xref: /kvm-unit-tests/x86/vmx_tests.c (revision c986dbe8670535e4f88871b1e8b8480bdc256ded)
1 /*
2  * All test cases of nested virtualization should be in this file
3  *
4  * Author : Arthur Chunqi Li <yzt356@gmail.com>
5  */
6 
7 #include <asm/debugreg.h>
8 
9 #include "vmx.h"
10 #include "msr.h"
11 #include "processor.h"
12 #include "vm.h"
13 #include "pci.h"
14 #include "fwcfg.h"
15 #include "isr.h"
16 #include "desc.h"
17 #include "apic.h"
18 #include "types.h"
19 #include "vmalloc.h"
20 #include "alloc_page.h"
21 #include "smp.h"
22 #include "delay.h"
23 
24 #define VPID_CAP_INVVPID_TYPES_SHIFT 40
25 
26 u64 ia32_pat;
27 u64 ia32_efer;
28 void *io_bitmap_a, *io_bitmap_b;
29 u16 ioport;
30 
31 unsigned long *pml4;
32 u64 eptp;
33 void *data_page1, *data_page2;
34 
35 phys_addr_t pci_physaddr;
36 
37 void *pml_log;
38 #define PML_INDEX 512
39 
40 static inline unsigned ffs(unsigned x)
41 {
42 	int pos = -1;
43 
44 	__asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0"
45 			     : "+r"(pos) : "rm"(x) : "eax");
46 	return pos + 1;
47 }
48 
49 static inline void vmcall(void)
50 {
51 	asm volatile("vmcall");
52 }
53 
54 static void basic_guest_main(void)
55 {
56 	report(1, "Basic VMX test");
57 }
58 
59 static int basic_exit_handler(union exit_reason exit_reason)
60 {
61 	report(0, "Basic VMX test");
62 	print_vmexit_info(exit_reason);
63 	return VMX_TEST_EXIT;
64 }
65 
66 static void vmenter_main(void)
67 {
68 	u64 rax;
69 	u64 rsp, resume_rsp;
70 
71 	report(1, "test vmlaunch");
72 
73 	asm volatile(
74 		"mov %%rsp, %0\n\t"
75 		"mov %3, %%rax\n\t"
76 		"vmcall\n\t"
77 		"mov %%rax, %1\n\t"
78 		"mov %%rsp, %2\n\t"
79 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
80 		: "g"(0xABCD));
81 	report((rax == 0xFFFF) && (rsp == resume_rsp), "test vmresume");
82 }
83 
84 static int vmenter_exit_handler(union exit_reason exit_reason)
85 {
86 	u64 guest_rip = vmcs_read(GUEST_RIP);
87 
88 	switch (exit_reason.basic) {
89 	case VMX_VMCALL:
90 		if (regs.rax != 0xABCD) {
91 			report(0, "test vmresume");
92 			return VMX_TEST_VMEXIT;
93 		}
94 		regs.rax = 0xFFFF;
95 		vmcs_write(GUEST_RIP, guest_rip + 3);
96 		return VMX_TEST_RESUME;
97 	default:
98 		report(0, "test vmresume");
99 		print_vmexit_info(exit_reason);
100 	}
101 	return VMX_TEST_VMEXIT;
102 }
103 
104 u32 preempt_scale;
105 volatile unsigned long long tsc_val;
106 volatile u32 preempt_val;
107 u64 saved_rip;
108 
109 static int preemption_timer_init(struct vmcs *vmcs)
110 {
111 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
112 		printf("\tPreemption timer is not supported\n");
113 		return VMX_TEST_EXIT;
114 	}
115 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
116 	preempt_val = 10000000;
117 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
118 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
119 
120 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
121 		printf("\tSave preemption value is not supported\n");
122 
123 	return VMX_TEST_START;
124 }
125 
126 static void preemption_timer_main(void)
127 {
128 	tsc_val = rdtsc();
129 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
130 		vmx_set_test_stage(0);
131 		vmcall();
132 		if (vmx_get_test_stage() == 1)
133 			vmcall();
134 	}
135 	vmx_set_test_stage(1);
136 	while (vmx_get_test_stage() == 1) {
137 		if (((rdtsc() - tsc_val) >> preempt_scale)
138 				> 10 * preempt_val) {
139 			vmx_set_test_stage(2);
140 			vmcall();
141 		}
142 	}
143 	tsc_val = rdtsc();
144 	asm volatile ("hlt");
145 	vmcall();
146 	vmx_set_test_stage(5);
147 	vmcall();
148 }
149 
150 static int preemption_timer_exit_handler(union exit_reason exit_reason)
151 {
152 	bool guest_halted;
153 	u64 guest_rip;
154 	u32 insn_len;
155 	u32 ctrl_exit;
156 
157 	guest_rip = vmcs_read(GUEST_RIP);
158 	insn_len = vmcs_read(EXI_INST_LEN);
159 	switch (exit_reason.basic) {
160 	case VMX_PREEMPT:
161 		switch (vmx_get_test_stage()) {
162 		case 1:
163 		case 2:
164 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val,
165 			       "busy-wait for preemption timer");
166 			vmx_set_test_stage(3);
167 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
168 			return VMX_TEST_RESUME;
169 		case 3:
170 			guest_halted =
171 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
172 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val
173 			        && guest_halted,
174 			       "preemption timer during hlt");
175 			vmx_set_test_stage(4);
176 			vmcs_write(PIN_CONTROLS,
177 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
178 			vmcs_write(EXI_CONTROLS,
179 				   vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT);
180 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
181 			return VMX_TEST_RESUME;
182 		case 4:
183 			report(saved_rip == guest_rip,
184 			       "preemption timer with 0 value");
185 			break;
186 		default:
187 			report(false, "Invalid stage.");
188 			print_vmexit_info(exit_reason);
189 			break;
190 		}
191 		break;
192 	case VMX_VMCALL:
193 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
194 		switch (vmx_get_test_stage()) {
195 		case 0:
196 			report(vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val,
197 			       "Keep preemption value");
198 			vmx_set_test_stage(1);
199 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
200 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
201 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
202 			vmcs_write(EXI_CONTROLS, ctrl_exit);
203 			return VMX_TEST_RESUME;
204 		case 1:
205 			report(vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val,
206 			       "Save preemption value");
207 			return VMX_TEST_RESUME;
208 		case 2:
209 			report(0, "busy-wait for preemption timer");
210 			vmx_set_test_stage(3);
211 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
212 			return VMX_TEST_RESUME;
213 		case 3:
214 			report(0, "preemption timer during hlt");
215 			vmx_set_test_stage(4);
216 			/* fall through */
217 		case 4:
218 			vmcs_write(PIN_CONTROLS,
219 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
220 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
221 			saved_rip = guest_rip + insn_len;
222 			return VMX_TEST_RESUME;
223 		case 5:
224 			report(0,
225 			       "preemption timer with 0 value (vmcall stage 5)");
226 			break;
227 		default:
228 			// Should not reach here
229 			report(false, "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(false, "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(0, "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(0, "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(0, "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(0, "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(0, "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(0, "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(0, "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(0, "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(false, "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(0, "Read shadowing CR0");
578 			vmx_inc_test_stage();
579 			break;
580 		case 5:
581 			report(0, "Read shadowing CR4");
582 			vmx_inc_test_stage();
583 			break;
584 		case 6:
585 			report(0, "Write shadowing CR0 (same value)");
586 			vmx_inc_test_stage();
587 			break;
588 		case 7:
589 			report(0, "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(false, "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(false, "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(false, "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(false, "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 gdt64_desc;ret\n\t"
802 	"insn_lgdt: lgdt gdt64_desc;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 	if (setup_ept(have_ad))
1152 		return VMX_TEST_EXIT;
1153 	data_page1 = alloc_page();
1154 	data_page2 = alloc_page();
1155 	*((u32 *)data_page1) = MAGIC_VAL_1;
1156 	*((u32 *)data_page2) = MAGIC_VAL_2;
1157 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1158 			EPT_RA | EPT_WA | EPT_EA);
1159 
1160 	apic_version = apic_read(APIC_LVR);
1161 
1162 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1163 	if (ret != PCIDEVADDR_INVALID) {
1164 		pci_dev_init(&pcidev, ret);
1165 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1166 	}
1167 
1168 	return VMX_TEST_START;
1169 }
1170 
1171 static int ept_init(struct vmcs *vmcs)
1172 {
1173 	return ept_init_common(false);
1174 }
1175 
1176 static void ept_common(void)
1177 {
1178 	vmx_set_test_stage(0);
1179 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1180 			*((u32 *)data_page1) != MAGIC_VAL_1)
1181 		report(0, "EPT basic framework - read");
1182 	else {
1183 		*((u32 *)data_page2) = MAGIC_VAL_3;
1184 		vmcall();
1185 		if (vmx_get_test_stage() == 1) {
1186 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1187 					*((u32 *)data_page2) == MAGIC_VAL_2)
1188 				report(1, "EPT basic framework");
1189 			else
1190 				report(1, "EPT basic framework - remap");
1191 		}
1192 	}
1193 	// Test EPT Misconfigurations
1194 	vmx_set_test_stage(1);
1195 	vmcall();
1196 	*((u32 *)data_page1) = MAGIC_VAL_1;
1197 	if (vmx_get_test_stage() != 2) {
1198 		report(0, "EPT misconfigurations");
1199 		goto t1;
1200 	}
1201 	vmx_set_test_stage(2);
1202 	vmcall();
1203 	*((u32 *)data_page1) = MAGIC_VAL_1;
1204 	report(vmx_get_test_stage() == 3, "EPT misconfigurations");
1205 t1:
1206 	// Test EPT violation
1207 	vmx_set_test_stage(3);
1208 	vmcall();
1209 	*((u32 *)data_page1) = MAGIC_VAL_1;
1210 	report(vmx_get_test_stage() == 4, "EPT violation - page permission");
1211 	// Violation caused by EPT paging structure
1212 	vmx_set_test_stage(4);
1213 	vmcall();
1214 	*((u32 *)data_page1) = MAGIC_VAL_2;
1215 	report(vmx_get_test_stage() == 5, "EPT violation - paging structure");
1216 
1217 	// MMIO Read/Write
1218 	vmx_set_test_stage(5);
1219 	vmcall();
1220 
1221 	*(u32 volatile *)pci_physaddr;
1222 	report(vmx_get_test_stage() == 6, "MMIO EPT violation - read");
1223 
1224 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1225 	report(vmx_get_test_stage() == 7, "MMIO EPT violation - write");
1226 }
1227 
1228 static void ept_main(void)
1229 {
1230 	ept_common();
1231 
1232 	// Test EPT access to L1 MMIO
1233 	vmx_set_test_stage(7);
1234 	report(*((u32 *)0xfee00030UL) == apic_version, "EPT - MMIO access");
1235 
1236 	// Test invalid operand for INVEPT
1237 	vmcall();
1238 	report(vmx_get_test_stage() == 8, "EPT - unsupported INVEPT");
1239 }
1240 
1241 static bool invept_test(int type, u64 eptp)
1242 {
1243 	bool ret, supported;
1244 
1245 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1246 	ret = invept(type, eptp);
1247 
1248 	if (ret == !supported)
1249 		return false;
1250 
1251 	if (!supported)
1252 		printf("WARNING: unsupported invept passed!\n");
1253 	else
1254 		printf("WARNING: invept failed!\n");
1255 
1256 	return true;
1257 }
1258 
1259 static int pml_exit_handler(union exit_reason exit_reason)
1260 {
1261 	u16 index, count;
1262 	u64 *pmlbuf = pml_log;
1263 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1264 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1265 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1266 
1267 	switch (exit_reason.basic) {
1268 	case VMX_VMCALL:
1269 		switch (vmx_get_test_stage()) {
1270 		case 0:
1271 			index = vmcs_read(GUEST_PML_INDEX);
1272 			for (count = index + 1; count < PML_INDEX; count++) {
1273 				if (pmlbuf[count] == (u64)data_page2) {
1274 					vmx_inc_test_stage();
1275 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1276 					break;
1277 				}
1278 			}
1279 			break;
1280 		case 1:
1281 			index = vmcs_read(GUEST_PML_INDEX);
1282 			/* Keep clearing the dirty bit till a overflow */
1283 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1284 			break;
1285 		default:
1286 			report(false, "unexpected stage, %d.",
1287 			       vmx_get_test_stage());
1288 			print_vmexit_info(exit_reason);
1289 			return VMX_TEST_VMEXIT;
1290 		}
1291 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1292 		return VMX_TEST_RESUME;
1293 	case VMX_PML_FULL:
1294 		vmx_inc_test_stage();
1295 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1296 		return VMX_TEST_RESUME;
1297 	default:
1298 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
1299 		print_vmexit_info(exit_reason);
1300 	}
1301 	return VMX_TEST_VMEXIT;
1302 }
1303 
1304 static int ept_exit_handler_common(union exit_reason exit_reason, bool have_ad)
1305 {
1306 	u64 guest_rip;
1307 	u64 guest_cr3;
1308 	u32 insn_len;
1309 	u32 exit_qual;
1310 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte,
1311 			     guest_pte_addr;
1312 
1313 	guest_rip = vmcs_read(GUEST_RIP);
1314 	guest_cr3 = vmcs_read(GUEST_CR3);
1315 	insn_len = vmcs_read(EXI_INST_LEN);
1316 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1317 	pteval_t *ptep;
1318 	switch (exit_reason.basic) {
1319 	case VMX_VMCALL:
1320 		switch (vmx_get_test_stage()) {
1321 		case 0:
1322 			check_ept_ad(pml4, guest_cr3,
1323 				     (unsigned long)data_page1,
1324 				     have_ad ? EPT_ACCESS_FLAG : 0,
1325 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1326 			check_ept_ad(pml4, guest_cr3,
1327 				     (unsigned long)data_page2,
1328 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1329 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1330 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1331 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1332 			if (have_ad)
1333 				ept_sync(INVEPT_SINGLE, eptp);;
1334 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1335 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1336 				vmx_inc_test_stage();
1337 				install_ept(pml4, (unsigned long)data_page2,
1338 						(unsigned long)data_page2,
1339 						EPT_RA | EPT_WA | EPT_EA);
1340 			} else
1341 				report(0, "EPT basic framework - write");
1342 			break;
1343 		case 1:
1344 			install_ept(pml4, (unsigned long)data_page1,
1345  				(unsigned long)data_page1, EPT_WA);
1346 			ept_sync(INVEPT_SINGLE, eptp);
1347 			break;
1348 		case 2:
1349 			install_ept(pml4, (unsigned long)data_page1,
1350  				(unsigned long)data_page1,
1351  				EPT_RA | EPT_WA | EPT_EA |
1352  				(2 << EPT_MEM_TYPE_SHIFT));
1353 			ept_sync(INVEPT_SINGLE, eptp);
1354 			break;
1355 		case 3:
1356 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1357 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1358 						1, &data_page1_pte));
1359 			set_ept_pte(pml4, (unsigned long)data_page1,
1360 				1, data_page1_pte & ~EPT_PRESENT);
1361 			ept_sync(INVEPT_SINGLE, eptp);
1362 			break;
1363 		case 4:
1364 			ptep = get_pte_level((pgd_t *)guest_cr3, data_page1, /*level=*/2);
1365 			guest_pte_addr = virt_to_phys(ptep) & PAGE_MASK;
1366 
1367 			TEST_ASSERT(get_ept_pte(pml4, guest_pte_addr, 2, &data_page1_pte_pte));
1368 			set_ept_pte(pml4, guest_pte_addr, 2,
1369 				data_page1_pte_pte & ~EPT_PRESENT);
1370 			ept_sync(INVEPT_SINGLE, eptp);
1371 			break;
1372 		case 5:
1373 			install_ept(pml4, (unsigned long)pci_physaddr,
1374 				(unsigned long)pci_physaddr, 0);
1375 			ept_sync(INVEPT_SINGLE, eptp);
1376 			break;
1377 		case 7:
1378 			if (!invept_test(0, eptp))
1379 				vmx_inc_test_stage();
1380 			break;
1381 		// Should not reach here
1382 		default:
1383 			report(false, "ERROR - unexpected stage, %d.",
1384 			       vmx_get_test_stage());
1385 			print_vmexit_info(exit_reason);
1386 			return VMX_TEST_VMEXIT;
1387 		}
1388 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1389 		return VMX_TEST_RESUME;
1390 	case VMX_EPT_MISCONFIG:
1391 		switch (vmx_get_test_stage()) {
1392 		case 1:
1393 		case 2:
1394 			vmx_inc_test_stage();
1395 			install_ept(pml4, (unsigned long)data_page1,
1396  				(unsigned long)data_page1,
1397  				EPT_RA | EPT_WA | EPT_EA);
1398 			ept_sync(INVEPT_SINGLE, eptp);
1399 			break;
1400 		// Should not reach here
1401 		default:
1402 			report(false, "ERROR - unexpected stage, %d.",
1403 			       vmx_get_test_stage());
1404 			print_vmexit_info(exit_reason);
1405 			return VMX_TEST_VMEXIT;
1406 		}
1407 		return VMX_TEST_RESUME;
1408 	case VMX_EPT_VIOLATION:
1409 		/*
1410 		 * Exit-qualifications are masked not to account for advanced
1411 		 * VM-exit information. Once KVM supports this feature, this
1412 		 * masking should be removed.
1413 		 */
1414 		exit_qual &= ~EPT_VLT_GUEST_MASK;
1415 
1416 		switch(vmx_get_test_stage()) {
1417 		case 3:
1418 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1419 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1420 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1421 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1422 					EPT_VLT_PADDR))
1423 				vmx_inc_test_stage();
1424 			set_ept_pte(pml4, (unsigned long)data_page1,
1425 				1, data_page1_pte | (EPT_PRESENT));
1426 			ept_sync(INVEPT_SINGLE, eptp);
1427 			break;
1428 		case 4:
1429 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1430 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1431 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1432 			if (exit_qual == (EPT_VLT_RD |
1433 					  (have_ad ? EPT_VLT_WR : 0) |
1434 					  EPT_VLT_LADDR_VLD))
1435 				vmx_inc_test_stage();
1436 			set_ept_pte(pml4, guest_pte_addr, 2,
1437 				data_page1_pte_pte | (EPT_PRESENT));
1438 			ept_sync(INVEPT_SINGLE, eptp);
1439 			break;
1440 		case 5:
1441 			if (exit_qual & EPT_VLT_RD)
1442 				vmx_inc_test_stage();
1443 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1444 						1, &memaddr_pte));
1445 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1446 			ept_sync(INVEPT_SINGLE, eptp);
1447 			break;
1448 		case 6:
1449 			if (exit_qual & EPT_VLT_WR)
1450 				vmx_inc_test_stage();
1451 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1452 						1, &memaddr_pte));
1453 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1454 			ept_sync(INVEPT_SINGLE, eptp);
1455 			break;
1456 		default:
1457 			// Should not reach here
1458 			report(false, "ERROR : unexpected stage, %d",
1459 			       vmx_get_test_stage());
1460 			print_vmexit_info(exit_reason);
1461 			return VMX_TEST_VMEXIT;
1462 		}
1463 		return VMX_TEST_RESUME;
1464 	default:
1465 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
1466 		print_vmexit_info(exit_reason);
1467 	}
1468 	return VMX_TEST_VMEXIT;
1469 }
1470 
1471 static int ept_exit_handler(union exit_reason exit_reason)
1472 {
1473 	return ept_exit_handler_common(exit_reason, false);
1474 }
1475 
1476 static int eptad_init(struct vmcs *vmcs)
1477 {
1478 	int r = ept_init_common(true);
1479 
1480 	if (r == VMX_TEST_EXIT)
1481 		return r;
1482 
1483 	if ((rdmsr(MSR_IA32_VMX_EPT_VPID_CAP) & EPT_CAP_AD_FLAG) == 0) {
1484 		printf("\tEPT A/D bits are not supported");
1485 		return VMX_TEST_EXIT;
1486 	}
1487 
1488 	return r;
1489 }
1490 
1491 static int pml_init(struct vmcs *vmcs)
1492 {
1493 	u32 ctrl_cpu;
1494 	int r = eptad_init(vmcs);
1495 
1496 	if (r == VMX_TEST_EXIT)
1497 		return r;
1498 
1499 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1500 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1501 		printf("\tPML is not supported");
1502 		return VMX_TEST_EXIT;
1503 	}
1504 
1505 	pml_log = alloc_page();
1506 	vmcs_write(PMLADDR, (u64)pml_log);
1507 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1508 
1509 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1510 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1511 
1512 	return VMX_TEST_START;
1513 }
1514 
1515 static void pml_main(void)
1516 {
1517 	int count = 0;
1518 
1519 	vmx_set_test_stage(0);
1520 	*((u32 *)data_page2) = 0x1;
1521 	vmcall();
1522 	report(vmx_get_test_stage() == 1, "PML - Dirty GPA Logging");
1523 
1524 	while (vmx_get_test_stage() == 1) {
1525 		vmcall();
1526 		*((u32 *)data_page2) = 0x1;
1527 		if (count++ > PML_INDEX)
1528 			break;
1529 	}
1530 	report(vmx_get_test_stage() == 2, "PML Full Event");
1531 }
1532 
1533 static void eptad_main(void)
1534 {
1535 	ept_common();
1536 }
1537 
1538 static int eptad_exit_handler(union exit_reason exit_reason)
1539 {
1540 	return ept_exit_handler_common(exit_reason, true);
1541 }
1542 
1543 static bool invvpid_test(int type, u16 vpid)
1544 {
1545 	bool ret, supported;
1546 
1547 	supported = ept_vpid.val &
1548 		(VPID_CAP_INVVPID_ADDR >> INVVPID_ADDR << type);
1549 	ret = invvpid(type, vpid, 0);
1550 
1551 	if (ret == !supported)
1552 		return false;
1553 
1554 	if (!supported)
1555 		printf("WARNING: unsupported invvpid passed!\n");
1556 	else
1557 		printf("WARNING: invvpid failed!\n");
1558 
1559 	return true;
1560 }
1561 
1562 static int vpid_init(struct vmcs *vmcs)
1563 {
1564 	u32 ctrl_cpu1;
1565 
1566 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1567 		!(ctrl_cpu_rev[1].clr & CPU_VPID)) {
1568 		printf("\tVPID is not supported");
1569 		return VMX_TEST_EXIT;
1570 	}
1571 
1572 	ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
1573 	ctrl_cpu1 |= CPU_VPID;
1574 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
1575 	return VMX_TEST_START;
1576 }
1577 
1578 static void vpid_main(void)
1579 {
1580 	vmx_set_test_stage(0);
1581 	vmcall();
1582 	report(vmx_get_test_stage() == 1, "INVVPID SINGLE ADDRESS");
1583 	vmx_set_test_stage(2);
1584 	vmcall();
1585 	report(vmx_get_test_stage() == 3, "INVVPID SINGLE");
1586 	vmx_set_test_stage(4);
1587 	vmcall();
1588 	report(vmx_get_test_stage() == 5, "INVVPID ALL");
1589 }
1590 
1591 static int vpid_exit_handler(union exit_reason exit_reason)
1592 {
1593 	u64 guest_rip;
1594 	u32 insn_len;
1595 
1596 	guest_rip = vmcs_read(GUEST_RIP);
1597 	insn_len = vmcs_read(EXI_INST_LEN);
1598 
1599 	switch (exit_reason.basic) {
1600 	case VMX_VMCALL:
1601 		switch(vmx_get_test_stage()) {
1602 		case 0:
1603 			if (!invvpid_test(INVVPID_ADDR, 1))
1604 				vmx_inc_test_stage();
1605 			break;
1606 		case 2:
1607 			if (!invvpid_test(INVVPID_CONTEXT_GLOBAL, 1))
1608 				vmx_inc_test_stage();
1609 			break;
1610 		case 4:
1611 			if (!invvpid_test(INVVPID_ALL, 1))
1612 				vmx_inc_test_stage();
1613 			break;
1614 		default:
1615 			report(false, "ERROR: unexpected stage, %d",
1616 					vmx_get_test_stage());
1617 			print_vmexit_info(exit_reason);
1618 			return VMX_TEST_VMEXIT;
1619 		}
1620 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1621 		return VMX_TEST_RESUME;
1622 	default:
1623 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
1624 		print_vmexit_info(exit_reason);
1625 	}
1626 	return VMX_TEST_VMEXIT;
1627 }
1628 
1629 #define TIMER_VECTOR	222
1630 
1631 static volatile bool timer_fired;
1632 
1633 static void timer_isr(isr_regs_t *regs)
1634 {
1635 	timer_fired = true;
1636 	apic_write(APIC_EOI, 0);
1637 }
1638 
1639 static int interrupt_init(struct vmcs *vmcs)
1640 {
1641 	msr_bmp_init();
1642 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1643 	handle_irq(TIMER_VECTOR, timer_isr);
1644 	return VMX_TEST_START;
1645 }
1646 
1647 static void interrupt_main(void)
1648 {
1649 	long long start, loops;
1650 
1651 	vmx_set_test_stage(0);
1652 
1653 	apic_write(APIC_LVTT, TIMER_VECTOR);
1654 	irq_enable();
1655 
1656 	apic_write(APIC_TMICT, 1);
1657 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1658 		asm volatile ("nop");
1659 	report(timer_fired, "direct interrupt while running guest");
1660 
1661 	apic_write(APIC_TMICT, 0);
1662 	irq_disable();
1663 	vmcall();
1664 	timer_fired = false;
1665 	apic_write(APIC_TMICT, 1);
1666 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1667 		asm volatile ("nop");
1668 	report(timer_fired, "intercepted interrupt while running guest");
1669 
1670 	irq_enable();
1671 	apic_write(APIC_TMICT, 0);
1672 	irq_disable();
1673 	vmcall();
1674 	timer_fired = false;
1675 	start = rdtsc();
1676 	apic_write(APIC_TMICT, 1000000);
1677 
1678 	asm volatile ("sti; hlt");
1679 
1680 	report(rdtsc() - start > 1000000 && timer_fired,
1681 	       "direct interrupt + hlt");
1682 
1683 	apic_write(APIC_TMICT, 0);
1684 	irq_disable();
1685 	vmcall();
1686 	timer_fired = false;
1687 	start = rdtsc();
1688 	apic_write(APIC_TMICT, 1000000);
1689 
1690 	asm volatile ("sti; hlt");
1691 
1692 	report(rdtsc() - start > 10000 && timer_fired,
1693 	       "intercepted interrupt + hlt");
1694 
1695 	apic_write(APIC_TMICT, 0);
1696 	irq_disable();
1697 	vmcall();
1698 	timer_fired = false;
1699 	start = rdtsc();
1700 	apic_write(APIC_TMICT, 1000000);
1701 
1702 	irq_enable();
1703 	asm volatile ("nop");
1704 	vmcall();
1705 
1706 	report(rdtsc() - start > 10000 && timer_fired,
1707 	       "direct interrupt + activity state hlt");
1708 
1709 	apic_write(APIC_TMICT, 0);
1710 	irq_disable();
1711 	vmcall();
1712 	timer_fired = false;
1713 	start = rdtsc();
1714 	apic_write(APIC_TMICT, 1000000);
1715 
1716 	irq_enable();
1717 	asm volatile ("nop");
1718 	vmcall();
1719 
1720 	report(rdtsc() - start > 10000 && timer_fired,
1721 	       "intercepted interrupt + activity state hlt");
1722 
1723 	apic_write(APIC_TMICT, 0);
1724 	irq_disable();
1725 	vmx_set_test_stage(7);
1726 	vmcall();
1727 	timer_fired = false;
1728 	apic_write(APIC_TMICT, 1);
1729 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1730 		asm volatile ("nop");
1731 	report(timer_fired,
1732 	       "running a guest with interrupt acknowledgement set");
1733 
1734 	apic_write(APIC_TMICT, 0);
1735 	irq_enable();
1736 	timer_fired = false;
1737 	vmcall();
1738 	report(timer_fired, "Inject an event to a halted guest");
1739 }
1740 
1741 static int interrupt_exit_handler(union exit_reason exit_reason)
1742 {
1743 	u64 guest_rip = vmcs_read(GUEST_RIP);
1744 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1745 
1746 	switch (exit_reason.basic) {
1747 	case VMX_VMCALL:
1748 		switch (vmx_get_test_stage()) {
1749 		case 0:
1750 		case 2:
1751 		case 5:
1752 			vmcs_write(PIN_CONTROLS,
1753 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1754 			break;
1755 		case 7:
1756 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1757 			vmcs_write(PIN_CONTROLS,
1758 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1759 			break;
1760 		case 1:
1761 		case 3:
1762 			vmcs_write(PIN_CONTROLS,
1763 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1764 			break;
1765 		case 4:
1766 		case 6:
1767 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1768 			break;
1769 
1770 		case 8:
1771 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1772 			vmcs_write(ENT_INTR_INFO,
1773 				   TIMER_VECTOR |
1774 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1775 				   INTR_INFO_VALID_MASK);
1776 			break;
1777 		}
1778 		vmx_inc_test_stage();
1779 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1780 		return VMX_TEST_RESUME;
1781 	case VMX_EXTINT:
1782 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1783 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1784 			handle_external_interrupt(vector);
1785 		} else {
1786 			irq_enable();
1787 			asm volatile ("nop");
1788 			irq_disable();
1789 		}
1790 		if (vmx_get_test_stage() >= 2)
1791 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1792 		return VMX_TEST_RESUME;
1793 	default:
1794 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
1795 		print_vmexit_info(exit_reason);
1796 	}
1797 
1798 	return VMX_TEST_VMEXIT;
1799 }
1800 
1801 
1802 static volatile int nmi_fired;
1803 
1804 #define NMI_DELAY 100000000ULL
1805 
1806 static void nmi_isr(isr_regs_t *regs)
1807 {
1808 	nmi_fired = true;
1809 }
1810 
1811 static int nmi_hlt_init(struct vmcs *vmcs)
1812 {
1813 	msr_bmp_init();
1814 	handle_irq(NMI_VECTOR, nmi_isr);
1815 	vmcs_write(PIN_CONTROLS,
1816 		   vmcs_read(PIN_CONTROLS) & ~PIN_NMI);
1817 	vmcs_write(PIN_CONTROLS,
1818 		   vmcs_read(PIN_CONTROLS) & ~PIN_VIRT_NMI);
1819 	return VMX_TEST_START;
1820 }
1821 
1822 static void nmi_message_thread(void *data)
1823 {
1824     while (vmx_get_test_stage() != 1)
1825         pause();
1826 
1827     delay(NMI_DELAY);
1828     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1829 
1830     while (vmx_get_test_stage() != 2)
1831         pause();
1832 
1833     delay(NMI_DELAY);
1834     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1835 }
1836 
1837 static void nmi_hlt_main(void)
1838 {
1839     long long start;
1840 
1841     if (cpu_count() < 2) {
1842         report_skip(__func__);
1843         vmx_set_test_stage(-1);
1844         return;
1845     }
1846 
1847     vmx_set_test_stage(0);
1848     on_cpu_async(1, nmi_message_thread, NULL);
1849     start = rdtsc();
1850     vmx_set_test_stage(1);
1851     asm volatile ("hlt");
1852     report((rdtsc() - start > NMI_DELAY) && nmi_fired,
1853             "direct NMI + hlt");
1854     if (!nmi_fired)
1855         vmx_set_test_stage(-1);
1856     nmi_fired = false;
1857 
1858     vmcall();
1859 
1860     start = rdtsc();
1861     vmx_set_test_stage(2);
1862     asm volatile ("hlt");
1863     report((rdtsc() - start > NMI_DELAY) && !nmi_fired,
1864             "intercepted NMI + hlt");
1865     if (nmi_fired) {
1866         report(!nmi_fired, "intercepted NMI was dispatched");
1867         vmx_set_test_stage(-1);
1868         return;
1869     }
1870     vmx_set_test_stage(3);
1871 }
1872 
1873 static int nmi_hlt_exit_handler(union exit_reason exit_reason)
1874 {
1875     u64 guest_rip = vmcs_read(GUEST_RIP);
1876     u32 insn_len = vmcs_read(EXI_INST_LEN);
1877 
1878     switch (vmx_get_test_stage()) {
1879     case 1:
1880         if (exit_reason.basic != VMX_VMCALL) {
1881             report(false, "VMEXIT not due to vmcall. Exit reason 0x%x",
1882                    exit_reason.full);
1883             print_vmexit_info(exit_reason);
1884             return VMX_TEST_VMEXIT;
1885         }
1886 
1887         vmcs_write(PIN_CONTROLS,
1888                vmcs_read(PIN_CONTROLS) | PIN_NMI);
1889         vmcs_write(PIN_CONTROLS,
1890                vmcs_read(PIN_CONTROLS) | PIN_VIRT_NMI);
1891         vmcs_write(GUEST_RIP, guest_rip + insn_len);
1892         break;
1893 
1894     case 2:
1895         if (exit_reason.basic != VMX_EXC_NMI) {
1896             report(false, "VMEXIT not due to NMI intercept. Exit reason 0x%x",
1897                    exit_reason.full);
1898             print_vmexit_info(exit_reason);
1899             return VMX_TEST_VMEXIT;
1900         }
1901         report(true, "NMI intercept while running guest");
1902         vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1903         break;
1904 
1905     case 3:
1906         break;
1907 
1908     default:
1909         return VMX_TEST_VMEXIT;
1910     }
1911 
1912     if (vmx_get_test_stage() == 3)
1913         return VMX_TEST_VMEXIT;
1914 
1915     return VMX_TEST_RESUME;
1916 }
1917 
1918 
1919 static int dbgctls_init(struct vmcs *vmcs)
1920 {
1921 	u64 dr7 = 0x402;
1922 	u64 zero = 0;
1923 
1924 	msr_bmp_init();
1925 	asm volatile(
1926 		"mov %0,%%dr0\n\t"
1927 		"mov %0,%%dr1\n\t"
1928 		"mov %0,%%dr2\n\t"
1929 		"mov %1,%%dr7\n\t"
1930 		: : "r" (zero), "r" (dr7));
1931 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1932 	vmcs_write(GUEST_DR7, 0x404);
1933 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1934 
1935 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1936 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1937 
1938 	return VMX_TEST_START;
1939 }
1940 
1941 static void dbgctls_main(void)
1942 {
1943 	u64 dr7, debugctl;
1944 
1945 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1946 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1947 	/* Commented out: KVM does not support DEBUGCTL so far */
1948 	(void)debugctl;
1949 	report(dr7 == 0x404, "Load debug controls" /* && debugctl == 0x2 */);
1950 
1951 	dr7 = 0x408;
1952 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1953 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1954 
1955 	vmx_set_test_stage(0);
1956 	vmcall();
1957 	report(vmx_get_test_stage() == 1, "Save debug controls");
1958 
1959 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1960 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1961 		printf("\tDebug controls are always loaded/saved\n");
1962 		return;
1963 	}
1964 	vmx_set_test_stage(2);
1965 	vmcall();
1966 
1967 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1968 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1969 	/* Commented out: KVM does not support DEBUGCTL so far */
1970 	(void)debugctl;
1971 	report(dr7 == 0x402,
1972 	       "Guest=host debug controls" /* && debugctl == 0x1 */);
1973 
1974 	dr7 = 0x408;
1975 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1976 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1977 
1978 	vmx_set_test_stage(3);
1979 	vmcall();
1980 	report(vmx_get_test_stage() == 4, "Don't save debug controls");
1981 }
1982 
1983 static int dbgctls_exit_handler(union exit_reason exit_reason)
1984 {
1985 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1986 	u64 guest_rip = vmcs_read(GUEST_RIP);
1987 	u64 dr7, debugctl;
1988 
1989 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1990 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1991 
1992 	switch (exit_reason.basic) {
1993 	case VMX_VMCALL:
1994 		switch (vmx_get_test_stage()) {
1995 		case 0:
1996 			if (dr7 == 0x400 && debugctl == 0 &&
1997 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1998 			    Commented out: KVM does not support DEBUGCTL so far
1999 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
2000 				vmx_inc_test_stage();
2001 			break;
2002 		case 2:
2003 			dr7 = 0x402;
2004 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
2005 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
2006 			vmcs_write(GUEST_DR7, 0x404);
2007 			vmcs_write(GUEST_DEBUGCTL, 0x2);
2008 
2009 			vmcs_write(ENT_CONTROLS,
2010 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
2011 			vmcs_write(EXI_CONTROLS,
2012 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
2013 			break;
2014 		case 3:
2015 			if (dr7 == 0x400 && debugctl == 0 &&
2016 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
2017 			    Commented out: KVM does not support DEBUGCTL so far
2018 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
2019 				vmx_inc_test_stage();
2020 			break;
2021 		}
2022 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
2023 		return VMX_TEST_RESUME;
2024 	default:
2025 		report(false, "Unknown exit reason, %d", exit_reason.full);
2026 		print_vmexit_info(exit_reason);
2027 	}
2028 	return VMX_TEST_VMEXIT;
2029 }
2030 
2031 struct vmx_msr_entry {
2032 	u32 index;
2033 	u32 reserved;
2034 	u64 value;
2035 } __attribute__((packed));
2036 
2037 #define MSR_MAGIC 0x31415926
2038 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
2039 
2040 static int msr_switch_init(struct vmcs *vmcs)
2041 {
2042 	msr_bmp_init();
2043 	exit_msr_store = alloc_page();
2044 	exit_msr_load = alloc_page();
2045 	entry_msr_load = alloc_page();
2046 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
2047 	entry_msr_load[0].value = MSR_MAGIC;
2048 
2049 	vmx_set_test_stage(1);
2050 	vmcs_write(ENT_MSR_LD_CNT, 1);
2051 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
2052 	vmcs_write(EXI_MSR_ST_CNT, 1);
2053 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
2054 	vmcs_write(EXI_MSR_LD_CNT, 1);
2055 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
2056 	return VMX_TEST_START;
2057 }
2058 
2059 static void msr_switch_main(void)
2060 {
2061 	if (vmx_get_test_stage() == 1) {
2062 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC,
2063 		       "VM entry MSR load");
2064 		vmx_set_test_stage(2);
2065 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
2066 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
2067 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
2068 		exit_msr_load[0].value = MSR_MAGIC + 2;
2069 	}
2070 	vmcall();
2071 }
2072 
2073 static int msr_switch_exit_handler(union exit_reason exit_reason)
2074 {
2075 	if (exit_reason.basic == VMX_VMCALL && vmx_get_test_stage() == 2) {
2076 		report(exit_msr_store[0].value == MSR_MAGIC + 1,
2077 		       "VM exit MSR store");
2078 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2,
2079 		       "VM exit MSR load");
2080 		vmx_set_test_stage(3);
2081 		entry_msr_load[0].index = MSR_FS_BASE;
2082 		return VMX_TEST_RESUME;
2083 	}
2084 	printf("ERROR %s: unexpected stage=%u or reason=0x%x\n",
2085 		__func__, vmx_get_test_stage(), exit_reason.full);
2086 	return VMX_TEST_EXIT;
2087 }
2088 
2089 static int msr_switch_entry_failure(struct vmentry_result *result)
2090 {
2091 	if (result->vm_fail) {
2092 		printf("ERROR %s: VM-Fail on %s\n", __func__, result->instr);
2093 		return VMX_TEST_EXIT;
2094 	}
2095 
2096 	if (result->exit_reason.failed_vmentry &&
2097 	    result->exit_reason.basic == VMX_FAIL_MSR &&
2098 	    vmx_get_test_stage() == 3) {
2099 		report(vmcs_read(EXI_QUALIFICATION) == 1,
2100 		       "VM entry MSR load: try to load FS_BASE");
2101 		return VMX_TEST_VMEXIT;
2102 	}
2103 	printf("ERROR %s: unexpected stage=%u or reason=%x\n",
2104 		__func__, vmx_get_test_stage(), result->exit_reason.full);
2105 	return VMX_TEST_EXIT;
2106 }
2107 
2108 static int vmmcall_init(struct vmcs *vmcs)
2109 {
2110 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
2111 	return VMX_TEST_START;
2112 }
2113 
2114 static void vmmcall_main(void)
2115 {
2116 	asm volatile(
2117 		"mov $0xABCD, %%rax\n\t"
2118 		"vmmcall\n\t"
2119 		::: "rax");
2120 
2121 	report(0, "VMMCALL");
2122 }
2123 
2124 static int vmmcall_exit_handler(union exit_reason exit_reason)
2125 {
2126 	switch (exit_reason.basic) {
2127 	case VMX_VMCALL:
2128 		printf("here\n");
2129 		report(0, "VMMCALL triggers #UD");
2130 		break;
2131 	case VMX_EXC_NMI:
2132 		report((vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR,
2133 		       "VMMCALL triggers #UD");
2134 		break;
2135 	default:
2136 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
2137 		print_vmexit_info(exit_reason);
2138 	}
2139 
2140 	return VMX_TEST_VMEXIT;
2141 }
2142 
2143 static int disable_rdtscp_init(struct vmcs *vmcs)
2144 {
2145 	u32 ctrl_cpu1;
2146 
2147 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
2148 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
2149 		ctrl_cpu1 &= ~CPU_RDTSCP;
2150 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
2151 	}
2152 
2153 	return VMX_TEST_START;
2154 }
2155 
2156 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
2157 {
2158 	switch (vmx_get_test_stage()) {
2159 	case 0:
2160 		report(true, "RDTSCP triggers #UD");
2161 		vmx_inc_test_stage();
2162 		regs->rip += 3;
2163 		break;
2164 	case 2:
2165 		report(true, "RDPID triggers #UD");
2166 		vmx_inc_test_stage();
2167 		regs->rip += 4;
2168 		break;
2169 	}
2170 	return;
2171 
2172 }
2173 
2174 static void disable_rdtscp_main(void)
2175 {
2176 	/* Test that #UD is properly injected in L2.  */
2177 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2178 
2179 	vmx_set_test_stage(0);
2180 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2181 	vmcall();
2182 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2183 
2184 	handle_exception(UD_VECTOR, 0);
2185 	vmcall();
2186 }
2187 
2188 static int disable_rdtscp_exit_handler(union exit_reason exit_reason)
2189 {
2190 	switch (exit_reason.basic) {
2191 	case VMX_VMCALL:
2192 		switch (vmx_get_test_stage()) {
2193 		case 0:
2194 			report(false, "RDTSCP triggers #UD");
2195 			vmx_inc_test_stage();
2196 			/* fallthrough */
2197 		case 1:
2198 			vmx_inc_test_stage();
2199 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2200 			return VMX_TEST_RESUME;
2201 		case 2:
2202 			report(false, "RDPID triggers #UD");
2203 			break;
2204 		}
2205 		break;
2206 
2207 	default:
2208 		report(false, "Unknown exit reason, 0x%x", exit_reason.full);
2209 		print_vmexit_info(exit_reason);
2210 	}
2211 	return VMX_TEST_VMEXIT;
2212 }
2213 
2214 static int int3_init(struct vmcs *vmcs)
2215 {
2216 	vmcs_write(EXC_BITMAP, ~0u);
2217 	return VMX_TEST_START;
2218 }
2219 
2220 static void int3_guest_main(void)
2221 {
2222 	asm volatile ("int3");
2223 }
2224 
2225 static int int3_exit_handler(union exit_reason exit_reason)
2226 {
2227 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2228 
2229 	report(exit_reason.basic == VMX_EXC_NMI &&
2230 	       (intr_info & INTR_INFO_VALID_MASK) &&
2231 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2232 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2233 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2234 	       "L1 intercepts #BP");
2235 
2236 	return VMX_TEST_VMEXIT;
2237 }
2238 
2239 static int into_init(struct vmcs *vmcs)
2240 {
2241 	vmcs_write(EXC_BITMAP, ~0u);
2242 	return VMX_TEST_START;
2243 }
2244 
2245 static void into_guest_main(void)
2246 {
2247 	struct far_pointer32 fp = {
2248 		.offset = (uintptr_t)&&into,
2249 		.selector = KERNEL_CS32,
2250 	};
2251 	uintptr_t rsp;
2252 
2253 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
2254 
2255 	if (fp.offset != (uintptr_t)&&into) {
2256 		printf("Code address too high.\n");
2257 		return;
2258 	}
2259 	if ((u32)rsp != rsp) {
2260 		printf("Stack address too high.\n");
2261 		return;
2262 	}
2263 
2264 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2265 	return;
2266 into:
2267 	asm volatile (".code32;"
2268 		      "movl $0x7fffffff, %eax;"
2269 		      "addl %eax, %eax;"
2270 		      "into;"
2271 		      "lret;"
2272 		      ".code64");
2273 	__builtin_unreachable();
2274 }
2275 
2276 static int into_exit_handler(union exit_reason exit_reason)
2277 {
2278 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2279 
2280 	report(exit_reason.basic == VMX_EXC_NMI &&
2281 	       (intr_info & INTR_INFO_VALID_MASK) &&
2282 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2283 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2284 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2285 	       "L1 intercepts #OF");
2286 
2287 	return VMX_TEST_VMEXIT;
2288 }
2289 
2290 static void exit_monitor_from_l2_main(void)
2291 {
2292 	printf("Calling exit(0) from l2...\n");
2293 	exit(0);
2294 }
2295 
2296 static int exit_monitor_from_l2_handler(union exit_reason exit_reason)
2297 {
2298 	report(false, "The guest should have killed the VMM");
2299 	return VMX_TEST_EXIT;
2300 }
2301 
2302 static void assert_exit_reason(u64 expected)
2303 {
2304 	u64 actual = vmcs_read(EXI_REASON);
2305 
2306 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2307 			   exit_reason_description(expected),
2308 			   exit_reason_description(actual));
2309 }
2310 
2311 static void skip_exit_insn(void)
2312 {
2313 	u64 guest_rip = vmcs_read(GUEST_RIP);
2314 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2315 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2316 }
2317 
2318 static void skip_exit_vmcall(void)
2319 {
2320 	assert_exit_reason(VMX_VMCALL);
2321 	skip_exit_insn();
2322 }
2323 
2324 static void v2_null_test_guest(void)
2325 {
2326 }
2327 
2328 static void v2_null_test(void)
2329 {
2330 	test_set_guest(v2_null_test_guest);
2331 	enter_guest();
2332 	report(1, __func__);
2333 }
2334 
2335 static void v2_multiple_entries_test_guest(void)
2336 {
2337 	vmx_set_test_stage(1);
2338 	vmcall();
2339 	vmx_set_test_stage(2);
2340 }
2341 
2342 static void v2_multiple_entries_test(void)
2343 {
2344 	test_set_guest(v2_multiple_entries_test_guest);
2345 	enter_guest();
2346 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2347 	skip_exit_vmcall();
2348 	enter_guest();
2349 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2350 	report(1, __func__);
2351 }
2352 
2353 static int fixture_test_data = 1;
2354 
2355 static void fixture_test_teardown(void *data)
2356 {
2357 	*((int *) data) = 1;
2358 }
2359 
2360 static void fixture_test_guest(void)
2361 {
2362 	fixture_test_data++;
2363 }
2364 
2365 
2366 static void fixture_test_setup(void)
2367 {
2368 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2369 			   "fixture_test_teardown didn't run?!");
2370 	fixture_test_data = 2;
2371 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2372 	test_set_guest(fixture_test_guest);
2373 }
2374 
2375 static void fixture_test_case1(void)
2376 {
2377 	fixture_test_setup();
2378 	TEST_ASSERT_EQ(2, fixture_test_data);
2379 	enter_guest();
2380 	TEST_ASSERT_EQ(3, fixture_test_data);
2381 	report(1, __func__);
2382 }
2383 
2384 static void fixture_test_case2(void)
2385 {
2386 	fixture_test_setup();
2387 	TEST_ASSERT_EQ(2, fixture_test_data);
2388 	enter_guest();
2389 	TEST_ASSERT_EQ(3, fixture_test_data);
2390 	report(1, __func__);
2391 }
2392 
2393 enum ept_access_op {
2394 	OP_READ,
2395 	OP_WRITE,
2396 	OP_EXEC,
2397 	OP_FLUSH_TLB,
2398 	OP_EXIT,
2399 };
2400 
2401 static struct ept_access_test_data {
2402 	unsigned long gpa;
2403 	unsigned long *gva;
2404 	unsigned long hpa;
2405 	unsigned long *hva;
2406 	enum ept_access_op op;
2407 } ept_access_test_data;
2408 
2409 extern unsigned char ret42_start;
2410 extern unsigned char ret42_end;
2411 
2412 /* Returns 42. */
2413 asm(
2414 	".align 64\n"
2415 	"ret42_start:\n"
2416 	"mov $42, %eax\n"
2417 	"ret\n"
2418 	"ret42_end:\n"
2419 );
2420 
2421 static void
2422 diagnose_ept_violation_qual(u64 expected, u64 actual)
2423 {
2424 
2425 #define DIAGNOSE(flag)							\
2426 do {									\
2427 	if ((expected & flag) != (actual & flag))			\
2428 		printf(#flag " %sexpected\n",				\
2429 		       (expected & flag) ? "" : "un");			\
2430 } while (0)
2431 
2432 	DIAGNOSE(EPT_VLT_RD);
2433 	DIAGNOSE(EPT_VLT_WR);
2434 	DIAGNOSE(EPT_VLT_FETCH);
2435 	DIAGNOSE(EPT_VLT_PERM_RD);
2436 	DIAGNOSE(EPT_VLT_PERM_WR);
2437 	DIAGNOSE(EPT_VLT_PERM_EX);
2438 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2439 	DIAGNOSE(EPT_VLT_PADDR);
2440 
2441 #undef DIAGNOSE
2442 }
2443 
2444 static void do_ept_access_op(enum ept_access_op op)
2445 {
2446 	ept_access_test_data.op = op;
2447 	enter_guest();
2448 }
2449 
2450 /*
2451  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2452  * needed by tests that modify guest PTEs.
2453  */
2454 static void ept_access_test_guest_flush_tlb(void)
2455 {
2456 	do_ept_access_op(OP_FLUSH_TLB);
2457 	skip_exit_vmcall();
2458 }
2459 
2460 /*
2461  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2462  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2463  * a huge page.
2464  */
2465 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2466 				 unsigned long clear, unsigned long set)
2467 {
2468 	struct ept_access_test_data *data = &ept_access_test_data;
2469 	unsigned long orig_pte;
2470 	unsigned long pte;
2471 
2472 	/* Screw with the mapping at the requested level. */
2473 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2474 	pte = orig_pte;
2475 	if (mkhuge)
2476 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2477 	else
2478 		pte = orig_pte;
2479 	pte = (pte & ~clear) | set;
2480 	set_ept_pte(pml4, gpa, level, pte);
2481 	ept_sync(INVEPT_SINGLE, eptp);
2482 
2483 	return orig_pte;
2484 }
2485 
2486 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2487 {
2488 	set_ept_pte(pml4, gpa, level, orig_pte);
2489 	ept_sync(INVEPT_SINGLE, eptp);
2490 }
2491 
2492 static void do_ept_violation(bool leaf, enum ept_access_op op,
2493 			     u64 expected_qual, u64 expected_paddr)
2494 {
2495 	u64 qual;
2496 
2497 	/* Try the access and observe the violation. */
2498 	do_ept_access_op(op);
2499 
2500 	assert_exit_reason(VMX_EPT_VIOLATION);
2501 
2502 	qual = vmcs_read(EXI_QUALIFICATION);
2503 
2504 	/* Mask undefined bits (which may later be defined in certain cases). */
2505 	qual &= ~(EPT_VLT_GUEST_USER | EPT_VLT_GUEST_RW | EPT_VLT_GUEST_EX |
2506 		 EPT_VLT_PERM_USER_EX);
2507 
2508 	diagnose_ept_violation_qual(expected_qual, qual);
2509 	TEST_EXPECT_EQ(expected_qual, qual);
2510 
2511 	#if 0
2512 	/* Disable for now otherwise every test will fail */
2513 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2514 		       (unsigned long) (
2515 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2516 	#endif
2517 	/*
2518 	 * TODO: tests that probe expected_paddr in pages other than the one at
2519 	 * the beginning of the 1g region.
2520 	 */
2521 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2522 }
2523 
2524 static void
2525 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2526 			      unsigned long set, enum ept_access_op op,
2527 			      u64 expected_qual)
2528 {
2529 	struct ept_access_test_data *data = &ept_access_test_data;
2530 	unsigned long orig_pte;
2531 
2532 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2533 
2534 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2535 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2536 					 data->gpa);
2537 
2538 	/* Fix the violation and resume the op loop. */
2539 	ept_untwiddle(data->gpa, level, orig_pte);
2540 	enter_guest();
2541 	skip_exit_vmcall();
2542 }
2543 
2544 static void
2545 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2546 		       enum ept_access_op op, u64 expected_qual)
2547 {
2548 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2549 				      expected_qual);
2550 	if (ept_huge_pages_supported(level))
2551 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2552 					      expected_qual);
2553 }
2554 
2555 static void ept_violation(unsigned long clear, unsigned long set,
2556 			  enum ept_access_op op, u64 expected_qual)
2557 {
2558 	ept_violation_at_level(1, clear, set, op, expected_qual);
2559 	ept_violation_at_level(2, clear, set, op, expected_qual);
2560 	ept_violation_at_level(3, clear, set, op, expected_qual);
2561 	ept_violation_at_level(4, clear, set, op, expected_qual);
2562 }
2563 
2564 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2565 				       u64 expected_qual)
2566 {
2567 	ept_violation(EPT_PRESENT, access, op,
2568 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2569 }
2570 
2571 /*
2572  * For translations that don't involve a GVA, that is physical address (paddr)
2573  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2574  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2575  * translations don't involve GVAs, such as when the hardware does the guest
2576  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2577  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2578  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2579  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2580  *
2581  * Note that paddr violations can also be triggered by loading PAE page tables
2582  * with wonky addresses. We don't test that yet.
2583  *
2584  * This function modifies the EPT entry that maps the GPA that the guest page
2585  * table entry mapping ept_access_test_data.gva resides on.
2586  *
2587  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2588  *
2589  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2590  *
2591  *	@op		Guest operation to perform with
2592  *			ept_access_test_data.gva.
2593  *
2594  *	@expect_violation
2595  *			Is a violation expected during the paddr access?
2596  *
2597  *	@expected_qual	Expected qualification for the EPT violation.
2598  *			EPT_VLT_PADDR should be clear.
2599  */
2600 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2601 			     enum ept_access_op op, bool expect_violation,
2602 			     u64 expected_qual)
2603 {
2604 	struct ept_access_test_data *data = &ept_access_test_data;
2605 	unsigned long *ptep;
2606 	unsigned long gpa;
2607 	unsigned long orig_epte;
2608 	unsigned long epte;
2609 	int i;
2610 
2611 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2612 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2613 	TEST_ASSERT(ptep);
2614 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2615 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2616 	ept_access_test_guest_flush_tlb();
2617 
2618 	/*
2619 	 * Now modify the access bits on the EPT entry for the GPA that the
2620 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2621 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2622 	 * constructed our test such that those other 511 PTEs aren't used by
2623 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2624 	 * PTE we're modifying is at the beginning of a 4K page and the
2625 	 * following 511 entires are also under our control (and not touched by
2626 	 * the guest).
2627 	 */
2628 	gpa = virt_to_phys(ptep);
2629 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2630 	/*
2631 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2632 	 * otherwise our level=1 twiddling below will fail. We use the
2633 	 * identity map (gpa = gpa) since page tables are shared with the host.
2634 	 */
2635 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2636 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2637 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2638 
2639 	if (expect_violation) {
2640 		do_ept_violation(/*leaf=*/true, op,
2641 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2642 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2643 		do_ept_access_op(op);
2644 	} else {
2645 		do_ept_access_op(op);
2646 		if (ept_ad_enabled()) {
2647 			for (i = EPT_PAGE_LEVEL; i > 0; i--) {
2648 				TEST_ASSERT(get_ept_pte(pml4, gpa, i, &epte));
2649 				TEST_ASSERT(epte & EPT_ACCESS_FLAG);
2650 				if (i == 1)
2651 					TEST_ASSERT(epte & EPT_DIRTY_FLAG);
2652 				else
2653 					TEST_ASSERT_EQ(epte & EPT_DIRTY_FLAG, 0);
2654 			}
2655 		}
2656 
2657 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2658 	}
2659 
2660 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2661 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2662 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2663 
2664 	skip_exit_vmcall();
2665 }
2666 
2667 static void ept_access_allowed_paddr(unsigned long ept_access,
2668 				     unsigned long pte_ad,
2669 				     enum ept_access_op op)
2670 {
2671 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2672 			 /*expected_qual=*/-1);
2673 }
2674 
2675 static void ept_access_violation_paddr(unsigned long ept_access,
2676 				       unsigned long pte_ad,
2677 				       enum ept_access_op op,
2678 				       u64 expected_qual)
2679 {
2680 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2681 			 expected_qual);
2682 }
2683 
2684 
2685 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2686 					unsigned long clear,
2687 					unsigned long set,
2688 					enum ept_access_op op)
2689 {
2690 	struct ept_access_test_data *data = &ept_access_test_data;
2691 	unsigned long orig_pte;
2692 
2693 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2694 
2695 	/* No violation. Should proceed to vmcall. */
2696 	do_ept_access_op(op);
2697 	skip_exit_vmcall();
2698 
2699 	ept_untwiddle(data->gpa, level, orig_pte);
2700 }
2701 
2702 static void ept_allowed_at_level(int level, unsigned long clear,
2703 				 unsigned long set, enum ept_access_op op)
2704 {
2705 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2706 	if (ept_huge_pages_supported(level))
2707 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2708 }
2709 
2710 static void ept_allowed(unsigned long clear, unsigned long set,
2711 			enum ept_access_op op)
2712 {
2713 	ept_allowed_at_level(1, clear, set, op);
2714 	ept_allowed_at_level(2, clear, set, op);
2715 	ept_allowed_at_level(3, clear, set, op);
2716 	ept_allowed_at_level(4, clear, set, op);
2717 }
2718 
2719 static void ept_ignored_bit(int bit)
2720 {
2721 	/* Set the bit. */
2722 	ept_allowed(0, 1ul << bit, OP_READ);
2723 	ept_allowed(0, 1ul << bit, OP_WRITE);
2724 	ept_allowed(0, 1ul << bit, OP_EXEC);
2725 
2726 	/* Clear the bit. */
2727 	ept_allowed(1ul << bit, 0, OP_READ);
2728 	ept_allowed(1ul << bit, 0, OP_WRITE);
2729 	ept_allowed(1ul << bit, 0, OP_EXEC);
2730 }
2731 
2732 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2733 {
2734 	ept_allowed(EPT_PRESENT, access, op);
2735 }
2736 
2737 
2738 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2739 					     unsigned long clear,
2740 					     unsigned long set,
2741 					     enum ept_access_op op)
2742 {
2743 	struct ept_access_test_data *data = &ept_access_test_data;
2744 	unsigned long orig_pte;
2745 
2746 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2747 
2748 	do_ept_access_op(op);
2749 	assert_exit_reason(VMX_EPT_MISCONFIG);
2750 
2751 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2752 	#if 0
2753 	/* broken: */
2754 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2755 	#endif
2756 	#if 0
2757 	/*
2758 	 * broken:
2759 	 * According to description of exit qual for EPT violation,
2760 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2761 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2762 	 * to be set for msiconfig.
2763 	 */
2764 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2765 		       (unsigned long) (
2766 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2767 	#endif
2768 
2769 	/* Fix the violation and resume the op loop. */
2770 	ept_untwiddle(data->gpa, level, orig_pte);
2771 	enter_guest();
2772 	skip_exit_vmcall();
2773 }
2774 
2775 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2776 					  unsigned long clear,
2777 					  unsigned long set)
2778 {
2779 	/* The op shouldn't matter (read, write, exec), so try them all! */
2780 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2781 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2782 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2783 }
2784 
2785 static void ept_misconfig_at_level(int level, unsigned long clear,
2786 				   unsigned long set)
2787 {
2788 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2789 	if (ept_huge_pages_supported(level))
2790 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2791 }
2792 
2793 static void ept_misconfig(unsigned long clear, unsigned long set)
2794 {
2795 	ept_misconfig_at_level(1, clear, set);
2796 	ept_misconfig_at_level(2, clear, set);
2797 	ept_misconfig_at_level(3, clear, set);
2798 	ept_misconfig_at_level(4, clear, set);
2799 }
2800 
2801 static void ept_access_misconfig(unsigned long access)
2802 {
2803 	ept_misconfig(EPT_PRESENT, access);
2804 }
2805 
2806 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2807 {
2808 	/* Setting the bit causes a misconfig. */
2809 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2810 
2811 	/* Making the entry non-present turns reserved bits into ignored. */
2812 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2813 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2814 }
2815 
2816 static void ept_reserved_bit_at_level_huge(int level, int bit)
2817 {
2818 	/* Setting the bit causes a misconfig. */
2819 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2820 
2821 	/* Making the entry non-present turns reserved bits into ignored. */
2822 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2823 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2824 }
2825 
2826 static void ept_reserved_bit_at_level(int level, int bit)
2827 {
2828 	/* Setting the bit causes a misconfig. */
2829 	ept_misconfig_at_level(level, 0, 1ul << bit);
2830 
2831 	/* Making the entry non-present turns reserved bits into ignored. */
2832 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2833 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2834 }
2835 
2836 static void ept_reserved_bit(int bit)
2837 {
2838 	ept_reserved_bit_at_level(1, bit);
2839 	ept_reserved_bit_at_level(2, bit);
2840 	ept_reserved_bit_at_level(3, bit);
2841 	ept_reserved_bit_at_level(4, bit);
2842 }
2843 
2844 #define PAGE_2M_ORDER 9
2845 #define PAGE_1G_ORDER 18
2846 
2847 static void *get_1g_page(void)
2848 {
2849 	static void *alloc;
2850 
2851 	if (!alloc)
2852 		alloc = alloc_pages(PAGE_1G_ORDER);
2853 	return alloc;
2854 }
2855 
2856 static void ept_access_test_teardown(void *unused)
2857 {
2858 	/* Exit the guest cleanly. */
2859 	do_ept_access_op(OP_EXIT);
2860 }
2861 
2862 static void ept_access_test_guest(void)
2863 {
2864 	struct ept_access_test_data *data = &ept_access_test_data;
2865 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2866 
2867 	while (true) {
2868 		switch (data->op) {
2869 		case OP_READ:
2870 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2871 			break;
2872 		case OP_WRITE:
2873 			*data->gva = MAGIC_VAL_2;
2874 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2875 			*data->gva = MAGIC_VAL_1;
2876 			break;
2877 		case OP_EXEC:
2878 			TEST_ASSERT_EQ(42, code());
2879 			break;
2880 		case OP_FLUSH_TLB:
2881 			write_cr3(read_cr3());
2882 			break;
2883 		case OP_EXIT:
2884 			return;
2885 		default:
2886 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2887 		}
2888 		vmcall();
2889 	}
2890 }
2891 
2892 static void ept_access_test_setup(void)
2893 {
2894 	struct ept_access_test_data *data = &ept_access_test_data;
2895 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2896 	unsigned long size = npages * PAGE_SIZE;
2897 	unsigned long *page_table = current_page_table();
2898 	unsigned long pte;
2899 
2900 	if (setup_ept(false))
2901 		test_skip("EPT not supported");
2902 
2903 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2904 	if (cpuid_maxphyaddr() < 40)
2905 		test_skip("Test needs MAXPHYADDR >= 40");
2906 
2907 	test_set_guest(ept_access_test_guest);
2908 	test_add_teardown(ept_access_test_teardown, NULL);
2909 
2910 	data->hva = get_1g_page();
2911 	TEST_ASSERT(data->hva);
2912 	data->hpa = virt_to_phys(data->hva);
2913 
2914 	data->gpa = 1ul << 39;
2915 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2916 				   size);
2917 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2918 	install_pages(page_table, data->gpa, size, data->gva);
2919 
2920 	/*
2921 	 * Make sure nothing's mapped here so the tests that screw with the
2922 	 * pml4 entry don't inadvertently break something.
2923 	 */
2924 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2925 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2926 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2927 
2928 	data->hva[0] = MAGIC_VAL_1;
2929 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2930 }
2931 
2932 static void ept_access_test_not_present(void)
2933 {
2934 	ept_access_test_setup();
2935 	/* --- */
2936 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2937 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2938 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2939 }
2940 
2941 static void ept_access_test_read_only(void)
2942 {
2943 	ept_access_test_setup();
2944 
2945 	/* r-- */
2946 	ept_access_allowed(EPT_RA, OP_READ);
2947 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2948 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2949 }
2950 
2951 static void ept_access_test_write_only(void)
2952 {
2953 	ept_access_test_setup();
2954 	/* -w- */
2955 	ept_access_misconfig(EPT_WA);
2956 }
2957 
2958 static void ept_access_test_read_write(void)
2959 {
2960 	ept_access_test_setup();
2961 	/* rw- */
2962 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2963 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2964 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2965 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2966 }
2967 
2968 
2969 static void ept_access_test_execute_only(void)
2970 {
2971 	ept_access_test_setup();
2972 	/* --x */
2973 	if (ept_execute_only_supported()) {
2974 		ept_access_violation(EPT_EA, OP_READ,
2975 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2976 		ept_access_violation(EPT_EA, OP_WRITE,
2977 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2978 		ept_access_allowed(EPT_EA, OP_EXEC);
2979 	} else {
2980 		ept_access_misconfig(EPT_EA);
2981 	}
2982 }
2983 
2984 static void ept_access_test_read_execute(void)
2985 {
2986 	ept_access_test_setup();
2987 	/* r-x */
2988 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2989 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2990 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2991 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2992 }
2993 
2994 static void ept_access_test_write_execute(void)
2995 {
2996 	ept_access_test_setup();
2997 	/* -wx */
2998 	ept_access_misconfig(EPT_WA | EPT_EA);
2999 }
3000 
3001 static void ept_access_test_read_write_execute(void)
3002 {
3003 	ept_access_test_setup();
3004 	/* rwx */
3005 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
3006 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
3007 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
3008 }
3009 
3010 static void ept_access_test_reserved_bits(void)
3011 {
3012 	int i;
3013 	int maxphyaddr;
3014 
3015 	ept_access_test_setup();
3016 
3017 	/* Reserved bits above maxphyaddr. */
3018 	maxphyaddr = cpuid_maxphyaddr();
3019 	for (i = maxphyaddr; i <= 51; i++) {
3020 		report_prefix_pushf("reserved_bit=%d", i);
3021 		ept_reserved_bit(i);
3022 		report_prefix_pop();
3023 	}
3024 
3025 	/* Level-specific reserved bits. */
3026 	ept_reserved_bit_at_level_nohuge(2, 3);
3027 	ept_reserved_bit_at_level_nohuge(2, 4);
3028 	ept_reserved_bit_at_level_nohuge(2, 5);
3029 	ept_reserved_bit_at_level_nohuge(2, 6);
3030 	/* 2M alignment. */
3031 	for (i = 12; i < 20; i++) {
3032 		report_prefix_pushf("reserved_bit=%d", i);
3033 		ept_reserved_bit_at_level_huge(2, i);
3034 		report_prefix_pop();
3035 	}
3036 	ept_reserved_bit_at_level_nohuge(3, 3);
3037 	ept_reserved_bit_at_level_nohuge(3, 4);
3038 	ept_reserved_bit_at_level_nohuge(3, 5);
3039 	ept_reserved_bit_at_level_nohuge(3, 6);
3040 	/* 1G alignment. */
3041 	for (i = 12; i < 29; i++) {
3042 		report_prefix_pushf("reserved_bit=%d", i);
3043 		ept_reserved_bit_at_level_huge(3, i);
3044 		report_prefix_pop();
3045 	}
3046 	ept_reserved_bit_at_level(4, 3);
3047 	ept_reserved_bit_at_level(4, 4);
3048 	ept_reserved_bit_at_level(4, 5);
3049 	ept_reserved_bit_at_level(4, 6);
3050 	ept_reserved_bit_at_level(4, 7);
3051 }
3052 
3053 static void ept_access_test_ignored_bits(void)
3054 {
3055 	ept_access_test_setup();
3056 	/*
3057 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
3058 	 * far as translation is concerned even if AD bits are enabled in the
3059 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
3060 	 * control is 0.
3061 	 */
3062 	ept_ignored_bit(8);
3063 	ept_ignored_bit(9);
3064 	ept_ignored_bit(10);
3065 	ept_ignored_bit(11);
3066 	ept_ignored_bit(52);
3067 	ept_ignored_bit(53);
3068 	ept_ignored_bit(54);
3069 	ept_ignored_bit(55);
3070 	ept_ignored_bit(56);
3071 	ept_ignored_bit(57);
3072 	ept_ignored_bit(58);
3073 	ept_ignored_bit(59);
3074 	ept_ignored_bit(60);
3075 	ept_ignored_bit(61);
3076 	ept_ignored_bit(62);
3077 	ept_ignored_bit(63);
3078 }
3079 
3080 static void ept_access_test_paddr_not_present_ad_disabled(void)
3081 {
3082 	ept_access_test_setup();
3083 	ept_disable_ad_bits();
3084 
3085 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
3086 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
3087 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
3088 }
3089 
3090 static void ept_access_test_paddr_not_present_ad_enabled(void)
3091 {
3092 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
3093 
3094 	ept_access_test_setup();
3095 	ept_enable_ad_bits_or_skip_test();
3096 
3097 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
3098 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
3099 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
3100 }
3101 
3102 static void ept_access_test_paddr_read_only_ad_disabled(void)
3103 {
3104 	/*
3105 	 * When EPT AD bits are disabled, all accesses to guest paging
3106 	 * structures are reported separately as a read and (after
3107 	 * translation of the GPA to host physical address) a read+write
3108 	 * if the A/D bits have to be set.
3109 	 */
3110 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3111 
3112 	ept_access_test_setup();
3113 	ept_disable_ad_bits();
3114 
3115 	/* Can't update A bit, so all accesses fail. */
3116 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3117 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3118 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3119 	/* AD bits disabled, so only writes try to update the D bit. */
3120 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
3121 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3122 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
3123 	/* Both A and D already set, so read-only is OK. */
3124 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
3125 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
3126 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
3127 }
3128 
3129 static void ept_access_test_paddr_read_only_ad_enabled(void)
3130 {
3131 	/*
3132 	 * When EPT AD bits are enabled, all accesses to guest paging
3133 	 * structures are considered writes as far as EPT translation
3134 	 * is concerned.
3135 	 */
3136 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3137 
3138 	ept_access_test_setup();
3139 	ept_enable_ad_bits_or_skip_test();
3140 
3141 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3142 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3143 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3144 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
3145 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3146 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
3147 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
3148 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
3149 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
3150 }
3151 
3152 static void ept_access_test_paddr_read_write(void)
3153 {
3154 	ept_access_test_setup();
3155 	/* Read-write access to paging structure. */
3156 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
3157 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
3158 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
3159 }
3160 
3161 static void ept_access_test_paddr_read_write_execute(void)
3162 {
3163 	ept_access_test_setup();
3164 	/* RWX access to paging structure. */
3165 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
3166 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
3167 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
3168 }
3169 
3170 static void ept_access_test_paddr_read_execute_ad_disabled(void)
3171 {
3172   	/*
3173 	 * When EPT AD bits are disabled, all accesses to guest paging
3174 	 * structures are reported separately as a read and (after
3175 	 * translation of the GPA to host physical address) a read+write
3176 	 * if the A/D bits have to be set.
3177 	 */
3178 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3179 
3180 	ept_access_test_setup();
3181 	ept_disable_ad_bits();
3182 
3183 	/* Can't update A bit, so all accesses fail. */
3184 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3185 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3186 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3187 	/* AD bits disabled, so only writes try to update the D bit. */
3188 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
3189 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3190 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3191 	/* Both A and D already set, so read-only is OK. */
3192 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3193 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3194 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3195 }
3196 
3197 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3198 {
3199 	/*
3200 	 * When EPT AD bits are enabled, all accesses to guest paging
3201 	 * structures are considered writes as far as EPT translation
3202 	 * is concerned.
3203 	 */
3204 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3205 
3206 	ept_access_test_setup();
3207 	ept_enable_ad_bits_or_skip_test();
3208 
3209 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3210 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3211 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3212 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3213 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3214 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3215 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3216 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3217 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3218 }
3219 
3220 static void ept_access_test_paddr_not_present_page_fault(void)
3221 {
3222 	ept_access_test_setup();
3223 	/*
3224 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3225 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3226 	 * Thus guest page fault before host takes EPT violation for trying to
3227 	 * update A bit.
3228 	 */
3229 }
3230 
3231 static void ept_access_test_force_2m_page(void)
3232 {
3233 	ept_access_test_setup();
3234 
3235 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3236 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3237 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3238 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3239 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3240 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3241 }
3242 
3243 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3244 {
3245 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3246 
3247 	TEST_ASSERT(msr & VPID_CAP_INVVPID);
3248 
3249 	if (type < INVVPID_ADDR || type > INVVPID_CONTEXT_LOCAL)
3250 		return false;
3251 
3252 	if (!(msr & (1ull << (type + VPID_CAP_INVVPID_TYPES_SHIFT))))
3253 		return false;
3254 
3255 	if (vpid >> 16)
3256 		return false;
3257 
3258 	if (type != INVVPID_ALL && !vpid)
3259 		return false;
3260 
3261 	if (type == INVVPID_ADDR && !is_canonical(gla))
3262 		return false;
3263 
3264 	return true;
3265 }
3266 
3267 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3268 {
3269 	int rc;
3270 	bool valid = invvpid_valid(type, vpid, gla);
3271 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3272 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3273 	/*
3274 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3275 	 * that we can tell if it is updated by INVVPID.
3276 	 */
3277 	vmcs_read(~0);
3278 	rc = invvpid(type, vpid, gla);
3279 	report(!rc == valid, "INVVPID type %ld VPID %lx GLA %lx %s", type,
3280 	       vpid, gla,
3281 	       valid ? "passes" : "fails");
3282 	report(vmcs_read(VMX_INST_ERROR) == expected,
3283 	       "After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3284 	       rc ? "failed" : "successful",
3285 	       expected, vmcs_read(VMX_INST_ERROR));
3286 }
3287 
3288 static void ds_invvpid(void *data)
3289 {
3290 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3291 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3292 
3293 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3294 	asm volatile("invvpid %0, %1"
3295 		     :
3296 		     : "m"(*(struct invvpid_operand *)data),
3297 		       "r"(type));
3298 }
3299 
3300 /*
3301  * The SS override is ignored in 64-bit mode, so we use an addressing
3302  * mode with %rsp as the base register to generate an implicit SS
3303  * reference.
3304  */
3305 static void ss_invvpid(void *data)
3306 {
3307 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3308 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3309 
3310 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3311 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3312 		     : "+r"(data)
3313 		     : "r"(type));
3314 }
3315 
3316 static void invvpid_test_gp(void)
3317 {
3318 	bool fault;
3319 
3320 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3321 				   (void *)NONCANONICAL);
3322 	report(fault, "INVVPID with non-canonical DS operand raises #GP");
3323 }
3324 
3325 static void invvpid_test_ss(void)
3326 {
3327 	bool fault;
3328 
3329 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3330 				   (void *)NONCANONICAL);
3331 	report(fault, "INVVPID with non-canonical SS operand raises #SS");
3332 }
3333 
3334 static void invvpid_test_pf(void)
3335 {
3336 	void *vpage = alloc_vpage();
3337 	bool fault;
3338 
3339 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3340 	report(fault, "INVVPID with unmapped operand raises #PF");
3341 }
3342 
3343 static void try_compat_invvpid(void *unused)
3344 {
3345 	struct far_pointer32 fp = {
3346 		.offset = (uintptr_t)&&invvpid,
3347 		.selector = KERNEL_CS32,
3348 	};
3349 	uintptr_t rsp;
3350 
3351 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
3352 
3353 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3354 			"Code address too high.");
3355 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3356 
3357 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3358 	return;
3359 invvpid:
3360 	asm volatile (".code32;"
3361 		      "invvpid (%eax), %eax;"
3362 		      "lret;"
3363 		      ".code64");
3364 	__builtin_unreachable();
3365 }
3366 
3367 static void invvpid_test_compatibility_mode(void)
3368 {
3369 	bool fault;
3370 
3371 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3372 	report(fault, "Compatibility mode INVVPID raises #UD");
3373 }
3374 
3375 static void invvpid_test_not_in_vmx_operation(void)
3376 {
3377 	bool fault;
3378 
3379 	TEST_ASSERT(!vmx_off());
3380 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3381 	report(fault, "INVVPID outside of VMX operation raises #UD");
3382 	TEST_ASSERT(!vmx_on());
3383 }
3384 
3385 /*
3386  * This does not test real-address mode, virtual-8086 mode, protected mode,
3387  * or CPL > 0.
3388  */
3389 static void invvpid_test_v2(void)
3390 {
3391 	u64 msr;
3392 	int i;
3393 	unsigned types = 0;
3394 	unsigned type;
3395 
3396 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
3397 	    !(ctrl_cpu_rev[1].clr & CPU_VPID))
3398 		test_skip("VPID not supported");
3399 
3400 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3401 
3402 	if (!(msr & VPID_CAP_INVVPID))
3403 		test_skip("INVVPID not supported.\n");
3404 
3405 	if (msr & VPID_CAP_INVVPID_ADDR)
3406 		types |= 1u << INVVPID_ADDR;
3407 	if (msr & VPID_CAP_INVVPID_CXTGLB)
3408 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3409 	if (msr & VPID_CAP_INVVPID_ALL)
3410 		types |= 1u << INVVPID_ALL;
3411 	if (msr & VPID_CAP_INVVPID_CXTLOC)
3412 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3413 
3414 	if (!types)
3415 		test_skip("No INVVPID types supported.\n");
3416 
3417 	for (i = -127; i < 128; i++)
3418 		try_invvpid(i, 0xffff, 0);
3419 
3420 	/*
3421 	 * VPID must not be more than 16 bits.
3422 	 */
3423 	for (i = 0; i < 64; i++)
3424 		for (type = 0; type < 4; type++)
3425 			if (types & (1u << type))
3426 				try_invvpid(type, 1ul << i, 0);
3427 
3428 	/*
3429 	 * VPID must not be zero, except for "all contexts."
3430 	 */
3431 	for (type = 0; type < 4; type++)
3432 		if (types & (1u << type))
3433 			try_invvpid(type, 0, 0);
3434 
3435 	/*
3436 	 * The gla operand is only validated for single-address INVVPID.
3437 	 */
3438 	if (types & (1u << INVVPID_ADDR))
3439 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3440 
3441 	invvpid_test_gp();
3442 	invvpid_test_ss();
3443 	invvpid_test_pf();
3444 	invvpid_test_compatibility_mode();
3445 	invvpid_test_not_in_vmx_operation();
3446 }
3447 
3448 /*
3449  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3450  * at least as far as the guest-state checks. Returns false if the
3451  * VMLAUNCH fails early and execution falls through to the next
3452  * instruction.
3453  */
3454 static bool vmlaunch_succeeds(void)
3455 {
3456 	u32 exit_reason;
3457 
3458 	/*
3459 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3460 	 * unsupported VMCS component"). The caller can then check
3461 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3462 	 */
3463 	vmcs_write(~0u, 0);
3464 
3465 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3466 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3467 				   :
3468 				   : "r" ((u64)HOST_RSP)
3469 				   : "cc", "memory"
3470 				   : success);
3471 	return false;
3472 success:
3473 	exit_reason = vmcs_read(EXI_REASON);
3474 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3475 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3476 	return true;
3477 }
3478 
3479 /*
3480  * Try to launch the current VMCS.
3481  */
3482 static void test_vmx_vmlaunch(u32 xerror)
3483 {
3484 	bool success = vmlaunch_succeeds();
3485 	u32 vmx_inst_err;
3486 
3487 	report(success == !xerror, "vmlaunch %s",
3488 	       !xerror ? "succeeds" : "fails");
3489 	if (!success && xerror) {
3490 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3491 		report(vmx_inst_err == xerror,
3492 		       "VMX inst error is %d (actual %d)", xerror,
3493 		       vmx_inst_err);
3494 	}
3495 }
3496 
3497 /*
3498  * Try to launch the current VMCS, and expect one of two possible
3499  * errors (or success) codes.
3500  */
3501 static void test_vmx_vmlaunch2(u32 xerror1, u32 xerror2)
3502 {
3503 	bool success = vmlaunch_succeeds();
3504 	u32 vmx_inst_err;
3505 
3506 	if (!xerror1 == !xerror2)
3507 		report(success == !xerror1, "vmlaunch %s",
3508 		       !xerror1 ? "succeeds" : "fails");
3509 
3510 	if (!success && (xerror1 || xerror2)) {
3511 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3512 		report(vmx_inst_err == xerror1 || vmx_inst_err == xerror2,
3513 		       "VMX inst error is %d or %d (actual %d)", xerror1,
3514 		       xerror2, vmx_inst_err);
3515 	}
3516 }
3517 
3518 static void test_vmx_invalid_controls(void)
3519 {
3520 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3521 }
3522 
3523 static void test_vmx_valid_controls(void)
3524 {
3525 	test_vmx_vmlaunch(0);
3526 }
3527 
3528 /*
3529  * Test a particular value of a VM-execution control bit, if the value
3530  * is required or if the value is zero.
3531  */
3532 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3533 				    enum Encoding encoding, unsigned bit,
3534 				    unsigned val)
3535 {
3536 	u32 mask = 1u << bit;
3537 	bool expected;
3538 	u32 controls;
3539 
3540 	if (msr.set & mask)
3541 		TEST_ASSERT(msr.clr & mask);
3542 
3543 	/*
3544 	 * We can't arbitrarily turn on a control bit, because it may
3545 	 * introduce dependencies on other VMCS fields. So, we only
3546 	 * test turning on bits that have a required setting.
3547 	 */
3548 	if (val && (msr.clr & mask) && !(msr.set & mask))
3549 		return;
3550 
3551 	report_prefix_pushf("%s %s bit %d",
3552 			    val ? "Set" : "Clear", name, bit);
3553 
3554 	controls = vmcs_read(encoding);
3555 	if (val) {
3556 		vmcs_write(encoding, msr.set | mask);
3557 		expected = (msr.clr & mask);
3558 	} else {
3559 		vmcs_write(encoding, msr.set & ~mask);
3560 		expected = !(msr.set & mask);
3561 	}
3562 	if (expected)
3563 		test_vmx_valid_controls();
3564 	else
3565 		test_vmx_invalid_controls();
3566 	vmcs_write(encoding, controls);
3567 	report_prefix_pop();
3568 }
3569 
3570 /*
3571  * Test reserved values of a VM-execution control bit, based on the
3572  * allowed bit settings from the corresponding VMX capability MSR.
3573  */
3574 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3575 			      enum Encoding encoding, unsigned bit)
3576 {
3577 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3578 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3579 }
3580 
3581 /*
3582  * Reserved bits in the pin-based VM-execution controls must be set
3583  * properly. Software may consult the VMX capability MSRs to determine
3584  * the proper settings.
3585  * [Intel SDM]
3586  */
3587 static void test_pin_based_ctls(void)
3588 {
3589 	unsigned bit;
3590 
3591 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3592 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3593 	for (bit = 0; bit < 32; bit++)
3594 		test_rsvd_ctl_bit("pin-based controls",
3595 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3596 }
3597 
3598 /*
3599  * Reserved bits in the primary processor-based VM-execution controls
3600  * must be set properly. Software may consult the VMX capability MSRs
3601  * to determine the proper settings.
3602  * [Intel SDM]
3603  */
3604 static void test_primary_processor_based_ctls(void)
3605 {
3606 	unsigned bit;
3607 
3608 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3609 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3610 	for (bit = 0; bit < 32; bit++)
3611 		test_rsvd_ctl_bit("primary processor-based controls",
3612 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3613 }
3614 
3615 /*
3616  * If the "activate secondary controls" primary processor-based
3617  * VM-execution control is 1, reserved bits in the secondary
3618  * processor-based VM-execution controls must be cleared. Software may
3619  * consult the VMX capability MSRs to determine which bits are
3620  * reserved.
3621  * If the "activate secondary controls" primary processor-based
3622  * VM-execution control is 0 (or if the processor does not support the
3623  * 1-setting of that control), no checks are performed on the
3624  * secondary processor-based VM-execution controls.
3625  * [Intel SDM]
3626  */
3627 static void test_secondary_processor_based_ctls(void)
3628 {
3629 	u32 primary;
3630 	u32 secondary;
3631 	unsigned bit;
3632 
3633 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3634 		return;
3635 
3636 	primary = vmcs_read(CPU_EXEC_CTRL0);
3637 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3638 
3639 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3640 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3641 	for (bit = 0; bit < 32; bit++)
3642 		test_rsvd_ctl_bit("secondary processor-based controls",
3643 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3644 
3645 	/*
3646 	 * When the "activate secondary controls" VM-execution control
3647 	 * is clear, there are no checks on the secondary controls.
3648 	 */
3649 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3650 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3651 	report(vmlaunch_succeeds(),
3652 	       "Secondary processor-based controls ignored");
3653 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3654 	vmcs_write(CPU_EXEC_CTRL0, primary);
3655 }
3656 
3657 static void try_cr3_target_count(unsigned i, unsigned max)
3658 {
3659 	report_prefix_pushf("CR3 target count 0x%x", i);
3660 	vmcs_write(CR3_TARGET_COUNT, i);
3661 	if (i <= max)
3662 		test_vmx_valid_controls();
3663 	else
3664 		test_vmx_invalid_controls();
3665 	report_prefix_pop();
3666 }
3667 
3668 /*
3669  * The CR3-target count must not be greater than 4. Future processors
3670  * may support a different number of CR3-target values. Software
3671  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3672  * number of values supported.
3673  * [Intel SDM]
3674  */
3675 static void test_cr3_targets(void)
3676 {
3677 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3678 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3679 	unsigned i;
3680 
3681 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3682 	TEST_ASSERT(supported_targets <= 256);
3683 
3684 	try_cr3_target_count(-1u, supported_targets);
3685 	try_cr3_target_count(0x80000000, supported_targets);
3686 	try_cr3_target_count(0x7fffffff, supported_targets);
3687 	for (i = 0; i <= supported_targets + 1; i++)
3688 		try_cr3_target_count(i, supported_targets);
3689 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3690 
3691 	/* VMWRITE to nonexistent target fields should fail. */
3692 	for (i = supported_targets; i < 256; i++)
3693 		TEST_ASSERT(vmcs_write(CR3_TARGET_0 + i*2, 0));
3694 }
3695 
3696 /*
3697  * Test a particular address setting in the VMCS
3698  */
3699 static void test_vmcs_addr(const char *name,
3700 			   enum Encoding encoding,
3701 			   u64 align,
3702 			   bool ignored,
3703 			   bool skip_beyond_mapped_ram,
3704 			   u64 addr)
3705 {
3706 	report_prefix_pushf("%s = %lx", name, addr);
3707 	vmcs_write(encoding, addr);
3708 	if (skip_beyond_mapped_ram &&
3709 	    addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3710 	    addr < (1ul << cpuid_maxphyaddr()))
3711 		printf("Skipping physical address beyond mapped RAM\n");
3712 	else if (ignored || (IS_ALIGNED(addr, align) &&
3713 	    addr < (1ul << cpuid_maxphyaddr())))
3714 		test_vmx_valid_controls();
3715 	else
3716 		test_vmx_invalid_controls();
3717 	report_prefix_pop();
3718 }
3719 
3720 /*
3721  * Test interesting values for a VMCS address
3722  */
3723 static void test_vmcs_addr_values(const char *name,
3724 				  enum Encoding encoding,
3725 				  u64 align,
3726 				  bool ignored,
3727 				  bool skip_beyond_mapped_ram,
3728 				  u32 bit_start, u32 bit_end)
3729 {
3730 	unsigned i;
3731 	u64 orig_val = vmcs_read(encoding);
3732 
3733 	for (i = bit_start; i <= bit_end; i++)
3734 		test_vmcs_addr(name, encoding, align, ignored,
3735 			       skip_beyond_mapped_ram, 1ul << i);
3736 
3737 	test_vmcs_addr(name, encoding, align, ignored,
3738 		       skip_beyond_mapped_ram, PAGE_SIZE - 1);
3739 	test_vmcs_addr(name, encoding, align, ignored,
3740 		       skip_beyond_mapped_ram, PAGE_SIZE);
3741 	test_vmcs_addr(name, encoding, align, ignored,
3742 		       skip_beyond_mapped_ram,
3743 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3744 	test_vmcs_addr(name, encoding, align, ignored,
3745 		       skip_beyond_mapped_ram, -1ul);
3746 
3747 	vmcs_write(encoding, orig_val);
3748 }
3749 
3750 /*
3751  * Test a physical address reference in the VMCS, when the corresponding
3752  * feature is enabled and when the corresponding feature is disabled.
3753  */
3754 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3755 				     const char *field_name,
3756 				     const char *control_name, u64 align,
3757 				     bool skip_beyond_mapped_ram,
3758 				     bool control_primary)
3759 {
3760 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3761 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3762 	u64 page_addr;
3763 
3764 	if (control_primary) {
3765 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3766 			return;
3767 	} else {
3768 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3769 			return;
3770 	}
3771 
3772 	page_addr = vmcs_read(field);
3773 
3774 	report_prefix_pushf("%s enabled", control_name);
3775 	if (control_primary) {
3776 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3777 	} else {
3778 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3779 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3780 	}
3781 
3782 	test_vmcs_addr_values(field_name, field, align, false,
3783 			      skip_beyond_mapped_ram, 0, 63);
3784 	report_prefix_pop();
3785 
3786 	report_prefix_pushf("%s disabled", control_name);
3787 	if (control_primary) {
3788 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3789 	} else {
3790 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3791 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3792 	}
3793 
3794 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3795 	report_prefix_pop();
3796 
3797 	vmcs_write(field, page_addr);
3798 	vmcs_write(CPU_EXEC_CTRL0, primary);
3799 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3800 }
3801 
3802 /*
3803  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3804  * each I/O-bitmap address must be 0. Neither address should set any
3805  * bits beyond the processor's physical-address width.
3806  * [Intel SDM]
3807  */
3808 static void test_io_bitmaps(void)
3809 {
3810 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3811 				 "I/O bitmap A", "Use I/O bitmaps",
3812 				 PAGE_SIZE, false, true);
3813 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3814 				 "I/O bitmap B", "Use I/O bitmaps",
3815 				 PAGE_SIZE, false, true);
3816 }
3817 
3818 /*
3819  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3820  * the MSR-bitmap address must be 0. The address should not set any
3821  * bits beyond the processor's physical-address width.
3822  * [Intel SDM]
3823  */
3824 static void test_msr_bitmap(void)
3825 {
3826 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3827 				 "MSR bitmap", "Use MSR bitmaps",
3828 				 PAGE_SIZE, false, true);
3829 }
3830 
3831 /*
3832  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3833  * address must satisfy the following checks:
3834  * - Bits 11:0 of the address must be 0.
3835  * - The address should not set any bits beyond the processor's
3836  *   physical-address width.
3837  * [Intel SDM]
3838  */
3839 static void test_apic_virt_addr(void)
3840 {
3841 	/*
3842 	 * Ensure the processor will never use the virtual-APIC page, since
3843 	 * we will point it to invalid RAM.  Otherwise KVM is puzzled about
3844 	 * what we're trying to achieve and fails vmentry.
3845 	 */
3846 	u32 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
3847 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0 | CPU_CR8_LOAD | CPU_CR8_STORE);
3848 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3849 				 "virtual-APIC address", "Use TPR shadow",
3850 				 PAGE_SIZE, false, true);
3851 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
3852 }
3853 
3854 /*
3855  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3856  * APIC-access address must satisfy the following checks:
3857  *  - Bits 11:0 of the address must be 0.
3858  *  - The address should not set any bits beyond the processor's
3859  *    physical-address width.
3860  * [Intel SDM]
3861  */
3862 static void test_apic_access_addr(void)
3863 {
3864 	void *apic_access_page = alloc_page();
3865 
3866 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3867 
3868 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3869 				 "APIC-access address",
3870 				 "virtualize APIC-accesses", PAGE_SIZE,
3871 				 true, false);
3872 }
3873 
3874 static bool set_bit_pattern(u8 mask, u32 *secondary)
3875 {
3876 	u8 i;
3877 	bool flag = false;
3878 	u32 test_bits[3] = {
3879 		CPU_VIRT_X2APIC,
3880 		CPU_APIC_REG_VIRT,
3881 		CPU_VINTD
3882 	};
3883 
3884         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3885 		if ((mask & (1u << i)) &&
3886 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3887 			*secondary |= test_bits[i];
3888 			flag = true;
3889 		}
3890 	}
3891 
3892 	return (flag);
3893 }
3894 
3895 /*
3896  * If the "use TPR shadow" VM-execution control is 0, the following
3897  * VM-execution controls must also be 0:
3898  * 	- virtualize x2APIC mode
3899  *	- APIC-register virtualization
3900  *	- virtual-interrupt delivery
3901  *    [Intel SDM]
3902  *
3903  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3904  *    "virtualize APIC accesses" VM-execution control must be 0.
3905  *    [Intel SDM]
3906  */
3907 static void test_apic_virtual_ctls(void)
3908 {
3909 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3910 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3911 	u32 primary = saved_primary;
3912 	u32 secondary = saved_secondary;
3913 	bool ctrl = false;
3914 	char str[10] = "disabled";
3915 	u8 i = 0, j;
3916 
3917 	/*
3918 	 * First test
3919 	 */
3920 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3921 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3922 		return;
3923 
3924 	primary |= CPU_SECONDARY;
3925 	primary &= ~CPU_TPR_SHADOW;
3926 	vmcs_write(CPU_EXEC_CTRL0, primary);
3927 
3928 	while (1) {
3929 		for (j = 1; j < 8; j++) {
3930 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3931 			if (primary & CPU_TPR_SHADOW) {
3932 				ctrl = true;
3933 			} else {
3934 				if (! set_bit_pattern(j, &secondary))
3935 					ctrl = true;
3936 				else
3937 					ctrl = false;
3938 			}
3939 
3940 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3941 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3942 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3943 			if (ctrl)
3944 				test_vmx_valid_controls();
3945 			else
3946 				test_vmx_invalid_controls();
3947 			report_prefix_pop();
3948 		}
3949 
3950 		if (i == 1)
3951 			break;
3952 		i++;
3953 
3954 		primary |= CPU_TPR_SHADOW;
3955 		vmcs_write(CPU_EXEC_CTRL0, primary);
3956 		strcpy(str, "enabled");
3957 	}
3958 
3959 	/*
3960 	 * Second test
3961 	 */
3962 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3963 
3964 	primary = saved_primary;
3965 	secondary = saved_secondary;
3966 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3967 		return;
3968 
3969 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3970 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3971 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3972 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3973 	test_vmx_valid_controls();
3974 	report_prefix_pop();
3975 
3976 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3977 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3978 	test_vmx_valid_controls();
3979 	report_prefix_pop();
3980 
3981 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3982 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3983 	test_vmx_invalid_controls();
3984 	report_prefix_pop();
3985 
3986 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3987 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3988 	test_vmx_valid_controls();
3989 	report_prefix_pop();
3990 
3991 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3992 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3993 }
3994 
3995 /*
3996  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3997  * "external-interrupt exiting" VM-execution control must be 1.
3998  * [Intel SDM]
3999  */
4000 static void test_virtual_intr_ctls(void)
4001 {
4002 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4003 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4004 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4005 	u32 primary = saved_primary;
4006 	u32 secondary = saved_secondary;
4007 	u32 pin = saved_pin;
4008 
4009 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
4010 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
4011 		return;
4012 
4013 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
4014 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
4015 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
4016 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
4017 	test_vmx_valid_controls();
4018 	report_prefix_pop();
4019 
4020 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
4021 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
4022 	test_vmx_invalid_controls();
4023 	report_prefix_pop();
4024 
4025 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
4026 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
4027 	test_vmx_valid_controls();
4028 	report_prefix_pop();
4029 
4030 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
4031 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
4032 	test_vmx_invalid_controls();
4033 	report_prefix_pop();
4034 
4035 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4036 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4037 	vmcs_write(PIN_CONTROLS, saved_pin);
4038 }
4039 
4040 static void test_pi_desc_addr(u64 addr, bool ctrl)
4041 {
4042 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
4043 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
4044 	if (ctrl)
4045 		test_vmx_valid_controls();
4046 	else
4047 		test_vmx_invalid_controls();
4048 	report_prefix_pop();
4049 }
4050 
4051 /*
4052  * If the "process posted interrupts" VM-execution control is 1, the
4053  * following must be true:
4054  *
4055  *	- The "virtual-interrupt delivery" VM-execution control is 1.
4056  *	- The "acknowledge interrupt on exit" VM-exit control is 1.
4057  *	- The posted-interrupt notification vector has a value in the
4058  *	- range 0 - 255 (bits 15:8 are all 0).
4059  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
4060  *	- The posted-interrupt descriptor address does not set any bits
4061  *	  beyond the processor's physical-address width.
4062  * [Intel SDM]
4063  */
4064 static void test_posted_intr(void)
4065 {
4066 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4067 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4068 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4069 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
4070 	u32 primary = saved_primary;
4071 	u32 secondary = saved_secondary;
4072 	u32 pin = saved_pin;
4073 	u32 exit_ctl = exit_ctl_saved;
4074 	u16 vec;
4075 	int i;
4076 
4077 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
4078 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
4079 	    (ctrl_exit_rev.clr & EXI_INTA)))
4080 		return;
4081 
4082 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
4083 
4084 	/*
4085 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
4086 	 */
4087 	pin |= PIN_POST_INTR;
4088 	vmcs_write(PIN_CONTROLS, pin);
4089 	secondary &= ~CPU_VINTD;
4090 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4091 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
4092 	test_vmx_invalid_controls();
4093 	report_prefix_pop();
4094 
4095 	secondary |= CPU_VINTD;
4096 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4097 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
4098 	test_vmx_invalid_controls();
4099 	report_prefix_pop();
4100 
4101 	exit_ctl &= ~EXI_INTA;
4102 	vmcs_write(EXI_CONTROLS, exit_ctl);
4103 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
4104 	test_vmx_invalid_controls();
4105 	report_prefix_pop();
4106 
4107 	exit_ctl |= EXI_INTA;
4108 	vmcs_write(EXI_CONTROLS, exit_ctl);
4109 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4110 	test_vmx_valid_controls();
4111 	report_prefix_pop();
4112 
4113 	secondary &= ~CPU_VINTD;
4114 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4115 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
4116 	test_vmx_invalid_controls();
4117 	report_prefix_pop();
4118 
4119 	secondary |= CPU_VINTD;
4120 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4121 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4122 	test_vmx_valid_controls();
4123 	report_prefix_pop();
4124 
4125 	/*
4126 	 * Test posted-interrupt notification vector
4127 	 */
4128 	for (i = 0; i < 8; i++) {
4129 		vec = (1ul << i);
4130 		vmcs_write(PINV, vec);
4131 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4132 		test_vmx_valid_controls();
4133 		report_prefix_pop();
4134 	}
4135 	for (i = 8; i < 16; i++) {
4136 		vec = (1ul << i);
4137 		vmcs_write(PINV, vec);
4138 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4139 		test_vmx_invalid_controls();
4140 		report_prefix_pop();
4141 	}
4142 
4143 	vec &= ~(0xff << 8);
4144 	vmcs_write(PINV, vec);
4145 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4146 	test_vmx_valid_controls();
4147 	report_prefix_pop();
4148 
4149 	/*
4150 	 * Test posted-interrupt descriptor addresss
4151 	 */
4152 	for (i = 0; i < 6; i++) {
4153 		test_pi_desc_addr(1ul << i, false);
4154 	}
4155 
4156 	test_pi_desc_addr(0xf0, false);
4157 	test_pi_desc_addr(0xff, false);
4158 	test_pi_desc_addr(0x0f, false);
4159 	test_pi_desc_addr(0x8000, true);
4160 	test_pi_desc_addr(0x00, true);
4161 	test_pi_desc_addr(0xc000, true);
4162 
4163 	test_vmcs_addr_values("process-posted interrupts",
4164 			       POSTED_INTR_DESC_ADDR, 64,
4165 			       false, false, 0, 63);
4166 
4167 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4168 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4169 	vmcs_write(PIN_CONTROLS, saved_pin);
4170 }
4171 
4172 static void test_apic_ctls(void)
4173 {
4174 	test_apic_virt_addr();
4175 	test_apic_access_addr();
4176 	test_apic_virtual_ctls();
4177 	test_virtual_intr_ctls();
4178 	test_posted_intr();
4179 }
4180 
4181 /*
4182  * If the "enable VPID" VM-execution control is 1, the value of the
4183  * of the VPID VM-execution control field must not be 0000H.
4184  * [Intel SDM]
4185  */
4186 static void test_vpid(void)
4187 {
4188 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4189 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4190 	u16 vpid = 0x0000;
4191 	int i;
4192 
4193 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4194 	    (ctrl_cpu_rev[1].clr & CPU_VPID))) {
4195 		printf("Secondary controls and/or VPID not supported\n");
4196 		return;
4197 	}
4198 
4199 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
4200 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
4201 	vmcs_write(VPID, vpid);
4202 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
4203 	test_vmx_valid_controls();
4204 	report_prefix_pop();
4205 
4206 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
4207 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4208 	test_vmx_invalid_controls();
4209 	report_prefix_pop();
4210 
4211 	for (i = 0; i < 16; i++) {
4212 		vpid = (short)1 << i;;
4213 		vmcs_write(VPID, vpid);
4214 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4215 		test_vmx_valid_controls();
4216 		report_prefix_pop();
4217 	}
4218 
4219 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4220 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4221 }
4222 
4223 static void set_vtpr(unsigned vtpr)
4224 {
4225 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
4226 }
4227 
4228 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
4229 {
4230 	bool valid = true;
4231 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4232 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4233 
4234 	if ((primary & CPU_TPR_SHADOW) &&
4235 	    (!(primary & CPU_SECONDARY) ||
4236 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
4237 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
4238 
4239 	set_vtpr(vtpr);
4240 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
4241 	    threshold, (vtpr >> 4) & 0xf);
4242 	if (valid)
4243 		test_vmx_valid_controls();
4244 	else
4245 		test_vmx_invalid_controls();
4246 	report_prefix_pop();
4247 }
4248 
4249 static void test_invalid_event_injection(void)
4250 {
4251 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
4252 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
4253 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
4254 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
4255 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
4256 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4257 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4258 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4259 	u32 cnt;
4260 
4261 	/* Setup */
4262 	report_prefix_push("invalid event injection");
4263 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4264 	vmcs_write(ENT_INST_LEN, 0x00000001);
4265 
4266 	/* The field's interruption type is not set to a reserved value. */
4267 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4268 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4269 			    "RESERVED interruption type invalid [-]",
4270 			    ent_intr_info);
4271 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4272 	test_vmx_invalid_controls();
4273 	report_prefix_pop();
4274 
4275 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4276 			DE_VECTOR;
4277 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4278 			    "RESERVED interruption type invalid [+]",
4279 			    ent_intr_info);
4280 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4281 	test_vmx_valid_controls();
4282 	report_prefix_pop();
4283 
4284 	/* If the interruption type is other event, the vector is 0. */
4285 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4286 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4287 			    "(OTHER EVENT && vector != 0) invalid [-]",
4288 			    ent_intr_info);
4289 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4290 	test_vmx_invalid_controls();
4291 	report_prefix_pop();
4292 
4293 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4294 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4295 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4296 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4297 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4298 	test_vmx_invalid_controls();
4299 	report_prefix_pop();
4300 
4301 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4302 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4303 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4304 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4305 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4306 	test_vmx_valid_controls();
4307 	report_prefix_pop();
4308 
4309 	/*
4310 	 * If the interruption type
4311 	 * is HW exception, the vector is at most 31.
4312 	 */
4313 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4314 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4315 			    "(HW exception && vector > 31) invalid [-]",
4316 			    ent_intr_info);
4317 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4318 	test_vmx_invalid_controls();
4319 	report_prefix_pop();
4320 
4321 	/*
4322 	 * deliver-error-code is 1 iff either
4323 	 * (a) the "unrestricted guest" VM-execution control is 0
4324 	 * (b) CR0.PE is set.
4325 	 */
4326 
4327 	/* Assert that unrestricted guest is disabled or unsupported */
4328 	assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
4329 	       !(secondary_save & CPU_URG));
4330 
4331 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4332 			GP_VECTOR;
4333 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4334 			    "error code <-> (!URG || prot_mode) [-]",
4335 			    ent_intr_info);
4336 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4337 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4338 	test_vmx_invalid_controls();
4339 	report_prefix_pop();
4340 
4341 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4342 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4343 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4344 			    "error code <-> (!URG || prot_mode) [+]",
4345 			    ent_intr_info);
4346 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4347 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4348 	test_vmx_valid_controls();
4349 	report_prefix_pop();
4350 
4351 	if (enable_unrestricted_guest(false))
4352 		goto skip_unrestricted_guest;
4353 
4354 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4355 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4356 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4357 			    "error code <-> (!URG || prot_mode) [-]",
4358 			    ent_intr_info);
4359 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4360 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4361 	test_vmx_invalid_controls();
4362 	report_prefix_pop();
4363 
4364 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4365 			GP_VECTOR;
4366 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4367 			    "error code <-> (!URG || prot_mode) [-]",
4368 			    ent_intr_info);
4369 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4370 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4371 	test_vmx_invalid_controls();
4372 	report_prefix_pop();
4373 
4374 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4375 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4376 
4377 skip_unrestricted_guest:
4378 	vmcs_write(GUEST_CR0, guest_cr0_save);
4379 
4380 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4381 	report_prefix_push("error code <-> HW exception");
4382 	for (cnt = 0; cnt < 8; cnt++) {
4383 		u32 exception_type_mask = cnt << 8;
4384 		u32 deliver_error_code_mask =
4385 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4386 			INTR_INFO_DELIVER_CODE_MASK : 0;
4387 
4388 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4389 				exception_type_mask | GP_VECTOR;
4390 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4391 				    ent_intr_info);
4392 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4393 		test_vmx_invalid_controls();
4394 		report_prefix_pop();
4395 	}
4396 	report_prefix_pop();
4397 
4398 	/*
4399 	 * deliver-error-code is 1 iff the the vector
4400 	 * indicates an exception that would normally deliver an error code
4401 	 */
4402 	report_prefix_push("error code <-> vector delivers error code");
4403 	for (cnt = 0; cnt < 32; cnt++) {
4404 		bool has_error_code = false;
4405 		u32 deliver_error_code_mask;
4406 
4407 		switch (cnt) {
4408 		case DF_VECTOR:
4409 		case TS_VECTOR:
4410 		case NP_VECTOR:
4411 		case SS_VECTOR:
4412 		case GP_VECTOR:
4413 		case PF_VECTOR:
4414 		case AC_VECTOR:
4415 			has_error_code = true;
4416 		case CP_VECTOR:
4417 			/* Some CPUs have error code and some do not, skip */
4418 			continue;
4419 		}
4420 
4421 		/* Negative case */
4422 		deliver_error_code_mask = has_error_code ?
4423 						0 :
4424 						INTR_INFO_DELIVER_CODE_MASK;
4425 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4426 				INTR_TYPE_HARD_EXCEPTION | cnt;
4427 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4428 				    ent_intr_info);
4429 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4430 		test_vmx_invalid_controls();
4431 		report_prefix_pop();
4432 
4433 		/* Positive case */
4434 		deliver_error_code_mask = has_error_code ?
4435 						INTR_INFO_DELIVER_CODE_MASK :
4436 						0;
4437 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4438 				INTR_TYPE_HARD_EXCEPTION | cnt;
4439 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4440 				    ent_intr_info);
4441 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4442 		test_vmx_valid_controls();
4443 		report_prefix_pop();
4444 	}
4445 	report_prefix_pop();
4446 
4447 	/* Reserved bits in the field (30:12) are 0. */
4448 	report_prefix_push("reserved bits clear");
4449 	for (cnt = 12; cnt <= 30; cnt++) {
4450 		ent_intr_info = ent_intr_info_base |
4451 				INTR_INFO_DELIVER_CODE_MASK |
4452 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4453 				(1U << cnt);
4454 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4455 				    ent_intr_info);
4456 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4457 		test_vmx_invalid_controls();
4458 		report_prefix_pop();
4459 	}
4460 	report_prefix_pop();
4461 
4462 	/*
4463 	 * If deliver-error-code is 1
4464 	 * bits 31:16 of the VM-entry exception error-code field are 0.
4465 	 */
4466 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4467 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4468 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4469 			    "VM-entry exception error code[31:16] clear",
4470 			    ent_intr_info);
4471 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4472 	for (cnt = 16; cnt <= 31; cnt++) {
4473 		ent_intr_err = 1U << cnt;
4474 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4475 				    ent_intr_err);
4476 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4477 		test_vmx_invalid_controls();
4478 		report_prefix_pop();
4479 	}
4480 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4481 	report_prefix_pop();
4482 
4483 	/*
4484 	 * If the interruption type is software interrupt, software exception,
4485 	 * or privileged software exception, the VM-entry instruction-length
4486 	 * field is in the range 0 - 15.
4487 	 */
4488 
4489 	for (cnt = 0; cnt < 3; cnt++) {
4490 		switch (cnt) {
4491 		case 0:
4492 			ent_intr_info = ent_intr_info_base |
4493 					INTR_TYPE_SOFT_INTR;
4494 			break;
4495 		case 1:
4496 			ent_intr_info = ent_intr_info_base |
4497 					INTR_TYPE_SOFT_EXCEPTION;
4498 			break;
4499 		case 2:
4500 			ent_intr_info = ent_intr_info_base |
4501 					INTR_TYPE_PRIV_SW_EXCEPTION;
4502 			break;
4503 		}
4504 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4505 				    "VM-entry instruction-length check",
4506 				    ent_intr_info);
4507 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4508 
4509 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4510 		ent_intr_len = -1;
4511 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4512 				    ent_intr_len);
4513 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4514 		test_vmx_invalid_controls();
4515 		report_prefix_pop();
4516 
4517 		/* Instruction length set to 16 should fail */
4518 		ent_intr_len = 0x00000010;
4519 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4520 				    ent_intr_len);
4521 		vmcs_write(ENT_INST_LEN, 0x00000010);
4522 		test_vmx_invalid_controls();
4523 		report_prefix_pop();
4524 
4525 		report_prefix_pop();
4526 	}
4527 
4528 	/* Cleanup */
4529 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4530 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4531 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4532 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4533 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4534 	vmcs_write(GUEST_CR0, guest_cr0_save);
4535 	report_prefix_pop();
4536 }
4537 
4538 /*
4539  * Test interesting vTPR values for a given TPR threshold.
4540  */
4541 static void test_vtpr_values(unsigned threshold)
4542 {
4543 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4544 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4545 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4546 }
4547 
4548 static void try_tpr_threshold(unsigned threshold)
4549 {
4550 	bool valid = true;
4551 
4552 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4553 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4554 
4555 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4556 	    (secondary & CPU_VINTD)))
4557 		valid = !(threshold >> 4);
4558 
4559 	set_vtpr(-1);
4560 	vmcs_write(TPR_THRESHOLD, threshold);
4561 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4562 	if (valid)
4563 		test_vmx_valid_controls();
4564 	else
4565 		test_vmx_invalid_controls();
4566 	report_prefix_pop();
4567 
4568 	if (valid)
4569 		test_vtpr_values(threshold);
4570 }
4571 
4572 /*
4573  * Test interesting TPR threshold values.
4574  */
4575 static void test_tpr_threshold_values(void)
4576 {
4577 	unsigned i;
4578 
4579 	for (i = 0; i < 0x10; i++)
4580 		try_tpr_threshold(i);
4581 	for (i = 4; i < 32; i++)
4582 		try_tpr_threshold(1u << i);
4583 	try_tpr_threshold(-1u);
4584 	try_tpr_threshold(0x7fffffff);
4585 }
4586 
4587 /*
4588  * This test covers the following two VM entry checks:
4589  *
4590  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4591  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4592  *         31:4 of the TPR threshold VM-execution control field must
4593 	   be 0.
4594  *         [Intel SDM]
4595  *
4596  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4597  *          "virtual-interrupt delivery" VM-execution control is 0
4598  *          and the "virtualize APIC accesses" VM-execution control
4599  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4600  *          control field must not be greater than the value of bits
4601  *          7:4 of VTPR.
4602  *          [Intel SDM]
4603  */
4604 static void test_tpr_threshold(void)
4605 {
4606 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4607 	u64 apic_virt_addr = vmcs_read(APIC_VIRT_ADDR);
4608 	u64 threshold = vmcs_read(TPR_THRESHOLD);
4609 	void *virtual_apic_page;
4610 
4611 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4612 		return;
4613 
4614 	virtual_apic_page = alloc_page();
4615 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4616 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4617 
4618 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4619 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4620 	test_tpr_threshold_values();
4621 	report_prefix_pop();
4622 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4623 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4624 	test_tpr_threshold_values();
4625 	report_prefix_pop();
4626 
4627 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4628 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES))))
4629 		goto out;
4630 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4631 
4632 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4633 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4634 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4635 		test_tpr_threshold_values();
4636 		report_prefix_pop();
4637 
4638 		vmcs_write(CPU_EXEC_CTRL0,
4639 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4640 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4641 		test_tpr_threshold_values();
4642 		report_prefix_pop();
4643 	}
4644 
4645 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4646 		vmcs_write(CPU_EXEC_CTRL0,
4647 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4648 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4649 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4650 		test_tpr_threshold_values();
4651 		report_prefix_pop();
4652 
4653 		vmcs_write(CPU_EXEC_CTRL0,
4654 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4655 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4656 		test_tpr_threshold_values();
4657 		report_prefix_pop();
4658 	}
4659 
4660 	if ((ctrl_cpu_rev[1].clr &
4661 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4662 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4663 		vmcs_write(CPU_EXEC_CTRL0,
4664 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4665 		vmcs_write(CPU_EXEC_CTRL1,
4666 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4667 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4668 		test_tpr_threshold_values();
4669 		report_prefix_pop();
4670 
4671 		vmcs_write(CPU_EXEC_CTRL0,
4672 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4673 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4674 		test_tpr_threshold_values();
4675 		report_prefix_pop();
4676 	}
4677 
4678 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4679 out:
4680 	vmcs_write(TPR_THRESHOLD, threshold);
4681 	vmcs_write(APIC_VIRT_ADDR, apic_virt_addr);
4682 	vmcs_write(CPU_EXEC_CTRL0, primary);
4683 }
4684 
4685 /*
4686  * This test verifies the following two vmentry checks:
4687  *
4688  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4689  *  VM-execution control must be 0.
4690  *  [Intel SDM]
4691  *
4692  *  If the "virtual NMIs" VM-execution control is 0, the "NMI-window
4693  *  exiting" VM-execution control must be 0.
4694  *  [Intel SDM]
4695  */
4696 static void test_nmi_ctrls(void)
4697 {
4698 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4699 
4700 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4701 	    (PIN_NMI | PIN_VIRT_NMI)) {
4702 		printf("NMI exiting and Virtual NMIs are not supported !\n");
4703 		return;
4704 	}
4705 
4706 	/* Save the controls so that we can restore them after our tests */
4707 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4708 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4709 
4710 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4711 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4712 
4713 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4714 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4715 	test_vmx_valid_controls();
4716 	report_prefix_pop();
4717 
4718 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4719 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4720 	test_vmx_invalid_controls();
4721 	report_prefix_pop();
4722 
4723 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4724 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4725 	test_vmx_valid_controls();
4726 	report_prefix_pop();
4727 
4728 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4729 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4730 	test_vmx_valid_controls();
4731 	report_prefix_pop();
4732 
4733 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4734 		report_info("NMI-window exiting is not supported, skipping...");
4735 		goto done;
4736 	}
4737 
4738 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4739 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4740 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4741 	test_vmx_invalid_controls();
4742 	report_prefix_pop();
4743 
4744 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4745 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4746 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4747 	test_vmx_valid_controls();
4748 	report_prefix_pop();
4749 
4750 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4751 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4752 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4753 	test_vmx_valid_controls();
4754 	report_prefix_pop();
4755 
4756 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4757 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4758 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4759 	test_vmx_valid_controls();
4760 	report_prefix_pop();
4761 
4762 	/* Restore the controls to their original values */
4763 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4764 done:
4765 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4766 }
4767 
4768 static void test_eptp_ad_bit(u64 eptp, bool ctrl)
4769 {
4770 	vmcs_write(EPTP, eptp);
4771 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4772 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4773 	if (ctrl)
4774 		test_vmx_valid_controls();
4775 	else
4776 		test_vmx_invalid_controls();
4777 	report_prefix_pop();
4778 
4779 }
4780 
4781 /*
4782  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4783  *    control field must satisfy the following checks:
4784  *
4785  *     - The EPT memory type (bits 2:0) must be a value supported by the
4786  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4787  *     - Bits 5:3 (1 less than the EPT page-walk length) must indicate a
4788  *	 supported EPT page-walk length.
4789  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4790  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4791  *	 indicating that the processor does not support accessed and dirty
4792  *	 dirty flags for EPT.
4793  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4794  *	 physical-address width) must all be 0.
4795  *
4796  * 2. If the "unrestricted guest" VM-execution control is 1, the
4797  *    "enable EPT" VM-execution control must also be 1.
4798  */
4799 static void test_ept_eptp(void)
4800 {
4801 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4802 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4803 	u64 eptp_saved = vmcs_read(EPTP);
4804 	u32 primary = primary_saved;
4805 	u32 secondary = secondary_saved;
4806 	u64 msr, eptp = eptp_saved;
4807 	bool un_cache = false;
4808 	bool wr_bk = false;
4809 	bool ctrl;
4810 	u32 i, maxphysaddr;
4811 	u64 j, resv_bits_mask = 0;
4812 
4813 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4814 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4815 		printf("\"CPU secondary\" and/or \"enable EPT\" execution controls are not supported !\n");
4816 		return;
4817 	}
4818 
4819 	/*
4820 	 * Memory type (bits 2:0)
4821 	 */
4822 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
4823 	if (msr & EPT_CAP_UC)
4824 		un_cache = true;
4825 	if (msr & EPT_CAP_WB)
4826 		wr_bk = true;
4827 
4828 	/* Support for 4-level EPT is mandatory. */
4829 	report(msr & EPT_CAP_PWL4, "4-level EPT support check");
4830 
4831 	primary |= CPU_SECONDARY;
4832 	vmcs_write(CPU_EXEC_CTRL0, primary);
4833 	secondary |= CPU_EPT;
4834 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4835 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4836 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4837 	vmcs_write(EPTP, eptp);
4838 
4839 	for (i = 0; i < 8; i++) {
4840 		if (i == 0) {
4841 			if (un_cache) {
4842 				report_info("EPT paging structure memory-type is Un-cacheable\n");
4843 				ctrl = true;
4844 			} else {
4845 				ctrl = false;
4846 			}
4847 		} else if (i == 6) {
4848 			if (wr_bk) {
4849 				report_info("EPT paging structure memory-type is Write-back\n");
4850 				ctrl = true;
4851 			} else {
4852 				ctrl = false;
4853 			}
4854 		} else {
4855 			ctrl = false;
4856 		}
4857 
4858 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4859 		vmcs_write(EPTP, eptp);
4860 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4861 		    eptp & EPT_MEM_TYPE_MASK);
4862 		if (ctrl)
4863 			test_vmx_valid_controls();
4864 		else
4865 			test_vmx_invalid_controls();
4866 		report_prefix_pop();
4867 	}
4868 
4869 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4870 
4871 	/*
4872 	 * Page walk length (bits 5:3).  Note, the value in VMCS.EPTP "is 1
4873 	 * less than the EPT page-walk length".
4874 	 */
4875 	for (i = 0; i < 8; i++) {
4876 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4877 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4878 		if (i == 3 || (i == 4 && (msr & EPT_CAP_PWL5)))
4879 			ctrl = true;
4880 		else
4881 			ctrl = false;
4882 
4883 		vmcs_write(EPTP, eptp);
4884 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4885 		    eptp & EPTP_PG_WALK_LEN_MASK);
4886 		if (ctrl)
4887 			test_vmx_valid_controls();
4888 		else
4889 			test_vmx_invalid_controls();
4890 		report_prefix_pop();
4891 	}
4892 
4893 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4894 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4895 
4896 	/*
4897 	 * Accessed and dirty flag (bit 6)
4898 	 */
4899 	if (msr & EPT_CAP_AD_FLAG) {
4900 		report_info("Processor supports accessed and dirty flag");
4901 		eptp &= ~EPTP_AD_FLAG;
4902 		test_eptp_ad_bit(eptp, true);
4903 
4904 		eptp |= EPTP_AD_FLAG;
4905 		test_eptp_ad_bit(eptp, true);
4906 	} else {
4907 		report_info("Processor does not supports accessed and dirty flag");
4908 		eptp &= ~EPTP_AD_FLAG;
4909 		test_eptp_ad_bit(eptp, true);
4910 
4911 		eptp |= EPTP_AD_FLAG;
4912 		test_eptp_ad_bit(eptp, false);
4913 	}
4914 
4915 	/*
4916 	 * Reserved bits [11:7] and [63:N]
4917 	 */
4918 	for (i = 0; i < 32; i++) {
4919 		eptp = (eptp &
4920 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4921 		    (i << EPTP_RESERV_BITS_SHIFT);
4922 		vmcs_write(EPTP, eptp);
4923 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4924 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4925 		    EPTP_RESERV_BITS_MASK);
4926 		if (i == 0)
4927 			test_vmx_valid_controls();
4928 		else
4929 			test_vmx_invalid_controls();
4930 		report_prefix_pop();
4931 	}
4932 
4933 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4934 
4935 	maxphysaddr = cpuid_maxphyaddr();
4936 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4937 		resv_bits_mask |= 1ul << i;
4938 	}
4939 
4940 	for (j = maxphysaddr - 1; j <= 63; j++) {
4941 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4942 		    (j < maxphysaddr ? 0 : 1ul << j);
4943 		vmcs_write(EPTP, eptp);
4944 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4945 		    (eptp >> maxphysaddr) & resv_bits_mask);
4946 		if (j < maxphysaddr)
4947 			test_vmx_valid_controls();
4948 		else
4949 			test_vmx_invalid_controls();
4950 		report_prefix_pop();
4951 	}
4952 
4953 	secondary &= ~(CPU_EPT | CPU_URG);
4954 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4955 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4956 	test_vmx_valid_controls();
4957 	report_prefix_pop();
4958 
4959 	if (!(ctrl_cpu_rev[1].clr & CPU_URG))
4960 		goto skip_unrestricted_guest;
4961 
4962 	secondary |= CPU_URG;
4963 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4964 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4965 	test_vmx_invalid_controls();
4966 	report_prefix_pop();
4967 
4968 	secondary |= CPU_EPT;
4969 	setup_dummy_ept();
4970 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4971 	test_vmx_valid_controls();
4972 	report_prefix_pop();
4973 
4974 skip_unrestricted_guest:
4975 	secondary &= ~CPU_URG;
4976 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4977 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4978 	test_vmx_valid_controls();
4979 	report_prefix_pop();
4980 
4981 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4982 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4983 	vmcs_write(EPTP, eptp_saved);
4984 }
4985 
4986 /*
4987  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4988  * VM-execution control must also be 1. In addition, the PML address
4989  * must satisfy the following checks:
4990  *
4991  *    * Bits 11:0 of the address must be 0.
4992  *    * The address should not set any bits beyond the processor's
4993  *	physical-address width.
4994  *
4995  *  [Intel SDM]
4996  */
4997 static void test_pml(void)
4998 {
4999 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
5000 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
5001 	u32 primary = primary_saved;
5002 	u32 secondary = secondary_saved;
5003 
5004 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
5005 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
5006 		printf("\"Secondary execution\" control or \"enable EPT\" control or \"enable PML\" control is not supported !\n");
5007 		return;
5008 	}
5009 
5010 	primary |= CPU_SECONDARY;
5011 	vmcs_write(CPU_EXEC_CTRL0, primary);
5012 	secondary &= ~(CPU_PML | CPU_EPT);
5013 	vmcs_write(CPU_EXEC_CTRL1, secondary);
5014 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
5015 	test_vmx_valid_controls();
5016 	report_prefix_pop();
5017 
5018 	secondary |= CPU_PML;
5019 	vmcs_write(CPU_EXEC_CTRL1, secondary);
5020 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
5021 	test_vmx_invalid_controls();
5022 	report_prefix_pop();
5023 
5024 	secondary |= CPU_EPT;
5025 	setup_dummy_ept();
5026 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
5027 	test_vmx_valid_controls();
5028 	report_prefix_pop();
5029 
5030 	secondary &= ~CPU_PML;
5031 	vmcs_write(CPU_EXEC_CTRL1, secondary);
5032 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
5033 	test_vmx_valid_controls();
5034 	report_prefix_pop();
5035 
5036 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
5037 				 PAGE_SIZE, false, false);
5038 
5039 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
5040 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
5041 }
5042 
5043  /*
5044  * If the "activate VMX-preemption timer" VM-execution control is 0, the
5045  * the "save VMX-preemption timer value" VM-exit control must also be 0.
5046  *
5047  *  [Intel SDM]
5048  */
5049 static void test_vmx_preemption_timer(void)
5050 {
5051 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
5052 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
5053 	u32 pin = saved_pin;
5054 	u32 exit = saved_exit;
5055 
5056 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
5057 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
5058 		printf("\"Save-VMX-preemption-timer\" control and/or \"Enable-VMX-preemption-timer\" control is not supported\n");
5059 		return;
5060 	}
5061 
5062 	pin |= PIN_PREEMPT;
5063 	vmcs_write(PIN_CONTROLS, pin);
5064 	exit &= ~EXI_SAVE_PREEMPT;
5065 	vmcs_write(EXI_CONTROLS, exit);
5066 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
5067 	test_vmx_valid_controls();
5068 	report_prefix_pop();
5069 
5070 	exit |= EXI_SAVE_PREEMPT;
5071 	vmcs_write(EXI_CONTROLS, exit);
5072 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
5073 	test_vmx_valid_controls();
5074 	report_prefix_pop();
5075 
5076 	pin &= ~PIN_PREEMPT;
5077 	vmcs_write(PIN_CONTROLS, pin);
5078 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
5079 	test_vmx_invalid_controls();
5080 	report_prefix_pop();
5081 
5082 	exit &= ~EXI_SAVE_PREEMPT;
5083 	vmcs_write(EXI_CONTROLS, exit);
5084 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
5085 	test_vmx_valid_controls();
5086 	report_prefix_pop();
5087 
5088 	vmcs_write(PIN_CONTROLS, saved_pin);
5089 	vmcs_write(EXI_CONTROLS, saved_exit);
5090 }
5091 
5092 extern unsigned char test_mtf1;
5093 extern unsigned char test_mtf2;
5094 extern unsigned char test_mtf3;
5095 extern unsigned char test_mtf4;
5096 
5097 static void test_mtf_guest(void)
5098 {
5099 	asm ("vmcall;\n\t"
5100 	     "out %al, $0x80;\n\t"
5101 	     "test_mtf1:\n\t"
5102 	     "vmcall;\n\t"
5103 	     "out %al, $0x80;\n\t"
5104 	     "test_mtf2:\n\t"
5105 	     /*
5106 	      * Prepare for the 'MOV CR3' test. Attempt to induce a
5107 	      * general-protection fault by moving a non-canonical address into
5108 	      * CR3. The 'MOV CR3' instruction does not take an imm64 operand,
5109 	      * so we must MOV the desired value into a register first.
5110 	      *
5111 	      * MOV RAX is done before the VMCALL such that MTF is only enabled
5112 	      * for the instruction under test.
5113 	      */
5114 	     "mov $0xaaaaaaaaaaaaaaaa, %rax;\n\t"
5115 	     "vmcall;\n\t"
5116 	     "mov %rax, %cr3;\n\t"
5117 	     "test_mtf3:\n\t"
5118 	     "vmcall;\n\t"
5119 	     /*
5120 	      * ICEBP/INT1 instruction. Though the instruction is now
5121 	      * documented, don't rely on assemblers enumerating the
5122 	      * instruction. Resort to hand assembly.
5123 	      */
5124 	     ".byte 0xf1;\n\t"
5125 	     "vmcall;\n\t"
5126 	     "test_mtf4:\n\t"
5127 	     "mov $0, %eax;\n\t");
5128 }
5129 
5130 static void test_mtf_gp_handler(struct ex_regs *regs)
5131 {
5132 	regs->rip = (unsigned long) &test_mtf3;
5133 }
5134 
5135 static void test_mtf_db_handler(struct ex_regs *regs)
5136 {
5137 }
5138 
5139 static void enable_mtf(void)
5140 {
5141 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5142 
5143 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 | CPU_MTF);
5144 }
5145 
5146 static void disable_mtf(void)
5147 {
5148 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5149 
5150 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 & ~CPU_MTF);
5151 }
5152 
5153 static void enable_tf(void)
5154 {
5155 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5156 
5157 	vmcs_write(GUEST_RFLAGS, rflags | X86_EFLAGS_TF);
5158 }
5159 
5160 static void disable_tf(void)
5161 {
5162 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5163 
5164 	vmcs_write(GUEST_RFLAGS, rflags & ~X86_EFLAGS_TF);
5165 }
5166 
5167 static void report_mtf(const char *insn_name, unsigned long exp_rip)
5168 {
5169 	unsigned long rip = vmcs_read(GUEST_RIP);
5170 
5171 	assert_exit_reason(VMX_MTF);
5172 	report(rip == exp_rip, "MTF VM-exit after %s. RIP: 0x%lx (expected 0x%lx)",
5173 	       insn_name, rip, exp_rip);
5174 }
5175 
5176 static void vmx_mtf_test(void)
5177 {
5178 	unsigned long pending_dbg;
5179 	handler old_gp, old_db;
5180 
5181 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5182 		printf("CPU does not support the 'monitor trap flag' processor-based VM-execution control.\n");
5183 		return;
5184 	}
5185 
5186 	test_set_guest(test_mtf_guest);
5187 
5188 	/* Expect an MTF VM-exit after OUT instruction */
5189 	enter_guest();
5190 	skip_exit_vmcall();
5191 
5192 	enable_mtf();
5193 	enter_guest();
5194 	report_mtf("OUT", (unsigned long) &test_mtf1);
5195 	disable_mtf();
5196 
5197 	/*
5198 	 * Concurrent #DB trap and MTF on instruction boundary. Expect MTF
5199 	 * VM-exit with populated 'pending debug exceptions' VMCS field.
5200 	 */
5201 	enter_guest();
5202 	skip_exit_vmcall();
5203 
5204 	enable_mtf();
5205 	enable_tf();
5206 
5207 	enter_guest();
5208 	report_mtf("OUT", (unsigned long) &test_mtf2);
5209 	pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
5210 	report(pending_dbg & DR_STEP,
5211 	       "'pending debug exceptions' field after MTF VM-exit: 0x%lx (expected 0x%lx)",
5212 	       pending_dbg, (unsigned long) DR_STEP);
5213 
5214 	disable_mtf();
5215 	disable_tf();
5216 	vmcs_write(GUEST_PENDING_DEBUG, 0);
5217 
5218 	/*
5219 	 * #GP exception takes priority over MTF. Expect MTF VM-exit with RIP
5220 	 * advanced to first instruction of #GP handler.
5221 	 */
5222 	enter_guest();
5223 	skip_exit_vmcall();
5224 
5225 	old_gp = handle_exception(GP_VECTOR, test_mtf_gp_handler);
5226 
5227 	enable_mtf();
5228 	enter_guest();
5229 	report_mtf("MOV CR3", (unsigned long) get_idt_addr(&boot_idt[GP_VECTOR]));
5230 	disable_mtf();
5231 
5232 	/*
5233 	 * Concurrent MTF and privileged software exception (i.e. ICEBP/INT1).
5234 	 * MTF should follow the delivery of #DB trap, though the SDM doesn't
5235 	 * provide clear indication of the relative priority.
5236 	 */
5237 	enter_guest();
5238 	skip_exit_vmcall();
5239 
5240 	handle_exception(GP_VECTOR, old_gp);
5241 	old_db = handle_exception(DB_VECTOR, test_mtf_db_handler);
5242 
5243 	enable_mtf();
5244 	enter_guest();
5245 	report_mtf("INT1", (unsigned long) get_idt_addr(&boot_idt[DB_VECTOR]));
5246 	disable_mtf();
5247 
5248 	enter_guest();
5249 	skip_exit_vmcall();
5250 	handle_exception(DB_VECTOR, old_db);
5251 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK | INTR_TYPE_OTHER_EVENT);
5252 	enter_guest();
5253 	report_mtf("injected MTF", (unsigned long) &test_mtf4);
5254 	enter_guest();
5255 }
5256 
5257 extern char vmx_mtf_pdpte_guest_begin;
5258 extern char vmx_mtf_pdpte_guest_end;
5259 
5260 asm("vmx_mtf_pdpte_guest_begin:\n\t"
5261     "mov %cr0, %rax\n\t"    /* save CR0 with PG=1                 */
5262     "vmcall\n\t"            /* on return from this CR0.PG=0       */
5263     "mov %rax, %cr0\n\t"    /* restore CR0.PG=1 to enter PAE mode */
5264     "vmcall\n\t"
5265     "retq\n\t"
5266     "vmx_mtf_pdpte_guest_end:");
5267 
5268 static void vmx_mtf_pdpte_test(void)
5269 {
5270 	void *test_mtf_pdpte_guest;
5271 	pteval_t *pdpt;
5272 	u32 guest_ar_cs;
5273 	u64 guest_efer;
5274 	pteval_t *pte;
5275 	u64 guest_cr0;
5276 	u64 guest_cr3;
5277 	u64 guest_cr4;
5278 	u64 ent_ctls;
5279 	int i;
5280 
5281 	if (setup_ept(false))
5282 		return;
5283 
5284 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5285 		printf("CPU does not support 'monitor trap flag.'\n");
5286 		return;
5287 	}
5288 
5289 	if (!(ctrl_cpu_rev[1].clr & CPU_URG)) {
5290 		printf("CPU does not support 'unrestricted guest.'\n");
5291 		return;
5292 	}
5293 
5294 	vmcs_write(EXC_BITMAP, ~0);
5295 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
5296 
5297 	/*
5298 	 * Copy the guest code to an identity-mapped page.
5299 	 */
5300 	test_mtf_pdpte_guest = alloc_page();
5301 	memcpy(test_mtf_pdpte_guest, &vmx_mtf_pdpte_guest_begin,
5302 	       &vmx_mtf_pdpte_guest_end - &vmx_mtf_pdpte_guest_begin);
5303 
5304 	test_set_guest(test_mtf_pdpte_guest);
5305 
5306 	enter_guest();
5307 	skip_exit_vmcall();
5308 
5309 	/*
5310 	 * Put the guest in non-paged 32-bit protected mode, ready to enter
5311 	 * PAE mode when CR0.PG is set. CR4.PAE will already have been set
5312 	 * when the guest started out in long mode.
5313 	 */
5314 	ent_ctls = vmcs_read(ENT_CONTROLS);
5315 	vmcs_write(ENT_CONTROLS, ent_ctls & ~ENT_GUEST_64);
5316 
5317 	guest_efer = vmcs_read(GUEST_EFER);
5318 	vmcs_write(GUEST_EFER, guest_efer & ~(EFER_LMA | EFER_LME));
5319 
5320 	/*
5321 	 * Set CS access rights bits for 32-bit protected mode:
5322 	 * 3:0    B execute/read/accessed
5323 	 * 4      1 code or data
5324 	 * 6:5    0 descriptor privilege level
5325 	 * 7      1 present
5326 	 * 11:8   0 reserved
5327 	 * 12     0 available for use by system software
5328 	 * 13     0 64 bit mode not active
5329 	 * 14     1 default operation size 32-bit segment
5330 	 * 15     1 page granularity: segment limit in 4K units
5331 	 * 16     0 segment usable
5332 	 * 31:17  0 reserved
5333 	 */
5334 	guest_ar_cs = vmcs_read(GUEST_AR_CS);
5335 	vmcs_write(GUEST_AR_CS, 0xc09b);
5336 
5337 	guest_cr0 = vmcs_read(GUEST_CR0);
5338 	vmcs_write(GUEST_CR0, guest_cr0 & ~X86_CR0_PG);
5339 
5340 	guest_cr4 = vmcs_read(GUEST_CR4);
5341 	vmcs_write(GUEST_CR4, guest_cr4 & ~X86_CR4_PCIDE);
5342 
5343 	guest_cr3 = vmcs_read(GUEST_CR3);
5344 
5345 	/*
5346 	 * Turn the 4-level page table into a PAE page table by following the 0th
5347 	 * PML4 entry to a PDPT page, and grab the first four PDPTEs from that
5348 	 * page.
5349 	 *
5350 	 * Why does this work?
5351 	 *
5352 	 * PAE uses 32-bit addressing which implies:
5353 	 * Bits 11:0   page offset
5354 	 * Bits 20:12  entry into 512-entry page table
5355 	 * Bits 29:21  entry into a 512-entry directory table
5356 	 * Bits 31:30  entry into the page directory pointer table.
5357 	 * Bits 63:32  zero
5358 	 *
5359 	 * As only 2 bits are needed to select the PDPTEs for the entire
5360 	 * 32-bit address space, take the first 4 PDPTEs in the level 3 page
5361 	 * directory pointer table. It doesn't matter which of these PDPTEs
5362 	 * are present because they must cover the guest code given that it
5363 	 * has already run successfully.
5364 	 *
5365 	 * Get a pointer to PTE for GVA=0 in the page directory pointer table
5366 	 */
5367 	pte = get_pte_level(
5368             (pgd_t *)phys_to_virt(guest_cr3 & ~X86_CR3_PCID_MASK), 0,
5369             PDPT_LEVEL);
5370 
5371 	/*
5372 	 * Need some memory for the 4-entry PAE page directory pointer
5373 	 * table. Use the end of the identity-mapped page where the guest code
5374 	 * is stored. There is definitely space as the guest code is only a
5375 	 * few bytes.
5376 	 */
5377 	pdpt = test_mtf_pdpte_guest + PAGE_SIZE - 4 * sizeof(pteval_t);
5378 
5379 	/*
5380 	 * Copy the first four PDPTEs into the PAE page table with reserved
5381 	 * bits cleared. Note that permission bits from the PML4E and PDPTE
5382 	 * are not propagated.
5383 	 */
5384 	for (i = 0; i < 4; i++) {
5385 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_RSVD_MASK),
5386 				   "PDPTE has invalid reserved bits");
5387 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_PAGE_SIZE_MASK),
5388 				   "Cannot use 1GB super pages for PAE");
5389 		pdpt[i] = pte[i] & ~(PAE_PDPTE_RSVD_MASK);
5390 	}
5391 	vmcs_write(GUEST_CR3, virt_to_phys(pdpt));
5392 
5393 	enable_mtf();
5394 	enter_guest();
5395 	assert_exit_reason(VMX_MTF);
5396 	disable_mtf();
5397 
5398 	/*
5399 	 * The four PDPTEs should have been loaded into the VMCS when
5400 	 * the guest set CR0.PG to enter PAE mode.
5401 	 */
5402 	for (i = 0; i < 4; i++) {
5403 		u64 pdpte = vmcs_read(GUEST_PDPTE + 2 * i);
5404 
5405 		report(pdpte == pdpt[i], "PDPTE%d is 0x%lx (expected 0x%lx)",
5406 		       i, pdpte, pdpt[i]);
5407 	}
5408 
5409 	/*
5410 	 * Now, try to enter the guest in PAE mode. If the PDPTEs in the
5411 	 * vmcs are wrong, this will fail.
5412 	 */
5413 	enter_guest();
5414 	skip_exit_vmcall();
5415 
5416 	/*
5417 	 * Return guest to 64-bit mode and wrap up.
5418 	 */
5419 	vmcs_write(ENT_CONTROLS, ent_ctls);
5420 	vmcs_write(GUEST_EFER, guest_efer);
5421 	vmcs_write(GUEST_AR_CS, guest_ar_cs);
5422 	vmcs_write(GUEST_CR0, guest_cr0);
5423 	vmcs_write(GUEST_CR4, guest_cr4);
5424 	vmcs_write(GUEST_CR3, guest_cr3);
5425 
5426 	enter_guest();
5427 }
5428 
5429 /*
5430  * Tests for VM-execution control fields
5431  */
5432 static void test_vm_execution_ctls(void)
5433 {
5434 	test_pin_based_ctls();
5435 	test_primary_processor_based_ctls();
5436 	test_secondary_processor_based_ctls();
5437 	test_cr3_targets();
5438 	test_io_bitmaps();
5439 	test_msr_bitmap();
5440 	test_apic_ctls();
5441 	test_tpr_threshold();
5442 	test_nmi_ctrls();
5443 	test_pml();
5444 	test_vpid();
5445 	test_ept_eptp();
5446 	test_vmx_preemption_timer();
5447 }
5448 
5449  /*
5450   * The following checks are performed for the VM-entry MSR-load address if
5451   * the VM-entry MSR-load count field is non-zero:
5452   *
5453   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
5454   *      The address should not set any bits beyond the processor's
5455   *      physical-address width.
5456   *
5457   *    - The address of the last byte in the VM-entry MSR-load area
5458   *      should not set any bits beyond the processor's physical-address
5459   *      width. The address of this last byte is VM-entry MSR-load address
5460   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5461   *      uses more bits than the processor's physical-address width.)
5462   *
5463   *
5464   *  [Intel SDM]
5465   */
5466 static void test_entry_msr_load(void)
5467 {
5468 	entry_msr_load = alloc_page();
5469 	u64 tmp;
5470 	u32 entry_msr_ld_cnt = 1;
5471 	int i;
5472 	u32 addr_len = 64;
5473 
5474 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
5475 
5476 	/* Check first 4 bits of VM-entry MSR-load address */
5477 	for (i = 0; i < 4; i++) {
5478 		tmp = (u64)entry_msr_load | 1ull << i;
5479 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
5480 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
5481 				    tmp & 0xf);
5482 		test_vmx_invalid_controls();
5483 		report_prefix_pop();
5484 	}
5485 
5486 	if (basic.val & (1ul << 48))
5487 		addr_len = 32;
5488 
5489 	test_vmcs_addr_values("VM-entry-MSR-load address",
5490 				ENTER_MSR_LD_ADDR, 16, false, false,
5491 				4, addr_len - 1);
5492 
5493 	/*
5494 	 * Check last byte of VM-entry MSR-load address
5495 	 */
5496 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
5497 
5498 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5499 							i < 64; i++) {
5500 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
5501 			1ul << i;
5502 		vmcs_write(ENTER_MSR_LD_ADDR,
5503 			   tmp - (entry_msr_ld_cnt * 16 - 1));
5504 		test_vmx_invalid_controls();
5505 	}
5506 
5507 	vmcs_write(ENT_MSR_LD_CNT, 2);
5508 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5509 	test_vmx_invalid_controls();
5510 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5511 	test_vmx_valid_controls();
5512 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5513 	test_vmx_valid_controls();
5514 }
5515 
5516 static struct vmx_state_area_test_data {
5517 	u32 msr;
5518 	u64 exp;
5519 	bool enabled;
5520 } vmx_state_area_test_data;
5521 
5522 static void guest_state_test_main(void)
5523 {
5524 	u64 obs;
5525 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
5526 
5527 	while (1) {
5528 		if (vmx_get_test_stage() == 2)
5529 			break;
5530 
5531 		if (data->enabled) {
5532 			obs = rdmsr(data->msr);
5533 			report(data->exp == obs,
5534 			       "Guest state is 0x%lx (expected 0x%lx)",
5535 			       obs, data->exp);
5536 		}
5537 
5538 		vmcall();
5539 	}
5540 
5541 	asm volatile("fnop");
5542 }
5543 
5544 static void test_guest_state(const char *test, bool xfail, u64 field,
5545 			     const char * field_name)
5546 {
5547 	struct vmentry_result result;
5548 	u8 abort_flags;
5549 
5550 	abort_flags = ABORT_ON_EARLY_VMENTRY_FAIL;
5551 	if (!xfail)
5552 		abort_flags = ABORT_ON_INVALID_GUEST_STATE;
5553 
5554 	__enter_guest(abort_flags, &result);
5555 
5556 	report(result.exit_reason.failed_vmentry == xfail &&
5557 	       ((xfail && result.exit_reason.basic == VMX_FAIL_STATE) ||
5558 	        (!xfail && result.exit_reason.basic == VMX_VMCALL)) &&
5559 		(!xfail || vmcs_read(EXI_QUALIFICATION) == ENTRY_FAIL_DEFAULT),
5560 	        "%s, %s = %lx", test, field_name, field);
5561 
5562 	if (!result.exit_reason.failed_vmentry)
5563 		skip_exit_insn();
5564 }
5565 
5566 /*
5567  * Tests for VM-entry control fields
5568  */
5569 static void test_vm_entry_ctls(void)
5570 {
5571 	test_invalid_event_injection();
5572 	test_entry_msr_load();
5573 }
5574 
5575 /*
5576  * The following checks are performed for the VM-exit MSR-store address if
5577  * the VM-exit MSR-store count field is non-zero:
5578  *
5579  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
5580  *      The address should not set any bits beyond the processor's
5581  *      physical-address width.
5582  *
5583  *    - The address of the last byte in the VM-exit MSR-store area
5584  *      should not set any bits beyond the processor's physical-address
5585  *      width. The address of this last byte is VM-exit MSR-store address
5586  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5587  *      uses more bits than the processor's physical-address width.)
5588  *
5589  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
5590  * in the range 63:32.
5591  *
5592  *  [Intel SDM]
5593  */
5594 static void test_exit_msr_store(void)
5595 {
5596 	exit_msr_store = alloc_page();
5597 	u64 tmp;
5598 	u32 exit_msr_st_cnt = 1;
5599 	int i;
5600 	u32 addr_len = 64;
5601 
5602 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
5603 
5604 	/* Check first 4 bits of VM-exit MSR-store address */
5605 	for (i = 0; i < 4; i++) {
5606 		tmp = (u64)exit_msr_store | 1ull << i;
5607 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
5608 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
5609 				    tmp & 0xf);
5610 		test_vmx_invalid_controls();
5611 		report_prefix_pop();
5612 	}
5613 
5614 	if (basic.val & (1ul << 48))
5615 		addr_len = 32;
5616 
5617 	test_vmcs_addr_values("VM-exit-MSR-store address",
5618 				EXIT_MSR_ST_ADDR, 16, false, false,
5619 				4, addr_len - 1);
5620 
5621 	/*
5622 	 * Check last byte of VM-exit MSR-store address
5623 	 */
5624 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
5625 
5626 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5627 							i < 64; i++) {
5628 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
5629 			1ul << i;
5630 		vmcs_write(EXIT_MSR_ST_ADDR,
5631 			   tmp - (exit_msr_st_cnt * 16 - 1));
5632 		test_vmx_invalid_controls();
5633 	}
5634 
5635 	vmcs_write(EXI_MSR_ST_CNT, 2);
5636 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5637 	test_vmx_invalid_controls();
5638 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5639 	test_vmx_valid_controls();
5640 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5641 	test_vmx_valid_controls();
5642 }
5643 
5644 /*
5645  * Tests for VM-exit controls
5646  */
5647 static void test_vm_exit_ctls(void)
5648 {
5649 	test_exit_msr_store();
5650 }
5651 
5652 /*
5653  * Check that the virtual CPU checks all of the VMX controls as
5654  * documented in the Intel SDM.
5655  */
5656 static void vmx_controls_test(void)
5657 {
5658 	/*
5659 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
5660 	 * fail due to invalid guest state, should we make it that
5661 	 * far.
5662 	 */
5663 	vmcs_write(GUEST_RFLAGS, 0);
5664 
5665 	test_vm_execution_ctls();
5666 	test_vm_exit_ctls();
5667 	test_vm_entry_ctls();
5668 }
5669 
5670 struct apic_reg_virt_config {
5671 	bool apic_register_virtualization;
5672 	bool use_tpr_shadow;
5673 	bool virtualize_apic_accesses;
5674 	bool virtualize_x2apic_mode;
5675 	bool activate_secondary_controls;
5676 };
5677 
5678 struct apic_reg_test {
5679 	const char *name;
5680 	struct apic_reg_virt_config apic_reg_virt_config;
5681 };
5682 
5683 struct apic_reg_virt_expectation {
5684 	enum Reason rd_exit_reason;
5685 	enum Reason wr_exit_reason;
5686 	u32 val;
5687 	u32 (*virt_fn)(u32);
5688 
5689 	/*
5690 	 * If false, accessing the APIC access address from L2 is treated as a
5691 	 * normal memory operation, rather than triggering virtualization.
5692 	 */
5693 	bool virtualize_apic_accesses;
5694 };
5695 
5696 static u32 apic_virt_identity(u32 val)
5697 {
5698 	return val;
5699 }
5700 
5701 static u32 apic_virt_nibble1(u32 val)
5702 {
5703 	return val & 0xf0;
5704 }
5705 
5706 static u32 apic_virt_byte3(u32 val)
5707 {
5708 	return val & (0xff << 24);
5709 }
5710 
5711 static bool apic_reg_virt_exit_expectation(
5712 	u32 reg, struct apic_reg_virt_config *config,
5713 	struct apic_reg_virt_expectation *expectation)
5714 {
5715 	/* Good configs, where some L2 APIC accesses are virtualized. */
5716 	bool virtualize_apic_accesses_only =
5717 		config->virtualize_apic_accesses &&
5718 		!config->use_tpr_shadow &&
5719 		!config->apic_register_virtualization &&
5720 		!config->virtualize_x2apic_mode &&
5721 		config->activate_secondary_controls;
5722 	bool virtualize_apic_accesses_and_use_tpr_shadow =
5723 		config->virtualize_apic_accesses &&
5724 		config->use_tpr_shadow &&
5725 		!config->apic_register_virtualization &&
5726 		!config->virtualize_x2apic_mode &&
5727 		config->activate_secondary_controls;
5728 	bool apic_register_virtualization =
5729 		config->virtualize_apic_accesses &&
5730 		config->use_tpr_shadow &&
5731 		config->apic_register_virtualization &&
5732 		!config->virtualize_x2apic_mode &&
5733 		config->activate_secondary_controls;
5734 
5735 	expectation->val = MAGIC_VAL_1;
5736 	expectation->virt_fn = apic_virt_identity;
5737 	expectation->virtualize_apic_accesses =
5738 		config->virtualize_apic_accesses &&
5739 		config->activate_secondary_controls;
5740 	if (virtualize_apic_accesses_only) {
5741 		expectation->rd_exit_reason = VMX_APIC_ACCESS;
5742 		expectation->wr_exit_reason = VMX_APIC_ACCESS;
5743 	} else if (virtualize_apic_accesses_and_use_tpr_shadow) {
5744 		switch (reg) {
5745 		case APIC_TASKPRI:
5746 			expectation->rd_exit_reason = VMX_VMCALL;
5747 			expectation->wr_exit_reason = VMX_VMCALL;
5748 			expectation->virt_fn = apic_virt_nibble1;
5749 			break;
5750 		default:
5751 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5752 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5753 		}
5754 	} else if (apic_register_virtualization) {
5755 		expectation->rd_exit_reason = VMX_VMCALL;
5756 
5757 		switch (reg) {
5758 		case APIC_ID:
5759 		case APIC_EOI:
5760 		case APIC_LDR:
5761 		case APIC_DFR:
5762 		case APIC_SPIV:
5763 		case APIC_ESR:
5764 		case APIC_ICR:
5765 		case APIC_LVTT:
5766 		case APIC_LVTTHMR:
5767 		case APIC_LVTPC:
5768 		case APIC_LVT0:
5769 		case APIC_LVT1:
5770 		case APIC_LVTERR:
5771 		case APIC_TMICT:
5772 		case APIC_TDCR:
5773 			expectation->wr_exit_reason = VMX_APIC_WRITE;
5774 			break;
5775 		case APIC_LVR:
5776 		case APIC_ISR ... APIC_ISR + 0x70:
5777 		case APIC_TMR ... APIC_TMR + 0x70:
5778 		case APIC_IRR ... APIC_IRR + 0x70:
5779 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5780 			break;
5781 		case APIC_TASKPRI:
5782 			expectation->wr_exit_reason = VMX_VMCALL;
5783 			expectation->virt_fn = apic_virt_nibble1;
5784 			break;
5785 		case APIC_ICR2:
5786 			expectation->wr_exit_reason = VMX_VMCALL;
5787 			expectation->virt_fn = apic_virt_byte3;
5788 			break;
5789 		default:
5790 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5791 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5792 		}
5793 	} else if (!expectation->virtualize_apic_accesses) {
5794 		/*
5795 		 * No APIC registers are directly virtualized. This includes
5796 		 * VTPR, which can be virtualized through MOV to/from CR8 via
5797 		 * the use TPR shadow control, but not through directly
5798 		 * accessing VTPR.
5799 		 */
5800 		expectation->rd_exit_reason = VMX_VMCALL;
5801 		expectation->wr_exit_reason = VMX_VMCALL;
5802 	} else {
5803 		printf("Cannot parse APIC register virtualization config:\n"
5804 		       "\tvirtualize_apic_accesses: %d\n"
5805 		       "\tuse_tpr_shadow: %d\n"
5806 		       "\tapic_register_virtualization: %d\n"
5807 		       "\tvirtualize_x2apic_mode: %d\n"
5808 		       "\tactivate_secondary_controls: %d\n",
5809 		       config->virtualize_apic_accesses,
5810 		       config->use_tpr_shadow,
5811 		       config->apic_register_virtualization,
5812 		       config->virtualize_x2apic_mode,
5813 		       config->activate_secondary_controls);
5814 
5815 		return false;
5816 	}
5817 
5818 	return true;
5819 }
5820 
5821 struct apic_reg_test apic_reg_tests[] = {
5822 	/* Good configs, where some L2 APIC accesses are virtualized. */
5823 	{
5824 		.name = "Virtualize APIC accesses",
5825 		.apic_reg_virt_config = {
5826 			.virtualize_apic_accesses = true,
5827 			.use_tpr_shadow = false,
5828 			.apic_register_virtualization = false,
5829 			.virtualize_x2apic_mode = false,
5830 			.activate_secondary_controls = true,
5831 		},
5832 	},
5833 	{
5834 		.name = "Virtualize APIC accesses + Use TPR shadow",
5835 		.apic_reg_virt_config = {
5836 			.virtualize_apic_accesses = true,
5837 			.use_tpr_shadow = true,
5838 			.apic_register_virtualization = false,
5839 			.virtualize_x2apic_mode = false,
5840 			.activate_secondary_controls = true,
5841 		},
5842 	},
5843 	{
5844 		.name = "APIC-register virtualization",
5845 		.apic_reg_virt_config = {
5846 			.virtualize_apic_accesses = true,
5847 			.use_tpr_shadow = true,
5848 			.apic_register_virtualization = true,
5849 			.virtualize_x2apic_mode = false,
5850 			.activate_secondary_controls = true,
5851 		},
5852 	},
5853 
5854 	/*
5855 	 * Test that the secondary processor-based VM-execution controls are
5856 	 * correctly ignored when "activate secondary controls" is disabled.
5857 	 */
5858 	{
5859 		.name = "Activate secondary controls off",
5860 		.apic_reg_virt_config = {
5861 			.virtualize_apic_accesses = true,
5862 			.use_tpr_shadow = false,
5863 			.apic_register_virtualization = true,
5864 			.virtualize_x2apic_mode = true,
5865 			.activate_secondary_controls = false,
5866 		},
5867 	},
5868 	{
5869 		.name = "Activate secondary controls off + Use TPR shadow",
5870 		.apic_reg_virt_config = {
5871 			.virtualize_apic_accesses = true,
5872 			.use_tpr_shadow = true,
5873 			.apic_register_virtualization = true,
5874 			.virtualize_x2apic_mode = true,
5875 			.activate_secondary_controls = false,
5876 		},
5877 	},
5878 
5879 	/*
5880 	 * Test that the APIC access address is treated like an arbitrary memory
5881 	 * address when "virtualize APIC accesses" is disabled.
5882 	 */
5883 	{
5884 		.name = "Virtualize APIC accesses off + Use TPR shadow",
5885 		.apic_reg_virt_config = {
5886 			.virtualize_apic_accesses = false,
5887 			.use_tpr_shadow = true,
5888 			.apic_register_virtualization = true,
5889 			.virtualize_x2apic_mode = true,
5890 			.activate_secondary_controls = true,
5891 		},
5892 	},
5893 
5894 	/*
5895 	 * Test that VM entry fails due to invalid controls when
5896 	 * "APIC-register virtualization" is enabled while "use TPR shadow" is
5897 	 * disabled.
5898 	 */
5899 	{
5900 		.name = "APIC-register virtualization + Use TPR shadow off",
5901 		.apic_reg_virt_config = {
5902 			.virtualize_apic_accesses = true,
5903 			.use_tpr_shadow = false,
5904 			.apic_register_virtualization = true,
5905 			.virtualize_x2apic_mode = false,
5906 			.activate_secondary_controls = true,
5907 		},
5908 	},
5909 
5910 	/*
5911 	 * Test that VM entry fails due to invalid controls when
5912 	 * "Virtualize x2APIC mode" is enabled while "use TPR shadow" is
5913 	 * disabled.
5914 	 */
5915 	{
5916 		.name = "Virtualize x2APIC mode + Use TPR shadow off",
5917 		.apic_reg_virt_config = {
5918 			.virtualize_apic_accesses = false,
5919 			.use_tpr_shadow = false,
5920 			.apic_register_virtualization = false,
5921 			.virtualize_x2apic_mode = true,
5922 			.activate_secondary_controls = true,
5923 		},
5924 	},
5925 	{
5926 		.name = "Virtualize x2APIC mode + Use TPR shadow off v2",
5927 		.apic_reg_virt_config = {
5928 			.virtualize_apic_accesses = false,
5929 			.use_tpr_shadow = false,
5930 			.apic_register_virtualization = true,
5931 			.virtualize_x2apic_mode = true,
5932 			.activate_secondary_controls = true,
5933 		},
5934 	},
5935 
5936 	/*
5937 	 * Test that VM entry fails due to invalid controls when
5938 	 * "virtualize x2APIC mode" is enabled while "virtualize APIC accesses"
5939 	 * is enabled.
5940 	 */
5941 	{
5942 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses",
5943 		.apic_reg_virt_config = {
5944 			.virtualize_apic_accesses = true,
5945 			.use_tpr_shadow = true,
5946 			.apic_register_virtualization = false,
5947 			.virtualize_x2apic_mode = true,
5948 			.activate_secondary_controls = true,
5949 		},
5950 	},
5951 	{
5952 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses v2",
5953 		.apic_reg_virt_config = {
5954 			.virtualize_apic_accesses = true,
5955 			.use_tpr_shadow = true,
5956 			.apic_register_virtualization = true,
5957 			.virtualize_x2apic_mode = true,
5958 			.activate_secondary_controls = true,
5959 		},
5960 	},
5961 };
5962 
5963 enum Apic_op {
5964 	APIC_OP_XAPIC_RD,
5965 	APIC_OP_XAPIC_WR,
5966 	TERMINATE,
5967 };
5968 
5969 static u32 vmx_xapic_read(u32 *apic_access_address, u32 reg)
5970 {
5971 	return *(volatile u32 *)((uintptr_t)apic_access_address + reg);
5972 }
5973 
5974 static void vmx_xapic_write(u32 *apic_access_address, u32 reg, u32 val)
5975 {
5976 	*(volatile u32 *)((uintptr_t)apic_access_address + reg) = val;
5977 }
5978 
5979 struct apic_reg_virt_guest_args {
5980 	enum Apic_op op;
5981 	u32 *apic_access_address;
5982 	u32 reg;
5983 	u32 val;
5984 	bool check_rd;
5985 	u32 (*virt_fn)(u32);
5986 } apic_reg_virt_guest_args;
5987 
5988 static void apic_reg_virt_guest(void)
5989 {
5990 	volatile struct apic_reg_virt_guest_args *args =
5991 		&apic_reg_virt_guest_args;
5992 
5993 	for (;;) {
5994 		enum Apic_op op = args->op;
5995 		u32 *apic_access_address = args->apic_access_address;
5996 		u32 reg = args->reg;
5997 		u32 val = args->val;
5998 		bool check_rd = args->check_rd;
5999 		u32 (*virt_fn)(u32) = args->virt_fn;
6000 
6001 		if (op == TERMINATE)
6002 			break;
6003 
6004 		if (op == APIC_OP_XAPIC_RD) {
6005 			u32 ret = vmx_xapic_read(apic_access_address, reg);
6006 
6007 			if (check_rd) {
6008 				u32 want = virt_fn(val);
6009 				u32 got = virt_fn(ret);
6010 
6011 				report(got == want,
6012 				       "read 0x%x, expected 0x%x.", got, want);
6013 			}
6014 		} else if (op == APIC_OP_XAPIC_WR) {
6015 			vmx_xapic_write(apic_access_address, reg, val);
6016 		}
6017 
6018 		/*
6019 		 * The L1 should always execute a vmcall after it's done testing
6020 		 * an individual APIC operation. This helps to validate that the
6021 		 * L1 and L2 are in sync with each other, as expected.
6022 		 */
6023 		vmcall();
6024 	}
6025 }
6026 
6027 static void test_xapic_rd(
6028 	u32 reg, struct apic_reg_virt_expectation *expectation,
6029 	u32 *apic_access_address, u32 *virtual_apic_page)
6030 {
6031 	u32 val = expectation->val;
6032 	u32 exit_reason_want = expectation->rd_exit_reason;
6033 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
6034 
6035 	report_prefix_pushf("xapic - reading 0x%03x", reg);
6036 
6037 	/* Configure guest to do an xapic read */
6038 	args->op = APIC_OP_XAPIC_RD;
6039 	args->apic_access_address = apic_access_address;
6040 	args->reg = reg;
6041 	args->val = val;
6042 	args->check_rd = exit_reason_want == VMX_VMCALL;
6043 	args->virt_fn = expectation->virt_fn;
6044 
6045 	/* Setup virtual APIC page */
6046 	if (!expectation->virtualize_apic_accesses) {
6047 		apic_access_address[apic_reg_index(reg)] = val;
6048 		virtual_apic_page[apic_reg_index(reg)] = 0;
6049 	} else if (exit_reason_want == VMX_VMCALL) {
6050 		apic_access_address[apic_reg_index(reg)] = 0;
6051 		virtual_apic_page[apic_reg_index(reg)] = val;
6052 	}
6053 
6054 	/* Enter guest */
6055 	enter_guest();
6056 
6057 	/*
6058 	 * Validate the behavior and
6059 	 * pass a magic value back to the guest.
6060 	 */
6061 	if (exit_reason_want == VMX_APIC_ACCESS) {
6062 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
6063 
6064 		assert_exit_reason(exit_reason_want);
6065 		report(apic_page_offset == reg,
6066 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
6067 		       apic_page_offset, reg);
6068 		skip_exit_insn();
6069 
6070 		/* Reenter guest so it can consume/check rcx and exit again. */
6071 		enter_guest();
6072 	} else if (exit_reason_want != VMX_VMCALL) {
6073 		report(false, "Oops, bad exit expectation: %u.",
6074 		       exit_reason_want);
6075 	}
6076 
6077 	skip_exit_vmcall();
6078 	report_prefix_pop();
6079 }
6080 
6081 static void test_xapic_wr(
6082 	u32 reg, struct apic_reg_virt_expectation *expectation,
6083 	u32 *apic_access_address, u32 *virtual_apic_page)
6084 {
6085 	u32 val = expectation->val;
6086 	u32 exit_reason_want = expectation->wr_exit_reason;
6087 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
6088 	bool virtualized =
6089 		expectation->virtualize_apic_accesses &&
6090 		(exit_reason_want == VMX_APIC_WRITE ||
6091 		 exit_reason_want == VMX_VMCALL);
6092 	bool checked = false;
6093 
6094 	report_prefix_pushf("xapic - writing 0x%x to 0x%03x", val, reg);
6095 
6096 	/* Configure guest to do an xapic read */
6097 	args->op = APIC_OP_XAPIC_WR;
6098 	args->apic_access_address = apic_access_address;
6099 	args->reg = reg;
6100 	args->val = val;
6101 
6102 	/* Setup virtual APIC page */
6103 	if (virtualized || !expectation->virtualize_apic_accesses) {
6104 		apic_access_address[apic_reg_index(reg)] = 0;
6105 		virtual_apic_page[apic_reg_index(reg)] = 0;
6106 	}
6107 
6108 	/* Enter guest */
6109 	enter_guest();
6110 
6111 	/*
6112 	 * Validate the behavior and
6113 	 * pass a magic value back to the guest.
6114 	 */
6115 	if (exit_reason_want == VMX_APIC_ACCESS) {
6116 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
6117 
6118 		assert_exit_reason(exit_reason_want);
6119 		report(apic_page_offset == reg,
6120 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
6121 		       apic_page_offset, reg);
6122 		skip_exit_insn();
6123 
6124 		/* Reenter guest so it can consume/check rcx and exit again. */
6125 		enter_guest();
6126 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6127 		assert_exit_reason(exit_reason_want);
6128 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6129 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%x",
6130 		       apic_reg_index(reg),
6131 		       virtual_apic_page[apic_reg_index(reg)], val);
6132 		checked = true;
6133 
6134 		/* Reenter guest so it can consume/check rcx and exit again. */
6135 		enter_guest();
6136 	} else if (exit_reason_want != VMX_VMCALL) {
6137 		report(false, "Oops, bad exit expectation: %u.",
6138 		       exit_reason_want);
6139 	}
6140 
6141 	assert_exit_reason(VMX_VMCALL);
6142 	if (virtualized && !checked) {
6143 		u32 want = expectation->virt_fn(val);
6144 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6145 		got = expectation->virt_fn(got);
6146 
6147 		report(got == want, "exitless write; val is 0x%x, want 0x%x",
6148 		       got, want);
6149 	} else if (!expectation->virtualize_apic_accesses && !checked) {
6150 		u32 got = apic_access_address[apic_reg_index(reg)];
6151 
6152 		report(got == val,
6153 		       "non-virtualized write; val is 0x%x, want 0x%x", got,
6154 		       val);
6155 	} else if (!expectation->virtualize_apic_accesses && checked) {
6156 		report(false,
6157 		       "Non-virtualized write was prematurely checked!");
6158 	}
6159 
6160 	skip_exit_vmcall();
6161 	report_prefix_pop();
6162 }
6163 
6164 enum Config_type {
6165 	CONFIG_TYPE_GOOD,
6166 	CONFIG_TYPE_UNSUPPORTED,
6167 	CONFIG_TYPE_VMENTRY_FAILS_EARLY,
6168 };
6169 
6170 static enum Config_type configure_apic_reg_virt_test(
6171 	struct apic_reg_virt_config *apic_reg_virt_config)
6172 {
6173 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6174 	u32 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6175 	/* Configs where L2 entry fails early, due to invalid controls. */
6176 	bool use_tpr_shadow_incorrectly_off =
6177 		!apic_reg_virt_config->use_tpr_shadow &&
6178 		(apic_reg_virt_config->apic_register_virtualization ||
6179 		 apic_reg_virt_config->virtualize_x2apic_mode) &&
6180 		apic_reg_virt_config->activate_secondary_controls;
6181 	bool virtualize_apic_accesses_incorrectly_on =
6182 		apic_reg_virt_config->virtualize_apic_accesses &&
6183 		apic_reg_virt_config->virtualize_x2apic_mode &&
6184 		apic_reg_virt_config->activate_secondary_controls;
6185 	bool vmentry_fails_early =
6186 		use_tpr_shadow_incorrectly_off ||
6187 		virtualize_apic_accesses_incorrectly_on;
6188 
6189 	if (apic_reg_virt_config->activate_secondary_controls) {
6190 		if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
6191 			printf("VM-execution control \"activate secondary controls\" NOT supported.\n");
6192 			return CONFIG_TYPE_UNSUPPORTED;
6193 		}
6194 		cpu_exec_ctrl0 |= CPU_SECONDARY;
6195 	} else {
6196 		cpu_exec_ctrl0 &= ~CPU_SECONDARY;
6197 	}
6198 
6199 	if (apic_reg_virt_config->virtualize_apic_accesses) {
6200 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES)) {
6201 			printf("VM-execution control \"virtualize APIC accesses\" NOT supported.\n");
6202 			return CONFIG_TYPE_UNSUPPORTED;
6203 		}
6204 		cpu_exec_ctrl1 |= CPU_VIRT_APIC_ACCESSES;
6205 	} else {
6206 		cpu_exec_ctrl1 &= ~CPU_VIRT_APIC_ACCESSES;
6207 	}
6208 
6209 	if (apic_reg_virt_config->use_tpr_shadow) {
6210 		if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6211 			printf("VM-execution control \"use TPR shadow\" NOT supported.\n");
6212 			return CONFIG_TYPE_UNSUPPORTED;
6213 		}
6214 		cpu_exec_ctrl0 |= CPU_TPR_SHADOW;
6215 	} else {
6216 		cpu_exec_ctrl0 &= ~CPU_TPR_SHADOW;
6217 	}
6218 
6219 	if (apic_reg_virt_config->apic_register_virtualization) {
6220 		if (!(ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT)) {
6221 			printf("VM-execution control \"APIC-register virtualization\" NOT supported.\n");
6222 			return CONFIG_TYPE_UNSUPPORTED;
6223 		}
6224 		cpu_exec_ctrl1 |= CPU_APIC_REG_VIRT;
6225 	} else {
6226 		cpu_exec_ctrl1 &= ~CPU_APIC_REG_VIRT;
6227 	}
6228 
6229 	if (apic_reg_virt_config->virtualize_x2apic_mode) {
6230 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_X2APIC)) {
6231 			printf("VM-execution control \"virtualize x2APIC mode\" NOT supported.\n");
6232 			return CONFIG_TYPE_UNSUPPORTED;
6233 		}
6234 		cpu_exec_ctrl1 |= CPU_VIRT_X2APIC;
6235 	} else {
6236 		cpu_exec_ctrl1 &= ~CPU_VIRT_X2APIC;
6237 	}
6238 
6239 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6240 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6241 
6242 	if (vmentry_fails_early)
6243 		return CONFIG_TYPE_VMENTRY_FAILS_EARLY;
6244 
6245 	return CONFIG_TYPE_GOOD;
6246 }
6247 
6248 static bool cpu_has_apicv(void)
6249 {
6250 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
6251 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
6252 		(ctrl_pin_rev.clr & PIN_POST_INTR));
6253 }
6254 
6255 /* Validates APIC register access across valid virtualization configurations. */
6256 static void apic_reg_virt_test(void)
6257 {
6258 	u32 *apic_access_address;
6259 	u32 *virtual_apic_page;
6260 	u64 control;
6261 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6262 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6263 	int i;
6264 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
6265 
6266 	if (!cpu_has_apicv()) {
6267 		report_skip(__func__);
6268 		return;
6269 	}
6270 
6271 	control = cpu_exec_ctrl1;
6272 	control &= ~CPU_VINTD;
6273 	vmcs_write(CPU_EXEC_CTRL1, control);
6274 
6275 	test_set_guest(apic_reg_virt_guest);
6276 
6277 	/*
6278 	 * From the SDM: The 1-setting of the "virtualize APIC accesses"
6279 	 * VM-execution is guaranteed to apply only if translations to the
6280 	 * APIC-access address use a 4-KByte page.
6281 	 */
6282 	apic_access_address = alloc_page();
6283 	force_4k_page(apic_access_address);
6284 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_address));
6285 
6286 	virtual_apic_page = alloc_page();
6287 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6288 
6289 	for (i = 0; i < ARRAY_SIZE(apic_reg_tests); i++) {
6290 		struct apic_reg_test *apic_reg_test = &apic_reg_tests[i];
6291 		struct apic_reg_virt_config *apic_reg_virt_config =
6292 				&apic_reg_test->apic_reg_virt_config;
6293 		enum Config_type config_type;
6294 		u32 reg;
6295 
6296 		printf("--- %s test ---\n", apic_reg_test->name);
6297 		config_type =
6298 			configure_apic_reg_virt_test(apic_reg_virt_config);
6299 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6300 			printf("Skip because of missing features.\n");
6301 			continue;
6302 		}
6303 
6304 		if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6305 			enter_guest_with_bad_controls();
6306 			continue;
6307 		}
6308 
6309 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6310 			struct apic_reg_virt_expectation expectation = {};
6311 			bool ok;
6312 
6313 			ok = apic_reg_virt_exit_expectation(
6314 				reg, apic_reg_virt_config, &expectation);
6315 			if (!ok) {
6316 				report(false, "Malformed test.");
6317 				break;
6318 			}
6319 
6320 			test_xapic_rd(reg, &expectation, apic_access_address,
6321 				      virtual_apic_page);
6322 			test_xapic_wr(reg, &expectation, apic_access_address,
6323 				      virtual_apic_page);
6324 		}
6325 	}
6326 
6327 	/* Terminate the guest */
6328 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6329 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6330 	args->op = TERMINATE;
6331 	enter_guest();
6332 	assert_exit_reason(VMX_VMCALL);
6333 }
6334 
6335 struct virt_x2apic_mode_config {
6336 	struct apic_reg_virt_config apic_reg_virt_config;
6337 	bool virtual_interrupt_delivery;
6338 	bool use_msr_bitmaps;
6339 	bool disable_x2apic_msr_intercepts;
6340 	bool disable_x2apic;
6341 };
6342 
6343 struct virt_x2apic_mode_test_case {
6344 	const char *name;
6345 	struct virt_x2apic_mode_config virt_x2apic_mode_config;
6346 };
6347 
6348 enum Virt_x2apic_mode_behavior_type {
6349 	X2APIC_ACCESS_VIRTUALIZED,
6350 	X2APIC_ACCESS_PASSED_THROUGH,
6351 	X2APIC_ACCESS_TRIGGERS_GP,
6352 };
6353 
6354 struct virt_x2apic_mode_expectation {
6355 	enum Reason rd_exit_reason;
6356 	enum Reason wr_exit_reason;
6357 
6358 	/*
6359 	 * RDMSR and WRMSR handle 64-bit values. However, except for ICR, all of
6360 	 * the x2APIC registers are 32 bits. Notice:
6361 	 *   1. vmx_x2apic_read() clears the upper 32 bits for 32-bit registers.
6362 	 *   2. vmx_x2apic_write() expects the val arg to be well-formed.
6363 	 */
6364 	u64 rd_val;
6365 	u64 wr_val;
6366 
6367 	/*
6368 	 * Compares input to virtualized output;
6369 	 * 1st arg is pointer to return expected virtualization output.
6370 	 */
6371 	u64 (*virt_fn)(u64);
6372 
6373 	enum Virt_x2apic_mode_behavior_type rd_behavior;
6374 	enum Virt_x2apic_mode_behavior_type wr_behavior;
6375 	bool wr_only;
6376 };
6377 
6378 static u64 virt_x2apic_mode_identity(u64 val)
6379 {
6380 	return val;
6381 }
6382 
6383 static u64 virt_x2apic_mode_nibble1(u64 val)
6384 {
6385 	return val & 0xf0;
6386 }
6387 
6388 static void virt_x2apic_mode_rd_expectation(
6389 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6390 	bool apic_register_virtualization, bool virtual_interrupt_delivery,
6391 	struct virt_x2apic_mode_expectation *expectation)
6392 {
6393 	bool readable =
6394 		!x2apic_reg_reserved(reg) &&
6395 		reg != APIC_EOI;
6396 
6397 	expectation->rd_exit_reason = VMX_VMCALL;
6398 	expectation->virt_fn = virt_x2apic_mode_identity;
6399 	if (virt_x2apic_mode_on && apic_register_virtualization) {
6400 		expectation->rd_val = MAGIC_VAL_1;
6401 		if (reg == APIC_PROCPRI && virtual_interrupt_delivery)
6402 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6403 		else if (reg == APIC_TASKPRI)
6404 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6405 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6406 	} else if (virt_x2apic_mode_on && !apic_register_virtualization &&
6407 		   reg == APIC_TASKPRI) {
6408 		expectation->rd_val = MAGIC_VAL_1;
6409 		expectation->virt_fn = virt_x2apic_mode_nibble1;
6410 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6411 	} else if (!disable_x2apic && readable) {
6412 		expectation->rd_val = apic_read(reg);
6413 		expectation->rd_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6414 	} else {
6415 		expectation->rd_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6416 	}
6417 }
6418 
6419 /*
6420  * get_x2apic_wr_val() creates an innocuous write value for an x2APIC register.
6421  *
6422  * For writable registers, get_x2apic_wr_val() deposits the write value into the
6423  * val pointer arg and returns true. For non-writable registers, val is not
6424  * modified and get_x2apic_wr_val() returns false.
6425  */
6426 static bool get_x2apic_wr_val(u32 reg, u64 *val)
6427 {
6428 	switch (reg) {
6429 	case APIC_TASKPRI:
6430 		/* Bits 31:8 are reserved. */
6431 		*val &= 0xff;
6432 		break;
6433 	case APIC_EOI:
6434 	case APIC_ESR:
6435 	case APIC_TMICT:
6436 		/*
6437 		 * EOI, ESR: WRMSR of a non-zero value causes #GP(0).
6438 		 * TMICT: A write of 0 to the initial-count register effectively
6439 		 *        stops the local APIC timer, in both one-shot and
6440 		 *        periodic mode.
6441 		 */
6442 		*val = 0;
6443 		break;
6444 	case APIC_SPIV:
6445 	case APIC_LVTT:
6446 	case APIC_LVTTHMR:
6447 	case APIC_LVTPC:
6448 	case APIC_LVT0:
6449 	case APIC_LVT1:
6450 	case APIC_LVTERR:
6451 	case APIC_TDCR:
6452 		/*
6453 		 * To avoid writing a 1 to a reserved bit or causing some other
6454 		 * unintended side effect, read the current value and use it as
6455 		 * the write value.
6456 		 */
6457 		*val = apic_read(reg);
6458 		break;
6459 	case APIC_CMCI:
6460 		if (!apic_lvt_entry_supported(6))
6461 			return false;
6462 		*val = apic_read(reg);
6463 		break;
6464 	case APIC_ICR:
6465 		*val = 0x40000 | 0xf1;
6466 		break;
6467 	case APIC_SELF_IPI:
6468 		/*
6469 		 * With special processing (i.e., virtualize x2APIC mode +
6470 		 * virtual interrupt delivery), writing zero causes an
6471 		 * APIC-write VM exit. We plan to add a test for enabling
6472 		 * "virtual-interrupt delivery" in VMCS12, and that's where we
6473 		 * will test a self IPI with special processing.
6474 		 */
6475 		*val = 0x0;
6476 		break;
6477 	default:
6478 		return false;
6479 	}
6480 
6481 	return true;
6482 }
6483 
6484 static bool special_processing_applies(u32 reg, u64 *val,
6485 				       bool virt_int_delivery)
6486 {
6487 	bool special_processing =
6488 		(reg == APIC_TASKPRI) ||
6489 		(virt_int_delivery &&
6490 		 (reg == APIC_EOI || reg == APIC_SELF_IPI));
6491 
6492 	if (special_processing) {
6493 		TEST_ASSERT(get_x2apic_wr_val(reg, val));
6494 		return true;
6495 	}
6496 
6497 	return false;
6498 }
6499 
6500 static void virt_x2apic_mode_wr_expectation(
6501 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6502 	bool virt_int_delivery,
6503 	struct virt_x2apic_mode_expectation *expectation)
6504 {
6505 	expectation->wr_exit_reason = VMX_VMCALL;
6506 	expectation->wr_val = MAGIC_VAL_1;
6507 	expectation->wr_only = false;
6508 
6509 	if (virt_x2apic_mode_on &&
6510 	    special_processing_applies(reg, &expectation->wr_val,
6511 				       virt_int_delivery)) {
6512 		expectation->wr_behavior = X2APIC_ACCESS_VIRTUALIZED;
6513 		if (reg == APIC_SELF_IPI)
6514 			expectation->wr_exit_reason = VMX_APIC_WRITE;
6515 	} else if (!disable_x2apic &&
6516 		   get_x2apic_wr_val(reg, &expectation->wr_val)) {
6517 		expectation->wr_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6518 		if (reg == APIC_EOI || reg == APIC_SELF_IPI)
6519 			expectation->wr_only = true;
6520 		if (reg == APIC_ICR)
6521 			expectation->wr_exit_reason = VMX_EXTINT;
6522 	} else {
6523 		expectation->wr_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6524 		/*
6525 		 * Writing 1 to a reserved bit triggers a #GP.
6526 		 * Thus, set the write value to 0, which seems
6527 		 * the most likely to detect a missed #GP.
6528 		 */
6529 		expectation->wr_val = 0;
6530 	}
6531 }
6532 
6533 static void virt_x2apic_mode_exit_expectation(
6534 	u32 reg, struct virt_x2apic_mode_config *config,
6535 	struct virt_x2apic_mode_expectation *expectation)
6536 {
6537 	struct apic_reg_virt_config *base_config =
6538 		&config->apic_reg_virt_config;
6539 	bool virt_x2apic_mode_on =
6540 		base_config->virtualize_x2apic_mode &&
6541 		config->use_msr_bitmaps &&
6542 		config->disable_x2apic_msr_intercepts &&
6543 		base_config->activate_secondary_controls;
6544 
6545 	virt_x2apic_mode_wr_expectation(
6546 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6547 		config->virtual_interrupt_delivery, expectation);
6548 	virt_x2apic_mode_rd_expectation(
6549 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6550 		base_config->apic_register_virtualization,
6551 		config->virtual_interrupt_delivery, expectation);
6552 }
6553 
6554 struct virt_x2apic_mode_test_case virt_x2apic_mode_tests[] = {
6555 	/*
6556 	 * Baseline "virtualize x2APIC mode" configuration:
6557 	 *   - virtualize x2APIC mode
6558 	 *   - virtual-interrupt delivery
6559 	 *   - APIC-register virtualization
6560 	 *   - x2APIC MSR intercepts disabled
6561 	 *
6562 	 * Reads come from virtual APIC page, special processing applies to
6563 	 * VTPR, EOI, and SELF IPI, and all other writes pass through to L1
6564 	 * APIC.
6565 	 */
6566 	{
6567 		.name = "Baseline",
6568 		.virt_x2apic_mode_config = {
6569 			.virtual_interrupt_delivery = true,
6570 			.use_msr_bitmaps = true,
6571 			.disable_x2apic_msr_intercepts = true,
6572 			.disable_x2apic = false,
6573 			.apic_reg_virt_config = {
6574 				.apic_register_virtualization = true,
6575 				.use_tpr_shadow = true,
6576 				.virtualize_apic_accesses = false,
6577 				.virtualize_x2apic_mode = true,
6578 				.activate_secondary_controls = true,
6579 			},
6580 		},
6581 	},
6582 	{
6583 		.name = "Baseline w/ x2apic disabled",
6584 		.virt_x2apic_mode_config = {
6585 			.virtual_interrupt_delivery = true,
6586 			.use_msr_bitmaps = true,
6587 			.disable_x2apic_msr_intercepts = true,
6588 			.disable_x2apic = true,
6589 			.apic_reg_virt_config = {
6590 				.apic_register_virtualization = true,
6591 				.use_tpr_shadow = true,
6592 				.virtualize_apic_accesses = false,
6593 				.virtualize_x2apic_mode = true,
6594 				.activate_secondary_controls = true,
6595 			},
6596 		},
6597 	},
6598 
6599 	/*
6600 	 * Baseline, minus virtual-interrupt delivery. Reads come from virtual
6601 	 * APIC page, special processing applies to VTPR, and all other writes
6602 	 * pass through to L1 APIC.
6603 	 */
6604 	{
6605 		.name = "Baseline - virtual interrupt delivery",
6606 		.virt_x2apic_mode_config = {
6607 			.virtual_interrupt_delivery = false,
6608 			.use_msr_bitmaps = true,
6609 			.disable_x2apic_msr_intercepts = true,
6610 			.disable_x2apic = false,
6611 			.apic_reg_virt_config = {
6612 				.apic_register_virtualization = true,
6613 				.use_tpr_shadow = true,
6614 				.virtualize_apic_accesses = false,
6615 				.virtualize_x2apic_mode = true,
6616 				.activate_secondary_controls = true,
6617 			},
6618 		},
6619 	},
6620 
6621 	/*
6622 	 * Baseline, minus APIC-register virtualization. x2APIC reads pass
6623 	 * through to L1's APIC, unless reading VTPR
6624 	 */
6625 	{
6626 		.name = "Virtualize x2APIC mode, no APIC reg virt",
6627 		.virt_x2apic_mode_config = {
6628 			.virtual_interrupt_delivery = true,
6629 			.use_msr_bitmaps = true,
6630 			.disable_x2apic_msr_intercepts = true,
6631 			.disable_x2apic = false,
6632 			.apic_reg_virt_config = {
6633 				.apic_register_virtualization = false,
6634 				.use_tpr_shadow = true,
6635 				.virtualize_apic_accesses = false,
6636 				.virtualize_x2apic_mode = true,
6637 				.activate_secondary_controls = true,
6638 			},
6639 		},
6640 	},
6641 	{
6642 		.name = "Virtualize x2APIC mode, no APIC reg virt, x2APIC off",
6643 		.virt_x2apic_mode_config = {
6644 			.virtual_interrupt_delivery = true,
6645 			.use_msr_bitmaps = true,
6646 			.disable_x2apic_msr_intercepts = true,
6647 			.disable_x2apic = true,
6648 			.apic_reg_virt_config = {
6649 				.apic_register_virtualization = false,
6650 				.use_tpr_shadow = true,
6651 				.virtualize_apic_accesses = false,
6652 				.virtualize_x2apic_mode = true,
6653 				.activate_secondary_controls = true,
6654 			},
6655 		},
6656 	},
6657 
6658 	/*
6659 	 * Enable "virtualize x2APIC mode" and "APIC-register virtualization",
6660 	 * and disable intercepts for the x2APIC MSRs, but fail to enable
6661 	 * "activate secondary controls" (i.e. L2 gets access to L1's x2APIC
6662 	 * MSRs).
6663 	 */
6664 	{
6665 		.name = "Fail to enable activate secondary controls",
6666 		.virt_x2apic_mode_config = {
6667 			.virtual_interrupt_delivery = true,
6668 			.use_msr_bitmaps = true,
6669 			.disable_x2apic_msr_intercepts = true,
6670 			.disable_x2apic = false,
6671 			.apic_reg_virt_config = {
6672 				.apic_register_virtualization = true,
6673 				.use_tpr_shadow = true,
6674 				.virtualize_apic_accesses = false,
6675 				.virtualize_x2apic_mode = true,
6676 				.activate_secondary_controls = false,
6677 			},
6678 		},
6679 	},
6680 
6681 	/*
6682 	 * Enable "APIC-register virtualization" and enable "activate secondary
6683 	 * controls" and disable intercepts for the x2APIC MSRs, but do not
6684 	 * enable the "virtualize x2APIC mode" VM-execution control (i.e. L2
6685 	 * gets access to L1's x2APIC MSRs).
6686 	 */
6687 	{
6688 		.name = "Fail to enable virtualize x2APIC mode",
6689 		.virt_x2apic_mode_config = {
6690 			.virtual_interrupt_delivery = true,
6691 			.use_msr_bitmaps = true,
6692 			.disable_x2apic_msr_intercepts = true,
6693 			.disable_x2apic = false,
6694 			.apic_reg_virt_config = {
6695 				.apic_register_virtualization = true,
6696 				.use_tpr_shadow = true,
6697 				.virtualize_apic_accesses = false,
6698 				.virtualize_x2apic_mode = false,
6699 				.activate_secondary_controls = true,
6700 			},
6701 		},
6702 	},
6703 
6704 	/*
6705 	 * Disable "Virtualize x2APIC mode", disable x2APIC MSR intercepts, and
6706 	 * enable "APIC-register virtualization" --> L2 gets L1's x2APIC MSRs.
6707 	 */
6708 	{
6709 		.name = "Baseline",
6710 		.virt_x2apic_mode_config = {
6711 			.virtual_interrupt_delivery = true,
6712 			.use_msr_bitmaps = true,
6713 			.disable_x2apic_msr_intercepts = true,
6714 			.disable_x2apic = false,
6715 			.apic_reg_virt_config = {
6716 				.apic_register_virtualization = true,
6717 				.use_tpr_shadow = true,
6718 				.virtualize_apic_accesses = false,
6719 				.virtualize_x2apic_mode = false,
6720 				.activate_secondary_controls = true,
6721 			},
6722 		},
6723 	},
6724 };
6725 
6726 enum X2apic_op {
6727 	X2APIC_OP_RD,
6728 	X2APIC_OP_WR,
6729 	X2APIC_TERMINATE,
6730 };
6731 
6732 static u64 vmx_x2apic_read(u32 reg)
6733 {
6734 	u32 msr_addr = x2apic_msr(reg);
6735 	u64 val;
6736 
6737 	val = rdmsr(msr_addr);
6738 
6739 	return val;
6740 }
6741 
6742 static void vmx_x2apic_write(u32 reg, u64 val)
6743 {
6744 	u32 msr_addr = x2apic_msr(reg);
6745 
6746 	wrmsr(msr_addr, val);
6747 }
6748 
6749 struct virt_x2apic_mode_guest_args {
6750 	enum X2apic_op op;
6751 	u32 reg;
6752 	u64 val;
6753 	bool should_gp;
6754 	u64 (*virt_fn)(u64);
6755 } virt_x2apic_mode_guest_args;
6756 
6757 static volatile bool handle_x2apic_gp_ran;
6758 static volatile u32 handle_x2apic_gp_insn_len;
6759 static void handle_x2apic_gp(struct ex_regs *regs)
6760 {
6761 	handle_x2apic_gp_ran = true;
6762 	regs->rip += handle_x2apic_gp_insn_len;
6763 }
6764 
6765 static handler setup_x2apic_gp_handler(void)
6766 {
6767 	handler old_handler;
6768 
6769 	old_handler = handle_exception(GP_VECTOR, handle_x2apic_gp);
6770 	/* RDMSR and WRMSR are both 2 bytes, assuming no prefixes. */
6771 	handle_x2apic_gp_insn_len = 2;
6772 
6773 	return old_handler;
6774 }
6775 
6776 static void teardown_x2apic_gp_handler(handler old_handler)
6777 {
6778 	handle_exception(GP_VECTOR, old_handler);
6779 
6780 	/*
6781 	 * Defensively reset instruction length, so that if the handler is
6782 	 * incorrectly used, it will loop infinitely, rather than run off into
6783 	 * la la land.
6784 	 */
6785 	handle_x2apic_gp_insn_len = 0;
6786 	handle_x2apic_gp_ran = false;
6787 }
6788 
6789 static void virt_x2apic_mode_guest(void)
6790 {
6791 	volatile struct virt_x2apic_mode_guest_args *args =
6792 		&virt_x2apic_mode_guest_args;
6793 
6794 	for (;;) {
6795 		enum X2apic_op op = args->op;
6796 		u32 reg = args->reg;
6797 		u64 val = args->val;
6798 		bool should_gp = args->should_gp;
6799 		u64 (*virt_fn)(u64) = args->virt_fn;
6800 		handler old_handler;
6801 
6802 		if (op == X2APIC_TERMINATE)
6803 			break;
6804 
6805 		if (should_gp) {
6806 			TEST_ASSERT(!handle_x2apic_gp_ran);
6807 			old_handler = setup_x2apic_gp_handler();
6808 		}
6809 
6810 		if (op == X2APIC_OP_RD) {
6811 			u64 ret = vmx_x2apic_read(reg);
6812 
6813 			if (!should_gp) {
6814 				u64 want = virt_fn(val);
6815 				u64 got = virt_fn(ret);
6816 
6817 				report(got == want,
6818 				       "APIC read; got 0x%lx, want 0x%lx.",
6819 				       got, want);
6820 			}
6821 		} else if (op == X2APIC_OP_WR) {
6822 			vmx_x2apic_write(reg, val);
6823 		}
6824 
6825 		if (should_gp) {
6826 			report(handle_x2apic_gp_ran,
6827 			       "x2APIC op triggered GP.");
6828 			teardown_x2apic_gp_handler(old_handler);
6829 		}
6830 
6831 		/*
6832 		 * The L1 should always execute a vmcall after it's done testing
6833 		 * an individual APIC operation. This helps to validate that the
6834 		 * L1 and L2 are in sync with each other, as expected.
6835 		 */
6836 		vmcall();
6837 	}
6838 }
6839 
6840 static void test_x2apic_rd(
6841 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6842 	u32 *virtual_apic_page)
6843 {
6844 	u64 val = expectation->rd_val;
6845 	u32 exit_reason_want = expectation->rd_exit_reason;
6846 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6847 
6848 	report_prefix_pushf("x2apic - reading 0x%03x", reg);
6849 
6850 	/* Configure guest to do an x2apic read */
6851 	args->op = X2APIC_OP_RD;
6852 	args->reg = reg;
6853 	args->val = val;
6854 	args->should_gp = expectation->rd_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6855 	args->virt_fn = expectation->virt_fn;
6856 
6857 	/* Setup virtual APIC page */
6858 	if (expectation->rd_behavior == X2APIC_ACCESS_VIRTUALIZED)
6859 		virtual_apic_page[apic_reg_index(reg)] = (u32)val;
6860 
6861 	/* Enter guest */
6862 	enter_guest();
6863 
6864 	if (exit_reason_want != VMX_VMCALL) {
6865 		report(false, "Oops, bad exit expectation: %u.",
6866 		       exit_reason_want);
6867 	}
6868 
6869 	skip_exit_vmcall();
6870 	report_prefix_pop();
6871 }
6872 
6873 static volatile bool handle_x2apic_ipi_ran;
6874 static void handle_x2apic_ipi(isr_regs_t *regs)
6875 {
6876 	handle_x2apic_ipi_ran = true;
6877 	eoi();
6878 }
6879 
6880 static void test_x2apic_wr(
6881 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6882 	u32 *virtual_apic_page)
6883 {
6884 	u64 val = expectation->wr_val;
6885 	u32 exit_reason_want = expectation->wr_exit_reason;
6886 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6887 	int ipi_vector = 0xf1;
6888 	u32 restore_val = 0;
6889 
6890 	report_prefix_pushf("x2apic - writing 0x%lx to 0x%03x", val, reg);
6891 
6892 	/* Configure guest to do an x2apic read */
6893 	args->op = X2APIC_OP_WR;
6894 	args->reg = reg;
6895 	args->val = val;
6896 	args->should_gp = expectation->wr_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6897 
6898 	/* Setup virtual APIC page */
6899 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED)
6900 		virtual_apic_page[apic_reg_index(reg)] = 0;
6901 	if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH && !expectation->wr_only)
6902 		restore_val = apic_read(reg);
6903 
6904 	/* Setup IPI handler */
6905 	handle_x2apic_ipi_ran = false;
6906 	handle_irq(ipi_vector, handle_x2apic_ipi);
6907 
6908 	/* Enter guest */
6909 	enter_guest();
6910 
6911 	/*
6912 	 * Validate the behavior and
6913 	 * pass a magic value back to the guest.
6914 	 */
6915 	if (exit_reason_want == VMX_EXTINT) {
6916 		assert_exit_reason(exit_reason_want);
6917 
6918 		/* Clear the external interrupt. */
6919 		irq_enable();
6920 		asm volatile ("nop");
6921 		irq_disable();
6922 		report(handle_x2apic_ipi_ran,
6923 		       "Got pending interrupt after IRQ enabled.");
6924 
6925 		enter_guest();
6926 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6927 		assert_exit_reason(exit_reason_want);
6928 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6929 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%lx",
6930 		       apic_reg_index(reg),
6931 		       virtual_apic_page[apic_reg_index(reg)], val);
6932 
6933 		/* Reenter guest so it can consume/check rcx and exit again. */
6934 		enter_guest();
6935 	} else if (exit_reason_want != VMX_VMCALL) {
6936 		report(false, "Oops, bad exit expectation: %u.",
6937 		       exit_reason_want);
6938 	}
6939 
6940 	assert_exit_reason(VMX_VMCALL);
6941 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) {
6942 		u64 want = val;
6943 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6944 
6945 		report(got == want, "x2APIC write; got 0x%x, want 0x%lx", got,
6946 		       want);
6947 	} else if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH) {
6948 		if (!expectation->wr_only) {
6949 			u32 got = apic_read(reg);
6950 			bool ok;
6951 
6952 			/*
6953 			 * When L1's TPR is passed through to L2, the lower
6954 			 * nibble can be lost. For example, if L2 executes
6955 			 * WRMSR(0x808, 0x78), then, L1 might read 0x70.
6956 			 *
6957 			 * Here's how the lower nibble can get lost:
6958 			 *   1. L2 executes WRMSR(0x808, 0x78).
6959 			 *   2. L2 exits to L0 with a WRMSR exit.
6960 			 *   3. L0 emulates WRMSR, by writing L1's TPR.
6961 			 *   4. L0 re-enters L2.
6962 			 *   5. L2 exits to L0 (reason doesn't matter).
6963 			 *   6. L0 reflects L2's exit to L1.
6964 			 *   7. Before entering L1, L0 exits to user-space
6965 			 *      (e.g., to satisfy TPR access reporting).
6966 			 *   8. User-space executes KVM_SET_REGS ioctl, which
6967 			 *      clears the lower nibble of L1's TPR.
6968 			 */
6969 			if (reg == APIC_TASKPRI) {
6970 				got = apic_virt_nibble1(got);
6971 				val = apic_virt_nibble1(val);
6972 			}
6973 
6974 			ok = got == val;
6975 			report(ok,
6976 			       "non-virtualized write; val is 0x%x, want 0x%lx",
6977 			       got, val);
6978 			apic_write(reg, restore_val);
6979 		} else {
6980 			report(true, "non-virtualized and write-only OK");
6981 		}
6982 	}
6983 	skip_exit_insn();
6984 
6985 	report_prefix_pop();
6986 }
6987 
6988 static enum Config_type configure_virt_x2apic_mode_test(
6989 	struct virt_x2apic_mode_config *virt_x2apic_mode_config,
6990 	u8 *msr_bitmap_page)
6991 {
6992 	int msr;
6993 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6994 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6995 
6996 	/* x2apic-specific VMCS config */
6997 	if (virt_x2apic_mode_config->use_msr_bitmaps) {
6998 		/* virt_x2apic_mode_test() checks for MSR bitmaps support */
6999 		cpu_exec_ctrl0 |= CPU_MSR_BITMAP;
7000 	} else {
7001 		cpu_exec_ctrl0 &= ~CPU_MSR_BITMAP;
7002 	}
7003 
7004 	if (virt_x2apic_mode_config->virtual_interrupt_delivery) {
7005 		if (!(ctrl_cpu_rev[1].clr & CPU_VINTD)) {
7006 			report_skip("VM-execution control \"virtual-interrupt delivery\" NOT supported.\n");
7007 			return CONFIG_TYPE_UNSUPPORTED;
7008 		}
7009 		cpu_exec_ctrl1 |= CPU_VINTD;
7010 	} else {
7011 		cpu_exec_ctrl1 &= ~CPU_VINTD;
7012 	}
7013 
7014 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
7015 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
7016 
7017 	/* x2APIC MSR intercepts are usually off for "Virtualize x2APIC mode" */
7018 	for (msr = 0x800; msr <= 0x8ff; msr++) {
7019 		if (virt_x2apic_mode_config->disable_x2apic_msr_intercepts) {
7020 			clear_bit(msr, msr_bitmap_page + 0x000);
7021 			clear_bit(msr, msr_bitmap_page + 0x800);
7022 		} else {
7023 			set_bit(msr, msr_bitmap_page + 0x000);
7024 			set_bit(msr, msr_bitmap_page + 0x800);
7025 		}
7026 	}
7027 
7028 	/* x2APIC mode can impact virtualization */
7029 	reset_apic();
7030 	if (!virt_x2apic_mode_config->disable_x2apic)
7031 		enable_x2apic();
7032 
7033 	return configure_apic_reg_virt_test(
7034 		&virt_x2apic_mode_config->apic_reg_virt_config);
7035 }
7036 
7037 static void virt_x2apic_mode_test(void)
7038 {
7039 	u32 *virtual_apic_page;
7040 	u8 *msr_bitmap_page;
7041 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
7042 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
7043 	int i;
7044 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
7045 
7046 	if (!cpu_has_apicv()) {
7047 		report_skip(__func__);
7048 		return;
7049 	}
7050 
7051 	/*
7052 	 * This is to exercise an issue in KVM's logic to merge L0's and L1's
7053 	 * MSR bitmaps. Previously, an L1 could get at L0's x2APIC MSRs by
7054 	 * writing the IA32_SPEC_CTRL MSR or the IA32_PRED_CMD MSRs. KVM would
7055 	 * then proceed to manipulate the MSR bitmaps, as if VMCS12 had the
7056 	 * "Virtualize x2APIC mod" control set, even when it didn't.
7057 	 */
7058 	if (has_spec_ctrl())
7059 		wrmsr(MSR_IA32_SPEC_CTRL, 1);
7060 
7061 	/*
7062 	 * Check that VMCS12 supports:
7063 	 *   - "Virtual-APIC address", indicated by "use TPR shadow"
7064 	 *   - "MSR-bitmap address", indicated by "use MSR bitmaps"
7065 	 */
7066 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
7067 		report_skip("VM-execution control \"use TPR shadow\" NOT supported.\n");
7068 		return;
7069 	} else if (!(ctrl_cpu_rev[0].clr & CPU_MSR_BITMAP)) {
7070 		report_skip("VM-execution control \"use MSR bitmaps\" NOT supported.\n");
7071 		return;
7072 	}
7073 
7074 	test_set_guest(virt_x2apic_mode_guest);
7075 
7076 	virtual_apic_page = alloc_page();
7077 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
7078 
7079 	msr_bitmap_page = alloc_page();
7080 	memset(msr_bitmap_page, 0xff, PAGE_SIZE);
7081 	vmcs_write(MSR_BITMAP, virt_to_phys(msr_bitmap_page));
7082 
7083 	for (i = 0; i < ARRAY_SIZE(virt_x2apic_mode_tests); i++) {
7084 		struct virt_x2apic_mode_test_case *virt_x2apic_mode_test_case =
7085 			&virt_x2apic_mode_tests[i];
7086 		struct virt_x2apic_mode_config *virt_x2apic_mode_config =
7087 			&virt_x2apic_mode_test_case->virt_x2apic_mode_config;
7088 		enum Config_type config_type;
7089 		u32 reg;
7090 
7091 		printf("--- %s test ---\n", virt_x2apic_mode_test_case->name);
7092 		config_type =
7093 			configure_virt_x2apic_mode_test(virt_x2apic_mode_config,
7094 							msr_bitmap_page);
7095 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
7096 			report_skip("Skip because of missing features.\n");
7097 			continue;
7098 		} else if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
7099 			enter_guest_with_bad_controls();
7100 			continue;
7101 		}
7102 
7103 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
7104 			struct virt_x2apic_mode_expectation expectation;
7105 
7106 			virt_x2apic_mode_exit_expectation(
7107 				reg, virt_x2apic_mode_config, &expectation);
7108 
7109 			test_x2apic_rd(reg, &expectation, virtual_apic_page);
7110 			test_x2apic_wr(reg, &expectation, virtual_apic_page);
7111 		}
7112 	}
7113 
7114 
7115 	/* Terminate the guest */
7116 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
7117 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
7118 	args->op = X2APIC_TERMINATE;
7119 	enter_guest();
7120 	assert_exit_reason(VMX_VMCALL);
7121 }
7122 
7123 static void test_ctl_reg(const char *cr_name, u64 cr, u64 fixed0, u64 fixed1)
7124 {
7125 	u64 val;
7126 	u64 cr_saved = vmcs_read(cr);
7127 	int i;
7128 
7129 	val = fixed0 & fixed1;
7130 	if (cr == HOST_CR4)
7131 		vmcs_write(cr, val | X86_CR4_PAE);
7132 	else
7133 		vmcs_write(cr, val);
7134 	report_prefix_pushf("%s %lx", cr_name, val);
7135 	if (val == fixed0)
7136 		test_vmx_vmlaunch(0);
7137 	else
7138 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7139 	report_prefix_pop();
7140 
7141 	for (i = 0; i < 64; i++) {
7142 
7143 		/* Set a bit when the corresponding bit in fixed1 is 0 */
7144 		if ((fixed1 & (1ull << i)) == 0) {
7145 			if (cr == HOST_CR4 && ((1ull << i) & X86_CR4_SMEP ||
7146 					       (1ull << i) & X86_CR4_SMAP))
7147 				continue;
7148 
7149 			vmcs_write(cr, cr_saved | (1ull << i));
7150 			report_prefix_pushf("%s %llx", cr_name,
7151 						cr_saved | (1ull << i));
7152 			test_vmx_vmlaunch(
7153 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7154 			report_prefix_pop();
7155 		}
7156 
7157 		/* Unset a bit when the corresponding bit in fixed0 is 1 */
7158 		if (fixed0 & (1ull << i)) {
7159 			vmcs_write(cr, cr_saved & ~(1ull << i));
7160 			report_prefix_pushf("%s %llx", cr_name,
7161 						cr_saved & ~(1ull << i));
7162 			test_vmx_vmlaunch(
7163 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7164 			report_prefix_pop();
7165 		}
7166 	}
7167 
7168 	vmcs_write(cr, cr_saved);
7169 }
7170 
7171 /*
7172  * 1. The CR0 field must not set any bit to a value not supported in VMX
7173  *    operation.
7174  * 2. The CR4 field must not set any bit to a value not supported in VMX
7175  *    operation.
7176  * 3. On processors that support Intel 64 architecture, the CR3 field must
7177  *    be such that bits 63:52 and bits in the range 51:32 beyond the
7178  *    processor's physical-address width must be 0.
7179  *
7180  *  [Intel SDM]
7181  */
7182 static void test_host_ctl_regs(void)
7183 {
7184 	u64 fixed0, fixed1, cr3, cr3_saved;
7185 	int i;
7186 
7187 	/* Test CR0 */
7188 	fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0);
7189 	fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1);
7190 	test_ctl_reg("HOST_CR0", HOST_CR0, fixed0, fixed1);
7191 
7192 	/* Test CR4 */
7193 	fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0);
7194 	fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1) &
7195 		 ~(X86_CR4_SMEP | X86_CR4_SMAP);
7196 	test_ctl_reg("HOST_CR4", HOST_CR4, fixed0, fixed1);
7197 
7198 	/* Test CR3 */
7199 	cr3_saved = vmcs_read(HOST_CR3);
7200 	for (i = cpuid_maxphyaddr(); i < 64; i++) {
7201 		cr3 = cr3_saved | (1ul << i);
7202 		vmcs_write(HOST_CR3, cr3);
7203 		report_prefix_pushf("HOST_CR3 %lx", cr3);
7204 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7205 		report_prefix_pop();
7206 	}
7207 
7208 	vmcs_write(HOST_CR3, cr3_saved);
7209 }
7210 
7211 static void test_efer_vmlaunch(u32 fld, bool ok)
7212 {
7213 	if (fld == HOST_EFER) {
7214 		if (ok)
7215 			test_vmx_vmlaunch(0);
7216 		else
7217 			test_vmx_vmlaunch2(VMXERR_ENTRY_INVALID_CONTROL_FIELD,
7218 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7219 	} else {
7220 		test_guest_state("EFER test", !ok, GUEST_EFER, "GUEST_EFER");
7221 	}
7222 }
7223 
7224 static void test_efer_one(u32 fld, const char * fld_name, u64 efer,
7225 			  u32 ctrl_fld, u64 ctrl,
7226 			  int i, const char *efer_bit_name)
7227 {
7228 	bool ok;
7229 
7230 	ok = true;
7231 	if (ctrl_fld == EXI_CONTROLS && (ctrl & EXI_LOAD_EFER)) {
7232 		if (!!(efer & EFER_LMA) != !!(ctrl & EXI_HOST_64))
7233 			ok = false;
7234 		if (!!(efer & EFER_LME) != !!(ctrl & EXI_HOST_64))
7235 			ok = false;
7236 	}
7237 	if (ctrl_fld == ENT_CONTROLS && (ctrl & ENT_LOAD_EFER)) {
7238 		/* Check LMA too since CR0.PG is set.  */
7239 		if (!!(efer & EFER_LMA) != !!(ctrl & ENT_GUEST_64))
7240 			ok = false;
7241 		if (!!(efer & EFER_LME) != !!(ctrl & ENT_GUEST_64))
7242 			ok = false;
7243 	}
7244 
7245 	/*
7246 	 * Skip the test if it would enter the guest in 32-bit mode.
7247 	 * Perhaps write the test in assembly and make sure it
7248 	 * can be run in either mode?
7249 	 */
7250 	if (fld == GUEST_EFER && ok && !(ctrl & ENT_GUEST_64))
7251 		return;
7252 
7253 	vmcs_write(ctrl_fld, ctrl);
7254 	vmcs_write(fld, efer);
7255 	report_prefix_pushf("%s %s bit turned %s, controls %s",
7256 			    fld_name, efer_bit_name,
7257 			    (i & 1) ? "on" : "off",
7258 			    (i & 2) ? "on" : "off");
7259 
7260 	test_efer_vmlaunch(fld, ok);
7261 	report_prefix_pop();
7262 }
7263 
7264 static void test_efer_bit(u32 fld, const char * fld_name,
7265 			  u32 ctrl_fld, u64 ctrl_bit, u64 efer_bit,
7266 			  const char *efer_bit_name)
7267 {
7268 	u64 efer_saved = vmcs_read(fld);
7269 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7270 	int i;
7271 
7272 	for (i = 0; i < 4; i++) {
7273 		u64 efer = efer_saved & ~efer_bit;
7274 		u64 ctrl = ctrl_saved & ~ctrl_bit;
7275 
7276 		if (i & 1)
7277 			efer |= efer_bit;
7278 		if (i & 2)
7279 			ctrl |= ctrl_bit;
7280 
7281 		test_efer_one(fld, fld_name, efer, ctrl_fld, ctrl,
7282 			      i, efer_bit_name);
7283 	}
7284 
7285 	vmcs_write(ctrl_fld, ctrl_saved);
7286 	vmcs_write(fld, efer_saved);
7287 }
7288 
7289 static void test_efer(u32 fld, const char * fld_name, u32 ctrl_fld,
7290 		      u64 ctrl_bit1, u64 ctrl_bit2)
7291 {
7292 	u64 efer_saved = vmcs_read(fld);
7293 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7294 	u64 efer_reserved_bits =  ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
7295 	u64 i;
7296 	u64 efer;
7297 
7298 	if (cpu_has_efer_nx())
7299 		efer_reserved_bits &= ~EFER_NX;
7300 
7301 	if (!ctrl_bit1) {
7302 		printf("\"Load-IA32-EFER\" exit control not supported\n");
7303 		goto test_entry_exit_mode;
7304 	}
7305 
7306 	report_prefix_pushf("%s %lx", fld_name, efer_saved);
7307 	test_efer_vmlaunch(fld, true);
7308 	report_prefix_pop();
7309 
7310 	/*
7311 	 * Check reserved bits
7312 	 */
7313 	vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit1);
7314 	for (i = 0; i < 64; i++) {
7315 		if ((1ull << i) & efer_reserved_bits) {
7316 			efer = efer_saved | (1ull << i);
7317 			vmcs_write(fld, efer);
7318 			report_prefix_pushf("%s %lx", fld_name, efer);
7319 			test_efer_vmlaunch(fld, true);
7320 			report_prefix_pop();
7321 		}
7322 	}
7323 
7324 	vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit1);
7325 	for (i = 0; i < 64; i++) {
7326 		if ((1ull << i) & efer_reserved_bits) {
7327 			efer = efer_saved | (1ull << i);
7328 			vmcs_write(fld, efer);
7329 			report_prefix_pushf("%s %lx", fld_name, efer);
7330 			test_efer_vmlaunch(fld, false);
7331 			report_prefix_pop();
7332 		}
7333 	}
7334 
7335 	vmcs_write(ctrl_fld, ctrl_saved);
7336 	vmcs_write(fld, efer_saved);
7337 
7338 	/*
7339 	 * Check LMA and LME bits
7340 	 */
7341 	test_efer_bit(fld, fld_name,
7342 		      ctrl_fld, ctrl_bit1,
7343 		      EFER_LMA,
7344 		      "EFER_LMA");
7345 	test_efer_bit(fld, fld_name,
7346 		      ctrl_fld, ctrl_bit1,
7347 		      EFER_LME,
7348 		      "EFER_LME");
7349 
7350 test_entry_exit_mode:
7351 	test_efer_bit(fld, fld_name,
7352 		      ctrl_fld, ctrl_bit2,
7353 		      EFER_LMA,
7354 		      "EFER_LMA");
7355 	test_efer_bit(fld, fld_name,
7356 		      ctrl_fld, ctrl_bit2,
7357 		      EFER_LME,
7358 		      "EFER_LME");
7359 }
7360 
7361 /*
7362  * If the 'load IA32_EFER' VM-exit control is 1, bits reserved in the
7363  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7364  * the values of the LMA and LME bits in the field must each be that of
7365  * the 'host address-space size' VM-exit control.
7366  *
7367  *  [Intel SDM]
7368  */
7369 static void test_host_efer(void)
7370 {
7371 	test_efer(HOST_EFER, "HOST_EFER", EXI_CONTROLS,
7372 		  ctrl_exit_rev.clr & EXI_LOAD_EFER,
7373 		  EXI_HOST_64);
7374 }
7375 
7376 /*
7377  * If the 'load IA32_EFER' VM-enter control is 1, bits reserved in the
7378  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7379  * the values of the LMA and LME bits in the field must each be that of
7380  * the 'IA32e-mode guest' VM-exit control.
7381  */
7382 static void test_guest_efer(void)
7383 {
7384 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER)) {
7385 		printf("\"Load-IA32-EFER\" entry control not supported\n");
7386 		return;
7387 	}
7388 
7389 	vmcs_write(GUEST_EFER, rdmsr(MSR_EFER));
7390 	test_efer(GUEST_EFER, "GUEST_EFER", ENT_CONTROLS,
7391 		  ctrl_enter_rev.clr & ENT_LOAD_EFER,
7392 		  ENT_GUEST_64);
7393 }
7394 
7395 /*
7396  * PAT values higher than 8 are uninteresting since they're likely lumped
7397  * in with "8". We only test values above 8 one bit at a time,
7398  * in order to reduce the number of VM-Entries and keep the runtime reasonable.
7399  */
7400 #define	PAT_VAL_LIMIT	8
7401 
7402 static void test_pat(u32 field, const char * field_name, u32 ctrl_field,
7403 		     u64 ctrl_bit)
7404 {
7405 	u32 ctrl_saved = vmcs_read(ctrl_field);
7406 	u64 pat_saved = vmcs_read(field);
7407 	u64 i, val;
7408 	u32 j;
7409 	int error;
7410 
7411 	vmcs_clear_bits(ctrl_field, ctrl_bit);
7412 
7413 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7414 		/* Test PAT0..PAT7 fields */
7415 		for (j = 0; j < (i ? 8 : 1); j++) {
7416 			val = i << j * 8;
7417 			vmcs_write(field, val);
7418 			if (field == HOST_PAT) {
7419 				report_prefix_pushf("%s %lx", field_name, val);
7420 				test_vmx_vmlaunch(0);
7421 				report_prefix_pop();
7422 
7423 			} else {	// GUEST_PAT
7424 				test_guest_state("ENT_LOAD_PAT enabled", false,
7425 						 val, "GUEST_PAT");
7426 			}
7427 		}
7428 	}
7429 
7430 	vmcs_set_bits(ctrl_field, ctrl_bit);
7431 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7432 		/* Test PAT0..PAT7 fields */
7433 		for (j = 0; j < (i ? 8 : 1); j++) {
7434 			val = i << j * 8;
7435 			vmcs_write(field, val);
7436 
7437 			if (field == HOST_PAT) {
7438 				report_prefix_pushf("%s %lx", field_name, val);
7439 				if (i == 0x2 || i == 0x3 || i >= 0x8)
7440 					error =
7441 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
7442 				else
7443 					error = 0;
7444 
7445 				test_vmx_vmlaunch(error);
7446 				report_prefix_pop();
7447 
7448 			} else {	// GUEST_PAT
7449 				error = (i == 0x2 || i == 0x3 || i >= 0x8);
7450 				test_guest_state("ENT_LOAD_PAT enabled", !!error,
7451 						 val, "GUEST_PAT");
7452 			}
7453 
7454 		}
7455 	}
7456 
7457 	vmcs_write(ctrl_field, ctrl_saved);
7458 	vmcs_write(field, pat_saved);
7459 }
7460 
7461 /*
7462  *  If the "load IA32_PAT" VM-exit control is 1, the value of the field
7463  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7464  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7465  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7466  *  6 (WB), or 7 (UC-).
7467  *
7468  *  [Intel SDM]
7469  */
7470 static void test_load_host_pat(void)
7471 {
7472 	/*
7473 	 * "load IA32_PAT" VM-exit control
7474 	 */
7475 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) {
7476 		printf("\"Load-IA32-PAT\" exit control not supported\n");
7477 		return;
7478 	}
7479 
7480 	test_pat(HOST_PAT, "HOST_PAT", EXI_CONTROLS, EXI_LOAD_PAT);
7481 }
7482 
7483 union cpuidA_eax {
7484 	struct {
7485 		unsigned int version_id:8;
7486 		unsigned int num_counters_gp:8;
7487 		unsigned int bit_width:8;
7488 		unsigned int mask_length:8;
7489 	} split;
7490 	unsigned int full;
7491 };
7492 
7493 union cpuidA_edx {
7494 	struct {
7495 		unsigned int num_counters_fixed:5;
7496 		unsigned int bit_width_fixed:8;
7497 		unsigned int reserved:9;
7498 	} split;
7499 	unsigned int full;
7500 };
7501 
7502 static bool valid_pgc(u64 val)
7503 {
7504 	struct cpuid id;
7505 	union cpuidA_eax eax;
7506 	union cpuidA_edx edx;
7507 	u64 mask;
7508 
7509 	id = cpuid(0xA);
7510 	eax.full = id.a;
7511 	edx.full = id.d;
7512 	mask = ~(((1ull << eax.split.num_counters_gp) - 1) |
7513 		 (((1ull << edx.split.num_counters_fixed) - 1) << 32));
7514 
7515 	return !(val & mask);
7516 }
7517 
7518 static void test_pgc_vmlaunch(u32 xerror, u32 xreason, bool xfail, bool host)
7519 {
7520 	u32 inst_err;
7521 	u64 obs;
7522 	bool success;
7523 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7524 
7525 	if (host) {
7526 		success = vmlaunch_succeeds();
7527 		obs = rdmsr(data->msr);
7528 		if (!success) {
7529 			inst_err = vmcs_read(VMX_INST_ERROR);
7530 			report(xerror == inst_err, "vmlaunch failed, "
7531 			       "VMX Inst Error is %d (expected %d)",
7532 			       inst_err, xerror);
7533 		} else {
7534 			report(!data->enabled || data->exp == obs,
7535 			       "Host state is 0x%lx (expected 0x%lx)",
7536 			       obs, data->exp);
7537 			report(success != xfail, "vmlaunch succeeded");
7538 		}
7539 	} else {
7540 		test_guest_state("load GUEST_PERF_GLOBAL_CTRL", xfail,
7541 				 GUEST_PERF_GLOBAL_CTRL,
7542 				 "GUEST_PERF_GLOBAL_CTRL");
7543 	}
7544 }
7545 
7546 /*
7547  * test_load_perf_global_ctrl is a generic function for testing the
7548  * "load IA32_PERF_GLOBAL_CTRL" VM-{Entry,Exit} controls. This test function
7549  * tests the provided ctrl_val when disabled and enabled.
7550  *
7551  * @nr: VMCS field number corresponding to the host/guest state field
7552  * @name: Name of the above VMCS field for printing in test report
7553  * @ctrl_nr: VMCS field number corresponding to the VM-{Entry,Exit} control
7554  * @ctrl_val: Bit to set on the ctrl_field
7555  */
7556 static void test_perf_global_ctrl(u32 nr, const char *name, u32 ctrl_nr,
7557 				  const char *ctrl_name, u64 ctrl_val)
7558 {
7559 	u64 ctrl_saved = vmcs_read(ctrl_nr);
7560 	u64 pgc_saved = vmcs_read(nr);
7561 	u64 i, val;
7562 	bool host = nr == HOST_PERF_GLOBAL_CTRL;
7563 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7564 
7565 	data->msr = MSR_CORE_PERF_GLOBAL_CTRL;
7566 	msr_bmp_init();
7567 	vmcs_write(ctrl_nr, ctrl_saved & ~ctrl_val);
7568 	data->enabled = false;
7569 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=0 on %s",
7570 			    ctrl_name);
7571 
7572 	for (i = 0; i < 64; i++) {
7573 		val = 1ull << i;
7574 		vmcs_write(nr, val);
7575 		report_prefix_pushf("%s = 0x%lx", name, val);
7576 		test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7577 		report_prefix_pop();
7578 	}
7579 	report_prefix_pop();
7580 
7581 	vmcs_write(ctrl_nr, ctrl_saved | ctrl_val);
7582 	data->enabled = true;
7583 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=1 on %s",
7584 			    ctrl_name);
7585 	for (i = 0; i < 64; i++) {
7586 		val = 1ull << i;
7587 		data->exp = val;
7588 		vmcs_write(nr, val);
7589 		report_prefix_pushf("%s = 0x%lx", name, val);
7590 		if (valid_pgc(val)) {
7591 			test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7592 		} else {
7593 			if (host)
7594 				test_pgc_vmlaunch(
7595 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
7596 					0,
7597 					true,
7598 					host);
7599 			else
7600 				test_pgc_vmlaunch(
7601 					0,
7602 					VMX_ENTRY_FAILURE | VMX_FAIL_STATE,
7603 					true,
7604 					host);
7605 		}
7606 		report_prefix_pop();
7607 	}
7608 
7609 	data->enabled = false;
7610 	report_prefix_pop();
7611 	vmcs_write(ctrl_nr, ctrl_saved);
7612 	vmcs_write(nr, pgc_saved);
7613 }
7614 
7615 static void test_load_host_perf_global_ctrl(void)
7616 {
7617 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PERF)) {
7618 		printf("\"load IA32_PERF_GLOBAL_CTRL\" exit control not supported\n");
7619 		return;
7620 	}
7621 
7622 	test_perf_global_ctrl(HOST_PERF_GLOBAL_CTRL, "HOST_PERF_GLOBAL_CTRL",
7623 				   EXI_CONTROLS, "EXI_CONTROLS", EXI_LOAD_PERF);
7624 }
7625 
7626 
7627 static void test_load_guest_perf_global_ctrl(void)
7628 {
7629 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PERF)) {
7630 		printf("\"load IA32_PERF_GLOBAL_CTRL\" entry control not supported\n");
7631 		return;
7632 	}
7633 
7634 	test_perf_global_ctrl(GUEST_PERF_GLOBAL_CTRL, "GUEST_PERF_GLOBAL_CTRL",
7635 				   ENT_CONTROLS, "ENT_CONTROLS", ENT_LOAD_PERF);
7636 }
7637 
7638 
7639 /*
7640  * test_vmcs_field - test a value for the given VMCS field
7641  * @field: VMCS field
7642  * @field_name: string name of VMCS field
7643  * @bit_start: starting bit
7644  * @bit_end: ending bit
7645  * @val: value that the bit range must or must not contain
7646  * @valid_val: whether value given in 'val' must be valid or not
7647  * @error: expected VMCS error when vmentry fails for an invalid value
7648  */
7649 static void test_vmcs_field(u64 field, const char *field_name, u32 bit_start,
7650 			    u32 bit_end, u64 val, bool valid_val, u32 error)
7651 {
7652 	u64 field_saved = vmcs_read(field);
7653 	u32 i;
7654 	u64 tmp;
7655 	u32 bit_on;
7656 	u64 mask = ~0ull;
7657 
7658 	mask = (mask >> bit_end) << bit_end;
7659 	mask = mask | ((1 << bit_start) - 1);
7660 	tmp = (field_saved & mask) | (val << bit_start);
7661 
7662 	vmcs_write(field, tmp);
7663 	report_prefix_pushf("%s %lx", field_name, tmp);
7664 	if (valid_val)
7665 		test_vmx_vmlaunch(0);
7666 	else
7667 		test_vmx_vmlaunch(error);
7668 	report_prefix_pop();
7669 
7670 	for (i = bit_start; i <= bit_end; i = i + 2) {
7671 		bit_on = ((1ull < i) & (val << bit_start)) ? 0 : 1;
7672 		if (bit_on)
7673 			tmp = field_saved | (1ull << i);
7674 		else
7675 			tmp = field_saved & ~(1ull << i);
7676 		vmcs_write(field, tmp);
7677 		report_prefix_pushf("%s %lx", field_name, tmp);
7678 		if (valid_val)
7679 			test_vmx_vmlaunch(error);
7680 		else
7681 			test_vmx_vmlaunch(0);
7682 		report_prefix_pop();
7683 	}
7684 
7685 	vmcs_write(field, field_saved);
7686 }
7687 
7688 static void test_canonical(u64 field, const char * field_name, bool host)
7689 {
7690 	u64 addr_saved = vmcs_read(field);
7691 
7692 	/*
7693 	 * Use the existing value if possible.  Writing a random canonical
7694 	 * value is not an option as doing so would corrupt the field being
7695 	 * tested and likely hose the test.
7696 	 */
7697 	if (is_canonical(addr_saved)) {
7698 		if (host) {
7699 			report_prefix_pushf("%s %lx", field_name, addr_saved);
7700 			test_vmx_vmlaunch(0);
7701 			report_prefix_pop();
7702 		} else {
7703 			test_guest_state("Test canonical address", false,
7704 					 addr_saved, field_name);
7705 		}
7706 	}
7707 
7708 	vmcs_write(field, NONCANONICAL);
7709 
7710 	if (host) {
7711 		report_prefix_pushf("%s %llx", field_name, NONCANONICAL);
7712 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7713 		report_prefix_pop();
7714 	} else {
7715 		test_guest_state("Test non-canonical address", true,
7716 				 NONCANONICAL, field_name);
7717 	}
7718 
7719 	vmcs_write(field, addr_saved);
7720 }
7721 
7722 #define TEST_RPL_TI_FLAGS(reg, name)				\
7723 	test_vmcs_field(reg, name, 0, 2, 0x0, true,		\
7724 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7725 
7726 #define TEST_CS_TR_FLAGS(reg, name)				\
7727 	test_vmcs_field(reg, name, 3, 15, 0x0000, false,	\
7728 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7729 
7730 /*
7731  * 1. In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the
7732  *    RPL (bits 1:0) and the TI flag (bit 2) must be 0.
7733  * 2. The selector fields for CS and TR cannot be 0000H.
7734  * 3. The selector field for SS cannot be 0000H if the "host address-space
7735  *    size" VM-exit control is 0.
7736  * 4. On processors that support Intel 64 architecture, the base-address
7737  *    fields for FS, GS and TR must contain canonical addresses.
7738  */
7739 static void test_host_segment_regs(void)
7740 {
7741 	u16 selector_saved;
7742 
7743 	/*
7744 	 * Test RPL and TI flags
7745 	 */
7746 	TEST_RPL_TI_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7747 	TEST_RPL_TI_FLAGS(HOST_SEL_SS, "HOST_SEL_SS");
7748 	TEST_RPL_TI_FLAGS(HOST_SEL_DS, "HOST_SEL_DS");
7749 	TEST_RPL_TI_FLAGS(HOST_SEL_ES, "HOST_SEL_ES");
7750 	TEST_RPL_TI_FLAGS(HOST_SEL_FS, "HOST_SEL_FS");
7751 	TEST_RPL_TI_FLAGS(HOST_SEL_GS, "HOST_SEL_GS");
7752 	TEST_RPL_TI_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7753 
7754 	/*
7755 	 * Test that CS and TR fields can not be 0x0000
7756 	 */
7757 	TEST_CS_TR_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7758 	TEST_CS_TR_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7759 
7760 	/*
7761 	 * SS field can not be 0x0000 if "host address-space size" VM-exit
7762 	 * control is 0
7763 	 */
7764 	selector_saved = vmcs_read(HOST_SEL_SS);
7765 	vmcs_write(HOST_SEL_SS, 0);
7766 	report_prefix_pushf("HOST_SEL_SS 0");
7767 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7768 		test_vmx_vmlaunch(0);
7769 	} else {
7770 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7771 	}
7772 	report_prefix_pop();
7773 
7774 	vmcs_write(HOST_SEL_SS, selector_saved);
7775 
7776 	/*
7777 	 * Base address for FS, GS and TR must be canonical
7778 	 */
7779 	test_canonical(HOST_BASE_FS, "HOST_BASE_FS", true);
7780 	test_canonical(HOST_BASE_GS, "HOST_BASE_GS", true);
7781 	test_canonical(HOST_BASE_TR, "HOST_BASE_TR", true);
7782 }
7783 
7784 /*
7785  *  On processors that support Intel 64 architecture, the base-address
7786  *  fields for GDTR and IDTR must contain canonical addresses.
7787  */
7788 static void test_host_desc_tables(void)
7789 {
7790 	test_canonical(HOST_BASE_GDTR, "HOST_BASE_GDTR", true);
7791 	test_canonical(HOST_BASE_IDTR, "HOST_BASE_IDTR", true);
7792 }
7793 
7794 /*
7795  * If the "host address-space size" VM-exit control is 0, the following must
7796  * hold:
7797  *    - The "IA-32e mode guest" VM-entry control is 0.
7798  *    - Bit 17 of the CR4 field (corresponding to CR4.PCIDE) is 0.
7799  *    - Bits 63:32 in the RIP field are 0.
7800  *
7801  * If the "host address-space size" VM-exit control is 1, the following must
7802  * hold:
7803  *    - Bit 5 of the CR4 field (corresponding to CR4.PAE) is 1.
7804  *    - The RIP field contains a canonical address.
7805  *
7806  */
7807 static void test_host_addr_size(void)
7808 {
7809 	u64 cr4_saved = vmcs_read(HOST_CR4);
7810 	u64 rip_saved = vmcs_read(HOST_RIP);
7811 	u64 entry_ctrl_saved = vmcs_read(ENT_CONTROLS);
7812 	int i;
7813 	u64 tmp;
7814 
7815 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7816 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7817 		report_prefix_pushf("\"IA-32e mode guest\" enabled");
7818 		test_vmx_vmlaunch(0);
7819 		report_prefix_pop();
7820 
7821 		vmcs_write(HOST_CR4, cr4_saved | X86_CR4_PCIDE);
7822 		report_prefix_pushf("\"CR4.PCIDE\" set");
7823 		test_vmx_vmlaunch(0);
7824 		report_prefix_pop();
7825 
7826 		for (i = 32; i <= 63; i = i + 4) {
7827 			tmp = rip_saved | 1ull << i;
7828 			vmcs_write(HOST_RIP, tmp);
7829 			report_prefix_pushf("HOST_RIP %lx", tmp);
7830 			test_vmx_vmlaunch(0);
7831 			report_prefix_pop();
7832 		}
7833 
7834 		if (cr4_saved & X86_CR4_PAE) {
7835 			vmcs_write(HOST_CR4, cr4_saved  & ~X86_CR4_PAE);
7836 			report_prefix_pushf("\"CR4.PAE\" unset");
7837 			test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7838 		} else {
7839 			report_prefix_pushf("\"CR4.PAE\" set");
7840 			test_vmx_vmlaunch(0);
7841 		}
7842 		report_prefix_pop();
7843 
7844 		vmcs_write(HOST_RIP, NONCANONICAL);
7845 		report_prefix_pushf("HOST_RIP %llx", NONCANONICAL);
7846 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7847 		report_prefix_pop();
7848 
7849 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7850 		vmcs_write(HOST_RIP, rip_saved);
7851 		vmcs_write(HOST_CR4, cr4_saved);
7852 
7853 		/* Restore host's active RIP and CR4 values. */
7854 		report_prefix_pushf("restore host state");
7855 		test_vmx_vmlaunch(0);
7856 		report_prefix_pop();
7857 	}
7858 }
7859 
7860 /*
7861  * Check that the virtual CPU checks the VMX Host State Area as
7862  * documented in the Intel SDM.
7863  */
7864 static void vmx_host_state_area_test(void)
7865 {
7866 	/*
7867 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
7868 	 * fail due to invalid guest state, should we make it that
7869 	 * far.
7870 	 */
7871 	vmcs_write(GUEST_RFLAGS, 0);
7872 
7873 	test_host_ctl_regs();
7874 
7875 	test_canonical(HOST_SYSENTER_ESP, "HOST_SYSENTER_ESP", true);
7876 	test_canonical(HOST_SYSENTER_EIP, "HOST_SYSENTER_EIP", true);
7877 
7878 	test_host_efer();
7879 	test_load_host_pat();
7880 	test_host_segment_regs();
7881 	test_host_desc_tables();
7882 	test_host_addr_size();
7883 	test_load_host_perf_global_ctrl();
7884 }
7885 
7886 /*
7887  * If the "load debug controls" VM-entry control is 1, bits 63:32 in
7888  * the DR7 field must be 0.
7889  *
7890  * [Intel SDM]
7891  */
7892 static void test_guest_dr7(void)
7893 {
7894 	u32 ent_saved = vmcs_read(ENT_CONTROLS);
7895 	u64 dr7_saved = vmcs_read(GUEST_DR7);
7896 	u64 val;
7897 	int i;
7898 
7899 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS) {
7900 		vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7901 		for (i = 0; i < 64; i++) {
7902 			val = 1ull << i;
7903 			vmcs_write(GUEST_DR7, val);
7904 			test_guest_state("ENT_LOAD_DBGCTLS disabled", false,
7905 					 val, "GUEST_DR7");
7906 		}
7907 	}
7908 	if (ctrl_enter_rev.clr & ENT_LOAD_DBGCTLS) {
7909 		vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7910 		for (i = 0; i < 64; i++) {
7911 			val = 1ull << i;
7912 			vmcs_write(GUEST_DR7, val);
7913 			test_guest_state("ENT_LOAD_DBGCTLS enabled", i >= 32,
7914 					 val, "GUEST_DR7");
7915 		}
7916 	}
7917 	vmcs_write(GUEST_DR7, dr7_saved);
7918 	vmcs_write(ENT_CONTROLS, ent_saved);
7919 }
7920 
7921 /*
7922  *  If the "load IA32_PAT" VM-entry control is 1, the value of the field
7923  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7924  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7925  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7926  *  6 (WB), or 7 (UC-).
7927  *
7928  *  [Intel SDM]
7929  */
7930 static void test_load_guest_pat(void)
7931 {
7932 	/*
7933 	 * "load IA32_PAT" VM-entry control
7934 	 */
7935 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
7936 		printf("\"Load-IA32-PAT\" entry control not supported\n");
7937 		return;
7938 	}
7939 
7940 	test_pat(GUEST_PAT, "GUEST_PAT", ENT_CONTROLS, ENT_LOAD_PAT);
7941 }
7942 
7943 #define MSR_IA32_BNDCFGS_RSVD_MASK	0x00000ffc
7944 
7945 /*
7946  * If the "load IA32_BNDCFGS" VM-entry control is 1, the following
7947  * checks are performed on the field for the IA32_BNDCFGS MSR:
7948  *
7949  *   - Bits reserved in the IA32_BNDCFGS MSR must be 0.
7950  *   - The linear address in bits 63:12 must be canonical.
7951  *
7952  *  [Intel SDM]
7953  */
7954 static void test_load_guest_bndcfgs(void)
7955 {
7956 	u64 bndcfgs_saved = vmcs_read(GUEST_BNDCFGS);
7957 	u64 bndcfgs;
7958 
7959 	if (!(ctrl_enter_rev.clr & ENT_LOAD_BNDCFGS)) {
7960 		printf("\"Load-IA32-BNDCFGS\" entry control not supported\n");
7961 		return;
7962 	}
7963 
7964 	vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7965 
7966 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7967 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7968 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7969 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7970 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7971 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7972 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7973 
7974 	vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7975 
7976 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7977 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7978 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7979 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7980 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7981 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7982 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7983 
7984 	vmcs_write(GUEST_BNDCFGS, bndcfgs_saved);
7985 }
7986 
7987 #define	GUEST_SEG_UNUSABLE_MASK	(1u << 16)
7988 #define	GUEST_SEG_SEL_TI_MASK	(1u << 2)
7989 
7990 
7991 #define	TEST_SEGMENT_SEL(test, xfail, sel, val)				\
7992 do {									\
7993 	vmcs_write(sel, val);						\
7994 	test_guest_state(test " segment", xfail, val, xstr(sel));	\
7995 } while (0)
7996 
7997 #define	TEST_INVALID_SEG_SEL(sel, val) \
7998 	TEST_SEGMENT_SEL("Invalid: " xstr(val), true, sel, val);
7999 
8000 #define	TEST_VALID_SEG_SEL(sel, val) \
8001 	TEST_SEGMENT_SEL("Valid: " xstr(val), false, sel, val);
8002 
8003 /*
8004  * The following checks are done on the Selector field of the Guest Segment
8005  * Registers:
8006  *    - TR. The TI flag (bit 2) must be 0.
8007  *    - LDTR. If LDTR is usable, the TI flag (bit 2) must be 0.
8008  *    - SS. If the guest will not be virtual-8086 and the "unrestricted
8009  *	guest" VM-execution control is 0, the RPL (bits 1:0) must equal
8010  *	the RPL of the selector field for CS.
8011  *
8012  *  [Intel SDM]
8013  */
8014 static void test_guest_segment_sel_fields(void)
8015 {
8016 	u16 sel_saved;
8017 	u32 ar_saved;
8018 	u32 cpu_ctrl0_saved;
8019 	u32 cpu_ctrl1_saved;
8020 	u16 cs_rpl_bits;
8021 
8022 	/*
8023 	 * Test for GUEST_SEL_TR
8024 	 */
8025 	sel_saved = vmcs_read(GUEST_SEL_TR);
8026 	TEST_INVALID_SEG_SEL(GUEST_SEL_TR, sel_saved | GUEST_SEG_SEL_TI_MASK);
8027 	vmcs_write(GUEST_SEL_TR, sel_saved);
8028 
8029 	/*
8030 	 * Test for GUEST_SEL_LDTR
8031 	 */
8032 	sel_saved = vmcs_read(GUEST_SEL_LDTR);
8033 	ar_saved = vmcs_read(GUEST_AR_LDTR);
8034 	/* LDTR is set unusable */
8035 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8036 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
8037 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
8038 	/* LDTR is set usable */
8039 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8040 	TEST_INVALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
8041 
8042 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
8043 
8044 	vmcs_write(GUEST_AR_LDTR, ar_saved);
8045 	vmcs_write(GUEST_SEL_LDTR, sel_saved);
8046 
8047 	/*
8048 	 * Test for GUEST_SEL_SS
8049 	 */
8050 	cpu_ctrl0_saved = vmcs_read(CPU_EXEC_CTRL0);
8051 	cpu_ctrl1_saved = vmcs_read(CPU_EXEC_CTRL1);
8052 	ar_saved = vmcs_read(GUEST_AR_SS);
8053 	/* Turn off "unrestricted guest" vm-execution control */
8054 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved & ~CPU_URG);
8055 	cs_rpl_bits = vmcs_read(GUEST_SEL_CS) & 0x3;
8056 	sel_saved = vmcs_read(GUEST_SEL_SS);
8057 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
8058 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
8059 	/* Make SS usable if it's unusable or vice-versa */
8060 	if (ar_saved & GUEST_SEG_UNUSABLE_MASK)
8061 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8062 	else
8063 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8064 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
8065 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
8066 
8067 	/* Need a valid EPTP as the passing case fully enters the guest. */
8068 	if (enable_unrestricted_guest(true))
8069 		goto skip_ss_tests;
8070 
8071 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
8072 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
8073 
8074 	/* Make SS usable if it's unusable or vice-versa */
8075 	if (vmcs_read(GUEST_AR_SS) & GUEST_SEG_UNUSABLE_MASK)
8076 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8077 	else
8078 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8079 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
8080 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
8081 skip_ss_tests:
8082 
8083 	vmcs_write(GUEST_AR_SS, ar_saved);
8084 	vmcs_write(GUEST_SEL_SS, sel_saved);
8085 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrl0_saved);
8086 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved);
8087 }
8088 
8089 #define	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(xfail, seg_base)			\
8090 do {										\
8091 	addr_saved = vmcs_read(seg_base);					\
8092 	for (i = 32; i < 63; i = i + 4) {					\
8093 		addr = addr_saved | 1ull << i;					\
8094 		vmcs_write(seg_base, addr);					\
8095 		test_guest_state("seg.BASE[63:32] != 0, usable = " xstr(xfail),	\
8096 				 xfail, addr, xstr(seg_base));			\
8097 	}									\
8098 	vmcs_write(seg_base, addr_saved);					\
8099 } while (0)
8100 
8101 #define	TEST_SEGMENT_BASE_ADDR_CANONICAL(xfail, seg_base)		  \
8102 do {									  \
8103 	addr_saved = vmcs_read(seg_base);				  \
8104 	vmcs_write(seg_base, NONCANONICAL);				  \
8105 	test_guest_state("seg.BASE non-canonical, usable = " xstr(xfail), \
8106 			 xfail, NONCANONICAL, xstr(seg_base));		  \
8107 	vmcs_write(seg_base, addr_saved);				  \
8108 } while (0)
8109 
8110 /*
8111  * The following checks are done on the Base Address field of the Guest
8112  * Segment Registers on processors that support Intel 64 architecture:
8113  *    - TR, FS, GS : The address must be canonical.
8114  *    - LDTR : If LDTR is usable, the address must be canonical.
8115  *    - CS : Bits 63:32 of the address must be zero.
8116  *    - SS, DS, ES : If the register is usable, bits 63:32 of the address
8117  *	must be zero.
8118  *
8119  *  [Intel SDM]
8120  */
8121 static void test_guest_segment_base_addr_fields(void)
8122 {
8123 	u64 addr_saved;
8124 	u64 addr;
8125 	u32 ar_saved;
8126 	int i;
8127 
8128 	/*
8129 	 * The address of TR, FS, GS and LDTR must be canonical.
8130 	 */
8131 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_TR);
8132 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_FS);
8133 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_GS);
8134 	ar_saved = vmcs_read(GUEST_AR_LDTR);
8135 	/* Make LDTR unusable */
8136 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8137 	TEST_SEGMENT_BASE_ADDR_CANONICAL(false, GUEST_BASE_LDTR);
8138 	/* Make LDTR usable */
8139 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8140 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_LDTR);
8141 
8142 	vmcs_write(GUEST_AR_LDTR, ar_saved);
8143 
8144 	/*
8145 	 * Bits 63:32 in CS, SS, DS and ES base address must be zero
8146 	 */
8147 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_CS);
8148 	ar_saved = vmcs_read(GUEST_AR_SS);
8149 	/* Make SS unusable */
8150 	vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8151 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_SS);
8152 	/* Make SS usable */
8153 	vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8154 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_SS);
8155 	vmcs_write(GUEST_AR_SS, ar_saved);
8156 
8157 	ar_saved = vmcs_read(GUEST_AR_DS);
8158 	/* Make DS unusable */
8159 	vmcs_write(GUEST_AR_DS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8160 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_DS);
8161 	/* Make DS usable */
8162 	vmcs_write(GUEST_AR_DS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8163 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_DS);
8164 	vmcs_write(GUEST_AR_DS, ar_saved);
8165 
8166 	ar_saved = vmcs_read(GUEST_AR_ES);
8167 	/* Make ES unusable */
8168 	vmcs_write(GUEST_AR_ES, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8169 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_ES);
8170 	/* Make ES usable */
8171 	vmcs_write(GUEST_AR_ES, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8172 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_ES);
8173 	vmcs_write(GUEST_AR_ES, ar_saved);
8174 }
8175 
8176 /*
8177  * Check that the virtual CPU checks the VMX Guest State Area as
8178  * documented in the Intel SDM.
8179  */
8180 static void vmx_guest_state_area_test(void)
8181 {
8182 	vmx_set_test_stage(1);
8183 	test_set_guest(guest_state_test_main);
8184 
8185 	/*
8186 	 * The IA32_SYSENTER_ESP field and the IA32_SYSENTER_EIP field
8187 	 * must each contain a canonical address.
8188 	 */
8189 	test_canonical(GUEST_SYSENTER_ESP, "GUEST_SYSENTER_ESP", false);
8190 	test_canonical(GUEST_SYSENTER_EIP, "GUEST_SYSENTER_EIP", false);
8191 
8192 	test_guest_dr7();
8193 	test_load_guest_pat();
8194 	test_guest_efer();
8195 	test_load_guest_perf_global_ctrl();
8196 	test_load_guest_bndcfgs();
8197 
8198 	test_guest_segment_sel_fields();
8199 	test_guest_segment_base_addr_fields();
8200 
8201 	test_canonical(GUEST_BASE_GDTR, "GUEST_BASE_GDTR", false);
8202 	test_canonical(GUEST_BASE_IDTR, "GUEST_BASE_IDTR", false);
8203 
8204 	u32 guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_GDTR);
8205 	int i;
8206 	for (i = 16; i <= 31; i++) {
8207 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8208 		vmcs_write(GUEST_LIMIT_GDTR, tmp);
8209 		test_guest_state("GDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_GDTR");
8210 	}
8211 	vmcs_write(GUEST_LIMIT_GDTR, guest_desc_limit_saved);
8212 
8213 	guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_IDTR);
8214 	for (i = 16; i <= 31; i++) {
8215 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8216 		vmcs_write(GUEST_LIMIT_IDTR, tmp);
8217 		test_guest_state("IDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_IDTR");
8218 	}
8219 	vmcs_write(GUEST_LIMIT_IDTR, guest_desc_limit_saved);
8220 
8221 	/*
8222 	 * Let the guest finish execution
8223 	 */
8224 	vmx_set_test_stage(2);
8225 	enter_guest();
8226 }
8227 
8228 extern void unrestricted_guest_main(void);
8229 asm (".code32\n"
8230 	"unrestricted_guest_main:\n"
8231 	"vmcall\n"
8232 	"nop\n"
8233 	"mov $1, %edi\n"
8234 	"call hypercall\n"
8235 	".code64\n");
8236 
8237 static void setup_unrestricted_guest(void)
8238 {
8239 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) & ~(X86_CR0_PG));
8240 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) & ~ENT_GUEST_64);
8241 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) & ~EFER_LMA);
8242 	vmcs_write(GUEST_RIP, virt_to_phys(unrestricted_guest_main));
8243 }
8244 
8245 static void unsetup_unrestricted_guest(void)
8246 {
8247 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) | X86_CR0_PG);
8248 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_GUEST_64);
8249 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) | EFER_LMA);
8250 	vmcs_write(GUEST_RIP, (u64) phys_to_virt(vmcs_read(GUEST_RIP)));
8251 	vmcs_write(GUEST_RSP, (u64) phys_to_virt(vmcs_read(GUEST_RSP)));
8252 }
8253 
8254 /*
8255  * If "unrestricted guest" secondary VM-execution control is set, guests
8256  * can run in unpaged protected mode.
8257  */
8258 static void vmentry_unrestricted_guest_test(void)
8259 {
8260 	if (enable_unrestricted_guest(true)) {
8261 		report_skip("Unrestricted guest not supported");
8262 		return;
8263 	}
8264 
8265 	test_set_guest(unrestricted_guest_main);
8266 	setup_unrestricted_guest();
8267 	test_guest_state("Unrestricted guest test", false, CPU_URG, "CPU_URG");
8268 
8269 	/*
8270 	 * Let the guest finish execution as a regular guest
8271 	 */
8272 	unsetup_unrestricted_guest();
8273 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) & ~CPU_URG);
8274 	enter_guest();
8275 }
8276 
8277 static bool valid_vmcs_for_vmentry(void)
8278 {
8279 	struct vmcs *current_vmcs = NULL;
8280 
8281 	if (vmcs_save(&current_vmcs))
8282 		return false;
8283 
8284 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
8285 }
8286 
8287 static void try_vmentry_in_movss_shadow(void)
8288 {
8289 	u32 vm_inst_err;
8290 	u32 flags;
8291 	bool early_failure = false;
8292 	u32 expected_flags = X86_EFLAGS_FIXED;
8293 	bool valid_vmcs = valid_vmcs_for_vmentry();
8294 
8295 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
8296 
8297 	/*
8298 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
8299 	 * unsupported VMCS component").
8300 	 */
8301 	vmcs_write(~0u, 0);
8302 
8303 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
8304 			      "vmwrite %%rsp, %%rdx;"
8305 			      "mov 0f, %%rax;"
8306 			      "mov %[host_rip], %%edx;"
8307 			      "vmwrite %%rax, %%rdx;"
8308 			      "mov $-1, %%ah;"
8309 			      "sahf;"
8310 			      "mov %%ss, %%ax;"
8311 			      "mov %%ax, %%ss;"
8312 			      "vmlaunch;"
8313 			      "mov $1, %[early_failure];"
8314 			      "0: lahf;"
8315 			      "movzbl %%ah, %[flags]"
8316 			      : [early_failure] "+r" (early_failure),
8317 				[flags] "=&a" (flags)
8318 			      : [host_rsp] "i" (HOST_RSP),
8319 				[host_rip] "i" (HOST_RIP)
8320 			      : "rdx", "cc", "memory");
8321 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
8322 
8323 	report(early_failure, "Early VM-entry failure");
8324 	report(flags == expected_flags, "RFLAGS[8:0] is %x (actual %x)",
8325 	       expected_flags, flags);
8326 	if (valid_vmcs)
8327 		report(vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
8328 		       "VM-instruction error is %d (actual %d)",
8329 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
8330 }
8331 
8332 static void vmentry_movss_shadow_test(void)
8333 {
8334 	struct vmcs *orig_vmcs;
8335 
8336 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
8337 
8338 	/*
8339 	 * Set the launched flag on the current VMCS to verify the correct
8340 	 * error priority, below.
8341 	 */
8342 	test_set_guest(v2_null_test_guest);
8343 	enter_guest();
8344 
8345 	/*
8346 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
8347 	 * fail due to invalid guest state (if we make it that far).
8348 	 */
8349 	vmcs_write(GUEST_RFLAGS, 0);
8350 
8351 	/*
8352 	 * "VM entry with events blocked by MOV SS" takes precedence over
8353 	 * "VMLAUNCH with non-clear VMCS."
8354 	 */
8355 	report_prefix_push("valid current-VMCS");
8356 	try_vmentry_in_movss_shadow();
8357 	report_prefix_pop();
8358 
8359 	/*
8360 	 * VMfailInvalid takes precedence over "VM entry with events
8361 	 * blocked by MOV SS."
8362 	 */
8363 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
8364 	report_prefix_push("no current-VMCS");
8365 	try_vmentry_in_movss_shadow();
8366 	report_prefix_pop();
8367 
8368 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
8369 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8370 }
8371 
8372 static void vmx_cr_load_test(void)
8373 {
8374 	unsigned long cr3, cr4, orig_cr3, orig_cr4;
8375 
8376 	orig_cr4 = read_cr4();
8377 	orig_cr3 = read_cr3();
8378 
8379 	if (!this_cpu_has(X86_FEATURE_PCID)) {
8380 		report_skip("PCID not detected");
8381 		return;
8382 	}
8383 	if (!this_cpu_has(X86_FEATURE_MCE)) {
8384 		report_skip("MCE not detected");
8385 		return;
8386 	}
8387 
8388 	TEST_ASSERT(!(orig_cr3 & X86_CR3_PCID_MASK));
8389 
8390 	/* Enable PCID for L1. */
8391 	cr4 = orig_cr4 | X86_CR4_PCIDE;
8392 	cr3 = orig_cr3 | 0x1;
8393 	TEST_ASSERT(!write_cr4_checking(cr4));
8394 	write_cr3(cr3);
8395 
8396 	test_set_guest(v2_null_test_guest);
8397 	vmcs_write(HOST_CR4, cr4);
8398 	vmcs_write(HOST_CR3, cr3);
8399 	enter_guest();
8400 
8401 	/*
8402 	 * No exception is expected.
8403 	 *
8404 	 * NB. KVM loads the last guest write to CR4 into CR4 read
8405 	 *     shadow. In order to trigger an exit to KVM, we can toggle a
8406 	 *     bit that is owned by KVM. We use CR4.MCE, which shall
8407 	 *     have no side effect because normally no guest MCE (e.g., as the
8408 	 *     result of bad memory) would happen during this test.
8409 	 */
8410 	TEST_ASSERT(!write_cr4_checking(cr4 ^ X86_CR4_MCE));
8411 
8412 	/* Cleanup L1 state. */
8413 	write_cr3(orig_cr3);
8414 	TEST_ASSERT(!write_cr4_checking(orig_cr4));
8415 }
8416 
8417 static void vmx_cr4_osxsave_test_guest(void)
8418 {
8419 	write_cr4(read_cr4() & ~X86_CR4_OSXSAVE);
8420 }
8421 
8422 /*
8423  * Ensure that kvm recalculates the L1 guest's CPUID.01H:ECX.OSXSAVE
8424  * after VM-exit from an L2 guest that sets CR4.OSXSAVE to a different
8425  * value than in L1.
8426  */
8427 static void vmx_cr4_osxsave_test(void)
8428 {
8429 	if (!this_cpu_has(X86_FEATURE_XSAVE)) {
8430 		report_skip("XSAVE not detected");
8431 		return;
8432 	}
8433 
8434 	if (!(read_cr4() & X86_CR4_OSXSAVE)) {
8435 		unsigned long cr4 = read_cr4() | X86_CR4_OSXSAVE;
8436 
8437 		write_cr4(cr4);
8438 		vmcs_write(GUEST_CR4, cr4);
8439 		vmcs_write(HOST_CR4, cr4);
8440 	}
8441 
8442 	TEST_ASSERT(cpuid_osxsave());
8443 
8444 	test_set_guest(vmx_cr4_osxsave_test_guest);
8445 	enter_guest();
8446 
8447 	TEST_ASSERT(cpuid_osxsave());
8448 }
8449 
8450 static void vmx_nm_test_guest(void)
8451 {
8452 	write_cr0(read_cr0() | X86_CR0_TS);
8453 	asm volatile("fnop");
8454 }
8455 
8456 static void check_nm_exit(const char *test)
8457 {
8458 	u32 reason = vmcs_read(EXI_REASON);
8459 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
8460 	const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8461 		NM_VECTOR;
8462 
8463 	report(reason == VMX_EXC_NMI && intr_info == expected, "%s", test);
8464 }
8465 
8466 /*
8467  * This test checks that:
8468  *
8469  * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then
8470  *     a subsequent #NM VM-exit is reflected to L1.
8471  *
8472  * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a
8473  *     subsequent #NM VM-exit is reflected to L1.
8474  */
8475 static void vmx_nm_test(void)
8476 {
8477 	unsigned long cr0 = read_cr0();
8478 
8479 	test_set_guest(vmx_nm_test_guest);
8480 
8481 	/*
8482 	 * L1 wants to intercept #NM exceptions encountered in L2.
8483 	 */
8484 	vmcs_write(EXC_BITMAP, 1 << NM_VECTOR);
8485 
8486 	/*
8487 	 * Launch L2 with CR0.TS clear, but don't claim host ownership of
8488 	 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop,
8489 	 * which will raise #NM. L0 should reflect the #NM VM-exit to L1.
8490 	 */
8491 	vmcs_write(CR0_MASK, 0);
8492 	vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS);
8493 	enter_guest();
8494 	check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1");
8495 
8496 	/*
8497 	 * Re-enter L2 at the fnop instruction, with CR0.TS clear but
8498 	 * CR0.EM set. The fnop will still raise #NM, and L0 should
8499 	 * reflect the #NM VM-exit to L1.
8500 	 */
8501 	vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM);
8502 	enter_guest();
8503 	check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1");
8504 
8505 	/*
8506 	 * Re-enter L2 at the fnop instruction, with both CR0.TS and
8507 	 * CR0.EM clear. There will be no #NM, and the L2 guest should
8508 	 * exit normally.
8509 	 */
8510 	vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM));
8511 	enter_guest();
8512 }
8513 
8514 bool vmx_pending_event_ipi_fired;
8515 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
8516 {
8517 	vmx_pending_event_ipi_fired = true;
8518 	eoi();
8519 }
8520 
8521 bool vmx_pending_event_guest_run;
8522 static void vmx_pending_event_guest(void)
8523 {
8524 	vmcall();
8525 	vmx_pending_event_guest_run = true;
8526 }
8527 
8528 static void vmx_pending_event_test_core(bool guest_hlt)
8529 {
8530 	int ipi_vector = 0xf1;
8531 
8532 	vmx_pending_event_ipi_fired = false;
8533 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
8534 
8535 	vmx_pending_event_guest_run = false;
8536 	test_set_guest(vmx_pending_event_guest);
8537 
8538 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
8539 
8540 	enter_guest();
8541 	skip_exit_vmcall();
8542 
8543 	if (guest_hlt)
8544 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8545 
8546 	irq_disable();
8547 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
8548 				   APIC_DM_FIXED | ipi_vector,
8549 				   0);
8550 
8551 	enter_guest();
8552 
8553 	assert_exit_reason(VMX_EXTINT);
8554 	report(!vmx_pending_event_guest_run,
8555 	       "Guest did not run before host received IPI");
8556 
8557 	irq_enable();
8558 	asm volatile ("nop");
8559 	irq_disable();
8560 	report(vmx_pending_event_ipi_fired,
8561 	       "Got pending interrupt after IRQ enabled");
8562 
8563 	if (guest_hlt)
8564 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8565 
8566 	enter_guest();
8567 	report(vmx_pending_event_guest_run,
8568 	       "Guest finished running when no interrupt");
8569 }
8570 
8571 static void vmx_pending_event_test(void)
8572 {
8573 	vmx_pending_event_test_core(false);
8574 }
8575 
8576 static void vmx_pending_event_hlt_test(void)
8577 {
8578 	vmx_pending_event_test_core(true);
8579 }
8580 
8581 static int vmx_window_test_db_count;
8582 
8583 static void vmx_window_test_db_handler(struct ex_regs *regs)
8584 {
8585 	vmx_window_test_db_count++;
8586 }
8587 
8588 static void vmx_nmi_window_test_guest(void)
8589 {
8590 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8591 
8592 	asm volatile("vmcall\n\t"
8593 		     "nop\n\t");
8594 
8595 	handle_exception(DB_VECTOR, NULL);
8596 }
8597 
8598 static void verify_nmi_window_exit(u64 rip)
8599 {
8600 	u32 exit_reason = vmcs_read(EXI_REASON);
8601 
8602 	report(exit_reason == VMX_NMI_WINDOW,
8603 	       "Exit reason (%d) is 'NMI window'", exit_reason);
8604 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8605 	       vmcs_read(GUEST_RIP), rip);
8606 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8607 }
8608 
8609 static void vmx_nmi_window_test(void)
8610 {
8611 	u64 nop_addr;
8612 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8613 
8614 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
8615 		report_skip("CPU does not support the \"Virtual NMIs\" VM-execution control.");
8616 		return;
8617 	}
8618 
8619 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
8620 		report_skip("CPU does not support the \"NMI-window exiting\" VM-execution control.");
8621 		return;
8622 	}
8623 
8624 	vmx_window_test_db_count = 0;
8625 
8626 	report_prefix_push("NMI-window");
8627 	test_set_guest(vmx_nmi_window_test_guest);
8628 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
8629 	enter_guest();
8630 	skip_exit_vmcall();
8631 	nop_addr = vmcs_read(GUEST_RIP);
8632 
8633 	/*
8634 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
8635 	 * RIP will not advance.
8636 	 */
8637 	report_prefix_push("active, no blocking");
8638 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8639 	enter_guest();
8640 	verify_nmi_window_exit(nop_addr);
8641 	report_prefix_pop();
8642 
8643 	/*
8644 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
8645 	 * a VM-exit on the next instruction after the nop. (The nop
8646 	 * is one byte.)
8647 	 */
8648 	report_prefix_push("active, blocking by MOV-SS");
8649 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8650 	enter_guest();
8651 	verify_nmi_window_exit(nop_addr + 1);
8652 	report_prefix_pop();
8653 
8654 	/*
8655 	 * Ask for "NMI-window exiting" (with event injection), and
8656 	 * expect a VM-exit after the event is injected. (RIP should
8657 	 * be at the address specified in the IDT entry for #DB.)
8658 	 */
8659 	report_prefix_push("active, no blocking, injecting #DB");
8660 	vmcs_write(ENT_INTR_INFO,
8661 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8662 	enter_guest();
8663 	verify_nmi_window_exit((u64)db_fault_addr);
8664 	report_prefix_pop();
8665 
8666 	/*
8667 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
8668 	 * a VM-exit after the next IRET (i.e. after the #DB handler
8669 	 * returns). So, RIP should be back at one byte past the nop.
8670 	 */
8671 	report_prefix_push("active, blocking by NMI");
8672 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
8673 	enter_guest();
8674 	verify_nmi_window_exit(nop_addr + 1);
8675 	report(vmx_window_test_db_count == 1,
8676 	       "#DB handler executed once (actual %d times)",
8677 	       vmx_window_test_db_count);
8678 	report_prefix_pop();
8679 
8680 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8681 		report_skip("CPU does not support activity state HLT.");
8682 	} else {
8683 		/*
8684 		 * Ask for "NMI-window exiting" when entering activity
8685 		 * state HLT, and expect an immediate VM-exit. RIP is
8686 		 * still one byte past the nop.
8687 		 */
8688 		report_prefix_push("halted, no blocking");
8689 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8690 		enter_guest();
8691 		verify_nmi_window_exit(nop_addr + 1);
8692 		report_prefix_pop();
8693 
8694 		/*
8695 		 * Ask for "NMI-window exiting" when entering activity
8696 		 * state HLT (with event injection), and expect a
8697 		 * VM-exit after the event is injected. (RIP should be
8698 		 * at the address specified in the IDT entry for #DB.)
8699 		 */
8700 		report_prefix_push("halted, no blocking, injecting #DB");
8701 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8702 		vmcs_write(ENT_INTR_INFO,
8703 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8704 			   DB_VECTOR);
8705 		enter_guest();
8706 		verify_nmi_window_exit((u64)db_fault_addr);
8707 		report_prefix_pop();
8708 	}
8709 
8710 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8711 	enter_guest();
8712 	report_prefix_pop();
8713 }
8714 
8715 static void vmx_intr_window_test_guest(void)
8716 {
8717 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8718 
8719 	/*
8720 	 * The two consecutive STIs are to ensure that only the first
8721 	 * one has a shadow. Note that NOP and STI are one byte
8722 	 * instructions.
8723 	 */
8724 	asm volatile("vmcall\n\t"
8725 		     "nop\n\t"
8726 		     "sti\n\t"
8727 		     "sti\n\t");
8728 
8729 	handle_exception(DB_VECTOR, NULL);
8730 }
8731 
8732 static void verify_intr_window_exit(u64 rip)
8733 {
8734 	u32 exit_reason = vmcs_read(EXI_REASON);
8735 
8736 	report(exit_reason == VMX_INTR_WINDOW,
8737 	       "Exit reason (%d) is 'interrupt window'", exit_reason);
8738 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8739 	       vmcs_read(GUEST_RIP), rip);
8740 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8741 }
8742 
8743 static void vmx_intr_window_test(void)
8744 {
8745 	u64 vmcall_addr;
8746 	u64 nop_addr;
8747 	unsigned int orig_db_gate_type;
8748 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8749 
8750 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
8751 		report_skip("CPU does not support the \"interrupt-window exiting\" VM-execution control.");
8752 		return;
8753 	}
8754 
8755 	/*
8756 	 * Change the IDT entry for #DB from interrupt gate to trap gate,
8757 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
8758 	 * be disabled after vectoring a #DB.
8759 	 */
8760 	orig_db_gate_type = boot_idt[DB_VECTOR].type;
8761 	boot_idt[DB_VECTOR].type = 15;
8762 
8763 	report_prefix_push("interrupt-window");
8764 	test_set_guest(vmx_intr_window_test_guest);
8765 	enter_guest();
8766 	assert_exit_reason(VMX_VMCALL);
8767 	vmcall_addr = vmcs_read(GUEST_RIP);
8768 
8769 	/*
8770 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
8771 	 * no blocking; expect an immediate VM-exit. Note that we have
8772 	 * not advanced past the vmcall instruction yet, so RIP should
8773 	 * point to the vmcall instruction.
8774 	 */
8775 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
8776 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8777 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
8778 	enter_guest();
8779 	verify_intr_window_exit(vmcall_addr);
8780 	report_prefix_pop();
8781 
8782 	/*
8783 	 * Ask for "interrupt-window exiting" (with event injection)
8784 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
8785 	 * the event is injected. That is, RIP should should be at the
8786 	 * address specified in the IDT entry for #DB.
8787 	 */
8788 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #DB");
8789 	vmcs_write(ENT_INTR_INFO,
8790 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8791 	vmcall_addr = vmcs_read(GUEST_RIP);
8792 	enter_guest();
8793 	verify_intr_window_exit((u64)db_fault_addr);
8794 	report_prefix_pop();
8795 
8796 	/*
8797 	 * Let the L2 guest run through the IRET, back to the VMCALL.
8798 	 * We have to clear the "interrupt-window exiting"
8799 	 * VM-execution control, or it would just keep causing
8800 	 * VM-exits. Then, advance past the VMCALL and set the
8801 	 * "interrupt-window exiting" VM-execution control again.
8802 	 */
8803 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8804 	enter_guest();
8805 	skip_exit_vmcall();
8806 	nop_addr = vmcs_read(GUEST_RIP);
8807 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8808 
8809 	/*
8810 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
8811 	 * RFLAGS.IF set, and expect a VM-exit on the next
8812 	 * instruction. (NOP is one byte.)
8813 	 */
8814 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
8815 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8816 	enter_guest();
8817 	verify_intr_window_exit(nop_addr + 1);
8818 	report_prefix_pop();
8819 
8820 	/*
8821 	 * Back up to the NOP and ask for "interrupt-window exiting"
8822 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
8823 	 * on the next instruction. (NOP is one byte.)
8824 	 */
8825 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
8826 	vmcs_write(GUEST_RIP, nop_addr);
8827 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
8828 	enter_guest();
8829 	verify_intr_window_exit(nop_addr + 1);
8830 	report_prefix_pop();
8831 
8832 	/*
8833 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
8834 	 * and expect a VM-exit on the instruction following the STI
8835 	 * shadow. Only the first STI (which is one byte past the NOP)
8836 	 * should have a shadow. The second STI (which is two bytes
8837 	 * past the NOP) has no shadow. Therefore, the interrupt
8838 	 * window opens at three bytes past the NOP.
8839 	 */
8840 	report_prefix_push("active, RFLAGS.IF = 0");
8841 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8842 	enter_guest();
8843 	verify_intr_window_exit(nop_addr + 3);
8844 	report_prefix_pop();
8845 
8846 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8847 		report_skip("CPU does not support activity state HLT.");
8848 	} else {
8849 		/*
8850 		 * Ask for "interrupt-window exiting" when entering
8851 		 * activity state HLT, and expect an immediate
8852 		 * VM-exit. RIP is still three bytes past the nop.
8853 		 */
8854 		report_prefix_push("halted, no blocking");
8855 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8856 		enter_guest();
8857 		verify_intr_window_exit(nop_addr + 3);
8858 		report_prefix_pop();
8859 
8860 		/*
8861 		 * Ask for "interrupt-window exiting" when entering
8862 		 * activity state HLT (with event injection), and
8863 		 * expect a VM-exit after the event is injected. That
8864 		 * is, RIP should should be at the address specified
8865 		 * in the IDT entry for #DB.
8866 		 */
8867 		report_prefix_push("halted, no blocking, injecting #DB");
8868 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8869 		vmcs_write(ENT_INTR_INFO,
8870 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8871 			   DB_VECTOR);
8872 		enter_guest();
8873 		verify_intr_window_exit((u64)db_fault_addr);
8874 		report_prefix_pop();
8875 	}
8876 
8877 	boot_idt[DB_VECTOR].type = orig_db_gate_type;
8878 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8879 	enter_guest();
8880 	report_prefix_pop();
8881 }
8882 
8883 #define GUEST_TSC_OFFSET (1u << 30)
8884 
8885 static u64 guest_tsc;
8886 
8887 static void vmx_store_tsc_test_guest(void)
8888 {
8889 	guest_tsc = rdtsc();
8890 }
8891 
8892 /*
8893  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
8894  * list, the value saved is not subject to the TSC offset that is
8895  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
8896  */
8897 static void vmx_store_tsc_test(void)
8898 {
8899 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
8900 	u64 low, high;
8901 
8902 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
8903 		report_skip("'Use TSC offsetting' not supported");
8904 		return;
8905 	}
8906 
8907 	test_set_guest(vmx_store_tsc_test_guest);
8908 
8909 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
8910 	vmcs_write(EXI_MSR_ST_CNT, 1);
8911 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
8912 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
8913 
8914 	low = rdtsc();
8915 	enter_guest();
8916 	high = rdtsc();
8917 
8918 	report(low + GUEST_TSC_OFFSET <= guest_tsc &&
8919 	       guest_tsc <= high + GUEST_TSC_OFFSET,
8920 	       "RDTSC value in the guest (%lu) is in range [%lu, %lu]",
8921 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
8922 	report(low <= msr_entry.value && msr_entry.value <= high,
8923 	       "IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
8924 	       msr_entry.value, low, high);
8925 }
8926 
8927 static void vmx_preemption_timer_zero_test_db_handler(struct ex_regs *regs)
8928 {
8929 }
8930 
8931 static void vmx_preemption_timer_zero_test_guest(void)
8932 {
8933 	while (vmx_get_test_stage() < 3)
8934 		vmcall();
8935 }
8936 
8937 static void vmx_preemption_timer_zero_activate_preemption_timer(void)
8938 {
8939 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
8940 	vmcs_write(PREEMPT_TIMER_VALUE, 0);
8941 }
8942 
8943 static void vmx_preemption_timer_zero_advance_past_vmcall(void)
8944 {
8945 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
8946 	enter_guest();
8947 	skip_exit_vmcall();
8948 }
8949 
8950 static void vmx_preemption_timer_zero_inject_db(bool intercept_db)
8951 {
8952 	vmx_preemption_timer_zero_activate_preemption_timer();
8953 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK |
8954 		   INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8955 	vmcs_write(EXC_BITMAP, intercept_db ? 1 << DB_VECTOR : 0);
8956 	enter_guest();
8957 }
8958 
8959 static void vmx_preemption_timer_zero_set_pending_dbg(u32 exception_bitmap)
8960 {
8961 	vmx_preemption_timer_zero_activate_preemption_timer();
8962 	vmcs_write(GUEST_PENDING_DEBUG, BIT(12) | DR_TRAP1);
8963 	vmcs_write(EXC_BITMAP, exception_bitmap);
8964 	enter_guest();
8965 }
8966 
8967 static void vmx_preemption_timer_zero_expect_preempt_at_rip(u64 expected_rip)
8968 {
8969 	u32 reason = (u32)vmcs_read(EXI_REASON);
8970 	u64 guest_rip = vmcs_read(GUEST_RIP);
8971 
8972 	report(reason == VMX_PREEMPT && guest_rip == expected_rip,
8973 	       "Exit reason is 0x%x (expected 0x%x) and guest RIP is %lx (0x%lx expected).",
8974 	       reason, VMX_PREEMPT, guest_rip, expected_rip);
8975 }
8976 
8977 /*
8978  * This test ensures that when the VMX preemption timer is zero at
8979  * VM-entry, a VM-exit occurs after any event injection and after any
8980  * pending debug exceptions are raised, but before execution of any
8981  * guest instructions.
8982  */
8983 static void vmx_preemption_timer_zero_test(void)
8984 {
8985 	u64 db_fault_address = (u64)get_idt_addr(&boot_idt[DB_VECTOR]);
8986 	handler old_db;
8987 	u32 reason;
8988 
8989 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
8990 		report_skip("'Activate VMX-preemption timer' not supported");
8991 		return;
8992 	}
8993 
8994 	/*
8995 	 * Install a custom #DB handler that doesn't abort.
8996 	 */
8997 	old_db = handle_exception(DB_VECTOR,
8998 				  vmx_preemption_timer_zero_test_db_handler);
8999 
9000 	test_set_guest(vmx_preemption_timer_zero_test_guest);
9001 
9002 	/*
9003 	 * VMX-preemption timer should fire after event injection.
9004 	 */
9005 	vmx_set_test_stage(0);
9006 	vmx_preemption_timer_zero_inject_db(0);
9007 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
9008 	vmx_preemption_timer_zero_advance_past_vmcall();
9009 
9010 	/*
9011 	 * VMX-preemption timer should fire after event injection.
9012 	 * Exception bitmap is irrelevant, since you can't intercept
9013 	 * an event that you injected.
9014 	 */
9015 	vmx_set_test_stage(1);
9016 	vmx_preemption_timer_zero_inject_db(true);
9017 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
9018 	vmx_preemption_timer_zero_advance_past_vmcall();
9019 
9020 	/*
9021 	 * VMX-preemption timer should fire after pending debug exceptions
9022 	 * have delivered a #DB trap.
9023 	 */
9024 	vmx_set_test_stage(2);
9025 	vmx_preemption_timer_zero_set_pending_dbg(0);
9026 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
9027 	vmx_preemption_timer_zero_advance_past_vmcall();
9028 
9029 	/*
9030 	 * VMX-preemption timer would fire after pending debug exceptions
9031 	 * have delivered a #DB trap, but in this case, the #DB trap is
9032 	 * intercepted.
9033 	 */
9034 	vmx_set_test_stage(3);
9035 	vmx_preemption_timer_zero_set_pending_dbg(1 << DB_VECTOR);
9036 	reason = (u32)vmcs_read(EXI_REASON);
9037 	report(reason == VMX_EXC_NMI, "Exit reason is 0x%x (expected 0x%x)",
9038 	       reason, VMX_EXC_NMI);
9039 
9040 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9041 	enter_guest();
9042 
9043 	handle_exception(DB_VECTOR, old_db);
9044 }
9045 
9046 static u64 vmx_preemption_timer_tf_test_prev_rip;
9047 
9048 static void vmx_preemption_timer_tf_test_db_handler(struct ex_regs *regs)
9049 {
9050 	extern char vmx_preemption_timer_tf_test_endloop;
9051 
9052 	if (vmx_get_test_stage() == 2) {
9053 		/*
9054 		 * Stage 2 means that we're done, one way or another.
9055 		 * Arrange for the iret to drop us out of the wbinvd
9056 		 * loop and stop single-stepping.
9057 		 */
9058 		regs->rip = (u64)&vmx_preemption_timer_tf_test_endloop;
9059 		regs->rflags &= ~X86_EFLAGS_TF;
9060 	} else if (regs->rip == vmx_preemption_timer_tf_test_prev_rip) {
9061 		/*
9062 		 * The RIP should alternate between the wbinvd and the
9063 		 * jmp instruction in the code below. If we ever see
9064 		 * the same instruction twice in a row, that means a
9065 		 * single-step trap has been dropped. Let the
9066 		 * hypervisor know about the failure by executing a
9067 		 * VMCALL.
9068 		 */
9069 		vmcall();
9070 	}
9071 	vmx_preemption_timer_tf_test_prev_rip = regs->rip;
9072 }
9073 
9074 static void vmx_preemption_timer_tf_test_guest(void)
9075 {
9076 	/*
9077 	 * The hypervisor doesn't intercept WBINVD, so the loop below
9078 	 * shouldn't be a problem--it's just two instructions
9079 	 * executing in VMX non-root mode. However, when the
9080 	 * hypervisor is running in a virtual environment, the parent
9081 	 * hypervisor might intercept WBINVD and emulate it. If the
9082 	 * parent hypervisor is broken, the single-step trap after the
9083 	 * WBINVD might be lost.
9084 	 */
9085 	asm volatile("vmcall\n\t"
9086 		     "0: wbinvd\n\t"
9087 		     "1: jmp 0b\n\t"
9088 		     "vmx_preemption_timer_tf_test_endloop:");
9089 }
9090 
9091 /*
9092  * Ensure that the delivery of a "VMX-preemption timer expired"
9093  * VM-exit doesn't disrupt single-stepping in the guest. Note that
9094  * passing this test doesn't ensure correctness, because the test will
9095  * only fail if the VMX-preemtion timer fires at the right time (or
9096  * the wrong time, as it were).
9097  */
9098 static void vmx_preemption_timer_tf_test(void)
9099 {
9100 	handler old_db;
9101 	u32 reason;
9102 	int i;
9103 
9104 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9105 		report_skip("'Activate VMX-preemption timer' not supported");
9106 		return;
9107 	}
9108 
9109 	old_db = handle_exception(DB_VECTOR,
9110 				  vmx_preemption_timer_tf_test_db_handler);
9111 
9112 	test_set_guest(vmx_preemption_timer_tf_test_guest);
9113 
9114 	enter_guest();
9115 	skip_exit_vmcall();
9116 
9117 	vmx_set_test_stage(1);
9118 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9119 	vmcs_write(PREEMPT_TIMER_VALUE, 50000);
9120 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9121 
9122 	/*
9123 	 * The only exit we should see is "VMX-preemption timer
9124 	 * expired."  If we get a VMCALL exit, that means the #DB
9125 	 * handler has detected a missing single-step trap. It doesn't
9126 	 * matter where the guest RIP is when the VMX-preemption timer
9127 	 * expires (whether it's in the WBINVD loop or in the #DB
9128 	 * handler)--a single-step trap should never be discarded.
9129 	 */
9130 	for (i = 0; i < 10000; i++) {
9131 		enter_guest();
9132 		reason = (u32)vmcs_read(EXI_REASON);
9133 		if (reason == VMX_PREEMPT)
9134 			continue;
9135 		TEST_ASSERT(reason == VMX_VMCALL);
9136 		skip_exit_insn();
9137 		break;
9138 	}
9139 
9140 	report(reason == VMX_PREEMPT, "No single-step traps skipped");
9141 
9142 	vmx_set_test_stage(2);
9143 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9144 	enter_guest();
9145 
9146 	handle_exception(DB_VECTOR, old_db);
9147 }
9148 
9149 #define VMX_PREEMPTION_TIMER_EXPIRY_CYCLES 1000000
9150 
9151 static u64 vmx_preemption_timer_expiry_start;
9152 static u64 vmx_preemption_timer_expiry_finish;
9153 
9154 static void vmx_preemption_timer_expiry_test_guest(void)
9155 {
9156 	vmcall();
9157 	vmx_preemption_timer_expiry_start = fenced_rdtsc();
9158 
9159 	while (vmx_get_test_stage() == 0)
9160 		vmx_preemption_timer_expiry_finish = fenced_rdtsc();
9161 }
9162 
9163 /*
9164  * Test that the VMX-preemption timer is not excessively delayed.
9165  *
9166  * Per the SDM, volume 3, VM-entry starts the VMX-preemption timer
9167  * with the unsigned value in the VMX-preemption timer-value field,
9168  * and the VMX-preemption timer counts down by 1 every time bit X in
9169  * the TSC changes due to a TSC increment (where X is
9170  * IA32_VMX_MISC[4:0]). If the timer counts down to zero in any state
9171  * other than the wait-for-SIPI state, the logical processor
9172  * transitions to the C0 C-state and causes a VM-exit.
9173  *
9174  * The guest code above reads the starting TSC after VM-entry. At this
9175  * point, the VMX-preemption timer has already been activated. Next,
9176  * the guest code reads the current TSC in a loop, storing the value
9177  * read to memory.
9178  *
9179  * If the RDTSC in the loop reads a value past the VMX-preemption
9180  * timer deadline, then the VMX-preemption timer VM-exit must be
9181  * delivered before the next instruction retires. Even if a higher
9182  * priority SMI is delivered first, the VMX-preemption timer VM-exit
9183  * must be delivered before the next instruction retires. Hence, a TSC
9184  * value past the VMX-preemption timer deadline might be read, but it
9185  * cannot be stored. If a TSC value past the deadline *is* stored,
9186  * then the architectural specification has been violated.
9187  */
9188 static void vmx_preemption_timer_expiry_test(void)
9189 {
9190 	u32 preemption_timer_value;
9191 	union vmx_misc misc;
9192 	u64 tsc_deadline;
9193 	u32 reason;
9194 
9195 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9196 		report_skip("'Activate VMX-preemption timer' not supported");
9197 		return;
9198 	}
9199 
9200 	test_set_guest(vmx_preemption_timer_expiry_test_guest);
9201 
9202 	enter_guest();
9203 	skip_exit_vmcall();
9204 
9205 	misc.val = rdmsr(MSR_IA32_VMX_MISC);
9206 	preemption_timer_value =
9207 		VMX_PREEMPTION_TIMER_EXPIRY_CYCLES >> misc.pt_bit;
9208 
9209 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9210 	vmcs_write(PREEMPT_TIMER_VALUE, preemption_timer_value);
9211 	vmx_set_test_stage(0);
9212 
9213 	enter_guest();
9214 	reason = (u32)vmcs_read(EXI_REASON);
9215 	TEST_ASSERT(reason == VMX_PREEMPT);
9216 
9217 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9218 	vmx_set_test_stage(1);
9219 	enter_guest();
9220 
9221 	tsc_deadline = ((vmx_preemption_timer_expiry_start >> misc.pt_bit) <<
9222 			misc.pt_bit) + (preemption_timer_value << misc.pt_bit);
9223 
9224 	report(vmx_preemption_timer_expiry_finish < tsc_deadline,
9225 	       "Last stored guest TSC (%lu) < TSC deadline (%lu)",
9226 	       vmx_preemption_timer_expiry_finish, tsc_deadline);
9227 }
9228 
9229 static void vmx_db_test_guest(void)
9230 {
9231 	/*
9232 	 * For a hardware generated single-step #DB.
9233 	 */
9234 	asm volatile("vmcall;"
9235 		     "nop;"
9236 		     ".Lpost_nop:");
9237 	/*
9238 	 * ...in a MOVSS shadow, with pending debug exceptions.
9239 	 */
9240 	asm volatile("vmcall;"
9241 		     "nop;"
9242 		     ".Lpost_movss_nop:");
9243 	/*
9244 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
9245 	 * emulates it in software.)
9246 	 */
9247 	asm volatile("vmcall;"
9248 		     "wbinvd;"
9249 		     ".Lpost_wbinvd:");
9250 	/*
9251 	 * ...in a MOVSS shadow, with pending debug exceptions.
9252 	 */
9253 	asm volatile("vmcall;"
9254 		     "wbinvd;"
9255 		     ".Lpost_movss_wbinvd:");
9256 	/*
9257 	 * For a hardware generated single-step #DB in a transactional region.
9258 	 */
9259 	asm volatile("vmcall;"
9260 		     ".Lxbegin: xbegin .Lskip_rtm;"
9261 		     "xend;"
9262 		     ".Lskip_rtm:");
9263 }
9264 
9265 /*
9266  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
9267  * L2. No #DB is delivered and L2 continues to the next point of
9268  * interest.
9269  */
9270 static void dismiss_db(void)
9271 {
9272 	vmcs_write(GUEST_PENDING_DEBUG, 0);
9273 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
9274 	enter_guest();
9275 }
9276 
9277 /*
9278  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
9279  * Then throw away the #DB exception and resume L2.
9280  */
9281 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
9282 			  void *expected_rip, u64 expected_exit_qual,
9283 			  u64 expected_dr6)
9284 {
9285 	u32 reason = vmcs_read(EXI_REASON);
9286 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
9287 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
9288 	u64 guest_rip = vmcs_read(GUEST_RIP);
9289 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
9290 	u64 dr6 = read_dr6();
9291 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
9292 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
9293 
9294 	report(reason == VMX_EXC_NMI && intr_info == expected_intr_info,
9295 	       "Expected #DB VM-exit");
9296 	report((u64)expected_rip == guest_rip, "Expected RIP %p (actual %lx)",
9297 	       expected_rip, guest_rip);
9298 	report_xfail(xfail_pdbg, 0 == guest_pending_dbg,
9299 		     "Expected pending debug exceptions 0 (actual %lx)",
9300 		     guest_pending_dbg);
9301 	report_xfail(xfail_qual, expected_exit_qual == exit_qual,
9302 		     "Expected exit qualification %lx (actual %lx)",
9303 		     expected_exit_qual, exit_qual);
9304 	report_xfail(xfail_dr6, expected_dr6 == dr6,
9305 		     "Expected DR6 %lx (actual %lx)", expected_dr6, dr6);
9306 	dismiss_db();
9307 }
9308 
9309 /*
9310  * Assuming the guest has just exited on a VMCALL instruction, skip
9311  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
9312  * pending debug exceptions are non-zero, set the VMCS up as if the
9313  * previous instruction was a MOVSS that generated the indicated
9314  * pending debug exceptions. Then enter L2.
9315  */
9316 static void single_step_guest(const char *test_name, u64 starting_dr6,
9317 			      u64 pending_debug_exceptions)
9318 {
9319 	printf("\n%s\n", test_name);
9320 	skip_exit_vmcall();
9321 	write_dr6(starting_dr6);
9322 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9323 	if (pending_debug_exceptions) {
9324 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
9325 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
9326 	}
9327 	enter_guest();
9328 }
9329 
9330 /*
9331  * When L1 intercepts #DB, verify that a single-step trap clears
9332  * pending debug exceptions, populates the exit qualification field
9333  * properly, and that DR6 is not prematurely clobbered. In a
9334  * (simulated) MOVSS shadow, make sure that the pending debug
9335  * exception bits are properly accumulated into the exit qualification
9336  * field.
9337  */
9338 static void vmx_db_test(void)
9339 {
9340 	/*
9341 	 * We are going to set a few arbitrary bits in DR6 to verify that
9342 	 * (a) DR6 is not modified by an intercepted #DB, and
9343 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
9344          *     the exit qualification field for a subsequent #DB exception.
9345 	 */
9346 	const u64 starting_dr6 = DR6_RESERVED | BIT(13) | DR_TRAP3 | DR_TRAP1;
9347 	extern char post_nop asm(".Lpost_nop");
9348 	extern char post_movss_nop asm(".Lpost_movss_nop");
9349 	extern char post_wbinvd asm(".Lpost_wbinvd");
9350 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
9351 	extern char xbegin asm(".Lxbegin");
9352 	extern char skip_rtm asm(".Lskip_rtm");
9353 
9354 	/*
9355 	 * L1 wants to intercept #DB exceptions encountered in L2.
9356 	 */
9357 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
9358 
9359 	/*
9360 	 * Start L2 and run it up to the first point of interest.
9361 	 */
9362 	test_set_guest(vmx_db_test_guest);
9363 	enter_guest();
9364 
9365 	/*
9366 	 * Hardware-delivered #DB trap for single-step sets the
9367 	 * standard that L0 has to follow for emulated instructions.
9368 	 */
9369 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
9370 	check_db_exit(false, false, false, &post_nop, DR_STEP, starting_dr6);
9371 
9372 	/*
9373 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
9374 	 * also sets the standard that L0 has to follow for emulated
9375 	 * instructions. Here, we establish the VMCS pending debug
9376 	 * exceptions to indicate that the simulated MOVSS triggered a
9377 	 * data breakpoint as well as the single-step trap.
9378 	 */
9379 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
9380 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0 );
9381 	check_db_exit(false, false, false, &post_movss_nop, DR_STEP | DR_TRAP0,
9382 		      starting_dr6);
9383 
9384 	/*
9385 	 * L0 synthesized #DB trap for single-step is buggy, because
9386 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
9387 	 * reconstitute the exit qualification from the prematurely
9388 	 * modified DR6, but fails miserably.
9389 	 */
9390 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
9391 	check_db_exit(false, false, false, &post_wbinvd, DR_STEP, starting_dr6);
9392 
9393 	/*
9394 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
9395 	 * even worse, because L0 also leaves the pending debug
9396 	 * exceptions in the VMCS instead of accumulating them into
9397 	 * the exit qualification field for the #DB exception.
9398 	 */
9399 	single_step_guest("Software synthesized single-step in MOVSS shadow",
9400 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0);
9401 	check_db_exit(true, false, true, &post_movss_wbinvd, DR_STEP | DR_TRAP0,
9402 		      starting_dr6);
9403 
9404 	/*
9405 	 * Optional RTM test for hardware that supports RTM, to
9406 	 * demonstrate that the current volume 3 of the SDM
9407 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
9408 	 * qualification for debug exceptions is not reserved. It is
9409 	 * set to 1 if a debug exception (#DB) or a breakpoint
9410 	 * exception (#BP) occurs inside an RTM region while advanced
9411 	 * debugging of RTM transactional regions is enabled.
9412 	 */
9413 	if (this_cpu_has(X86_FEATURE_RTM)) {
9414 		vmcs_write(ENT_CONTROLS,
9415 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
9416 		/*
9417 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
9418 		 * in the guest to enable advanced debugging of RTM
9419 		 * transactional regions.
9420 		 */
9421 		vmcs_write(GUEST_DR7, BIT(11));
9422 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
9423 		single_step_guest("Hardware delivered single-step in "
9424 				  "transactional region", starting_dr6, 0);
9425 		check_db_exit(false, false, false, &xbegin, BIT(16),
9426 			      starting_dr6);
9427 	} else {
9428 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
9429 		enter_guest();
9430 	}
9431 }
9432 
9433 static void enable_vid(void)
9434 {
9435 	void *virtual_apic_page;
9436 
9437 	assert(cpu_has_apicv());
9438 
9439 	disable_intercept_for_x2apic_msrs();
9440 
9441 	virtual_apic_page = alloc_page();
9442 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
9443 
9444 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
9445 
9446 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
9447 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
9448 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
9449 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
9450 
9451 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
9452 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
9453 }
9454 
9455 static void trigger_ioapic_scan_thread(void *data)
9456 {
9457 	/* Wait until other CPU entered L2 */
9458 	while (vmx_get_test_stage() != 1)
9459 		;
9460 
9461 	/* Trigger ioapic scan */
9462 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
9463 	vmx_set_test_stage(2);
9464 }
9465 
9466 static void irq_79_handler_guest(isr_regs_t *regs)
9467 {
9468 	eoi();
9469 
9470 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
9471 	vmcall();
9472 }
9473 
9474 /*
9475  * Constant for num of busy-loop iterations after which
9476  * a timer interrupt should have happened in host
9477  */
9478 #define TIMER_INTERRUPT_DELAY 100000000
9479 
9480 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
9481 {
9482 	handle_irq(0x79, irq_79_handler_guest);
9483 	irq_enable();
9484 
9485 	/* Signal to L1 CPU to trigger ioapic scan */
9486 	vmx_set_test_stage(1);
9487 	/* Wait until L1 CPU to trigger ioapic scan */
9488 	while (vmx_get_test_stage() != 2)
9489 		;
9490 
9491 	/*
9492 	 * Wait for L0 timer interrupt to be raised while we run in L2
9493 	 * such that L0 will process the IOAPIC scan request before
9494 	 * resuming L2
9495 	 */
9496 	delay(TIMER_INTERRUPT_DELAY);
9497 
9498 	asm volatile ("int $0x79");
9499 }
9500 
9501 static void vmx_eoi_bitmap_ioapic_scan_test(void)
9502 {
9503 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
9504 		report_skip(__func__);
9505 		return;
9506 	}
9507 
9508 	enable_vid();
9509 
9510 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
9511 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
9512 
9513 	/*
9514 	 * Launch L2.
9515 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
9516 	 * In case the reason isn't VMX_VMCALL, the asserion inside
9517 	 * skip_exit_vmcall() will fail.
9518 	 */
9519 	enter_guest();
9520 	skip_exit_vmcall();
9521 
9522 	/* Let L2 finish */
9523 	enter_guest();
9524 	report(1, __func__);
9525 }
9526 
9527 #define HLT_WITH_RVI_VECTOR		(0xf1)
9528 
9529 bool vmx_hlt_with_rvi_guest_isr_fired;
9530 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
9531 {
9532 	vmx_hlt_with_rvi_guest_isr_fired = true;
9533 	eoi();
9534 }
9535 
9536 static void vmx_hlt_with_rvi_guest(void)
9537 {
9538 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
9539 
9540 	irq_enable();
9541 	asm volatile ("nop");
9542 
9543 	vmcall();
9544 }
9545 
9546 static void vmx_hlt_with_rvi_test(void)
9547 {
9548 	if (!cpu_has_apicv()) {
9549 		report_skip(__func__);
9550 		return;
9551 	}
9552 
9553 	enable_vid();
9554 
9555 	vmx_hlt_with_rvi_guest_isr_fired = false;
9556 	test_set_guest(vmx_hlt_with_rvi_guest);
9557 
9558 	enter_guest();
9559 	skip_exit_vmcall();
9560 
9561 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
9562 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
9563 	enter_guest();
9564 
9565 	report(vmx_hlt_with_rvi_guest_isr_fired, "Interrupt raised in guest");
9566 }
9567 
9568 static void set_irq_line_thread(void *data)
9569 {
9570 	/* Wait until other CPU entered L2 */
9571 	while (vmx_get_test_stage() != 1)
9572 		;
9573 
9574 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
9575 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9576 	vmx_set_test_stage(2);
9577 }
9578 
9579 static bool irq_78_handler_vmcall_before_eoi;
9580 static void irq_78_handler_guest(isr_regs_t *regs)
9581 {
9582 	set_irq_line(0xf, 0);
9583 	if (irq_78_handler_vmcall_before_eoi)
9584 		vmcall();
9585 	eoi();
9586 	vmcall();
9587 }
9588 
9589 static void vmx_apic_passthrough_guest(void)
9590 {
9591 	handle_irq(0x78, irq_78_handler_guest);
9592 	irq_enable();
9593 
9594 	/* If requested, wait for other CPU to trigger ioapic scan */
9595 	if (vmx_get_test_stage() < 1) {
9596 		vmx_set_test_stage(1);
9597 		while (vmx_get_test_stage() != 2)
9598 			;
9599 	}
9600 
9601 	set_irq_line(0xf, 1);
9602 }
9603 
9604 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
9605 {
9606 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
9607 		report_skip(__func__);
9608 		return;
9609 	}
9610 
9611 	/* Test device is required for generating IRQs */
9612 	if (!test_device_enabled()) {
9613 		report_skip(__func__);
9614 		return;
9615 	}
9616 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9617 	u64 cpu_ctrl_1 = 0;
9618 
9619 	disable_intercept_for_x2apic_msrs();
9620 
9621 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9622 
9623 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9624 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9625 
9626 	if (set_irq_line_from_thread) {
9627 		irq_78_handler_vmcall_before_eoi = false;
9628 		on_cpu_async(1, set_irq_line_thread, NULL);
9629 	} else {
9630 		irq_78_handler_vmcall_before_eoi = true;
9631 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9632 		vmx_set_test_stage(2);
9633 	}
9634 	test_set_guest(vmx_apic_passthrough_guest);
9635 
9636 	if (irq_78_handler_vmcall_before_eoi) {
9637 		/* Before EOI remote_irr should still be set */
9638 		enter_guest();
9639 		skip_exit_vmcall();
9640 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
9641 			"IOAPIC pass-through: remote_irr=1 before EOI");
9642 	}
9643 
9644 	/* After EOI remote_irr should be cleared */
9645 	enter_guest();
9646 	skip_exit_vmcall();
9647 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
9648 		"IOAPIC pass-through: remote_irr=0 after EOI");
9649 
9650 	/* Let L2 finish */
9651 	enter_guest();
9652 	report(1, __func__);
9653 }
9654 
9655 static void vmx_apic_passthrough_test(void)
9656 {
9657 	vmx_apic_passthrough(false);
9658 }
9659 
9660 static void vmx_apic_passthrough_thread_test(void)
9661 {
9662 	vmx_apic_passthrough(true);
9663 }
9664 
9665 static void vmx_apic_passthrough_tpr_threshold_guest(void)
9666 {
9667 	cli();
9668 	apic_set_tpr(0);
9669 }
9670 
9671 static bool vmx_apic_passthrough_tpr_threshold_ipi_isr_fired;
9672 static void vmx_apic_passthrough_tpr_threshold_ipi_isr(isr_regs_t *regs)
9673 {
9674 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = true;
9675 	eoi();
9676 }
9677 
9678 static void vmx_apic_passthrough_tpr_threshold_test(void)
9679 {
9680 	int ipi_vector = 0xe1;
9681 
9682 	disable_intercept_for_x2apic_msrs();
9683 	vmcs_clear_bits(PIN_CONTROLS, PIN_EXTINT);
9684 
9685 	/* Raise L0 TPR-threshold by queueing vector in LAPIC IRR */
9686 	cli();
9687 	apic_set_tpr((ipi_vector >> 4) + 1);
9688 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
9689 			APIC_DM_FIXED | ipi_vector,
9690 			0);
9691 
9692 	test_set_guest(vmx_apic_passthrough_tpr_threshold_guest);
9693 	enter_guest();
9694 
9695 	report(apic_get_tpr() == 0, "TPR was zero by guest");
9696 
9697 	/* Clean pending self-IPI */
9698 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = false;
9699 	handle_irq(ipi_vector, vmx_apic_passthrough_tpr_threshold_ipi_isr);
9700 	sti();
9701 	asm volatile ("nop");
9702 	report(vmx_apic_passthrough_tpr_threshold_ipi_isr_fired, "self-IPI fired");
9703 
9704 	report(1, __func__);
9705 }
9706 
9707 static u64 init_signal_test_exit_reason;
9708 static bool init_signal_test_thread_continued;
9709 
9710 static void init_signal_test_thread(void *data)
9711 {
9712 	struct vmcs *test_vmcs = data;
9713 
9714 	/* Enter VMX operation (i.e. exec VMXON) */
9715 	u64 *ap_vmxon_region = alloc_page();
9716 	enable_vmx();
9717 	init_vmx(ap_vmxon_region);
9718 	_vmx_on(ap_vmxon_region);
9719 
9720 	/* Signal CPU have entered VMX operation */
9721 	vmx_set_test_stage(1);
9722 
9723 	/* Wait for BSP CPU to send INIT signal */
9724 	while (vmx_get_test_stage() != 2)
9725 		;
9726 
9727 	/*
9728 	 * Signal that we continue as usual as INIT signal
9729 	 * should be blocked while CPU is in VMX operation
9730 	 */
9731 	vmx_set_test_stage(3);
9732 
9733 	/* Wait for signal to enter VMX non-root mode */
9734 	while (vmx_get_test_stage() != 4)
9735 		;
9736 
9737 	/* Enter VMX non-root mode */
9738 	test_set_guest(v2_null_test_guest);
9739 	make_vmcs_current(test_vmcs);
9740 	enter_guest();
9741 	/* Save exit reason for BSP CPU to compare to expected result */
9742 	init_signal_test_exit_reason = vmcs_read(EXI_REASON);
9743 	/* VMCLEAR test-vmcs so it could be loaded by BSP CPU */
9744 	vmcs_clear(test_vmcs);
9745 	launched = false;
9746 	/* Signal that CPU exited to VMX root mode */
9747 	vmx_set_test_stage(5);
9748 
9749 	/* Wait for BSP CPU to signal to exit VMX operation */
9750 	while (vmx_get_test_stage() != 6)
9751 		;
9752 
9753 	/* Exit VMX operation (i.e. exec VMXOFF) */
9754 	vmx_off();
9755 
9756 	/*
9757 	 * Signal to BSP CPU that we continue as usual as INIT signal
9758 	 * should have been consumed by VMX_INIT exit from guest
9759 	 */
9760 	vmx_set_test_stage(7);
9761 
9762 	/* Wait for BSP CPU to signal to enter VMX operation */
9763 	while (vmx_get_test_stage() != 8)
9764 		;
9765 	/* Enter VMX operation (i.e. exec VMXON) */
9766 	_vmx_on(ap_vmxon_region);
9767 	/* Signal to BSP we are in VMX operation */
9768 	vmx_set_test_stage(9);
9769 
9770 	/* Wait for BSP CPU to send INIT signal */
9771 	while (vmx_get_test_stage() != 10)
9772 		;
9773 
9774 	/* Exit VMX operation (i.e. exec VMXOFF) */
9775 	vmx_off();
9776 
9777 	/*
9778 	 * Exiting VMX operation should result in latched
9779 	 * INIT signal being processed. Therefore, we should
9780 	 * never reach the below code. Thus, signal to BSP
9781 	 * CPU if we have reached here so it is able to
9782 	 * report an issue if it happens.
9783 	 */
9784 	init_signal_test_thread_continued = true;
9785 }
9786 
9787 #define INIT_SIGNAL_TEST_DELAY	100000000ULL
9788 
9789 static void vmx_init_signal_test(void)
9790 {
9791 	struct vmcs *test_vmcs;
9792 
9793 	if (cpu_count() < 2) {
9794 		report_skip(__func__);
9795 		return;
9796 	}
9797 
9798 	/* VMCLEAR test-vmcs so it could be loaded by other CPU */
9799 	vmcs_save(&test_vmcs);
9800 	vmcs_clear(test_vmcs);
9801 
9802 	vmx_set_test_stage(0);
9803 	on_cpu_async(1, init_signal_test_thread, test_vmcs);
9804 
9805 	/* Wait for other CPU to enter VMX operation */
9806 	while (vmx_get_test_stage() != 1)
9807 		;
9808 
9809 	/* Send INIT signal to other CPU */
9810 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9811 				   id_map[1]);
9812 	/* Signal other CPU we have sent INIT signal */
9813 	vmx_set_test_stage(2);
9814 
9815 	/*
9816 	 * Wait reasonable amount of time for INIT signal to
9817 	 * be received on other CPU and verify that other CPU
9818 	 * have proceed as usual to next test stage as INIT
9819 	 * signal should be blocked while other CPU in
9820 	 * VMX operation
9821 	 */
9822 	delay(INIT_SIGNAL_TEST_DELAY);
9823 	report(vmx_get_test_stage() == 3,
9824 	       "INIT signal blocked when CPU in VMX operation");
9825 	/* No point to continue if we failed at this point */
9826 	if (vmx_get_test_stage() != 3)
9827 		return;
9828 
9829 	/* Signal other CPU to enter VMX non-root mode */
9830 	init_signal_test_exit_reason = -1ull;
9831 	vmx_set_test_stage(4);
9832 	/*
9833 	 * Wait reasonable amont of time for other CPU
9834 	 * to exit to VMX root mode
9835 	 */
9836 	delay(INIT_SIGNAL_TEST_DELAY);
9837 	if (vmx_get_test_stage() != 5) {
9838 		report(false, "Pending INIT signal didn't result in VMX exit");
9839 		return;
9840 	}
9841 	report(init_signal_test_exit_reason == VMX_INIT,
9842 			"INIT signal during VMX non-root mode result in exit-reason %s (%lu)",
9843 			exit_reason_description(init_signal_test_exit_reason),
9844 			init_signal_test_exit_reason);
9845 
9846 	/* Run guest to completion */
9847 	make_vmcs_current(test_vmcs);
9848 	enter_guest();
9849 
9850 	/* Signal other CPU to exit VMX operation */
9851 	init_signal_test_thread_continued = false;
9852 	vmx_set_test_stage(6);
9853 
9854 	/* Wait reasonable amount of time for other CPU to exit VMX operation */
9855 	delay(INIT_SIGNAL_TEST_DELAY);
9856 	report(vmx_get_test_stage() == 7,
9857 	       "INIT signal consumed on VMX_INIT exit");
9858 	/* No point to continue if we failed at this point */
9859 	if (vmx_get_test_stage() != 7)
9860 		return;
9861 
9862 	/* Signal other CPU to enter VMX operation */
9863 	vmx_set_test_stage(8);
9864 	/* Wait for other CPU to enter VMX operation */
9865 	while (vmx_get_test_stage() != 9)
9866 		;
9867 
9868 	/* Send INIT signal to other CPU */
9869 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9870 				   id_map[1]);
9871 	/* Signal other CPU we have sent INIT signal */
9872 	vmx_set_test_stage(10);
9873 
9874 	/*
9875 	 * Wait reasonable amount of time for other CPU
9876 	 * to exit VMX operation and process INIT signal
9877 	 */
9878 	delay(INIT_SIGNAL_TEST_DELAY);
9879 	report(!init_signal_test_thread_continued,
9880 	       "INIT signal processed after exit VMX operation");
9881 
9882 	/*
9883 	 * TODO: Send SIPI to other CPU to sipi_entry (See x86/cstart64.S)
9884 	 * to re-init it to kvm-unit-tests standard environment.
9885 	 * Somehow (?) verify that SIPI was indeed received.
9886 	 */
9887 }
9888 
9889 #define SIPI_SIGNAL_TEST_DELAY	100000000ULL
9890 
9891 static void vmx_sipi_test_guest(void)
9892 {
9893 	if (apic_id() == 0) {
9894 		/* wait AP enter guest with activity=WAIT_SIPI */
9895 		while (vmx_get_test_stage() != 1)
9896 			;
9897 		delay(SIPI_SIGNAL_TEST_DELAY);
9898 
9899 		/* First SIPI signal */
9900 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9901 		report(1, "BSP(L2): Send first SIPI to cpu[%d]", id_map[1]);
9902 
9903 		/* wait AP enter guest */
9904 		while (vmx_get_test_stage() != 2)
9905 			;
9906 		delay(SIPI_SIGNAL_TEST_DELAY);
9907 
9908 		/* Second SIPI signal should be ignored since AP is not in WAIT_SIPI state */
9909 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9910 		report(1, "BSP(L2): Send second SIPI to cpu[%d]", id_map[1]);
9911 
9912 		/* Delay a while to check whether second SIPI would cause VMExit */
9913 		delay(SIPI_SIGNAL_TEST_DELAY);
9914 
9915 		/* Test is done, notify AP to exit test */
9916 		vmx_set_test_stage(3);
9917 
9918 		/* wait AP exit non-root mode */
9919 		while (vmx_get_test_stage() != 5)
9920 			;
9921 	} else {
9922 		/* wait BSP notify test is done */
9923 		while (vmx_get_test_stage() != 3)
9924 			;
9925 
9926 		/* AP exit guest */
9927 		vmx_set_test_stage(4);
9928 	}
9929 }
9930 
9931 static void sipi_test_ap_thread(void *data)
9932 {
9933 	struct vmcs *ap_vmcs;
9934 	u64 *ap_vmxon_region;
9935 	void *ap_stack, *ap_syscall_stack;
9936 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9937 	u64 cpu_ctrl_1 = 0;
9938 
9939 	/* Enter VMX operation (i.e. exec VMXON) */
9940 	ap_vmxon_region = alloc_page();
9941 	enable_vmx();
9942 	init_vmx(ap_vmxon_region);
9943 	_vmx_on(ap_vmxon_region);
9944 	init_vmcs(&ap_vmcs);
9945 	make_vmcs_current(ap_vmcs);
9946 
9947 	/* Set stack for AP */
9948 	ap_stack = alloc_page();
9949 	ap_syscall_stack = alloc_page();
9950 	vmcs_write(GUEST_RSP, (u64)(ap_stack + PAGE_SIZE - 1));
9951 	vmcs_write(GUEST_SYSENTER_ESP, (u64)(ap_syscall_stack + PAGE_SIZE - 1));
9952 
9953 	/* passthrough lapic to L2 */
9954 	disable_intercept_for_x2apic_msrs();
9955 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9956 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9957 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9958 
9959 	/* Set guest activity state to wait-for-SIPI state */
9960 	vmcs_write(GUEST_ACTV_STATE, ACTV_WAIT_SIPI);
9961 
9962 	vmx_set_test_stage(1);
9963 
9964 	/* AP enter guest */
9965 	enter_guest();
9966 
9967 	if (vmcs_read(EXI_REASON) == VMX_SIPI) {
9968 		report(1, "AP: Handle SIPI VMExit");
9969 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
9970 		vmx_set_test_stage(2);
9971 	} else {
9972 		report(0, "AP: Unexpected VMExit, reason=%ld", vmcs_read(EXI_REASON));
9973 		vmx_off();
9974 		return;
9975 	}
9976 
9977 	/* AP enter guest */
9978 	enter_guest();
9979 
9980 	report(vmcs_read(EXI_REASON) != VMX_SIPI,
9981 		"AP: should no SIPI VMExit since activity is not in WAIT_SIPI state");
9982 
9983 	/* notify BSP that AP is already exit from non-root mode */
9984 	vmx_set_test_stage(5);
9985 
9986 	/* Leave VMX operation */
9987 	vmx_off();
9988 }
9989 
9990 static void vmx_sipi_signal_test(void)
9991 {
9992 	if (!(rdmsr(MSR_IA32_VMX_MISC) & MSR_IA32_VMX_MISC_ACTIVITY_WAIT_SIPI)) {
9993 		printf("\tACTIVITY_WAIT_SIPI state is not supported.\n");
9994 		return;
9995 	}
9996 
9997 	if (cpu_count() < 2) {
9998 		report_skip(__func__);
9999 		return;
10000 	}
10001 
10002 	u64 cpu_ctrl_0 = CPU_SECONDARY;
10003 	u64 cpu_ctrl_1 = 0;
10004 
10005 	/* passthrough lapic to L2 */
10006 	disable_intercept_for_x2apic_msrs();
10007 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
10008 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
10009 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
10010 
10011 	test_set_guest(vmx_sipi_test_guest);
10012 
10013 	/* update CR3 on AP */
10014 	on_cpu(1, update_cr3, (void *)read_cr3());
10015 
10016 	/* start AP */
10017 	on_cpu_async(1, sipi_test_ap_thread, NULL);
10018 
10019 	vmx_set_test_stage(0);
10020 
10021 	/* BSP enter guest */
10022 	enter_guest();
10023 }
10024 
10025 
10026 enum vmcs_access {
10027 	ACCESS_VMREAD,
10028 	ACCESS_VMWRITE,
10029 	ACCESS_NONE,
10030 };
10031 
10032 struct vmcs_shadow_test_common {
10033 	enum vmcs_access op;
10034 	enum Reason reason;
10035 	u64 field;
10036 	u64 value;
10037 	u64 flags;
10038 	u64 time;
10039 } l1_l2_common;
10040 
10041 static inline u64 vmread_flags(u64 field, u64 *val)
10042 {
10043 	u64 flags;
10044 
10045 	asm volatile ("vmread %2, %1; pushf; pop %0"
10046 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
10047 	return flags & X86_EFLAGS_ALU;
10048 }
10049 
10050 static inline u64 vmwrite_flags(u64 field, u64 val)
10051 {
10052 	u64 flags;
10053 
10054 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
10055 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
10056 	return flags & X86_EFLAGS_ALU;
10057 }
10058 
10059 static void vmx_vmcs_shadow_test_guest(void)
10060 {
10061 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10062 	u64 start;
10063 
10064 	while (c->op != ACCESS_NONE) {
10065 		start = rdtsc();
10066 		switch (c->op) {
10067 		default:
10068 			c->flags = -1ull;
10069 			break;
10070 		case ACCESS_VMREAD:
10071 			c->flags = vmread_flags(c->field, &c->value);
10072 			break;
10073 		case ACCESS_VMWRITE:
10074 			c->flags = vmwrite_flags(c->field, 0);
10075 			break;
10076 		}
10077 		c->time = rdtsc() - start;
10078 		vmcall();
10079 	}
10080 }
10081 
10082 static u64 vmread_from_shadow(u64 field)
10083 {
10084 	struct vmcs *primary;
10085 	struct vmcs *shadow;
10086 	u64 value;
10087 
10088 	TEST_ASSERT(!vmcs_save(&primary));
10089 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10090 	TEST_ASSERT(!make_vmcs_current(shadow));
10091 	value = vmcs_read(field);
10092 	TEST_ASSERT(!make_vmcs_current(primary));
10093 	return value;
10094 }
10095 
10096 static u64 vmwrite_to_shadow(u64 field, u64 value)
10097 {
10098 	struct vmcs *primary;
10099 	struct vmcs *shadow;
10100 
10101 	TEST_ASSERT(!vmcs_save(&primary));
10102 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10103 	TEST_ASSERT(!make_vmcs_current(shadow));
10104 	vmcs_write(field, value);
10105 	value = vmcs_read(field);
10106 	TEST_ASSERT(!make_vmcs_current(primary));
10107 	return value;
10108 }
10109 
10110 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
10111 {
10112 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10113 
10114 	c->op = access;
10115 	vmcs_write(VMX_INST_ERROR, 0);
10116 	enter_guest();
10117 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
10118 	if (c->reason != VMX_VMCALL) {
10119 		skip_exit_insn();
10120 		enter_guest();
10121 	}
10122 	skip_exit_vmcall();
10123 }
10124 
10125 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
10126 {
10127 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10128 	struct vmcs *shadow;
10129 	u64 value;
10130 	uintptr_t flags[2];
10131 	bool good_shadow;
10132 	u32 vmx_inst_error;
10133 
10134 	report_prefix_pushf("field %lx", field);
10135 	c->field = field;
10136 
10137 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10138 	if (shadow != (struct vmcs *)-1ull) {
10139 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
10140 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
10141 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
10142 	} else {
10143 		/*
10144 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
10145 		 * shadowed-fields should fail with setting RFLAGS.CF.
10146 		 */
10147 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
10148 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
10149 		good_shadow = false;
10150 	}
10151 
10152 	/* Intercept both VMREAD and VMWRITE. */
10153 	report_prefix_push("no VMREAD/VMWRITE permission");
10154 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
10155 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
10156 		set_bit(field, bitmap[ACCESS_VMREAD]);
10157 		set_bit(field, bitmap[ACCESS_VMWRITE]);
10158 	}
10159 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10160 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10161 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10162 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10163 	report_prefix_pop();
10164 
10165 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
10166 		goto out;
10167 
10168 	/* Permit shadowed VMREAD. */
10169 	report_prefix_push("VMREAD permission only");
10170 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10171 	set_bit(field, bitmap[ACCESS_VMWRITE]);
10172 	if (good_shadow)
10173 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10174 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10175 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10176 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10177 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10178 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10179 	       c->time);
10180 	report(c->flags == flags[ACCESS_VMREAD],
10181 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10182 	       c->flags, flags[ACCESS_VMREAD]);
10183 	if (good_shadow)
10184 		report(c->value == value,
10185 		       "value read from shadow (%lx) is as expected (%lx)",
10186 		       c->value, value);
10187 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10188 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10189 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10190 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10191 	report_prefix_pop();
10192 
10193 	/* Permit shadowed VMWRITE. */
10194 	report_prefix_push("VMWRITE permission only");
10195 	set_bit(field, bitmap[ACCESS_VMREAD]);
10196 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10197 	if (good_shadow)
10198 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10199 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10200 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10201 	report(c->reason == VMX_VMCALL,
10202 		"shadowed for VMWRITE (in %ld cycles)",
10203 		c->time);
10204 	report(c->flags == flags[ACCESS_VMREAD],
10205 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10206 	       c->flags, flags[ACCESS_VMREAD]);
10207 	if (good_shadow) {
10208 		value = vmread_from_shadow(field);
10209 		report(value == 0,
10210 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10211 		       0ul);
10212 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10213 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10214 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10215 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10216 	}
10217 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10218 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10219 	report_prefix_pop();
10220 
10221 	/* Permit shadowed VMREAD and VMWRITE. */
10222 	report_prefix_push("VMREAD and VMWRITE permission");
10223 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10224 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10225 	if (good_shadow)
10226 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10227 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10228 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10229 	report(c->reason == VMX_VMCALL,
10230 		"shadowed for VMWRITE (in %ld cycles)",
10231 		c->time);
10232 	report(c->flags == flags[ACCESS_VMREAD],
10233 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10234 	       c->flags, flags[ACCESS_VMREAD]);
10235 	if (good_shadow) {
10236 		value = vmread_from_shadow(field);
10237 		report(value == 0,
10238 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10239 		       0ul);
10240 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10241 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10242 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10243 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10244 	}
10245 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10246 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10247 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10248 	       c->time);
10249 	report(c->flags == flags[ACCESS_VMREAD],
10250 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10251 	       c->flags, flags[ACCESS_VMREAD]);
10252 	if (good_shadow)
10253 		report(c->value == 0,
10254 		       "value read from shadow (%lx) is as expected (%lx)",
10255 		       c->value, 0ul);
10256 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10257 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10258 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10259 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10260 	report_prefix_pop();
10261 
10262 out:
10263 	report_prefix_pop();
10264 }
10265 
10266 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
10267 {
10268 	unsigned base;
10269 	unsigned index;
10270 	unsigned bit;
10271 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
10272 
10273 	/* Run test on all possible valid VMCS fields */
10274 	for (base = 0;
10275 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
10276 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
10277 		for (index = 0; index <= highest_index; index++)
10278 			vmcs_shadow_test_field(bitmap, base + index);
10279 
10280 	/*
10281 	 * Run tests on some invalid VMCS fields
10282 	 * (Have reserved bit set).
10283 	 */
10284 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
10285 		vmcs_shadow_test_field(bitmap, (1ull << bit));
10286 }
10287 
10288 static void vmx_vmcs_shadow_test(void)
10289 {
10290 	u8 *bitmap[2];
10291 	struct vmcs *shadow;
10292 
10293 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
10294 		printf("\t'Activate secondary controls' not supported.\n");
10295 		return;
10296 	}
10297 
10298 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
10299 		printf("\t'VMCS shadowing' not supported.\n");
10300 		return;
10301 	}
10302 
10303 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
10304 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
10305 		printf("\tVMWRITE can't modify VM-exit information fields.\n");
10306 		return;
10307 	}
10308 
10309 	test_set_guest(vmx_vmcs_shadow_test_guest);
10310 
10311 	bitmap[ACCESS_VMREAD] = alloc_page();
10312 	bitmap[ACCESS_VMWRITE] = alloc_page();
10313 
10314 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
10315 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
10316 
10317 	shadow = alloc_page();
10318 	shadow->hdr.revision_id = basic.revision;
10319 	shadow->hdr.shadow_vmcs = 1;
10320 	TEST_ASSERT(!vmcs_clear(shadow));
10321 
10322 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
10323 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
10324 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
10325 
10326 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
10327 	report_prefix_push("valid link pointer");
10328 	vmx_vmcs_shadow_test_body(bitmap);
10329 	report_prefix_pop();
10330 
10331 	vmcs_write(VMCS_LINK_PTR, -1ull);
10332 	report_prefix_push("invalid link pointer");
10333 	vmx_vmcs_shadow_test_body(bitmap);
10334 	report_prefix_pop();
10335 
10336 	l1_l2_common.op = ACCESS_NONE;
10337 	enter_guest();
10338 }
10339 
10340 /*
10341  * This test monitors the difference between a guest RDTSC instruction
10342  * and the IA32_TIME_STAMP_COUNTER MSR value stored in the VMCS12
10343  * VM-exit MSR-store list when taking a VM-exit on the instruction
10344  * following RDTSC.
10345  */
10346 #define RDTSC_DIFF_ITERS 100000
10347 #define RDTSC_DIFF_FAILS 100
10348 #define HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD 750
10349 
10350 /*
10351  * Set 'use TSC offsetting' and set the guest offset to the
10352  * inverse of the host's current TSC value, so that the guest starts running
10353  * with an effective TSC value of 0.
10354  */
10355 static void reset_guest_tsc_to_zero(void)
10356 {
10357 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
10358 	vmcs_write(TSC_OFFSET, -rdtsc());
10359 }
10360 
10361 static void rdtsc_vmexit_diff_test_guest(void)
10362 {
10363 	int i;
10364 
10365 	for (i = 0; i < RDTSC_DIFF_ITERS; i++)
10366 		/* Ensure rdtsc is the last instruction before the vmcall. */
10367 		asm volatile("rdtsc; vmcall" : : : "eax", "edx");
10368 }
10369 
10370 /*
10371  * This function only considers the "use TSC offsetting" VM-execution
10372  * control.  It does not handle "use TSC scaling" (because the latter
10373  * isn't available to the host today.)
10374  */
10375 static unsigned long long host_time_to_guest_time(unsigned long long t)
10376 {
10377 	TEST_ASSERT(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
10378 		    !(vmcs_read(CPU_EXEC_CTRL1) & CPU_USE_TSC_SCALING));
10379 
10380 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_USE_TSC_OFFSET)
10381 		t += vmcs_read(TSC_OFFSET);
10382 
10383 	return t;
10384 }
10385 
10386 static unsigned long long rdtsc_vmexit_diff_test_iteration(void)
10387 {
10388 	unsigned long long guest_tsc, host_to_guest_tsc;
10389 
10390 	enter_guest();
10391 	skip_exit_vmcall();
10392 	guest_tsc = (u32) regs.rax + (regs.rdx << 32);
10393 	host_to_guest_tsc = host_time_to_guest_time(exit_msr_store[0].value);
10394 
10395 	return host_to_guest_tsc - guest_tsc;
10396 }
10397 
10398 static void rdtsc_vmexit_diff_test(void)
10399 {
10400 	unsigned long long delta;
10401 	int fail = 0;
10402 	int i;
10403 
10404 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET))
10405 		test_skip("CPU doesn't support the 'use TSC offsetting' processor-based VM-execution control.\n");
10406 
10407 	test_set_guest(rdtsc_vmexit_diff_test_guest);
10408 
10409 	reset_guest_tsc_to_zero();
10410 
10411 	/*
10412 	 * Set up the VMCS12 VM-exit MSR-store list to store just one
10413 	 * MSR: IA32_TIME_STAMP_COUNTER. Note that the value stored is
10414 	 * in the host time domain (i.e., it is not adjusted according
10415 	 * to the TSC multiplier and TSC offset fields in the VMCS12,
10416 	 * as a guest RDTSC would be.)
10417 	 */
10418 	exit_msr_store = alloc_page();
10419 	exit_msr_store[0].index = MSR_IA32_TSC;
10420 	vmcs_write(EXI_MSR_ST_CNT, 1);
10421 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(exit_msr_store));
10422 
10423 	for (i = 0; i < RDTSC_DIFF_ITERS && fail < RDTSC_DIFF_FAILS; i++) {
10424 		delta = rdtsc_vmexit_diff_test_iteration();
10425 		if (delta >= HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD)
10426 			fail++;
10427 	}
10428 
10429 	enter_guest();
10430 
10431 	report(fail < RDTSC_DIFF_FAILS,
10432 	       "RDTSC to VM-exit delta too high in %d of %d iterations, last = %llu",
10433 	       fail, i, delta);
10434 }
10435 
10436 static int invalid_msr_init(struct vmcs *vmcs)
10437 {
10438 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
10439 		printf("\tPreemption timer is not supported\n");
10440 		return VMX_TEST_EXIT;
10441 	}
10442 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
10443 	preempt_val = 10000000;
10444 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
10445 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
10446 
10447 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
10448 		printf("\tSave preemption value is not supported\n");
10449 
10450 	vmcs_write(ENT_MSR_LD_CNT, 1);
10451 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
10452 
10453 	return VMX_TEST_START;
10454 }
10455 
10456 
10457 static void invalid_msr_main(void)
10458 {
10459 	report(0, "Invalid MSR load");
10460 }
10461 
10462 static int invalid_msr_exit_handler(union exit_reason exit_reason)
10463 {
10464 	report(0, "Invalid MSR load");
10465 	print_vmexit_info(exit_reason);
10466 	return VMX_TEST_EXIT;
10467 }
10468 
10469 static int invalid_msr_entry_failure(struct vmentry_result *result)
10470 {
10471 	report(result->exit_reason.failed_vmentry &&
10472 	       result->exit_reason.basic == VMX_FAIL_MSR, "Invalid MSR load");
10473 	return VMX_TEST_VMEXIT;
10474 }
10475 
10476 /*
10477  * The max number of MSRs in an atomic switch MSR list is:
10478  * (111B + 1) * 512 = 4096
10479  *
10480  * Each list entry consumes:
10481  * 4-byte MSR index + 4 bytes reserved + 8-byte data = 16 bytes
10482  *
10483  * Allocate 128 kB to cover max_msr_list_size (i.e., 64 kB) and then some.
10484  */
10485 static const u32 msr_list_page_order = 5;
10486 
10487 static void atomic_switch_msr_limit_test_guest(void)
10488 {
10489 	vmcall();
10490 }
10491 
10492 static void populate_msr_list(struct vmx_msr_entry *msr_list,
10493 			      size_t byte_capacity, int count)
10494 {
10495 	int i;
10496 
10497 	for (i = 0; i < count; i++) {
10498 		msr_list[i].index = MSR_IA32_TSC;
10499 		msr_list[i].reserved = 0;
10500 		msr_list[i].value = 0x1234567890abcdef;
10501 	}
10502 
10503 	memset(msr_list + count, 0xff,
10504 	       byte_capacity - count * sizeof(*msr_list));
10505 }
10506 
10507 static int max_msr_list_size(void)
10508 {
10509 	u32 vmx_misc = rdmsr(MSR_IA32_VMX_MISC);
10510 	u32 factor = ((vmx_misc & GENMASK(27, 25)) >> 25) + 1;
10511 
10512 	return factor * 512;
10513 }
10514 
10515 static void atomic_switch_msrs_test(int count)
10516 {
10517 	struct vmx_msr_entry *vm_enter_load;
10518         struct vmx_msr_entry *vm_exit_load;
10519         struct vmx_msr_entry *vm_exit_store;
10520 	int max_allowed = max_msr_list_size();
10521 	int byte_capacity = 1ul << (msr_list_page_order + PAGE_SHIFT);
10522 	/* Exceeding the max MSR list size at exit trigers KVM to abort. */
10523 	int exit_count = count > max_allowed ? max_allowed : count;
10524 	int cleanup_count = count > max_allowed ? 2 : 1;
10525 	int i;
10526 
10527 	/*
10528 	 * Check for the IA32_TSC MSR,
10529 	 * available with the "TSC flag" and used to populate the MSR lists.
10530 	 */
10531 	if (!(cpuid(1).d & (1 << 4))) {
10532 		report_skip(__func__);
10533 		return;
10534 	}
10535 
10536 	/* Set L2 guest. */
10537 	test_set_guest(atomic_switch_msr_limit_test_guest);
10538 
10539 	/* Setup atomic MSR switch lists. */
10540 	vm_enter_load = alloc_pages(msr_list_page_order);
10541 	vm_exit_load = alloc_pages(msr_list_page_order);
10542 	vm_exit_store = alloc_pages(msr_list_page_order);
10543 
10544 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)vm_enter_load);
10545 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)vm_exit_load);
10546 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)vm_exit_store);
10547 
10548 	/*
10549 	 * VM-Enter should succeed up to the max number of MSRs per list, and
10550 	 * should not consume junk beyond the last entry.
10551 	 */
10552 	populate_msr_list(vm_enter_load, byte_capacity, count);
10553 	populate_msr_list(vm_exit_load, byte_capacity, exit_count);
10554 	populate_msr_list(vm_exit_store, byte_capacity, exit_count);
10555 
10556 	vmcs_write(ENT_MSR_LD_CNT, count);
10557 	vmcs_write(EXI_MSR_LD_CNT, exit_count);
10558 	vmcs_write(EXI_MSR_ST_CNT, exit_count);
10559 
10560 	if (count <= max_allowed) {
10561 		enter_guest();
10562 		assert_exit_reason(VMX_VMCALL);
10563 		skip_exit_vmcall();
10564 	} else {
10565 		u32 exit_qual;
10566 
10567 		test_guest_state("Invalid MSR Load Count", true, count,
10568 				 "ENT_MSR_LD_CNT");
10569 
10570 		exit_qual = vmcs_read(EXI_QUALIFICATION);
10571 		report(exit_qual == max_allowed + 1, "exit_qual, %u, is %u.",
10572 		       exit_qual, max_allowed + 1);
10573 	}
10574 
10575 	/* Cleanup. */
10576 	vmcs_write(ENT_MSR_LD_CNT, 0);
10577 	vmcs_write(EXI_MSR_LD_CNT, 0);
10578 	vmcs_write(EXI_MSR_ST_CNT, 0);
10579 	for (i = 0; i < cleanup_count; i++) {
10580 		enter_guest();
10581 		skip_exit_vmcall();
10582 	}
10583 	free_pages_by_order(vm_enter_load, msr_list_page_order);
10584 	free_pages_by_order(vm_exit_load, msr_list_page_order);
10585 	free_pages_by_order(vm_exit_store, msr_list_page_order);
10586 }
10587 
10588 static void atomic_switch_max_msrs_test(void)
10589 {
10590 	atomic_switch_msrs_test(max_msr_list_size());
10591 }
10592 
10593 static void atomic_switch_overflow_msrs_test(void)
10594 {
10595 	if (test_device_enabled())
10596 		atomic_switch_msrs_test(max_msr_list_size() + 1);
10597 	else
10598 		test_skip("Test is only supported on KVM");
10599 }
10600 
10601 #define TEST(name) { #name, .v2 = name }
10602 
10603 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
10604 struct vmx_test vmx_tests[] = {
10605 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
10606 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
10607 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
10608 		preemption_timer_exit_handler, NULL, {0} },
10609 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
10610 		test_ctrl_pat_exit_handler, NULL, {0} },
10611 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
10612 		test_ctrl_efer_exit_handler, NULL, {0} },
10613 	{ "CR shadowing", NULL, cr_shadowing_main,
10614 		cr_shadowing_exit_handler, NULL, {0} },
10615 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
10616 		NULL, {0} },
10617 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
10618 		insn_intercept_exit_handler, NULL, {0} },
10619 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
10620 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
10621 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
10622 	{ "VPID", vpid_init, vpid_main, vpid_exit_handler, NULL, {0} },
10623 	{ "interrupt", interrupt_init, interrupt_main,
10624 		interrupt_exit_handler, NULL, {0} },
10625 	{ "nmi_hlt", nmi_hlt_init, nmi_hlt_main,
10626 		nmi_hlt_exit_handler, NULL, {0} },
10627 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
10628 		NULL, {0} },
10629 	{ "MSR switch", msr_switch_init, msr_switch_main,
10630 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
10631 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
10632 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
10633 		disable_rdtscp_exit_handler, NULL, {0} },
10634 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
10635 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
10636 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
10637 		exit_monitor_from_l2_handler, NULL, {0} },
10638 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
10639 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
10640 	/* Basic V2 tests. */
10641 	TEST(v2_null_test),
10642 	TEST(v2_multiple_entries_test),
10643 	TEST(fixture_test_case1),
10644 	TEST(fixture_test_case2),
10645 	/* Opcode tests. */
10646 	TEST(invvpid_test_v2),
10647 	/* VM-entry tests */
10648 	TEST(vmx_controls_test),
10649 	TEST(vmx_host_state_area_test),
10650 	TEST(vmx_guest_state_area_test),
10651 	TEST(vmentry_movss_shadow_test),
10652 	TEST(vmentry_unrestricted_guest_test),
10653 	/* APICv tests */
10654 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
10655 	TEST(vmx_hlt_with_rvi_test),
10656 	TEST(apic_reg_virt_test),
10657 	TEST(virt_x2apic_mode_test),
10658 	/* APIC pass-through tests */
10659 	TEST(vmx_apic_passthrough_test),
10660 	TEST(vmx_apic_passthrough_thread_test),
10661 	TEST(vmx_apic_passthrough_tpr_threshold_test),
10662 	TEST(vmx_init_signal_test),
10663 	TEST(vmx_sipi_signal_test),
10664 	/* VMCS Shadowing tests */
10665 	TEST(vmx_vmcs_shadow_test),
10666 	/* Regression tests */
10667 	TEST(vmx_cr_load_test),
10668 	TEST(vmx_cr4_osxsave_test),
10669 	TEST(vmx_nm_test),
10670 	TEST(vmx_db_test),
10671 	TEST(vmx_nmi_window_test),
10672 	TEST(vmx_intr_window_test),
10673 	TEST(vmx_pending_event_test),
10674 	TEST(vmx_pending_event_hlt_test),
10675 	TEST(vmx_store_tsc_test),
10676 	TEST(vmx_preemption_timer_zero_test),
10677 	TEST(vmx_preemption_timer_tf_test),
10678 	TEST(vmx_preemption_timer_expiry_test),
10679 	/* EPT access tests. */
10680 	TEST(ept_access_test_not_present),
10681 	TEST(ept_access_test_read_only),
10682 	TEST(ept_access_test_write_only),
10683 	TEST(ept_access_test_read_write),
10684 	TEST(ept_access_test_execute_only),
10685 	TEST(ept_access_test_read_execute),
10686 	TEST(ept_access_test_write_execute),
10687 	TEST(ept_access_test_read_write_execute),
10688 	TEST(ept_access_test_reserved_bits),
10689 	TEST(ept_access_test_ignored_bits),
10690 	TEST(ept_access_test_paddr_not_present_ad_disabled),
10691 	TEST(ept_access_test_paddr_not_present_ad_enabled),
10692 	TEST(ept_access_test_paddr_read_only_ad_disabled),
10693 	TEST(ept_access_test_paddr_read_only_ad_enabled),
10694 	TEST(ept_access_test_paddr_read_write),
10695 	TEST(ept_access_test_paddr_read_write_execute),
10696 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
10697 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
10698 	TEST(ept_access_test_paddr_not_present_page_fault),
10699 	TEST(ept_access_test_force_2m_page),
10700 	/* Atomic MSR switch tests. */
10701 	TEST(atomic_switch_max_msrs_test),
10702 	TEST(atomic_switch_overflow_msrs_test),
10703 	TEST(rdtsc_vmexit_diff_test),
10704 	TEST(vmx_mtf_test),
10705 	TEST(vmx_mtf_pdpte_test),
10706 	{ NULL, NULL, NULL, NULL, NULL, {0} },
10707 };
10708