5489  * the VM-exit MSR-store count field is non-zero:
5490  *
5491  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
5492  *      The address should not set any bits beyond the processor's
5493  *      physical-address width.
5494  *
5495  *    - The address of the last byte in the VM-exit MSR-store area
5496  *      should not set any bits beyond the processor's physical-address
5497  *      width. The address of this last byte is VM-exit MSR-store address
5498  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5499  *      uses more bits than the processor's physical-address width.)
5500  *
5501  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
5502  * in the range 63:32.
5503  *
5504  *  [Intel SDM]
5505  */
5506 static void test_exit_msr_store(void)
5507 {
5508 	exit_msr_store = alloc_page();
5509 	u64 tmp;
5510 	u32 exit_msr_st_cnt = 1;
5511 	int i;
5512 	u32 addr_len = 64;
5513 
5514 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
5515 
5516 	/* Check first 4 bits of VM-exit MSR-store address */
5517 	for (i = 0; i < 4; i++) {
5518 		tmp = (u64)exit_msr_store | 1ull << i;
5519 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
5520 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
5521 				    tmp & 0xf);
5522 		test_vmx_invalid_controls();
5523 		report_prefix_pop();
5524 	}
5525 
5526 	if (basic.val & (1ul << 48))
5527 		addr_len = 32;
5528 
5529 	test_vmcs_addr_values("VM-exit-MSR-store address",
5530 				EXIT_MSR_ST_ADDR, 16, false, false,
5531 				4, addr_len - 1);
5532 
5533 	/*
5534 	 * Check last byte of VM-exit MSR-store address
5535 	 */
5536 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
5537 
5538 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5539 							i < 64; i++) {
5540 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
5541 			1ul << i;
5542 		vmcs_write(EXIT_MSR_ST_ADDR,
5543 			   tmp - (exit_msr_st_cnt * 16 - 1));
5544 		test_vmx_invalid_controls();
5545 	}
5546 
5547 	vmcs_write(EXI_MSR_ST_CNT, 2);
5548 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5549 	test_vmx_invalid_controls();
5550 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5551 	test_vmx_valid_controls();
5552 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5553 	test_vmx_valid_controls();
5554 }
5555 
5556 /*
5557  * Tests for VM-exit controls
5558  */
5559 static void test_vm_exit_ctls(void)
5560 {
5561 	test_exit_msr_store();
5562 }
5563 
5564 /*
5565  * Check that the virtual CPU checks all of the VMX controls as
5566  * documented in the Intel SDM.
5567  */
5568 static void vmx_controls_test(void)
5569 {
5570 	/*
5571 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
5572 	 * fail due to invalid guest state, should we make it that
5573 	 * far.
5574 	 */
5575 	vmcs_write(GUEST_RFLAGS, 0);
5576 
5577 	test_vm_execution_ctls();
5578 	test_vm_exit_ctls();
5579 	test_vm_entry_ctls();
5580 }
5581 
5582 struct apic_reg_virt_config {
5583 	bool apic_register_virtualization;
5584 	bool use_tpr_shadow;
5585 	bool virtualize_apic_accesses;
5586 	bool virtualize_x2apic_mode;
5587 	bool activate_secondary_controls;
5588 };
5589 
5590 struct apic_reg_test {
5591 	const char *name;
5592 	struct apic_reg_virt_config apic_reg_virt_config;
5593 };
5594 
5595 struct apic_reg_virt_expectation {
5596 	enum Reason rd_exit_reason;
5597 	enum Reason wr_exit_reason;
5598 	u32 val;
5599 	u32 (*virt_fn)(u32);
5600 
5601 	/*
5602 	 * If false, accessing the APIC access address from L2 is treated as a
5603 	 * normal memory operation, rather than triggering virtualization.
5604 	 */
5605 	bool virtualize_apic_accesses;
5606 };
5607 
5608 static u32 apic_virt_identity(u32 val)
5609 {
5610 	return val;
5611 }
5612 
5613 static u32 apic_virt_nibble1(u32 val)
5614 {
5615 	return val & 0xf0;
5616 }
5617 
5618 static u32 apic_virt_byte3(u32 val)
5619 {
5620 	return val & (0xff << 24);
5621 }
5622 
5623 static bool apic_reg_virt_exit_expectation(
5624 	u32 reg, struct apic_reg_virt_config *config,
5625 	struct apic_reg_virt_expectation *expectation)
5626 {
5627 	/* Good configs, where some L2 APIC accesses are virtualized. */
5628 	bool virtualize_apic_accesses_only =
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 	bool virtualize_apic_accesses_and_use_tpr_shadow =
5635 		config->virtualize_apic_accesses &&
5636 		config->use_tpr_shadow &&
5637 		!config->apic_register_virtualization &&
5638 		!config->virtualize_x2apic_mode &&
5639 		config->activate_secondary_controls;
5640 	bool apic_register_virtualization =
5641 		config->virtualize_apic_accesses &&
5642 		config->use_tpr_shadow &&
5643 		config->apic_register_virtualization &&
5644 		!config->virtualize_x2apic_mode &&
5645 		config->activate_secondary_controls;
5646 
5647 	expectation->val = MAGIC_VAL_1;
5648 	expectation->virt_fn = apic_virt_identity;
5649 	expectation->virtualize_apic_accesses =
5650 		config->virtualize_apic_accesses &&
5651 		config->activate_secondary_controls;
5652 	if (virtualize_apic_accesses_only) {
5653 		expectation->rd_exit_reason = VMX_APIC_ACCESS;
5654 		expectation->wr_exit_reason = VMX_APIC_ACCESS;
5655 	} else if (virtualize_apic_accesses_and_use_tpr_shadow) {
5656 		switch (reg) {
5657 		case APIC_TASKPRI:
5658 			expectation->rd_exit_reason = VMX_VMCALL;
5659 			expectation->wr_exit_reason = VMX_VMCALL;
5660 			expectation->virt_fn = apic_virt_nibble1;
5661 			break;
5662 		default:
5663 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5664 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5665 		}
5666 	} else if (apic_register_virtualization) {
5667 		expectation->rd_exit_reason = VMX_VMCALL;
5668 
5669 		switch (reg) {
5670 		case APIC_ID:
5671 		case APIC_EOI:
5672 		case APIC_LDR:
5673 		case APIC_DFR:
5674 		case APIC_SPIV:
5675 		case APIC_ESR:
5676 		case APIC_ICR:
5677 		case APIC_LVTT:
5678 		case APIC_LVTTHMR:
5679 		case APIC_LVTPC:
5680 		case APIC_LVT0:
5681 		case APIC_LVT1:
5682 		case APIC_LVTERR:
5683 		case APIC_TMICT:
5684 		case APIC_TDCR:
5685 			expectation->wr_exit_reason = VMX_APIC_WRITE;
5686 			break;
5687 		case APIC_LVR:
5688 		case APIC_ISR ... APIC_ISR + 0x70:
5689 		case APIC_TMR ... APIC_TMR + 0x70:
5690 		case APIC_IRR ... APIC_IRR + 0x70:
5691 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5692 			break;
5693 		case APIC_TASKPRI:
5694 			expectation->wr_exit_reason = VMX_VMCALL;
5695 			expectation->virt_fn = apic_virt_nibble1;
5696 			break;
5697 		case APIC_ICR2:
5698 			expectation->wr_exit_reason = VMX_VMCALL;
5699 			expectation->virt_fn = apic_virt_byte3;
5700 			break;
5701 		default:
5702 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5703 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5704 		}
5705 	} else if (!expectation->virtualize_apic_accesses) {
5706 		/*
5707 		 * No APIC registers are directly virtualized. This includes
5708 		 * VTPR, which can be virtualized through MOV to/from CR8 via
5709 		 * the use TPR shadow control, but not through directly
5710 		 * accessing VTPR.
5711 		 */
5712 		expectation->rd_exit_reason = VMX_VMCALL;
5713 		expectation->wr_exit_reason = VMX_VMCALL;
5714 	} else {
5715 		printf("Cannot parse APIC register virtualization config:\n"
5716 		       "\tvirtualize_apic_accesses: %d\n"
5717 		       "\tuse_tpr_shadow: %d\n"
5718 		       "\tapic_register_virtualization: %d\n"
5719 		       "\tvirtualize_x2apic_mode: %d\n"
5720 		       "\tactivate_secondary_controls: %d\n",
5721 		       config->virtualize_apic_accesses,
5722 		       config->use_tpr_shadow,
5723 		       config->apic_register_virtualization,
5724 		       config->virtualize_x2apic_mode,
5725 		       config->activate_secondary_controls);
5726 
5727 		return false;
5728 	}
5729 
5730 	return true;
5731 }
5732 
5733 struct apic_reg_test apic_reg_tests[] = {
5734 	/* Good configs, where some L2 APIC accesses are virtualized. */
5735 	{
5736 		.name = "Virtualize APIC accesses",
5737 		.apic_reg_virt_config = {
5738 			.virtualize_apic_accesses = true,
5739 			.use_tpr_shadow = false,
5740 			.apic_register_virtualization = false,
5741 			.virtualize_x2apic_mode = false,
5742 			.activate_secondary_controls = true,
5743 		},
5744 	},
5745 	{
5746 		.name = "Virtualize APIC accesses + Use TPR shadow",
5747 		.apic_reg_virt_config = {
5748 			.virtualize_apic_accesses = true,
5749 			.use_tpr_shadow = true,
5750 			.apic_register_virtualization = false,
5751 			.virtualize_x2apic_mode = false,
5752 			.activate_secondary_controls = true,
5753 		},
5754 	},
5755 	{
5756 		.name = "APIC-register virtualization",
5757 		.apic_reg_virt_config = {
5758 			.virtualize_apic_accesses = true,
5759 			.use_tpr_shadow = true,
5760 			.apic_register_virtualization = true,
5761 			.virtualize_x2apic_mode = false,
5762 			.activate_secondary_controls = true,
5763 		},
5764 	},
5765 
5766 	/*
5767 	 * Test that the secondary processor-based VM-execution controls are
5768 	 * correctly ignored when "activate secondary controls" is disabled.
5769 	 */
5770 	{
5771 		.name = "Activate secondary controls off",
5772 		.apic_reg_virt_config = {
5773 			.virtualize_apic_accesses = true,
5774 			.use_tpr_shadow = false,
5775 			.apic_register_virtualization = true,
5776 			.virtualize_x2apic_mode = true,
5777 			.activate_secondary_controls = false,
5778 		},
5779 	},
5780 	{
5781 		.name = "Activate secondary controls off + Use TPR shadow",
5782 		.apic_reg_virt_config = {
5783 			.virtualize_apic_accesses = true,
5784 			.use_tpr_shadow = true,
5785 			.apic_register_virtualization = true,
5786 			.virtualize_x2apic_mode = true,
5787 			.activate_secondary_controls = false,
5788 		},
5789 	},
5790 
5791 	/*
5792 	 * Test that the APIC access address is treated like an arbitrary memory
5793 	 * address when "virtualize APIC accesses" is disabled.
5794 	 */
5795 	{
5796 		.name = "Virtualize APIC accesses off + Use TPR shadow",
5797 		.apic_reg_virt_config = {
5798 			.virtualize_apic_accesses = false,
5799 			.use_tpr_shadow = true,
5800 			.apic_register_virtualization = true,
5801 			.virtualize_x2apic_mode = true,
5802 			.activate_secondary_controls = true,
5803 		},
5804 	},
5805 
5806 	/*
5807 	 * Test that VM entry fails due to invalid controls when
5808 	 * "APIC-register virtualization" is enabled while "use TPR shadow" is
5809 	 * disabled.
5810 	 */
5811 	{
5812 		.name = "APIC-register virtualization + Use TPR shadow off",
5813 		.apic_reg_virt_config = {
5814 			.virtualize_apic_accesses = true,
5815 			.use_tpr_shadow = false,
5816 			.apic_register_virtualization = true,
5817 			.virtualize_x2apic_mode = false,
5818 			.activate_secondary_controls = true,
5819 		},
5820 	},
5821 
5822 	/*
5823 	 * Test that VM entry fails due to invalid controls when
5824 	 * "Virtualize x2APIC mode" is enabled while "use TPR shadow" is
5825 	 * disabled.
5826 	 */
5827 	{
5828 		.name = "Virtualize x2APIC mode + Use TPR shadow off",
5829 		.apic_reg_virt_config = {
5830 			.virtualize_apic_accesses = false,
5831 			.use_tpr_shadow = false,
5832 			.apic_register_virtualization = false,
5833 			.virtualize_x2apic_mode = true,
5834 			.activate_secondary_controls = true,
5835 		},
5836 	},
5837 	{
5838 		.name = "Virtualize x2APIC mode + Use TPR shadow off v2",
5839 		.apic_reg_virt_config = {
5840 			.virtualize_apic_accesses = false,
5841 			.use_tpr_shadow = false,
5842 			.apic_register_virtualization = true,
5843 			.virtualize_x2apic_mode = true,
5844 			.activate_secondary_controls = true,
5845 		},
5846 	},
5847 
5848 	/*
5849 	 * Test that VM entry fails due to invalid controls when
5850 	 * "virtualize x2APIC mode" is enabled while "virtualize APIC accesses"
5851 	 * is enabled.
5852 	 */
5853 	{
5854 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses",
5855 		.apic_reg_virt_config = {
5856 			.virtualize_apic_accesses = true,
5857 			.use_tpr_shadow = true,
5858 			.apic_register_virtualization = false,
5859 			.virtualize_x2apic_mode = true,
5860 			.activate_secondary_controls = true,
5861 		},
5862 	},
5863 	{
5864 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses v2",
5865 		.apic_reg_virt_config = {
5866 			.virtualize_apic_accesses = true,
5867 			.use_tpr_shadow = true,
5868 			.apic_register_virtualization = true,
5869 			.virtualize_x2apic_mode = true,
5870 			.activate_secondary_controls = true,
5871 		},
5872 	},
5873 };
5874 
5875 enum Apic_op {
5876 	APIC_OP_XAPIC_RD,
5877 	APIC_OP_XAPIC_WR,
5878 	TERMINATE,
5879 };
5880 
5881 static u32 vmx_xapic_read(u32 *apic_access_address, u32 reg)
5882 {
5883 	return *(volatile u32 *)((uintptr_t)apic_access_address + reg);
5884 }
5885 
5886 static void vmx_xapic_write(u32 *apic_access_address, u32 reg, u32 val)
5887 {
5888 	*(volatile u32 *)((uintptr_t)apic_access_address + reg) = val;
5889 }
5890 
5891 struct apic_reg_virt_guest_args {
5892 	enum Apic_op op;
5893 	u32 *apic_access_address;
5894 	u32 reg;
5895 	u32 val;
5896 	bool check_rd;
5897 	u32 (*virt_fn)(u32);
5898 } apic_reg_virt_guest_args;
5899 
5900 static void apic_reg_virt_guest(void)
5901 {
5902 	volatile struct apic_reg_virt_guest_args *args =
5903 		&apic_reg_virt_guest_args;
5904 
5905 	for (;;) {
5906 		enum Apic_op op = args->op;
5907 		u32 *apic_access_address = args->apic_access_address;
5908 		u32 reg = args->reg;
5909 		u32 val = args->val;
5910 		bool check_rd = args->check_rd;
5911 		u32 (*virt_fn)(u32) = args->virt_fn;
5912 
5913 		if (op == TERMINATE)
5914 			break;
5915 
5916 		if (op == APIC_OP_XAPIC_RD) {
5917 			u32 ret = vmx_xapic_read(apic_access_address, reg);
5918 
5919 			if (check_rd) {
5920 				u32 want = virt_fn(val);
5921 				u32 got = virt_fn(ret);
5922 
5923 				report(got == want,
5924 				       "read 0x%x, expected 0x%x.", got, want);
5925 			}
5926 		} else if (op == APIC_OP_XAPIC_WR) {
5927 			vmx_xapic_write(apic_access_address, reg, val);
5928 		}
5929 
5930 		/*
5931 		 * The L1 should always execute a vmcall after it's done testing
5932 		 * an individual APIC operation. This helps to validate that the
5933 		 * L1 and L2 are in sync with each other, as expected.
5934 		 */
5935 		vmcall();
5936 	}
5937 }
5938 
5939 static void test_xapic_rd(
5940 	u32 reg, struct apic_reg_virt_expectation *expectation,
5941 	u32 *apic_access_address, u32 *virtual_apic_page)
5942 {
5943 	u32 val = expectation->val;
5944 	u32 exit_reason_want = expectation->rd_exit_reason;
5945 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5946 
5947 	report_prefix_pushf("xapic - reading 0x%03x", reg);
5948 
5949 	/* Configure guest to do an xapic read */
5950 	args->op = APIC_OP_XAPIC_RD;
5951 	args->apic_access_address = apic_access_address;
5952 	args->reg = reg;
5953 	args->val = val;
5954 	args->check_rd = exit_reason_want == VMX_VMCALL;
5955 	args->virt_fn = expectation->virt_fn;
5956 
5957 	/* Setup virtual APIC page */
5958 	if (!expectation->virtualize_apic_accesses) {
5959 		apic_access_address[apic_reg_index(reg)] = val;
5960 		virtual_apic_page[apic_reg_index(reg)] = 0;
5961 	} else if (exit_reason_want == VMX_VMCALL) {
5962 		apic_access_address[apic_reg_index(reg)] = 0;
5963 		virtual_apic_page[apic_reg_index(reg)] = val;
5964 	}
5965 
5966 	/* Enter guest */
5967 	enter_guest();
5968 
5969 	/*
5970 	 * Validate the behavior and
5971 	 * pass a magic value back to the guest.
5972 	 */
5973 	if (exit_reason_want == VMX_APIC_ACCESS) {
5974 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5975 
5976 		assert_exit_reason(exit_reason_want);
5977 		report(apic_page_offset == reg,
5978 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
5979 		       apic_page_offset, reg);
5980 		skip_exit_insn();
5981 
5982 		/* Reenter guest so it can consume/check rcx and exit again. */
5983 		enter_guest();
5984 	} else if (exit_reason_want != VMX_VMCALL) {
5985 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
5986 	}
5987 
5988 	skip_exit_vmcall();
5989 	report_prefix_pop();
5990 }
5991 
5992 static void test_xapic_wr(
5993 	u32 reg, struct apic_reg_virt_expectation *expectation,
5994 	u32 *apic_access_address, u32 *virtual_apic_page)
5995 {
5996 	u32 val = expectation->val;
5997 	u32 exit_reason_want = expectation->wr_exit_reason;
5998 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5999 	bool virtualized =
6000 		expectation->virtualize_apic_accesses &&
6001 		(exit_reason_want == VMX_APIC_WRITE ||
6002 		 exit_reason_want == VMX_VMCALL);
6003 	bool checked = false;
6004 
6005 	report_prefix_pushf("xapic - writing 0x%x to 0x%03x", val, reg);
6006 
6007 	/* Configure guest to do an xapic read */
6008 	args->op = APIC_OP_XAPIC_WR;
6009 	args->apic_access_address = apic_access_address;
6010 	args->reg = reg;
6011 	args->val = val;
6012 
6013 	/* Setup virtual APIC page */
6014 	if (virtualized || !expectation->virtualize_apic_accesses) {
6015 		apic_access_address[apic_reg_index(reg)] = 0;
6016 		virtual_apic_page[apic_reg_index(reg)] = 0;
6017 	}
6018 
6019 	/* Enter guest */
6020 	enter_guest();
6021 
6022 	/*
6023 	 * Validate the behavior and
6024 	 * pass a magic value back to the guest.
6025 	 */
6026 	if (exit_reason_want == VMX_APIC_ACCESS) {
6027 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
6028 
6029 		assert_exit_reason(exit_reason_want);
6030 		report(apic_page_offset == reg,
6031 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
6032 		       apic_page_offset, reg);
6033 		skip_exit_insn();
6034 
6035 		/* Reenter guest so it can consume/check rcx and exit again. */
6036 		enter_guest();
6037 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6038 		assert_exit_reason(exit_reason_want);
6039 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6040 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%x",
6041 		       apic_reg_index(reg),
6042 		       virtual_apic_page[apic_reg_index(reg)], val);
6043 		checked = true;
6044 
6045 		/* Reenter guest so it can consume/check rcx and exit again. */
6046 		enter_guest();
6047 	} else if (exit_reason_want != VMX_VMCALL) {
6048 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6049 	}
6050 
6051 	assert_exit_reason(VMX_VMCALL);
6052 	if (virtualized && !checked) {
6053 		u32 want = expectation->virt_fn(val);
6054 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6055 		got = expectation->virt_fn(got);
6056 
6057 		report(got == want, "exitless write; val is 0x%x, want 0x%x",
6058 		       got, want);
6059 	} else if (!expectation->virtualize_apic_accesses && !checked) {
6060 		u32 got = apic_access_address[apic_reg_index(reg)];
6061 
6062 		report(got == val,
6063 		       "non-virtualized write; val is 0x%x, want 0x%x", got,
6064 		       val);
6065 	} else if (!expectation->virtualize_apic_accesses && checked) {
6066 		report_fail("Non-virtualized write was prematurely checked!");
6067 	}
6068 
6069 	skip_exit_vmcall();
6070 	report_prefix_pop();
6071 }
6072 
6073 enum Config_type {
6074 	CONFIG_TYPE_GOOD,
6075 	CONFIG_TYPE_UNSUPPORTED,
6076 	CONFIG_TYPE_VMENTRY_FAILS_EARLY,
6077 };
6078 
6079 static enum Config_type configure_apic_reg_virt_test(
6080 	struct apic_reg_virt_config *apic_reg_virt_config)
6081 {
6082 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6083 	u32 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6084 	/* Configs where L2 entry fails early, due to invalid controls. */
6085 	bool use_tpr_shadow_incorrectly_off =
6086 		!apic_reg_virt_config->use_tpr_shadow &&
6087 		(apic_reg_virt_config->apic_register_virtualization ||
6088 		 apic_reg_virt_config->virtualize_x2apic_mode) &&
6089 		apic_reg_virt_config->activate_secondary_controls;
6090 	bool virtualize_apic_accesses_incorrectly_on =
6091 		apic_reg_virt_config->virtualize_apic_accesses &&
6092 		apic_reg_virt_config->virtualize_x2apic_mode &&
6093 		apic_reg_virt_config->activate_secondary_controls;
6094 	bool vmentry_fails_early =
6095 		use_tpr_shadow_incorrectly_off ||
6096 		virtualize_apic_accesses_incorrectly_on;
6097 
6098 	if (apic_reg_virt_config->activate_secondary_controls) {
6099 		if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
6100 			printf("VM-execution control \"activate secondary controls\" NOT supported.\n");
6101 			return CONFIG_TYPE_UNSUPPORTED;
6102 		}
6103 		cpu_exec_ctrl0 |= CPU_SECONDARY;
6104 	} else {
6105 		cpu_exec_ctrl0 &= ~CPU_SECONDARY;
6106 	}
6107 
6108 	if (apic_reg_virt_config->virtualize_apic_accesses) {
6109 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES)) {
6110 			printf("VM-execution control \"virtualize APIC accesses\" NOT supported.\n");
6111 			return CONFIG_TYPE_UNSUPPORTED;
6112 		}
6113 		cpu_exec_ctrl1 |= CPU_VIRT_APIC_ACCESSES;
6114 	} else {
6115 		cpu_exec_ctrl1 &= ~CPU_VIRT_APIC_ACCESSES;
6116 	}
6117 
6118 	if (apic_reg_virt_config->use_tpr_shadow) {
6119 		if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6120 			printf("VM-execution control \"use TPR shadow\" NOT supported.\n");
6121 			return CONFIG_TYPE_UNSUPPORTED;
6122 		}
6123 		cpu_exec_ctrl0 |= CPU_TPR_SHADOW;
6124 	} else {
6125 		cpu_exec_ctrl0 &= ~CPU_TPR_SHADOW;
6126 	}
6127 
6128 	if (apic_reg_virt_config->apic_register_virtualization) {
6129 		if (!(ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT)) {
6130 			printf("VM-execution control \"APIC-register virtualization\" NOT supported.\n");
6131 			return CONFIG_TYPE_UNSUPPORTED;
6132 		}
6133 		cpu_exec_ctrl1 |= CPU_APIC_REG_VIRT;
6134 	} else {
6135 		cpu_exec_ctrl1 &= ~CPU_APIC_REG_VIRT;
6136 	}
6137 
6138 	if (apic_reg_virt_config->virtualize_x2apic_mode) {
6139 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_X2APIC)) {
6140 			printf("VM-execution control \"virtualize x2APIC mode\" NOT supported.\n");
6141 			return CONFIG_TYPE_UNSUPPORTED;
6142 		}
6143 		cpu_exec_ctrl1 |= CPU_VIRT_X2APIC;
6144 	} else {
6145 		cpu_exec_ctrl1 &= ~CPU_VIRT_X2APIC;
6146 	}
6147 
6148 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6149 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6150 
6151 	if (vmentry_fails_early)
6152 		return CONFIG_TYPE_VMENTRY_FAILS_EARLY;
6153 
6154 	return CONFIG_TYPE_GOOD;
6155 }
6156 
6157 static bool cpu_has_apicv(void)
6158 {
6159 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
6160 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
6161 		(ctrl_pin_rev.clr & PIN_POST_INTR));
6162 }
6163 
6164 /* Validates APIC register access across valid virtualization configurations. */
6165 static void apic_reg_virt_test(void)
6166 {
6167 	u32 *apic_access_address;
6168 	u32 *virtual_apic_page;
6169 	u64 control;
6170 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6171 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6172 	int i;
6173 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
6174 
6175 	if (!cpu_has_apicv()) {
6176 		report_skip(__func__);
6177 		return;
6178 	}
6179 
6180 	control = cpu_exec_ctrl1;
6181 	control &= ~CPU_VINTD;
6182 	vmcs_write(CPU_EXEC_CTRL1, control);
6183 
6184 	test_set_guest(apic_reg_virt_guest);
6185 
6186 	/*
6187 	 * From the SDM: The 1-setting of the "virtualize APIC accesses"
6188 	 * VM-execution is guaranteed to apply only if translations to the
6189 	 * APIC-access address use a 4-KByte page.
6190 	 */
6191 	apic_access_address = alloc_page();
6192 	force_4k_page(apic_access_address);
6193 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_address));
6194 
6195 	virtual_apic_page = alloc_page();
6196 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6197 
6198 	for (i = 0; i < ARRAY_SIZE(apic_reg_tests); i++) {
6199 		struct apic_reg_test *apic_reg_test = &apic_reg_tests[i];
6200 		struct apic_reg_virt_config *apic_reg_virt_config =
6201 				&apic_reg_test->apic_reg_virt_config;
6202 		enum Config_type config_type;
6203 		u32 reg;
6204 
6205 		printf("--- %s test ---\n", apic_reg_test->name);
6206 		config_type =
6207 			configure_apic_reg_virt_test(apic_reg_virt_config);
6208 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6209 			printf("Skip because of missing features.\n");
6210 			continue;
6211 		}
6212 
6213 		if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6214 			enter_guest_with_bad_controls();
6215 			continue;
6216 		}
6217 
6218 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6219 			struct apic_reg_virt_expectation expectation = {};
6220 			bool ok;
6221 
6222 			ok = apic_reg_virt_exit_expectation(
6223 				reg, apic_reg_virt_config, &expectation);
6224 			if (!ok) {
6225 				report_fail("Malformed test.");
6226 				break;
6227 			}
6228 
6229 			test_xapic_rd(reg, &expectation, apic_access_address,
6230 				      virtual_apic_page);
6231 			test_xapic_wr(reg, &expectation, apic_access_address,
6232 				      virtual_apic_page);
6233 		}
6234 	}
6235 
6236 	/* Terminate the guest */
6237 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6238 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6239 	args->op = TERMINATE;
6240 	enter_guest();
6241 	assert_exit_reason(VMX_VMCALL);
6242 }
6243 
6244 struct virt_x2apic_mode_config {
6245 	struct apic_reg_virt_config apic_reg_virt_config;
6246 	bool virtual_interrupt_delivery;
6247 	bool use_msr_bitmaps;
6248 	bool disable_x2apic_msr_intercepts;
6249 	bool disable_x2apic;
6250 };
6251 
6252 struct virt_x2apic_mode_test_case {
6253 	const char *name;
6254 	struct virt_x2apic_mode_config virt_x2apic_mode_config;
6255 };
6256 
6257 enum Virt_x2apic_mode_behavior_type {
6258 	X2APIC_ACCESS_VIRTUALIZED,
6259 	X2APIC_ACCESS_PASSED_THROUGH,
6260 	X2APIC_ACCESS_TRIGGERS_GP,
6261 };
6262 
6263 struct virt_x2apic_mode_expectation {
6264 	enum Reason rd_exit_reason;
6265 	enum Reason wr_exit_reason;
6266 
6267 	/*
6268 	 * RDMSR and WRMSR handle 64-bit values. However, except for ICR, all of
6269 	 * the x2APIC registers are 32 bits. Notice:
6270 	 *   1. vmx_x2apic_read() clears the upper 32 bits for 32-bit registers.
6271 	 *   2. vmx_x2apic_write() expects the val arg to be well-formed.
6272 	 */
6273 	u64 rd_val;
6274 	u64 wr_val;
6275 
6276 	/*
6277 	 * Compares input to virtualized output;
6278 	 * 1st arg is pointer to return expected virtualization output.
6279 	 */
6280 	u64 (*virt_fn)(u64);
6281 
6282 	enum Virt_x2apic_mode_behavior_type rd_behavior;
6283 	enum Virt_x2apic_mode_behavior_type wr_behavior;
6284 	bool wr_only;
6285 };
6286 
6287 static u64 virt_x2apic_mode_identity(u64 val)
6288 {
6289 	return val;
6290 }
6291 
6292 static u64 virt_x2apic_mode_nibble1(u64 val)
6293 {
6294 	return val & 0xf0;
6295 }
6296 
6297 static void virt_x2apic_mode_rd_expectation(
6298 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6299 	bool apic_register_virtualization, bool virtual_interrupt_delivery,
6300 	struct virt_x2apic_mode_expectation *expectation)
6301 {
6302 	bool readable =
6303 		!x2apic_reg_reserved(reg) &&
6304 		reg != APIC_EOI;
6305 
6306 	expectation->rd_exit_reason = VMX_VMCALL;
6307 	expectation->virt_fn = virt_x2apic_mode_identity;
6308 	if (virt_x2apic_mode_on && apic_register_virtualization) {
6309 		expectation->rd_val = MAGIC_VAL_1;
6310 		if (reg == APIC_PROCPRI && virtual_interrupt_delivery)
6311 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6312 		else if (reg == APIC_TASKPRI)
6313 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6314 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6315 	} else if (virt_x2apic_mode_on && !apic_register_virtualization &&
6316 		   reg == APIC_TASKPRI) {
6317 		expectation->rd_val = MAGIC_VAL_1;
6318 		expectation->virt_fn = virt_x2apic_mode_nibble1;
6319 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6320 	} else if (!disable_x2apic && readable) {
6321 		expectation->rd_val = apic_read(reg);
6322 		expectation->rd_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6323 	} else {
6324 		expectation->rd_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6325 	}
6326 }
6327 
6328 /*
6329  * get_x2apic_wr_val() creates an innocuous write value for an x2APIC register.
6330  *
6331  * For writable registers, get_x2apic_wr_val() deposits the write value into the
6332  * val pointer arg and returns true. For non-writable registers, val is not
6333  * modified and get_x2apic_wr_val() returns false.
6334  */
6335 static bool get_x2apic_wr_val(u32 reg, u64 *val)
6336 {
6337 	switch (reg) {
6338 	case APIC_TASKPRI:
6339 		/* Bits 31:8 are reserved. */
6340 		*val &= 0xff;
6341 		break;
6342 	case APIC_EOI:
6343 	case APIC_ESR:
6344 	case APIC_TMICT:
6345 		/*
6346 		 * EOI, ESR: WRMSR of a non-zero value causes #GP(0).
6347 		 * TMICT: A write of 0 to the initial-count register effectively
6348 		 *        stops the local APIC timer, in both one-shot and
6349 		 *        periodic mode.
6350 		 */
6351 		*val = 0;
6352 		break;
6353 	case APIC_SPIV:
6354 	case APIC_LVTT:
6355 	case APIC_LVTTHMR:
6356 	case APIC_LVTPC:
6357 	case APIC_LVT0:
6358 	case APIC_LVT1:
6359 	case APIC_LVTERR:
6360 	case APIC_TDCR:
6361 		/*
6362 		 * To avoid writing a 1 to a reserved bit or causing some other
6363 		 * unintended side effect, read the current value and use it as
6364 		 * the write value.
6365 		 */
6366 		*val = apic_read(reg);
6367 		break;
6368 	case APIC_CMCI:
6369 		if (!apic_lvt_entry_supported(6))
6370 			return false;
6371 		*val = apic_read(reg);
6372 		break;
6373 	case APIC_ICR:
6374 		*val = 0x40000 | 0xf1;
6375 		break;
6376 	case APIC_SELF_IPI:
6377 		/*
6378 		 * With special processing (i.e., virtualize x2APIC mode +
6379 		 * virtual interrupt delivery), writing zero causes an
6380 		 * APIC-write VM exit. We plan to add a test for enabling
6381 		 * "virtual-interrupt delivery" in VMCS12, and that's where we
6382 		 * will test a self IPI with special processing.
6383 		 */
6384 		*val = 0x0;
6385 		break;
6386 	default:
6387 		return false;
6388 	}
6389 
6390 	return true;
6391 }
6392 
6393 static bool special_processing_applies(u32 reg, u64 *val,
6394 				       bool virt_int_delivery)
6395 {
6396 	bool special_processing =
6397 		(reg == APIC_TASKPRI) ||
6398 		(virt_int_delivery &&
6399 		 (reg == APIC_EOI || reg == APIC_SELF_IPI));
6400 
6401 	if (special_processing) {
6402 		TEST_ASSERT(get_x2apic_wr_val(reg, val));
6403 		return true;
6404 	}
6405 
6406 	return false;
6407 }
6408 
6409 static void virt_x2apic_mode_wr_expectation(
6410 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6411 	bool virt_int_delivery,
6412 	struct virt_x2apic_mode_expectation *expectation)
6413 {
6414 	expectation->wr_exit_reason = VMX_VMCALL;
6415 	expectation->wr_val = MAGIC_VAL_1;
6416 	expectation->wr_only = false;
6417 
6418 	if (virt_x2apic_mode_on &&
6419 	    special_processing_applies(reg, &expectation->wr_val,
6420 				       virt_int_delivery)) {
6421 		expectation->wr_behavior = X2APIC_ACCESS_VIRTUALIZED;
6422 		if (reg == APIC_SELF_IPI)
6423 			expectation->wr_exit_reason = VMX_APIC_WRITE;
6424 	} else if (!disable_x2apic &&
6425 		   get_x2apic_wr_val(reg, &expectation->wr_val)) {
6426 		expectation->wr_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6427 		if (reg == APIC_EOI || reg == APIC_SELF_IPI)
6428 			expectation->wr_only = true;
6429 		if (reg == APIC_ICR)
6430 			expectation->wr_exit_reason = VMX_EXTINT;
6431 	} else {
6432 		expectation->wr_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6433 		/*
6434 		 * Writing 1 to a reserved bit triggers a #GP.
6435 		 * Thus, set the write value to 0, which seems
6436 		 * the most likely to detect a missed #GP.
6437 		 */
6438 		expectation->wr_val = 0;
6439 	}
6440 }
6441 
6442 static void virt_x2apic_mode_exit_expectation(
6443 	u32 reg, struct virt_x2apic_mode_config *config,
6444 	struct virt_x2apic_mode_expectation *expectation)
6445 {
6446 	struct apic_reg_virt_config *base_config =
6447 		&config->apic_reg_virt_config;
6448 	bool virt_x2apic_mode_on =
6449 		base_config->virtualize_x2apic_mode &&
6450 		config->use_msr_bitmaps &&
6451 		config->disable_x2apic_msr_intercepts &&
6452 		base_config->activate_secondary_controls;
6453 
6454 	virt_x2apic_mode_wr_expectation(
6455 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6456 		config->virtual_interrupt_delivery, expectation);
6457 	virt_x2apic_mode_rd_expectation(
6458 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6459 		base_config->apic_register_virtualization,
6460 		config->virtual_interrupt_delivery, expectation);
6461 }
6462 
6463 struct virt_x2apic_mode_test_case virt_x2apic_mode_tests[] = {
6464 	/*
6465 	 * Baseline "virtualize x2APIC mode" configuration:
6466 	 *   - virtualize x2APIC mode
6467 	 *   - virtual-interrupt delivery
6468 	 *   - APIC-register virtualization
6469 	 *   - x2APIC MSR intercepts disabled
6470 	 *
6471 	 * Reads come from virtual APIC page, special processing applies to
6472 	 * VTPR, EOI, and SELF IPI, and all other writes pass through to L1
6473 	 * APIC.
6474 	 */
6475 	{
6476 		.name = "Baseline",
6477 		.virt_x2apic_mode_config = {
6478 			.virtual_interrupt_delivery = true,
6479 			.use_msr_bitmaps = true,
6480 			.disable_x2apic_msr_intercepts = true,
6481 			.disable_x2apic = false,
6482 			.apic_reg_virt_config = {
6483 				.apic_register_virtualization = true,
6484 				.use_tpr_shadow = true,
6485 				.virtualize_apic_accesses = false,
6486 				.virtualize_x2apic_mode = true,
6487 				.activate_secondary_controls = true,
6488 			},
6489 		},
6490 	},
6491 	{
6492 		.name = "Baseline w/ x2apic disabled",
6493 		.virt_x2apic_mode_config = {
6494 			.virtual_interrupt_delivery = true,
6495 			.use_msr_bitmaps = true,
6496 			.disable_x2apic_msr_intercepts = true,
6497 			.disable_x2apic = true,
6498 			.apic_reg_virt_config = {
6499 				.apic_register_virtualization = true,
6500 				.use_tpr_shadow = true,
6501 				.virtualize_apic_accesses = false,
6502 				.virtualize_x2apic_mode = true,
6503 				.activate_secondary_controls = true,
6504 			},
6505 		},
6506 	},
6507 
6508 	/*
6509 	 * Baseline, minus virtual-interrupt delivery. Reads come from virtual
6510 	 * APIC page, special processing applies to VTPR, and all other writes
6511 	 * pass through to L1 APIC.
6512 	 */
6513 	{
6514 		.name = "Baseline - virtual interrupt delivery",
6515 		.virt_x2apic_mode_config = {
6516 			.virtual_interrupt_delivery = false,
6517 			.use_msr_bitmaps = true,
6518 			.disable_x2apic_msr_intercepts = true,
6519 			.disable_x2apic = false,
6520 			.apic_reg_virt_config = {
6521 				.apic_register_virtualization = true,
6522 				.use_tpr_shadow = true,
6523 				.virtualize_apic_accesses = false,
6524 				.virtualize_x2apic_mode = true,
6525 				.activate_secondary_controls = true,
6526 			},
6527 		},
6528 	},
6529 
6530 	/*
6531 	 * Baseline, minus APIC-register virtualization. x2APIC reads pass
6532 	 * through to L1's APIC, unless reading VTPR
6533 	 */
6534 	{
6535 		.name = "Virtualize x2APIC mode, no APIC reg virt",
6536 		.virt_x2apic_mode_config = {
6537 			.virtual_interrupt_delivery = true,
6538 			.use_msr_bitmaps = true,
6539 			.disable_x2apic_msr_intercepts = true,
6540 			.disable_x2apic = false,
6541 			.apic_reg_virt_config = {
6542 				.apic_register_virtualization = false,
6543 				.use_tpr_shadow = true,
6544 				.virtualize_apic_accesses = false,
6545 				.virtualize_x2apic_mode = true,
6546 				.activate_secondary_controls = true,
6547 			},
6548 		},
6549 	},
6550 	{
6551 		.name = "Virtualize x2APIC mode, no APIC reg virt, x2APIC off",
6552 		.virt_x2apic_mode_config = {
6553 			.virtual_interrupt_delivery = true,
6554 			.use_msr_bitmaps = true,
6555 			.disable_x2apic_msr_intercepts = true,
6556 			.disable_x2apic = true,
6557 			.apic_reg_virt_config = {
6558 				.apic_register_virtualization = false,
6559 				.use_tpr_shadow = true,
6560 				.virtualize_apic_accesses = false,
6561 				.virtualize_x2apic_mode = true,
6562 				.activate_secondary_controls = true,
6563 			},
6564 		},
6565 	},
6566 
6567 	/*
6568 	 * Enable "virtualize x2APIC mode" and "APIC-register virtualization",
6569 	 * and disable intercepts for the x2APIC MSRs, but fail to enable
6570 	 * "activate secondary controls" (i.e. L2 gets access to L1's x2APIC
6571 	 * MSRs).
6572 	 */
6573 	{
6574 		.name = "Fail to enable activate secondary controls",
6575 		.virt_x2apic_mode_config = {
6576 			.virtual_interrupt_delivery = true,
6577 			.use_msr_bitmaps = true,
6578 			.disable_x2apic_msr_intercepts = true,
6579 			.disable_x2apic = false,
6580 			.apic_reg_virt_config = {
6581 				.apic_register_virtualization = true,
6582 				.use_tpr_shadow = true,
6583 				.virtualize_apic_accesses = false,
6584 				.virtualize_x2apic_mode = true,
6585 				.activate_secondary_controls = false,
6586 			},
6587 		},
6588 	},
6589 
6590 	/*
6591 	 * Enable "APIC-register virtualization" and enable "activate secondary
6592 	 * controls" and disable intercepts for the x2APIC MSRs, but do not
6593 	 * enable the "virtualize x2APIC mode" VM-execution control (i.e. L2
6594 	 * gets access to L1's x2APIC MSRs).
6595 	 */
6596 	{
6597 		.name = "Fail to enable virtualize x2APIC mode",
6598 		.virt_x2apic_mode_config = {
6599 			.virtual_interrupt_delivery = true,
6600 			.use_msr_bitmaps = true,
6601 			.disable_x2apic_msr_intercepts = true,
6602 			.disable_x2apic = false,
6603 			.apic_reg_virt_config = {
6604 				.apic_register_virtualization = true,
6605 				.use_tpr_shadow = true,
6606 				.virtualize_apic_accesses = false,
6607 				.virtualize_x2apic_mode = false,
6608 				.activate_secondary_controls = true,
6609 			},
6610 		},
6611 	},
6612 
6613 	/*
6614 	 * Disable "Virtualize x2APIC mode", disable x2APIC MSR intercepts, and
6615 	 * enable "APIC-register virtualization" --> L2 gets L1's x2APIC MSRs.
6616 	 */
6617 	{
6618 		.name = "Baseline",
6619 		.virt_x2apic_mode_config = {
6620 			.virtual_interrupt_delivery = true,
6621 			.use_msr_bitmaps = true,
6622 			.disable_x2apic_msr_intercepts = true,
6623 			.disable_x2apic = false,
6624 			.apic_reg_virt_config = {
6625 				.apic_register_virtualization = true,
6626 				.use_tpr_shadow = true,
6627 				.virtualize_apic_accesses = false,
6628 				.virtualize_x2apic_mode = false,
6629 				.activate_secondary_controls = true,
6630 			},
6631 		},
6632 	},
6633 };
6634 
6635 enum X2apic_op {
6636 	X2APIC_OP_RD,
6637 	X2APIC_OP_WR,
6638 	X2APIC_TERMINATE,
6639 };
6640 
6641 static u64 vmx_x2apic_read(u32 reg)
6642 {
6643 	u32 msr_addr = x2apic_msr(reg);
6644 	u64 val;
6645 
6646 	val = rdmsr(msr_addr);
6647 
6648 	return val;
6649 }
6650 
6651 static void vmx_x2apic_write(u32 reg, u64 val)
6652 {
6653 	u32 msr_addr = x2apic_msr(reg);
6654 
6655 	wrmsr(msr_addr, val);
6656 }
6657 
6658 struct virt_x2apic_mode_guest_args {
6659 	enum X2apic_op op;
6660 	u32 reg;
6661 	u64 val;
6662 	bool should_gp;
6663 	u64 (*virt_fn)(u64);
6664 } virt_x2apic_mode_guest_args;
6665 
6666 static volatile bool handle_x2apic_gp_ran;
6667 static volatile u32 handle_x2apic_gp_insn_len;
6668 static void handle_x2apic_gp(struct ex_regs *regs)
6669 {
6670 	handle_x2apic_gp_ran = true;
6671 	regs->rip += handle_x2apic_gp_insn_len;
6672 }
6673 
6674 static handler setup_x2apic_gp_handler(void)
6675 {
6676 	handler old_handler;
6677 
6678 	old_handler = handle_exception(GP_VECTOR, handle_x2apic_gp);
6679 	/* RDMSR and WRMSR are both 2 bytes, assuming no prefixes. */
6680 	handle_x2apic_gp_insn_len = 2;
6681 
6682 	return old_handler;
6683 }
6684 
6685 static void teardown_x2apic_gp_handler(handler old_handler)
6686 {
6687 	handle_exception(GP_VECTOR, old_handler);
6688 
6689 	/*
6690 	 * Defensively reset instruction length, so that if the handler is
6691 	 * incorrectly used, it will loop infinitely, rather than run off into
6692 	 * la la land.
6693 	 */
6694 	handle_x2apic_gp_insn_len = 0;
6695 	handle_x2apic_gp_ran = false;
6696 }
6697 
6698 static void virt_x2apic_mode_guest(void)
6699 {
6700 	volatile struct virt_x2apic_mode_guest_args *args =
6701 		&virt_x2apic_mode_guest_args;
6702 
6703 	for (;;) {
6704 		enum X2apic_op op = args->op;
6705 		u32 reg = args->reg;
6706 		u64 val = args->val;
6707 		bool should_gp = args->should_gp;
6708 		u64 (*virt_fn)(u64) = args->virt_fn;
6709 		handler old_handler;
6710 
6711 		if (op == X2APIC_TERMINATE)
6712 			break;
6713 
6714 		if (should_gp) {
6715 			TEST_ASSERT(!handle_x2apic_gp_ran);
6716 			old_handler = setup_x2apic_gp_handler();
6717 		}
6718 
6719 		if (op == X2APIC_OP_RD) {
6720 			u64 ret = vmx_x2apic_read(reg);
6721 
6722 			if (!should_gp) {
6723 				u64 want = virt_fn(val);
6724 				u64 got = virt_fn(ret);
6725 
6726 				report(got == want,
6727 				       "APIC read; got 0x%lx, want 0x%lx.",
6728 				       got, want);
6729 			}
6730 		} else if (op == X2APIC_OP_WR) {
6731 			vmx_x2apic_write(reg, val);
6732 		}
6733 
6734 		if (should_gp) {
6735 			report(handle_x2apic_gp_ran,
6736 			       "x2APIC op triggered GP.");
6737 			teardown_x2apic_gp_handler(old_handler);
6738 		}
6739 
6740 		/*
6741 		 * The L1 should always execute a vmcall after it's done testing
6742 		 * an individual APIC operation. This helps to validate that the
6743 		 * L1 and L2 are in sync with each other, as expected.
6744 		 */
6745 		vmcall();
6746 	}
6747 }
6748 
6749 static void test_x2apic_rd(
6750 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6751 	u32 *virtual_apic_page)
6752 {
6753 	u64 val = expectation->rd_val;
6754 	u32 exit_reason_want = expectation->rd_exit_reason;
6755 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6756 
6757 	report_prefix_pushf("x2apic - reading 0x%03x", reg);
6758 
6759 	/* Configure guest to do an x2apic read */
6760 	args->op = X2APIC_OP_RD;
6761 	args->reg = reg;
6762 	args->val = val;
6763 	args->should_gp = expectation->rd_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6764 	args->virt_fn = expectation->virt_fn;
6765 
6766 	/* Setup virtual APIC page */
6767 	if (expectation->rd_behavior == X2APIC_ACCESS_VIRTUALIZED)
6768 		virtual_apic_page[apic_reg_index(reg)] = (u32)val;
6769 
6770 	/* Enter guest */
6771 	enter_guest();
6772 
6773 	if (exit_reason_want != VMX_VMCALL) {
6774 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6775 	}
6776 
6777 	skip_exit_vmcall();
6778 	report_prefix_pop();
6779 }
6780 
6781 static volatile bool handle_x2apic_ipi_ran;
6782 static void handle_x2apic_ipi(isr_regs_t *regs)
6783 {
6784 	handle_x2apic_ipi_ran = true;
6785 	eoi();
6786 }
6787 
6788 static void test_x2apic_wr(
6789 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6790 	u32 *virtual_apic_page)
6791 {
6792 	u64 val = expectation->wr_val;
6793 	u32 exit_reason_want = expectation->wr_exit_reason;
6794 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6795 	int ipi_vector = 0xf1;
6796 	u32 restore_val = 0;
6797 
6798 	report_prefix_pushf("x2apic - writing 0x%lx to 0x%03x", val, reg);
6799 
6800 	/* Configure guest to do an x2apic read */
6801 	args->op = X2APIC_OP_WR;
6802 	args->reg = reg;
6803 	args->val = val;
6804 	args->should_gp = expectation->wr_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6805 
6806 	/* Setup virtual APIC page */
6807 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED)
6808 		virtual_apic_page[apic_reg_index(reg)] = 0;
6809 	if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH && !expectation->wr_only)
6810 		restore_val = apic_read(reg);
6811 
6812 	/* Setup IPI handler */
6813 	handle_x2apic_ipi_ran = false;
6814 	handle_irq(ipi_vector, handle_x2apic_ipi);
6815 
6816 	/* Enter guest */
6817 	enter_guest();
6818 
6819 	/*
6820 	 * Validate the behavior and
6821 	 * pass a magic value back to the guest.
6822 	 */
6823 	if (exit_reason_want == VMX_EXTINT) {
6824 		assert_exit_reason(exit_reason_want);
6825 
6826 		/* Clear the external interrupt. */
6827 		irq_enable();
6828 		asm volatile ("nop");
6829 		irq_disable();
6830 		report(handle_x2apic_ipi_ran,
6831 		       "Got pending interrupt after IRQ enabled.");
6832 
6833 		enter_guest();
6834 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6835 		assert_exit_reason(exit_reason_want);
6836 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6837 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%lx",
6838 		       apic_reg_index(reg),
6839 		       virtual_apic_page[apic_reg_index(reg)], val);
6840 
6841 		/* Reenter guest so it can consume/check rcx and exit again. */
6842 		enter_guest();
6843 	} else if (exit_reason_want != VMX_VMCALL) {
6844 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6845 	}
6846 
6847 	assert_exit_reason(VMX_VMCALL);
6848 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) {
6849 		u64 want = val;
6850 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6851 
6852 		report(got == want, "x2APIC write; got 0x%x, want 0x%lx", got,
6853 		       want);
6854 	} else if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH) {
6855 		if (!expectation->wr_only) {
6856 			u32 got = apic_read(reg);
6857 			bool ok;
6858 
6859 			/*
6860 			 * When L1's TPR is passed through to L2, the lower
6861 			 * nibble can be lost. For example, if L2 executes
6862 			 * WRMSR(0x808, 0x78), then, L1 might read 0x70.
6863 			 *
6864 			 * Here's how the lower nibble can get lost:
6865 			 *   1. L2 executes WRMSR(0x808, 0x78).
6866 			 *   2. L2 exits to L0 with a WRMSR exit.
6867 			 *   3. L0 emulates WRMSR, by writing L1's TPR.
6868 			 *   4. L0 re-enters L2.
6869 			 *   5. L2 exits to L0 (reason doesn't matter).
6870 			 *   6. L0 reflects L2's exit to L1.
6871 			 *   7. Before entering L1, L0 exits to user-space
6872 			 *      (e.g., to satisfy TPR access reporting).
6873 			 *   8. User-space executes KVM_SET_REGS ioctl, which
6874 			 *      clears the lower nibble of L1's TPR.
6875 			 */
6876 			if (reg == APIC_TASKPRI) {
6877 				got = apic_virt_nibble1(got);
6878 				val = apic_virt_nibble1(val);
6879 			}
6880 
6881 			ok = got == val;
6882 			report(ok,
6883 			       "non-virtualized write; val is 0x%x, want 0x%lx",
6884 			       got, val);
6885 			apic_write(reg, restore_val);
6886 		} else {
6887 			report_pass("non-virtualized and write-only OK");
6888 		}
6889 	}
6890 	skip_exit_insn();
6891 
6892 	report_prefix_pop();
6893 }
6894 
6895 static enum Config_type configure_virt_x2apic_mode_test(
6896 	struct virt_x2apic_mode_config *virt_x2apic_mode_config,
6897 	u8 *msr_bitmap_page)
6898 {
6899 	int msr;
6900 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6901 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6902 
6903 	/* x2apic-specific VMCS config */
6904 	if (virt_x2apic_mode_config->use_msr_bitmaps) {
6905 		/* virt_x2apic_mode_test() checks for MSR bitmaps support */
6906 		cpu_exec_ctrl0 |= CPU_MSR_BITMAP;
6907 	} else {
6908 		cpu_exec_ctrl0 &= ~CPU_MSR_BITMAP;
6909 	}
6910 
6911 	if (virt_x2apic_mode_config->virtual_interrupt_delivery) {
6912 		if (!(ctrl_cpu_rev[1].clr & CPU_VINTD)) {
6913 			report_skip("VM-execution control \"virtual-interrupt delivery\" NOT supported.\n");
6914 			return CONFIG_TYPE_UNSUPPORTED;
6915 		}
6916 		cpu_exec_ctrl1 |= CPU_VINTD;
6917 	} else {
6918 		cpu_exec_ctrl1 &= ~CPU_VINTD;
6919 	}
6920 
6921 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6922 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6923 
6924 	/* x2APIC MSR intercepts are usually off for "Virtualize x2APIC mode" */
6925 	for (msr = 0x800; msr <= 0x8ff; msr++) {
6926 		if (virt_x2apic_mode_config->disable_x2apic_msr_intercepts) {
6927 			clear_bit(msr, msr_bitmap_page + 0x000);
6928 			clear_bit(msr, msr_bitmap_page + 0x800);
6929 		} else {
6930 			set_bit(msr, msr_bitmap_page + 0x000);
6931 			set_bit(msr, msr_bitmap_page + 0x800);
6932 		}
6933 	}
6934 
6935 	/* x2APIC mode can impact virtualization */
6936 	reset_apic();
6937 	if (!virt_x2apic_mode_config->disable_x2apic)
6938 		enable_x2apic();
6939 
6940 	return configure_apic_reg_virt_test(
6941 		&virt_x2apic_mode_config->apic_reg_virt_config);
6942 }
6943 
6944 static void virt_x2apic_mode_test(void)
6945 {
6946 	u32 *virtual_apic_page;
6947 	u8 *msr_bitmap_page;
6948 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6949 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6950 	int i;
6951 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6952 
6953 	if (!cpu_has_apicv()) {
6954 		report_skip(__func__);
6955 		return;
6956 	}
6957 
6958 	/*
6959 	 * This is to exercise an issue in KVM's logic to merge L0's and L1's
6960 	 * MSR bitmaps. Previously, an L1 could get at L0's x2APIC MSRs by
6961 	 * writing the IA32_SPEC_CTRL MSR or the IA32_PRED_CMD MSRs. KVM would
6962 	 * then proceed to manipulate the MSR bitmaps, as if VMCS12 had the
6963 	 * "Virtualize x2APIC mod" control set, even when it didn't.
6964 	 */
6965 	if (has_spec_ctrl())
6966 		wrmsr(MSR_IA32_SPEC_CTRL, 1);
6967 
6968 	/*
6969 	 * Check that VMCS12 supports:
6970 	 *   - "Virtual-APIC address", indicated by "use TPR shadow"
6971 	 *   - "MSR-bitmap address", indicated by "use MSR bitmaps"
6972 	 */
6973 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6974 		report_skip("VM-execution control \"use TPR shadow\" NOT supported.\n");
6975 		return;
6976 	} else if (!(ctrl_cpu_rev[0].clr & CPU_MSR_BITMAP)) {
6977 		report_skip("VM-execution control \"use MSR bitmaps\" NOT supported.\n");
6978 		return;
6979 	}
6980 
6981 	test_set_guest(virt_x2apic_mode_guest);
6982 
6983 	virtual_apic_page = alloc_page();
6984 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6985 
6986 	msr_bitmap_page = alloc_page();
6987 	memset(msr_bitmap_page, 0xff, PAGE_SIZE);
6988 	vmcs_write(MSR_BITMAP, virt_to_phys(msr_bitmap_page));
6989 
6990 	for (i = 0; i < ARRAY_SIZE(virt_x2apic_mode_tests); i++) {
6991 		struct virt_x2apic_mode_test_case *virt_x2apic_mode_test_case =
6992 			&virt_x2apic_mode_tests[i];
6993 		struct virt_x2apic_mode_config *virt_x2apic_mode_config =
6994 			&virt_x2apic_mode_test_case->virt_x2apic_mode_config;
6995 		enum Config_type config_type;
6996 		u32 reg;
6997 
6998 		printf("--- %s test ---\n", virt_x2apic_mode_test_case->name);
6999 		config_type =
7000 			configure_virt_x2apic_mode_test(virt_x2apic_mode_config,
7001 							msr_bitmap_page);
7002 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
7003 			report_skip("Skip because of missing features.\n");
7004 			continue;
7005 		} else if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
7006 			enter_guest_with_bad_controls();
7007 			continue;
7008 		}
7009 
7010 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
7011 			struct virt_x2apic_mode_expectation expectation;
7012 
7013 			virt_x2apic_mode_exit_expectation(
7014 				reg, virt_x2apic_mode_config, &expectation);
7015 
7016 			test_x2apic_rd(reg, &expectation, virtual_apic_page);
7017 			test_x2apic_wr(reg, &expectation, virtual_apic_page);
7018 		}
7019 	}
7020 
7021 
7022 	/* Terminate the guest */
7023 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
7024 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
7025 	args->op = X2APIC_TERMINATE;
7026 	enter_guest();
7027 	assert_exit_reason(VMX_VMCALL);
7028 }
7029 
7030 static void test_ctl_reg(const char *cr_name, u64 cr, u64 fixed0, u64 fixed1)
7031 {
7032 	u64 val;
7033 	u64 cr_saved = vmcs_read(cr);
7034 	int i;
7035 
7036 	val = fixed0 & fixed1;
7037 	if (cr == HOST_CR4)
7038 		vmcs_write(cr, val | X86_CR4_PAE);
7039 	else
7040 		vmcs_write(cr, val);
7041 	report_prefix_pushf("%s %lx", cr_name, val);
7042 	if (val == fixed0)
7043 		test_vmx_vmlaunch(0);
7044 	else
7045 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7046 	report_prefix_pop();
7047 
7048 	for (i = 0; i < 64; i++) {
7049 
7050 		/* Set a bit when the corresponding bit in fixed1 is 0 */
7051 		if ((fixed1 & (1ull << i)) == 0) {
7052 			if (cr == HOST_CR4 && ((1ull << i) & X86_CR4_SMEP ||
7053 					       (1ull << i) & X86_CR4_SMAP))
7054 				continue;
7055 
7056 			vmcs_write(cr, cr_saved | (1ull << i));
7057 			report_prefix_pushf("%s %llx", cr_name,
7058 						cr_saved | (1ull << i));
7059 			test_vmx_vmlaunch(
7060 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7061 			report_prefix_pop();
7062 		}
7063 
7064 		/* Unset a bit when the corresponding bit in fixed0 is 1 */
7065 		if (fixed0 & (1ull << i)) {
7066 			vmcs_write(cr, cr_saved & ~(1ull << i));
7067 			report_prefix_pushf("%s %llx", cr_name,
7068 						cr_saved & ~(1ull << i));
7069 			test_vmx_vmlaunch(
7070 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7071 			report_prefix_pop();
7072 		}
7073 	}
7074 
7075 	vmcs_write(cr, cr_saved);
7076 }
7077 
7078 /*
7079  * 1. The CR0 field must not set any bit to a value not supported in VMX
7080  *    operation.
7081  * 2. The CR4 field must not set any bit to a value not supported in VMX
7082  *    operation.
7083  * 3. On processors that support Intel 64 architecture, the CR3 field must
7084  *    be such that bits 63:52 and bits in the range 51:32 beyond the
7085  *    processor's physical-address width must be 0.
7086  *
7087  *  [Intel SDM]
7088  */
7089 static void test_host_ctl_regs(void)
7090 {
7091 	u64 fixed0, fixed1, cr3, cr3_saved;
7092 	int i;
7093 
7094 	/* Test CR0 */
7095 	fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0);
7096 	fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1);
7097 	test_ctl_reg("HOST_CR0", HOST_CR0, fixed0, fixed1);
7098 
7099 	/* Test CR4 */
7100 	fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0);
7101 	fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1) &
7102 		 ~(X86_CR4_SMEP | X86_CR4_SMAP);
7103 	test_ctl_reg("HOST_CR4", HOST_CR4, fixed0, fixed1);
7104 
7105 	/* Test CR3 */
7106 	cr3_saved = vmcs_read(HOST_CR3);
7107 	for (i = cpuid_maxphyaddr(); i < 64; i++) {
7108 		cr3 = cr3_saved | (1ul << i);
7109 		vmcs_write(HOST_CR3, cr3);
7110 		report_prefix_pushf("HOST_CR3 %lx", cr3);
7111 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7112 		report_prefix_pop();
7113 	}
7114 
7115 	vmcs_write(HOST_CR3, cr3_saved);
7116 }
7117 
7118 static void test_efer_vmlaunch(u32 fld, bool ok)
7119 {
7120 	if (fld == HOST_EFER) {
7121 		if (ok)
7122 			test_vmx_vmlaunch(0);
7123 		else
7124 			test_vmx_vmlaunch2(VMXERR_ENTRY_INVALID_CONTROL_FIELD,
7125 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7126 	} else {
7127 		test_guest_state("EFER test", !ok, GUEST_EFER, "GUEST_EFER");
7128 	}
7129 }
7130 
7131 static void test_efer_one(u32 fld, const char * fld_name, u64 efer,
7132 			  u32 ctrl_fld, u64 ctrl,
7133 			  int i, const char *efer_bit_name)
7134 {
7135 	bool ok;
7136 
7137 	ok = true;
7138 	if (ctrl_fld == EXI_CONTROLS && (ctrl & EXI_LOAD_EFER)) {
7139 		if (!!(efer & EFER_LMA) != !!(ctrl & EXI_HOST_64))
7140 			ok = false;
7141 		if (!!(efer & EFER_LME) != !!(ctrl & EXI_HOST_64))
7142 			ok = false;
7143 	}
7144 	if (ctrl_fld == ENT_CONTROLS && (ctrl & ENT_LOAD_EFER)) {
7145 		/* Check LMA too since CR0.PG is set.  */
7146 		if (!!(efer & EFER_LMA) != !!(ctrl & ENT_GUEST_64))
7147 			ok = false;
7148 		if (!!(efer & EFER_LME) != !!(ctrl & ENT_GUEST_64))
7149 			ok = false;
7150 	}
7151 
7152 	/*
7153 	 * Skip the test if it would enter the guest in 32-bit mode.
7154 	 * Perhaps write the test in assembly and make sure it
7155 	 * can be run in either mode?
7156 	 */
7157 	if (fld == GUEST_EFER && ok && !(ctrl & ENT_GUEST_64))
7158 		return;
7159 
7160 	vmcs_write(ctrl_fld, ctrl);
7161 	vmcs_write(fld, efer);
7162 	report_prefix_pushf("%s %s bit turned %s, controls %s",
7163 			    fld_name, efer_bit_name,
7164 			    (i & 1) ? "on" : "off",
7165 			    (i & 2) ? "on" : "off");
7166 
7167 	test_efer_vmlaunch(fld, ok);
7168 	report_prefix_pop();
7169 }
7170 
7171 static void test_efer_bit(u32 fld, const char * fld_name,
7172 			  u32 ctrl_fld, u64 ctrl_bit, u64 efer_bit,
7173 			  const char *efer_bit_name)
7174 {
7175 	u64 efer_saved = vmcs_read(fld);
7176 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7177 	int i;
7178 
7179 	for (i = 0; i < 4; i++) {
7180 		u64 efer = efer_saved & ~efer_bit;
7181 		u64 ctrl = ctrl_saved & ~ctrl_bit;
7182 
7183 		if (i & 1)
7184 			efer |= efer_bit;
7185 		if (i & 2)
7186 			ctrl |= ctrl_bit;
7187 
7188 		test_efer_one(fld, fld_name, efer, ctrl_fld, ctrl,
7189 			      i, efer_bit_name);
7190 	}
7191 
7192 	vmcs_write(ctrl_fld, ctrl_saved);
7193 	vmcs_write(fld, efer_saved);
7194 }
7195 
7196 static void test_efer(u32 fld, const char * fld_name, u32 ctrl_fld,
7197 		      u64 ctrl_bit1, u64 ctrl_bit2)
7198 {
7199 	u64 efer_saved = vmcs_read(fld);
7200 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7201 	u64 efer_reserved_bits =  ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
7202 	u64 i;
7203 	u64 efer;
7204 
7205 	if (cpu_has_efer_nx())
7206 		efer_reserved_bits &= ~EFER_NX;
7207 
7208 	if (!ctrl_bit1) {
7209 		printf("\"Load-IA32-EFER\" exit control not supported\n");
7210 		goto test_entry_exit_mode;
7211 	}
7212 
7213 	report_prefix_pushf("%s %lx", fld_name, efer_saved);
7214 	test_efer_vmlaunch(fld, true);
7215 	report_prefix_pop();
7216 
7217 	/*
7218 	 * Check reserved bits
7219 	 */
7220 	vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit1);
7221 	for (i = 0; i < 64; i++) {
7222 		if ((1ull << i) & efer_reserved_bits) {
7223 			efer = efer_saved | (1ull << i);
7224 			vmcs_write(fld, efer);
7225 			report_prefix_pushf("%s %lx", fld_name, efer);
7226 			test_efer_vmlaunch(fld, true);
7227 			report_prefix_pop();
7228 		}
7229 	}
7230 
7231 	vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit1);
7232 	for (i = 0; i < 64; i++) {
7233 		if ((1ull << i) & efer_reserved_bits) {
7234 			efer = efer_saved | (1ull << i);
7235 			vmcs_write(fld, efer);
7236 			report_prefix_pushf("%s %lx", fld_name, efer);
7237 			test_efer_vmlaunch(fld, false);
7238 			report_prefix_pop();
7239 		}
7240 	}
7241 
7242 	vmcs_write(ctrl_fld, ctrl_saved);
7243 	vmcs_write(fld, efer_saved);
7244 
7245 	/*
7246 	 * Check LMA and LME bits
7247 	 */
7248 	test_efer_bit(fld, fld_name,
7249 		      ctrl_fld, ctrl_bit1,
7250 		      EFER_LMA,
7251 		      "EFER_LMA");
7252 	test_efer_bit(fld, fld_name,
7253 		      ctrl_fld, ctrl_bit1,
7254 		      EFER_LME,
7255 		      "EFER_LME");
7256 
7257 test_entry_exit_mode:
7258 	test_efer_bit(fld, fld_name,
7259 		      ctrl_fld, ctrl_bit2,
7260 		      EFER_LMA,
7261 		      "EFER_LMA");
7262 	test_efer_bit(fld, fld_name,
7263 		      ctrl_fld, ctrl_bit2,
7264 		      EFER_LME,
7265 		      "EFER_LME");
7266 }
7267 
7268 /*
7269  * If the 'load IA32_EFER' VM-exit control is 1, bits reserved in the
7270  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7271  * the values of the LMA and LME bits in the field must each be that of
7272  * the 'host address-space size' VM-exit control.
7273  *
7274  *  [Intel SDM]
7275  */
7276 static void test_host_efer(void)
7277 {
7278 	test_efer(HOST_EFER, "HOST_EFER", EXI_CONTROLS,
7279 		  ctrl_exit_rev.clr & EXI_LOAD_EFER,
7280 		  EXI_HOST_64);
7281 }
7282 
7283 /*
7284  * If the 'load IA32_EFER' VM-enter control is 1, bits reserved in the
7285  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7286  * the values of the LMA and LME bits in the field must each be that of
7287  * the 'IA32e-mode guest' VM-exit control.
7288  */
7289 static void test_guest_efer(void)
7290 {
7291 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER)) {
7292 		printf("\"Load-IA32-EFER\" entry control not supported\n");
7293 		return;
7294 	}
7295 
7296 	vmcs_write(GUEST_EFER, rdmsr(MSR_EFER));
7297 	test_efer(GUEST_EFER, "GUEST_EFER", ENT_CONTROLS,
7298 		  ctrl_enter_rev.clr & ENT_LOAD_EFER,
7299 		  ENT_GUEST_64);
7300 }
7301 
7302 /*
7303  * PAT values higher than 8 are uninteresting since they're likely lumped
7304  * in with "8". We only test values above 8 one bit at a time,
7305  * in order to reduce the number of VM-Entries and keep the runtime reasonable.
7306  */
7307 #define	PAT_VAL_LIMIT	8
7308 
7309 static void test_pat(u32 field, const char * field_name, u32 ctrl_field,
7310 		     u64 ctrl_bit)
7311 {
7312 	u32 ctrl_saved = vmcs_read(ctrl_field);
7313 	u64 pat_saved = vmcs_read(field);
7314 	u64 i, val;
7315 	u32 j;
7316 	int error;
7317 
7318 	vmcs_clear_bits(ctrl_field, ctrl_bit);
7319 
7320 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7321 		/* Test PAT0..PAT7 fields */
7322 		for (j = 0; j < (i ? 8 : 1); j++) {
7323 			val = i << j * 8;
7324 			vmcs_write(field, val);
7325 			if (field == HOST_PAT) {
7326 				report_prefix_pushf("%s %lx", field_name, val);
7327 				test_vmx_vmlaunch(0);
7328 				report_prefix_pop();
7329 
7330 			} else {	// GUEST_PAT
7331 				test_guest_state("ENT_LOAD_PAT enabled", false,
7332 						 val, "GUEST_PAT");
7333 			}
7334 		}
7335 	}
7336 
7337 	vmcs_set_bits(ctrl_field, ctrl_bit);
7338 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7339 		/* Test PAT0..PAT7 fields */
7340 		for (j = 0; j < (i ? 8 : 1); j++) {
7341 			val = i << j * 8;
7342 			vmcs_write(field, val);
7343 
7344 			if (field == HOST_PAT) {
7345 				report_prefix_pushf("%s %lx", field_name, val);
7346 				if (i == 0x2 || i == 0x3 || i >= 0x8)
7347 					error =
7348 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
7349 				else
7350 					error = 0;
7351 
7352 				test_vmx_vmlaunch(error);
7353 				report_prefix_pop();
7354 
7355 			} else {	// GUEST_PAT
7356 				error = (i == 0x2 || i == 0x3 || i >= 0x8);
7357 				test_guest_state("ENT_LOAD_PAT enabled", !!error,
7358 						 val, "GUEST_PAT");
7359 			}
7360 
7361 		}
7362 	}
7363 
7364 	vmcs_write(ctrl_field, ctrl_saved);
7365 	vmcs_write(field, pat_saved);
7366 }
7367 
7368 /*
7369  *  If the "load IA32_PAT" VM-exit control is 1, the value of the field
7370  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7371  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7372  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7373  *  6 (WB), or 7 (UC-).
7374  *
7375  *  [Intel SDM]
7376  */
7377 static void test_load_host_pat(void)
7378 {
7379 	/*
7380 	 * "load IA32_PAT" VM-exit control
7381 	 */
7382 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) {
7383 		printf("\"Load-IA32-PAT\" exit control not supported\n");
7384 		return;
7385 	}
7386 
7387 	test_pat(HOST_PAT, "HOST_PAT", EXI_CONTROLS, EXI_LOAD_PAT);
7388 }
7389 
7390 union cpuidA_eax {
7391 	struct {
7392 		unsigned int version_id:8;
7393 		unsigned int num_counters_gp:8;
7394 		unsigned int bit_width:8;
7395 		unsigned int mask_length:8;
7396 	} split;
7397 	unsigned int full;
7398 };
7399 
7400 union cpuidA_edx {
7401 	struct {
7402 		unsigned int num_counters_fixed:5;
7403 		unsigned int bit_width_fixed:8;
7404 		unsigned int reserved:9;
7405 	} split;
7406 	unsigned int full;
7407 };
7408 
7409 static bool valid_pgc(u64 val)
7410 {
7411 	struct cpuid id;
7412 	union cpuidA_eax eax;
7413 	union cpuidA_edx edx;
7414 	u64 mask;
7415 
7416 	id = cpuid(0xA);
7417 	eax.full = id.a;
7418 	edx.full = id.d;
7419 	mask = ~(((1ull << eax.split.num_counters_gp) - 1) |
7420 		 (((1ull << edx.split.num_counters_fixed) - 1) << 32));
7421 
7422 	return !(val & mask);
7423 }
7424 
7425 static void test_pgc_vmlaunch(u32 xerror, u32 xreason, bool xfail, bool host)
7426 {
7427 	u32 inst_err;
7428 	u64 obs;
7429 	bool success;
7430 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7431 
7432 	if (host) {
7433 		success = vmlaunch_succeeds();
7434 		obs = rdmsr(data->msr);
7435 		if (!success) {
7436 			inst_err = vmcs_read(VMX_INST_ERROR);
7437 			report(xerror == inst_err, "vmlaunch failed, "
7438 			       "VMX Inst Error is %d (expected %d)",
7439 			       inst_err, xerror);
7440 		} else {
7441 			report(!data->enabled || data->exp == obs,
7442 			       "Host state is 0x%lx (expected 0x%lx)",
7443 			       obs, data->exp);
7444 			report(success != xfail, "vmlaunch succeeded");
7445 		}
7446 	} else {
7447 		test_guest_state("load GUEST_PERF_GLOBAL_CTRL", xfail,
7448 				 GUEST_PERF_GLOBAL_CTRL,
7449 				 "GUEST_PERF_GLOBAL_CTRL");
7450 	}
7451 }
7452 
7453 /*
7454  * test_load_perf_global_ctrl is a generic function for testing the
7455  * "load IA32_PERF_GLOBAL_CTRL" VM-{Entry,Exit} controls. This test function
7456  * tests the provided ctrl_val when disabled and enabled.
7457  *
7458  * @nr: VMCS field number corresponding to the host/guest state field
7459  * @name: Name of the above VMCS field for printing in test report
7460  * @ctrl_nr: VMCS field number corresponding to the VM-{Entry,Exit} control
7461  * @ctrl_val: Bit to set on the ctrl_field
7462  */
7463 static void test_perf_global_ctrl(u32 nr, const char *name, u32 ctrl_nr,
7464 				  const char *ctrl_name, u64 ctrl_val)
7465 {
7466 	u64 ctrl_saved = vmcs_read(ctrl_nr);
7467 	u64 pgc_saved = vmcs_read(nr);
7468 	u64 i, val;
7469 	bool host = nr == HOST_PERF_GLOBAL_CTRL;
7470 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7471 
7472 	data->msr = MSR_CORE_PERF_GLOBAL_CTRL;
7473 	msr_bmp_init();
7474 	vmcs_write(ctrl_nr, ctrl_saved & ~ctrl_val);
7475 	data->enabled = false;
7476 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=0 on %s",
7477 			    ctrl_name);
7478 
7479 	for (i = 0; i < 64; i++) {
7480 		val = 1ull << i;
7481 		vmcs_write(nr, val);
7482 		report_prefix_pushf("%s = 0x%lx", name, val);
7483 		test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7484 		report_prefix_pop();
7485 	}
7486 	report_prefix_pop();
7487 
7488 	vmcs_write(ctrl_nr, ctrl_saved | ctrl_val);
7489 	data->enabled = true;
7490 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=1 on %s",
7491 			    ctrl_name);
7492 	for (i = 0; i < 64; i++) {
7493 		val = 1ull << i;
7494 		data->exp = val;
7495 		vmcs_write(nr, val);
7496 		report_prefix_pushf("%s = 0x%lx", name, val);
7497 		if (valid_pgc(val)) {
7498 			test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7499 		} else {
7500 			if (host)
7501 				test_pgc_vmlaunch(
7502 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
7503 					0,
7504 					true,
7505 					host);
7506 			else
7507 				test_pgc_vmlaunch(
7508 					0,
7509 					VMX_ENTRY_FAILURE | VMX_FAIL_STATE,
7510 					true,
7511 					host);
7512 		}
7513 		report_prefix_pop();
7514 	}
7515 
7516 	data->enabled = false;
7517 	report_prefix_pop();
7518 	vmcs_write(ctrl_nr, ctrl_saved);
7519 	vmcs_write(nr, pgc_saved);
7520 }
7521 
7522 static void test_load_host_perf_global_ctrl(void)
7523 {
7524 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PERF)) {
7525 		printf("\"load IA32_PERF_GLOBAL_CTRL\" exit control not supported\n");
7526 		return;
7527 	}
7528 
7529 	test_perf_global_ctrl(HOST_PERF_GLOBAL_CTRL, "HOST_PERF_GLOBAL_CTRL",
7530 				   EXI_CONTROLS, "EXI_CONTROLS", EXI_LOAD_PERF);
7531 }
7532 
7533 
7534 static void test_load_guest_perf_global_ctrl(void)
7535 {
7536 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PERF)) {
7537 		printf("\"load IA32_PERF_GLOBAL_CTRL\" entry control not supported\n");
7538 		return;
7539 	}
7540 
7541 	test_perf_global_ctrl(GUEST_PERF_GLOBAL_CTRL, "GUEST_PERF_GLOBAL_CTRL",
7542 				   ENT_CONTROLS, "ENT_CONTROLS", ENT_LOAD_PERF);
7543 }
7544 
7545 
7546 /*
7547  * test_vmcs_field - test a value for the given VMCS field
7548  * @field: VMCS field
7549  * @field_name: string name of VMCS field
7550  * @bit_start: starting bit
7551  * @bit_end: ending bit
7552  * @val: value that the bit range must or must not contain
7553  * @valid_val: whether value given in 'val' must be valid or not
7554  * @error: expected VMCS error when vmentry fails for an invalid value
7555  */
7556 static void test_vmcs_field(u64 field, const char *field_name, u32 bit_start,
7557 			    u32 bit_end, u64 val, bool valid_val, u32 error)
7558 {
7559 	u64 field_saved = vmcs_read(field);
7560 	u32 i;
7561 	u64 tmp;
7562 	u32 bit_on;
7563 	u64 mask = ~0ull;
7564 
7565 	mask = (mask >> bit_end) << bit_end;
7566 	mask = mask | ((1 << bit_start) - 1);
7567 	tmp = (field_saved & mask) | (val << bit_start);
7568 
7569 	vmcs_write(field, tmp);
7570 	report_prefix_pushf("%s %lx", field_name, tmp);
7571 	if (valid_val)
7572 		test_vmx_vmlaunch(0);
7573 	else
7574 		test_vmx_vmlaunch(error);
7575 	report_prefix_pop();
7576 
7577 	for (i = bit_start; i <= bit_end; i = i + 2) {
7578 		bit_on = ((1ull < i) & (val << bit_start)) ? 0 : 1;
7579 		if (bit_on)
7580 			tmp = field_saved | (1ull << i);
7581 		else
7582 			tmp = field_saved & ~(1ull << i);
7583 		vmcs_write(field, tmp);
7584 		report_prefix_pushf("%s %lx", field_name, tmp);
7585 		if (valid_val)
7586 			test_vmx_vmlaunch(error);
7587 		else
7588 			test_vmx_vmlaunch(0);
7589 		report_prefix_pop();
7590 	}
7591 
7592 	vmcs_write(field, field_saved);
7593 }
7594 
7595 static void test_canonical(u64 field, const char * field_name, bool host)
7596 {
7597 	u64 addr_saved = vmcs_read(field);
7598 
7599 	/*
7600 	 * Use the existing value if possible.  Writing a random canonical
7601 	 * value is not an option as doing so would corrupt the field being
7602 	 * tested and likely hose the test.
7603 	 */
7604 	if (is_canonical(addr_saved)) {
7605 		if (host) {
7606 			report_prefix_pushf("%s %lx", field_name, addr_saved);
7607 			test_vmx_vmlaunch(0);
7608 			report_prefix_pop();
7609 		} else {
7610 			test_guest_state("Test canonical address", false,
7611 					 addr_saved, field_name);
7612 		}
7613 	}
7614 
7615 	vmcs_write(field, NONCANONICAL);
7616 
7617 	if (host) {
7618 		report_prefix_pushf("%s %llx", field_name, NONCANONICAL);
7619 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7620 		report_prefix_pop();
7621 	} else {
7622 		test_guest_state("Test non-canonical address", true,
7623 				 NONCANONICAL, field_name);
7624 	}
7625 
7626 	vmcs_write(field, addr_saved);
7627 }
7628 
7629 #define TEST_RPL_TI_FLAGS(reg, name)				\
7630 	test_vmcs_field(reg, name, 0, 2, 0x0, true,		\
7631 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7632 
7633 #define TEST_CS_TR_FLAGS(reg, name)				\
7634 	test_vmcs_field(reg, name, 3, 15, 0x0000, false,	\
7635 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7636 
7637 /*
7638  * 1. In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the
7639  *    RPL (bits 1:0) and the TI flag (bit 2) must be 0.
7640  * 2. The selector fields for CS and TR cannot be 0000H.
7641  * 3. The selector field for SS cannot be 0000H if the "host address-space
7642  *    size" VM-exit control is 0.
7643  * 4. On processors that support Intel 64 architecture, the base-address
7644  *    fields for FS, GS and TR must contain canonical addresses.
7645  */
7646 static void test_host_segment_regs(void)
7647 {
7648 	u16 selector_saved;
7649 
7650 	/*
7651 	 * Test RPL and TI flags
7652 	 */
7653 	TEST_RPL_TI_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7654 	TEST_RPL_TI_FLAGS(HOST_SEL_SS, "HOST_SEL_SS");
7655 	TEST_RPL_TI_FLAGS(HOST_SEL_DS, "HOST_SEL_DS");
7656 	TEST_RPL_TI_FLAGS(HOST_SEL_ES, "HOST_SEL_ES");
7657 	TEST_RPL_TI_FLAGS(HOST_SEL_FS, "HOST_SEL_FS");
7658 	TEST_RPL_TI_FLAGS(HOST_SEL_GS, "HOST_SEL_GS");
7659 	TEST_RPL_TI_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7660 
7661 	/*
7662 	 * Test that CS and TR fields can not be 0x0000
7663 	 */
7664 	TEST_CS_TR_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7665 	TEST_CS_TR_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7666 
7667 	/*
7668 	 * SS field can not be 0x0000 if "host address-space size" VM-exit
7669 	 * control is 0
7670 	 */
7671 	selector_saved = vmcs_read(HOST_SEL_SS);
7672 	vmcs_write(HOST_SEL_SS, 0);
7673 	report_prefix_pushf("HOST_SEL_SS 0");
7674 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7675 		test_vmx_vmlaunch(0);
7676 	} else {
7677 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7678 	}
7679 	report_prefix_pop();
7680 
7681 	vmcs_write(HOST_SEL_SS, selector_saved);
7682 
7683 	/*
7684 	 * Base address for FS, GS and TR must be canonical
7685 	 */
7686 	test_canonical(HOST_BASE_FS, "HOST_BASE_FS", true);
7687 	test_canonical(HOST_BASE_GS, "HOST_BASE_GS", true);
7688 	test_canonical(HOST_BASE_TR, "HOST_BASE_TR", true);
7689 }
7690 
7691 /*
7692  *  On processors that support Intel 64 architecture, the base-address
7693  *  fields for GDTR and IDTR must contain canonical addresses.
7694  */
7695 static void test_host_desc_tables(void)
7696 {
7697 	test_canonical(HOST_BASE_GDTR, "HOST_BASE_GDTR", true);
7698 	test_canonical(HOST_BASE_IDTR, "HOST_BASE_IDTR", true);
7699 }
7700 
7701 /*
7702  * If the "host address-space size" VM-exit control is 0, the following must
7703  * hold:
7704  *    - The "IA-32e mode guest" VM-entry control is 0.
7705  *    - Bit 17 of the CR4 field (corresponding to CR4.PCIDE) is 0.
7706  *    - Bits 63:32 in the RIP field are 0.
7707  *
7708  * If the "host address-space size" VM-exit control is 1, the following must
7709  * hold:
7710  *    - Bit 5 of the CR4 field (corresponding to CR4.PAE) is 1.
7711  *    - The RIP field contains a canonical address.
7712  *
7713  */
7714 static void test_host_addr_size(void)
7715 {
7716 	u64 cr4_saved = vmcs_read(HOST_CR4);
7717 	u64 rip_saved = vmcs_read(HOST_RIP);
7718 	u64 entry_ctrl_saved = vmcs_read(ENT_CONTROLS);
7719 	int i;
7720 	u64 tmp;
7721 
7722 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7723 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7724 		report_prefix_pushf("\"IA-32e mode guest\" enabled");
7725 		test_vmx_vmlaunch(0);
7726 		report_prefix_pop();
7727 
7728 		vmcs_write(HOST_CR4, cr4_saved | X86_CR4_PCIDE);
7729 		report_prefix_pushf("\"CR4.PCIDE\" set");
7730 		test_vmx_vmlaunch(0);
7731 		report_prefix_pop();
7732 
7733 		for (i = 32; i <= 63; i = i + 4) {
7734 			tmp = rip_saved | 1ull << i;
7735 			vmcs_write(HOST_RIP, tmp);
7736 			report_prefix_pushf("HOST_RIP %lx", tmp);
7737 			test_vmx_vmlaunch(0);
7738 			report_prefix_pop();
7739 		}
7740 
7741 		if (cr4_saved & X86_CR4_PAE) {
7742 			vmcs_write(HOST_CR4, cr4_saved  & ~X86_CR4_PAE);
7743 			report_prefix_pushf("\"CR4.PAE\" unset");
7744 			test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7745 		} else {
7746 			report_prefix_pushf("\"CR4.PAE\" set");
7747 			test_vmx_vmlaunch(0);
7748 		}
7749 		report_prefix_pop();
7750 
7751 		vmcs_write(HOST_RIP, NONCANONICAL);
7752 		report_prefix_pushf("HOST_RIP %llx", NONCANONICAL);
7753 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7754 		report_prefix_pop();
7755 
7756 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7757 		vmcs_write(HOST_RIP, rip_saved);
7758 		vmcs_write(HOST_CR4, cr4_saved);
7759 
7760 		/* Restore host's active RIP and CR4 values. */
7761 		report_prefix_pushf("restore host state");
7762 		test_vmx_vmlaunch(0);
7763 		report_prefix_pop();
7764 	}
7765 }
7766 
7767 /*
7768  * Check that the virtual CPU checks the VMX Host State Area as
7769  * documented in the Intel SDM.
7770  */
7771 static void vmx_host_state_area_test(void)
7772 {
7773 	/*
7774 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
7775 	 * fail due to invalid guest state, should we make it that
7776 	 * far.
7777 	 */
7778 	vmcs_write(GUEST_RFLAGS, 0);
7779 
7780 	test_host_ctl_regs();
7781 
7782 	test_canonical(HOST_SYSENTER_ESP, "HOST_SYSENTER_ESP", true);
7783 	test_canonical(HOST_SYSENTER_EIP, "HOST_SYSENTER_EIP", true);
7784 
7785 	test_host_efer();
7786 	test_load_host_pat();
7787 	test_host_segment_regs();
7788 	test_host_desc_tables();
7789 	test_host_addr_size();
7790 	test_load_host_perf_global_ctrl();
7791 }
7792 
7793 /*
7794  * If the "load debug controls" VM-entry control is 1, bits 63:32 in
7795  * the DR7 field must be 0.
7796  *
7797  * [Intel SDM]
7798  */
7799 static void test_guest_dr7(void)
7800 {
7801 	u32 ent_saved = vmcs_read(ENT_CONTROLS);
7802 	u64 dr7_saved = vmcs_read(GUEST_DR7);
7803 	u64 val;
7804 	int i;
7805 
7806 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS) {
7807 		vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7808 		for (i = 0; i < 64; i++) {
7809 			val = 1ull << i;
7810 			vmcs_write(GUEST_DR7, val);
7811 			test_guest_state("ENT_LOAD_DBGCTLS disabled", false,
7812 					 val, "GUEST_DR7");
7813 		}
7814 	}
7815 	if (ctrl_enter_rev.clr & ENT_LOAD_DBGCTLS) {
7816 		vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7817 		for (i = 0; i < 64; i++) {
7818 			val = 1ull << i;
7819 			vmcs_write(GUEST_DR7, val);
7820 			test_guest_state("ENT_LOAD_DBGCTLS enabled", i >= 32,
7821 					 val, "GUEST_DR7");
7822 		}
7823 	}
7824 	vmcs_write(GUEST_DR7, dr7_saved);
7825 	vmcs_write(ENT_CONTROLS, ent_saved);
7826 }
7827 
7828 /*
7829  *  If the "load IA32_PAT" VM-entry control is 1, the value of the field
7830  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7831  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7832  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7833  *  6 (WB), or 7 (UC-).
7834  *
7835  *  [Intel SDM]
7836  */
7837 static void test_load_guest_pat(void)
7838 {
7839 	/*
7840 	 * "load IA32_PAT" VM-entry control
7841 	 */
7842 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
7843 		printf("\"Load-IA32-PAT\" entry control not supported\n");
7844 		return;
7845 	}
7846 
7847 	test_pat(GUEST_PAT, "GUEST_PAT", ENT_CONTROLS, ENT_LOAD_PAT);
7848 }
7849 
7850 #define MSR_IA32_BNDCFGS_RSVD_MASK	0x00000ffc
7851 
7852 /*
7853  * If the "load IA32_BNDCFGS" VM-entry control is 1, the following
7854  * checks are performed on the field for the IA32_BNDCFGS MSR:
7855  *
7856  *   - Bits reserved in the IA32_BNDCFGS MSR must be 0.
7857  *   - The linear address in bits 63:12 must be canonical.
7858  *
7859  *  [Intel SDM]
7860  */
7861 static void test_load_guest_bndcfgs(void)
7862 {
7863 	u64 bndcfgs_saved = vmcs_read(GUEST_BNDCFGS);
7864 	u64 bndcfgs;
7865 
7866 	if (!(ctrl_enter_rev.clr & ENT_LOAD_BNDCFGS)) {
7867 		printf("\"Load-IA32-BNDCFGS\" entry control not supported\n");
7868 		return;
7869 	}
7870 
7871 	vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7872 
7873 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7874 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7875 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7876 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7877 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7878 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7879 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7880 
7881 	vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7882 
7883 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7884 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7885 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7886 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7887 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7888 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7889 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7890 
7891 	vmcs_write(GUEST_BNDCFGS, bndcfgs_saved);
7892 }
7893 
7894 #define	GUEST_SEG_UNUSABLE_MASK	(1u << 16)
7895 #define	GUEST_SEG_SEL_TI_MASK	(1u << 2)
7896 
7897 
7898 #define	TEST_SEGMENT_SEL(test, xfail, sel, val)				\
7899 do {									\
7900 	vmcs_write(sel, val);						\
7901 	test_guest_state(test " segment", xfail, val, xstr(sel));	\
7902 } while (0)
7903 
7904 #define	TEST_INVALID_SEG_SEL(sel, val) \
7905 	TEST_SEGMENT_SEL("Invalid: " xstr(val), true, sel, val);
7906 
7907 #define	TEST_VALID_SEG_SEL(sel, val) \
7908 	TEST_SEGMENT_SEL("Valid: " xstr(val), false, sel, val);
7909 
7910 /*
7911  * The following checks are done on the Selector field of the Guest Segment
7912  * Registers:
7913  *    - TR. The TI flag (bit 2) must be 0.
7914  *    - LDTR. If LDTR is usable, the TI flag (bit 2) must be 0.
7915  *    - SS. If the guest will not be virtual-8086 and the "unrestricted
7916  *	guest" VM-execution control is 0, the RPL (bits 1:0) must equal
7917  *	the RPL of the selector field for CS.
7918  *
7919  *  [Intel SDM]
7920  */
7921 static void test_guest_segment_sel_fields(void)
7922 {
7923 	u16 sel_saved;
7924 	u32 ar_saved;
7925 	u32 cpu_ctrl0_saved;
7926 	u32 cpu_ctrl1_saved;
7927 	u16 cs_rpl_bits;
7928 
7929 	/*
7930 	 * Test for GUEST_SEL_TR
7931 	 */
7932 	sel_saved = vmcs_read(GUEST_SEL_TR);
7933 	TEST_INVALID_SEG_SEL(GUEST_SEL_TR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7934 	vmcs_write(GUEST_SEL_TR, sel_saved);
7935 
7936 	/*
7937 	 * Test for GUEST_SEL_LDTR
7938 	 */
7939 	sel_saved = vmcs_read(GUEST_SEL_LDTR);
7940 	ar_saved = vmcs_read(GUEST_AR_LDTR);
7941 	/* LDTR is set unusable */
7942 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7943 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7944 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7945 	/* LDTR is set usable */
7946 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7947 	TEST_INVALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7948 
7949 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7950 
7951 	vmcs_write(GUEST_AR_LDTR, ar_saved);
7952 	vmcs_write(GUEST_SEL_LDTR, sel_saved);
7953 
7954 	/*
7955 	 * Test for GUEST_SEL_SS
7956 	 */
7957 	cpu_ctrl0_saved = vmcs_read(CPU_EXEC_CTRL0);
7958 	cpu_ctrl1_saved = vmcs_read(CPU_EXEC_CTRL1);
7959 	ar_saved = vmcs_read(GUEST_AR_SS);
7960 	/* Turn off "unrestricted guest" vm-execution control */
7961 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved & ~CPU_URG);
7962 	cs_rpl_bits = vmcs_read(GUEST_SEL_CS) & 0x3;
7963 	sel_saved = vmcs_read(GUEST_SEL_SS);
7964 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7965 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7966 	/* Make SS usable if it's unusable or vice-versa */
7967 	if (ar_saved & GUEST_SEG_UNUSABLE_MASK)
7968 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7969 	else
7970 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7971 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7972 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7973 
7974 	/* Need a valid EPTP as the passing case fully enters the guest. */
7975 	if (enable_unrestricted_guest(true))
7976 		goto skip_ss_tests;
7977 
7978 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7979 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7980 
7981 	/* Make SS usable if it's unusable or vice-versa */
7982 	if (vmcs_read(GUEST_AR_SS) & GUEST_SEG_UNUSABLE_MASK)
7983 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7984 	else
7985 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7986 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7987 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7988 skip_ss_tests:
7989 
7990 	vmcs_write(GUEST_AR_SS, ar_saved);
7991 	vmcs_write(GUEST_SEL_SS, sel_saved);
7992 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrl0_saved);
7993 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved);
7994 }
7995 
7996 #define	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(xfail, seg_base)			\
7997 do {										\
7998 	addr_saved = vmcs_read(seg_base);					\
7999 	for (i = 32; i < 63; i = i + 4) {					\
8000 		addr = addr_saved | 1ull << i;					\
8001 		vmcs_write(seg_base, addr);					\
8002 		test_guest_state("seg.BASE[63:32] != 0, usable = " xstr(xfail),	\
8003 				 xfail, addr, xstr(seg_base));			\
8004 	}									\
8005 	vmcs_write(seg_base, addr_saved);					\
8006 } while (0)
8007 
8008 #define	TEST_SEGMENT_BASE_ADDR_CANONICAL(xfail, seg_base)		  \
8009 do {									  \
8010 	addr_saved = vmcs_read(seg_base);				  \
8011 	vmcs_write(seg_base, NONCANONICAL);				  \
8012 	test_guest_state("seg.BASE non-canonical, usable = " xstr(xfail), \
8013 			 xfail, NONCANONICAL, xstr(seg_base));		  \
8014 	vmcs_write(seg_base, addr_saved);				  \
8015 } while (0)
8016 
8017 /*
8018  * The following checks are done on the Base Address field of the Guest
8019  * Segment Registers on processors that support Intel 64 architecture:
8020  *    - TR, FS, GS : The address must be canonical.
8021  *    - LDTR : If LDTR is usable, the address must be canonical.
8022  *    - CS : Bits 63:32 of the address must be zero.
8023  *    - SS, DS, ES : If the register is usable, bits 63:32 of the address
8024  *	must be zero.
8025  *
8026  *  [Intel SDM]
8027  */
8028 static void test_guest_segment_base_addr_fields(void)
8029 {
8030 	u64 addr_saved;
8031 	u64 addr;
8032 	u32 ar_saved;
8033 	int i;
8034 
8035 	/*
8036 	 * The address of TR, FS, GS and LDTR must be canonical.
8037 	 */
8038 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_TR);
8039 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_FS);
8040 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_GS);
8041 	ar_saved = vmcs_read(GUEST_AR_LDTR);
8042 	/* Make LDTR unusable */
8043 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8044 	TEST_SEGMENT_BASE_ADDR_CANONICAL(false, GUEST_BASE_LDTR);
8045 	/* Make LDTR usable */
8046 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8047 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_LDTR);
8048 
8049 	vmcs_write(GUEST_AR_LDTR, ar_saved);
8050 
8051 	/*
8052 	 * Bits 63:32 in CS, SS, DS and ES base address must be zero
8053 	 */
8054 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_CS);
8055 	ar_saved = vmcs_read(GUEST_AR_SS);
8056 	/* Make SS unusable */
8057 	vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8058 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_SS);
8059 	/* Make SS usable */
8060 	vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8061 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_SS);
8062 	vmcs_write(GUEST_AR_SS, ar_saved);
8063 
8064 	ar_saved = vmcs_read(GUEST_AR_DS);
8065 	/* Make DS unusable */
8066 	vmcs_write(GUEST_AR_DS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8067 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_DS);
8068 	/* Make DS usable */
8069 	vmcs_write(GUEST_AR_DS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8070 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_DS);
8071 	vmcs_write(GUEST_AR_DS, ar_saved);
8072 
8073 	ar_saved = vmcs_read(GUEST_AR_ES);
8074 	/* Make ES unusable */
8075 	vmcs_write(GUEST_AR_ES, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8076 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_ES);
8077 	/* Make ES usable */
8078 	vmcs_write(GUEST_AR_ES, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8079 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_ES);
8080 	vmcs_write(GUEST_AR_ES, ar_saved);
8081 }
8082 
8083 /*
8084  * Check that the virtual CPU checks the VMX Guest State Area as
8085  * documented in the Intel SDM.
8086  */
8087 static void vmx_guest_state_area_test(void)
8088 {
8089 	vmx_set_test_stage(1);
8090 	test_set_guest(guest_state_test_main);
8091 
8092 	/*
8093 	 * The IA32_SYSENTER_ESP field and the IA32_SYSENTER_EIP field
8094 	 * must each contain a canonical address.
8095 	 */
8096 	test_canonical(GUEST_SYSENTER_ESP, "GUEST_SYSENTER_ESP", false);
8097 	test_canonical(GUEST_SYSENTER_EIP, "GUEST_SYSENTER_EIP", false);
8098 
8099 	test_guest_dr7();
8100 	test_load_guest_pat();
8101 	test_guest_efer();
8102 	test_load_guest_perf_global_ctrl();
8103 	test_load_guest_bndcfgs();
8104 
8105 	test_guest_segment_sel_fields();
8106 	test_guest_segment_base_addr_fields();
8107 
8108 	test_canonical(GUEST_BASE_GDTR, "GUEST_BASE_GDTR", false);
8109 	test_canonical(GUEST_BASE_IDTR, "GUEST_BASE_IDTR", false);
8110 
8111 	u32 guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_GDTR);
8112 	int i;
8113 	for (i = 16; i <= 31; i++) {
8114 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8115 		vmcs_write(GUEST_LIMIT_GDTR, tmp);
8116 		test_guest_state("GDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_GDTR");
8117 	}
8118 	vmcs_write(GUEST_LIMIT_GDTR, guest_desc_limit_saved);
8119 
8120 	guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_IDTR);
8121 	for (i = 16; i <= 31; i++) {
8122 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8123 		vmcs_write(GUEST_LIMIT_IDTR, tmp);
8124 		test_guest_state("IDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_IDTR");
8125 	}
8126 	vmcs_write(GUEST_LIMIT_IDTR, guest_desc_limit_saved);
8127 
8128 	/*
8129 	 * Let the guest finish execution
8130 	 */
8131 	vmx_set_test_stage(2);
8132 	enter_guest();
8133 }
8134 
8135 extern void unrestricted_guest_main(void);
8136 asm (".code32\n"
8137 	"unrestricted_guest_main:\n"
8138 	"vmcall\n"
8139 	"nop\n"
8140 	"mov $1, %edi\n"
8141 	"call hypercall\n"
8142 	".code64\n");
8143 
8144 static void setup_unrestricted_guest(void)
8145 {
8146 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) & ~(X86_CR0_PG));
8147 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) & ~ENT_GUEST_64);
8148 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) & ~EFER_LMA);
8149 	vmcs_write(GUEST_RIP, virt_to_phys(unrestricted_guest_main));
8150 }
8151 
8152 static void unsetup_unrestricted_guest(void)
8153 {
8154 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) | X86_CR0_PG);
8155 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_GUEST_64);
8156 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) | EFER_LMA);
8157 	vmcs_write(GUEST_RIP, (u64) phys_to_virt(vmcs_read(GUEST_RIP)));
8158 	vmcs_write(GUEST_RSP, (u64) phys_to_virt(vmcs_read(GUEST_RSP)));
8159 }
8160 
8161 /*
8162  * If "unrestricted guest" secondary VM-execution control is set, guests
8163  * can run in unpaged protected mode.
8164  */
8165 static void vmentry_unrestricted_guest_test(void)
8166 {
8167 	if (enable_unrestricted_guest(true)) {
8168 		report_skip("Unrestricted guest not supported");
8169 		return;
8170 	}
8171 
8172 	test_set_guest(unrestricted_guest_main);
8173 	setup_unrestricted_guest();
8174 	test_guest_state("Unrestricted guest test", false, CPU_URG, "CPU_URG");
8175 
8176 	/*
8177 	 * Let the guest finish execution as a regular guest
8178 	 */
8179 	unsetup_unrestricted_guest();
8180 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) & ~CPU_URG);
8181 	enter_guest();
8182 }
8183 
8184 static bool valid_vmcs_for_vmentry(void)
8185 {
8186 	struct vmcs *current_vmcs = NULL;
8187 
8188 	if (vmcs_save(&current_vmcs))
8189 		return false;
8190 
8191 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
8192 }
8193 
8194 static void try_vmentry_in_movss_shadow(void)
8195 {
8196 	u32 vm_inst_err;
8197 	u32 flags;
8198 	bool early_failure = false;
8199 	u32 expected_flags = X86_EFLAGS_FIXED;
8200 	bool valid_vmcs = valid_vmcs_for_vmentry();
8201 
8202 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
8203 
8204 	/*
8205 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
8206 	 * unsupported VMCS component").
8207 	 */
8208 	vmcs_write(~0u, 0);
8209 
8210 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
8211 			      "vmwrite %%rsp, %%rdx;"
8212 			      "mov 0f, %%rax;"
8213 			      "mov %[host_rip], %%edx;"
8214 			      "vmwrite %%rax, %%rdx;"
8215 			      "mov $-1, %%ah;"
8216 			      "sahf;"
8217 			      "mov %%ss, %%ax;"
8218 			      "mov %%ax, %%ss;"
8219 			      "vmlaunch;"
8220 			      "mov $1, %[early_failure];"
8221 			      "0: lahf;"
8222 			      "movzbl %%ah, %[flags]"
8223 			      : [early_failure] "+r" (early_failure),
8224 				[flags] "=&a" (flags)
8225 			      : [host_rsp] "i" (HOST_RSP),
8226 				[host_rip] "i" (HOST_RIP)
8227 			      : "rdx", "cc", "memory");
8228 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
8229 
8230 	report(early_failure, "Early VM-entry failure");
8231 	report(flags == expected_flags, "RFLAGS[8:0] is %x (actual %x)",
8232 	       expected_flags, flags);
8233 	if (valid_vmcs)
8234 		report(vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
8235 		       "VM-instruction error is %d (actual %d)",
8236 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
8237 }
8238 
8239 static void vmentry_movss_shadow_test(void)
8240 {
8241 	struct vmcs *orig_vmcs;
8242 
8243 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
8244 
8245 	/*
8246 	 * Set the launched flag on the current VMCS to verify the correct
8247 	 * error priority, below.
8248 	 */
8249 	test_set_guest(v2_null_test_guest);
8250 	enter_guest();
8251 
8252 	/*
8253 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
8254 	 * fail due to invalid guest state (if we make it that far).
8255 	 */
8256 	vmcs_write(GUEST_RFLAGS, 0);
8257 
8258 	/*
8259 	 * "VM entry with events blocked by MOV SS" takes precedence over
8260 	 * "VMLAUNCH with non-clear VMCS."
8261 	 */
8262 	report_prefix_push("valid current-VMCS");
8263 	try_vmentry_in_movss_shadow();
8264 	report_prefix_pop();
8265 
8266 	/*
8267 	 * VMfailInvalid takes precedence over "VM entry with events
8268 	 * blocked by MOV SS."
8269 	 */
8270 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
8271 	report_prefix_push("no current-VMCS");
8272 	try_vmentry_in_movss_shadow();
8273 	report_prefix_pop();
8274 
8275 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
8276 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8277 }
8278 
8279 static void vmx_ldtr_test_guest(void)
8280 {
8281 	u16 ldtr = sldt();
8282 
8283 	report(ldtr == NP_SEL, "Expected %x for L2 LDTR selector (got %x)",
8284 	       NP_SEL, ldtr);
8285 }
8286 
8287 /*
8288  * Ensure that the L1 LDTR is set to 0 on VM-exit.
8289  */
8290 static void vmx_ldtr_test(void)
8291 {
8292 	const u8 ldt_ar = 0x82; /* Present LDT */
8293 	u16 sel = FIRST_SPARE_SEL;
8294 
8295 	/* Set up a non-zero L1 LDTR prior to VM-entry. */
8296 	set_gdt_entry(sel, 0, 0, ldt_ar, 0);
8297 	lldt(sel);
8298 
8299 	test_set_guest(vmx_ldtr_test_guest);
8300 	/*
8301 	 * Set up a different LDTR for L2. The actual GDT contents are
8302 	 * irrelevant, since we stuff the hidden descriptor state
8303 	 * straight into the VMCS rather than reading it from the GDT.
8304 	 */
8305 	vmcs_write(GUEST_SEL_LDTR, NP_SEL);
8306 	vmcs_write(GUEST_AR_LDTR, ldt_ar);
8307 	enter_guest();
8308 
8309 	/*
8310 	 * VM-exit should clear LDTR (and make it unusable, but we
8311 	 * won't verify that here).
8312 	 */
8313 	sel = sldt();
8314 	report(!sel, "Expected 0 for L1 LDTR selector (got %x)", sel);
8315 }
8316 
8317 static void vmx_single_vmcall_guest(void)
8318 {
8319 	vmcall();
8320 }
8321 
8322 static void vmx_cr_load_test(void)
8323 {
8324 	unsigned long cr3, cr4, orig_cr3, orig_cr4;
8325 	u32 ctrls[2] = {0};
8326 	pgd_t *pml5;
8327 
8328 	orig_cr4 = read_cr4();
8329 	orig_cr3 = read_cr3();
8330 
8331 	if (!this_cpu_has(X86_FEATURE_PCID)) {
8332 		report_skip("PCID not detected");
8333 		return;
8334 	}
8335 	if (!this_cpu_has(X86_FEATURE_MCE)) {
8336 		report_skip("MCE not detected");
8337 		return;
8338 	}
8339 
8340 	TEST_ASSERT(!(orig_cr3 & X86_CR3_PCID_MASK));
8341 
8342 	/* Enable PCID for L1. */
8343 	cr4 = orig_cr4 | X86_CR4_PCIDE;
8344 	cr3 = orig_cr3 | 0x1;
8345 	TEST_ASSERT(!write_cr4_checking(cr4));
8346 	write_cr3(cr3);
8347 
8348 	test_set_guest(vmx_single_vmcall_guest);
8349 	vmcs_write(HOST_CR4, cr4);
8350 	vmcs_write(HOST_CR3, cr3);
8351 	enter_guest();
8352 
8353 	/*
8354 	 * No exception is expected.
8355 	 *
8356 	 * NB. KVM loads the last guest write to CR4 into CR4 read
8357 	 *     shadow. In order to trigger an exit to KVM, we can toggle a
8358 	 *     bit that is owned by KVM. We use CR4.MCE, which shall
8359 	 *     have no side effect because normally no guest MCE (e.g., as the
8360 	 *     result of bad memory) would happen during this test.
8361 	 */
8362 	TEST_ASSERT(!write_cr4_checking(cr4 ^ X86_CR4_MCE));
8363 
8364 	/* Cleanup L1 state. */
8365 	write_cr3(orig_cr3);
8366 	TEST_ASSERT(!write_cr4_checking(orig_cr4));
8367 
8368 	if (!this_cpu_has(X86_FEATURE_LA57))
8369 		goto done;
8370 
8371 	/*
8372 	 * Allocate a full page for PML5 to guarantee alignment, though only
8373 	 * the first entry needs to be filled (the test's virtual addresses
8374 	 * most definitely do not have any of bits 56:48 set).
8375 	 */
8376 	pml5 = alloc_page();
8377 	*pml5 = orig_cr3 | PT_PRESENT_MASK | PT_WRITABLE_MASK;
8378 
8379 	/*
8380 	 * Transition to/from 5-level paging in the host via VM-Exit.  CR4.LA57
8381 	 * can't be toggled while long is active via MOV CR4, but there are no
8382 	 * such restrictions on VM-Exit.
8383 	 */
8384 lol_5level:
8385 	vmcs_write(HOST_CR4, orig_cr4 | X86_CR4_LA57);
8386 	vmcs_write(HOST_CR3, virt_to_phys(pml5));
8387 	enter_guest();
8388 
8389 	/*
8390 	 * VMREAD with a memory operand to verify KVM detects the LA57 change,
8391 	 * e.g. uses the correct guest root level in gva_to_gpa().
8392 	 */
8393 	TEST_ASSERT(vmcs_readm(HOST_CR3) == virt_to_phys(pml5));
8394 	TEST_ASSERT(vmcs_readm(HOST_CR4) == (orig_cr4 | X86_CR4_LA57));
8395 
8396 	vmcs_write(HOST_CR4, orig_cr4);
8397 	vmcs_write(HOST_CR3, orig_cr3);
8398 	enter_guest();
8399 
8400 	TEST_ASSERT(vmcs_readm(HOST_CR3) == orig_cr3);
8401 	TEST_ASSERT(vmcs_readm(HOST_CR4) == orig_cr4);
8402 
8403 	/*
8404 	 * And now do the same LA57 shenanigans with EPT enabled.  KVM uses
8405 	 * two separate MMUs when L1 uses TDP, whereas the above shadow paging
8406 	 * version shares an MMU between L1 and L2.
8407 	 *
8408 	 * If the saved execution controls are non-zero then the EPT version
8409 	 * has already run.  In that case, restore the old controls.  If EPT
8410 	 * setup fails, e.g. EPT isn't supported, fall through and finish up.
8411 	 */
8412 	if (ctrls[0]) {
8413 		vmcs_write(CPU_EXEC_CTRL0, ctrls[0]);
8414 		vmcs_write(CPU_EXEC_CTRL1, ctrls[1]);
8415 	} else if (!setup_ept(false)) {
8416 		ctrls[0] = vmcs_read(CPU_EXEC_CTRL0);
8417 		ctrls[1]  = vmcs_read(CPU_EXEC_CTRL1);
8418 		goto lol_5level;
8419 	}
8420 
8421 	free_page(pml5);
8422 
8423 done:
8424 	skip_exit_vmcall();
8425 	enter_guest();
8426 }
8427 
8428 static void vmx_cr4_osxsave_test_guest(void)
8429 {
8430 	write_cr4(read_cr4() & ~X86_CR4_OSXSAVE);
8431 }
8432 
8433 /*
8434  * Ensure that kvm recalculates the L1 guest's CPUID.01H:ECX.OSXSAVE
8435  * after VM-exit from an L2 guest that sets CR4.OSXSAVE to a different
8436  * value than in L1.
8437  */
8438 static void vmx_cr4_osxsave_test(void)
8439 {
8440 	if (!this_cpu_has(X86_FEATURE_XSAVE)) {
8441 		report_skip("XSAVE not detected");
8442 		return;
8443 	}
8444 
8445 	if (!(read_cr4() & X86_CR4_OSXSAVE)) {
8446 		unsigned long cr4 = read_cr4() | X86_CR4_OSXSAVE;
8447 
8448 		write_cr4(cr4);
8449 		vmcs_write(GUEST_CR4, cr4);
8450 		vmcs_write(HOST_CR4, cr4);
8451 	}
8452 
8453 	TEST_ASSERT(cpuid_osxsave());
8454 
8455 	test_set_guest(vmx_cr4_osxsave_test_guest);
8456 	enter_guest();
8457 
8458 	TEST_ASSERT(cpuid_osxsave());
8459 }
8460 
8461 static void vmx_nm_test_guest(void)
8462 {
8463 	write_cr0(read_cr0() | X86_CR0_TS);
8464 	asm volatile("fnop");
8465 }
8466 
8467 static void check_nm_exit(const char *test)
8468 {
8469 	u32 reason = vmcs_read(EXI_REASON);
8470 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
8471 	const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8472 		NM_VECTOR;
8473 
8474 	report(reason == VMX_EXC_NMI && intr_info == expected, "%s", test);
8475 }
8476 
8477 /*
8478  * This test checks that:
8479  *
8480  * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then
8481  *     a subsequent #NM VM-exit is reflected to L1.
8482  *
8483  * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a
8484  *     subsequent #NM VM-exit is reflected to L1.
8485  */
8486 static void vmx_nm_test(void)
8487 {
8488 	unsigned long cr0 = read_cr0();
8489 
8490 	test_set_guest(vmx_nm_test_guest);
8491 
8492 	/*
8493 	 * L1 wants to intercept #NM exceptions encountered in L2.
8494 	 */
8495 	vmcs_write(EXC_BITMAP, 1 << NM_VECTOR);
8496 
8497 	/*
8498 	 * Launch L2 with CR0.TS clear, but don't claim host ownership of
8499 	 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop,
8500 	 * which will raise #NM. L0 should reflect the #NM VM-exit to L1.
8501 	 */
8502 	vmcs_write(CR0_MASK, 0);
8503 	vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS);
8504 	enter_guest();
8505 	check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1");
8506 
8507 	/*
8508 	 * Re-enter L2 at the fnop instruction, with CR0.TS clear but
8509 	 * CR0.EM set. The fnop will still raise #NM, and L0 should
8510 	 * reflect the #NM VM-exit to L1.
8511 	 */
8512 	vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM);
8513 	enter_guest();
8514 	check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1");
8515 
8516 	/*
8517 	 * Re-enter L2 at the fnop instruction, with both CR0.TS and
8518 	 * CR0.EM clear. There will be no #NM, and the L2 guest should
8519 	 * exit normally.
8520 	 */
8521 	vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM));
8522 	enter_guest();
8523 }
8524 
8525 bool vmx_pending_event_ipi_fired;
8526 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
8527 {
8528 	vmx_pending_event_ipi_fired = true;
8529 	eoi();
8530 }
8531 
8532 bool vmx_pending_event_guest_run;
8533 static void vmx_pending_event_guest(void)
8534 {
8535 	vmcall();
8536 	vmx_pending_event_guest_run = true;
8537 }
8538 
8539 static void vmx_pending_event_test_core(bool guest_hlt)
8540 {
8541 	int ipi_vector = 0xf1;
8542 
8543 	vmx_pending_event_ipi_fired = false;
8544 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
8545 
8546 	vmx_pending_event_guest_run = false;
8547 	test_set_guest(vmx_pending_event_guest);
8548 
8549 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
8550 
8551 	enter_guest();
8552 	skip_exit_vmcall();
8553 
8554 	if (guest_hlt)
8555 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8556 
8557 	irq_disable();
8558 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
8559 				   APIC_DM_FIXED | ipi_vector,
8560 				   0);
8561 
8562 	enter_guest();
8563 
8564 	assert_exit_reason(VMX_EXTINT);
8565 	report(!vmx_pending_event_guest_run,
8566 	       "Guest did not run before host received IPI");
8567 
8568 	irq_enable();
8569 	asm volatile ("nop");
8570 	irq_disable();
8571 	report(vmx_pending_event_ipi_fired,
8572 	       "Got pending interrupt after IRQ enabled");
8573 
8574 	if (guest_hlt)
8575 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8576 
8577 	enter_guest();
8578 	report(vmx_pending_event_guest_run,
8579 	       "Guest finished running when no interrupt");
8580 }
8581 
8582 static void vmx_pending_event_test(void)
8583 {
8584 	vmx_pending_event_test_core(false);
8585 }
8586 
8587 static void vmx_pending_event_hlt_test(void)
8588 {
8589 	vmx_pending_event_test_core(true);
8590 }
8591 
8592 static int vmx_window_test_db_count;
8593 
8594 static void vmx_window_test_db_handler(struct ex_regs *regs)
8595 {
8596 	vmx_window_test_db_count++;
8597 }
8598 
8599 static void vmx_nmi_window_test_guest(void)
8600 {
8601 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8602 
8603 	asm volatile("vmcall\n\t"
8604 		     "nop\n\t");
8605 
8606 	handle_exception(DB_VECTOR, NULL);
8607 }
8608 
8609 static void verify_nmi_window_exit(u64 rip)
8610 {
8611 	u32 exit_reason = vmcs_read(EXI_REASON);
8612 
8613 	report(exit_reason == VMX_NMI_WINDOW,
8614 	       "Exit reason (%d) is 'NMI window'", exit_reason);
8615 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8616 	       vmcs_read(GUEST_RIP), rip);
8617 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8618 }
8619 
8620 static void vmx_nmi_window_test(void)
8621 {
8622 	u64 nop_addr;
8623 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8624 
8625 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
8626 		report_skip("CPU does not support the \"Virtual NMIs\" VM-execution control.");
8627 		return;
8628 	}
8629 
8630 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
8631 		report_skip("CPU does not support the \"NMI-window exiting\" VM-execution control.");
8632 		return;
8633 	}
8634 
8635 	vmx_window_test_db_count = 0;
8636 
8637 	report_prefix_push("NMI-window");
8638 	test_set_guest(vmx_nmi_window_test_guest);
8639 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
8640 	enter_guest();
8641 	skip_exit_vmcall();
8642 	nop_addr = vmcs_read(GUEST_RIP);
8643 
8644 	/*
8645 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
8646 	 * RIP will not advance.
8647 	 */
8648 	report_prefix_push("active, no blocking");
8649 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8650 	enter_guest();
8651 	verify_nmi_window_exit(nop_addr);
8652 	report_prefix_pop();
8653 
8654 	/*
8655 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
8656 	 * a VM-exit on the next instruction after the nop. (The nop
8657 	 * is one byte.)
8658 	 */
8659 	report_prefix_push("active, blocking by MOV-SS");
8660 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8661 	enter_guest();
8662 	verify_nmi_window_exit(nop_addr + 1);
8663 	report_prefix_pop();
8664 
8665 	/*
8666 	 * Ask for "NMI-window exiting" (with event injection), and
8667 	 * expect a VM-exit after the event is injected. (RIP should
8668 	 * be at the address specified in the IDT entry for #DB.)
8669 	 */
8670 	report_prefix_push("active, no blocking, injecting #DB");
8671 	vmcs_write(ENT_INTR_INFO,
8672 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8673 	enter_guest();
8674 	verify_nmi_window_exit((u64)db_fault_addr);
8675 	report_prefix_pop();
8676 
8677 	/*
8678 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
8679 	 * a VM-exit after the next IRET (i.e. after the #DB handler
8680 	 * returns). So, RIP should be back at one byte past the nop.
8681 	 */
8682 	report_prefix_push("active, blocking by NMI");
8683 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
8684 	enter_guest();
8685 	verify_nmi_window_exit(nop_addr + 1);
8686 	report(vmx_window_test_db_count == 1,
8687 	       "#DB handler executed once (actual %d times)",
8688 	       vmx_window_test_db_count);
8689 	report_prefix_pop();
8690 
8691 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8692 		report_skip("CPU does not support activity state HLT.");
8693 	} else {
8694 		/*
8695 		 * Ask for "NMI-window exiting" when entering activity
8696 		 * state HLT, and expect an immediate VM-exit. RIP is
8697 		 * still one byte past the nop.
8698 		 */
8699 		report_prefix_push("halted, no blocking");
8700 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8701 		enter_guest();
8702 		verify_nmi_window_exit(nop_addr + 1);
8703 		report_prefix_pop();
8704 
8705 		/*
8706 		 * Ask for "NMI-window exiting" when entering activity
8707 		 * state HLT (with event injection), and expect a
8708 		 * VM-exit after the event is injected. (RIP should be
8709 		 * at the address specified in the IDT entry for #DB.)
8710 		 */
8711 		report_prefix_push("halted, no blocking, injecting #DB");
8712 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8713 		vmcs_write(ENT_INTR_INFO,
8714 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8715 			   DB_VECTOR);
8716 		enter_guest();
8717 		verify_nmi_window_exit((u64)db_fault_addr);
8718 		report_prefix_pop();
8719 	}
8720 
8721 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8722 	enter_guest();
8723 	report_prefix_pop();
8724 }
8725 
8726 static void vmx_intr_window_test_guest(void)
8727 {
8728 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8729 
8730 	/*
8731 	 * The two consecutive STIs are to ensure that only the first
8732 	 * one has a shadow. Note that NOP and STI are one byte
8733 	 * instructions.
8734 	 */
8735 	asm volatile("vmcall\n\t"
8736 		     "nop\n\t"
8737 		     "sti\n\t"
8738 		     "sti\n\t");
8739 
8740 	handle_exception(DB_VECTOR, NULL);
8741 }
8742 
8743 static void verify_intr_window_exit(u64 rip)
8744 {
8745 	u32 exit_reason = vmcs_read(EXI_REASON);
8746 
8747 	report(exit_reason == VMX_INTR_WINDOW,
8748 	       "Exit reason (%d) is 'interrupt window'", exit_reason);
8749 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8750 	       vmcs_read(GUEST_RIP), rip);
8751 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8752 }
8753 
8754 static void vmx_intr_window_test(void)
8755 {
8756 	u64 vmcall_addr;
8757 	u64 nop_addr;
8758 	unsigned int orig_db_gate_type;
8759 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8760 
8761 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
8762 		report_skip("CPU does not support the \"interrupt-window exiting\" VM-execution control.");
8763 		return;
8764 	}
8765 
8766 	/*
8767 	 * Change the IDT entry for #DB from interrupt gate to trap gate,
8768 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
8769 	 * be disabled after vectoring a #DB.
8770 	 */
8771 	orig_db_gate_type = boot_idt[DB_VECTOR].type;
8772 	boot_idt[DB_VECTOR].type = 15;
8773 
8774 	report_prefix_push("interrupt-window");
8775 	test_set_guest(vmx_intr_window_test_guest);
8776 	enter_guest();
8777 	assert_exit_reason(VMX_VMCALL);
8778 	vmcall_addr = vmcs_read(GUEST_RIP);
8779 
8780 	/*
8781 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
8782 	 * no blocking; expect an immediate VM-exit. Note that we have
8783 	 * not advanced past the vmcall instruction yet, so RIP should
8784 	 * point to the vmcall instruction.
8785 	 */
8786 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
8787 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8788 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
8789 	enter_guest();
8790 	verify_intr_window_exit(vmcall_addr);
8791 	report_prefix_pop();
8792 
8793 	/*
8794 	 * Ask for "interrupt-window exiting" (with event injection)
8795 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
8796 	 * the event is injected. That is, RIP should should be at the
8797 	 * address specified in the IDT entry for #DB.
8798 	 */
8799 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #DB");
8800 	vmcs_write(ENT_INTR_INFO,
8801 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8802 	vmcall_addr = vmcs_read(GUEST_RIP);
8803 	enter_guest();
8804 	verify_intr_window_exit((u64)db_fault_addr);
8805 	report_prefix_pop();
8806 
8807 	/*
8808 	 * Let the L2 guest run through the IRET, back to the VMCALL.
8809 	 * We have to clear the "interrupt-window exiting"
8810 	 * VM-execution control, or it would just keep causing
8811 	 * VM-exits. Then, advance past the VMCALL and set the
8812 	 * "interrupt-window exiting" VM-execution control again.
8813 	 */
8814 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8815 	enter_guest();
8816 	skip_exit_vmcall();
8817 	nop_addr = vmcs_read(GUEST_RIP);
8818 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8819 
8820 	/*
8821 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
8822 	 * RFLAGS.IF set, and expect a VM-exit on the next
8823 	 * instruction. (NOP is one byte.)
8824 	 */
8825 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
8826 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8827 	enter_guest();
8828 	verify_intr_window_exit(nop_addr + 1);
8829 	report_prefix_pop();
8830 
8831 	/*
8832 	 * Back up to the NOP and ask for "interrupt-window exiting"
8833 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
8834 	 * on the next instruction. (NOP is one byte.)
8835 	 */
8836 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
8837 	vmcs_write(GUEST_RIP, nop_addr);
8838 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
8839 	enter_guest();
8840 	verify_intr_window_exit(nop_addr + 1);
8841 	report_prefix_pop();
8842 
8843 	/*
8844 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
8845 	 * and expect a VM-exit on the instruction following the STI
8846 	 * shadow. Only the first STI (which is one byte past the NOP)
8847 	 * should have a shadow. The second STI (which is two bytes
8848 	 * past the NOP) has no shadow. Therefore, the interrupt
8849 	 * window opens at three bytes past the NOP.
8850 	 */
8851 	report_prefix_push("active, RFLAGS.IF = 0");
8852 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8853 	enter_guest();
8854 	verify_intr_window_exit(nop_addr + 3);
8855 	report_prefix_pop();
8856 
8857 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8858 		report_skip("CPU does not support activity state HLT.");
8859 	} else {
8860 		/*
8861 		 * Ask for "interrupt-window exiting" when entering
8862 		 * activity state HLT, and expect an immediate
8863 		 * VM-exit. RIP is still three bytes past the nop.
8864 		 */
8865 		report_prefix_push("halted, no blocking");
8866 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8867 		enter_guest();
8868 		verify_intr_window_exit(nop_addr + 3);
8869 		report_prefix_pop();
8870 
8871 		/*
8872 		 * Ask for "interrupt-window exiting" when entering
8873 		 * activity state HLT (with event injection), and
8874 		 * expect a VM-exit after the event is injected. That
8875 		 * is, RIP should should be at the address specified
8876 		 * in the IDT entry for #DB.
8877 		 */
8878 		report_prefix_push("halted, no blocking, injecting #DB");
8879 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8880 		vmcs_write(ENT_INTR_INFO,
8881 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8882 			   DB_VECTOR);
8883 		enter_guest();
8884 		verify_intr_window_exit((u64)db_fault_addr);
8885 		report_prefix_pop();
8886 	}
8887 
8888 	boot_idt[DB_VECTOR].type = orig_db_gate_type;
8889 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8890 	enter_guest();
8891 	report_prefix_pop();
8892 }
8893 
8894 #define GUEST_TSC_OFFSET (1u << 30)
8895 
8896 static u64 guest_tsc;
8897 
8898 static void vmx_store_tsc_test_guest(void)
8899 {
8900 	guest_tsc = rdtsc();
8901 }
8902 
8903 /*
8904  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
8905  * list, the value saved is not subject to the TSC offset that is
8906  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
8907  */
8908 static void vmx_store_tsc_test(void)
8909 {
8910 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
8911 	u64 low, high;
8912 
8913 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
8914 		report_skip("'Use TSC offsetting' not supported");
8915 		return;
8916 	}
8917 
8918 	test_set_guest(vmx_store_tsc_test_guest);
8919 
8920 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
8921 	vmcs_write(EXI_MSR_ST_CNT, 1);
8922 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
8923 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
8924 
8925 	low = rdtsc();
8926 	enter_guest();
8927 	high = rdtsc();
8928 
8929 	report(low + GUEST_TSC_OFFSET <= guest_tsc &&
8930 	       guest_tsc <= high + GUEST_TSC_OFFSET,
8931 	       "RDTSC value in the guest (%lu) is in range [%lu, %lu]",
8932 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
8933 	report(low <= msr_entry.value && msr_entry.value <= high,
8934 	       "IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
8935 	       msr_entry.value, low, high);
8936 }
8937 
8938 static void vmx_preemption_timer_zero_test_db_handler(struct ex_regs *regs)
8939 {
8940 }
8941 
8942 static void vmx_preemption_timer_zero_test_guest(void)
8943 {
8944 	while (vmx_get_test_stage() < 3)
8945 		vmcall();
8946 }
8947 
8948 static void vmx_preemption_timer_zero_activate_preemption_timer(void)
8949 {
8950 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
8951 	vmcs_write(PREEMPT_TIMER_VALUE, 0);
8952 }
8953 
8954 static void vmx_preemption_timer_zero_advance_past_vmcall(void)
8955 {
8956 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
8957 	enter_guest();
8958 	skip_exit_vmcall();
8959 }
8960 
8961 static void vmx_preemption_timer_zero_inject_db(bool intercept_db)
8962 {
8963 	vmx_preemption_timer_zero_activate_preemption_timer();
8964 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK |
8965 		   INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8966 	vmcs_write(EXC_BITMAP, intercept_db ? 1 << DB_VECTOR : 0);
8967 	enter_guest();
8968 }
8969 
8970 static void vmx_preemption_timer_zero_set_pending_dbg(u32 exception_bitmap)
8971 {
8972 	vmx_preemption_timer_zero_activate_preemption_timer();
8973 	vmcs_write(GUEST_PENDING_DEBUG, BIT(12) | DR_TRAP1);
8974 	vmcs_write(EXC_BITMAP, exception_bitmap);
8975 	enter_guest();
8976 }
8977 
8978 static void vmx_preemption_timer_zero_expect_preempt_at_rip(u64 expected_rip)
8979 {
8980 	u32 reason = (u32)vmcs_read(EXI_REASON);
8981 	u64 guest_rip = vmcs_read(GUEST_RIP);
8982 
8983 	report(reason == VMX_PREEMPT && guest_rip == expected_rip,
8984 	       "Exit reason is 0x%x (expected 0x%x) and guest RIP is %lx (0x%lx expected).",
8985 	       reason, VMX_PREEMPT, guest_rip, expected_rip);
8986 }
8987 
8988 /*
8989  * This test ensures that when the VMX preemption timer is zero at
8990  * VM-entry, a VM-exit occurs after any event injection and after any
8991  * pending debug exceptions are raised, but before execution of any
8992  * guest instructions.
8993  */
8994 static void vmx_preemption_timer_zero_test(void)
8995 {
8996 	u64 db_fault_address = (u64)get_idt_addr(&boot_idt[DB_VECTOR]);
8997 	handler old_db;
8998 	u32 reason;
8999 
9000 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9001 		report_skip("'Activate VMX-preemption timer' not supported");
9002 		return;
9003 	}
9004 
9005 	/*
9006 	 * Install a custom #DB handler that doesn't abort.
9007 	 */
9008 	old_db = handle_exception(DB_VECTOR,
9009 				  vmx_preemption_timer_zero_test_db_handler);
9010 
9011 	test_set_guest(vmx_preemption_timer_zero_test_guest);
9012 
9013 	/*
9014 	 * VMX-preemption timer should fire after event injection.
9015 	 */
9016 	vmx_set_test_stage(0);
9017 	vmx_preemption_timer_zero_inject_db(0);
9018 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
9019 	vmx_preemption_timer_zero_advance_past_vmcall();
9020 
9021 	/*
9022 	 * VMX-preemption timer should fire after event injection.
9023 	 * Exception bitmap is irrelevant, since you can't intercept
9024 	 * an event that you injected.
9025 	 */
9026 	vmx_set_test_stage(1);
9027 	vmx_preemption_timer_zero_inject_db(true);
9028 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
9029 	vmx_preemption_timer_zero_advance_past_vmcall();
9030 
9031 	/*
9032 	 * VMX-preemption timer should fire after pending debug exceptions
9033 	 * have delivered a #DB trap.
9034 	 */
9035 	vmx_set_test_stage(2);
9036 	vmx_preemption_timer_zero_set_pending_dbg(0);
9037 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
9038 	vmx_preemption_timer_zero_advance_past_vmcall();
9039 
9040 	/*
9041 	 * VMX-preemption timer would fire after pending debug exceptions
9042 	 * have delivered a #DB trap, but in this case, the #DB trap is
9043 	 * intercepted.
9044 	 */
9045 	vmx_set_test_stage(3);
9046 	vmx_preemption_timer_zero_set_pending_dbg(1 << DB_VECTOR);
9047 	reason = (u32)vmcs_read(EXI_REASON);
9048 	report(reason == VMX_EXC_NMI, "Exit reason is 0x%x (expected 0x%x)",
9049 	       reason, VMX_EXC_NMI);
9050 
9051 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9052 	enter_guest();
9053 
9054 	handle_exception(DB_VECTOR, old_db);
9055 }
9056 
9057 static u64 vmx_preemption_timer_tf_test_prev_rip;
9058 
9059 static void vmx_preemption_timer_tf_test_db_handler(struct ex_regs *regs)
9060 {
9061 	extern char vmx_preemption_timer_tf_test_endloop;
9062 
9063 	if (vmx_get_test_stage() == 2) {
9064 		/*
9065 		 * Stage 2 means that we're done, one way or another.
9066 		 * Arrange for the iret to drop us out of the wbinvd
9067 		 * loop and stop single-stepping.
9068 		 */
9069 		regs->rip = (u64)&vmx_preemption_timer_tf_test_endloop;
9070 		regs->rflags &= ~X86_EFLAGS_TF;
9071 	} else if (regs->rip == vmx_preemption_timer_tf_test_prev_rip) {
9072 		/*
9073 		 * The RIP should alternate between the wbinvd and the
9074 		 * jmp instruction in the code below. If we ever see
9075 		 * the same instruction twice in a row, that means a
9076 		 * single-step trap has been dropped. Let the
9077 		 * hypervisor know about the failure by executing a
9078 		 * VMCALL.
9079 		 */
9080 		vmcall();
9081 	}
9082 	vmx_preemption_timer_tf_test_prev_rip = regs->rip;
9083 }
9084 
9085 static void vmx_preemption_timer_tf_test_guest(void)
9086 {
9087 	/*
9088 	 * The hypervisor doesn't intercept WBINVD, so the loop below
9089 	 * shouldn't be a problem--it's just two instructions
9090 	 * executing in VMX non-root mode. However, when the
9091 	 * hypervisor is running in a virtual environment, the parent
9092 	 * hypervisor might intercept WBINVD and emulate it. If the
9093 	 * parent hypervisor is broken, the single-step trap after the
9094 	 * WBINVD might be lost.
9095 	 */
9096 	asm volatile("vmcall\n\t"
9097 		     "0: wbinvd\n\t"
9098 		     "1: jmp 0b\n\t"
9099 		     "vmx_preemption_timer_tf_test_endloop:");
9100 }
9101 
9102 /*
9103  * Ensure that the delivery of a "VMX-preemption timer expired"
9104  * VM-exit doesn't disrupt single-stepping in the guest. Note that
9105  * passing this test doesn't ensure correctness, because the test will
9106  * only fail if the VMX-preemtion timer fires at the right time (or
9107  * the wrong time, as it were).
9108  */
9109 static void vmx_preemption_timer_tf_test(void)
9110 {
9111 	handler old_db;
9112 	u32 reason;
9113 	int i;
9114 
9115 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9116 		report_skip("'Activate VMX-preemption timer' not supported");
9117 		return;
9118 	}
9119 
9120 	old_db = handle_exception(DB_VECTOR,
9121 				  vmx_preemption_timer_tf_test_db_handler);
9122 
9123 	test_set_guest(vmx_preemption_timer_tf_test_guest);
9124 
9125 	enter_guest();
9126 	skip_exit_vmcall();
9127 
9128 	vmx_set_test_stage(1);
9129 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9130 	vmcs_write(PREEMPT_TIMER_VALUE, 50000);
9131 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9132 
9133 	/*
9134 	 * The only exit we should see is "VMX-preemption timer
9135 	 * expired."  If we get a VMCALL exit, that means the #DB
9136 	 * handler has detected a missing single-step trap. It doesn't
9137 	 * matter where the guest RIP is when the VMX-preemption timer
9138 	 * expires (whether it's in the WBINVD loop or in the #DB
9139 	 * handler)--a single-step trap should never be discarded.
9140 	 */
9141 	for (i = 0; i < 10000; i++) {
9142 		enter_guest();
9143 		reason = (u32)vmcs_read(EXI_REASON);
9144 		if (reason == VMX_PREEMPT)
9145 			continue;
9146 		TEST_ASSERT(reason == VMX_VMCALL);
9147 		skip_exit_insn();
9148 		break;
9149 	}
9150 
9151 	report(reason == VMX_PREEMPT, "No single-step traps skipped");
9152 
9153 	vmx_set_test_stage(2);
9154 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9155 	enter_guest();
9156 
9157 	handle_exception(DB_VECTOR, old_db);
9158 }
9159 
9160 #define VMX_PREEMPTION_TIMER_EXPIRY_CYCLES 1000000
9161 
9162 static u64 vmx_preemption_timer_expiry_start;
9163 static u64 vmx_preemption_timer_expiry_finish;
9164 
9165 static void vmx_preemption_timer_expiry_test_guest(void)
9166 {
9167 	vmcall();
9168 	vmx_preemption_timer_expiry_start = fenced_rdtsc();
9169 
9170 	while (vmx_get_test_stage() == 0)
9171 		vmx_preemption_timer_expiry_finish = fenced_rdtsc();
9172 }
9173 
9174 /*
9175  * Test that the VMX-preemption timer is not excessively delayed.
9176  *
9177  * Per the SDM, volume 3, VM-entry starts the VMX-preemption timer
9178  * with the unsigned value in the VMX-preemption timer-value field,
9179  * and the VMX-preemption timer counts down by 1 every time bit X in
9180  * the TSC changes due to a TSC increment (where X is
9181  * IA32_VMX_MISC[4:0]). If the timer counts down to zero in any state
9182  * other than the wait-for-SIPI state, the logical processor
9183  * transitions to the C0 C-state and causes a VM-exit.
9184  *
9185  * The guest code above reads the starting TSC after VM-entry. At this
9186  * point, the VMX-preemption timer has already been activated. Next,
9187  * the guest code reads the current TSC in a loop, storing the value
9188  * read to memory.
9189  *
9190  * If the RDTSC in the loop reads a value past the VMX-preemption
9191  * timer deadline, then the VMX-preemption timer VM-exit must be
9192  * delivered before the next instruction retires. Even if a higher
9193  * priority SMI is delivered first, the VMX-preemption timer VM-exit
9194  * must be delivered before the next instruction retires. Hence, a TSC
9195  * value past the VMX-preemption timer deadline might be read, but it
9196  * cannot be stored. If a TSC value past the deadline *is* stored,
9197  * then the architectural specification has been violated.
9198  */
9199 static void vmx_preemption_timer_expiry_test(void)
9200 {
9201 	u32 preemption_timer_value;
9202 	union vmx_misc misc;
9203 	u64 tsc_deadline;
9204 	u32 reason;
9205 
9206 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9207 		report_skip("'Activate VMX-preemption timer' not supported");
9208 		return;
9209 	}
9210 
9211 	test_set_guest(vmx_preemption_timer_expiry_test_guest);
9212 
9213 	enter_guest();
9214 	skip_exit_vmcall();
9215 
9216 	misc.val = rdmsr(MSR_IA32_VMX_MISC);
9217 	preemption_timer_value =
9218 		VMX_PREEMPTION_TIMER_EXPIRY_CYCLES >> misc.pt_bit;
9219 
9220 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9221 	vmcs_write(PREEMPT_TIMER_VALUE, preemption_timer_value);
9222 	vmx_set_test_stage(0);
9223 
9224 	enter_guest();
9225 	reason = (u32)vmcs_read(EXI_REASON);
9226 	TEST_ASSERT(reason == VMX_PREEMPT);
9227 
9228 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9229 	vmx_set_test_stage(1);
9230 	enter_guest();
9231 
9232 	tsc_deadline = ((vmx_preemption_timer_expiry_start >> misc.pt_bit) <<
9233 			misc.pt_bit) + (preemption_timer_value << misc.pt_bit);
9234 
9235 	report(vmx_preemption_timer_expiry_finish < tsc_deadline,
9236 	       "Last stored guest TSC (%lu) < TSC deadline (%lu)",
9237 	       vmx_preemption_timer_expiry_finish, tsc_deadline);
9238 }
9239 
9240 static void vmx_db_test_guest(void)
9241 {
9242 	/*
9243 	 * For a hardware generated single-step #DB.
9244 	 */
9245 	asm volatile("vmcall;"
9246 		     "nop;"
9247 		     ".Lpost_nop:");
9248 	/*
9249 	 * ...in a MOVSS shadow, with pending debug exceptions.
9250 	 */
9251 	asm volatile("vmcall;"
9252 		     "nop;"
9253 		     ".Lpost_movss_nop:");
9254 	/*
9255 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
9256 	 * emulates it in software.)
9257 	 */
9258 	asm volatile("vmcall;"
9259 		     "wbinvd;"
9260 		     ".Lpost_wbinvd:");
9261 	/*
9262 	 * ...in a MOVSS shadow, with pending debug exceptions.
9263 	 */
9264 	asm volatile("vmcall;"
9265 		     "wbinvd;"
9266 		     ".Lpost_movss_wbinvd:");
9267 	/*
9268 	 * For a hardware generated single-step #DB in a transactional region.
9269 	 */
9270 	asm volatile("vmcall;"
9271 		     ".Lxbegin: xbegin .Lskip_rtm;"
9272 		     "xend;"
9273 		     ".Lskip_rtm:");
9274 }
9275 
9276 /*
9277  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
9278  * L2. No #DB is delivered and L2 continues to the next point of
9279  * interest.
9280  */
9281 static void dismiss_db(void)
9282 {
9283 	vmcs_write(GUEST_PENDING_DEBUG, 0);
9284 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
9285 	enter_guest();
9286 }
9287 
9288 /*
9289  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
9290  * Then throw away the #DB exception and resume L2.
9291  */
9292 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
9293 			  void *expected_rip, u64 expected_exit_qual,
9294 			  u64 expected_dr6)
9295 {
9296 	u32 reason = vmcs_read(EXI_REASON);
9297 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
9298 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
9299 	u64 guest_rip = vmcs_read(GUEST_RIP);
9300 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
9301 	u64 dr6 = read_dr6();
9302 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
9303 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
9304 
9305 	report(reason == VMX_EXC_NMI && intr_info == expected_intr_info,
9306 	       "Expected #DB VM-exit");
9307 	report((u64)expected_rip == guest_rip, "Expected RIP %p (actual %lx)",
9308 	       expected_rip, guest_rip);
9309 	report_xfail(xfail_pdbg, 0 == guest_pending_dbg,
9310 		     "Expected pending debug exceptions 0 (actual %lx)",
9311 		     guest_pending_dbg);
9312 	report_xfail(xfail_qual, expected_exit_qual == exit_qual,
9313 		     "Expected exit qualification %lx (actual %lx)",
9314 		     expected_exit_qual, exit_qual);
9315 	report_xfail(xfail_dr6, expected_dr6 == dr6,
9316 		     "Expected DR6 %lx (actual %lx)", expected_dr6, dr6);
9317 	dismiss_db();
9318 }
9319 
9320 /*
9321  * Assuming the guest has just exited on a VMCALL instruction, skip
9322  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
9323  * pending debug exceptions are non-zero, set the VMCS up as if the
9324  * previous instruction was a MOVSS that generated the indicated
9325  * pending debug exceptions. Then enter L2.
9326  */
9327 static void single_step_guest(const char *test_name, u64 starting_dr6,
9328 			      u64 pending_debug_exceptions)
9329 {
9330 	printf("\n%s\n", test_name);
9331 	skip_exit_vmcall();
9332 	write_dr6(starting_dr6);
9333 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9334 	if (pending_debug_exceptions) {
9335 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
9336 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
9337 	}
9338 	enter_guest();
9339 }
9340 
9341 /*
9342  * When L1 intercepts #DB, verify that a single-step trap clears
9343  * pending debug exceptions, populates the exit qualification field
9344  * properly, and that DR6 is not prematurely clobbered. In a
9345  * (simulated) MOVSS shadow, make sure that the pending debug
9346  * exception bits are properly accumulated into the exit qualification
9347  * field.
9348  */
9349 static void vmx_db_test(void)
9350 {
9351 	/*
9352 	 * We are going to set a few arbitrary bits in DR6 to verify that
9353 	 * (a) DR6 is not modified by an intercepted #DB, and
9354 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
9355          *     the exit qualification field for a subsequent #DB exception.
9356 	 */
9357 	const u64 starting_dr6 = DR6_RESERVED | BIT(13) | DR_TRAP3 | DR_TRAP1;
9358 	extern char post_nop asm(".Lpost_nop");
9359 	extern char post_movss_nop asm(".Lpost_movss_nop");
9360 	extern char post_wbinvd asm(".Lpost_wbinvd");
9361 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
9362 	extern char xbegin asm(".Lxbegin");
9363 	extern char skip_rtm asm(".Lskip_rtm");
9364 
9365 	/*
9366 	 * L1 wants to intercept #DB exceptions encountered in L2.
9367 	 */
9368 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
9369 
9370 	/*
9371 	 * Start L2 and run it up to the first point of interest.
9372 	 */
9373 	test_set_guest(vmx_db_test_guest);
9374 	enter_guest();
9375 
9376 	/*
9377 	 * Hardware-delivered #DB trap for single-step sets the
9378 	 * standard that L0 has to follow for emulated instructions.
9379 	 */
9380 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
9381 	check_db_exit(false, false, false, &post_nop, DR_STEP, starting_dr6);
9382 
9383 	/*
9384 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
9385 	 * also sets the standard that L0 has to follow for emulated
9386 	 * instructions. Here, we establish the VMCS pending debug
9387 	 * exceptions to indicate that the simulated MOVSS triggered a
9388 	 * data breakpoint as well as the single-step trap.
9389 	 */
9390 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
9391 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0 );
9392 	check_db_exit(false, false, false, &post_movss_nop, DR_STEP | DR_TRAP0,
9393 		      starting_dr6);
9394 
9395 	/*
9396 	 * L0 synthesized #DB trap for single-step is buggy, because
9397 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
9398 	 * reconstitute the exit qualification from the prematurely
9399 	 * modified DR6, but fails miserably.
9400 	 */
9401 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
9402 	check_db_exit(false, false, false, &post_wbinvd, DR_STEP, starting_dr6);
9403 
9404 	/*
9405 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
9406 	 * even worse, because L0 also leaves the pending debug
9407 	 * exceptions in the VMCS instead of accumulating them into
9408 	 * the exit qualification field for the #DB exception.
9409 	 */
9410 	single_step_guest("Software synthesized single-step in MOVSS shadow",
9411 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0);
9412 	check_db_exit(true, false, true, &post_movss_wbinvd, DR_STEP | DR_TRAP0,
9413 		      starting_dr6);
9414 
9415 	/*
9416 	 * Optional RTM test for hardware that supports RTM, to
9417 	 * demonstrate that the current volume 3 of the SDM
9418 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
9419 	 * qualification for debug exceptions is not reserved. It is
9420 	 * set to 1 if a debug exception (#DB) or a breakpoint
9421 	 * exception (#BP) occurs inside an RTM region while advanced
9422 	 * debugging of RTM transactional regions is enabled.
9423 	 */
9424 	if (this_cpu_has(X86_FEATURE_RTM)) {
9425 		vmcs_write(ENT_CONTROLS,
9426 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
9427 		/*
9428 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
9429 		 * in the guest to enable advanced debugging of RTM
9430 		 * transactional regions.
9431 		 */
9432 		vmcs_write(GUEST_DR7, BIT(11));
9433 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
9434 		single_step_guest("Hardware delivered single-step in "
9435 				  "transactional region", starting_dr6, 0);
9436 		check_db_exit(false, false, false, &xbegin, BIT(16),
9437 			      starting_dr6);
9438 	} else {
9439 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
9440 		enter_guest();
9441 	}
9442 }
9443 
9444 static void enable_vid(void)
9445 {
9446 	void *virtual_apic_page;
9447 
9448 	assert(cpu_has_apicv());
9449 
9450 	disable_intercept_for_x2apic_msrs();
9451 
9452 	virtual_apic_page = alloc_page();
9453 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
9454 
9455 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
9456 
9457 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
9458 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
9459 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
9460 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
9461 
9462 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
9463 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
9464 }
9465 
9466 static void trigger_ioapic_scan_thread(void *data)
9467 {
9468 	/* Wait until other CPU entered L2 */
9469 	while (vmx_get_test_stage() != 1)
9470 		;
9471 
9472 	/* Trigger ioapic scan */
9473 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
9474 	vmx_set_test_stage(2);
9475 }
9476 
9477 static void irq_79_handler_guest(isr_regs_t *regs)
9478 {
9479 	eoi();
9480 
9481 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
9482 	vmcall();
9483 }
9484 
9485 /*
9486  * Constant for num of busy-loop iterations after which
9487  * a timer interrupt should have happened in host
9488  */
9489 #define TIMER_INTERRUPT_DELAY 100000000
9490 
9491 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
9492 {
9493 	handle_irq(0x79, irq_79_handler_guest);
9494 	irq_enable();
9495 
9496 	/* Signal to L1 CPU to trigger ioapic scan */
9497 	vmx_set_test_stage(1);
9498 	/* Wait until L1 CPU to trigger ioapic scan */
9499 	while (vmx_get_test_stage() != 2)
9500 		;
9501 
9502 	/*
9503 	 * Wait for L0 timer interrupt to be raised while we run in L2
9504 	 * such that L0 will process the IOAPIC scan request before
9505 	 * resuming L2
9506 	 */
9507 	delay(TIMER_INTERRUPT_DELAY);
9508 
9509 	asm volatile ("int $0x79");
9510 }
9511 
9512 static void vmx_eoi_bitmap_ioapic_scan_test(void)
9513 {
9514 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
9515 		report_skip(__func__);
9516 		return;
9517 	}
9518 
9519 	enable_vid();
9520 
9521 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
9522 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
9523 
9524 	/*
9525 	 * Launch L2.
9526 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
9527 	 * In case the reason isn't VMX_VMCALL, the asserion inside
9528 	 * skip_exit_vmcall() will fail.
9529 	 */
9530 	enter_guest();
9531 	skip_exit_vmcall();
9532 
9533 	/* Let L2 finish */
9534 	enter_guest();
9535 	report_pass(__func__);
9536 }
9537 
9538 #define HLT_WITH_RVI_VECTOR		(0xf1)
9539 
9540 bool vmx_hlt_with_rvi_guest_isr_fired;
9541 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
9542 {
9543 	vmx_hlt_with_rvi_guest_isr_fired = true;
9544 	eoi();
9545 }
9546 
9547 static void vmx_hlt_with_rvi_guest(void)
9548 {
9549 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
9550 
9551 	irq_enable();
9552 	asm volatile ("nop");
9553 
9554 	vmcall();
9555 }
9556 
9557 static void vmx_hlt_with_rvi_test(void)
9558 {
9559 	if (!cpu_has_apicv()) {
9560 		report_skip(__func__);
9561 		return;
9562 	}
9563 
9564 	enable_vid();
9565 
9566 	vmx_hlt_with_rvi_guest_isr_fired = false;
9567 	test_set_guest(vmx_hlt_with_rvi_guest);
9568 
9569 	enter_guest();
9570 	skip_exit_vmcall();
9571 
9572 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
9573 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
9574 	enter_guest();
9575 
9576 	report(vmx_hlt_with_rvi_guest_isr_fired, "Interrupt raised in guest");
9577 }
9578 
9579 static void set_irq_line_thread(void *data)
9580 {
9581 	/* Wait until other CPU entered L2 */
9582 	while (vmx_get_test_stage() != 1)
9583 		;
9584 
9585 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
9586 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9587 	vmx_set_test_stage(2);
9588 }
9589 
9590 static bool irq_78_handler_vmcall_before_eoi;
9591 static void irq_78_handler_guest(isr_regs_t *regs)
9592 {
9593 	set_irq_line(0xf, 0);
9594 	if (irq_78_handler_vmcall_before_eoi)
9595 		vmcall();
9596 	eoi();
9597 	vmcall();
9598 }
9599 
9600 static void vmx_apic_passthrough_guest(void)
9601 {
9602 	handle_irq(0x78, irq_78_handler_guest);
9603 	irq_enable();
9604 
9605 	/* If requested, wait for other CPU to trigger ioapic scan */
9606 	if (vmx_get_test_stage() < 1) {
9607 		vmx_set_test_stage(1);
9608 		while (vmx_get_test_stage() != 2)
9609 			;
9610 	}
9611 
9612 	set_irq_line(0xf, 1);
9613 }
9614 
9615 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
9616 {
9617 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
9618 		report_skip(__func__);
9619 		return;
9620 	}
9621 
9622 	/* Test device is required for generating IRQs */
9623 	if (!test_device_enabled()) {
9624 		report_skip(__func__);
9625 		return;
9626 	}
9627 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9628 	u64 cpu_ctrl_1 = 0;
9629 
9630 	disable_intercept_for_x2apic_msrs();
9631 
9632 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9633 
9634 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9635 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9636 
9637 	if (set_irq_line_from_thread) {
9638 		irq_78_handler_vmcall_before_eoi = false;
9639 		on_cpu_async(1, set_irq_line_thread, NULL);
9640 	} else {
9641 		irq_78_handler_vmcall_before_eoi = true;
9642 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9643 		vmx_set_test_stage(2);
9644 	}
9645 	test_set_guest(vmx_apic_passthrough_guest);
9646 
9647 	if (irq_78_handler_vmcall_before_eoi) {
9648 		/* Before EOI remote_irr should still be set */
9649 		enter_guest();
9650 		skip_exit_vmcall();
9651 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
9652 			"IOAPIC pass-through: remote_irr=1 before EOI");
9653 	}
9654 
9655 	/* After EOI remote_irr should be cleared */
9656 	enter_guest();
9657 	skip_exit_vmcall();
9658 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
9659 		"IOAPIC pass-through: remote_irr=0 after EOI");
9660 
9661 	/* Let L2 finish */
9662 	enter_guest();
9663 	report_pass(__func__);
9664 }
9665 
9666 static void vmx_apic_passthrough_test(void)
9667 {
9668 	vmx_apic_passthrough(false);
9669 }
9670 
9671 static void vmx_apic_passthrough_thread_test(void)
9672 {
9673 	vmx_apic_passthrough(true);
9674 }
9675 
9676 static void vmx_apic_passthrough_tpr_threshold_guest(void)
9677 {
9678 	cli();
9679 	apic_set_tpr(0);
9680 }
9681 
9682 static bool vmx_apic_passthrough_tpr_threshold_ipi_isr_fired;
9683 static void vmx_apic_passthrough_tpr_threshold_ipi_isr(isr_regs_t *regs)
9684 {
9685 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = true;
9686 	eoi();
9687 }
9688 
9689 static void vmx_apic_passthrough_tpr_threshold_test(void)
9690 {
9691 	int ipi_vector = 0xe1;
9692 
9693 	disable_intercept_for_x2apic_msrs();
9694 	vmcs_clear_bits(PIN_CONTROLS, PIN_EXTINT);
9695 
9696 	/* Raise L0 TPR-threshold by queueing vector in LAPIC IRR */
9697 	cli();
9698 	apic_set_tpr((ipi_vector >> 4) + 1);
9699 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
9700 			APIC_DM_FIXED | ipi_vector,
9701 			0);
9702 
9703 	test_set_guest(vmx_apic_passthrough_tpr_threshold_guest);
9704 	enter_guest();
9705 
9706 	report(apic_get_tpr() == 0, "TPR was zero by guest");
9707 
9708 	/* Clean pending self-IPI */
9709 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = false;
9710 	handle_irq(ipi_vector, vmx_apic_passthrough_tpr_threshold_ipi_isr);
9711 	sti();
9712 	asm volatile ("nop");
9713 	report(vmx_apic_passthrough_tpr_threshold_ipi_isr_fired, "self-IPI fired");
9714 
9715 	report_pass(__func__);
9716 }
9717 
9718 static u64 init_signal_test_exit_reason;
9719 static bool init_signal_test_thread_continued;
9720 
9721 static void init_signal_test_thread(void *data)
9722 {
9723 	struct vmcs *test_vmcs = data;
9724 
9725 	/* Enter VMX operation (i.e. exec VMXON) */
9726 	u64 *ap_vmxon_region = alloc_page();
9727 	enable_vmx();
9728 	init_vmx(ap_vmxon_region);
9729 	_vmx_on(ap_vmxon_region);
9730 
9731 	/* Signal CPU have entered VMX operation */
9732 	vmx_set_test_stage(1);
9733 
9734 	/* Wait for BSP CPU to send INIT signal */
9735 	while (vmx_get_test_stage() != 2)
9736 		;
9737 
9738 	/*
9739 	 * Signal that we continue as usual as INIT signal
9740 	 * should be blocked while CPU is in VMX operation
9741 	 */
9742 	vmx_set_test_stage(3);
9743 
9744 	/* Wait for signal to enter VMX non-root mode */
9745 	while (vmx_get_test_stage() != 4)
9746 		;
9747 
9748 	/* Enter VMX non-root mode */
9749 	test_set_guest(v2_null_test_guest);
9750 	make_vmcs_current(test_vmcs);
9751 	enter_guest();
9752 	/* Save exit reason for BSP CPU to compare to expected result */
9753 	init_signal_test_exit_reason = vmcs_read(EXI_REASON);
9754 	/* VMCLEAR test-vmcs so it could be loaded by BSP CPU */
9755 	vmcs_clear(test_vmcs);
9756 	launched = false;
9757 	/* Signal that CPU exited to VMX root mode */
9758 	vmx_set_test_stage(5);
9759 
9760 	/* Wait for BSP CPU to signal to exit VMX operation */
9761 	while (vmx_get_test_stage() != 6)
9762 		;
9763 
9764 	/* Exit VMX operation (i.e. exec VMXOFF) */
9765 	vmx_off();
9766 
9767 	/*
9768 	 * Signal to BSP CPU that we continue as usual as INIT signal
9769 	 * should have been consumed by VMX_INIT exit from guest
9770 	 */
9771 	vmx_set_test_stage(7);
9772 
9773 	/* Wait for BSP CPU to signal to enter VMX operation */
9774 	while (vmx_get_test_stage() != 8)
9775 		;
9776 	/* Enter VMX operation (i.e. exec VMXON) */
9777 	_vmx_on(ap_vmxon_region);
9778 	/* Signal to BSP we are in VMX operation */
9779 	vmx_set_test_stage(9);
9780 
9781 	/* Wait for BSP CPU to send INIT signal */
9782 	while (vmx_get_test_stage() != 10)
9783 		;
9784 
9785 	/* Exit VMX operation (i.e. exec VMXOFF) */
9786 	vmx_off();
9787 
9788 	/*
9789 	 * Exiting VMX operation should result in latched
9790 	 * INIT signal being processed. Therefore, we should
9791 	 * never reach the below code. Thus, signal to BSP
9792 	 * CPU if we have reached here so it is able to
9793 	 * report an issue if it happens.
9794 	 */
9795 	init_signal_test_thread_continued = true;
9796 }
9797 
9798 #define INIT_SIGNAL_TEST_DELAY	100000000ULL
9799 
9800 static void vmx_init_signal_test(void)
9801 {
9802 	struct vmcs *test_vmcs;
9803 
9804 	if (cpu_count() < 2) {
9805 		report_skip(__func__);
9806 		return;
9807 	}
9808 
9809 	/* VMCLEAR test-vmcs so it could be loaded by other CPU */
9810 	vmcs_save(&test_vmcs);
9811 	vmcs_clear(test_vmcs);
9812 
9813 	vmx_set_test_stage(0);
9814 	on_cpu_async(1, init_signal_test_thread, test_vmcs);
9815 
9816 	/* Wait for other CPU to enter VMX operation */
9817 	while (vmx_get_test_stage() != 1)
9818 		;
9819 
9820 	/* Send INIT signal to other CPU */
9821 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9822 				   id_map[1]);
9823 	/* Signal other CPU we have sent INIT signal */
9824 	vmx_set_test_stage(2);
9825 
9826 	/*
9827 	 * Wait reasonable amount of time for INIT signal to
9828 	 * be received on other CPU and verify that other CPU
9829 	 * have proceed as usual to next test stage as INIT
9830 	 * signal should be blocked while other CPU in
9831 	 * VMX operation
9832 	 */
9833 	delay(INIT_SIGNAL_TEST_DELAY);
9834 	report(vmx_get_test_stage() == 3,
9835 	       "INIT signal blocked when CPU in VMX operation");
9836 	/* No point to continue if we failed at this point */
9837 	if (vmx_get_test_stage() != 3)
9838 		return;
9839 
9840 	/* Signal other CPU to enter VMX non-root mode */
9841 	init_signal_test_exit_reason = -1ull;
9842 	vmx_set_test_stage(4);
9843 	/*
9844 	 * Wait reasonable amont of time for other CPU
9845 	 * to exit to VMX root mode
9846 	 */
9847 	delay(INIT_SIGNAL_TEST_DELAY);
9848 	if (vmx_get_test_stage() != 5) {
9849 		report_fail("Pending INIT signal didn't result in VMX exit");
9850 		return;
9851 	}
9852 	report(init_signal_test_exit_reason == VMX_INIT,
9853 			"INIT signal during VMX non-root mode result in exit-reason %s (%lu)",
9854 			exit_reason_description(init_signal_test_exit_reason),
9855 			init_signal_test_exit_reason);
9856 
9857 	/* Run guest to completion */
9858 	make_vmcs_current(test_vmcs);
9859 	enter_guest();
9860 
9861 	/* Signal other CPU to exit VMX operation */
9862 	init_signal_test_thread_continued = false;
9863 	vmx_set_test_stage(6);
9864 
9865 	/* Wait reasonable amount of time for other CPU to exit VMX operation */
9866 	delay(INIT_SIGNAL_TEST_DELAY);
9867 	report(vmx_get_test_stage() == 7,
9868 	       "INIT signal consumed on VMX_INIT exit");
9869 	/* No point to continue if we failed at this point */
9870 	if (vmx_get_test_stage() != 7)
9871 		return;
9872 
9873 	/* Signal other CPU to enter VMX operation */
9874 	vmx_set_test_stage(8);
9875 	/* Wait for other CPU to enter VMX operation */
9876 	while (vmx_get_test_stage() != 9)
9877 		;
9878 
9879 	/* Send INIT signal to other CPU */
9880 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9881 				   id_map[1]);
9882 	/* Signal other CPU we have sent INIT signal */
9883 	vmx_set_test_stage(10);
9884 
9885 	/*
9886 	 * Wait reasonable amount of time for other CPU
9887 	 * to exit VMX operation and process INIT signal
9888 	 */
9889 	delay(INIT_SIGNAL_TEST_DELAY);
9890 	report(!init_signal_test_thread_continued,
9891 	       "INIT signal processed after exit VMX operation");
9892 
9893 	/*
9894 	 * TODO: Send SIPI to other CPU to sipi_entry (See x86/cstart64.S)
9895 	 * to re-init it to kvm-unit-tests standard environment.
9896 	 * Somehow (?) verify that SIPI was indeed received.
9897 	 */
9898 }
9899 
9900 #define SIPI_SIGNAL_TEST_DELAY	100000000ULL
9901 
9902 static void vmx_sipi_test_guest(void)
9903 {
9904 	if (apic_id() == 0) {
9905 		/* wait AP enter guest with activity=WAIT_SIPI */
9906 		while (vmx_get_test_stage() != 1)
9907 			;
9908 		delay(SIPI_SIGNAL_TEST_DELAY);
9909 
9910 		/* First SIPI signal */
9911 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9912 		report_pass("BSP(L2): Send first SIPI to cpu[%d]", id_map[1]);
9913 
9914 		/* wait AP enter guest */
9915 		while (vmx_get_test_stage() != 2)
9916 			;
9917 		delay(SIPI_SIGNAL_TEST_DELAY);
9918 
9919 		/* Second SIPI signal should be ignored since AP is not in WAIT_SIPI state */
9920 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9921 		report_pass("BSP(L2): Send second SIPI to cpu[%d]", id_map[1]);
9922 
9923 		/* Delay a while to check whether second SIPI would cause VMExit */
9924 		delay(SIPI_SIGNAL_TEST_DELAY);
9925 
9926 		/* Test is done, notify AP to exit test */
9927 		vmx_set_test_stage(3);
9928 
9929 		/* wait AP exit non-root mode */
9930 		while (vmx_get_test_stage() != 5)
9931 			;
9932 	} else {
9933 		/* wait BSP notify test is done */
9934 		while (vmx_get_test_stage() != 3)
9935 			;
9936 
9937 		/* AP exit guest */
9938 		vmx_set_test_stage(4);
9939 	}
9940 }
9941 
9942 static void sipi_test_ap_thread(void *data)
9943 {
9944 	struct vmcs *ap_vmcs;
9945 	u64 *ap_vmxon_region;
9946 	void *ap_stack, *ap_syscall_stack;
9947 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9948 	u64 cpu_ctrl_1 = 0;
9949 
9950 	/* Enter VMX operation (i.e. exec VMXON) */
9951 	ap_vmxon_region = alloc_page();
9952 	enable_vmx();
9953 	init_vmx(ap_vmxon_region);
9954 	_vmx_on(ap_vmxon_region);
9955 	init_vmcs(&ap_vmcs);
9956 	make_vmcs_current(ap_vmcs);
9957 
9958 	/* Set stack for AP */
9959 	ap_stack = alloc_page();
9960 	ap_syscall_stack = alloc_page();
9961 	vmcs_write(GUEST_RSP, (u64)(ap_stack + PAGE_SIZE - 1));
9962 	vmcs_write(GUEST_SYSENTER_ESP, (u64)(ap_syscall_stack + PAGE_SIZE - 1));
9963 
9964 	/* passthrough lapic to L2 */
9965 	disable_intercept_for_x2apic_msrs();
9966 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9967 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9968 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9969 
9970 	/* Set guest activity state to wait-for-SIPI state */
9971 	vmcs_write(GUEST_ACTV_STATE, ACTV_WAIT_SIPI);
9972 
9973 	vmx_set_test_stage(1);
9974 
9975 	/* AP enter guest */
9976 	enter_guest();
9977 
9978 	if (vmcs_read(EXI_REASON) == VMX_SIPI) {
9979 		report_pass("AP: Handle SIPI VMExit");
9980 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
9981 		vmx_set_test_stage(2);
9982 	} else {
9983 		report_fail("AP: Unexpected VMExit, reason=%ld", vmcs_read(EXI_REASON));
9984 		vmx_off();
9985 		return;
9986 	}
9987 
9988 	/* AP enter guest */
9989 	enter_guest();
9990 
9991 	report(vmcs_read(EXI_REASON) != VMX_SIPI,
9992 		"AP: should no SIPI VMExit since activity is not in WAIT_SIPI state");
9993 
9994 	/* notify BSP that AP is already exit from non-root mode */
9995 	vmx_set_test_stage(5);
9996 
9997 	/* Leave VMX operation */
9998 	vmx_off();
9999 }
10000 
10001 static void vmx_sipi_signal_test(void)
10002 {
10003 	if (!(rdmsr(MSR_IA32_VMX_MISC) & MSR_IA32_VMX_MISC_ACTIVITY_WAIT_SIPI)) {
10004 		printf("\tACTIVITY_WAIT_SIPI state is not supported.\n");
10005 		return;
10006 	}
10007 
10008 	if (cpu_count() < 2) {
10009 		report_skip(__func__);
10010 		return;
10011 	}
10012 
10013 	u64 cpu_ctrl_0 = CPU_SECONDARY;
10014 	u64 cpu_ctrl_1 = 0;
10015 
10016 	/* passthrough lapic to L2 */
10017 	disable_intercept_for_x2apic_msrs();
10018 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
10019 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
10020 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
10021 
10022 	test_set_guest(vmx_sipi_test_guest);
10023 
10024 	/* update CR3 on AP */
10025 	on_cpu(1, update_cr3, (void *)read_cr3());
10026 
10027 	/* start AP */
10028 	on_cpu_async(1, sipi_test_ap_thread, NULL);
10029 
10030 	vmx_set_test_stage(0);
10031 
10032 	/* BSP enter guest */
10033 	enter_guest();
10034 }
10035 
10036 
10037 enum vmcs_access {
10038 	ACCESS_VMREAD,
10039 	ACCESS_VMWRITE,
10040 	ACCESS_NONE,
10041 };
10042 
10043 struct vmcs_shadow_test_common {
10044 	enum vmcs_access op;
10045 	enum Reason reason;
10046 	u64 field;
10047 	u64 value;
10048 	u64 flags;
10049 	u64 time;
10050 } l1_l2_common;
10051 
10052 static inline u64 vmread_flags(u64 field, u64 *val)
10053 {
10054 	u64 flags;
10055 
10056 	asm volatile ("vmread %2, %1; pushf; pop %0"
10057 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
10058 	return flags & X86_EFLAGS_ALU;
10059 }
10060 
10061 static inline u64 vmwrite_flags(u64 field, u64 val)
10062 {
10063 	u64 flags;
10064 
10065 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
10066 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
10067 	return flags & X86_EFLAGS_ALU;
10068 }
10069 
10070 static void vmx_vmcs_shadow_test_guest(void)
10071 {
10072 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10073 	u64 start;
10074 
10075 	while (c->op != ACCESS_NONE) {
10076 		start = rdtsc();
10077 		switch (c->op) {
10078 		default:
10079 			c->flags = -1ull;
10080 			break;
10081 		case ACCESS_VMREAD:
10082 			c->flags = vmread_flags(c->field, &c->value);
10083 			break;
10084 		case ACCESS_VMWRITE:
10085 			c->flags = vmwrite_flags(c->field, 0);
10086 			break;
10087 		}
10088 		c->time = rdtsc() - start;
10089 		vmcall();
10090 	}
10091 }
10092 
10093 static u64 vmread_from_shadow(u64 field)
10094 {
10095 	struct vmcs *primary;
10096 	struct vmcs *shadow;
10097 	u64 value;
10098 
10099 	TEST_ASSERT(!vmcs_save(&primary));
10100 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10101 	TEST_ASSERT(!make_vmcs_current(shadow));
10102 	value = vmcs_read(field);
10103 	TEST_ASSERT(!make_vmcs_current(primary));
10104 	return value;
10105 }
10106 
10107 static u64 vmwrite_to_shadow(u64 field, u64 value)
10108 {
10109 	struct vmcs *primary;
10110 	struct vmcs *shadow;
10111 
10112 	TEST_ASSERT(!vmcs_save(&primary));
10113 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10114 	TEST_ASSERT(!make_vmcs_current(shadow));
10115 	vmcs_write(field, value);
10116 	value = vmcs_read(field);
10117 	TEST_ASSERT(!make_vmcs_current(primary));
10118 	return value;
10119 }
10120 
10121 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
10122 {
10123 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10124 
10125 	c->op = access;
10126 	vmcs_write(VMX_INST_ERROR, 0);
10127 	enter_guest();
10128 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
10129 	if (c->reason != VMX_VMCALL) {
10130 		skip_exit_insn();
10131 		enter_guest();
10132 	}
10133 	skip_exit_vmcall();
10134 }
10135 
10136 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
10137 {
10138 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10139 	struct vmcs *shadow;
10140 	u64 value;
10141 	uintptr_t flags[2];
10142 	bool good_shadow;
10143 	u32 vmx_inst_error;
10144 
10145 	report_prefix_pushf("field %lx", field);
10146 	c->field = field;
10147 
10148 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10149 	if (shadow != (struct vmcs *)-1ull) {
10150 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
10151 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
10152 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
10153 	} else {
10154 		/*
10155 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
10156 		 * shadowed-fields should fail with setting RFLAGS.CF.
10157 		 */
10158 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
10159 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
10160 		good_shadow = false;
10161 	}
10162 
10163 	/* Intercept both VMREAD and VMWRITE. */
10164 	report_prefix_push("no VMREAD/VMWRITE permission");
10165 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
10166 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
10167 		set_bit(field, bitmap[ACCESS_VMREAD]);
10168 		set_bit(field, bitmap[ACCESS_VMWRITE]);
10169 	}
10170 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10171 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10172 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10173 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10174 	report_prefix_pop();
10175 
10176 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
10177 		goto out;
10178 
10179 	/* Permit shadowed VMREAD. */
10180 	report_prefix_push("VMREAD permission only");
10181 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10182 	set_bit(field, bitmap[ACCESS_VMWRITE]);
10183 	if (good_shadow)
10184 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10185 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10186 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10187 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10188 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10189 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10190 	       c->time);
10191 	report(c->flags == flags[ACCESS_VMREAD],
10192 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10193 	       c->flags, flags[ACCESS_VMREAD]);
10194 	if (good_shadow)
10195 		report(c->value == value,
10196 		       "value read from shadow (%lx) is as expected (%lx)",
10197 		       c->value, value);
10198 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10199 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10200 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10201 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10202 	report_prefix_pop();
10203 
10204 	/* Permit shadowed VMWRITE. */
10205 	report_prefix_push("VMWRITE permission only");
10206 	set_bit(field, bitmap[ACCESS_VMREAD]);
10207 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10208 	if (good_shadow)
10209 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10210 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10211 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10212 	report(c->reason == VMX_VMCALL,
10213 		"shadowed for VMWRITE (in %ld cycles)",
10214 		c->time);
10215 	report(c->flags == flags[ACCESS_VMREAD],
10216 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10217 	       c->flags, flags[ACCESS_VMREAD]);
10218 	if (good_shadow) {
10219 		value = vmread_from_shadow(field);
10220 		report(value == 0,
10221 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10222 		       0ul);
10223 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10224 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10225 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10226 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10227 	}
10228 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10229 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10230 	report_prefix_pop();
10231 
10232 	/* Permit shadowed VMREAD and VMWRITE. */
10233 	report_prefix_push("VMREAD and VMWRITE permission");
10234 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10235 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10236 	if (good_shadow)
10237 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10238 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10239 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10240 	report(c->reason == VMX_VMCALL,
10241 		"shadowed for VMWRITE (in %ld cycles)",
10242 		c->time);
10243 	report(c->flags == flags[ACCESS_VMREAD],
10244 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10245 	       c->flags, flags[ACCESS_VMREAD]);
10246 	if (good_shadow) {
10247 		value = vmread_from_shadow(field);
10248 		report(value == 0,
10249 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10250 		       0ul);
10251 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10252 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10253 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10254 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10255 	}
10256 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10257 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10258 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10259 	       c->time);
10260 	report(c->flags == flags[ACCESS_VMREAD],
10261 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10262 	       c->flags, flags[ACCESS_VMREAD]);
10263 	if (good_shadow)
10264 		report(c->value == 0,
10265 		       "value read from shadow (%lx) is as expected (%lx)",
10266 		       c->value, 0ul);
10267 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10268 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10269 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10270 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10271 	report_prefix_pop();
10272 
10273 out:
10274 	report_prefix_pop();
10275 }
10276 
10277 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
10278 {
10279 	unsigned base;
10280 	unsigned index;
10281 	unsigned bit;
10282 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
10283 
10284 	/* Run test on all possible valid VMCS fields */
10285 	for (base = 0;
10286 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
10287 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
10288 		for (index = 0; index <= highest_index; index++)
10289 			vmcs_shadow_test_field(bitmap, base + index);
10290 
10291 	/*
10292 	 * Run tests on some invalid VMCS fields
10293 	 * (Have reserved bit set).
10294 	 */
10295 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
10296 		vmcs_shadow_test_field(bitmap, (1ull << bit));
10297 }
10298 
10299 static void vmx_vmcs_shadow_test(void)
10300 {
10301 	u8 *bitmap[2];
10302 	struct vmcs *shadow;
10303 
10304 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
10305 		printf("\t'Activate secondary controls' not supported.\n");
10306 		return;
10307 	}
10308 
10309 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
10310 		printf("\t'VMCS shadowing' not supported.\n");
10311 		return;
10312 	}
10313 
10314 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
10315 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
10316 		printf("\tVMWRITE can't modify VM-exit information fields.\n");
10317 		return;
10318 	}
10319 
10320 	test_set_guest(vmx_vmcs_shadow_test_guest);
10321 
10322 	bitmap[ACCESS_VMREAD] = alloc_page();
10323 	bitmap[ACCESS_VMWRITE] = alloc_page();
10324 
10325 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
10326 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
10327 
10328 	shadow = alloc_page();
10329 	shadow->hdr.revision_id = basic.revision;
10330 	shadow->hdr.shadow_vmcs = 1;
10331 	TEST_ASSERT(!vmcs_clear(shadow));
10332 
10333 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
10334 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
10335 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
10336 
10337 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
10338 	report_prefix_push("valid link pointer");
10339 	vmx_vmcs_shadow_test_body(bitmap);
10340 	report_prefix_pop();
10341 
10342 	vmcs_write(VMCS_LINK_PTR, -1ull);
10343 	report_prefix_push("invalid link pointer");
10344 	vmx_vmcs_shadow_test_body(bitmap);
10345 	report_prefix_pop();
10346 
10347 	l1_l2_common.op = ACCESS_NONE;
10348 	enter_guest();
10349 }
10350 
10351 /*
10352  * This test monitors the difference between a guest RDTSC instruction
10353  * and the IA32_TIME_STAMP_COUNTER MSR value stored in the VMCS12
10354  * VM-exit MSR-store list when taking a VM-exit on the instruction
10355  * following RDTSC.
10356  */
10357 #define RDTSC_DIFF_ITERS 100000
10358 #define RDTSC_DIFF_FAILS 100
10359 #define HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD 750
10360 
10361 /*
10362  * Set 'use TSC offsetting' and set the guest offset to the
10363  * inverse of the host's current TSC value, so that the guest starts running
10364  * with an effective TSC value of 0.
10365  */
10366 static void reset_guest_tsc_to_zero(void)
10367 {
10368 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
10369 	vmcs_write(TSC_OFFSET, -rdtsc());
10370 }
10371 
10372 static void rdtsc_vmexit_diff_test_guest(void)
10373 {
10374 	int i;
10375 
10376 	for (i = 0; i < RDTSC_DIFF_ITERS; i++)
10377 		/* Ensure rdtsc is the last instruction before the vmcall. */
10378 		asm volatile("rdtsc; vmcall" : : : "eax", "edx");
10379 }
10380 
10381 /*
10382  * This function only considers the "use TSC offsetting" VM-execution
10383  * control.  It does not handle "use TSC scaling" (because the latter
10384  * isn't available to the host today.)
10385  */
10386 static unsigned long long host_time_to_guest_time(unsigned long long t)
10387 {
10388 	TEST_ASSERT(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
10389 		    !(vmcs_read(CPU_EXEC_CTRL1) & CPU_USE_TSC_SCALING));
10390 
10391 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_USE_TSC_OFFSET)
10392 		t += vmcs_read(TSC_OFFSET);
10393 
10394 	return t;
10395 }
10396 
10397 static unsigned long long rdtsc_vmexit_diff_test_iteration(void)
10398 {
10399 	unsigned long long guest_tsc, host_to_guest_tsc;
10400 
10401 	enter_guest();
10402 	skip_exit_vmcall();
10403 	guest_tsc = (u32) regs.rax + (regs.rdx << 32);
10404 	host_to_guest_tsc = host_time_to_guest_time(exit_msr_store[0].value);
10405 
10406 	return host_to_guest_tsc - guest_tsc;
10407 }
10408 
10409 static void rdtsc_vmexit_diff_test(void)
10410 {
10411 	unsigned long long delta;
10412 	int fail = 0;
10413 	int i;
10414 
10415 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET))
10416 		test_skip("CPU doesn't support the 'use TSC offsetting' processor-based VM-execution control.\n");
10417 
10418 	test_set_guest(rdtsc_vmexit_diff_test_guest);
10419 
10420 	reset_guest_tsc_to_zero();
10421 
10422 	/*
10423 	 * Set up the VMCS12 VM-exit MSR-store list to store just one
10424 	 * MSR: IA32_TIME_STAMP_COUNTER. Note that the value stored is
10425 	 * in the host time domain (i.e., it is not adjusted according
10426 	 * to the TSC multiplier and TSC offset fields in the VMCS12,
10427 	 * as a guest RDTSC would be.)
10428 	 */
10429 	exit_msr_store = alloc_page();
10430 	exit_msr_store[0].index = MSR_IA32_TSC;
10431 	vmcs_write(EXI_MSR_ST_CNT, 1);
10432 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(exit_msr_store));
10433 
10434 	for (i = 0; i < RDTSC_DIFF_ITERS && fail < RDTSC_DIFF_FAILS; i++) {
10435 		delta = rdtsc_vmexit_diff_test_iteration();
10436 		if (delta >= HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD)
10437 			fail++;
10438 	}
10439 
10440 	enter_guest();
10441 
10442 	report(fail < RDTSC_DIFF_FAILS,
10443 	       "RDTSC to VM-exit delta too high in %d of %d iterations, last = %llu",
10444 	       fail, i, delta);
10445 }
10446 
10447 static int invalid_msr_init(struct vmcs *vmcs)
10448 {
10449 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
10450 		printf("\tPreemption timer is not supported\n");
10451 		return VMX_TEST_EXIT;
10452 	}
10453 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
10454 	preempt_val = 10000000;
10455 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
10456 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
10457 
10458 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
10459 		printf("\tSave preemption value is not supported\n");
10460 
10461 	vmcs_write(ENT_MSR_LD_CNT, 1);
10462 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
10463 
10464 	return VMX_TEST_START;
10465 }
10466 
10467 
10468 static void invalid_msr_main(void)
10469 {
10470 	report_fail("Invalid MSR load");
10471 }
10472 
10473 static int invalid_msr_exit_handler(union exit_reason exit_reason)
10474 {
10475 	report_fail("Invalid MSR load");
10476 	print_vmexit_info(exit_reason);
10477 	return VMX_TEST_EXIT;
10478 }
10479 
10480 static int invalid_msr_entry_failure(struct vmentry_result *result)
10481 {
10482 	report(result->exit_reason.failed_vmentry &&
10483 	       result->exit_reason.basic == VMX_FAIL_MSR, "Invalid MSR load");
10484 	return VMX_TEST_VMEXIT;
10485 }
10486 
10487 /*
10488  * The max number of MSRs in an atomic switch MSR list is:
10489  * (111B + 1) * 512 = 4096
10490  *
10491  * Each list entry consumes:
10492  * 4-byte MSR index + 4 bytes reserved + 8-byte data = 16 bytes
10493  *
10494  * Allocate 128 kB to cover max_msr_list_size (i.e., 64 kB) and then some.
10495  */
10496 static const u32 msr_list_page_order = 5;
10497 
10498 static void atomic_switch_msr_limit_test_guest(void)
10499 {
10500 	vmcall();
10501 }
10502 
10503 static void populate_msr_list(struct vmx_msr_entry *msr_list,
10504 			      size_t byte_capacity, int count)
10505 {
10506 	int i;
10507 
10508 	for (i = 0; i < count; i++) {
10509 		msr_list[i].index = MSR_IA32_TSC;
10510 		msr_list[i].reserved = 0;
10511 		msr_list[i].value = 0x1234567890abcdef;
10512 	}
10513 
10514 	memset(msr_list + count, 0xff,
10515 	       byte_capacity - count * sizeof(*msr_list));
10516 }
10517 
10518 static int max_msr_list_size(void)
10519 {
10520 	u32 vmx_misc = rdmsr(MSR_IA32_VMX_MISC);
10521 	u32 factor = ((vmx_misc & GENMASK(27, 25)) >> 25) + 1;
10522 
10523 	return factor * 512;
10524 }
10525 
10526 static void atomic_switch_msrs_test(int count)
10527 {
10528 	struct vmx_msr_entry *vm_enter_load;
10529         struct vmx_msr_entry *vm_exit_load;
10530         struct vmx_msr_entry *vm_exit_store;
10531 	int max_allowed = max_msr_list_size();
10532 	int byte_capacity = 1ul << (msr_list_page_order + PAGE_SHIFT);
10533 	/* Exceeding the max MSR list size at exit trigers KVM to abort. */
10534 	int exit_count = count > max_allowed ? max_allowed : count;
10535 	int cleanup_count = count > max_allowed ? 2 : 1;
10536 	int i;
10537 
10538 	/*
10539 	 * Check for the IA32_TSC MSR,
10540 	 * available with the "TSC flag" and used to populate the MSR lists.
10541 	 */
10542 	if (!(cpuid(1).d & (1 << 4))) {
10543 		report_skip(__func__);
10544 		return;
10545 	}
10546 
10547 	/* Set L2 guest. */
10548 	test_set_guest(atomic_switch_msr_limit_test_guest);
10549 
10550 	/* Setup atomic MSR switch lists. */
10551 	vm_enter_load = alloc_pages(msr_list_page_order);
10552 	vm_exit_load = alloc_pages(msr_list_page_order);
10553 	vm_exit_store = alloc_pages(msr_list_page_order);
10554 
10555 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)vm_enter_load);
10556 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)vm_exit_load);
10557 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)vm_exit_store);
10558 
10559 	/*
10560 	 * VM-Enter should succeed up to the max number of MSRs per list, and
10561 	 * should not consume junk beyond the last entry.
10562 	 */
10563 	populate_msr_list(vm_enter_load, byte_capacity, count);
10564 	populate_msr_list(vm_exit_load, byte_capacity, exit_count);
10565 	populate_msr_list(vm_exit_store, byte_capacity, exit_count);
10566 
10567 	vmcs_write(ENT_MSR_LD_CNT, count);
10568 	vmcs_write(EXI_MSR_LD_CNT, exit_count);
10569 	vmcs_write(EXI_MSR_ST_CNT, exit_count);
10570 
10571 	if (count <= max_allowed) {
10572 		enter_guest();
10573 		assert_exit_reason(VMX_VMCALL);
10574 		skip_exit_vmcall();
10575 	} else {
10576 		u32 exit_qual;
10577 
10578 		test_guest_state("Invalid MSR Load Count", true, count,
10579 				 "ENT_MSR_LD_CNT");
10580 
10581 		exit_qual = vmcs_read(EXI_QUALIFICATION);
10582 		report(exit_qual == max_allowed + 1, "exit_qual, %u, is %u.",
10583 		       exit_qual, max_allowed + 1);
10584 	}
10585 
10586 	/* Cleanup. */
10587 	vmcs_write(ENT_MSR_LD_CNT, 0);
10588 	vmcs_write(EXI_MSR_LD_CNT, 0);
10589 	vmcs_write(EXI_MSR_ST_CNT, 0);
10590 	for (i = 0; i < cleanup_count; i++) {
10591 		enter_guest();
10592 		skip_exit_vmcall();
10593 	}
10594 	free_pages_by_order(vm_enter_load, msr_list_page_order);
10595 	free_pages_by_order(vm_exit_load, msr_list_page_order);
10596 	free_pages_by_order(vm_exit_store, msr_list_page_order);
10597 }
10598 
10599 static void atomic_switch_max_msrs_test(void)
10600 {
10601 	atomic_switch_msrs_test(max_msr_list_size());
10602 }
10603 
10604 static void atomic_switch_overflow_msrs_test(void)
10605 {
10606 	if (test_device_enabled())
10607 		atomic_switch_msrs_test(max_msr_list_size() + 1);
10608 	else
10609 		test_skip("Test is only supported on KVM");
10610 }
10611 
10612 static void vmx_pf_exception_test_guest(void)
10613 {
10614 	ac_test_run(PT_LEVEL_PML4);
10615 }
10616 
10617 static void vmx_pf_exception_test(void)
10618 {
10619 	u64 efer;
10620 	struct cpuid cpuid;
10621 
10622 	test_set_guest(vmx_pf_exception_test_guest);
10623 
10624 	enter_guest();
10625 
10626 	while (vmcs_read(EXI_REASON) != VMX_VMCALL) {
10627 		switch (vmcs_read(EXI_REASON)) {
10628 		case VMX_RDMSR:
10629 			assert(regs.rcx == MSR_EFER);
10630 			efer = vmcs_read(GUEST_EFER);
10631 			regs.rdx = efer >> 32;
10632 			regs.rax = efer & 0xffffffff;
10633 			break;
10634 		case VMX_WRMSR:
10635 			assert(regs.rcx == MSR_EFER);
10636 			efer = regs.rdx << 32 | (regs.rax & 0xffffffff);
10637 			vmcs_write(GUEST_EFER, efer);
10638 			break;
10639 		case VMX_CPUID:
10640 			cpuid = (struct cpuid) {0, 0, 0, 0};
10641 			cpuid = raw_cpuid(regs.rax, regs.rcx);
10642 			regs.rax = cpuid.a;
10643 			regs.rbx = cpuid.b;
10644 			regs.rcx = cpuid.c;
10645 			regs.rdx = cpuid.d;
10646 			break;
10647 		default:
10648 			assert_msg(false,
10649 				"Unexpected exit to L1, exit_reason: %s (0x%lx)",
10650 				exit_reason_description(vmcs_read(EXI_REASON)),
10651 				vmcs_read(EXI_REASON));
10652 		}
10653 		skip_exit_insn();
10654 		enter_guest();
10655 	}
10656 
10657 	assert_exit_reason(VMX_VMCALL);
10658 }
10659 #define TEST(name) { #name, .v2 = name }
10660 
10661 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
10662 struct vmx_test vmx_tests[] = {
10663 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
10664 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
10665 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
10666 		preemption_timer_exit_handler, NULL, {0} },
10667 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
10668 		test_ctrl_pat_exit_handler, NULL, {0} },
10669 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
10670 		test_ctrl_efer_exit_handler, NULL, {0} },
10671 	{ "CR shadowing", NULL, cr_shadowing_main,
10672 		cr_shadowing_exit_handler, NULL, {0} },
10673 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
10674 		NULL, {0} },
10675 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
10676 		insn_intercept_exit_handler, NULL, {0} },
10677 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
10678 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
10679 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
10680 	{ "interrupt", interrupt_init, interrupt_main,
10681 		interrupt_exit_handler, NULL, {0} },
10682 	{ "nmi_hlt", nmi_hlt_init, nmi_hlt_main,
10683 		nmi_hlt_exit_handler, NULL, {0} },
10684 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
10685 		NULL, {0} },
10686 	{ "MSR switch", msr_switch_init, msr_switch_main,
10687 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
10688 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
10689 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
10690 		disable_rdtscp_exit_handler, NULL, {0} },
10691 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
10692 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
10693 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
10694 		exit_monitor_from_l2_handler, NULL, {0} },
10695 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
10696 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
10697 	/* Basic V2 tests. */
10698 	TEST(v2_null_test),
10699 	TEST(v2_multiple_entries_test),
10700 	TEST(fixture_test_case1),
10701 	TEST(fixture_test_case2),
10702 	/* Opcode tests. */
10703 	TEST(invvpid_test),
10704 	/* VM-entry tests */
10705 	TEST(vmx_controls_test),
10706 	TEST(vmx_host_state_area_test),
10707 	TEST(vmx_guest_state_area_test),
10708 	TEST(vmentry_movss_shadow_test),
10709 	TEST(vmentry_unrestricted_guest_test),
10710 	/* APICv tests */
10711 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
10712 	TEST(vmx_hlt_with_rvi_test),
10713 	TEST(apic_reg_virt_test),
10714 	TEST(virt_x2apic_mode_test),
10715 	/* APIC pass-through tests */
10716 	TEST(vmx_apic_passthrough_test),
10717 	TEST(vmx_apic_passthrough_thread_test),
10718 	TEST(vmx_apic_passthrough_tpr_threshold_test),
10719 	TEST(vmx_init_signal_test),
10720 	TEST(vmx_sipi_signal_test),
10721 	/* VMCS Shadowing tests */
10722 	TEST(vmx_vmcs_shadow_test),
10723 	/* Regression tests */
10724 	TEST(vmx_ldtr_test),
10725 	TEST(vmx_cr_load_test),
10726 	TEST(vmx_cr4_osxsave_test),
10727 	TEST(vmx_nm_test),
10728 	TEST(vmx_db_test),
10729 	TEST(vmx_nmi_window_test),
10730 	TEST(vmx_intr_window_test),
10731 	TEST(vmx_pending_event_test),
10732 	TEST(vmx_pending_event_hlt_test),
10733 	TEST(vmx_store_tsc_test),
10734 	TEST(vmx_preemption_timer_zero_test),
10735 	TEST(vmx_preemption_timer_tf_test),
10736 	TEST(vmx_preemption_timer_expiry_test),
10737 	/* EPT access tests. */
10738 	TEST(ept_access_test_not_present),
10739 	TEST(ept_access_test_read_only),
10740 	TEST(ept_access_test_write_only),
10741 	TEST(ept_access_test_read_write),
10742 	TEST(ept_access_test_execute_only),
10743 	TEST(ept_access_test_read_execute),
10744 	TEST(ept_access_test_write_execute),
10745 	TEST(ept_access_test_read_write_execute),
10746 	TEST(ept_access_test_reserved_bits),
10747 	TEST(ept_access_test_ignored_bits),
10748 	TEST(ept_access_test_paddr_not_present_ad_disabled),
10749 	TEST(ept_access_test_paddr_not_present_ad_enabled),
10750 	TEST(ept_access_test_paddr_read_only_ad_disabled),
10751 	TEST(ept_access_test_paddr_read_only_ad_enabled),
10752 	TEST(ept_access_test_paddr_read_write),
10753 	TEST(ept_access_test_paddr_read_write_execute),
10754 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
10755 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
10756 	TEST(ept_access_test_paddr_not_present_page_fault),
10757 	TEST(ept_access_test_force_2m_page),
10758 	/* Atomic MSR switch tests. */
10759 	TEST(atomic_switch_max_msrs_test),
10760 	TEST(atomic_switch_overflow_msrs_test),
10761 	TEST(rdtsc_vmexit_diff_test),
10762 	TEST(vmx_mtf_test),
10763 	TEST(vmx_mtf_pdpte_test),
10764 	TEST(vmx_pf_exception_test),
10765 	{ NULL, NULL, NULL, NULL, NULL, {0} },
10766 };
10767