xref: /kvm-unit-tests/x86/vmx_tests.c (revision e31194e75c7650b85dcc7fd88025ad59e7c09b1e)
1 /*
2  * All test cases of nested virtualization should be in this file
3  *
4  * Author : Arthur Chunqi Li <yzt356@gmail.com>
5  */
6 
7 #include <asm/debugreg.h>
8 
9 #include "vmx.h"
10 #include "msr.h"
11 #include "processor.h"
12 #include "vm.h"
13 #include "pci.h"
14 #include "fwcfg.h"
15 #include "isr.h"
16 #include "desc.h"
17 #include "apic.h"
18 #include "types.h"
19 #include "vmalloc.h"
20 #include "alloc_page.h"
21 #include "smp.h"
22 #include "delay.h"
23 
24 #define NONCANONICAL            0xaaaaaaaaaaaaaaaaull
25 
26 #define VPID_CAP_INVVPID_TYPES_SHIFT 40
27 
28 u64 ia32_pat;
29 u64 ia32_efer;
30 void *io_bitmap_a, *io_bitmap_b;
31 u16 ioport;
32 
33 unsigned long *pml4;
34 u64 eptp;
35 void *data_page1, *data_page2;
36 
37 phys_addr_t pci_physaddr;
38 
39 void *pml_log;
40 #define PML_INDEX 512
41 
42 static inline unsigned ffs(unsigned x)
43 {
44 	int pos = -1;
45 
46 	__asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0"
47 			     : "+r"(pos) : "rm"(x) : "eax");
48 	return pos + 1;
49 }
50 
51 static inline void vmcall(void)
52 {
53 	asm volatile("vmcall");
54 }
55 
56 static void basic_guest_main(void)
57 {
58 	report("Basic VMX test", 1);
59 }
60 
61 static int basic_exit_handler(void)
62 {
63 	report("Basic VMX test", 0);
64 	print_vmexit_info();
65 	return VMX_TEST_EXIT;
66 }
67 
68 static void vmenter_main(void)
69 {
70 	u64 rax;
71 	u64 rsp, resume_rsp;
72 
73 	report("test vmlaunch", 1);
74 
75 	asm volatile(
76 		"mov %%rsp, %0\n\t"
77 		"mov %3, %%rax\n\t"
78 		"vmcall\n\t"
79 		"mov %%rax, %1\n\t"
80 		"mov %%rsp, %2\n\t"
81 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
82 		: "g"(0xABCD));
83 	report("test vmresume", (rax == 0xFFFF) && (rsp == resume_rsp));
84 }
85 
86 static int vmenter_exit_handler(void)
87 {
88 	u64 guest_rip;
89 	ulong reason;
90 
91 	guest_rip = vmcs_read(GUEST_RIP);
92 	reason = vmcs_read(EXI_REASON) & 0xff;
93 	switch (reason) {
94 	case VMX_VMCALL:
95 		if (regs.rax != 0xABCD) {
96 			report("test vmresume", 0);
97 			return VMX_TEST_VMEXIT;
98 		}
99 		regs.rax = 0xFFFF;
100 		vmcs_write(GUEST_RIP, guest_rip + 3);
101 		return VMX_TEST_RESUME;
102 	default:
103 		report("test vmresume", 0);
104 		print_vmexit_info();
105 	}
106 	return VMX_TEST_VMEXIT;
107 }
108 
109 u32 preempt_scale;
110 volatile unsigned long long tsc_val;
111 volatile u32 preempt_val;
112 u64 saved_rip;
113 
114 static int preemption_timer_init(struct vmcs *vmcs)
115 {
116 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
117 		printf("\tPreemption timer is not supported\n");
118 		return VMX_TEST_EXIT;
119 	}
120 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
121 	preempt_val = 10000000;
122 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
123 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
124 
125 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
126 		printf("\tSave preemption value is not supported\n");
127 
128 	return VMX_TEST_START;
129 }
130 
131 static void preemption_timer_main(void)
132 {
133 	tsc_val = rdtsc();
134 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
135 		vmx_set_test_stage(0);
136 		vmcall();
137 		if (vmx_get_test_stage() == 1)
138 			vmcall();
139 	}
140 	vmx_set_test_stage(1);
141 	while (vmx_get_test_stage() == 1) {
142 		if (((rdtsc() - tsc_val) >> preempt_scale)
143 				> 10 * preempt_val) {
144 			vmx_set_test_stage(2);
145 			vmcall();
146 		}
147 	}
148 	tsc_val = rdtsc();
149 	asm volatile ("hlt");
150 	vmcall();
151 	vmx_set_test_stage(5);
152 	vmcall();
153 }
154 
155 static int preemption_timer_exit_handler(void)
156 {
157 	bool guest_halted;
158 	u64 guest_rip;
159 	ulong reason;
160 	u32 insn_len;
161 	u32 ctrl_exit;
162 
163 	guest_rip = vmcs_read(GUEST_RIP);
164 	reason = vmcs_read(EXI_REASON) & 0xff;
165 	insn_len = vmcs_read(EXI_INST_LEN);
166 	switch (reason) {
167 	case VMX_PREEMPT:
168 		switch (vmx_get_test_stage()) {
169 		case 1:
170 		case 2:
171 			report("busy-wait for preemption timer",
172 			       ((rdtsc() - tsc_val) >> preempt_scale) >=
173 			       preempt_val);
174 			vmx_set_test_stage(3);
175 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
176 			return VMX_TEST_RESUME;
177 		case 3:
178 			guest_halted =
179 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
180 			report("preemption timer during hlt",
181 			       ((rdtsc() - tsc_val) >> preempt_scale) >=
182 			       preempt_val && guest_halted);
183 			vmx_set_test_stage(4);
184 			vmcs_write(PIN_CONTROLS,
185 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
186 			vmcs_write(EXI_CONTROLS,
187 				   vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT);
188 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
189 			return VMX_TEST_RESUME;
190 		case 4:
191 			report("preemption timer with 0 value",
192 			       saved_rip == guest_rip);
193 			break;
194 		default:
195 			report("Invalid stage.", false);
196 			print_vmexit_info();
197 			break;
198 		}
199 		break;
200 	case VMX_VMCALL:
201 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
202 		switch (vmx_get_test_stage()) {
203 		case 0:
204 			report("Keep preemption value",
205 			       vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val);
206 			vmx_set_test_stage(1);
207 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
208 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
209 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
210 			vmcs_write(EXI_CONTROLS, ctrl_exit);
211 			return VMX_TEST_RESUME;
212 		case 1:
213 			report("Save preemption value",
214 			       vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val);
215 			return VMX_TEST_RESUME;
216 		case 2:
217 			report("busy-wait for preemption timer", 0);
218 			vmx_set_test_stage(3);
219 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
220 			return VMX_TEST_RESUME;
221 		case 3:
222 			report("preemption timer during hlt", 0);
223 			vmx_set_test_stage(4);
224 			/* fall through */
225 		case 4:
226 			vmcs_write(PIN_CONTROLS,
227 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
228 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
229 			saved_rip = guest_rip + insn_len;
230 			return VMX_TEST_RESUME;
231 		case 5:
232 			report("preemption timer with 0 value (vmcall stage 5)", 0);
233 			break;
234 		default:
235 			// Should not reach here
236 			report("unexpected stage, %d", false,
237 			       vmx_get_test_stage());
238 			print_vmexit_info();
239 			return VMX_TEST_VMEXIT;
240 		}
241 		break;
242 	default:
243 		report("Unknown exit reason, %ld", false, reason);
244 		print_vmexit_info();
245 	}
246 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
247 	return VMX_TEST_VMEXIT;
248 }
249 
250 static void msr_bmp_init(void)
251 {
252 	void *msr_bitmap;
253 	u32 ctrl_cpu0;
254 
255 	msr_bitmap = alloc_page();
256 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
257 	ctrl_cpu0 |= CPU_MSR_BITMAP;
258 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
259 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
260 }
261 
262 static void *get_msr_bitmap(void)
263 {
264 	void *msr_bitmap;
265 
266 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
267 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
268 	} else {
269 		msr_bitmap = alloc_page();
270 		memset(msr_bitmap, 0xff, PAGE_SIZE);
271 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
272 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
273 	}
274 
275 	return msr_bitmap;
276 }
277 
278 static void disable_intercept_for_x2apic_msrs(void)
279 {
280 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
281 	u32 msr;
282 
283 	for (msr = APIC_BASE_MSR;
284 		 msr < (APIC_BASE_MSR+0xff);
285 		 msr += BITS_PER_LONG) {
286 		unsigned int word = msr / BITS_PER_LONG;
287 
288 		msr_bitmap[word] = 0;
289 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
290 	}
291 }
292 
293 static int test_ctrl_pat_init(struct vmcs *vmcs)
294 {
295 	u64 ctrl_ent;
296 	u64 ctrl_exi;
297 
298 	msr_bmp_init();
299 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
300 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
301 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
302 		printf("\tSave/load PAT is not supported\n");
303 		return 1;
304 	}
305 
306 	ctrl_ent = vmcs_read(ENT_CONTROLS);
307 	ctrl_exi = vmcs_read(EXI_CONTROLS);
308 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
309 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
310 	vmcs_write(ENT_CONTROLS, ctrl_ent);
311 	vmcs_write(EXI_CONTROLS, ctrl_exi);
312 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
313 	vmcs_write(GUEST_PAT, 0x0);
314 	vmcs_write(HOST_PAT, ia32_pat);
315 	return VMX_TEST_START;
316 }
317 
318 static void test_ctrl_pat_main(void)
319 {
320 	u64 guest_ia32_pat;
321 
322 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
323 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
324 		printf("\tENT_LOAD_PAT is not supported.\n");
325 	else {
326 		if (guest_ia32_pat != 0) {
327 			report("Entry load PAT", 0);
328 			return;
329 		}
330 	}
331 	wrmsr(MSR_IA32_CR_PAT, 0x6);
332 	vmcall();
333 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
334 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
335 		report("Entry load PAT", guest_ia32_pat == ia32_pat);
336 }
337 
338 static int test_ctrl_pat_exit_handler(void)
339 {
340 	u64 guest_rip;
341 	ulong reason;
342 	u64 guest_pat;
343 
344 	guest_rip = vmcs_read(GUEST_RIP);
345 	reason = vmcs_read(EXI_REASON) & 0xff;
346 	switch (reason) {
347 	case VMX_VMCALL:
348 		guest_pat = vmcs_read(GUEST_PAT);
349 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
350 			printf("\tEXI_SAVE_PAT is not supported\n");
351 			vmcs_write(GUEST_PAT, 0x6);
352 		} else {
353 			report("Exit save PAT", guest_pat == 0x6);
354 		}
355 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
356 			printf("\tEXI_LOAD_PAT is not supported\n");
357 		else
358 			report("Exit load PAT", rdmsr(MSR_IA32_CR_PAT) == ia32_pat);
359 		vmcs_write(GUEST_PAT, ia32_pat);
360 		vmcs_write(GUEST_RIP, guest_rip + 3);
361 		return VMX_TEST_RESUME;
362 	default:
363 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
364 		break;
365 	}
366 	return VMX_TEST_VMEXIT;
367 }
368 
369 static int test_ctrl_efer_init(struct vmcs *vmcs)
370 {
371 	u64 ctrl_ent;
372 	u64 ctrl_exi;
373 
374 	msr_bmp_init();
375 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
376 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
377 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
378 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
379 	ia32_efer = rdmsr(MSR_EFER);
380 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
381 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
382 	return VMX_TEST_START;
383 }
384 
385 static void test_ctrl_efer_main(void)
386 {
387 	u64 guest_ia32_efer;
388 
389 	guest_ia32_efer = rdmsr(MSR_EFER);
390 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
391 		printf("\tENT_LOAD_EFER is not supported.\n");
392 	else {
393 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
394 			report("Entry load EFER", 0);
395 			return;
396 		}
397 	}
398 	wrmsr(MSR_EFER, ia32_efer);
399 	vmcall();
400 	guest_ia32_efer = rdmsr(MSR_EFER);
401 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
402 		report("Entry load EFER", guest_ia32_efer == ia32_efer);
403 }
404 
405 static int test_ctrl_efer_exit_handler(void)
406 {
407 	u64 guest_rip;
408 	ulong reason;
409 	u64 guest_efer;
410 
411 	guest_rip = vmcs_read(GUEST_RIP);
412 	reason = vmcs_read(EXI_REASON) & 0xff;
413 	switch (reason) {
414 	case VMX_VMCALL:
415 		guest_efer = vmcs_read(GUEST_EFER);
416 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
417 			printf("\tEXI_SAVE_EFER is not supported\n");
418 			vmcs_write(GUEST_EFER, ia32_efer);
419 		} else {
420 			report("Exit save EFER", guest_efer == ia32_efer);
421 		}
422 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
423 			printf("\tEXI_LOAD_EFER is not supported\n");
424 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
425 		} else {
426 			report("Exit load EFER", rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX));
427 		}
428 		vmcs_write(GUEST_PAT, ia32_efer);
429 		vmcs_write(GUEST_RIP, guest_rip + 3);
430 		return VMX_TEST_RESUME;
431 	default:
432 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
433 		break;
434 	}
435 	return VMX_TEST_VMEXIT;
436 }
437 
438 u32 guest_cr0, guest_cr4;
439 
440 static void cr_shadowing_main(void)
441 {
442 	u32 cr0, cr4, tmp;
443 
444 	// Test read through
445 	vmx_set_test_stage(0);
446 	guest_cr0 = read_cr0();
447 	if (vmx_get_test_stage() == 1)
448 		report("Read through CR0", 0);
449 	else
450 		vmcall();
451 	vmx_set_test_stage(1);
452 	guest_cr4 = read_cr4();
453 	if (vmx_get_test_stage() == 2)
454 		report("Read through CR4", 0);
455 	else
456 		vmcall();
457 	// Test write through
458 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
459 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
460 	vmx_set_test_stage(2);
461 	write_cr0(guest_cr0);
462 	if (vmx_get_test_stage() == 3)
463 		report("Write throuth CR0", 0);
464 	else
465 		vmcall();
466 	vmx_set_test_stage(3);
467 	write_cr4(guest_cr4);
468 	if (vmx_get_test_stage() == 4)
469 		report("Write through CR4", 0);
470 	else
471 		vmcall();
472 	// Test read shadow
473 	vmx_set_test_stage(4);
474 	vmcall();
475 	cr0 = read_cr0();
476 	if (vmx_get_test_stage() != 5)
477 		report("Read shadowing CR0", cr0 == guest_cr0);
478 	vmx_set_test_stage(5);
479 	cr4 = read_cr4();
480 	if (vmx_get_test_stage() != 6)
481 		report("Read shadowing CR4", cr4 == guest_cr4);
482 	// Test write shadow (same value with shadow)
483 	vmx_set_test_stage(6);
484 	write_cr0(guest_cr0);
485 	if (vmx_get_test_stage() == 7)
486 		report("Write shadowing CR0 (same value with shadow)", 0);
487 	else
488 		vmcall();
489 	vmx_set_test_stage(7);
490 	write_cr4(guest_cr4);
491 	if (vmx_get_test_stage() == 8)
492 		report("Write shadowing CR4 (same value with shadow)", 0);
493 	else
494 		vmcall();
495 	// Test write shadow (different value)
496 	vmx_set_test_stage(8);
497 	tmp = guest_cr0 ^ X86_CR0_TS;
498 	asm volatile("mov %0, %%rsi\n\t"
499 		"mov %%rsi, %%cr0\n\t"
500 		::"m"(tmp)
501 		:"rsi", "memory", "cc");
502 	report("Write shadowing different X86_CR0_TS", vmx_get_test_stage() == 9);
503 	vmx_set_test_stage(9);
504 	tmp = guest_cr0 ^ X86_CR0_MP;
505 	asm volatile("mov %0, %%rsi\n\t"
506 		"mov %%rsi, %%cr0\n\t"
507 		::"m"(tmp)
508 		:"rsi", "memory", "cc");
509 	report("Write shadowing different X86_CR0_MP", vmx_get_test_stage() == 10);
510 	vmx_set_test_stage(10);
511 	tmp = guest_cr4 ^ X86_CR4_TSD;
512 	asm volatile("mov %0, %%rsi\n\t"
513 		"mov %%rsi, %%cr4\n\t"
514 		::"m"(tmp)
515 		:"rsi", "memory", "cc");
516 	report("Write shadowing different X86_CR4_TSD", vmx_get_test_stage() == 11);
517 	vmx_set_test_stage(11);
518 	tmp = guest_cr4 ^ X86_CR4_DE;
519 	asm volatile("mov %0, %%rsi\n\t"
520 		"mov %%rsi, %%cr4\n\t"
521 		::"m"(tmp)
522 		:"rsi", "memory", "cc");
523 	report("Write shadowing different X86_CR4_DE", vmx_get_test_stage() == 12);
524 }
525 
526 static int cr_shadowing_exit_handler(void)
527 {
528 	u64 guest_rip;
529 	ulong reason;
530 	u32 insn_len;
531 	u32 exit_qual;
532 
533 	guest_rip = vmcs_read(GUEST_RIP);
534 	reason = vmcs_read(EXI_REASON) & 0xff;
535 	insn_len = vmcs_read(EXI_INST_LEN);
536 	exit_qual = vmcs_read(EXI_QUALIFICATION);
537 	switch (reason) {
538 	case VMX_VMCALL:
539 		switch (vmx_get_test_stage()) {
540 		case 0:
541 			report("Read through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
542 			break;
543 		case 1:
544 			report("Read through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
545 			break;
546 		case 2:
547 			report("Write through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
548 			break;
549 		case 3:
550 			report("Write through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
551 			break;
552 		case 4:
553 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
554 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
555 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
556 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
557 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
558 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
559 			break;
560 		case 6:
561 			report("Write shadowing CR0 (same value)",
562 					guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)));
563 			break;
564 		case 7:
565 			report("Write shadowing CR4 (same value)",
566 					guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)));
567 			break;
568 		default:
569 			// Should not reach here
570 			report("unexpected stage, %d", false,
571 			       vmx_get_test_stage());
572 			print_vmexit_info();
573 			return VMX_TEST_VMEXIT;
574 		}
575 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
576 		return VMX_TEST_RESUME;
577 	case VMX_CR:
578 		switch (vmx_get_test_stage()) {
579 		case 4:
580 			report("Read shadowing CR0", 0);
581 			vmx_inc_test_stage();
582 			break;
583 		case 5:
584 			report("Read shadowing CR4", 0);
585 			vmx_inc_test_stage();
586 			break;
587 		case 6:
588 			report("Write shadowing CR0 (same value)", 0);
589 			vmx_inc_test_stage();
590 			break;
591 		case 7:
592 			report("Write shadowing CR4 (same value)", 0);
593 			vmx_inc_test_stage();
594 			break;
595 		case 8:
596 		case 9:
597 			// 0x600 encodes "mov %esi, %cr0"
598 			if (exit_qual == 0x600)
599 				vmx_inc_test_stage();
600 			break;
601 		case 10:
602 		case 11:
603 			// 0x604 encodes "mov %esi, %cr4"
604 			if (exit_qual == 0x604)
605 				vmx_inc_test_stage();
606 			break;
607 		default:
608 			// Should not reach here
609 			report("unexpected stage, %d", false,
610 			       vmx_get_test_stage());
611 			print_vmexit_info();
612 			return VMX_TEST_VMEXIT;
613 		}
614 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
615 		return VMX_TEST_RESUME;
616 	default:
617 		report("Unknown exit reason, %ld", false, reason);
618 		print_vmexit_info();
619 	}
620 	return VMX_TEST_VMEXIT;
621 }
622 
623 static int iobmp_init(struct vmcs *vmcs)
624 {
625 	u32 ctrl_cpu0;
626 
627 	io_bitmap_a = alloc_page();
628 	io_bitmap_b = alloc_page();
629 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
630 	ctrl_cpu0 |= CPU_IO_BITMAP;
631 	ctrl_cpu0 &= (~CPU_IO);
632 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
633 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
634 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
635 	return VMX_TEST_START;
636 }
637 
638 static void iobmp_main(void)
639 {
640 	// stage 0, test IO pass
641 	vmx_set_test_stage(0);
642 	inb(0x5000);
643 	outb(0x0, 0x5000);
644 	report("I/O bitmap - I/O pass", vmx_get_test_stage() == 0);
645 	// test IO width, in/out
646 	((u8 *)io_bitmap_a)[0] = 0xFF;
647 	vmx_set_test_stage(2);
648 	inb(0x0);
649 	report("I/O bitmap - trap in", vmx_get_test_stage() == 3);
650 	vmx_set_test_stage(3);
651 	outw(0x0, 0x0);
652 	report("I/O bitmap - trap out", vmx_get_test_stage() == 4);
653 	vmx_set_test_stage(4);
654 	inl(0x0);
655 	report("I/O bitmap - I/O width, long", vmx_get_test_stage() == 5);
656 	// test low/high IO port
657 	vmx_set_test_stage(5);
658 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
659 	inb(0x5000);
660 	report("I/O bitmap - I/O port, low part", vmx_get_test_stage() == 6);
661 	vmx_set_test_stage(6);
662 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
663 	inb(0x9000);
664 	report("I/O bitmap - I/O port, high part", vmx_get_test_stage() == 7);
665 	// test partial pass
666 	vmx_set_test_stage(7);
667 	inl(0x4FFF);
668 	report("I/O bitmap - partial pass", vmx_get_test_stage() == 8);
669 	// test overrun
670 	vmx_set_test_stage(8);
671 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
672 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
673 	inl(0xFFFF);
674 	report("I/O bitmap - overrun", vmx_get_test_stage() == 9);
675 	vmx_set_test_stage(9);
676 	vmcall();
677 	outb(0x0, 0x0);
678 	report("I/O bitmap - ignore unconditional exiting",
679 	       vmx_get_test_stage() == 9);
680 	vmx_set_test_stage(10);
681 	vmcall();
682 	outb(0x0, 0x0);
683 	report("I/O bitmap - unconditional exiting",
684 	       vmx_get_test_stage() == 11);
685 }
686 
687 static int iobmp_exit_handler(void)
688 {
689 	u64 guest_rip;
690 	ulong reason, exit_qual;
691 	u32 insn_len, ctrl_cpu0;
692 
693 	guest_rip = vmcs_read(GUEST_RIP);
694 	reason = vmcs_read(EXI_REASON) & 0xff;
695 	exit_qual = vmcs_read(EXI_QUALIFICATION);
696 	insn_len = vmcs_read(EXI_INST_LEN);
697 	switch (reason) {
698 	case VMX_IO:
699 		switch (vmx_get_test_stage()) {
700 		case 0:
701 		case 1:
702 			vmx_inc_test_stage();
703 			break;
704 		case 2:
705 			report("I/O bitmap - I/O width, byte",
706 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE);
707 			report("I/O bitmap - I/O direction, in", exit_qual & VMX_IO_IN);
708 			vmx_inc_test_stage();
709 			break;
710 		case 3:
711 			report("I/O bitmap - I/O width, word",
712 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD);
713 			report("I/O bitmap - I/O direction, out",
714 					!(exit_qual & VMX_IO_IN));
715 			vmx_inc_test_stage();
716 			break;
717 		case 4:
718 			report("I/O bitmap - I/O width, long",
719 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG);
720 			vmx_inc_test_stage();
721 			break;
722 		case 5:
723 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
724 				vmx_inc_test_stage();
725 			break;
726 		case 6:
727 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
728 				vmx_inc_test_stage();
729 			break;
730 		case 7:
731 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
732 				vmx_inc_test_stage();
733 			break;
734 		case 8:
735 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
736 				vmx_inc_test_stage();
737 			break;
738 		case 9:
739 		case 10:
740 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
741 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
742 			vmx_inc_test_stage();
743 			break;
744 		default:
745 			// Should not reach here
746 			report("unexpected stage, %d", false,
747 			       vmx_get_test_stage());
748 			print_vmexit_info();
749 			return VMX_TEST_VMEXIT;
750 		}
751 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
752 		return VMX_TEST_RESUME;
753 	case VMX_VMCALL:
754 		switch (vmx_get_test_stage()) {
755 		case 9:
756 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
757 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
758 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
759 			break;
760 		case 10:
761 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
762 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
763 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
764 			break;
765 		default:
766 			// Should not reach here
767 			report("unexpected stage, %d", false,
768 			       vmx_get_test_stage());
769 			print_vmexit_info();
770 			return VMX_TEST_VMEXIT;
771 		}
772 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
773 		return VMX_TEST_RESUME;
774 	default:
775 		printf("guest_rip = %#lx\n", guest_rip);
776 		printf("\tERROR : Undefined exit reason, reason = %ld.\n", reason);
777 		break;
778 	}
779 	return VMX_TEST_VMEXIT;
780 }
781 
782 #define INSN_CPU0		0
783 #define INSN_CPU1		1
784 #define INSN_ALWAYS_TRAP	2
785 
786 #define FIELD_EXIT_QUAL		(1 << 0)
787 #define FIELD_INSN_INFO		(1 << 1)
788 
789 asm(
790 	"insn_hlt: hlt;ret\n\t"
791 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
792 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
793 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
794 	"insn_rdtsc: rdtsc;ret\n\t"
795 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
796 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
797 #ifdef __x86_64__
798 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
799 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
800 #endif
801 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
802 	"insn_pause: pause;ret\n\t"
803 	"insn_wbinvd: wbinvd;ret\n\t"
804 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
805 	"insn_invd: invd;ret\n\t"
806 	"insn_sgdt: sgdt gdt64_desc;ret\n\t"
807 	"insn_lgdt: lgdt gdt64_desc;ret\n\t"
808 	"insn_sidt: sidt idt_descr;ret\n\t"
809 	"insn_lidt: lidt idt_descr;ret\n\t"
810 	"insn_sldt: sldt %ax;ret\n\t"
811 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
812 	"insn_str: str %ax;ret\n\t"
813 	"insn_rdrand: rdrand %rax;ret\n\t"
814 	"insn_rdseed: rdseed %rax;ret\n\t"
815 );
816 extern void insn_hlt(void);
817 extern void insn_invlpg(void);
818 extern void insn_mwait(void);
819 extern void insn_rdpmc(void);
820 extern void insn_rdtsc(void);
821 extern void insn_cr3_load(void);
822 extern void insn_cr3_store(void);
823 #ifdef __x86_64__
824 extern void insn_cr8_load(void);
825 extern void insn_cr8_store(void);
826 #endif
827 extern void insn_monitor(void);
828 extern void insn_pause(void);
829 extern void insn_wbinvd(void);
830 extern void insn_sgdt(void);
831 extern void insn_lgdt(void);
832 extern void insn_sidt(void);
833 extern void insn_lidt(void);
834 extern void insn_sldt(void);
835 extern void insn_lldt(void);
836 extern void insn_str(void);
837 extern void insn_cpuid(void);
838 extern void insn_invd(void);
839 extern void insn_rdrand(void);
840 extern void insn_rdseed(void);
841 
842 u32 cur_insn;
843 u64 cr3;
844 
845 #define X86_FEATURE_MONITOR	(1 << 3)
846 
847 typedef bool (*supported_fn)(void);
848 
849 static bool monitor_supported(void)
850 {
851 	return this_cpu_has(X86_FEATURE_MWAIT);
852 }
853 
854 struct insn_table {
855 	const char *name;
856 	u32 flag;
857 	void (*insn_func)(void);
858 	u32 type;
859 	u32 reason;
860 	ulong exit_qual;
861 	u32 insn_info;
862 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
863 	// which field need to be tested, reason is always tested
864 	u32 test_field;
865 	const supported_fn supported_fn;
866 	u8 disabled;
867 };
868 
869 /*
870  * Add more test cases of instruction intercept here. Elements in this
871  * table is:
872  *	name/control flag/insn function/type/exit reason/exit qulification/
873  *	instruction info/field to test
874  * The last field defines which fields (exit_qual and insn_info) need to be
875  * tested in exit handler. If set to 0, only "reason" is checked.
876  */
877 static struct insn_table insn_table[] = {
878 	// Flags for Primary Processor-Based VM-Execution Controls
879 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
880 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
881 		0x12345678, 0, FIELD_EXIT_QUAL},
882 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0, &monitor_supported},
883 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0},
884 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
885 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
886 		FIELD_EXIT_QUAL},
887 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
888 		FIELD_EXIT_QUAL},
889 #ifdef __x86_64__
890 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
891 		FIELD_EXIT_QUAL},
892 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
893 		FIELD_EXIT_QUAL},
894 #endif
895 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0, &monitor_supported},
896 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
897 	// Flags for Secondary Processor-Based VM-Execution Controls
898 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
899 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
900 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
901 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
902 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
903 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
904 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
905 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
906 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
907 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
908 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
909 	// Instructions always trap
910 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
911 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
912 	// Instructions never trap
913 	{NULL},
914 };
915 
916 static int insn_intercept_init(struct vmcs *vmcs)
917 {
918 	u32 ctrl_cpu, cur_insn;
919 
920 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
921 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
922 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
923 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
924 	cr3 = read_cr3();
925 
926 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
927 		if (insn_table[cur_insn].supported_fn == NULL)
928 			continue;
929 		insn_table[cur_insn].disabled = !insn_table[cur_insn].supported_fn();
930 	}
931 	return VMX_TEST_START;
932 }
933 
934 static void insn_intercept_main(void)
935 {
936 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
937 		vmx_set_test_stage(cur_insn * 2);
938 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
939 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
940 		    (insn_table[cur_insn].type == INSN_CPU1 &&
941 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
942 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
943 			       insn_table[cur_insn].type - INSN_CPU0,
944 			       insn_table[cur_insn].name);
945 			continue;
946 		}
947 
948 		if (insn_table[cur_insn].disabled) {
949 			printf("\tFeature required for %s is not supported.\n",
950 			       insn_table[cur_insn].name);
951 			continue;
952 		}
953 
954 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
955 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
956 		    (insn_table[cur_insn].type == INSN_CPU1 &&
957 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
958 			/* skip hlt, it stalls the guest and is tested below */
959 			if (insn_table[cur_insn].insn_func != insn_hlt)
960 				insn_table[cur_insn].insn_func();
961 			report("execute %s", vmx_get_test_stage() == cur_insn * 2,
962 					insn_table[cur_insn].name);
963 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
964 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
965 			       insn_table[cur_insn].type - INSN_CPU0,
966 			       insn_table[cur_insn].name);
967 
968 		vmcall();
969 
970 		insn_table[cur_insn].insn_func();
971 		report("intercept %s", vmx_get_test_stage() == cur_insn * 2 + 1,
972 				insn_table[cur_insn].name);
973 
974 		vmx_set_test_stage(cur_insn * 2 + 1);
975 		vmcall();
976 	}
977 }
978 
979 static int insn_intercept_exit_handler(void)
980 {
981 	u64 guest_rip;
982 	u32 reason;
983 	ulong exit_qual;
984 	u32 insn_len;
985 	u32 insn_info;
986 	bool pass;
987 
988 	guest_rip = vmcs_read(GUEST_RIP);
989 	reason = vmcs_read(EXI_REASON) & 0xff;
990 	exit_qual = vmcs_read(EXI_QUALIFICATION);
991 	insn_len = vmcs_read(EXI_INST_LEN);
992 	insn_info = vmcs_read(EXI_INST_INFO);
993 
994 	if (reason == VMX_VMCALL) {
995 		u32 val = 0;
996 
997 		if (insn_table[cur_insn].type == INSN_CPU0)
998 			val = vmcs_read(CPU_EXEC_CTRL0);
999 		else if (insn_table[cur_insn].type == INSN_CPU1)
1000 			val = vmcs_read(CPU_EXEC_CTRL1);
1001 
1002 		if (vmx_get_test_stage() & 1)
1003 			val &= ~insn_table[cur_insn].flag;
1004 		else
1005 			val |= insn_table[cur_insn].flag;
1006 
1007 		if (insn_table[cur_insn].type == INSN_CPU0)
1008 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
1009 		else if (insn_table[cur_insn].type == INSN_CPU1)
1010 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
1011 	} else {
1012 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
1013 			insn_table[cur_insn].reason == reason;
1014 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
1015 		    insn_table[cur_insn].exit_qual != exit_qual)
1016 			pass = false;
1017 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
1018 		    insn_table[cur_insn].insn_info != insn_info)
1019 			pass = false;
1020 		if (pass)
1021 			vmx_inc_test_stage();
1022 	}
1023 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1024 	return VMX_TEST_RESUME;
1025 }
1026 
1027 /**
1028  * __setup_ept - Setup the VMCS fields to enable Extended Page Tables (EPT)
1029  * @hpa:	Host physical address of the top-level, a.k.a. root, EPT table
1030  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1031  *
1032  * Returns 0 on success, 1 on failure.
1033  *
1034  * Note that @hpa doesn't need to point at actual memory if VM-Launch is
1035  * expected to fail, e.g. setup_dummy_ept() arbitrarily passes '0' to satisfy
1036  * the various EPTP consistency checks, but doesn't ensure backing for HPA '0'.
1037  */
1038 static int __setup_ept(u64 hpa, bool enable_ad)
1039 {
1040 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1041 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1042 		printf("\tEPT is not supported");
1043 		return 1;
1044 	}
1045 	if (!(ept_vpid.val & EPT_CAP_WB)) {
1046 		printf("WB memtype for EPT walks not supported\n");
1047 		return 1;
1048 	}
1049 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
1050 		printf("\tPWL4 is not supported\n");
1051 		return 1;
1052 	}
1053 
1054 	eptp = EPT_MEM_TYPE_WB;
1055 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1056 	eptp |= hpa;
1057 	if (enable_ad)
1058 		eptp |= EPTP_AD_FLAG;
1059 
1060 	vmcs_write(EPTP, eptp);
1061 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY);
1062 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT);
1063 
1064 	return 0;
1065 }
1066 
1067 /**
1068  * setup_ept - Enable Extended Page Tables (EPT) and setup an identity map
1069  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1070  *
1071  * Returns 0 on success, 1 on failure.
1072  *
1073  * This is the "real" function for setting up EPT tables, i.e. use this for
1074  * tests that need to run code in the guest with EPT enabled.
1075  */
1076 static int setup_ept(bool enable_ad)
1077 {
1078 	unsigned long end_of_memory;
1079 
1080 	pml4 = alloc_page();
1081 
1082 	if (__setup_ept(virt_to_phys(pml4), enable_ad))
1083 		return 1;
1084 
1085 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1086 	if (end_of_memory < (1ul << 32))
1087 		end_of_memory = (1ul << 32);
1088 	/* Cannot use large EPT pages if we need to track EPT
1089 	 * accessed/dirty bits at 4K granularity.
1090 	 */
1091 	setup_ept_range(pml4, 0, end_of_memory, 0,
1092 			!enable_ad && ept_2m_supported(),
1093 			EPT_WA | EPT_RA | EPT_EA);
1094 	return 0;
1095 }
1096 
1097 /**
1098  * setup_dummy_ept - Enable Extended Page Tables (EPT) with a dummy root HPA
1099  *
1100  * Setup EPT using a semi-arbitrary dummy root HPA.  This function is intended
1101  * for use by tests that need EPT enabled to verify dependent VMCS controls
1102  * but never expect to fully enter the guest, i.e. don't need setup the actual
1103  * EPT tables.
1104  */
1105 static void setup_dummy_ept(void)
1106 {
1107 	if (__setup_ept(0, false))
1108 		report_abort("EPT setup unexpectedly failed");
1109 }
1110 
1111 static int enable_unrestricted_guest(void)
1112 {
1113 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1114 	    !(ctrl_cpu_rev[1].clr & CPU_URG) ||
1115 	    !(ctrl_cpu_rev[1].clr & CPU_EPT))
1116 		return 1;
1117 
1118 	setup_dummy_ept();
1119 
1120 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
1121 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
1122 
1123 	return 0;
1124 }
1125 
1126 static void ept_enable_ad_bits(void)
1127 {
1128 	eptp |= EPTP_AD_FLAG;
1129 	vmcs_write(EPTP, eptp);
1130 }
1131 
1132 static void ept_disable_ad_bits(void)
1133 {
1134 	eptp &= ~EPTP_AD_FLAG;
1135 	vmcs_write(EPTP, eptp);
1136 }
1137 
1138 static void ept_enable_ad_bits_or_skip_test(void)
1139 {
1140 	if (!ept_ad_bits_supported())
1141 		test_skip("EPT AD bits not supported.");
1142 	ept_enable_ad_bits();
1143 }
1144 
1145 static int apic_version;
1146 
1147 static int ept_init_common(bool have_ad)
1148 {
1149 	int ret;
1150 	struct pci_dev pcidev;
1151 
1152 	if (setup_ept(have_ad))
1153 		return VMX_TEST_EXIT;
1154 	data_page1 = alloc_page();
1155 	data_page2 = alloc_page();
1156 	*((u32 *)data_page1) = MAGIC_VAL_1;
1157 	*((u32 *)data_page2) = MAGIC_VAL_2;
1158 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1159 			EPT_RA | EPT_WA | EPT_EA);
1160 
1161 	apic_version = apic_read(APIC_LVR);
1162 
1163 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1164 	if (ret != PCIDEVADDR_INVALID) {
1165 		pci_dev_init(&pcidev, ret);
1166 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1167 	}
1168 
1169 	return VMX_TEST_START;
1170 }
1171 
1172 static int ept_init(struct vmcs *vmcs)
1173 {
1174 	return ept_init_common(false);
1175 }
1176 
1177 static void ept_common(void)
1178 {
1179 	vmx_set_test_stage(0);
1180 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1181 			*((u32 *)data_page1) != MAGIC_VAL_1)
1182 		report("EPT basic framework - read", 0);
1183 	else {
1184 		*((u32 *)data_page2) = MAGIC_VAL_3;
1185 		vmcall();
1186 		if (vmx_get_test_stage() == 1) {
1187 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1188 					*((u32 *)data_page2) == MAGIC_VAL_2)
1189 				report("EPT basic framework", 1);
1190 			else
1191 				report("EPT basic framework - remap", 1);
1192 		}
1193 	}
1194 	// Test EPT Misconfigurations
1195 	vmx_set_test_stage(1);
1196 	vmcall();
1197 	*((u32 *)data_page1) = MAGIC_VAL_1;
1198 	if (vmx_get_test_stage() != 2) {
1199 		report("EPT misconfigurations", 0);
1200 		goto t1;
1201 	}
1202 	vmx_set_test_stage(2);
1203 	vmcall();
1204 	*((u32 *)data_page1) = MAGIC_VAL_1;
1205 	report("EPT misconfigurations", vmx_get_test_stage() == 3);
1206 t1:
1207 	// Test EPT violation
1208 	vmx_set_test_stage(3);
1209 	vmcall();
1210 	*((u32 *)data_page1) = MAGIC_VAL_1;
1211 	report("EPT violation - page permission", vmx_get_test_stage() == 4);
1212 	// Violation caused by EPT paging structure
1213 	vmx_set_test_stage(4);
1214 	vmcall();
1215 	*((u32 *)data_page1) = MAGIC_VAL_2;
1216 	report("EPT violation - paging structure", vmx_get_test_stage() == 5);
1217 
1218 	// MMIO Read/Write
1219 	vmx_set_test_stage(5);
1220 	vmcall();
1221 
1222 	*(u32 volatile *)pci_physaddr;
1223 	report("MMIO EPT violation - read", vmx_get_test_stage() == 6);
1224 
1225 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1226 	report("MMIO EPT violation - write", vmx_get_test_stage() == 7);
1227 }
1228 
1229 static void ept_main(void)
1230 {
1231 	ept_common();
1232 
1233 	// Test EPT access to L1 MMIO
1234 	vmx_set_test_stage(7);
1235 	report("EPT - MMIO access", *((u32 *)0xfee00030UL) == apic_version);
1236 
1237 	// Test invalid operand for INVEPT
1238 	vmcall();
1239 	report("EPT - unsupported INVEPT", vmx_get_test_stage() == 8);
1240 }
1241 
1242 static bool invept_test(int type, u64 eptp)
1243 {
1244 	bool ret, supported;
1245 
1246 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1247 	ret = invept(type, eptp);
1248 
1249 	if (ret == !supported)
1250 		return false;
1251 
1252 	if (!supported)
1253 		printf("WARNING: unsupported invept passed!\n");
1254 	else
1255 		printf("WARNING: invept failed!\n");
1256 
1257 	return true;
1258 }
1259 
1260 static int pml_exit_handler(void)
1261 {
1262 	u16 index, count;
1263 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1264 	u64 *pmlbuf = pml_log;
1265 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1266 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1267 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1268 
1269 	switch (reason) {
1270 	case VMX_VMCALL:
1271 		switch (vmx_get_test_stage()) {
1272 		case 0:
1273 			index = vmcs_read(GUEST_PML_INDEX);
1274 			for (count = index + 1; count < PML_INDEX; count++) {
1275 				if (pmlbuf[count] == (u64)data_page2) {
1276 					vmx_inc_test_stage();
1277 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1278 					break;
1279 				}
1280 			}
1281 			break;
1282 		case 1:
1283 			index = vmcs_read(GUEST_PML_INDEX);
1284 			/* Keep clearing the dirty bit till a overflow */
1285 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1286 			break;
1287 		default:
1288 			report("unexpected stage, %d.", false,
1289 			       vmx_get_test_stage());
1290 			print_vmexit_info();
1291 			return VMX_TEST_VMEXIT;
1292 		}
1293 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1294 		return VMX_TEST_RESUME;
1295 	case VMX_PML_FULL:
1296 		vmx_inc_test_stage();
1297 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1298 		return VMX_TEST_RESUME;
1299 	default:
1300 		report("Unknown exit reason, %ld", false, reason);
1301 		print_vmexit_info();
1302 	}
1303 	return VMX_TEST_VMEXIT;
1304 }
1305 
1306 static int ept_exit_handler_common(bool have_ad)
1307 {
1308 	u64 guest_rip;
1309 	u64 guest_cr3;
1310 	ulong reason;
1311 	u32 insn_len;
1312 	u32 exit_qual;
1313 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte;
1314 
1315 	guest_rip = vmcs_read(GUEST_RIP);
1316 	guest_cr3 = vmcs_read(GUEST_CR3);
1317 	reason = vmcs_read(EXI_REASON) & 0xff;
1318 	insn_len = vmcs_read(EXI_INST_LEN);
1319 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1320 	switch (reason) {
1321 	case VMX_VMCALL:
1322 		switch (vmx_get_test_stage()) {
1323 		case 0:
1324 			check_ept_ad(pml4, guest_cr3,
1325 				     (unsigned long)data_page1,
1326 				     have_ad ? EPT_ACCESS_FLAG : 0,
1327 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1328 			check_ept_ad(pml4, guest_cr3,
1329 				     (unsigned long)data_page2,
1330 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1331 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1332 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1333 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1334 			if (have_ad)
1335 				ept_sync(INVEPT_SINGLE, eptp);;
1336 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1337 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1338 				vmx_inc_test_stage();
1339 				install_ept(pml4, (unsigned long)data_page2,
1340 						(unsigned long)data_page2,
1341 						EPT_RA | EPT_WA | EPT_EA);
1342 			} else
1343 				report("EPT basic framework - write", 0);
1344 			break;
1345 		case 1:
1346 			install_ept(pml4, (unsigned long)data_page1,
1347  				(unsigned long)data_page1, EPT_WA);
1348 			ept_sync(INVEPT_SINGLE, eptp);
1349 			break;
1350 		case 2:
1351 			install_ept(pml4, (unsigned long)data_page1,
1352  				(unsigned long)data_page1,
1353  				EPT_RA | EPT_WA | EPT_EA |
1354  				(2 << EPT_MEM_TYPE_SHIFT));
1355 			ept_sync(INVEPT_SINGLE, eptp);
1356 			break;
1357 		case 3:
1358 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1359 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1360 						1, &data_page1_pte));
1361 			set_ept_pte(pml4, (unsigned long)data_page1,
1362 				1, data_page1_pte & ~EPT_PRESENT);
1363 			ept_sync(INVEPT_SINGLE, eptp);
1364 			break;
1365 		case 4:
1366 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1367 						2, &data_page1_pte));
1368 			data_page1_pte &= PAGE_MASK;
1369 			TEST_ASSERT(get_ept_pte(pml4, data_page1_pte,
1370 						2, &data_page1_pte_pte));
1371 			set_ept_pte(pml4, data_page1_pte, 2,
1372 				data_page1_pte_pte & ~EPT_PRESENT);
1373 			ept_sync(INVEPT_SINGLE, eptp);
1374 			break;
1375 		case 5:
1376 			install_ept(pml4, (unsigned long)pci_physaddr,
1377 				(unsigned long)pci_physaddr, 0);
1378 			ept_sync(INVEPT_SINGLE, eptp);
1379 			break;
1380 		case 7:
1381 			if (!invept_test(0, eptp))
1382 				vmx_inc_test_stage();
1383 			break;
1384 		// Should not reach here
1385 		default:
1386 			report("ERROR - unexpected stage, %d.", false,
1387 			       vmx_get_test_stage());
1388 			print_vmexit_info();
1389 			return VMX_TEST_VMEXIT;
1390 		}
1391 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1392 		return VMX_TEST_RESUME;
1393 	case VMX_EPT_MISCONFIG:
1394 		switch (vmx_get_test_stage()) {
1395 		case 1:
1396 		case 2:
1397 			vmx_inc_test_stage();
1398 			install_ept(pml4, (unsigned long)data_page1,
1399  				(unsigned long)data_page1,
1400  				EPT_RA | EPT_WA | EPT_EA);
1401 			ept_sync(INVEPT_SINGLE, eptp);
1402 			break;
1403 		// Should not reach here
1404 		default:
1405 			report("ERROR - unexpected stage, %d.", false,
1406 			       vmx_get_test_stage());
1407 			print_vmexit_info();
1408 			return VMX_TEST_VMEXIT;
1409 		}
1410 		return VMX_TEST_RESUME;
1411 	case VMX_EPT_VIOLATION:
1412 		/*
1413 		 * Exit-qualifications are masked not to account for advanced
1414 		 * VM-exit information. Once KVM supports this feature, this
1415 		 * masking should be removed.
1416 		 */
1417 		exit_qual &= ~EPT_VLT_GUEST_MASK;
1418 
1419 		switch(vmx_get_test_stage()) {
1420 		case 3:
1421 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1422 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1423 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1424 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1425 					EPT_VLT_PADDR))
1426 				vmx_inc_test_stage();
1427 			set_ept_pte(pml4, (unsigned long)data_page1,
1428 				1, data_page1_pte | (EPT_PRESENT));
1429 			ept_sync(INVEPT_SINGLE, eptp);
1430 			break;
1431 		case 4:
1432 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1433 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1434 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1435 			if (exit_qual == (EPT_VLT_RD |
1436 					  (have_ad ? EPT_VLT_WR : 0) |
1437 					  EPT_VLT_LADDR_VLD))
1438 				vmx_inc_test_stage();
1439 			set_ept_pte(pml4, data_page1_pte, 2,
1440 				data_page1_pte_pte | (EPT_PRESENT));
1441 			ept_sync(INVEPT_SINGLE, eptp);
1442 			break;
1443 		case 5:
1444 			if (exit_qual & EPT_VLT_RD)
1445 				vmx_inc_test_stage();
1446 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1447 						1, &memaddr_pte));
1448 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1449 			ept_sync(INVEPT_SINGLE, eptp);
1450 			break;
1451 		case 6:
1452 			if (exit_qual & EPT_VLT_WR)
1453 				vmx_inc_test_stage();
1454 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1455 						1, &memaddr_pte));
1456 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1457 			ept_sync(INVEPT_SINGLE, eptp);
1458 			break;
1459 		default:
1460 			// Should not reach here
1461 			report("ERROR : unexpected stage, %d", false,
1462 			       vmx_get_test_stage());
1463 			print_vmexit_info();
1464 			return VMX_TEST_VMEXIT;
1465 		}
1466 		return VMX_TEST_RESUME;
1467 	default:
1468 		report("Unknown exit reason, %ld", false, reason);
1469 		print_vmexit_info();
1470 	}
1471 	return VMX_TEST_VMEXIT;
1472 }
1473 
1474 static int ept_exit_handler(void)
1475 {
1476 	return ept_exit_handler_common(false);
1477 }
1478 
1479 static int eptad_init(struct vmcs *vmcs)
1480 {
1481 	int r = ept_init_common(true);
1482 
1483 	if (r == VMX_TEST_EXIT)
1484 		return r;
1485 
1486 	if ((rdmsr(MSR_IA32_VMX_EPT_VPID_CAP) & EPT_CAP_AD_FLAG) == 0) {
1487 		printf("\tEPT A/D bits are not supported");
1488 		return VMX_TEST_EXIT;
1489 	}
1490 
1491 	return r;
1492 }
1493 
1494 static int pml_init(struct vmcs *vmcs)
1495 {
1496 	u32 ctrl_cpu;
1497 	int r = eptad_init(vmcs);
1498 
1499 	if (r == VMX_TEST_EXIT)
1500 		return r;
1501 
1502 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1503 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1504 		printf("\tPML is not supported");
1505 		return VMX_TEST_EXIT;
1506 	}
1507 
1508 	pml_log = alloc_page();
1509 	vmcs_write(PMLADDR, (u64)pml_log);
1510 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1511 
1512 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1513 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1514 
1515 	return VMX_TEST_START;
1516 }
1517 
1518 static void pml_main(void)
1519 {
1520 	int count = 0;
1521 
1522 	vmx_set_test_stage(0);
1523 	*((u32 *)data_page2) = 0x1;
1524 	vmcall();
1525 	report("PML - Dirty GPA Logging", vmx_get_test_stage() == 1);
1526 
1527 	while (vmx_get_test_stage() == 1) {
1528 		vmcall();
1529 		*((u32 *)data_page2) = 0x1;
1530 		if (count++ > PML_INDEX)
1531 			break;
1532 	}
1533 	report("PML Full Event", vmx_get_test_stage() == 2);
1534 }
1535 
1536 static void eptad_main(void)
1537 {
1538 	ept_common();
1539 }
1540 
1541 static int eptad_exit_handler(void)
1542 {
1543 	return ept_exit_handler_common(true);
1544 }
1545 
1546 static bool invvpid_test(int type, u16 vpid)
1547 {
1548 	bool ret, supported;
1549 
1550 	supported = ept_vpid.val &
1551 		(VPID_CAP_INVVPID_ADDR >> INVVPID_ADDR << type);
1552 	ret = invvpid(type, vpid, 0);
1553 
1554 	if (ret == !supported)
1555 		return false;
1556 
1557 	if (!supported)
1558 		printf("WARNING: unsupported invvpid passed!\n");
1559 	else
1560 		printf("WARNING: invvpid failed!\n");
1561 
1562 	return true;
1563 }
1564 
1565 static int vpid_init(struct vmcs *vmcs)
1566 {
1567 	u32 ctrl_cpu1;
1568 
1569 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1570 		!(ctrl_cpu_rev[1].clr & CPU_VPID)) {
1571 		printf("\tVPID is not supported");
1572 		return VMX_TEST_EXIT;
1573 	}
1574 
1575 	ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
1576 	ctrl_cpu1 |= CPU_VPID;
1577 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
1578 	return VMX_TEST_START;
1579 }
1580 
1581 static void vpid_main(void)
1582 {
1583 	vmx_set_test_stage(0);
1584 	vmcall();
1585 	report("INVVPID SINGLE ADDRESS", vmx_get_test_stage() == 1);
1586 	vmx_set_test_stage(2);
1587 	vmcall();
1588 	report("INVVPID SINGLE", vmx_get_test_stage() == 3);
1589 	vmx_set_test_stage(4);
1590 	vmcall();
1591 	report("INVVPID ALL", vmx_get_test_stage() == 5);
1592 }
1593 
1594 static int vpid_exit_handler(void)
1595 {
1596 	u64 guest_rip;
1597 	ulong reason;
1598 	u32 insn_len;
1599 
1600 	guest_rip = vmcs_read(GUEST_RIP);
1601 	reason = vmcs_read(EXI_REASON) & 0xff;
1602 	insn_len = vmcs_read(EXI_INST_LEN);
1603 
1604 	switch (reason) {
1605 	case VMX_VMCALL:
1606 		switch(vmx_get_test_stage()) {
1607 		case 0:
1608 			if (!invvpid_test(INVVPID_ADDR, 1))
1609 				vmx_inc_test_stage();
1610 			break;
1611 		case 2:
1612 			if (!invvpid_test(INVVPID_CONTEXT_GLOBAL, 1))
1613 				vmx_inc_test_stage();
1614 			break;
1615 		case 4:
1616 			if (!invvpid_test(INVVPID_ALL, 1))
1617 				vmx_inc_test_stage();
1618 			break;
1619 		default:
1620 			report("ERROR: unexpected stage, %d", false,
1621 					vmx_get_test_stage());
1622 			print_vmexit_info();
1623 			return VMX_TEST_VMEXIT;
1624 		}
1625 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1626 		return VMX_TEST_RESUME;
1627 	default:
1628 		report("Unknown exit reason, %ld", false, reason);
1629 		print_vmexit_info();
1630 	}
1631 	return VMX_TEST_VMEXIT;
1632 }
1633 
1634 #define TIMER_VECTOR	222
1635 
1636 static volatile bool timer_fired;
1637 
1638 static void timer_isr(isr_regs_t *regs)
1639 {
1640 	timer_fired = true;
1641 	apic_write(APIC_EOI, 0);
1642 }
1643 
1644 static int interrupt_init(struct vmcs *vmcs)
1645 {
1646 	msr_bmp_init();
1647 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1648 	handle_irq(TIMER_VECTOR, timer_isr);
1649 	return VMX_TEST_START;
1650 }
1651 
1652 static void interrupt_main(void)
1653 {
1654 	long long start, loops;
1655 
1656 	vmx_set_test_stage(0);
1657 
1658 	apic_write(APIC_LVTT, TIMER_VECTOR);
1659 	irq_enable();
1660 
1661 	apic_write(APIC_TMICT, 1);
1662 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1663 		asm volatile ("nop");
1664 	report("direct interrupt while running guest", timer_fired);
1665 
1666 	apic_write(APIC_TMICT, 0);
1667 	irq_disable();
1668 	vmcall();
1669 	timer_fired = false;
1670 	apic_write(APIC_TMICT, 1);
1671 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1672 		asm volatile ("nop");
1673 	report("intercepted interrupt while running guest", timer_fired);
1674 
1675 	irq_enable();
1676 	apic_write(APIC_TMICT, 0);
1677 	irq_disable();
1678 	vmcall();
1679 	timer_fired = false;
1680 	start = rdtsc();
1681 	apic_write(APIC_TMICT, 1000000);
1682 
1683 	asm volatile ("sti; hlt");
1684 
1685 	report("direct interrupt + hlt",
1686 	       rdtsc() - start > 1000000 && timer_fired);
1687 
1688 	apic_write(APIC_TMICT, 0);
1689 	irq_disable();
1690 	vmcall();
1691 	timer_fired = false;
1692 	start = rdtsc();
1693 	apic_write(APIC_TMICT, 1000000);
1694 
1695 	asm volatile ("sti; hlt");
1696 
1697 	report("intercepted interrupt + hlt",
1698 	       rdtsc() - start > 10000 && timer_fired);
1699 
1700 	apic_write(APIC_TMICT, 0);
1701 	irq_disable();
1702 	vmcall();
1703 	timer_fired = false;
1704 	start = rdtsc();
1705 	apic_write(APIC_TMICT, 1000000);
1706 
1707 	irq_enable();
1708 	asm volatile ("nop");
1709 	vmcall();
1710 
1711 	report("direct interrupt + activity state hlt",
1712 	       rdtsc() - start > 10000 && timer_fired);
1713 
1714 	apic_write(APIC_TMICT, 0);
1715 	irq_disable();
1716 	vmcall();
1717 	timer_fired = false;
1718 	start = rdtsc();
1719 	apic_write(APIC_TMICT, 1000000);
1720 
1721 	irq_enable();
1722 	asm volatile ("nop");
1723 	vmcall();
1724 
1725 	report("intercepted interrupt + activity state hlt",
1726 	       rdtsc() - start > 10000 && timer_fired);
1727 
1728 	apic_write(APIC_TMICT, 0);
1729 	irq_disable();
1730 	vmx_set_test_stage(7);
1731 	vmcall();
1732 	timer_fired = false;
1733 	apic_write(APIC_TMICT, 1);
1734 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1735 		asm volatile ("nop");
1736 	report("running a guest with interrupt acknowledgement set", timer_fired);
1737 
1738 	apic_write(APIC_TMICT, 0);
1739 	irq_enable();
1740 	timer_fired = false;
1741 	vmcall();
1742 	report("Inject an event to a halted guest", timer_fired);
1743 }
1744 
1745 static int interrupt_exit_handler(void)
1746 {
1747 	u64 guest_rip = vmcs_read(GUEST_RIP);
1748 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1749 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1750 
1751 	switch (reason) {
1752 	case VMX_VMCALL:
1753 		switch (vmx_get_test_stage()) {
1754 		case 0:
1755 		case 2:
1756 		case 5:
1757 			vmcs_write(PIN_CONTROLS,
1758 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1759 			break;
1760 		case 7:
1761 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1762 			vmcs_write(PIN_CONTROLS,
1763 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1764 			break;
1765 		case 1:
1766 		case 3:
1767 			vmcs_write(PIN_CONTROLS,
1768 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1769 			break;
1770 		case 4:
1771 		case 6:
1772 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1773 			break;
1774 
1775 		case 8:
1776 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1777 			vmcs_write(ENT_INTR_INFO,
1778 				   TIMER_VECTOR |
1779 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1780 				   INTR_INFO_VALID_MASK);
1781 			break;
1782 		}
1783 		vmx_inc_test_stage();
1784 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1785 		return VMX_TEST_RESUME;
1786 	case VMX_EXTINT:
1787 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1788 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1789 			handle_external_interrupt(vector);
1790 		} else {
1791 			irq_enable();
1792 			asm volatile ("nop");
1793 			irq_disable();
1794 		}
1795 		if (vmx_get_test_stage() >= 2)
1796 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1797 		return VMX_TEST_RESUME;
1798 	default:
1799 		report("Unknown exit reason, %ld", false, reason);
1800 		print_vmexit_info();
1801 	}
1802 
1803 	return VMX_TEST_VMEXIT;
1804 }
1805 
1806 static int dbgctls_init(struct vmcs *vmcs)
1807 {
1808 	u64 dr7 = 0x402;
1809 	u64 zero = 0;
1810 
1811 	msr_bmp_init();
1812 	asm volatile(
1813 		"mov %0,%%dr0\n\t"
1814 		"mov %0,%%dr1\n\t"
1815 		"mov %0,%%dr2\n\t"
1816 		"mov %1,%%dr7\n\t"
1817 		: : "r" (zero), "r" (dr7));
1818 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1819 	vmcs_write(GUEST_DR7, 0x404);
1820 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1821 
1822 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1823 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1824 
1825 	return VMX_TEST_START;
1826 }
1827 
1828 static void dbgctls_main(void)
1829 {
1830 	u64 dr7, debugctl;
1831 
1832 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1833 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1834 	/* Commented out: KVM does not support DEBUGCTL so far */
1835 	(void)debugctl;
1836 	report("Load debug controls", dr7 == 0x404 /* && debugctl == 0x2 */);
1837 
1838 	dr7 = 0x408;
1839 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1840 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1841 
1842 	vmx_set_test_stage(0);
1843 	vmcall();
1844 	report("Save debug controls", vmx_get_test_stage() == 1);
1845 
1846 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1847 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1848 		printf("\tDebug controls are always loaded/saved\n");
1849 		return;
1850 	}
1851 	vmx_set_test_stage(2);
1852 	vmcall();
1853 
1854 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1855 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1856 	/* Commented out: KVM does not support DEBUGCTL so far */
1857 	(void)debugctl;
1858 	report("Guest=host debug controls", dr7 == 0x402 /* && debugctl == 0x1 */);
1859 
1860 	dr7 = 0x408;
1861 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1862 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1863 
1864 	vmx_set_test_stage(3);
1865 	vmcall();
1866 	report("Don't save debug controls", vmx_get_test_stage() == 4);
1867 }
1868 
1869 static int dbgctls_exit_handler(void)
1870 {
1871 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
1872 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1873 	u64 guest_rip = vmcs_read(GUEST_RIP);
1874 	u64 dr7, debugctl;
1875 
1876 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1877 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1878 
1879 	switch (reason) {
1880 	case VMX_VMCALL:
1881 		switch (vmx_get_test_stage()) {
1882 		case 0:
1883 			if (dr7 == 0x400 && debugctl == 0 &&
1884 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1885 			    Commented out: KVM does not support DEBUGCTL so far
1886 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1887 				vmx_inc_test_stage();
1888 			break;
1889 		case 2:
1890 			dr7 = 0x402;
1891 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1892 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1893 			vmcs_write(GUEST_DR7, 0x404);
1894 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1895 
1896 			vmcs_write(ENT_CONTROLS,
1897 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1898 			vmcs_write(EXI_CONTROLS,
1899 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1900 			break;
1901 		case 3:
1902 			if (dr7 == 0x400 && debugctl == 0 &&
1903 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1904 			    Commented out: KVM does not support DEBUGCTL so far
1905 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1906 				vmx_inc_test_stage();
1907 			break;
1908 		}
1909 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1910 		return VMX_TEST_RESUME;
1911 	default:
1912 		report("Unknown exit reason, %d", false, reason);
1913 		print_vmexit_info();
1914 	}
1915 	return VMX_TEST_VMEXIT;
1916 }
1917 
1918 struct vmx_msr_entry {
1919 	u32 index;
1920 	u32 reserved;
1921 	u64 value;
1922 } __attribute__((packed));
1923 
1924 #define MSR_MAGIC 0x31415926
1925 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1926 
1927 static int msr_switch_init(struct vmcs *vmcs)
1928 {
1929 	msr_bmp_init();
1930 	exit_msr_store = alloc_page();
1931 	exit_msr_load = alloc_page();
1932 	entry_msr_load = alloc_page();
1933 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1934 	entry_msr_load[0].value = MSR_MAGIC;
1935 
1936 	vmx_set_test_stage(1);
1937 	vmcs_write(ENT_MSR_LD_CNT, 1);
1938 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1939 	vmcs_write(EXI_MSR_ST_CNT, 1);
1940 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1941 	vmcs_write(EXI_MSR_LD_CNT, 1);
1942 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1943 	return VMX_TEST_START;
1944 }
1945 
1946 static void msr_switch_main(void)
1947 {
1948 	if (vmx_get_test_stage() == 1) {
1949 		report("VM entry MSR load",
1950 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC);
1951 		vmx_set_test_stage(2);
1952 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
1953 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
1954 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
1955 		exit_msr_load[0].value = MSR_MAGIC + 2;
1956 	}
1957 	vmcall();
1958 }
1959 
1960 static int msr_switch_exit_handler(void)
1961 {
1962 	ulong reason;
1963 
1964 	reason = vmcs_read(EXI_REASON);
1965 	if (reason == VMX_VMCALL && vmx_get_test_stage() == 2) {
1966 		report("VM exit MSR store",
1967 			exit_msr_store[0].value == MSR_MAGIC + 1);
1968 		report("VM exit MSR load",
1969 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2);
1970 		vmx_set_test_stage(3);
1971 		entry_msr_load[0].index = MSR_FS_BASE;
1972 		return VMX_TEST_RESUME;
1973 	}
1974 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1975 		__func__, vmx_get_test_stage(), reason);
1976 	return VMX_TEST_EXIT;
1977 }
1978 
1979 static int msr_switch_entry_failure(struct vmentry_failure *failure)
1980 {
1981 	ulong reason;
1982 
1983 	if (failure->early) {
1984 		printf("ERROR %s: early exit\n", __func__);
1985 		return VMX_TEST_EXIT;
1986 	}
1987 
1988 	reason = vmcs_read(EXI_REASON);
1989 	if (reason == (VMX_ENTRY_FAILURE | VMX_FAIL_MSR) &&
1990 	    vmx_get_test_stage() == 3) {
1991 		report("VM entry MSR load: try to load FS_BASE",
1992 			vmcs_read(EXI_QUALIFICATION) == 1);
1993 		return VMX_TEST_VMEXIT;
1994 	}
1995 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1996 		__func__, vmx_get_test_stage(), reason);
1997 	return VMX_TEST_EXIT;
1998 }
1999 
2000 static int vmmcall_init(struct vmcs *vmcs)
2001 {
2002 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
2003 	return VMX_TEST_START;
2004 }
2005 
2006 static void vmmcall_main(void)
2007 {
2008 	asm volatile(
2009 		"mov $0xABCD, %%rax\n\t"
2010 		"vmmcall\n\t"
2011 		::: "rax");
2012 
2013 	report("VMMCALL", 0);
2014 }
2015 
2016 static int vmmcall_exit_handler(void)
2017 {
2018 	ulong reason;
2019 
2020 	reason = vmcs_read(EXI_REASON);
2021 	switch (reason) {
2022 	case VMX_VMCALL:
2023 		printf("here\n");
2024 		report("VMMCALL triggers #UD", 0);
2025 		break;
2026 	case VMX_EXC_NMI:
2027 		report("VMMCALL triggers #UD",
2028 		       (vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR);
2029 		break;
2030 	default:
2031 		report("Unknown exit reason, %ld", false, reason);
2032 		print_vmexit_info();
2033 	}
2034 
2035 	return VMX_TEST_VMEXIT;
2036 }
2037 
2038 static int disable_rdtscp_init(struct vmcs *vmcs)
2039 {
2040 	u32 ctrl_cpu1;
2041 
2042 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
2043 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
2044 		ctrl_cpu1 &= ~CPU_RDTSCP;
2045 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
2046 	}
2047 
2048 	return VMX_TEST_START;
2049 }
2050 
2051 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
2052 {
2053 	switch (vmx_get_test_stage()) {
2054 	case 0:
2055 		report("RDTSCP triggers #UD", true);
2056 		vmx_inc_test_stage();
2057 		regs->rip += 3;
2058 		break;
2059 	case 2:
2060 		report("RDPID triggers #UD", true);
2061 		vmx_inc_test_stage();
2062 		regs->rip += 4;
2063 		break;
2064 	}
2065 	return;
2066 
2067 }
2068 
2069 static void disable_rdtscp_main(void)
2070 {
2071 	/* Test that #UD is properly injected in L2.  */
2072 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2073 
2074 	vmx_set_test_stage(0);
2075 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2076 	vmcall();
2077 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2078 
2079 	handle_exception(UD_VECTOR, 0);
2080 	vmcall();
2081 }
2082 
2083 static int disable_rdtscp_exit_handler(void)
2084 {
2085 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
2086 
2087 	switch (reason) {
2088 	case VMX_VMCALL:
2089 		switch (vmx_get_test_stage()) {
2090 		case 0:
2091 			report("RDTSCP triggers #UD", false);
2092 			vmx_inc_test_stage();
2093 			/* fallthrough */
2094 		case 1:
2095 			vmx_inc_test_stage();
2096 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2097 			return VMX_TEST_RESUME;
2098 		case 2:
2099 			report("RDPID triggers #UD", false);
2100 			break;
2101 		}
2102 		break;
2103 
2104 	default:
2105 		report("Unknown exit reason, %d", false, reason);
2106 		print_vmexit_info();
2107 	}
2108 	return VMX_TEST_VMEXIT;
2109 }
2110 
2111 static int int3_init(struct vmcs *vmcs)
2112 {
2113 	vmcs_write(EXC_BITMAP, ~0u);
2114 	return VMX_TEST_START;
2115 }
2116 
2117 static void int3_guest_main(void)
2118 {
2119 	asm volatile ("int3");
2120 }
2121 
2122 static int int3_exit_handler(void)
2123 {
2124 	u32 reason = vmcs_read(EXI_REASON);
2125 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2126 
2127 	report("L1 intercepts #BP", reason == VMX_EXC_NMI &&
2128 	       (intr_info & INTR_INFO_VALID_MASK) &&
2129 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2130 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2131 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2132 
2133 	return VMX_TEST_VMEXIT;
2134 }
2135 
2136 static int into_init(struct vmcs *vmcs)
2137 {
2138 	vmcs_write(EXC_BITMAP, ~0u);
2139 	return VMX_TEST_START;
2140 }
2141 
2142 static void into_guest_main(void)
2143 {
2144 	struct far_pointer32 fp = {
2145 		.offset = (uintptr_t)&&into,
2146 		.selector = KERNEL_CS32,
2147 	};
2148 	register uintptr_t rsp asm("rsp");
2149 
2150 	if (fp.offset != (uintptr_t)&&into) {
2151 		printf("Code address too high.\n");
2152 		return;
2153 	}
2154 	if ((u32)rsp != rsp) {
2155 		printf("Stack address too high.\n");
2156 		return;
2157 	}
2158 
2159 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2160 	return;
2161 into:
2162 	asm volatile (".code32;"
2163 		      "movl $0x7fffffff, %eax;"
2164 		      "addl %eax, %eax;"
2165 		      "into;"
2166 		      "lret;"
2167 		      ".code64");
2168 	__builtin_unreachable();
2169 }
2170 
2171 static int into_exit_handler(void)
2172 {
2173 	u32 reason = vmcs_read(EXI_REASON);
2174 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2175 
2176 	report("L1 intercepts #OF", reason == VMX_EXC_NMI &&
2177 	       (intr_info & INTR_INFO_VALID_MASK) &&
2178 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2179 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2180 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2181 
2182 	return VMX_TEST_VMEXIT;
2183 }
2184 
2185 static void exit_monitor_from_l2_main(void)
2186 {
2187 	printf("Calling exit(0) from l2...\n");
2188 	exit(0);
2189 }
2190 
2191 static int exit_monitor_from_l2_handler(void)
2192 {
2193 	report("The guest should have killed the VMM", false);
2194 	return VMX_TEST_EXIT;
2195 }
2196 
2197 static void assert_exit_reason(u64 expected)
2198 {
2199 	u64 actual = vmcs_read(EXI_REASON);
2200 
2201 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2202 			   exit_reason_description(expected),
2203 			   exit_reason_description(actual));
2204 }
2205 
2206 static void skip_exit_insn(void)
2207 {
2208 	u64 guest_rip = vmcs_read(GUEST_RIP);
2209 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2210 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2211 }
2212 
2213 static void skip_exit_vmcall(void)
2214 {
2215 	assert_exit_reason(VMX_VMCALL);
2216 	skip_exit_insn();
2217 }
2218 
2219 static void v2_null_test_guest(void)
2220 {
2221 }
2222 
2223 static void v2_null_test(void)
2224 {
2225 	test_set_guest(v2_null_test_guest);
2226 	enter_guest();
2227 	report(__func__, 1);
2228 }
2229 
2230 static void v2_multiple_entries_test_guest(void)
2231 {
2232 	vmx_set_test_stage(1);
2233 	vmcall();
2234 	vmx_set_test_stage(2);
2235 }
2236 
2237 static void v2_multiple_entries_test(void)
2238 {
2239 	test_set_guest(v2_multiple_entries_test_guest);
2240 	enter_guest();
2241 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2242 	skip_exit_vmcall();
2243 	enter_guest();
2244 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2245 	report(__func__, 1);
2246 }
2247 
2248 static int fixture_test_data = 1;
2249 
2250 static void fixture_test_teardown(void *data)
2251 {
2252 	*((int *) data) = 1;
2253 }
2254 
2255 static void fixture_test_guest(void)
2256 {
2257 	fixture_test_data++;
2258 }
2259 
2260 
2261 static void fixture_test_setup(void)
2262 {
2263 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2264 			   "fixture_test_teardown didn't run?!");
2265 	fixture_test_data = 2;
2266 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2267 	test_set_guest(fixture_test_guest);
2268 }
2269 
2270 static void fixture_test_case1(void)
2271 {
2272 	fixture_test_setup();
2273 	TEST_ASSERT_EQ(2, fixture_test_data);
2274 	enter_guest();
2275 	TEST_ASSERT_EQ(3, fixture_test_data);
2276 	report(__func__, 1);
2277 }
2278 
2279 static void fixture_test_case2(void)
2280 {
2281 	fixture_test_setup();
2282 	TEST_ASSERT_EQ(2, fixture_test_data);
2283 	enter_guest();
2284 	TEST_ASSERT_EQ(3, fixture_test_data);
2285 	report(__func__, 1);
2286 }
2287 
2288 enum ept_access_op {
2289 	OP_READ,
2290 	OP_WRITE,
2291 	OP_EXEC,
2292 	OP_FLUSH_TLB,
2293 	OP_EXIT,
2294 };
2295 
2296 static struct ept_access_test_data {
2297 	unsigned long gpa;
2298 	unsigned long *gva;
2299 	unsigned long hpa;
2300 	unsigned long *hva;
2301 	enum ept_access_op op;
2302 } ept_access_test_data;
2303 
2304 extern unsigned char ret42_start;
2305 extern unsigned char ret42_end;
2306 
2307 /* Returns 42. */
2308 asm(
2309 	".align 64\n"
2310 	"ret42_start:\n"
2311 	"mov $42, %eax\n"
2312 	"ret\n"
2313 	"ret42_end:\n"
2314 );
2315 
2316 static void
2317 diagnose_ept_violation_qual(u64 expected, u64 actual)
2318 {
2319 
2320 #define DIAGNOSE(flag)							\
2321 do {									\
2322 	if ((expected & flag) != (actual & flag))			\
2323 		printf(#flag " %sexpected\n",				\
2324 		       (expected & flag) ? "" : "un");			\
2325 } while (0)
2326 
2327 	DIAGNOSE(EPT_VLT_RD);
2328 	DIAGNOSE(EPT_VLT_WR);
2329 	DIAGNOSE(EPT_VLT_FETCH);
2330 	DIAGNOSE(EPT_VLT_PERM_RD);
2331 	DIAGNOSE(EPT_VLT_PERM_WR);
2332 	DIAGNOSE(EPT_VLT_PERM_EX);
2333 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2334 	DIAGNOSE(EPT_VLT_PADDR);
2335 
2336 #undef DIAGNOSE
2337 }
2338 
2339 static void do_ept_access_op(enum ept_access_op op)
2340 {
2341 	ept_access_test_data.op = op;
2342 	enter_guest();
2343 }
2344 
2345 /*
2346  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2347  * needed by tests that modify guest PTEs.
2348  */
2349 static void ept_access_test_guest_flush_tlb(void)
2350 {
2351 	do_ept_access_op(OP_FLUSH_TLB);
2352 	skip_exit_vmcall();
2353 }
2354 
2355 /*
2356  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2357  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2358  * a huge page.
2359  */
2360 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2361 				 unsigned long clear, unsigned long set)
2362 {
2363 	struct ept_access_test_data *data = &ept_access_test_data;
2364 	unsigned long orig_pte;
2365 	unsigned long pte;
2366 
2367 	/* Screw with the mapping at the requested level. */
2368 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2369 	pte = orig_pte;
2370 	if (mkhuge)
2371 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2372 	else
2373 		pte = orig_pte;
2374 	pte = (pte & ~clear) | set;
2375 	set_ept_pte(pml4, gpa, level, pte);
2376 	ept_sync(INVEPT_SINGLE, eptp);
2377 
2378 	return orig_pte;
2379 }
2380 
2381 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2382 {
2383 	set_ept_pte(pml4, gpa, level, orig_pte);
2384 	ept_sync(INVEPT_SINGLE, eptp);
2385 }
2386 
2387 static void do_ept_violation(bool leaf, enum ept_access_op op,
2388 			     u64 expected_qual, u64 expected_paddr)
2389 {
2390 	u64 qual;
2391 
2392 	/* Try the access and observe the violation. */
2393 	do_ept_access_op(op);
2394 
2395 	assert_exit_reason(VMX_EPT_VIOLATION);
2396 
2397 	qual = vmcs_read(EXI_QUALIFICATION);
2398 
2399 	/* Mask undefined bits (which may later be defined in certain cases). */
2400 	qual &= ~(EPT_VLT_GUEST_USER | EPT_VLT_GUEST_RW | EPT_VLT_GUEST_EX |
2401 		 EPT_VLT_PERM_USER_EX);
2402 
2403 	diagnose_ept_violation_qual(expected_qual, qual);
2404 	TEST_EXPECT_EQ(expected_qual, qual);
2405 
2406 	#if 0
2407 	/* Disable for now otherwise every test will fail */
2408 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2409 		       (unsigned long) (
2410 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2411 	#endif
2412 	/*
2413 	 * TODO: tests that probe expected_paddr in pages other than the one at
2414 	 * the beginning of the 1g region.
2415 	 */
2416 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2417 }
2418 
2419 static void
2420 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2421 			      unsigned long set, enum ept_access_op op,
2422 			      u64 expected_qual)
2423 {
2424 	struct ept_access_test_data *data = &ept_access_test_data;
2425 	unsigned long orig_pte;
2426 
2427 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2428 
2429 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2430 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2431 					 data->gpa);
2432 
2433 	/* Fix the violation and resume the op loop. */
2434 	ept_untwiddle(data->gpa, level, orig_pte);
2435 	enter_guest();
2436 	skip_exit_vmcall();
2437 }
2438 
2439 static void
2440 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2441 		       enum ept_access_op op, u64 expected_qual)
2442 {
2443 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2444 				      expected_qual);
2445 	if (ept_huge_pages_supported(level))
2446 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2447 					      expected_qual);
2448 }
2449 
2450 static void ept_violation(unsigned long clear, unsigned long set,
2451 			  enum ept_access_op op, u64 expected_qual)
2452 {
2453 	ept_violation_at_level(1, clear, set, op, expected_qual);
2454 	ept_violation_at_level(2, clear, set, op, expected_qual);
2455 	ept_violation_at_level(3, clear, set, op, expected_qual);
2456 	ept_violation_at_level(4, clear, set, op, expected_qual);
2457 }
2458 
2459 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2460 				       u64 expected_qual)
2461 {
2462 	ept_violation(EPT_PRESENT, access, op,
2463 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2464 }
2465 
2466 /*
2467  * For translations that don't involve a GVA, that is physical address (paddr)
2468  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2469  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2470  * translations don't involve GVAs, such as when the hardware does the guest
2471  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2472  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2473  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2474  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2475  *
2476  * Note that paddr violations can also be triggered by loading PAE page tables
2477  * with wonky addresses. We don't test that yet.
2478  *
2479  * This function modifies the EPT entry that maps the GPA that the guest page
2480  * table entry mapping ept_access_test_data.gva resides on.
2481  *
2482  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2483  *
2484  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2485  *
2486  *	@op		Guest operation to perform with
2487  *			ept_access_test_data.gva.
2488  *
2489  *	@expect_violation
2490  *			Is a violation expected during the paddr access?
2491  *
2492  *	@expected_qual	Expected qualification for the EPT violation.
2493  *			EPT_VLT_PADDR should be clear.
2494  */
2495 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2496 			     enum ept_access_op op, bool expect_violation,
2497 			     u64 expected_qual)
2498 {
2499 	struct ept_access_test_data *data = &ept_access_test_data;
2500 	unsigned long *ptep;
2501 	unsigned long gpa;
2502 	unsigned long orig_epte;
2503 
2504 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2505 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2506 	TEST_ASSERT(ptep);
2507 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2508 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2509 	ept_access_test_guest_flush_tlb();
2510 
2511 	/*
2512 	 * Now modify the access bits on the EPT entry for the GPA that the
2513 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2514 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2515 	 * constructed our test such that those other 511 PTEs aren't used by
2516 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2517 	 * PTE we're modifying is at the beginning of a 4K page and the
2518 	 * following 511 entires are also under our control (and not touched by
2519 	 * the guest).
2520 	 */
2521 	gpa = virt_to_phys(ptep);
2522 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2523 	/*
2524 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2525 	 * otherwise our level=1 twiddling below will fail. We use the
2526 	 * identity map (gpa = gpa) since page tables are shared with the host.
2527 	 */
2528 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2529 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2530 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2531 
2532 	if (expect_violation) {
2533 		do_ept_violation(/*leaf=*/true, op,
2534 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2535 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2536 		do_ept_access_op(op);
2537 	} else {
2538 		do_ept_access_op(op);
2539 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2540 	}
2541 
2542 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2543 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2544 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2545 
2546 	skip_exit_vmcall();
2547 }
2548 
2549 static void ept_access_allowed_paddr(unsigned long ept_access,
2550 				     unsigned long pte_ad,
2551 				     enum ept_access_op op)
2552 {
2553 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2554 			 /*expected_qual=*/-1);
2555 }
2556 
2557 static void ept_access_violation_paddr(unsigned long ept_access,
2558 				       unsigned long pte_ad,
2559 				       enum ept_access_op op,
2560 				       u64 expected_qual)
2561 {
2562 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2563 			 expected_qual);
2564 }
2565 
2566 
2567 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2568 					unsigned long clear,
2569 					unsigned long set,
2570 					enum ept_access_op op)
2571 {
2572 	struct ept_access_test_data *data = &ept_access_test_data;
2573 	unsigned long orig_pte;
2574 
2575 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2576 
2577 	/* No violation. Should proceed to vmcall. */
2578 	do_ept_access_op(op);
2579 	skip_exit_vmcall();
2580 
2581 	ept_untwiddle(data->gpa, level, orig_pte);
2582 }
2583 
2584 static void ept_allowed_at_level(int level, unsigned long clear,
2585 				 unsigned long set, enum ept_access_op op)
2586 {
2587 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2588 	if (ept_huge_pages_supported(level))
2589 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2590 }
2591 
2592 static void ept_allowed(unsigned long clear, unsigned long set,
2593 			enum ept_access_op op)
2594 {
2595 	ept_allowed_at_level(1, clear, set, op);
2596 	ept_allowed_at_level(2, clear, set, op);
2597 	ept_allowed_at_level(3, clear, set, op);
2598 	ept_allowed_at_level(4, clear, set, op);
2599 }
2600 
2601 static void ept_ignored_bit(int bit)
2602 {
2603 	/* Set the bit. */
2604 	ept_allowed(0, 1ul << bit, OP_READ);
2605 	ept_allowed(0, 1ul << bit, OP_WRITE);
2606 	ept_allowed(0, 1ul << bit, OP_EXEC);
2607 
2608 	/* Clear the bit. */
2609 	ept_allowed(1ul << bit, 0, OP_READ);
2610 	ept_allowed(1ul << bit, 0, OP_WRITE);
2611 	ept_allowed(1ul << bit, 0, OP_EXEC);
2612 }
2613 
2614 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2615 {
2616 	ept_allowed(EPT_PRESENT, access, op);
2617 }
2618 
2619 
2620 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2621 					     unsigned long clear,
2622 					     unsigned long set,
2623 					     enum ept_access_op op)
2624 {
2625 	struct ept_access_test_data *data = &ept_access_test_data;
2626 	unsigned long orig_pte;
2627 
2628 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2629 
2630 	do_ept_access_op(op);
2631 	assert_exit_reason(VMX_EPT_MISCONFIG);
2632 
2633 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2634 	#if 0
2635 	/* broken: */
2636 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2637 	#endif
2638 	#if 0
2639 	/*
2640 	 * broken:
2641 	 * According to description of exit qual for EPT violation,
2642 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2643 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2644 	 * to be set for msiconfig.
2645 	 */
2646 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2647 		       (unsigned long) (
2648 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2649 	#endif
2650 
2651 	/* Fix the violation and resume the op loop. */
2652 	ept_untwiddle(data->gpa, level, orig_pte);
2653 	enter_guest();
2654 	skip_exit_vmcall();
2655 }
2656 
2657 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2658 					  unsigned long clear,
2659 					  unsigned long set)
2660 {
2661 	/* The op shouldn't matter (read, write, exec), so try them all! */
2662 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2663 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2664 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2665 }
2666 
2667 static void ept_misconfig_at_level(int level, unsigned long clear,
2668 				   unsigned long set)
2669 {
2670 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2671 	if (ept_huge_pages_supported(level))
2672 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2673 }
2674 
2675 static void ept_misconfig(unsigned long clear, unsigned long set)
2676 {
2677 	ept_misconfig_at_level(1, clear, set);
2678 	ept_misconfig_at_level(2, clear, set);
2679 	ept_misconfig_at_level(3, clear, set);
2680 	ept_misconfig_at_level(4, clear, set);
2681 }
2682 
2683 static void ept_access_misconfig(unsigned long access)
2684 {
2685 	ept_misconfig(EPT_PRESENT, access);
2686 }
2687 
2688 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2689 {
2690 	/* Setting the bit causes a misconfig. */
2691 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2692 
2693 	/* Making the entry non-present turns reserved bits into ignored. */
2694 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2695 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2696 }
2697 
2698 static void ept_reserved_bit_at_level_huge(int level, int bit)
2699 {
2700 	/* Setting the bit causes a misconfig. */
2701 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2702 
2703 	/* Making the entry non-present turns reserved bits into ignored. */
2704 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2705 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2706 }
2707 
2708 static void ept_reserved_bit_at_level(int level, int bit)
2709 {
2710 	/* Setting the bit causes a misconfig. */
2711 	ept_misconfig_at_level(level, 0, 1ul << bit);
2712 
2713 	/* Making the entry non-present turns reserved bits into ignored. */
2714 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2715 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2716 }
2717 
2718 static void ept_reserved_bit(int bit)
2719 {
2720 	ept_reserved_bit_at_level(1, bit);
2721 	ept_reserved_bit_at_level(2, bit);
2722 	ept_reserved_bit_at_level(3, bit);
2723 	ept_reserved_bit_at_level(4, bit);
2724 }
2725 
2726 #define PAGE_2M_ORDER 9
2727 #define PAGE_1G_ORDER 18
2728 
2729 static void *get_1g_page(void)
2730 {
2731 	static void *alloc;
2732 
2733 	if (!alloc)
2734 		alloc = alloc_pages(PAGE_1G_ORDER);
2735 	return alloc;
2736 }
2737 
2738 static void ept_access_test_teardown(void *unused)
2739 {
2740 	/* Exit the guest cleanly. */
2741 	do_ept_access_op(OP_EXIT);
2742 }
2743 
2744 static void ept_access_test_guest(void)
2745 {
2746 	struct ept_access_test_data *data = &ept_access_test_data;
2747 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2748 
2749 	while (true) {
2750 		switch (data->op) {
2751 		case OP_READ:
2752 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2753 			break;
2754 		case OP_WRITE:
2755 			*data->gva = MAGIC_VAL_2;
2756 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2757 			*data->gva = MAGIC_VAL_1;
2758 			break;
2759 		case OP_EXEC:
2760 			TEST_ASSERT_EQ(42, code());
2761 			break;
2762 		case OP_FLUSH_TLB:
2763 			write_cr3(read_cr3());
2764 			break;
2765 		case OP_EXIT:
2766 			return;
2767 		default:
2768 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2769 		}
2770 		vmcall();
2771 	}
2772 }
2773 
2774 static void ept_access_test_setup(void)
2775 {
2776 	struct ept_access_test_data *data = &ept_access_test_data;
2777 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2778 	unsigned long size = npages * PAGE_SIZE;
2779 	unsigned long *page_table = current_page_table();
2780 	unsigned long pte;
2781 
2782 	if (setup_ept(false))
2783 		test_skip("EPT not supported");
2784 
2785 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2786 	if (cpuid_maxphyaddr() < 40)
2787 		test_skip("Test needs MAXPHYADDR >= 40");
2788 
2789 	test_set_guest(ept_access_test_guest);
2790 	test_add_teardown(ept_access_test_teardown, NULL);
2791 
2792 	data->hva = get_1g_page();
2793 	TEST_ASSERT(data->hva);
2794 	data->hpa = virt_to_phys(data->hva);
2795 
2796 	data->gpa = 1ul << 39;
2797 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2798 				   size);
2799 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2800 	install_pages(page_table, data->gpa, size, data->gva);
2801 
2802 	/*
2803 	 * Make sure nothing's mapped here so the tests that screw with the
2804 	 * pml4 entry don't inadvertently break something.
2805 	 */
2806 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2807 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2808 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2809 
2810 	data->hva[0] = MAGIC_VAL_1;
2811 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2812 }
2813 
2814 static void ept_access_test_not_present(void)
2815 {
2816 	ept_access_test_setup();
2817 	/* --- */
2818 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2819 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2820 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2821 }
2822 
2823 static void ept_access_test_read_only(void)
2824 {
2825 	ept_access_test_setup();
2826 
2827 	/* r-- */
2828 	ept_access_allowed(EPT_RA, OP_READ);
2829 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2830 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2831 }
2832 
2833 static void ept_access_test_write_only(void)
2834 {
2835 	ept_access_test_setup();
2836 	/* -w- */
2837 	ept_access_misconfig(EPT_WA);
2838 }
2839 
2840 static void ept_access_test_read_write(void)
2841 {
2842 	ept_access_test_setup();
2843 	/* rw- */
2844 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2845 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2846 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2847 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2848 }
2849 
2850 
2851 static void ept_access_test_execute_only(void)
2852 {
2853 	ept_access_test_setup();
2854 	/* --x */
2855 	if (ept_execute_only_supported()) {
2856 		ept_access_violation(EPT_EA, OP_READ,
2857 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2858 		ept_access_violation(EPT_EA, OP_WRITE,
2859 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2860 		ept_access_allowed(EPT_EA, OP_EXEC);
2861 	} else {
2862 		ept_access_misconfig(EPT_EA);
2863 	}
2864 }
2865 
2866 static void ept_access_test_read_execute(void)
2867 {
2868 	ept_access_test_setup();
2869 	/* r-x */
2870 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2871 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2872 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2873 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2874 }
2875 
2876 static void ept_access_test_write_execute(void)
2877 {
2878 	ept_access_test_setup();
2879 	/* -wx */
2880 	ept_access_misconfig(EPT_WA | EPT_EA);
2881 }
2882 
2883 static void ept_access_test_read_write_execute(void)
2884 {
2885 	ept_access_test_setup();
2886 	/* rwx */
2887 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2888 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2889 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2890 }
2891 
2892 static void ept_access_test_reserved_bits(void)
2893 {
2894 	int i;
2895 	int maxphyaddr;
2896 
2897 	ept_access_test_setup();
2898 
2899 	/* Reserved bits above maxphyaddr. */
2900 	maxphyaddr = cpuid_maxphyaddr();
2901 	for (i = maxphyaddr; i <= 51; i++) {
2902 		report_prefix_pushf("reserved_bit=%d", i);
2903 		ept_reserved_bit(i);
2904 		report_prefix_pop();
2905 	}
2906 
2907 	/* Level-specific reserved bits. */
2908 	ept_reserved_bit_at_level_nohuge(2, 3);
2909 	ept_reserved_bit_at_level_nohuge(2, 4);
2910 	ept_reserved_bit_at_level_nohuge(2, 5);
2911 	ept_reserved_bit_at_level_nohuge(2, 6);
2912 	/* 2M alignment. */
2913 	for (i = 12; i < 20; i++) {
2914 		report_prefix_pushf("reserved_bit=%d", i);
2915 		ept_reserved_bit_at_level_huge(2, i);
2916 		report_prefix_pop();
2917 	}
2918 	ept_reserved_bit_at_level_nohuge(3, 3);
2919 	ept_reserved_bit_at_level_nohuge(3, 4);
2920 	ept_reserved_bit_at_level_nohuge(3, 5);
2921 	ept_reserved_bit_at_level_nohuge(3, 6);
2922 	/* 1G alignment. */
2923 	for (i = 12; i < 29; i++) {
2924 		report_prefix_pushf("reserved_bit=%d", i);
2925 		ept_reserved_bit_at_level_huge(3, i);
2926 		report_prefix_pop();
2927 	}
2928 	ept_reserved_bit_at_level(4, 3);
2929 	ept_reserved_bit_at_level(4, 4);
2930 	ept_reserved_bit_at_level(4, 5);
2931 	ept_reserved_bit_at_level(4, 6);
2932 	ept_reserved_bit_at_level(4, 7);
2933 }
2934 
2935 static void ept_access_test_ignored_bits(void)
2936 {
2937 	ept_access_test_setup();
2938 	/*
2939 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2940 	 * far as translation is concerned even if AD bits are enabled in the
2941 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2942 	 * control is 0.
2943 	 */
2944 	ept_ignored_bit(8);
2945 	ept_ignored_bit(9);
2946 	ept_ignored_bit(10);
2947 	ept_ignored_bit(11);
2948 	ept_ignored_bit(52);
2949 	ept_ignored_bit(53);
2950 	ept_ignored_bit(54);
2951 	ept_ignored_bit(55);
2952 	ept_ignored_bit(56);
2953 	ept_ignored_bit(57);
2954 	ept_ignored_bit(58);
2955 	ept_ignored_bit(59);
2956 	ept_ignored_bit(60);
2957 	ept_ignored_bit(61);
2958 	ept_ignored_bit(62);
2959 	ept_ignored_bit(63);
2960 }
2961 
2962 static void ept_access_test_paddr_not_present_ad_disabled(void)
2963 {
2964 	ept_access_test_setup();
2965 	ept_disable_ad_bits();
2966 
2967 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
2968 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
2969 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
2970 }
2971 
2972 static void ept_access_test_paddr_not_present_ad_enabled(void)
2973 {
2974 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
2975 
2976 	ept_access_test_setup();
2977 	ept_enable_ad_bits_or_skip_test();
2978 
2979 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
2980 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
2981 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
2982 }
2983 
2984 static void ept_access_test_paddr_read_only_ad_disabled(void)
2985 {
2986 	/*
2987 	 * When EPT AD bits are disabled, all accesses to guest paging
2988 	 * structures are reported separately as a read and (after
2989 	 * translation of the GPA to host physical address) a read+write
2990 	 * if the A/D bits have to be set.
2991 	 */
2992 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2993 
2994 	ept_access_test_setup();
2995 	ept_disable_ad_bits();
2996 
2997 	/* Can't update A bit, so all accesses fail. */
2998 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2999 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3000 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3001 	/* AD bits disabled, so only writes try to update the D bit. */
3002 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
3003 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3004 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
3005 	/* Both A and D already set, so read-only is OK. */
3006 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
3007 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
3008 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
3009 }
3010 
3011 static void ept_access_test_paddr_read_only_ad_enabled(void)
3012 {
3013 	/*
3014 	 * When EPT AD bits are enabled, all accesses to guest paging
3015 	 * structures are considered writes as far as EPT translation
3016 	 * is concerned.
3017 	 */
3018 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3019 
3020 	ept_access_test_setup();
3021 	ept_enable_ad_bits_or_skip_test();
3022 
3023 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3024 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3025 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3026 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
3027 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3028 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
3029 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
3030 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
3031 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
3032 }
3033 
3034 static void ept_access_test_paddr_read_write(void)
3035 {
3036 	ept_access_test_setup();
3037 	/* Read-write access to paging structure. */
3038 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
3039 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
3040 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
3041 }
3042 
3043 static void ept_access_test_paddr_read_write_execute(void)
3044 {
3045 	ept_access_test_setup();
3046 	/* RWX access to paging structure. */
3047 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
3048 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
3049 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
3050 }
3051 
3052 static void ept_access_test_paddr_read_execute_ad_disabled(void)
3053 {
3054   	/*
3055 	 * When EPT AD bits are disabled, all accesses to guest paging
3056 	 * structures are reported separately as a read and (after
3057 	 * translation of the GPA to host physical address) a read+write
3058 	 * if the A/D bits have to be set.
3059 	 */
3060 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3061 
3062 	ept_access_test_setup();
3063 	ept_disable_ad_bits();
3064 
3065 	/* Can't update A bit, so all accesses fail. */
3066 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3067 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3068 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3069 	/* AD bits disabled, so only writes try to update the D bit. */
3070 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
3071 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3072 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3073 	/* Both A and D already set, so read-only is OK. */
3074 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3075 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3076 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3077 }
3078 
3079 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3080 {
3081 	/*
3082 	 * When EPT AD bits are enabled, all accesses to guest paging
3083 	 * structures are considered writes as far as EPT translation
3084 	 * is concerned.
3085 	 */
3086 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3087 
3088 	ept_access_test_setup();
3089 	ept_enable_ad_bits_or_skip_test();
3090 
3091 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3092 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3093 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3094 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3095 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3096 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3097 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3098 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3099 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3100 }
3101 
3102 static void ept_access_test_paddr_not_present_page_fault(void)
3103 {
3104 	ept_access_test_setup();
3105 	/*
3106 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3107 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3108 	 * Thus guest page fault before host takes EPT violation for trying to
3109 	 * update A bit.
3110 	 */
3111 }
3112 
3113 static void ept_access_test_force_2m_page(void)
3114 {
3115 	ept_access_test_setup();
3116 
3117 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3118 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3119 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3120 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3121 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3122 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3123 }
3124 
3125 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3126 {
3127 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3128 
3129 	TEST_ASSERT(msr & VPID_CAP_INVVPID);
3130 
3131 	if (type < INVVPID_ADDR || type > INVVPID_CONTEXT_LOCAL)
3132 		return false;
3133 
3134 	if (!(msr & (1ull << (type + VPID_CAP_INVVPID_TYPES_SHIFT))))
3135 		return false;
3136 
3137 	if (vpid >> 16)
3138 		return false;
3139 
3140 	if (type != INVVPID_ALL && !vpid)
3141 		return false;
3142 
3143 	if (type == INVVPID_ADDR && !is_canonical(gla))
3144 		return false;
3145 
3146 	return true;
3147 }
3148 
3149 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3150 {
3151 	int rc;
3152 	bool valid = invvpid_valid(type, vpid, gla);
3153 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3154 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3155 	/*
3156 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3157 	 * that we can tell if it is updated by INVVPID.
3158 	 */
3159 	vmcs_read(~0);
3160 	rc = invvpid(type, vpid, gla);
3161 	report("INVVPID type %ld VPID %lx GLA %lx %s",
3162 	       !rc == valid, type, vpid, gla,
3163 	       valid ? "passes" : "fails");
3164 	report("After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3165 	       vmcs_read(VMX_INST_ERROR) == expected,
3166 	       rc ? "failed" : "successful",
3167 	       expected, vmcs_read(VMX_INST_ERROR));
3168 }
3169 
3170 static void ds_invvpid(void *data)
3171 {
3172 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3173 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3174 
3175 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3176 	asm volatile("invvpid %0, %1"
3177 		     :
3178 		     : "m"(*(struct invvpid_operand *)data),
3179 		       "r"(type));
3180 }
3181 
3182 /*
3183  * The SS override is ignored in 64-bit mode, so we use an addressing
3184  * mode with %rsp as the base register to generate an implicit SS
3185  * reference.
3186  */
3187 static void ss_invvpid(void *data)
3188 {
3189 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3190 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3191 
3192 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3193 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3194 		     : "+r"(data)
3195 		     : "r"(type));
3196 }
3197 
3198 static void invvpid_test_gp(void)
3199 {
3200 	bool fault;
3201 
3202 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3203 				   (void *)NONCANONICAL);
3204 	report("INVVPID with non-canonical DS operand raises #GP", fault);
3205 }
3206 
3207 static void invvpid_test_ss(void)
3208 {
3209 	bool fault;
3210 
3211 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3212 				   (void *)NONCANONICAL);
3213 	report("INVVPID with non-canonical SS operand raises #SS", fault);
3214 }
3215 
3216 static void invvpid_test_pf(void)
3217 {
3218 	void *vpage = alloc_vpage();
3219 	bool fault;
3220 
3221 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3222 	report("INVVPID with unmapped operand raises #PF", fault);
3223 }
3224 
3225 static void try_compat_invvpid(void *unused)
3226 {
3227 	struct far_pointer32 fp = {
3228 		.offset = (uintptr_t)&&invvpid,
3229 		.selector = KERNEL_CS32,
3230 	};
3231 	register uintptr_t rsp asm("rsp");
3232 
3233 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3234 			"Code address too high.");
3235 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3236 
3237 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3238 	return;
3239 invvpid:
3240 	asm volatile (".code32;"
3241 		      "invvpid (%eax), %eax;"
3242 		      "lret;"
3243 		      ".code64");
3244 	__builtin_unreachable();
3245 }
3246 
3247 static void invvpid_test_compatibility_mode(void)
3248 {
3249 	bool fault;
3250 
3251 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3252 	report("Compatibility mode INVVPID raises #UD", fault);
3253 }
3254 
3255 static void invvpid_test_not_in_vmx_operation(void)
3256 {
3257 	bool fault;
3258 
3259 	TEST_ASSERT(!vmx_off());
3260 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3261 	report("INVVPID outside of VMX operation raises #UD", fault);
3262 	TEST_ASSERT(!vmx_on());
3263 }
3264 
3265 /*
3266  * This does not test real-address mode, virtual-8086 mode, protected mode,
3267  * or CPL > 0.
3268  */
3269 static void invvpid_test_v2(void)
3270 {
3271 	u64 msr;
3272 	int i;
3273 	unsigned types = 0;
3274 	unsigned type;
3275 
3276 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
3277 	    !(ctrl_cpu_rev[1].clr & CPU_VPID))
3278 		test_skip("VPID not supported");
3279 
3280 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3281 
3282 	if (!(msr & VPID_CAP_INVVPID))
3283 		test_skip("INVVPID not supported.\n");
3284 
3285 	if (msr & VPID_CAP_INVVPID_ADDR)
3286 		types |= 1u << INVVPID_ADDR;
3287 	if (msr & VPID_CAP_INVVPID_CXTGLB)
3288 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3289 	if (msr & VPID_CAP_INVVPID_ALL)
3290 		types |= 1u << INVVPID_ALL;
3291 	if (msr & VPID_CAP_INVVPID_CXTLOC)
3292 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3293 
3294 	if (!types)
3295 		test_skip("No INVVPID types supported.\n");
3296 
3297 	for (i = -127; i < 128; i++)
3298 		try_invvpid(i, 0xffff, 0);
3299 
3300 	/*
3301 	 * VPID must not be more than 16 bits.
3302 	 */
3303 	for (i = 0; i < 64; i++)
3304 		for (type = 0; type < 4; type++)
3305 			if (types & (1u << type))
3306 				try_invvpid(type, 1ul << i, 0);
3307 
3308 	/*
3309 	 * VPID must not be zero, except for "all contexts."
3310 	 */
3311 	for (type = 0; type < 4; type++)
3312 		if (types & (1u << type))
3313 			try_invvpid(type, 0, 0);
3314 
3315 	/*
3316 	 * The gla operand is only validated for single-address INVVPID.
3317 	 */
3318 	if (types & (1u << INVVPID_ADDR))
3319 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3320 
3321 	invvpid_test_gp();
3322 	invvpid_test_ss();
3323 	invvpid_test_pf();
3324 	invvpid_test_compatibility_mode();
3325 	invvpid_test_not_in_vmx_operation();
3326 }
3327 
3328 /*
3329  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3330  * at least as far as the guest-state checks. Returns false if the
3331  * VMLAUNCH fails early and execution falls through to the next
3332  * instruction.
3333  */
3334 static bool vmlaunch_succeeds(void)
3335 {
3336 	u32 exit_reason;
3337 
3338 	/*
3339 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3340 	 * unsupported VMCS component"). The caller can then check
3341 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3342 	 */
3343 	vmcs_write(~0u, 0);
3344 
3345 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3346 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3347 				   :
3348 				   : "r" ((u64)HOST_RSP)
3349 				   : "cc", "memory"
3350 				   : success);
3351 	return false;
3352 success:
3353 	exit_reason = vmcs_read(EXI_REASON);
3354 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3355 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3356 	return true;
3357 }
3358 
3359 /*
3360  * Try to launch the current VMCS.
3361  */
3362 static void test_vmx_vmlaunch(u32 xerror)
3363 {
3364 	bool success = vmlaunch_succeeds();
3365 	u32 vmx_inst_err;
3366 
3367 	report("vmlaunch %s", success == !xerror,
3368 	       !xerror ? "succeeds" : "fails");
3369 	if (!success && xerror) {
3370 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3371 		report("VMX inst error is %d (actual %d)",
3372 		       vmx_inst_err == xerror, xerror, vmx_inst_err);
3373 	}
3374 }
3375 
3376 static void test_vmx_invalid_controls(void)
3377 {
3378 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3379 }
3380 
3381 static void test_vmx_valid_controls(void)
3382 {
3383 	test_vmx_vmlaunch(0);
3384 }
3385 
3386 /*
3387  * Test a particular value of a VM-execution control bit, if the value
3388  * is required or if the value is zero.
3389  */
3390 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3391 				    enum Encoding encoding, unsigned bit,
3392 				    unsigned val)
3393 {
3394 	u32 mask = 1u << bit;
3395 	bool expected;
3396 	u32 controls;
3397 
3398 	if (msr.set & mask)
3399 		TEST_ASSERT(msr.clr & mask);
3400 
3401 	/*
3402 	 * We can't arbitrarily turn on a control bit, because it may
3403 	 * introduce dependencies on other VMCS fields. So, we only
3404 	 * test turning on bits that have a required setting.
3405 	 */
3406 	if (val && (msr.clr & mask) && !(msr.set & mask))
3407 		return;
3408 
3409 	report_prefix_pushf("%s %s bit %d",
3410 			    val ? "Set" : "Clear", name, bit);
3411 
3412 	controls = vmcs_read(encoding);
3413 	if (val) {
3414 		vmcs_write(encoding, msr.set | mask);
3415 		expected = (msr.clr & mask);
3416 	} else {
3417 		vmcs_write(encoding, msr.set & ~mask);
3418 		expected = !(msr.set & mask);
3419 	}
3420 	if (expected)
3421 		test_vmx_valid_controls();
3422 	else
3423 		test_vmx_invalid_controls();
3424 	vmcs_write(encoding, controls);
3425 	report_prefix_pop();
3426 }
3427 
3428 /*
3429  * Test reserved values of a VM-execution control bit, based on the
3430  * allowed bit settings from the corresponding VMX capability MSR.
3431  */
3432 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3433 			      enum Encoding encoding, unsigned bit)
3434 {
3435 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3436 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3437 }
3438 
3439 /*
3440  * Reserved bits in the pin-based VM-execution controls must be set
3441  * properly. Software may consult the VMX capability MSRs to determine
3442  * the proper settings.
3443  * [Intel SDM]
3444  */
3445 static void test_pin_based_ctls(void)
3446 {
3447 	unsigned bit;
3448 
3449 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3450 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3451 	for (bit = 0; bit < 32; bit++)
3452 		test_rsvd_ctl_bit("pin-based controls",
3453 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3454 }
3455 
3456 /*
3457  * Reserved bits in the primary processor-based VM-execution controls
3458  * must be set properly. Software may consult the VMX capability MSRs
3459  * to determine the proper settings.
3460  * [Intel SDM]
3461  */
3462 static void test_primary_processor_based_ctls(void)
3463 {
3464 	unsigned bit;
3465 
3466 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3467 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3468 	for (bit = 0; bit < 32; bit++)
3469 		test_rsvd_ctl_bit("primary processor-based controls",
3470 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3471 }
3472 
3473 /*
3474  * If the "activate secondary controls" primary processor-based
3475  * VM-execution control is 1, reserved bits in the secondary
3476  * processor-based VM-execution controls must be cleared. Software may
3477  * consult the VMX capability MSRs to determine which bits are
3478  * reserved.
3479  * If the "activate secondary controls" primary processor-based
3480  * VM-execution control is 0 (or if the processor does not support the
3481  * 1-setting of that control), no checks are performed on the
3482  * secondary processor-based VM-execution controls.
3483  * [Intel SDM]
3484  */
3485 static void test_secondary_processor_based_ctls(void)
3486 {
3487 	u32 primary;
3488 	u32 secondary;
3489 	unsigned bit;
3490 
3491 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3492 		return;
3493 
3494 	primary = vmcs_read(CPU_EXEC_CTRL0);
3495 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3496 
3497 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3498 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3499 	for (bit = 0; bit < 32; bit++)
3500 		test_rsvd_ctl_bit("secondary processor-based controls",
3501 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3502 
3503 	/*
3504 	 * When the "activate secondary controls" VM-execution control
3505 	 * is clear, there are no checks on the secondary controls.
3506 	 */
3507 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3508 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3509 	report("Secondary processor-based controls ignored",
3510 	       vmlaunch_succeeds());
3511 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3512 	vmcs_write(CPU_EXEC_CTRL0, primary);
3513 }
3514 
3515 static void try_cr3_target_count(unsigned i, unsigned max)
3516 {
3517 	report_prefix_pushf("CR3 target count 0x%x", i);
3518 	vmcs_write(CR3_TARGET_COUNT, i);
3519 	if (i <= max)
3520 		test_vmx_valid_controls();
3521 	else
3522 		test_vmx_invalid_controls();
3523 	report_prefix_pop();
3524 }
3525 
3526 /*
3527  * The CR3-target count must not be greater than 4. Future processors
3528  * may support a different number of CR3-target values. Software
3529  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3530  * number of values supported.
3531  * [Intel SDM]
3532  */
3533 static void test_cr3_targets(void)
3534 {
3535 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3536 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3537 	unsigned i;
3538 
3539 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3540 	TEST_ASSERT(supported_targets <= 256);
3541 
3542 	try_cr3_target_count(-1u, supported_targets);
3543 	try_cr3_target_count(0x80000000, supported_targets);
3544 	try_cr3_target_count(0x7fffffff, supported_targets);
3545 	for (i = 0; i <= supported_targets + 1; i++)
3546 		try_cr3_target_count(i, supported_targets);
3547 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3548 }
3549 
3550 /*
3551  * Test a particular address setting in the VMCS
3552  */
3553 static void test_vmcs_addr(const char *name,
3554 			   enum Encoding encoding,
3555 			   u64 align,
3556 			   bool ignored,
3557 			   bool skip_beyond_mapped_ram,
3558 			   u64 addr)
3559 {
3560 	report_prefix_pushf("%s = %lx", name, addr);
3561 	vmcs_write(encoding, addr);
3562 	if (skip_beyond_mapped_ram &&
3563 	    addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3564 	    addr < (1ul << cpuid_maxphyaddr()))
3565 		printf("Skipping physical address beyond mapped RAM\n");
3566 	else if (ignored || (IS_ALIGNED(addr, align) &&
3567 	    addr < (1ul << cpuid_maxphyaddr())))
3568 		test_vmx_valid_controls();
3569 	else
3570 		test_vmx_invalid_controls();
3571 	report_prefix_pop();
3572 }
3573 
3574 /*
3575  * Test interesting values for a VMCS address
3576  */
3577 static void test_vmcs_addr_values(const char *name,
3578 				  enum Encoding encoding,
3579 				  u64 align,
3580 				  bool ignored,
3581 				  bool skip_beyond_mapped_ram,
3582 				  u32 bit_start, u32 bit_end)
3583 {
3584 	unsigned i;
3585 	u64 orig_val = vmcs_read(encoding);
3586 
3587 	for (i = bit_start; i <= bit_end; i++)
3588 		test_vmcs_addr(name, encoding, align, ignored,
3589 			       skip_beyond_mapped_ram, 1ul << i);
3590 
3591 	test_vmcs_addr(name, encoding, align, ignored,
3592 		       skip_beyond_mapped_ram, PAGE_SIZE - 1);
3593 	test_vmcs_addr(name, encoding, align, ignored,
3594 		       skip_beyond_mapped_ram, PAGE_SIZE);
3595 	test_vmcs_addr(name, encoding, align, ignored,
3596 		       skip_beyond_mapped_ram,
3597 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3598 	test_vmcs_addr(name, encoding, align, ignored,
3599 		       skip_beyond_mapped_ram, -1ul);
3600 
3601 	vmcs_write(encoding, orig_val);
3602 }
3603 
3604 /*
3605  * Test a physical address reference in the VMCS, when the corresponding
3606  * feature is enabled and when the corresponding feature is disabled.
3607  */
3608 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3609 				     const char *field_name,
3610 				     const char *control_name, u64 align,
3611 				     bool skip_beyond_mapped_ram,
3612 				     bool control_primary)
3613 {
3614 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3615 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3616 	u64 page_addr;
3617 
3618 	if (control_primary) {
3619 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3620 			return;
3621 	} else {
3622 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3623 			return;
3624 	}
3625 
3626 	page_addr = vmcs_read(field);
3627 
3628 	report_prefix_pushf("%s enabled", control_name);
3629 	if (control_primary) {
3630 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3631 	} else {
3632 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3633 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3634 	}
3635 
3636 	test_vmcs_addr_values(field_name, field, align, false,
3637 			      skip_beyond_mapped_ram, 0, 63);
3638 	report_prefix_pop();
3639 
3640 	report_prefix_pushf("%s disabled", control_name);
3641 	if (control_primary) {
3642 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3643 	} else {
3644 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3645 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3646 	}
3647 
3648 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3649 	report_prefix_pop();
3650 
3651 	vmcs_write(field, page_addr);
3652 	vmcs_write(CPU_EXEC_CTRL0, primary);
3653 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3654 }
3655 
3656 /*
3657  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3658  * each I/O-bitmap address must be 0. Neither address should set any
3659  * bits beyond the processor's physical-address width.
3660  * [Intel SDM]
3661  */
3662 static void test_io_bitmaps(void)
3663 {
3664 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3665 				 "I/O bitmap A", "Use I/O bitmaps",
3666 				 PAGE_SIZE, false, true);
3667 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3668 				 "I/O bitmap B", "Use I/O bitmaps",
3669 				 PAGE_SIZE, false, true);
3670 }
3671 
3672 /*
3673  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3674  * the MSR-bitmap address must be 0. The address should not set any
3675  * bits beyond the processor's physical-address width.
3676  * [Intel SDM]
3677  */
3678 static void test_msr_bitmap(void)
3679 {
3680 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3681 				 "MSR bitmap", "Use MSR bitmaps",
3682 				 PAGE_SIZE, false, true);
3683 }
3684 
3685 /*
3686  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3687  * address must satisfy the following checks:
3688  * - Bits 11:0 of the address must be 0.
3689  * - The address should not set any bits beyond the processor's
3690  *   physical-address width.
3691  * [Intel SDM]
3692  */
3693 static void test_apic_virt_addr(void)
3694 {
3695 	/*
3696 	 * Ensure the processor will never use the virtual-APIC page, since
3697 	 * we will point it to invalid RAM.  Otherwise KVM is puzzled about
3698 	 * what we're trying to achieve and fails vmentry.
3699 	 */
3700 	u32 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
3701 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0 | CPU_CR8_LOAD | CPU_CR8_STORE);
3702 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3703 				 "virtual-APIC address", "Use TPR shadow",
3704 				 PAGE_SIZE, false, true);
3705 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
3706 }
3707 
3708 /*
3709  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3710  * APIC-access address must satisfy the following checks:
3711  *  - Bits 11:0 of the address must be 0.
3712  *  - The address should not set any bits beyond the processor's
3713  *    physical-address width.
3714  * [Intel SDM]
3715  */
3716 static void test_apic_access_addr(void)
3717 {
3718 	void *apic_access_page = alloc_page();
3719 
3720 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3721 
3722 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3723 				 "APIC-access address",
3724 				 "virtualize APIC-accesses", PAGE_SIZE,
3725 				 true, false);
3726 }
3727 
3728 static bool set_bit_pattern(u8 mask, u32 *secondary)
3729 {
3730 	u8 i;
3731 	bool flag = false;
3732 	u32 test_bits[3] = {
3733 		CPU_VIRT_X2APIC,
3734 		CPU_APIC_REG_VIRT,
3735 		CPU_VINTD
3736 	};
3737 
3738         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3739 		if ((mask & (1u << i)) &&
3740 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3741 			*secondary |= test_bits[i];
3742 			flag = true;
3743 		}
3744 	}
3745 
3746 	return (flag);
3747 }
3748 
3749 /*
3750  * If the "use TPR shadow" VM-execution control is 0, the following
3751  * VM-execution controls must also be 0:
3752  * 	- virtualize x2APIC mode
3753  *	- APIC-register virtualization
3754  *	- virtual-interrupt delivery
3755  *    [Intel SDM]
3756  *
3757  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3758  *    "virtualize APIC accesses" VM-execution control must be 0.
3759  *    [Intel SDM]
3760  */
3761 static void test_apic_virtual_ctls(void)
3762 {
3763 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3764 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3765 	u32 primary = saved_primary;
3766 	u32 secondary = saved_secondary;
3767 	bool ctrl = false;
3768 	char str[10] = "disabled";
3769 	u8 i = 0, j;
3770 
3771 	/*
3772 	 * First test
3773 	 */
3774 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3775 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3776 		return;
3777 
3778 	primary |= CPU_SECONDARY;
3779 	primary &= ~CPU_TPR_SHADOW;
3780 	vmcs_write(CPU_EXEC_CTRL0, primary);
3781 
3782 	while (1) {
3783 		for (j = 1; j < 8; j++) {
3784 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3785 			if (primary & CPU_TPR_SHADOW) {
3786 				ctrl = true;
3787 			} else {
3788 				if (! set_bit_pattern(j, &secondary))
3789 					ctrl = true;
3790 				else
3791 					ctrl = false;
3792 			}
3793 
3794 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3795 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3796 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3797 			if (ctrl)
3798 				test_vmx_valid_controls();
3799 			else
3800 				test_vmx_invalid_controls();
3801 			report_prefix_pop();
3802 		}
3803 
3804 		if (i == 1)
3805 			break;
3806 		i++;
3807 
3808 		primary |= CPU_TPR_SHADOW;
3809 		vmcs_write(CPU_EXEC_CTRL0, primary);
3810 		strcpy(str, "enabled");
3811 	}
3812 
3813 	/*
3814 	 * Second test
3815 	 */
3816 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3817 
3818 	primary = saved_primary;
3819 	secondary = saved_secondary;
3820 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3821 		return;
3822 
3823 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3824 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3825 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3826 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3827 	test_vmx_valid_controls();
3828 	report_prefix_pop();
3829 
3830 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3831 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3832 	test_vmx_valid_controls();
3833 	report_prefix_pop();
3834 
3835 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3836 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3837 	test_vmx_invalid_controls();
3838 	report_prefix_pop();
3839 
3840 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3841 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3842 	test_vmx_valid_controls();
3843 	report_prefix_pop();
3844 
3845 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3846 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3847 }
3848 
3849 /*
3850  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3851  * "external-interrupt exiting" VM-execution control must be 1.
3852  * [Intel SDM]
3853  */
3854 static void test_virtual_intr_ctls(void)
3855 {
3856 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3857 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3858 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3859 	u32 primary = saved_primary;
3860 	u32 secondary = saved_secondary;
3861 	u32 pin = saved_pin;
3862 
3863 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3864 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
3865 		return;
3866 
3867 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3868 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
3869 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3870 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
3871 	test_vmx_valid_controls();
3872 	report_prefix_pop();
3873 
3874 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
3875 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3876 	test_vmx_invalid_controls();
3877 	report_prefix_pop();
3878 
3879 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
3880 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
3881 	test_vmx_valid_controls();
3882 	report_prefix_pop();
3883 
3884 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3885 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3886 	test_vmx_invalid_controls();
3887 	report_prefix_pop();
3888 
3889 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3890 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3891 	vmcs_write(PIN_CONTROLS, saved_pin);
3892 }
3893 
3894 static void test_pi_desc_addr(u64 addr, bool ctrl)
3895 {
3896 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
3897 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
3898 	if (ctrl)
3899 		test_vmx_valid_controls();
3900 	else
3901 		test_vmx_invalid_controls();
3902 	report_prefix_pop();
3903 }
3904 
3905 /*
3906  * If the “process posted interrupts†VM-execution control is 1, the
3907  * following must be true:
3908  *
3909  *	- The “virtual-interrupt delivery†VM-execution control is 1.
3910  *	- The “acknowledge interrupt on exit†VM-exit control is 1.
3911  *	- The posted-interrupt notification vector has a value in the
3912  *	- range 0–255 (bits 15:8 are all 0).
3913  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
3914  *	- The posted-interrupt descriptor address does not set any bits
3915  *	  beyond the processor's physical-address width.
3916  * [Intel SDM]
3917  */
3918 static void test_posted_intr(void)
3919 {
3920 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3921 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3922 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3923 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
3924 	u32 primary = saved_primary;
3925 	u32 secondary = saved_secondary;
3926 	u32 pin = saved_pin;
3927 	u32 exit_ctl = exit_ctl_saved;
3928 	u16 vec;
3929 	int i;
3930 
3931 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
3932 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3933 	    (ctrl_exit_rev.clr & EXI_INTA)))
3934 		return;
3935 
3936 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3937 
3938 	/*
3939 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
3940 	 */
3941 	pin |= PIN_POST_INTR;
3942 	vmcs_write(PIN_CONTROLS, pin);
3943 	secondary &= ~CPU_VINTD;
3944 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3945 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
3946 	test_vmx_invalid_controls();
3947 	report_prefix_pop();
3948 
3949 	secondary |= CPU_VINTD;
3950 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3951 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
3952 	test_vmx_invalid_controls();
3953 	report_prefix_pop();
3954 
3955 	exit_ctl &= ~EXI_INTA;
3956 	vmcs_write(EXI_CONTROLS, exit_ctl);
3957 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
3958 	test_vmx_invalid_controls();
3959 	report_prefix_pop();
3960 
3961 	exit_ctl |= EXI_INTA;
3962 	vmcs_write(EXI_CONTROLS, exit_ctl);
3963 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3964 	test_vmx_valid_controls();
3965 	report_prefix_pop();
3966 
3967 	secondary &= ~CPU_VINTD;
3968 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3969 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
3970 	test_vmx_invalid_controls();
3971 	report_prefix_pop();
3972 
3973 	secondary |= CPU_VINTD;
3974 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3975 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3976 	test_vmx_valid_controls();
3977 	report_prefix_pop();
3978 
3979 	/*
3980 	 * Test posted-interrupt notification vector
3981 	 */
3982 	for (i = 0; i < 8; i++) {
3983 		vec = (1ul << i);
3984 		vmcs_write(PINV, vec);
3985 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3986 		test_vmx_valid_controls();
3987 		report_prefix_pop();
3988 	}
3989 	for (i = 8; i < 16; i++) {
3990 		vec = (1ul << i);
3991 		vmcs_write(PINV, vec);
3992 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3993 		test_vmx_invalid_controls();
3994 		report_prefix_pop();
3995 	}
3996 
3997 	vec &= ~(0xff << 8);
3998 	vmcs_write(PINV, vec);
3999 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4000 	test_vmx_valid_controls();
4001 	report_prefix_pop();
4002 
4003 	/*
4004 	 * Test posted-interrupt descriptor addresss
4005 	 */
4006 	for (i = 0; i < 6; i++) {
4007 		test_pi_desc_addr(1ul << i, false);
4008 	}
4009 
4010 	test_pi_desc_addr(0xf0, false);
4011 	test_pi_desc_addr(0xff, false);
4012 	test_pi_desc_addr(0x0f, false);
4013 	test_pi_desc_addr(0x8000, true);
4014 	test_pi_desc_addr(0x00, true);
4015 	test_pi_desc_addr(0xc000, true);
4016 
4017 	test_vmcs_addr_values("process-posted interrupts",
4018 			       POSTED_INTR_DESC_ADDR, 64,
4019 			       false, false, 0, 63);
4020 
4021 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4022 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4023 	vmcs_write(PIN_CONTROLS, saved_pin);
4024 }
4025 
4026 static void test_apic_ctls(void)
4027 {
4028 	test_apic_virt_addr();
4029 	test_apic_access_addr();
4030 	test_apic_virtual_ctls();
4031 	test_virtual_intr_ctls();
4032 	test_posted_intr();
4033 }
4034 
4035 /*
4036  * If the “enable VPID†VM-execution control is 1, the value of the
4037  * of the VPID VM-execution control field must not be 0000H.
4038  * [Intel SDM]
4039  */
4040 static void test_vpid(void)
4041 {
4042 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4043 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4044 	u16 vpid = 0x0000;
4045 	int i;
4046 
4047 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4048 	    (ctrl_cpu_rev[1].clr & CPU_VPID))) {
4049 		printf("Secondary controls and/or VPID not supported\n");
4050 		return;
4051 	}
4052 
4053 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
4054 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
4055 	vmcs_write(VPID, vpid);
4056 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
4057 	test_vmx_valid_controls();
4058 	report_prefix_pop();
4059 
4060 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
4061 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4062 	test_vmx_invalid_controls();
4063 	report_prefix_pop();
4064 
4065 	for (i = 0; i < 16; i++) {
4066 		vpid = (short)1 << i;;
4067 		vmcs_write(VPID, vpid);
4068 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4069 		test_vmx_valid_controls();
4070 		report_prefix_pop();
4071 	}
4072 
4073 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4074 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4075 }
4076 
4077 static void set_vtpr(unsigned vtpr)
4078 {
4079 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
4080 }
4081 
4082 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
4083 {
4084 	bool valid = true;
4085 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4086 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4087 
4088 	if ((primary & CPU_TPR_SHADOW) &&
4089 	    (!(primary & CPU_SECONDARY) ||
4090 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
4091 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
4092 
4093 	set_vtpr(vtpr);
4094 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
4095 	    threshold, (vtpr >> 4) & 0xf);
4096 	if (valid)
4097 		test_vmx_valid_controls();
4098 	else
4099 		test_vmx_invalid_controls();
4100 	report_prefix_pop();
4101 }
4102 
4103 static void test_invalid_event_injection(void)
4104 {
4105 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
4106 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
4107 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
4108 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
4109 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
4110 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4111 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4112 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4113 	u32 cnt;
4114 
4115 	/* Setup */
4116 	report_prefix_push("invalid event injection");
4117 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4118 	vmcs_write(ENT_INST_LEN, 0x00000001);
4119 
4120 	/* The field’s interruption type is not set to a reserved value. */
4121 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4122 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4123 			    "RESERVED interruption type invalid [-]",
4124 			    ent_intr_info);
4125 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4126 	test_vmx_invalid_controls();
4127 	report_prefix_pop();
4128 
4129 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4130 			DE_VECTOR;
4131 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4132 			    "RESERVED interruption type invalid [+]",
4133 			    ent_intr_info);
4134 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4135 	test_vmx_valid_controls();
4136 	report_prefix_pop();
4137 
4138 	/* If the interruption type is other event, the vector is 0. */
4139 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4140 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4141 			    "(OTHER EVENT && vector != 0) invalid [-]",
4142 			    ent_intr_info);
4143 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4144 	test_vmx_invalid_controls();
4145 	report_prefix_pop();
4146 
4147 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4148 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4149 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4150 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4151 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4152 	test_vmx_invalid_controls();
4153 	report_prefix_pop();
4154 
4155 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4156 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4157 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4158 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4159 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4160 	test_vmx_valid_controls();
4161 	report_prefix_pop();
4162 
4163 	/*
4164 	 * If the interruption type
4165 	 * is HW exception, the vector is at most 31.
4166 	 */
4167 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4168 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4169 			    "(HW exception && vector > 31) invalid [-]",
4170 			    ent_intr_info);
4171 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4172 	test_vmx_invalid_controls();
4173 	report_prefix_pop();
4174 
4175 	/*
4176 	 * deliver-error-code is 1 iff either
4177 	 * (a) the "unrestricted guest" VM-execution control is 0
4178 	 * (b) CR0.PE is set.
4179 	 */
4180 
4181 	/* Assert that unrestricted guest is disabled or unsupported */
4182 	assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
4183 	       !(secondary_save & CPU_URG));
4184 
4185 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4186 			GP_VECTOR;
4187 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4188 			    "error code <-> (!URG || prot_mode) [-]",
4189 			    ent_intr_info);
4190 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4191 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4192 	test_vmx_invalid_controls();
4193 	report_prefix_pop();
4194 
4195 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4196 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4197 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4198 			    "error code <-> (!URG || prot_mode) [+]",
4199 			    ent_intr_info);
4200 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4201 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4202 	test_vmx_valid_controls();
4203 	report_prefix_pop();
4204 
4205 	if (enable_unrestricted_guest())
4206 		goto skip_unrestricted_guest;
4207 
4208 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4209 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4210 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4211 			    "error code <-> (!URG || prot_mode) [-]",
4212 			    ent_intr_info);
4213 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4214 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4215 	test_vmx_invalid_controls();
4216 	report_prefix_pop();
4217 
4218 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4219 			GP_VECTOR;
4220 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4221 			    "error code <-> (!URG || prot_mode) [-]",
4222 			    ent_intr_info);
4223 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4224 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4225 	test_vmx_invalid_controls();
4226 	report_prefix_pop();
4227 
4228 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4229 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4230 
4231 skip_unrestricted_guest:
4232 	vmcs_write(GUEST_CR0, guest_cr0_save);
4233 
4234 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4235 	report_prefix_push("error code <-> HW exception");
4236 	for (cnt = 0; cnt < 8; cnt++) {
4237 		u32 exception_type_mask = cnt << 8;
4238 		u32 deliver_error_code_mask =
4239 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4240 			INTR_INFO_DELIVER_CODE_MASK : 0;
4241 
4242 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4243 				exception_type_mask | GP_VECTOR;
4244 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4245 				    ent_intr_info);
4246 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4247 		test_vmx_invalid_controls();
4248 		report_prefix_pop();
4249 	}
4250 	report_prefix_pop();
4251 
4252 	/*
4253 	 * deliver-error-code is 1 iff the the vector
4254 	 * indicates an exception that would normally deliver an error code
4255 	 */
4256 	report_prefix_push("error code <-> vector delivers error code");
4257 	for (cnt = 0; cnt < 32; cnt++) {
4258 		bool has_error_code = false;
4259 		u32 deliver_error_code_mask;
4260 
4261 		switch (cnt) {
4262 		case DF_VECTOR:
4263 		case TS_VECTOR:
4264 		case NP_VECTOR:
4265 		case SS_VECTOR:
4266 		case GP_VECTOR:
4267 		case PF_VECTOR:
4268 		case AC_VECTOR:
4269 			has_error_code = true;
4270 		}
4271 
4272 		/* Negative case */
4273 		deliver_error_code_mask = has_error_code ?
4274 						0 :
4275 						INTR_INFO_DELIVER_CODE_MASK;
4276 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4277 				INTR_TYPE_HARD_EXCEPTION | cnt;
4278 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4279 				    ent_intr_info);
4280 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4281 		test_vmx_invalid_controls();
4282 		report_prefix_pop();
4283 
4284 		/* Positive case */
4285 		deliver_error_code_mask = has_error_code ?
4286 						INTR_INFO_DELIVER_CODE_MASK :
4287 						0;
4288 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4289 				INTR_TYPE_HARD_EXCEPTION | cnt;
4290 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4291 				    ent_intr_info);
4292 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4293 		test_vmx_valid_controls();
4294 		report_prefix_pop();
4295 	}
4296 	report_prefix_pop();
4297 
4298 	/* Reserved bits in the field (30:12) are 0. */
4299 	report_prefix_push("reserved bits clear");
4300 	for (cnt = 12; cnt <= 30; cnt++) {
4301 		ent_intr_info = ent_intr_info_base |
4302 				INTR_INFO_DELIVER_CODE_MASK |
4303 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4304 				(1U << cnt);
4305 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4306 				    ent_intr_info);
4307 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4308 		test_vmx_invalid_controls();
4309 		report_prefix_pop();
4310 	}
4311 	report_prefix_pop();
4312 
4313 	/*
4314 	 * If deliver-error-code is 1
4315 	 * bits 31:16 of the VM-entry exception error-code field are 0.
4316 	 */
4317 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4318 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4319 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4320 			    "VM-entry exception error code[31:16] clear",
4321 			    ent_intr_info);
4322 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4323 	for (cnt = 16; cnt <= 31; cnt++) {
4324 		ent_intr_err = 1U << cnt;
4325 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4326 				    ent_intr_err);
4327 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4328 		test_vmx_invalid_controls();
4329 		report_prefix_pop();
4330 	}
4331 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4332 	report_prefix_pop();
4333 
4334 	/*
4335 	 * If the interruption type is software interrupt, software exception,
4336 	 * or privileged software exception, the VM-entry instruction-length
4337 	 * field is in the range 0–15.
4338 	 */
4339 
4340 	for (cnt = 0; cnt < 3; cnt++) {
4341 		switch (cnt) {
4342 		case 0:
4343 			ent_intr_info = ent_intr_info_base |
4344 					INTR_TYPE_SOFT_INTR;
4345 			break;
4346 		case 1:
4347 			ent_intr_info = ent_intr_info_base |
4348 					INTR_TYPE_SOFT_EXCEPTION;
4349 			break;
4350 		case 2:
4351 			ent_intr_info = ent_intr_info_base |
4352 					INTR_TYPE_PRIV_SW_EXCEPTION;
4353 			break;
4354 		}
4355 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4356 				    "VM-entry instruction-length check",
4357 				    ent_intr_info);
4358 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4359 
4360 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4361 		ent_intr_len = -1;
4362 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4363 				    ent_intr_len);
4364 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4365 		test_vmx_invalid_controls();
4366 		report_prefix_pop();
4367 
4368 		/* Instruction length set to 16 should fail */
4369 		ent_intr_len = 0x00000010;
4370 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4371 				    ent_intr_len);
4372 		vmcs_write(ENT_INST_LEN, 0x00000010);
4373 		test_vmx_invalid_controls();
4374 		report_prefix_pop();
4375 
4376 		report_prefix_pop();
4377 	}
4378 
4379 	/* Cleanup */
4380 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4381 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4382 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4383 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4384 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4385 	vmcs_write(GUEST_CR0, guest_cr0_save);
4386 	report_prefix_pop();
4387 }
4388 
4389 /*
4390  * Test interesting vTPR values for a given TPR threshold.
4391  */
4392 static void test_vtpr_values(unsigned threshold)
4393 {
4394 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4395 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4396 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4397 }
4398 
4399 static void try_tpr_threshold(unsigned threshold)
4400 {
4401 	bool valid = true;
4402 
4403 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4404 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4405 
4406 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4407 	    (secondary & CPU_VINTD)))
4408 		valid = !(threshold >> 4);
4409 
4410 	set_vtpr(-1);
4411 	vmcs_write(TPR_THRESHOLD, threshold);
4412 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4413 	if (valid)
4414 		test_vmx_valid_controls();
4415 	else
4416 		test_vmx_invalid_controls();
4417 	report_prefix_pop();
4418 
4419 	if (valid)
4420 		test_vtpr_values(threshold);
4421 }
4422 
4423 /*
4424  * Test interesting TPR threshold values.
4425  */
4426 static void test_tpr_threshold_values(void)
4427 {
4428 	unsigned i;
4429 
4430 	for (i = 0; i < 0x10; i++)
4431 		try_tpr_threshold(i);
4432 	for (i = 4; i < 32; i++)
4433 		try_tpr_threshold(1u << i);
4434 	try_tpr_threshold(-1u);
4435 	try_tpr_threshold(0x7fffffff);
4436 }
4437 
4438 /*
4439  * This test covers the following two VM entry checks:
4440  *
4441  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4442  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4443  *         31:4 of the TPR threshold VM-execution control field must
4444 	   be 0.
4445  *         [Intel SDM]
4446  *
4447  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4448  *          "virtual-interrupt delivery" VM-execution control is 0
4449  *          and the "virtualize APIC accesses" VM-execution control
4450  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4451  *          control field must not be greater than the value of bits
4452  *          7:4 of VTPR.
4453  *          [Intel SDM]
4454  */
4455 static void test_tpr_threshold(void)
4456 {
4457 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4458 	u64 apic_virt_addr = vmcs_read(APIC_VIRT_ADDR);
4459 	u64 threshold = vmcs_read(TPR_THRESHOLD);
4460 	void *virtual_apic_page;
4461 
4462 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4463 		return;
4464 
4465 	virtual_apic_page = alloc_page();
4466 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4467 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4468 
4469 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4470 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4471 	test_tpr_threshold_values();
4472 	report_prefix_pop();
4473 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4474 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4475 	test_tpr_threshold_values();
4476 	report_prefix_pop();
4477 
4478 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4479 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES))))
4480 		goto out;
4481 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4482 
4483 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4484 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4485 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4486 		test_tpr_threshold_values();
4487 		report_prefix_pop();
4488 
4489 		vmcs_write(CPU_EXEC_CTRL0,
4490 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4491 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4492 		test_tpr_threshold_values();
4493 		report_prefix_pop();
4494 	}
4495 
4496 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4497 		vmcs_write(CPU_EXEC_CTRL0,
4498 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4499 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4500 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4501 		test_tpr_threshold_values();
4502 		report_prefix_pop();
4503 
4504 		vmcs_write(CPU_EXEC_CTRL0,
4505 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4506 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4507 		test_tpr_threshold_values();
4508 		report_prefix_pop();
4509 	}
4510 
4511 	if ((ctrl_cpu_rev[1].clr &
4512 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4513 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4514 		vmcs_write(CPU_EXEC_CTRL0,
4515 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4516 		vmcs_write(CPU_EXEC_CTRL1,
4517 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4518 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4519 		test_tpr_threshold_values();
4520 		report_prefix_pop();
4521 
4522 		vmcs_write(CPU_EXEC_CTRL0,
4523 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4524 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4525 		test_tpr_threshold_values();
4526 		report_prefix_pop();
4527 	}
4528 
4529 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4530 out:
4531 	vmcs_write(TPR_THRESHOLD, threshold);
4532 	vmcs_write(APIC_VIRT_ADDR, apic_virt_addr);
4533 	vmcs_write(CPU_EXEC_CTRL0, primary);
4534 }
4535 
4536 /*
4537  * This test verifies the following two vmentry checks:
4538  *
4539  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4540  *  VM-execution control must be 0.
4541  *  [Intel SDM]
4542  *
4543  *  If the “virtual NMIs” VM-execution control is 0, the “NMI-window
4544  *  exiting” VM-execution control must be 0.
4545  *  [Intel SDM]
4546  */
4547 static void test_nmi_ctrls(void)
4548 {
4549 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4550 
4551 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4552 	    (PIN_NMI | PIN_VIRT_NMI)) {
4553 		printf("NMI exiting and Virtual NMIs are not supported !\n");
4554 		return;
4555 	}
4556 
4557 	/* Save the controls so that we can restore them after our tests */
4558 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4559 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4560 
4561 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4562 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4563 
4564 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4565 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4566 	test_vmx_valid_controls();
4567 	report_prefix_pop();
4568 
4569 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4570 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4571 	test_vmx_invalid_controls();
4572 	report_prefix_pop();
4573 
4574 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4575 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4576 	test_vmx_valid_controls();
4577 	report_prefix_pop();
4578 
4579 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4580 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4581 	test_vmx_valid_controls();
4582 	report_prefix_pop();
4583 
4584 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4585 		report_info("NMI-window exiting is not supported, skipping...");
4586 		goto done;
4587 	}
4588 
4589 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4590 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4591 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4592 	test_vmx_invalid_controls();
4593 	report_prefix_pop();
4594 
4595 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4596 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4597 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4598 	test_vmx_valid_controls();
4599 	report_prefix_pop();
4600 
4601 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4602 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4603 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4604 	test_vmx_valid_controls();
4605 	report_prefix_pop();
4606 
4607 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4608 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4609 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4610 	test_vmx_valid_controls();
4611 	report_prefix_pop();
4612 
4613 	/* Restore the controls to their original values */
4614 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4615 done:
4616 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4617 }
4618 
4619 static void test_eptp_ad_bit(u64 eptp, bool ctrl)
4620 {
4621 	vmcs_write(EPTP, eptp);
4622 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4623 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4624 	if (ctrl)
4625 		test_vmx_valid_controls();
4626 	else
4627 		test_vmx_invalid_controls();
4628 	report_prefix_pop();
4629 
4630 }
4631 
4632 /*
4633  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4634  *    control field must satisfy the following checks:
4635  *
4636  *     - The EPT memory type (bits 2:0) must be a value supported by the
4637  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4638  *     - Bits 5:3 (1 less than the EPT page-walk length) must be 3,
4639  *	 indicating an EPT page-walk length of 4.
4640  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4641  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4642  *	 indicating that the processor does not support accessed and dirty
4643  *	 dirty flags for EPT.
4644  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4645  *	 physical-address width) must all be 0.
4646  *
4647  * 2. If the "unrestricted guest" VM-execution control is 1, the
4648  *    "enable EPT" VM-execution control must also be 1.
4649  */
4650 static void test_ept_eptp(void)
4651 {
4652 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4653 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4654 	u64 eptp_saved = vmcs_read(EPTP);
4655 	u32 primary = primary_saved;
4656 	u32 secondary = secondary_saved;
4657 	u64 msr, eptp = eptp_saved;
4658 	bool un_cache = false;
4659 	bool wr_bk = false;
4660 	bool ctrl;
4661 	u32 i, maxphysaddr;
4662 	u64 j, resv_bits_mask = 0;
4663 
4664 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4665 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4666 		printf("\"CPU secondary\" and/or \"enable EPT\" execution controls are not supported !\n");
4667 		return;
4668 	}
4669 
4670 	/*
4671 	 * Memory type (bits 2:0)
4672 	 */
4673 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
4674 	if (msr & EPT_CAP_UC)
4675 		un_cache = true;
4676 	if (msr & EPT_CAP_WB)
4677 		wr_bk = true;
4678 
4679 	primary |= CPU_SECONDARY;
4680 	vmcs_write(CPU_EXEC_CTRL0, primary);
4681 	secondary |= CPU_EPT;
4682 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4683 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4684 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4685 	vmcs_write(EPTP, eptp);
4686 
4687 	for (i = 0; i < 8; i++) {
4688 		if (i == 0) {
4689 			if (un_cache) {
4690 				report_info("EPT paging structure memory-type is Un-cacheable\n");
4691 				ctrl = true;
4692 			} else {
4693 				ctrl = false;
4694 			}
4695 		} else if (i == 6) {
4696 			if (wr_bk) {
4697 				report_info("EPT paging structure memory-type is Write-back\n");
4698 				ctrl = true;
4699 			} else {
4700 				ctrl = false;
4701 			}
4702 		} else {
4703 			ctrl = false;
4704 		}
4705 
4706 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4707 		vmcs_write(EPTP, eptp);
4708 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4709 		    eptp & EPT_MEM_TYPE_MASK);
4710 		if (ctrl)
4711 			test_vmx_valid_controls();
4712 		else
4713 			test_vmx_invalid_controls();
4714 		report_prefix_pop();
4715 	}
4716 
4717 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4718 
4719 	/*
4720 	 * Page walk length (bits 5:3)
4721 	 */
4722 	for (i = 0; i < 8; i++) {
4723 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4724 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4725 		if (i == 3)
4726 			ctrl = true;
4727 		else
4728 			ctrl = false;
4729 
4730 		vmcs_write(EPTP, eptp);
4731 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4732 		    eptp & EPTP_PG_WALK_LEN_MASK);
4733 		if (ctrl)
4734 			test_vmx_valid_controls();
4735 		else
4736 			test_vmx_invalid_controls();
4737 		report_prefix_pop();
4738 	}
4739 
4740 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4741 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4742 
4743 	/*
4744 	 * Accessed and dirty flag (bit 6)
4745 	 */
4746 	if (msr & EPT_CAP_AD_FLAG) {
4747 		report_info("Processor supports accessed and dirty flag");
4748 		eptp &= ~EPTP_AD_FLAG;
4749 		test_eptp_ad_bit(eptp, true);
4750 
4751 		eptp |= EPTP_AD_FLAG;
4752 		test_eptp_ad_bit(eptp, true);
4753 	} else {
4754 		report_info("Processor does not supports accessed and dirty flag");
4755 		eptp &= ~EPTP_AD_FLAG;
4756 		test_eptp_ad_bit(eptp, true);
4757 
4758 		eptp |= EPTP_AD_FLAG;
4759 		test_eptp_ad_bit(eptp, false);
4760 	}
4761 
4762 	/*
4763 	 * Reserved bits [11:7] and [63:N]
4764 	 */
4765 	for (i = 0; i < 32; i++) {
4766 		eptp = (eptp &
4767 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4768 		    (i << EPTP_RESERV_BITS_SHIFT);
4769 		vmcs_write(EPTP, eptp);
4770 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4771 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4772 		    EPTP_RESERV_BITS_MASK);
4773 		if (i == 0)
4774 			test_vmx_valid_controls();
4775 		else
4776 			test_vmx_invalid_controls();
4777 		report_prefix_pop();
4778 	}
4779 
4780 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4781 
4782 	maxphysaddr = cpuid_maxphyaddr();
4783 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4784 		resv_bits_mask |= 1ul << i;
4785 	}
4786 
4787 	for (j = maxphysaddr - 1; j <= 63; j++) {
4788 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4789 		    (j < maxphysaddr ? 0 : 1ul << j);
4790 		vmcs_write(EPTP, eptp);
4791 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4792 		    (eptp >> maxphysaddr) & resv_bits_mask);
4793 		if (j < maxphysaddr)
4794 			test_vmx_valid_controls();
4795 		else
4796 			test_vmx_invalid_controls();
4797 		report_prefix_pop();
4798 	}
4799 
4800 	secondary &= ~(CPU_EPT | CPU_URG);
4801 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4802 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4803 	test_vmx_valid_controls();
4804 	report_prefix_pop();
4805 
4806 	if (!(ctrl_cpu_rev[1].clr & CPU_URG))
4807 		goto skip_unrestricted_guest;
4808 
4809 	secondary |= CPU_URG;
4810 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4811 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4812 	test_vmx_invalid_controls();
4813 	report_prefix_pop();
4814 
4815 	secondary |= CPU_EPT;
4816 	setup_dummy_ept();
4817 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4818 	test_vmx_valid_controls();
4819 	report_prefix_pop();
4820 
4821 skip_unrestricted_guest:
4822 	secondary &= ~CPU_URG;
4823 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4824 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4825 	test_vmx_valid_controls();
4826 	report_prefix_pop();
4827 
4828 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4829 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4830 	vmcs_write(EPTP, eptp_saved);
4831 }
4832 
4833 /*
4834  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4835  * VM-execution control must also be 1. In addition, the PML address
4836  * must satisfy the following checks:
4837  *
4838  *    * Bits 11:0 of the address must be 0.
4839  *    * The address should not set any bits beyond the processor's
4840  *	physical-address width.
4841  *
4842  *  [Intel SDM]
4843  */
4844 static void test_pml(void)
4845 {
4846 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4847 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4848 	u32 primary = primary_saved;
4849 	u32 secondary = secondary_saved;
4850 
4851 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4852 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
4853 		printf("\"Secondary execution\" control or \"enable EPT\" control or \"enable PML\" control is not supported !\n");
4854 		return;
4855 	}
4856 
4857 	primary |= CPU_SECONDARY;
4858 	vmcs_write(CPU_EXEC_CTRL0, primary);
4859 	secondary &= ~(CPU_PML | CPU_EPT);
4860 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4861 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
4862 	test_vmx_valid_controls();
4863 	report_prefix_pop();
4864 
4865 	secondary |= CPU_PML;
4866 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4867 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
4868 	test_vmx_invalid_controls();
4869 	report_prefix_pop();
4870 
4871 	secondary |= CPU_EPT;
4872 	setup_dummy_ept();
4873 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
4874 	test_vmx_valid_controls();
4875 	report_prefix_pop();
4876 
4877 	secondary &= ~CPU_PML;
4878 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4879 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
4880 	test_vmx_valid_controls();
4881 	report_prefix_pop();
4882 
4883 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
4884 				 PAGE_SIZE, false, false);
4885 
4886 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4887 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4888 }
4889 
4890  /*
4891  * If the "activate VMX-preemption timer" VM-execution control is 0, the
4892  * the "save VMX-preemption timer value" VM-exit control must also be 0.
4893  *
4894  *  [Intel SDM]
4895  */
4896 static void test_vmx_preemption_timer(void)
4897 {
4898 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4899 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
4900 	u32 pin = saved_pin;
4901 	u32 exit = saved_exit;
4902 
4903 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
4904 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
4905 		printf("\"Save-VMX-preemption-timer\" control and/or \"Enable-VMX-preemption-timer\" control is not supported\n");
4906 		return;
4907 	}
4908 
4909 	pin |= PIN_PREEMPT;
4910 	vmcs_write(PIN_CONTROLS, pin);
4911 	exit &= ~EXI_SAVE_PREEMPT;
4912 	vmcs_write(EXI_CONTROLS, exit);
4913 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
4914 	test_vmx_valid_controls();
4915 	report_prefix_pop();
4916 
4917 	exit |= EXI_SAVE_PREEMPT;
4918 	vmcs_write(EXI_CONTROLS, exit);
4919 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
4920 	test_vmx_valid_controls();
4921 	report_prefix_pop();
4922 
4923 	pin &= ~PIN_PREEMPT;
4924 	vmcs_write(PIN_CONTROLS, pin);
4925 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
4926 	test_vmx_invalid_controls();
4927 	report_prefix_pop();
4928 
4929 	exit &= ~EXI_SAVE_PREEMPT;
4930 	vmcs_write(EXI_CONTROLS, exit);
4931 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
4932 	test_vmx_valid_controls();
4933 	report_prefix_pop();
4934 
4935 	vmcs_write(PIN_CONTROLS, saved_pin);
4936 	vmcs_write(EXI_CONTROLS, saved_exit);
4937 }
4938 
4939 /*
4940  * Tests for VM-execution control fields
4941  */
4942 static void test_vm_execution_ctls(void)
4943 {
4944 	test_pin_based_ctls();
4945 	test_primary_processor_based_ctls();
4946 	test_secondary_processor_based_ctls();
4947 	test_cr3_targets();
4948 	test_io_bitmaps();
4949 	test_msr_bitmap();
4950 	test_apic_ctls();
4951 	test_tpr_threshold();
4952 	test_nmi_ctrls();
4953 	test_pml();
4954 	test_vpid();
4955 	test_ept_eptp();
4956 	test_vmx_preemption_timer();
4957 }
4958 
4959  /*
4960   * The following checks are performed for the VM-entry MSR-load address if
4961   * the VM-entry MSR-load count field is non-zero:
4962   *
4963   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
4964   *      The address should not set any bits beyond the processor’s
4965   *      physical-address width.
4966   *
4967   *    - The address of the last byte in the VM-entry MSR-load area
4968   *      should not set any bits beyond the processor’s physical-address
4969   *      width. The address of this last byte is VM-entry MSR-load address
4970   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
4971   *      uses more bits than the processor’s physical-address width.)
4972   *
4973   *
4974   *  [Intel SDM]
4975   */
4976 static void test_entry_msr_load(void)
4977 {
4978 	entry_msr_load = alloc_page();
4979 	u64 tmp;
4980 	u32 entry_msr_ld_cnt = 1;
4981 	int i;
4982 	u32 addr_len = 64;
4983 
4984 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
4985 
4986 	/* Check first 4 bits of VM-entry MSR-load address */
4987 	for (i = 0; i < 4; i++) {
4988 		tmp = (u64)entry_msr_load | 1ull << i;
4989 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
4990 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
4991 				    tmp & 0xf);
4992 		test_vmx_invalid_controls();
4993 		report_prefix_pop();
4994 	}
4995 
4996 	if (basic.val & (1ul << 48))
4997 		addr_len = 32;
4998 
4999 	test_vmcs_addr_values("VM-entry-MSR-load address",
5000 				ENTER_MSR_LD_ADDR, 16, false, false,
5001 				4, addr_len - 1);
5002 
5003 	/*
5004 	 * Check last byte of VM-entry MSR-load address
5005 	 */
5006 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
5007 
5008 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5009 							i < 64; i++) {
5010 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
5011 			1ul << i;
5012 		vmcs_write(ENTER_MSR_LD_ADDR,
5013 			   tmp - (entry_msr_ld_cnt * 16 - 1));
5014 		test_vmx_invalid_controls();
5015 	}
5016 
5017 	vmcs_write(ENT_MSR_LD_CNT, 2);
5018 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5019 	test_vmx_invalid_controls();
5020 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5021 	test_vmx_valid_controls();
5022 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5023 	test_vmx_valid_controls();
5024 }
5025 
5026 static struct vmx_state_area_test_data {
5027 	u32 msr;
5028 	u64 exp;
5029 	bool enabled;
5030 } vmx_state_area_test_data;
5031 
5032 static void guest_state_test_main(void)
5033 {
5034 	u64 obs;
5035 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
5036 
5037 	while (1) {
5038 		if (vmx_get_test_stage() == 2)
5039 			break;
5040 
5041 		if (data->enabled) {
5042 			obs = rdmsr(data->msr);
5043 			report("Guest state is 0x%lx (expected 0x%lx)",
5044 			       data->exp == obs, obs, data->exp);
5045 		}
5046 
5047 		vmcall();
5048 	}
5049 
5050 	asm volatile("fnop");
5051 }
5052 
5053 static void advance_guest_state_test(void)
5054 {
5055 	u32 reason = vmcs_read(EXI_REASON);
5056 	if (! (reason & 0x80000000)) {
5057 		u64 guest_rip = vmcs_read(GUEST_RIP);
5058 		u32 insn_len = vmcs_read(EXI_INST_LEN);
5059 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
5060 	}
5061 }
5062 
5063 static void report_guest_state_test(const char *test, u32 xreason,
5064 				    u64 field, const char * field_name)
5065 {
5066 	u32 reason = vmcs_read(EXI_REASON);
5067 
5068 	report("%s, %s %lx", reason == xreason, test, field_name, field);
5069 	advance_guest_state_test();
5070 }
5071 
5072 /*
5073  * Tests for VM-entry control fields
5074  */
5075 static void test_vm_entry_ctls(void)
5076 {
5077 	test_invalid_event_injection();
5078 	test_entry_msr_load();
5079 }
5080 
5081 /*
5082  * The following checks are performed for the VM-exit MSR-store address if
5083  * the VM-exit MSR-store count field is non-zero:
5084  *
5085  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
5086  *      The address should not set any bits beyond the processor’s
5087  *      physical-address width.
5088  *
5089  *    - The address of the last byte in the VM-exit MSR-store area
5090  *      should not set any bits beyond the processor’s physical-address
5091  *      width. The address of this last byte is VM-exit MSR-store address
5092  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5093  *      uses more bits than the processor’s physical-address width.)
5094  *
5095  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
5096  * in the range 63:32.
5097  *
5098  *  [Intel SDM]
5099  */
5100 static void test_exit_msr_store(void)
5101 {
5102 	exit_msr_store = alloc_page();
5103 	u64 tmp;
5104 	u32 exit_msr_st_cnt = 1;
5105 	int i;
5106 	u32 addr_len = 64;
5107 
5108 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
5109 
5110 	/* Check first 4 bits of VM-exit MSR-store address */
5111 	for (i = 0; i < 4; i++) {
5112 		tmp = (u64)exit_msr_store | 1ull << i;
5113 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
5114 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
5115 				    tmp & 0xf);
5116 		test_vmx_invalid_controls();
5117 		report_prefix_pop();
5118 	}
5119 
5120 	if (basic.val & (1ul << 48))
5121 		addr_len = 32;
5122 
5123 	test_vmcs_addr_values("VM-exit-MSR-store address",
5124 				EXIT_MSR_ST_ADDR, 16, false, false,
5125 				4, addr_len - 1);
5126 
5127 	/*
5128 	 * Check last byte of VM-exit MSR-store address
5129 	 */
5130 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
5131 
5132 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5133 							i < 64; i++) {
5134 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
5135 			1ul << i;
5136 		vmcs_write(EXIT_MSR_ST_ADDR,
5137 			   tmp - (exit_msr_st_cnt * 16 - 1));
5138 		test_vmx_invalid_controls();
5139 	}
5140 
5141 	vmcs_write(EXI_MSR_ST_CNT, 2);
5142 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5143 	test_vmx_invalid_controls();
5144 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5145 	test_vmx_valid_controls();
5146 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5147 	test_vmx_valid_controls();
5148 }
5149 
5150 /*
5151  * Tests for VM-exit controls
5152  */
5153 static void test_vm_exit_ctls(void)
5154 {
5155 	test_exit_msr_store();
5156 }
5157 
5158 /*
5159  * Check that the virtual CPU checks all of the VMX controls as
5160  * documented in the Intel SDM.
5161  */
5162 static void vmx_controls_test(void)
5163 {
5164 	/*
5165 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
5166 	 * fail due to invalid guest state, should we make it that
5167 	 * far.
5168 	 */
5169 	vmcs_write(GUEST_RFLAGS, 0);
5170 
5171 	test_vm_execution_ctls();
5172 	test_vm_exit_ctls();
5173 	test_vm_entry_ctls();
5174 }
5175 
5176 struct apic_reg_virt_config {
5177 	bool apic_register_virtualization;
5178 	bool use_tpr_shadow;
5179 	bool virtualize_apic_accesses;
5180 	bool virtualize_x2apic_mode;
5181 	bool activate_secondary_controls;
5182 };
5183 
5184 struct apic_reg_test {
5185 	const char *name;
5186 	struct apic_reg_virt_config apic_reg_virt_config;
5187 };
5188 
5189 struct apic_reg_virt_expectation {
5190 	enum Reason rd_exit_reason;
5191 	enum Reason wr_exit_reason;
5192 	u32 val;
5193 	u32 (*virt_fn)(u32);
5194 
5195 	/*
5196 	 * If false, accessing the APIC access address from L2 is treated as a
5197 	 * normal memory operation, rather than triggering virtualization.
5198 	 */
5199 	bool virtualize_apic_accesses;
5200 };
5201 
5202 static u32 apic_virt_identity(u32 val)
5203 {
5204 	return val;
5205 }
5206 
5207 static u32 apic_virt_nibble1(u32 val)
5208 {
5209 	return val & 0xf0;
5210 }
5211 
5212 static u32 apic_virt_byte3(u32 val)
5213 {
5214 	return val & (0xff << 24);
5215 }
5216 
5217 static bool apic_reg_virt_exit_expectation(
5218 	u32 reg, struct apic_reg_virt_config *config,
5219 	struct apic_reg_virt_expectation *expectation)
5220 {
5221 	/* Good configs, where some L2 APIC accesses are virtualized. */
5222 	bool virtualize_apic_accesses_only =
5223 		config->virtualize_apic_accesses &&
5224 		!config->use_tpr_shadow &&
5225 		!config->apic_register_virtualization &&
5226 		!config->virtualize_x2apic_mode &&
5227 		config->activate_secondary_controls;
5228 	bool virtualize_apic_accesses_and_use_tpr_shadow =
5229 		config->virtualize_apic_accesses &&
5230 		config->use_tpr_shadow &&
5231 		!config->apic_register_virtualization &&
5232 		!config->virtualize_x2apic_mode &&
5233 		config->activate_secondary_controls;
5234 	bool apic_register_virtualization =
5235 		config->virtualize_apic_accesses &&
5236 		config->use_tpr_shadow &&
5237 		config->apic_register_virtualization &&
5238 		!config->virtualize_x2apic_mode &&
5239 		config->activate_secondary_controls;
5240 
5241 	expectation->val = MAGIC_VAL_1;
5242 	expectation->virt_fn = apic_virt_identity;
5243 	expectation->virtualize_apic_accesses =
5244 		config->virtualize_apic_accesses &&
5245 		config->activate_secondary_controls;
5246 	if (virtualize_apic_accesses_only) {
5247 		expectation->rd_exit_reason = VMX_APIC_ACCESS;
5248 		expectation->wr_exit_reason = VMX_APIC_ACCESS;
5249 	} else if (virtualize_apic_accesses_and_use_tpr_shadow) {
5250 		switch (reg) {
5251 		case APIC_TASKPRI:
5252 			expectation->rd_exit_reason = VMX_VMCALL;
5253 			expectation->wr_exit_reason = VMX_VMCALL;
5254 			expectation->virt_fn = apic_virt_nibble1;
5255 			break;
5256 		default:
5257 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5258 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5259 		}
5260 	} else if (apic_register_virtualization) {
5261 		expectation->rd_exit_reason = VMX_VMCALL;
5262 
5263 		switch (reg) {
5264 		case APIC_ID:
5265 		case APIC_EOI:
5266 		case APIC_LDR:
5267 		case APIC_DFR:
5268 		case APIC_SPIV:
5269 		case APIC_ESR:
5270 		case APIC_ICR:
5271 		case APIC_LVTT:
5272 		case APIC_LVTTHMR:
5273 		case APIC_LVTPC:
5274 		case APIC_LVT0:
5275 		case APIC_LVT1:
5276 		case APIC_LVTERR:
5277 		case APIC_TMICT:
5278 		case APIC_TDCR:
5279 			expectation->wr_exit_reason = VMX_APIC_WRITE;
5280 			break;
5281 		case APIC_LVR:
5282 		case APIC_ISR ... APIC_ISR + 0x70:
5283 		case APIC_TMR ... APIC_TMR + 0x70:
5284 		case APIC_IRR ... APIC_IRR + 0x70:
5285 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5286 			break;
5287 		case APIC_TASKPRI:
5288 			expectation->wr_exit_reason = VMX_VMCALL;
5289 			expectation->virt_fn = apic_virt_nibble1;
5290 			break;
5291 		case APIC_ICR2:
5292 			expectation->wr_exit_reason = VMX_VMCALL;
5293 			expectation->virt_fn = apic_virt_byte3;
5294 			break;
5295 		default:
5296 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5297 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5298 		}
5299 	} else if (!expectation->virtualize_apic_accesses) {
5300 		/*
5301 		 * No APIC registers are directly virtualized. This includes
5302 		 * VTPR, which can be virtualized through MOV to/from CR8 via
5303 		 * the use TPR shadow control, but not through directly
5304 		 * accessing VTPR.
5305 		 */
5306 		expectation->rd_exit_reason = VMX_VMCALL;
5307 		expectation->wr_exit_reason = VMX_VMCALL;
5308 	} else {
5309 		printf("Cannot parse APIC register virtualization config:\n"
5310 		       "\tvirtualize_apic_accesses: %d\n"
5311 		       "\tuse_tpr_shadow: %d\n"
5312 		       "\tapic_register_virtualization: %d\n"
5313 		       "\tvirtualize_x2apic_mode: %d\n"
5314 		       "\tactivate_secondary_controls: %d\n",
5315 		       config->virtualize_apic_accesses,
5316 		       config->use_tpr_shadow,
5317 		       config->apic_register_virtualization,
5318 		       config->virtualize_x2apic_mode,
5319 		       config->activate_secondary_controls);
5320 
5321 		return false;
5322 	}
5323 
5324 	return true;
5325 }
5326 
5327 struct apic_reg_test apic_reg_tests[] = {
5328 	/* Good configs, where some L2 APIC accesses are virtualized. */
5329 	{
5330 		.name = "Virtualize APIC accesses",
5331 		.apic_reg_virt_config = {
5332 			.virtualize_apic_accesses = true,
5333 			.use_tpr_shadow = false,
5334 			.apic_register_virtualization = false,
5335 			.virtualize_x2apic_mode = false,
5336 			.activate_secondary_controls = true,
5337 		},
5338 	},
5339 	{
5340 		.name = "Virtualize APIC accesses + Use TPR shadow",
5341 		.apic_reg_virt_config = {
5342 			.virtualize_apic_accesses = true,
5343 			.use_tpr_shadow = true,
5344 			.apic_register_virtualization = false,
5345 			.virtualize_x2apic_mode = false,
5346 			.activate_secondary_controls = true,
5347 		},
5348 	},
5349 	{
5350 		.name = "APIC-register virtualization",
5351 		.apic_reg_virt_config = {
5352 			.virtualize_apic_accesses = true,
5353 			.use_tpr_shadow = true,
5354 			.apic_register_virtualization = true,
5355 			.virtualize_x2apic_mode = false,
5356 			.activate_secondary_controls = true,
5357 		},
5358 	},
5359 
5360 	/*
5361 	 * Test that the secondary processor-based VM-execution controls are
5362 	 * correctly ignored when "activate secondary controls" is disabled.
5363 	 */
5364 	{
5365 		.name = "Activate secondary controls off",
5366 		.apic_reg_virt_config = {
5367 			.virtualize_apic_accesses = true,
5368 			.use_tpr_shadow = false,
5369 			.apic_register_virtualization = true,
5370 			.virtualize_x2apic_mode = true,
5371 			.activate_secondary_controls = false,
5372 		},
5373 	},
5374 	{
5375 		.name = "Activate secondary controls off + Use TPR shadow",
5376 		.apic_reg_virt_config = {
5377 			.virtualize_apic_accesses = true,
5378 			.use_tpr_shadow = true,
5379 			.apic_register_virtualization = true,
5380 			.virtualize_x2apic_mode = true,
5381 			.activate_secondary_controls = false,
5382 		},
5383 	},
5384 
5385 	/*
5386 	 * Test that the APIC access address is treated like an arbitrary memory
5387 	 * address when "virtualize APIC accesses" is disabled.
5388 	 */
5389 	{
5390 		.name = "Virtualize APIC accesses off + Use TPR shadow",
5391 		.apic_reg_virt_config = {
5392 			.virtualize_apic_accesses = false,
5393 			.use_tpr_shadow = true,
5394 			.apic_register_virtualization = true,
5395 			.virtualize_x2apic_mode = true,
5396 			.activate_secondary_controls = true,
5397 		},
5398 	},
5399 
5400 	/*
5401 	 * Test that VM entry fails due to invalid controls when
5402 	 * "APIC-register virtualization" is enabled while "use TPR shadow" is
5403 	 * disabled.
5404 	 */
5405 	{
5406 		.name = "APIC-register virtualization + Use TPR shadow off",
5407 		.apic_reg_virt_config = {
5408 			.virtualize_apic_accesses = true,
5409 			.use_tpr_shadow = false,
5410 			.apic_register_virtualization = true,
5411 			.virtualize_x2apic_mode = false,
5412 			.activate_secondary_controls = true,
5413 		},
5414 	},
5415 
5416 	/*
5417 	 * Test that VM entry fails due to invalid controls when
5418 	 * "Virtualize x2APIC mode" is enabled while "use TPR shadow" is
5419 	 * disabled.
5420 	 */
5421 	{
5422 		.name = "Virtualize x2APIC mode + Use TPR shadow off",
5423 		.apic_reg_virt_config = {
5424 			.virtualize_apic_accesses = false,
5425 			.use_tpr_shadow = false,
5426 			.apic_register_virtualization = false,
5427 			.virtualize_x2apic_mode = true,
5428 			.activate_secondary_controls = true,
5429 		},
5430 	},
5431 	{
5432 		.name = "Virtualize x2APIC mode + Use TPR shadow off v2",
5433 		.apic_reg_virt_config = {
5434 			.virtualize_apic_accesses = false,
5435 			.use_tpr_shadow = false,
5436 			.apic_register_virtualization = true,
5437 			.virtualize_x2apic_mode = true,
5438 			.activate_secondary_controls = true,
5439 		},
5440 	},
5441 
5442 	/*
5443 	 * Test that VM entry fails due to invalid controls when
5444 	 * "virtualize x2APIC mode" is enabled while "virtualize APIC accesses"
5445 	 * is enabled.
5446 	 */
5447 	{
5448 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses",
5449 		.apic_reg_virt_config = {
5450 			.virtualize_apic_accesses = true,
5451 			.use_tpr_shadow = true,
5452 			.apic_register_virtualization = false,
5453 			.virtualize_x2apic_mode = true,
5454 			.activate_secondary_controls = true,
5455 		},
5456 	},
5457 	{
5458 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses v2",
5459 		.apic_reg_virt_config = {
5460 			.virtualize_apic_accesses = true,
5461 			.use_tpr_shadow = true,
5462 			.apic_register_virtualization = true,
5463 			.virtualize_x2apic_mode = true,
5464 			.activate_secondary_controls = true,
5465 		},
5466 	},
5467 };
5468 
5469 enum Apic_op {
5470 	APIC_OP_XAPIC_RD,
5471 	APIC_OP_XAPIC_WR,
5472 	TERMINATE,
5473 };
5474 
5475 static u32 vmx_xapic_read(u32 *apic_access_address, u32 reg)
5476 {
5477 	return *(volatile u32 *)((uintptr_t)apic_access_address + reg);
5478 }
5479 
5480 static void vmx_xapic_write(u32 *apic_access_address, u32 reg, u32 val)
5481 {
5482 	*(volatile u32 *)((uintptr_t)apic_access_address + reg) = val;
5483 }
5484 
5485 struct apic_reg_virt_guest_args {
5486 	enum Apic_op op;
5487 	u32 *apic_access_address;
5488 	u32 reg;
5489 	u32 val;
5490 	bool check_rd;
5491 	u32 (*virt_fn)(u32);
5492 } apic_reg_virt_guest_args;
5493 
5494 static void apic_reg_virt_guest(void)
5495 {
5496 	volatile struct apic_reg_virt_guest_args *args =
5497 		&apic_reg_virt_guest_args;
5498 
5499 	for (;;) {
5500 		enum Apic_op op = args->op;
5501 		u32 *apic_access_address = args->apic_access_address;
5502 		u32 reg = args->reg;
5503 		u32 val = args->val;
5504 		bool check_rd = args->check_rd;
5505 		u32 (*virt_fn)(u32) = args->virt_fn;
5506 
5507 		if (op == TERMINATE)
5508 			break;
5509 
5510 		if (op == APIC_OP_XAPIC_RD) {
5511 			u32 ret = vmx_xapic_read(apic_access_address, reg);
5512 
5513 			if (check_rd) {
5514 				u32 want = virt_fn(val);
5515 				u32 got = virt_fn(ret);
5516 
5517 				report("read 0x%x, expected 0x%x.",
5518 				       got == want, got, want);
5519 			}
5520 		} else if (op == APIC_OP_XAPIC_WR) {
5521 			vmx_xapic_write(apic_access_address, reg, val);
5522 		}
5523 
5524 		/*
5525 		 * The L1 should always execute a vmcall after it's done testing
5526 		 * an individual APIC operation. This helps to validate that the
5527 		 * L1 and L2 are in sync with each other, as expected.
5528 		 */
5529 		vmcall();
5530 	}
5531 }
5532 
5533 static void test_xapic_rd(
5534 	u32 reg, struct apic_reg_virt_expectation *expectation,
5535 	u32 *apic_access_address, u32 *virtual_apic_page)
5536 {
5537 	u32 val = expectation->val;
5538 	u32 exit_reason_want = expectation->rd_exit_reason;
5539 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5540 
5541 	report_prefix_pushf("xapic - reading 0x%03x", reg);
5542 
5543 	/* Configure guest to do an xapic read */
5544 	args->op = APIC_OP_XAPIC_RD;
5545 	args->apic_access_address = apic_access_address;
5546 	args->reg = reg;
5547 	args->val = val;
5548 	args->check_rd = exit_reason_want == VMX_VMCALL;
5549 	args->virt_fn = expectation->virt_fn;
5550 
5551 	/* Setup virtual APIC page */
5552 	if (!expectation->virtualize_apic_accesses) {
5553 		apic_access_address[apic_reg_index(reg)] = val;
5554 		virtual_apic_page[apic_reg_index(reg)] = 0;
5555 	} else if (exit_reason_want == VMX_VMCALL) {
5556 		apic_access_address[apic_reg_index(reg)] = 0;
5557 		virtual_apic_page[apic_reg_index(reg)] = val;
5558 	}
5559 
5560 	/* Enter guest */
5561 	enter_guest();
5562 
5563 	/*
5564 	 * Validate the behavior and
5565 	 * pass a magic value back to the guest.
5566 	 */
5567 	if (exit_reason_want == VMX_APIC_ACCESS) {
5568 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5569 
5570 		assert_exit_reason(exit_reason_want);
5571 		report("got APIC access exit @ page offset 0x%03x, want 0x%03x",
5572 		       apic_page_offset == reg, apic_page_offset, reg);
5573 		skip_exit_insn();
5574 
5575 		/* Reenter guest so it can consume/check rcx and exit again. */
5576 		enter_guest();
5577 	} else if (exit_reason_want != VMX_VMCALL) {
5578 		report("Oops, bad exit expectation: %u.", false,
5579 		       exit_reason_want);
5580 	}
5581 
5582 	skip_exit_vmcall();
5583 	report_prefix_pop();
5584 }
5585 
5586 static void test_xapic_wr(
5587 	u32 reg, struct apic_reg_virt_expectation *expectation,
5588 	u32 *apic_access_address, u32 *virtual_apic_page)
5589 {
5590 	u32 val = expectation->val;
5591 	u32 exit_reason_want = expectation->wr_exit_reason;
5592 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5593 	bool virtualized =
5594 		expectation->virtualize_apic_accesses &&
5595 		(exit_reason_want == VMX_APIC_WRITE ||
5596 		 exit_reason_want == VMX_VMCALL);
5597 	bool checked = false;
5598 
5599 	report_prefix_pushf("xapic - writing 0x%x to 0x%03x", val, reg);
5600 
5601 	/* Configure guest to do an xapic read */
5602 	args->op = APIC_OP_XAPIC_WR;
5603 	args->apic_access_address = apic_access_address;
5604 	args->reg = reg;
5605 	args->val = val;
5606 
5607 	/* Setup virtual APIC page */
5608 	if (virtualized || !expectation->virtualize_apic_accesses) {
5609 		apic_access_address[apic_reg_index(reg)] = 0;
5610 		virtual_apic_page[apic_reg_index(reg)] = 0;
5611 	}
5612 
5613 	/* Enter guest */
5614 	enter_guest();
5615 
5616 	/*
5617 	 * Validate the behavior and
5618 	 * pass a magic value back to the guest.
5619 	 */
5620 	if (exit_reason_want == VMX_APIC_ACCESS) {
5621 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5622 
5623 		assert_exit_reason(exit_reason_want);
5624 		report("got APIC access exit @ page offset 0x%03x, want 0x%03x",
5625 		       apic_page_offset == reg, apic_page_offset, reg);
5626 		skip_exit_insn();
5627 
5628 		/* Reenter guest so it can consume/check rcx and exit again. */
5629 		enter_guest();
5630 	} else if (exit_reason_want == VMX_APIC_WRITE) {
5631 		assert_exit_reason(exit_reason_want);
5632 		report("got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%x",
5633 		       virtual_apic_page[apic_reg_index(reg)] == val,
5634 		       apic_reg_index(reg),
5635 		       virtual_apic_page[apic_reg_index(reg)], val);
5636 		checked = true;
5637 
5638 		/* Reenter guest so it can consume/check rcx and exit again. */
5639 		enter_guest();
5640 	} else if (exit_reason_want != VMX_VMCALL) {
5641 		report("Oops, bad exit expectation: %u.", false,
5642 		       exit_reason_want);
5643 	}
5644 
5645 	assert_exit_reason(VMX_VMCALL);
5646 	if (virtualized && !checked) {
5647 		u32 want = expectation->virt_fn(val);
5648 		u32 got = virtual_apic_page[apic_reg_index(reg)];
5649 		got = expectation->virt_fn(got);
5650 
5651 		report("exitless write; val is 0x%x, want 0x%x",
5652 		       got == want, got, want);
5653 	} else if (!expectation->virtualize_apic_accesses && !checked) {
5654 		u32 got = apic_access_address[apic_reg_index(reg)];
5655 
5656 		report("non-virtualized write; val is 0x%x, want 0x%x",
5657 		       got == val, got, val);
5658 	} else if (!expectation->virtualize_apic_accesses && checked) {
5659 		report("Non-virtualized write was prematurely checked!", false);
5660 	}
5661 
5662 	skip_exit_vmcall();
5663 	report_prefix_pop();
5664 }
5665 
5666 enum Config_type {
5667 	CONFIG_TYPE_GOOD,
5668 	CONFIG_TYPE_UNSUPPORTED,
5669 	CONFIG_TYPE_VMENTRY_FAILS_EARLY,
5670 };
5671 
5672 static enum Config_type configure_apic_reg_virt_test(
5673 	struct apic_reg_virt_config *apic_reg_virt_config)
5674 {
5675 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5676 	u32 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
5677 	/* Configs where L2 entry fails early, due to invalid controls. */
5678 	bool use_tpr_shadow_incorrectly_off =
5679 		!apic_reg_virt_config->use_tpr_shadow &&
5680 		(apic_reg_virt_config->apic_register_virtualization ||
5681 		 apic_reg_virt_config->virtualize_x2apic_mode) &&
5682 		apic_reg_virt_config->activate_secondary_controls;
5683 	bool virtualize_apic_accesses_incorrectly_on =
5684 		apic_reg_virt_config->virtualize_apic_accesses &&
5685 		apic_reg_virt_config->virtualize_x2apic_mode &&
5686 		apic_reg_virt_config->activate_secondary_controls;
5687 	bool vmentry_fails_early =
5688 		use_tpr_shadow_incorrectly_off ||
5689 		virtualize_apic_accesses_incorrectly_on;
5690 
5691 	if (apic_reg_virt_config->activate_secondary_controls) {
5692 		if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
5693 			printf("VM-execution control \"activate secondary controls\" NOT supported.\n");
5694 			return CONFIG_TYPE_UNSUPPORTED;
5695 		}
5696 		cpu_exec_ctrl0 |= CPU_SECONDARY;
5697 	} else {
5698 		cpu_exec_ctrl0 &= ~CPU_SECONDARY;
5699 	}
5700 
5701 	if (apic_reg_virt_config->virtualize_apic_accesses) {
5702 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES)) {
5703 			printf("VM-execution control \"virtualize APIC accesses\" NOT supported.\n");
5704 			return CONFIG_TYPE_UNSUPPORTED;
5705 		}
5706 		cpu_exec_ctrl1 |= CPU_VIRT_APIC_ACCESSES;
5707 	} else {
5708 		cpu_exec_ctrl1 &= ~CPU_VIRT_APIC_ACCESSES;
5709 	}
5710 
5711 	if (apic_reg_virt_config->use_tpr_shadow) {
5712 		if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
5713 			printf("VM-execution control \"use TPR shadow\" NOT supported.\n");
5714 			return CONFIG_TYPE_UNSUPPORTED;
5715 		}
5716 		cpu_exec_ctrl0 |= CPU_TPR_SHADOW;
5717 	} else {
5718 		cpu_exec_ctrl0 &= ~CPU_TPR_SHADOW;
5719 	}
5720 
5721 	if (apic_reg_virt_config->apic_register_virtualization) {
5722 		if (!(ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT)) {
5723 			printf("VM-execution control \"APIC-register virtualization\" NOT supported.\n");
5724 			return CONFIG_TYPE_UNSUPPORTED;
5725 		}
5726 		cpu_exec_ctrl1 |= CPU_APIC_REG_VIRT;
5727 	} else {
5728 		cpu_exec_ctrl1 &= ~CPU_APIC_REG_VIRT;
5729 	}
5730 
5731 	if (apic_reg_virt_config->virtualize_x2apic_mode) {
5732 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_X2APIC)) {
5733 			printf("VM-execution control \"virtualize x2APIC mode\" NOT supported.\n");
5734 			return CONFIG_TYPE_UNSUPPORTED;
5735 		}
5736 		cpu_exec_ctrl1 |= CPU_VIRT_X2APIC;
5737 	} else {
5738 		cpu_exec_ctrl1 &= ~CPU_VIRT_X2APIC;
5739 	}
5740 
5741 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
5742 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
5743 
5744 	if (vmentry_fails_early)
5745 		return CONFIG_TYPE_VMENTRY_FAILS_EARLY;
5746 
5747 	return CONFIG_TYPE_GOOD;
5748 }
5749 
5750 static bool cpu_has_apicv(void)
5751 {
5752 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
5753 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
5754 		(ctrl_pin_rev.clr & PIN_POST_INTR));
5755 }
5756 
5757 /* Validates APIC register access across valid virtualization configurations. */
5758 static void apic_reg_virt_test(void)
5759 {
5760 	u32 *apic_access_address;
5761 	u32 *virtual_apic_page;
5762 	u64 control;
5763 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5764 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
5765 	int i;
5766 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5767 
5768 	if (!cpu_has_apicv()) {
5769 		report_skip(__func__);
5770 		return;
5771 	}
5772 
5773 	control = cpu_exec_ctrl1;
5774 	control &= ~CPU_VINTD;
5775 	vmcs_write(CPU_EXEC_CTRL1, control);
5776 
5777 	test_set_guest(apic_reg_virt_guest);
5778 
5779 	/*
5780 	 * From the SDM: The 1-setting of the "virtualize APIC accesses"
5781 	 * VM-execution is guaranteed to apply only if translations to the
5782 	 * APIC-access address use a 4-KByte page.
5783 	 */
5784 	apic_access_address = alloc_page();
5785 	force_4k_page(apic_access_address);
5786 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_address));
5787 
5788 	virtual_apic_page = alloc_page();
5789 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
5790 
5791 	for (i = 0; i < ARRAY_SIZE(apic_reg_tests); i++) {
5792 		struct apic_reg_test *apic_reg_test = &apic_reg_tests[i];
5793 		struct apic_reg_virt_config *apic_reg_virt_config =
5794 				&apic_reg_test->apic_reg_virt_config;
5795 		enum Config_type config_type;
5796 		u32 reg;
5797 
5798 		printf("--- %s test ---\n", apic_reg_test->name);
5799 		config_type =
5800 			configure_apic_reg_virt_test(apic_reg_virt_config);
5801 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
5802 			printf("Skip because of missing features.\n");
5803 			continue;
5804 		}
5805 
5806 		if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
5807 			enter_guest_with_bad_controls();
5808 			continue;
5809 		}
5810 
5811 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
5812 			struct apic_reg_virt_expectation expectation = {};
5813 			bool ok;
5814 
5815 			ok = apic_reg_virt_exit_expectation(
5816 				reg, apic_reg_virt_config, &expectation);
5817 			if (!ok) {
5818 				report("Malformed test.", false);
5819 				break;
5820 			}
5821 
5822 			test_xapic_rd(reg, &expectation, apic_access_address,
5823 				      virtual_apic_page);
5824 			test_xapic_wr(reg, &expectation, apic_access_address,
5825 				      virtual_apic_page);
5826 		}
5827 	}
5828 
5829 	/* Terminate the guest */
5830 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
5831 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
5832 	args->op = TERMINATE;
5833 	enter_guest();
5834 	assert_exit_reason(VMX_VMCALL);
5835 }
5836 
5837 struct virt_x2apic_mode_config {
5838 	struct apic_reg_virt_config apic_reg_virt_config;
5839 	bool virtual_interrupt_delivery;
5840 	bool use_msr_bitmaps;
5841 	bool disable_x2apic_msr_intercepts;
5842 	bool disable_x2apic;
5843 };
5844 
5845 struct virt_x2apic_mode_test_case {
5846 	const char *name;
5847 	struct virt_x2apic_mode_config virt_x2apic_mode_config;
5848 };
5849 
5850 enum Virt_x2apic_mode_behavior_type {
5851 	X2APIC_ACCESS_VIRTUALIZED,
5852 	X2APIC_ACCESS_PASSED_THROUGH,
5853 	X2APIC_ACCESS_TRIGGERS_GP,
5854 };
5855 
5856 struct virt_x2apic_mode_expectation {
5857 	enum Reason rd_exit_reason;
5858 	enum Reason wr_exit_reason;
5859 
5860 	/*
5861 	 * RDMSR and WRMSR handle 64-bit values. However, except for ICR, all of
5862 	 * the x2APIC registers are 32 bits. Notice:
5863 	 *   1. vmx_x2apic_read() clears the upper 32 bits for 32-bit registers.
5864 	 *   2. vmx_x2apic_write() expects the val arg to be well-formed.
5865 	 */
5866 	u64 rd_val;
5867 	u64 wr_val;
5868 
5869 	/*
5870 	 * Compares input to virtualized output;
5871 	 * 1st arg is pointer to return expected virtualization output.
5872 	 */
5873 	u64 (*virt_fn)(u64);
5874 
5875 	enum Virt_x2apic_mode_behavior_type rd_behavior;
5876 	enum Virt_x2apic_mode_behavior_type wr_behavior;
5877 	bool wr_only;
5878 };
5879 
5880 static u64 virt_x2apic_mode_identity(u64 val)
5881 {
5882 	return val;
5883 }
5884 
5885 static u64 virt_x2apic_mode_nibble1(u64 val)
5886 {
5887 	return val & 0xf0;
5888 }
5889 
5890 static bool is_cmci_enabled(void)
5891 {
5892 	return rdmsr(MSR_IA32_MCG_CAP) & BIT_ULL(10);
5893 }
5894 
5895 static void virt_x2apic_mode_rd_expectation(
5896 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
5897 	bool apic_register_virtualization, bool virtual_interrupt_delivery,
5898 	struct virt_x2apic_mode_expectation *expectation)
5899 {
5900 	bool readable =
5901 		!x2apic_reg_reserved(reg) &&
5902 		reg != APIC_EOI;
5903 
5904 	if (reg == APIC_CMCI && !is_cmci_enabled())
5905 		readable = false;
5906 
5907 	expectation->rd_exit_reason = VMX_VMCALL;
5908 	expectation->virt_fn = virt_x2apic_mode_identity;
5909 	if (virt_x2apic_mode_on && apic_register_virtualization) {
5910 		expectation->rd_val = MAGIC_VAL_1;
5911 		if (reg == APIC_PROCPRI && virtual_interrupt_delivery)
5912 			expectation->virt_fn = virt_x2apic_mode_nibble1;
5913 		else if (reg == APIC_TASKPRI)
5914 			expectation->virt_fn = virt_x2apic_mode_nibble1;
5915 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
5916 	} else if (virt_x2apic_mode_on && !apic_register_virtualization &&
5917 		   reg == APIC_TASKPRI) {
5918 		expectation->rd_val = MAGIC_VAL_1;
5919 		expectation->virt_fn = virt_x2apic_mode_nibble1;
5920 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
5921 	} else if (!disable_x2apic && readable) {
5922 		expectation->rd_val = apic_read(reg);
5923 		expectation->rd_behavior = X2APIC_ACCESS_PASSED_THROUGH;
5924 	} else {
5925 		expectation->rd_behavior = X2APIC_ACCESS_TRIGGERS_GP;
5926 	}
5927 }
5928 
5929 /*
5930  * get_x2apic_wr_val() creates an innocuous write value for an x2APIC register.
5931  *
5932  * For writable registers, get_x2apic_wr_val() deposits the write value into the
5933  * val pointer arg and returns true. For non-writable registers, val is not
5934  * modified and get_x2apic_wr_val() returns false.
5935  */
5936 static bool get_x2apic_wr_val(u32 reg, u64 *val)
5937 {
5938 	switch (reg) {
5939 	case APIC_TASKPRI:
5940 		/* Bits 31:8 are reserved. */
5941 		*val &= 0xff;
5942 		break;
5943 	case APIC_EOI:
5944 	case APIC_ESR:
5945 	case APIC_TMICT:
5946 		/*
5947 		 * EOI, ESR: WRMSR of a non-zero value causes #GP(0).
5948 		 * TMICT: A write of 0 to the initial-count register effectively
5949 		 *        stops the local APIC timer, in both one-shot and
5950 		 *        periodic mode.
5951 		 */
5952 		*val = 0;
5953 		break;
5954 	case APIC_SPIV:
5955 	case APIC_LVTT:
5956 	case APIC_LVTTHMR:
5957 	case APIC_LVTPC:
5958 	case APIC_LVT0:
5959 	case APIC_LVT1:
5960 	case APIC_LVTERR:
5961 	case APIC_TDCR:
5962 		/*
5963 		 * To avoid writing a 1 to a reserved bit or causing some other
5964 		 * unintended side effect, read the current value and use it as
5965 		 * the write value.
5966 		 */
5967 		*val = apic_read(reg);
5968 		break;
5969 	case APIC_CMCI:
5970 		if (!is_cmci_enabled())
5971 			return false;
5972 		*val = apic_read(reg);
5973 		break;
5974 	case APIC_ICR:
5975 		*val = 0x40000 | 0xf1;
5976 		break;
5977 	case APIC_SELF_IPI:
5978 		/*
5979 		 * With special processing (i.e., virtualize x2APIC mode +
5980 		 * virtual interrupt delivery), writing zero causes an
5981 		 * APIC-write VM exit. We plan to add a test for enabling
5982 		 * "virtual-interrupt delivery" in VMCS12, and that's where we
5983 		 * will test a self IPI with special processing.
5984 		 */
5985 		*val = 0x0;
5986 		break;
5987 	default:
5988 		return false;
5989 	}
5990 
5991 	return true;
5992 }
5993 
5994 static bool special_processing_applies(u32 reg, u64 *val,
5995 				       bool virt_int_delivery)
5996 {
5997 	bool special_processing =
5998 		(reg == APIC_TASKPRI) ||
5999 		(virt_int_delivery &&
6000 		 (reg == APIC_EOI || reg == APIC_SELF_IPI));
6001 
6002 	if (special_processing) {
6003 		TEST_ASSERT(get_x2apic_wr_val(reg, val));
6004 		return true;
6005 	}
6006 
6007 	return false;
6008 }
6009 
6010 static void virt_x2apic_mode_wr_expectation(
6011 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6012 	bool virt_int_delivery,
6013 	struct virt_x2apic_mode_expectation *expectation)
6014 {
6015 	expectation->wr_exit_reason = VMX_VMCALL;
6016 	expectation->wr_val = MAGIC_VAL_1;
6017 	expectation->wr_only = false;
6018 
6019 	if (virt_x2apic_mode_on &&
6020 	    special_processing_applies(reg, &expectation->wr_val,
6021 				       virt_int_delivery)) {
6022 		expectation->wr_behavior = X2APIC_ACCESS_VIRTUALIZED;
6023 		if (reg == APIC_SELF_IPI)
6024 			expectation->wr_exit_reason = VMX_APIC_WRITE;
6025 	} else if (!disable_x2apic &&
6026 		   get_x2apic_wr_val(reg, &expectation->wr_val)) {
6027 		expectation->wr_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6028 		if (reg == APIC_EOI || reg == APIC_SELF_IPI)
6029 			expectation->wr_only = true;
6030 		if (reg == APIC_ICR)
6031 			expectation->wr_exit_reason = VMX_EXTINT;
6032 	} else {
6033 		expectation->wr_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6034 		/*
6035 		 * Writing 1 to a reserved bit triggers a #GP.
6036 		 * Thus, set the write value to 0, which seems
6037 		 * the most likely to detect a missed #GP.
6038 		 */
6039 		expectation->wr_val = 0;
6040 	}
6041 }
6042 
6043 static void virt_x2apic_mode_exit_expectation(
6044 	u32 reg, struct virt_x2apic_mode_config *config,
6045 	struct virt_x2apic_mode_expectation *expectation)
6046 {
6047 	struct apic_reg_virt_config *base_config =
6048 		&config->apic_reg_virt_config;
6049 	bool virt_x2apic_mode_on =
6050 		base_config->virtualize_x2apic_mode &&
6051 		config->use_msr_bitmaps &&
6052 		config->disable_x2apic_msr_intercepts &&
6053 		base_config->activate_secondary_controls;
6054 
6055 	virt_x2apic_mode_wr_expectation(
6056 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6057 		config->virtual_interrupt_delivery, expectation);
6058 	virt_x2apic_mode_rd_expectation(
6059 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6060 		base_config->apic_register_virtualization,
6061 		config->virtual_interrupt_delivery, expectation);
6062 }
6063 
6064 struct virt_x2apic_mode_test_case virt_x2apic_mode_tests[] = {
6065 	/*
6066 	 * Baseline "virtualize x2APIC mode" configuration:
6067 	 *   - virtualize x2APIC mode
6068 	 *   - virtual-interrupt delivery
6069 	 *   - APIC-register virtualization
6070 	 *   - x2APIC MSR intercepts disabled
6071 	 *
6072 	 * Reads come from virtual APIC page, special processing applies to
6073 	 * VTPR, EOI, and SELF IPI, and all other writes pass through to L1
6074 	 * APIC.
6075 	 */
6076 	{
6077 		.name = "Baseline",
6078 		.virt_x2apic_mode_config = {
6079 			.virtual_interrupt_delivery = true,
6080 			.use_msr_bitmaps = true,
6081 			.disable_x2apic_msr_intercepts = true,
6082 			.disable_x2apic = false,
6083 			.apic_reg_virt_config = {
6084 				.apic_register_virtualization = true,
6085 				.use_tpr_shadow = true,
6086 				.virtualize_apic_accesses = false,
6087 				.virtualize_x2apic_mode = true,
6088 				.activate_secondary_controls = true,
6089 			},
6090 		},
6091 	},
6092 	{
6093 		.name = "Baseline w/ x2apic disabled",
6094 		.virt_x2apic_mode_config = {
6095 			.virtual_interrupt_delivery = true,
6096 			.use_msr_bitmaps = true,
6097 			.disable_x2apic_msr_intercepts = true,
6098 			.disable_x2apic = true,
6099 			.apic_reg_virt_config = {
6100 				.apic_register_virtualization = true,
6101 				.use_tpr_shadow = true,
6102 				.virtualize_apic_accesses = false,
6103 				.virtualize_x2apic_mode = true,
6104 				.activate_secondary_controls = true,
6105 			},
6106 		},
6107 	},
6108 
6109 	/*
6110 	 * Baseline, minus virtual-interrupt delivery. Reads come from virtual
6111 	 * APIC page, special processing applies to VTPR, and all other writes
6112 	 * pass through to L1 APIC.
6113 	 */
6114 	{
6115 		.name = "Baseline - virtual interrupt delivery",
6116 		.virt_x2apic_mode_config = {
6117 			.virtual_interrupt_delivery = false,
6118 			.use_msr_bitmaps = true,
6119 			.disable_x2apic_msr_intercepts = true,
6120 			.disable_x2apic = false,
6121 			.apic_reg_virt_config = {
6122 				.apic_register_virtualization = true,
6123 				.use_tpr_shadow = true,
6124 				.virtualize_apic_accesses = false,
6125 				.virtualize_x2apic_mode = true,
6126 				.activate_secondary_controls = true,
6127 			},
6128 		},
6129 	},
6130 
6131 	/*
6132 	 * Baseline, minus APIC-register virtualization. x2APIC reads pass
6133 	 * through to L1's APIC, unless reading VTPR
6134 	 */
6135 	{
6136 		.name = "Virtualize x2APIC mode, no APIC reg virt",
6137 		.virt_x2apic_mode_config = {
6138 			.virtual_interrupt_delivery = true,
6139 			.use_msr_bitmaps = true,
6140 			.disable_x2apic_msr_intercepts = true,
6141 			.disable_x2apic = false,
6142 			.apic_reg_virt_config = {
6143 				.apic_register_virtualization = false,
6144 				.use_tpr_shadow = true,
6145 				.virtualize_apic_accesses = false,
6146 				.virtualize_x2apic_mode = true,
6147 				.activate_secondary_controls = true,
6148 			},
6149 		},
6150 	},
6151 	{
6152 		.name = "Virtualize x2APIC mode, no APIC reg virt, x2APIC off",
6153 		.virt_x2apic_mode_config = {
6154 			.virtual_interrupt_delivery = true,
6155 			.use_msr_bitmaps = true,
6156 			.disable_x2apic_msr_intercepts = true,
6157 			.disable_x2apic = true,
6158 			.apic_reg_virt_config = {
6159 				.apic_register_virtualization = false,
6160 				.use_tpr_shadow = true,
6161 				.virtualize_apic_accesses = false,
6162 				.virtualize_x2apic_mode = true,
6163 				.activate_secondary_controls = true,
6164 			},
6165 		},
6166 	},
6167 
6168 	/*
6169 	 * Enable "virtualize x2APIC mode" and "APIC-register virtualization",
6170 	 * and disable intercepts for the x2APIC MSRs, but fail to enable
6171 	 * "activate secondary controls" (i.e. L2 gets access to L1's x2APIC
6172 	 * MSRs).
6173 	 */
6174 	{
6175 		.name = "Fail to enable activate secondary controls",
6176 		.virt_x2apic_mode_config = {
6177 			.virtual_interrupt_delivery = true,
6178 			.use_msr_bitmaps = true,
6179 			.disable_x2apic_msr_intercepts = true,
6180 			.disable_x2apic = false,
6181 			.apic_reg_virt_config = {
6182 				.apic_register_virtualization = true,
6183 				.use_tpr_shadow = true,
6184 				.virtualize_apic_accesses = false,
6185 				.virtualize_x2apic_mode = true,
6186 				.activate_secondary_controls = false,
6187 			},
6188 		},
6189 	},
6190 
6191 	/*
6192 	 * Enable "APIC-register virtualization" and enable "activate secondary
6193 	 * controls" and disable intercepts for the x2APIC MSRs, but do not
6194 	 * enable the "virtualize x2APIC mode" VM-execution control (i.e. L2
6195 	 * gets access to L1's x2APIC MSRs).
6196 	 */
6197 	{
6198 		.name = "Fail to enable virtualize x2APIC mode",
6199 		.virt_x2apic_mode_config = {
6200 			.virtual_interrupt_delivery = true,
6201 			.use_msr_bitmaps = true,
6202 			.disable_x2apic_msr_intercepts = true,
6203 			.disable_x2apic = false,
6204 			.apic_reg_virt_config = {
6205 				.apic_register_virtualization = true,
6206 				.use_tpr_shadow = true,
6207 				.virtualize_apic_accesses = false,
6208 				.virtualize_x2apic_mode = false,
6209 				.activate_secondary_controls = true,
6210 			},
6211 		},
6212 	},
6213 
6214 	/*
6215 	 * Disable "Virtualize x2APIC mode", disable x2APIC MSR intercepts, and
6216 	 * enable "APIC-register virtualization" --> L2 gets L1's x2APIC MSRs.
6217 	 */
6218 	{
6219 		.name = "Baseline",
6220 		.virt_x2apic_mode_config = {
6221 			.virtual_interrupt_delivery = true,
6222 			.use_msr_bitmaps = true,
6223 			.disable_x2apic_msr_intercepts = true,
6224 			.disable_x2apic = false,
6225 			.apic_reg_virt_config = {
6226 				.apic_register_virtualization = true,
6227 				.use_tpr_shadow = true,
6228 				.virtualize_apic_accesses = false,
6229 				.virtualize_x2apic_mode = false,
6230 				.activate_secondary_controls = true,
6231 			},
6232 		},
6233 	},
6234 };
6235 
6236 enum X2apic_op {
6237 	X2APIC_OP_RD,
6238 	X2APIC_OP_WR,
6239 	X2APIC_TERMINATE,
6240 };
6241 
6242 static u64 vmx_x2apic_read(u32 reg)
6243 {
6244 	u32 msr_addr = x2apic_msr(reg);
6245 	u64 val;
6246 
6247 	val = rdmsr(msr_addr);
6248 
6249 	return val;
6250 }
6251 
6252 static void vmx_x2apic_write(u32 reg, u64 val)
6253 {
6254 	u32 msr_addr = x2apic_msr(reg);
6255 
6256 	wrmsr(msr_addr, val);
6257 }
6258 
6259 struct virt_x2apic_mode_guest_args {
6260 	enum X2apic_op op;
6261 	u32 reg;
6262 	u64 val;
6263 	bool should_gp;
6264 	u64 (*virt_fn)(u64);
6265 } virt_x2apic_mode_guest_args;
6266 
6267 static volatile bool handle_x2apic_gp_ran;
6268 static volatile u32 handle_x2apic_gp_insn_len;
6269 static void handle_x2apic_gp(struct ex_regs *regs)
6270 {
6271 	handle_x2apic_gp_ran = true;
6272 	regs->rip += handle_x2apic_gp_insn_len;
6273 }
6274 
6275 static handler setup_x2apic_gp_handler(void)
6276 {
6277 	handler old_handler;
6278 
6279 	old_handler = handle_exception(GP_VECTOR, handle_x2apic_gp);
6280 	/* RDMSR and WRMSR are both 2 bytes, assuming no prefixes. */
6281 	handle_x2apic_gp_insn_len = 2;
6282 
6283 	return old_handler;
6284 }
6285 
6286 static void teardown_x2apic_gp_handler(handler old_handler)
6287 {
6288 	handle_exception(GP_VECTOR, old_handler);
6289 
6290 	/*
6291 	 * Defensively reset instruction length, so that if the handler is
6292 	 * incorrectly used, it will loop infinitely, rather than run off into
6293 	 * la la land.
6294 	 */
6295 	handle_x2apic_gp_insn_len = 0;
6296 	handle_x2apic_gp_ran = false;
6297 }
6298 
6299 static void virt_x2apic_mode_guest(void)
6300 {
6301 	volatile struct virt_x2apic_mode_guest_args *args =
6302 		&virt_x2apic_mode_guest_args;
6303 
6304 	for (;;) {
6305 		enum X2apic_op op = args->op;
6306 		u32 reg = args->reg;
6307 		u64 val = args->val;
6308 		bool should_gp = args->should_gp;
6309 		u64 (*virt_fn)(u64) = args->virt_fn;
6310 		handler old_handler;
6311 
6312 		if (op == X2APIC_TERMINATE)
6313 			break;
6314 
6315 		if (should_gp) {
6316 			TEST_ASSERT(!handle_x2apic_gp_ran);
6317 			old_handler = setup_x2apic_gp_handler();
6318 		}
6319 
6320 		if (op == X2APIC_OP_RD) {
6321 			u64 ret = vmx_x2apic_read(reg);
6322 
6323 			if (!should_gp) {
6324 				u64 want = virt_fn(val);
6325 				u64 got = virt_fn(ret);
6326 
6327 				report("APIC read; got 0x%lx, want 0x%lx.",
6328 				       got == want, got, want);
6329 			}
6330 		} else if (op == X2APIC_OP_WR) {
6331 			vmx_x2apic_write(reg, val);
6332 		}
6333 
6334 		if (should_gp) {
6335 			report("x2APIC op triggered GP.",
6336 			       handle_x2apic_gp_ran);
6337 			teardown_x2apic_gp_handler(old_handler);
6338 		}
6339 
6340 		/*
6341 		 * The L1 should always execute a vmcall after it's done testing
6342 		 * an individual APIC operation. This helps to validate that the
6343 		 * L1 and L2 are in sync with each other, as expected.
6344 		 */
6345 		vmcall();
6346 	}
6347 }
6348 
6349 static void test_x2apic_rd(
6350 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6351 	u32 *virtual_apic_page)
6352 {
6353 	u64 val = expectation->rd_val;
6354 	u32 exit_reason_want = expectation->rd_exit_reason;
6355 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6356 
6357 	report_prefix_pushf("x2apic - reading 0x%03x", reg);
6358 
6359 	/* Configure guest to do an x2apic read */
6360 	args->op = X2APIC_OP_RD;
6361 	args->reg = reg;
6362 	args->val = val;
6363 	args->should_gp = expectation->rd_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6364 	args->virt_fn = expectation->virt_fn;
6365 
6366 	/* Setup virtual APIC page */
6367 	if (expectation->rd_behavior == X2APIC_ACCESS_VIRTUALIZED)
6368 		virtual_apic_page[apic_reg_index(reg)] = (u32)val;
6369 
6370 	/* Enter guest */
6371 	enter_guest();
6372 
6373 	if (exit_reason_want != VMX_VMCALL) {
6374 		report("Oops, bad exit expectation: %u.", false,
6375 		       exit_reason_want);
6376 	}
6377 
6378 	skip_exit_vmcall();
6379 	report_prefix_pop();
6380 }
6381 
6382 static volatile bool handle_x2apic_ipi_ran;
6383 static void handle_x2apic_ipi(isr_regs_t *regs)
6384 {
6385 	handle_x2apic_ipi_ran = true;
6386 	eoi();
6387 }
6388 
6389 static void test_x2apic_wr(
6390 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6391 	u32 *virtual_apic_page)
6392 {
6393 	u64 val = expectation->wr_val;
6394 	u32 exit_reason_want = expectation->wr_exit_reason;
6395 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6396 	int ipi_vector = 0xf1;
6397 	u32 restore_val = 0;
6398 
6399 	report_prefix_pushf("x2apic - writing 0x%lx to 0x%03x", val, reg);
6400 
6401 	/* Configure guest to do an x2apic read */
6402 	args->op = X2APIC_OP_WR;
6403 	args->reg = reg;
6404 	args->val = val;
6405 	args->should_gp = expectation->wr_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6406 
6407 	/* Setup virtual APIC page */
6408 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED)
6409 		virtual_apic_page[apic_reg_index(reg)] = 0;
6410 	if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH && !expectation->wr_only)
6411 		restore_val = apic_read(reg);
6412 
6413 	/* Setup IPI handler */
6414 	handle_x2apic_ipi_ran = false;
6415 	handle_irq(ipi_vector, handle_x2apic_ipi);
6416 
6417 	/* Enter guest */
6418 	enter_guest();
6419 
6420 	/*
6421 	 * Validate the behavior and
6422 	 * pass a magic value back to the guest.
6423 	 */
6424 	if (exit_reason_want == VMX_EXTINT) {
6425 		assert_exit_reason(exit_reason_want);
6426 
6427 		/* Clear the external interrupt. */
6428 		irq_enable();
6429 		asm volatile ("nop");
6430 		irq_disable();
6431 		report("Got pending interrupt after IRQ enabled.",
6432 		       handle_x2apic_ipi_ran);
6433 
6434 		enter_guest();
6435 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6436 		assert_exit_reason(exit_reason_want);
6437 		report("got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%lx",
6438 		       virtual_apic_page[apic_reg_index(reg)] == val,
6439 		       apic_reg_index(reg),
6440 		       virtual_apic_page[apic_reg_index(reg)], val);
6441 
6442 		/* Reenter guest so it can consume/check rcx and exit again. */
6443 		enter_guest();
6444 	} else if (exit_reason_want != VMX_VMCALL) {
6445 		report("Oops, bad exit expectation: %u.", false,
6446 		       exit_reason_want);
6447 	}
6448 
6449 	assert_exit_reason(VMX_VMCALL);
6450 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) {
6451 		u64 want = val;
6452 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6453 
6454 		report("x2APIC write; got 0x%x, want 0x%lx",
6455 		       got == want, got, want);
6456 	} else if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH) {
6457 		if (!expectation->wr_only) {
6458 			u32 got = apic_read(reg);
6459 			bool ok;
6460 
6461 			/*
6462 			 * When L1's TPR is passed through to L2, the lower
6463 			 * nibble can be lost. For example, if L2 executes
6464 			 * WRMSR(0x808, 0x78), then, L1 might read 0x70.
6465 			 *
6466 			 * Here's how the lower nibble can get lost:
6467 			 *   1. L2 executes WRMSR(0x808, 0x78).
6468 			 *   2. L2 exits to L0 with a WRMSR exit.
6469 			 *   3. L0 emulates WRMSR, by writing L1's TPR.
6470 			 *   4. L0 re-enters L2.
6471 			 *   5. L2 exits to L0 (reason doesn't matter).
6472 			 *   6. L0 reflects L2's exit to L1.
6473 			 *   7. Before entering L1, L0 exits to user-space
6474 			 *      (e.g., to satisfy TPR access reporting).
6475 			 *   8. User-space executes KVM_SET_REGS ioctl, which
6476 			 *      clears the lower nibble of L1's TPR.
6477 			 */
6478 			if (reg == APIC_TASKPRI) {
6479 				got = apic_virt_nibble1(got);
6480 				val = apic_virt_nibble1(val);
6481 			}
6482 
6483 			ok = got == val;
6484 			report("non-virtualized write; val is 0x%x, want 0x%lx",
6485 			       ok, got, val);
6486 			apic_write(reg, restore_val);
6487 		} else {
6488 			report("non-virtualized and write-only OK", true);
6489 		}
6490 	}
6491 	skip_exit_insn();
6492 
6493 	report_prefix_pop();
6494 }
6495 
6496 static enum Config_type configure_virt_x2apic_mode_test(
6497 	struct virt_x2apic_mode_config *virt_x2apic_mode_config,
6498 	u8 *msr_bitmap_page)
6499 {
6500 	int msr;
6501 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6502 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6503 
6504 	/* x2apic-specific VMCS config */
6505 	if (virt_x2apic_mode_config->use_msr_bitmaps) {
6506 		/* virt_x2apic_mode_test() checks for MSR bitmaps support */
6507 		cpu_exec_ctrl0 |= CPU_MSR_BITMAP;
6508 	} else {
6509 		cpu_exec_ctrl0 &= ~CPU_MSR_BITMAP;
6510 	}
6511 
6512 	if (virt_x2apic_mode_config->virtual_interrupt_delivery) {
6513 		if (!(ctrl_cpu_rev[1].clr & CPU_VINTD)) {
6514 			report_skip("VM-execution control \"virtual-interrupt delivery\" NOT supported.\n");
6515 			return CONFIG_TYPE_UNSUPPORTED;
6516 		}
6517 		cpu_exec_ctrl1 |= CPU_VINTD;
6518 	} else {
6519 		cpu_exec_ctrl1 &= ~CPU_VINTD;
6520 	}
6521 
6522 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6523 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6524 
6525 	/* x2APIC MSR intercepts are usually off for "Virtualize x2APIC mode" */
6526 	for (msr = 0x800; msr <= 0x8ff; msr++) {
6527 		if (virt_x2apic_mode_config->disable_x2apic_msr_intercepts) {
6528 			clear_bit(msr, msr_bitmap_page + 0x000);
6529 			clear_bit(msr, msr_bitmap_page + 0x800);
6530 		} else {
6531 			set_bit(msr, msr_bitmap_page + 0x000);
6532 			set_bit(msr, msr_bitmap_page + 0x800);
6533 		}
6534 	}
6535 
6536 	/* x2APIC mode can impact virtualization */
6537 	reset_apic();
6538 	if (!virt_x2apic_mode_config->disable_x2apic)
6539 		enable_x2apic();
6540 
6541 	return configure_apic_reg_virt_test(
6542 		&virt_x2apic_mode_config->apic_reg_virt_config);
6543 }
6544 
6545 static void virt_x2apic_mode_test(void)
6546 {
6547 	u32 *virtual_apic_page;
6548 	u8 *msr_bitmap_page;
6549 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6550 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6551 	int i;
6552 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6553 
6554 	if (!cpu_has_apicv()) {
6555 		report_skip(__func__);
6556 		return;
6557 	}
6558 
6559 	/*
6560 	 * This is to exercise an issue in KVM's logic to merge L0's and L1's
6561 	 * MSR bitmaps. Previously, an L1 could get at L0's x2APIC MSRs by
6562 	 * writing the IA32_SPEC_CTRL MSR or the IA32_PRED_CMD MSRs. KVM would
6563 	 * then proceed to manipulate the MSR bitmaps, as if VMCS12 had the
6564 	 * "Virtualize x2APIC mod" control set, even when it didn't.
6565 	 */
6566 	if (has_spec_ctrl())
6567 		wrmsr(MSR_IA32_SPEC_CTRL, 1);
6568 
6569 	/*
6570 	 * Check that VMCS12 supports:
6571 	 *   - "Virtual-APIC address", indicated by "use TPR shadow"
6572 	 *   - "MSR-bitmap address", indicated by "use MSR bitmaps"
6573 	 */
6574 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6575 		report_skip("VM-execution control \"use TPR shadow\" NOT supported.\n");
6576 		return;
6577 	} else if (!(ctrl_cpu_rev[0].clr & CPU_MSR_BITMAP)) {
6578 		report_skip("VM-execution control \"use MSR bitmaps\" NOT supported.\n");
6579 		return;
6580 	}
6581 
6582 	test_set_guest(virt_x2apic_mode_guest);
6583 
6584 	virtual_apic_page = alloc_page();
6585 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6586 
6587 	msr_bitmap_page = alloc_page();
6588 	memset(msr_bitmap_page, 0xff, PAGE_SIZE);
6589 	vmcs_write(MSR_BITMAP, virt_to_phys(msr_bitmap_page));
6590 
6591 	for (i = 0; i < ARRAY_SIZE(virt_x2apic_mode_tests); i++) {
6592 		struct virt_x2apic_mode_test_case *virt_x2apic_mode_test_case =
6593 			&virt_x2apic_mode_tests[i];
6594 		struct virt_x2apic_mode_config *virt_x2apic_mode_config =
6595 			&virt_x2apic_mode_test_case->virt_x2apic_mode_config;
6596 		enum Config_type config_type;
6597 		u32 reg;
6598 
6599 		printf("--- %s test ---\n", virt_x2apic_mode_test_case->name);
6600 		config_type =
6601 			configure_virt_x2apic_mode_test(virt_x2apic_mode_config,
6602 							msr_bitmap_page);
6603 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6604 			report_skip("Skip because of missing features.\n");
6605 			continue;
6606 		} else if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6607 			enter_guest_with_bad_controls();
6608 			continue;
6609 		}
6610 
6611 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6612 			struct virt_x2apic_mode_expectation expectation;
6613 
6614 			virt_x2apic_mode_exit_expectation(
6615 				reg, virt_x2apic_mode_config, &expectation);
6616 
6617 			test_x2apic_rd(reg, &expectation, virtual_apic_page);
6618 			test_x2apic_wr(reg, &expectation, virtual_apic_page);
6619 		}
6620 	}
6621 
6622 
6623 	/* Terminate the guest */
6624 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6625 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6626 	args->op = X2APIC_TERMINATE;
6627 	enter_guest();
6628 	assert_exit_reason(VMX_VMCALL);
6629 }
6630 
6631 /*
6632  * On processors that support Intel 64 architecture, the IA32_SYSENTER_ESP
6633  * field and the IA32_SYSENTER_EIP field must each contain a canonical
6634  * address.
6635  *
6636  *  [Intel SDM]
6637  */
6638 static void test_sysenter_field(u32 field, const char *name)
6639 {
6640 	u64 addr_saved = vmcs_read(field);
6641 
6642 	vmcs_write(field, NONCANONICAL);
6643 	report_prefix_pushf("%s non-canonical", name);
6644 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6645 	report_prefix_pop();
6646 
6647 	vmcs_write(field, 0xffffffff);
6648 	report_prefix_pushf("%s canonical", name);
6649 	test_vmx_vmlaunch(0);
6650 	report_prefix_pop();
6651 
6652 	vmcs_write(field, addr_saved);
6653 }
6654 
6655 static void test_ctl_reg(const char *cr_name, u64 cr, u64 fixed0, u64 fixed1)
6656 {
6657 	u64 val;
6658 	u64 cr_saved = vmcs_read(cr);
6659 	int i;
6660 
6661 	val = fixed0 & fixed1;
6662 	if (cr == HOST_CR4)
6663 		vmcs_write(cr, val | X86_CR4_PAE);
6664 	else
6665 		vmcs_write(cr, val);
6666 	report_prefix_pushf("%s %lx", cr_name, val);
6667 	if (val == fixed0)
6668 		test_vmx_vmlaunch(0);
6669 	else
6670 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6671 	report_prefix_pop();
6672 
6673 	for (i = 0; i < 64; i++) {
6674 
6675 		/* Set a bit when the corresponding bit in fixed1 is 0 */
6676 		if ((fixed1 & (1ull << i)) == 0) {
6677 			if (cr == HOST_CR4 && ((1ull << i) & X86_CR4_SMEP ||
6678 					       (1ull << i) & X86_CR4_SMAP))
6679 				continue;
6680 
6681 			vmcs_write(cr, cr_saved | (1ull << i));
6682 			report_prefix_pushf("%s %llx", cr_name,
6683 						cr_saved | (1ull << i));
6684 			test_vmx_vmlaunch(
6685 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6686 			report_prefix_pop();
6687 		}
6688 
6689 		/* Unset a bit when the corresponding bit in fixed0 is 1 */
6690 		if (fixed0 & (1ull << i)) {
6691 			vmcs_write(cr, cr_saved & ~(1ull << i));
6692 			report_prefix_pushf("%s %llx", cr_name,
6693 						cr_saved & ~(1ull << i));
6694 			test_vmx_vmlaunch(
6695 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6696 			report_prefix_pop();
6697 		}
6698 	}
6699 
6700 	vmcs_write(cr, cr_saved);
6701 }
6702 
6703 /*
6704  * 1. The CR0 field must not set any bit to a value not supported in VMX
6705  *    operation.
6706  * 2. The CR4 field must not set any bit to a value not supported in VMX
6707  *    operation.
6708  * 3. On processors that support Intel 64 architecture, the CR3 field must
6709  *    be such that bits 63:52 and bits in the range 51:32 beyond the
6710  *    processor’s physical-address width must be 0.
6711  *
6712  *  [Intel SDM]
6713  */
6714 static void test_host_ctl_regs(void)
6715 {
6716 	u64 fixed0, fixed1, cr3, cr3_saved;
6717 	int i;
6718 
6719 	/* Test CR0 */
6720 	fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0);
6721 	fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1);
6722 	test_ctl_reg("HOST_CR0", HOST_CR0, fixed0, fixed1);
6723 
6724 	/* Test CR4 */
6725 	fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0);
6726 	fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1) &
6727 		 ~(X86_CR4_SMEP | X86_CR4_SMAP);
6728 	test_ctl_reg("HOST_CR4", HOST_CR4, fixed0, fixed1);
6729 
6730 	/* Test CR3 */
6731 	cr3_saved = vmcs_read(HOST_CR3);
6732 	for (i = cpuid_maxphyaddr(); i < 64; i++) {
6733 		cr3 = cr3_saved | (1ul << i);
6734 		vmcs_write(HOST_CR3, cr3);
6735 		report_prefix_pushf("HOST_CR3 %lx", cr3);
6736 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6737 		report_prefix_pop();
6738 	}
6739 
6740 	vmcs_write(HOST_CR3, cr3_saved);
6741 }
6742 
6743 static void test_efer_vmlaunch(u32 fld, bool ok)
6744 {
6745 	if (fld == HOST_EFER) {
6746 		if (ok)
6747 			test_vmx_vmlaunch(0);
6748 		else
6749 			test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6750 	} else {
6751 		if (ok) {
6752 			enter_guest();
6753 			report("vmlaunch succeeds", vmcs_read(EXI_REASON) == VMX_VMCALL);
6754 		} else {
6755 			enter_guest_with_invalid_guest_state();
6756 			report("vmlaunch fails", vmcs_read(EXI_REASON) == (VMX_ENTRY_FAILURE | VMX_FAIL_STATE));
6757 		}
6758 		advance_guest_state_test();
6759 	}
6760 }
6761 
6762 static void test_efer_one(u32 fld, const char * fld_name, u64 efer,
6763 			  u32 ctrl_fld, u64 ctrl,
6764 			  int i, const char *efer_bit_name)
6765 {
6766 	bool ok;
6767 
6768 	ok = true;
6769 	if (ctrl_fld == EXI_CONTROLS && (ctrl & EXI_LOAD_EFER)) {
6770 		if (!!(efer & EFER_LMA) != !!(ctrl & EXI_HOST_64))
6771 			ok = false;
6772 		if (!!(efer & EFER_LME) != !!(ctrl & EXI_HOST_64))
6773 			ok = false;
6774 	}
6775 	if (ctrl_fld == ENT_CONTROLS && (ctrl & ENT_LOAD_EFER)) {
6776 		/* Check LMA too since CR0.PG is set.  */
6777 		if (!!(efer & EFER_LMA) != !!(ctrl & ENT_GUEST_64))
6778 			ok = false;
6779 		if (!!(efer & EFER_LME) != !!(ctrl & ENT_GUEST_64))
6780 			ok = false;
6781 	}
6782 
6783 	/*
6784 	 * Skip the test if it would enter the guest in 32-bit mode.
6785 	 * Perhaps write the test in assembly and make sure it
6786 	 * can be run in either mode?
6787 	 */
6788 	if (fld == GUEST_EFER && ok && !(ctrl & ENT_GUEST_64))
6789 		return;
6790 
6791 	vmcs_write(ctrl_fld, ctrl);
6792 	vmcs_write(fld, efer);
6793 	report_prefix_pushf("%s %s bit turned %s, controls %s",
6794 			    fld_name, efer_bit_name,
6795 			    (i & 1) ? "on" : "off",
6796 			    (i & 2) ? "on" : "off");
6797 
6798 	test_efer_vmlaunch(fld, ok);
6799 	report_prefix_pop();
6800 }
6801 
6802 static void test_efer_bit(u32 fld, const char * fld_name,
6803 			  u32 ctrl_fld, u64 ctrl_bit, u64 efer_bit,
6804 			  const char *efer_bit_name)
6805 {
6806 	u64 efer_saved = vmcs_read(fld);
6807 	u32 ctrl_saved = vmcs_read(ctrl_fld);
6808 	int i;
6809 
6810 	for (i = 0; i < 4; i++) {
6811 		u64 efer = efer_saved & ~efer_bit;
6812 		u64 ctrl = ctrl_saved & ~ctrl_bit;
6813 
6814 		if (i & 1)
6815 			efer |= efer_bit;
6816 		if (i & 2)
6817 			ctrl |= ctrl_bit;
6818 
6819 		test_efer_one(fld, fld_name, efer, ctrl_fld, ctrl,
6820 			      i, efer_bit_name);
6821 	}
6822 
6823 	vmcs_write(ctrl_fld, ctrl_saved);
6824 	vmcs_write(fld, efer_saved);
6825 }
6826 
6827 static void test_efer(u32 fld, const char * fld_name, u32 ctrl_fld,
6828 		      u64 ctrl_bit1, u64 ctrl_bit2)
6829 {
6830 	u64 efer_saved = vmcs_read(fld);
6831 	u32 ctrl_saved = vmcs_read(ctrl_fld);
6832 	u64 efer_reserved_bits =  ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
6833 	u64 i;
6834 	u64 efer;
6835 
6836 	if (cpu_has_efer_nx())
6837 		efer_reserved_bits &= ~EFER_NX;
6838 
6839 	if (!ctrl_bit1) {
6840 		printf("\"Load-IA32-EFER\" exit control not supported\n");
6841 		goto test_entry_exit_mode;
6842 	}
6843 
6844 	report_prefix_pushf("%s %lx", fld_name, efer_saved);
6845 	test_efer_vmlaunch(fld, true);
6846 	report_prefix_pop();
6847 
6848 	/*
6849 	 * Check reserved bits
6850 	 */
6851 	vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit1);
6852 	for (i = 0; i < 64; i++) {
6853 		if ((1ull << i) & efer_reserved_bits) {
6854 			efer = efer_saved | (1ull << i);
6855 			vmcs_write(fld, efer);
6856 			report_prefix_pushf("%s %lx", fld_name, efer);
6857 			test_efer_vmlaunch(fld, true);
6858 			report_prefix_pop();
6859 		}
6860 	}
6861 
6862 	vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit1);
6863 	for (i = 0; i < 64; i++) {
6864 		if ((1ull << i) & efer_reserved_bits) {
6865 			efer = efer_saved | (1ull << i);
6866 			vmcs_write(fld, efer);
6867 			report_prefix_pushf("%s %lx", fld_name, efer);
6868 			test_efer_vmlaunch(fld, false);
6869 			report_prefix_pop();
6870 		}
6871 	}
6872 
6873 	vmcs_write(ctrl_fld, ctrl_saved);
6874 	vmcs_write(fld, efer_saved);
6875 
6876 	/*
6877 	 * Check LMA and LME bits
6878 	 */
6879 	test_efer_bit(fld, fld_name,
6880 		      ctrl_fld, ctrl_bit1,
6881 		      EFER_LMA,
6882 		      "EFER_LMA");
6883 	test_efer_bit(fld, fld_name,
6884 		      ctrl_fld, ctrl_bit1,
6885 		      EFER_LME,
6886 		      "EFER_LME");
6887 
6888 test_entry_exit_mode:
6889 	test_efer_bit(fld, fld_name,
6890 		      ctrl_fld, ctrl_bit2,
6891 		      EFER_LMA,
6892 		      "EFER_LMA");
6893 	test_efer_bit(fld, fld_name,
6894 		      ctrl_fld, ctrl_bit2,
6895 		      EFER_LME,
6896 		      "EFER_LME");
6897 }
6898 
6899 /*
6900  * If the 'load IA32_EFER' VM-exit control is 1, bits reserved in the
6901  * IA32_EFER MSR must be 0 in the field for that register. In addition,
6902  * the values of the LMA and LME bits in the field must each be that of
6903  * the 'host address-space size' VM-exit control.
6904  *
6905  *  [Intel SDM]
6906  */
6907 static void test_host_efer(void)
6908 {
6909 	test_efer(HOST_EFER, "HOST_EFER", EXI_CONTROLS,
6910 		  ctrl_exit_rev.clr & EXI_LOAD_EFER,
6911 		  EXI_HOST_64);
6912 }
6913 
6914 /*
6915  * If the 'load IA32_EFER' VM-enter control is 1, bits reserved in the
6916  * IA32_EFER MSR must be 0 in the field for that register. In addition,
6917  * the values of the LMA and LME bits in the field must each be that of
6918  * the 'IA32e-mode guest' VM-exit control.
6919  */
6920 static void test_guest_efer(void)
6921 {
6922 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER)) {
6923 		printf("\"Load-IA32-EFER\" entry control not supported\n");
6924 		return;
6925 	}
6926 
6927 	vmcs_write(GUEST_EFER, rdmsr(MSR_EFER));
6928 	test_efer(GUEST_EFER, "GUEST_EFER", ENT_CONTROLS,
6929 		  ctrl_enter_rev.clr & ENT_LOAD_EFER,
6930 		  ENT_GUEST_64);
6931 }
6932 
6933 /*
6934  * PAT values higher than 8 are uninteresting since they're likely lumped
6935  * in with "8". We only test values above 8 one bit at a time,
6936  * in order to reduce the number of VM-Entries and keep the runtime reasonable.
6937  */
6938 #define	PAT_VAL_LIMIT	8
6939 
6940 static void test_pat(u32 field, const char * field_name, u32 ctrl_field,
6941 		     u64 ctrl_bit)
6942 {
6943 	u32 ctrl_saved = vmcs_read(ctrl_field);
6944 	u64 pat_saved = vmcs_read(field);
6945 	u64 i, val;
6946 	u32 j;
6947 	int error;
6948 
6949 	vmcs_clear_bits(ctrl_field, ctrl_bit);
6950 
6951 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
6952 		/* Test PAT0..PAT7 fields */
6953 		for (j = 0; j < (i ? 8 : 1); j++) {
6954 			val = i << j * 8;
6955 			vmcs_write(field, val);
6956 			if (field == HOST_PAT) {
6957 				report_prefix_pushf("%s %lx", field_name, val);
6958 				test_vmx_vmlaunch(0);
6959 				report_prefix_pop();
6960 
6961 			} else {	// GUEST_PAT
6962 				enter_guest();
6963 				report_guest_state_test("ENT_LOAD_PAT enabled",
6964 							VMX_VMCALL, val,
6965 							"GUEST_PAT");
6966 			}
6967 		}
6968 	}
6969 
6970 	vmcs_set_bits(ctrl_field, ctrl_bit);
6971 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
6972 		/* Test PAT0..PAT7 fields */
6973 		for (j = 0; j < (i ? 8 : 1); j++) {
6974 			val = i << j * 8;
6975 			vmcs_write(field, val);
6976 
6977 			if (field == HOST_PAT) {
6978 				report_prefix_pushf("%s %lx", field_name, val);
6979 				if (i == 0x2 || i == 0x3 || i >= 0x8)
6980 					error =
6981 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
6982 				else
6983 					error = 0;
6984 
6985 				test_vmx_vmlaunch(error);
6986 				report_prefix_pop();
6987 
6988 			} else {	// GUEST_PAT
6989 				if (i == 0x2 || i == 0x3 || i >= 0x8) {
6990 					enter_guest_with_invalid_guest_state();
6991 					report_guest_state_test("ENT_LOAD_PAT "
6992 							        "enabled",
6993 							        VMX_FAIL_STATE | VMX_ENTRY_FAILURE,
6994 							        val,
6995 							        "GUEST_PAT");
6996 				} else {
6997 					enter_guest();
6998 					report_guest_state_test("ENT_LOAD_PAT "
6999 							        "enabled",
7000 							        VMX_VMCALL,
7001 							        val,
7002 							        "GUEST_PAT");
7003 				}
7004 			}
7005 
7006 		}
7007 	}
7008 
7009 	vmcs_write(ctrl_field, ctrl_saved);
7010 	vmcs_write(field, pat_saved);
7011 }
7012 
7013 /*
7014  *  If the "load IA32_PAT" VM-exit control is 1, the value of the field
7015  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7016  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7017  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7018  *  6 (WB), or 7 (UC-).
7019  *
7020  *  [Intel SDM]
7021  */
7022 static void test_load_host_pat(void)
7023 {
7024 	/*
7025 	 * "load IA32_PAT" VM-exit control
7026 	 */
7027 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) {
7028 		printf("\"Load-IA32-PAT\" exit control not supported\n");
7029 		return;
7030 	}
7031 
7032 	test_pat(HOST_PAT, "HOST_PAT", EXI_CONTROLS, EXI_LOAD_PAT);
7033 }
7034 
7035 /*
7036  * test_vmcs_field - test a value for the given VMCS field
7037  * @field: VMCS field
7038  * @field_name: string name of VMCS field
7039  * @bit_start: starting bit
7040  * @bit_end: ending bit
7041  * @val: value that the bit range must or must not contain
7042  * @valid_val: whether value given in 'val' must be valid or not
7043  * @error: expected VMCS error when vmentry fails for an invalid value
7044  */
7045 static void test_vmcs_field(u64 field, const char *field_name, u32 bit_start,
7046 			    u32 bit_end, u64 val, bool valid_val, u32 error)
7047 {
7048 	u64 field_saved = vmcs_read(field);
7049 	u32 i;
7050 	u64 tmp;
7051 	u32 bit_on;
7052 	u64 mask = ~0ull;
7053 
7054 	mask = (mask >> bit_end) << bit_end;
7055 	mask = mask | ((1 << bit_start) - 1);
7056 	tmp = (field_saved & mask) | (val << bit_start);
7057 
7058 	vmcs_write(field, tmp);
7059 	report_prefix_pushf("%s %lx", field_name, tmp);
7060 	if (valid_val)
7061 		test_vmx_vmlaunch(0);
7062 	else
7063 		test_vmx_vmlaunch(error);
7064 	report_prefix_pop();
7065 
7066 	for (i = bit_start; i <= bit_end; i = i + 2) {
7067 		bit_on = ((1ull < i) & (val << bit_start)) ? 0 : 1;
7068 		if (bit_on)
7069 			tmp = field_saved | (1ull << i);
7070 		else
7071 			tmp = field_saved & ~(1ull << i);
7072 		vmcs_write(field, tmp);
7073 		report_prefix_pushf("%s %lx", field_name, tmp);
7074 		if (valid_val)
7075 			test_vmx_vmlaunch(error);
7076 		else
7077 			test_vmx_vmlaunch(0);
7078 		report_prefix_pop();
7079 	}
7080 
7081 	vmcs_write(field, field_saved);
7082 }
7083 
7084 static void test_canonical(u64 field, const char * field_name)
7085 {
7086 	u64 addr_saved = vmcs_read(field);
7087 
7088 	report_prefix_pushf("%s %lx", field_name, addr_saved);
7089 	if (is_canonical(addr_saved)) {
7090 		test_vmx_vmlaunch(0);
7091 		report_prefix_pop();
7092 
7093 		vmcs_write(field, NONCANONICAL);
7094 		report_prefix_pushf("%s %llx", field_name, NONCANONICAL);
7095 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7096 
7097 		vmcs_write(field, addr_saved);
7098 	} else {
7099 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7100 	}
7101 	report_prefix_pop();
7102 }
7103 
7104 #define TEST_RPL_TI_FLAGS(reg, name)				\
7105 	test_vmcs_field(reg, name, 0, 2, 0x0, true,		\
7106 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7107 
7108 #define TEST_CS_TR_FLAGS(reg, name)				\
7109 	test_vmcs_field(reg, name, 3, 15, 0x0000, false,	\
7110 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7111 
7112 /*
7113  * 1. In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the
7114  *    RPL (bits 1:0) and the TI flag (bit 2) must be 0.
7115  * 2. The selector fields for CS and TR cannot be 0000H.
7116  * 3. The selector field for SS cannot be 0000H if the "host address-space
7117  *    size" VM-exit control is 0.
7118  * 4. On processors that support Intel 64 architecture, the base-address
7119  *    fields for FS, GS and TR must contain canonical addresses.
7120  */
7121 static void test_host_segment_regs(void)
7122 {
7123 	u16 selector_saved;
7124 
7125 	/*
7126 	 * Test RPL and TI flags
7127 	 */
7128 	TEST_RPL_TI_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7129 	TEST_RPL_TI_FLAGS(HOST_SEL_SS, "HOST_SEL_SS");
7130 	TEST_RPL_TI_FLAGS(HOST_SEL_DS, "HOST_SEL_DS");
7131 	TEST_RPL_TI_FLAGS(HOST_SEL_ES, "HOST_SEL_ES");
7132 	TEST_RPL_TI_FLAGS(HOST_SEL_FS, "HOST_SEL_FS");
7133 	TEST_RPL_TI_FLAGS(HOST_SEL_GS, "HOST_SEL_GS");
7134 	TEST_RPL_TI_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7135 
7136 	/*
7137 	 * Test that CS and TR fields can not be 0x0000
7138 	 */
7139 	TEST_CS_TR_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7140 	TEST_CS_TR_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7141 
7142 	/*
7143 	 * SS field can not be 0x0000 if "host address-space size" VM-exit
7144 	 * control is 0
7145 	 */
7146 	selector_saved = vmcs_read(HOST_SEL_SS);
7147 	vmcs_write(HOST_SEL_SS, 0);
7148 	report_prefix_pushf("HOST_SEL_SS 0");
7149 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7150 		test_vmx_vmlaunch(0);
7151 	} else {
7152 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7153 	}
7154 	report_prefix_pop();
7155 
7156 	vmcs_write(HOST_SEL_SS, selector_saved);
7157 
7158 #ifdef __x86_64__
7159 	/*
7160 	 * Base address for FS, GS and TR must be canonical
7161 	 */
7162 	test_canonical(HOST_BASE_FS, "HOST_BASE_FS");
7163 	test_canonical(HOST_BASE_GS, "HOST_BASE_GS");
7164 	test_canonical(HOST_BASE_TR, "HOST_BASE_TR");
7165 #endif
7166 }
7167 
7168 /*
7169  *  On processors that support Intel 64 architecture, the base-address
7170  *  fields for GDTR and IDTR must contain canonical addresses.
7171  */
7172 static void test_host_desc_tables(void)
7173 {
7174 #ifdef __x86_64__
7175 	test_canonical(HOST_BASE_GDTR, "HOST_BASE_GDTR");
7176 	test_canonical(HOST_BASE_IDTR, "HOST_BASE_IDTR");
7177 #endif
7178 }
7179 
7180 /*
7181  * If the "host address-space size" VM-exit control is 0, the following must
7182  * hold:
7183  *    - The "IA-32e mode guest" VM-entry control is 0.
7184  *    - Bit 17 of the CR4 field (corresponding to CR4.PCIDE) is 0.
7185  *    - Bits 63:32 in the RIP field are 0.
7186  *
7187  * If the "host address-space size" VM-exit control is 1, the following must
7188  * hold:
7189  *    - Bit 5 of the CR4 field (corresponding to CR4.PAE) is 1.
7190  *    - The RIP field contains a canonical address.
7191  *
7192  */
7193 static void test_host_addr_size(void)
7194 {
7195 	u64 cr4_saved = vmcs_read(HOST_CR4);
7196 	u64 rip_saved = vmcs_read(HOST_RIP);
7197 	u64 entry_ctrl_saved = vmcs_read(ENT_CONTROLS);
7198 	int i;
7199 	u64 tmp;
7200 
7201 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7202 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7203 		report_prefix_pushf("\"IA-32e mode guest\" enabled");
7204 		test_vmx_vmlaunch(0);
7205 		report_prefix_pop();
7206 
7207 		vmcs_write(HOST_CR4, cr4_saved | X86_CR4_PCIDE);
7208 		report_prefix_pushf("\"CR4.PCIDE\" set");
7209 		test_vmx_vmlaunch(0);
7210 		report_prefix_pop();
7211 
7212 		for (i = 32; i <= 63; i = i + 4) {
7213 			tmp = rip_saved | 1ull << i;
7214 			vmcs_write(HOST_RIP, tmp);
7215 			report_prefix_pushf("HOST_RIP %lx", tmp);
7216 			test_vmx_vmlaunch(0);
7217 			report_prefix_pop();
7218 		}
7219 
7220 		if (cr4_saved & X86_CR4_PAE) {
7221 			vmcs_write(HOST_CR4, cr4_saved  & ~X86_CR4_PAE);
7222 			report_prefix_pushf("\"CR4.PAE\" unset");
7223 			test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7224 		} else {
7225 			report_prefix_pushf("\"CR4.PAE\" set");
7226 			test_vmx_vmlaunch(0);
7227 		}
7228 		report_prefix_pop();
7229 
7230 		vmcs_write(HOST_RIP, NONCANONICAL);
7231 		report_prefix_pushf("HOST_RIP %llx", NONCANONICAL);
7232 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7233 		report_prefix_pop();
7234 
7235 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7236 		vmcs_write(HOST_RIP, rip_saved);
7237 		vmcs_write(HOST_CR4, cr4_saved);
7238 	}
7239 }
7240 
7241 /*
7242  * Check that the virtual CPU checks the VMX Host State Area as
7243  * documented in the Intel SDM.
7244  */
7245 static void vmx_host_state_area_test(void)
7246 {
7247 	/*
7248 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
7249 	 * fail due to invalid guest state, should we make it that
7250 	 * far.
7251 	 */
7252 	vmcs_write(GUEST_RFLAGS, 0);
7253 
7254 	test_host_ctl_regs();
7255 
7256 	test_sysenter_field(HOST_SYSENTER_ESP, "HOST_SYSENTER_ESP");
7257 	test_sysenter_field(HOST_SYSENTER_EIP, "HOST_SYSENTER_EIP");
7258 
7259 	test_host_efer();
7260 	test_load_host_pat();
7261 	test_host_segment_regs();
7262 	test_host_desc_tables();
7263 	test_host_addr_size();
7264 }
7265 
7266 /*
7267  *  If the "load IA32_PAT" VM-entry control is 1, the value of the field
7268  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7269  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7270  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7271  *  6 (WB), or 7 (UC-).
7272  *
7273  *  [Intel SDM]
7274  */
7275 static void test_load_guest_pat(void)
7276 {
7277 	/*
7278 	 * "load IA32_PAT" VM-entry control
7279 	 */
7280 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
7281 		printf("\"Load-IA32-PAT\" entry control not supported\n");
7282 		return;
7283 	}
7284 
7285 	test_pat(GUEST_PAT, "GUEST_PAT", ENT_CONTROLS, ENT_LOAD_PAT);
7286 }
7287 
7288 /*
7289  * Check that the virtual CPU checks the VMX Guest State Area as
7290  * documented in the Intel SDM.
7291  */
7292 static void vmx_guest_state_area_test(void)
7293 {
7294 	vmx_set_test_stage(1);
7295 	test_set_guest(guest_state_test_main);
7296 
7297 	test_load_guest_pat();
7298 	test_guest_efer();
7299 
7300 	/*
7301 	 * Let the guest finish execution
7302 	 */
7303 	vmx_set_test_stage(2);
7304 	enter_guest();
7305 }
7306 
7307 static bool valid_vmcs_for_vmentry(void)
7308 {
7309 	struct vmcs *current_vmcs = NULL;
7310 
7311 	if (vmcs_save(&current_vmcs))
7312 		return false;
7313 
7314 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
7315 }
7316 
7317 static void try_vmentry_in_movss_shadow(void)
7318 {
7319 	u32 vm_inst_err;
7320 	u32 flags;
7321 	bool early_failure = false;
7322 	u32 expected_flags = X86_EFLAGS_FIXED;
7323 	bool valid_vmcs = valid_vmcs_for_vmentry();
7324 
7325 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
7326 
7327 	/*
7328 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
7329 	 * unsupported VMCS component").
7330 	 */
7331 	vmcs_write(~0u, 0);
7332 
7333 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
7334 			      "vmwrite %%rsp, %%rdx;"
7335 			      "mov 0f, %%rax;"
7336 			      "mov %[host_rip], %%edx;"
7337 			      "vmwrite %%rax, %%rdx;"
7338 			      "mov $-1, %%ah;"
7339 			      "sahf;"
7340 			      "mov %%ss, %%ax;"
7341 			      "mov %%ax, %%ss;"
7342 			      "vmlaunch;"
7343 			      "mov $1, %[early_failure];"
7344 			      "0: lahf;"
7345 			      "movzbl %%ah, %[flags]"
7346 			      : [early_failure] "+r" (early_failure),
7347 				[flags] "=&a" (flags)
7348 			      : [host_rsp] "i" (HOST_RSP),
7349 				[host_rip] "i" (HOST_RIP)
7350 			      : "rdx", "cc", "memory");
7351 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
7352 
7353 	report("Early VM-entry failure", early_failure);
7354 	report("RFLAGS[8:0] is %x (actual %x)", flags == expected_flags,
7355 	       expected_flags, flags);
7356 	if (valid_vmcs)
7357 		report("VM-instruction error is %d (actual %d)",
7358 		       vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
7359 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
7360 }
7361 
7362 static void vmentry_movss_shadow_test(void)
7363 {
7364 	struct vmcs *orig_vmcs;
7365 
7366 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
7367 
7368 	/*
7369 	 * Set the launched flag on the current VMCS to verify the correct
7370 	 * error priority, below.
7371 	 */
7372 	test_set_guest(v2_null_test_guest);
7373 	enter_guest();
7374 
7375 	/*
7376 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
7377 	 * fail due to invalid guest state (if we make it that far).
7378 	 */
7379 	vmcs_write(GUEST_RFLAGS, 0);
7380 
7381 	/*
7382 	 * "VM entry with events blocked by MOV SS" takes precedence over
7383 	 * "VMLAUNCH with non-clear VMCS."
7384 	 */
7385 	report_prefix_push("valid current-VMCS");
7386 	try_vmentry_in_movss_shadow();
7387 	report_prefix_pop();
7388 
7389 	/*
7390 	 * VMfailInvalid takes precedence over "VM entry with events
7391 	 * blocked by MOV SS."
7392 	 */
7393 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
7394 	report_prefix_push("no current-VMCS");
7395 	try_vmentry_in_movss_shadow();
7396 	report_prefix_pop();
7397 
7398 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
7399 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
7400 }
7401 
7402 static void vmx_cr_load_test(void)
7403 {
7404 	unsigned long cr3, cr4, orig_cr3, orig_cr4;
7405 
7406 	orig_cr4 = read_cr4();
7407 	orig_cr3 = read_cr3();
7408 
7409 	if (!this_cpu_has(X86_FEATURE_PCID)) {
7410 		report_skip("PCID not detected");
7411 		return;
7412 	}
7413 	if (!this_cpu_has(X86_FEATURE_MCE)) {
7414 		report_skip("MCE not detected");
7415 		return;
7416 	}
7417 
7418 	TEST_ASSERT(!(orig_cr3 & X86_CR3_PCID_MASK));
7419 
7420 	/* Enable PCID for L1. */
7421 	cr4 = orig_cr4 | X86_CR4_PCIDE;
7422 	cr3 = orig_cr3 | 0x1;
7423 	TEST_ASSERT(!write_cr4_checking(cr4));
7424 	write_cr3(cr3);
7425 
7426 	test_set_guest(v2_null_test_guest);
7427 	vmcs_write(HOST_CR4, cr4);
7428 	vmcs_write(HOST_CR3, cr3);
7429 	enter_guest();
7430 
7431 	/*
7432 	 * No exception is expected.
7433 	 *
7434 	 * NB. KVM loads the last guest write to CR4 into CR4 read
7435 	 *     shadow. In order to trigger an exit to KVM, we can toggle a
7436 	 *     bit that is owned by KVM. We use CR4.MCE, which shall
7437 	 *     have no side effect because normally no guest MCE (e.g., as the
7438 	 *     result of bad memory) would happen during this test.
7439 	 */
7440 	TEST_ASSERT(!write_cr4_checking(cr4 ^ X86_CR4_MCE));
7441 
7442 	/* Cleanup L1 state. */
7443 	write_cr3(orig_cr3);
7444 	TEST_ASSERT(!write_cr4_checking(orig_cr4));
7445 }
7446 
7447 static void vmx_nm_test_guest(void)
7448 {
7449 	write_cr0(read_cr0() | X86_CR0_TS);
7450 	asm volatile("fnop");
7451 }
7452 
7453 static void check_nm_exit(const char *test)
7454 {
7455 	u32 reason = vmcs_read(EXI_REASON);
7456 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
7457 	const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
7458 		NM_VECTOR;
7459 
7460 	report("%s", reason == VMX_EXC_NMI && intr_info == expected, test);
7461 }
7462 
7463 /*
7464  * This test checks that:
7465  *
7466  * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then
7467  *     a subsequent #NM VM-exit is reflected to L1.
7468  *
7469  * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a
7470  *     subsequent #NM VM-exit is reflected to L1.
7471  */
7472 static void vmx_nm_test(void)
7473 {
7474 	unsigned long cr0 = read_cr0();
7475 
7476 	test_set_guest(vmx_nm_test_guest);
7477 
7478 	/*
7479 	 * L1 wants to intercept #NM exceptions encountered in L2.
7480 	 */
7481 	vmcs_write(EXC_BITMAP, 1 << NM_VECTOR);
7482 
7483 	/*
7484 	 * Launch L2 with CR0.TS clear, but don't claim host ownership of
7485 	 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop,
7486 	 * which will raise #NM. L0 should reflect the #NM VM-exit to L1.
7487 	 */
7488 	vmcs_write(CR0_MASK, 0);
7489 	vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS);
7490 	enter_guest();
7491 	check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1");
7492 
7493 	/*
7494 	 * Re-enter L2 at the fnop instruction, with CR0.TS clear but
7495 	 * CR0.EM set. The fnop will still raise #NM, and L0 should
7496 	 * reflect the #NM VM-exit to L1.
7497 	 */
7498 	vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM);
7499 	enter_guest();
7500 	check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1");
7501 
7502 	/*
7503 	 * Re-enter L2 at the fnop instruction, with both CR0.TS and
7504 	 * CR0.EM clear. There will be no #NM, and the L2 guest should
7505 	 * exit normally.
7506 	 */
7507 	vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM));
7508 	enter_guest();
7509 }
7510 
7511 bool vmx_pending_event_ipi_fired;
7512 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
7513 {
7514 	vmx_pending_event_ipi_fired = true;
7515 	eoi();
7516 }
7517 
7518 bool vmx_pending_event_guest_run;
7519 static void vmx_pending_event_guest(void)
7520 {
7521 	vmcall();
7522 	vmx_pending_event_guest_run = true;
7523 }
7524 
7525 static void vmx_pending_event_test_core(bool guest_hlt)
7526 {
7527 	int ipi_vector = 0xf1;
7528 
7529 	vmx_pending_event_ipi_fired = false;
7530 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
7531 
7532 	vmx_pending_event_guest_run = false;
7533 	test_set_guest(vmx_pending_event_guest);
7534 
7535 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
7536 
7537 	enter_guest();
7538 	skip_exit_vmcall();
7539 
7540 	if (guest_hlt)
7541 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7542 
7543 	irq_disable();
7544 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
7545 				   APIC_DM_FIXED | ipi_vector,
7546 				   0);
7547 
7548 	enter_guest();
7549 
7550 	assert_exit_reason(VMX_EXTINT);
7551 	report("Guest did not run before host received IPI",
7552 		   !vmx_pending_event_guest_run);
7553 
7554 	irq_enable();
7555 	asm volatile ("nop");
7556 	irq_disable();
7557 	report("Got pending interrupt after IRQ enabled",
7558 		   vmx_pending_event_ipi_fired);
7559 
7560 	if (guest_hlt)
7561 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
7562 
7563 	enter_guest();
7564 	report("Guest finished running when no interrupt",
7565 		   vmx_pending_event_guest_run);
7566 }
7567 
7568 static void vmx_pending_event_test(void)
7569 {
7570 	vmx_pending_event_test_core(false);
7571 }
7572 
7573 static void vmx_pending_event_hlt_test(void)
7574 {
7575 	vmx_pending_event_test_core(true);
7576 }
7577 
7578 static int vmx_window_test_db_count;
7579 
7580 static void vmx_window_test_db_handler(struct ex_regs *regs)
7581 {
7582 	vmx_window_test_db_count++;
7583 }
7584 
7585 static void vmx_nmi_window_test_guest(void)
7586 {
7587 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
7588 
7589 	asm volatile("vmcall\n\t"
7590 		     "nop\n\t");
7591 
7592 	handle_exception(DB_VECTOR, NULL);
7593 }
7594 
7595 static void verify_nmi_window_exit(u64 rip)
7596 {
7597 	u32 exit_reason = vmcs_read(EXI_REASON);
7598 
7599 	report("Exit reason (%d) is 'NMI window'",
7600 	       exit_reason == VMX_NMI_WINDOW, exit_reason);
7601 	report("RIP (%#lx) is %#lx", vmcs_read(GUEST_RIP) == rip,
7602 	       vmcs_read(GUEST_RIP), rip);
7603 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
7604 }
7605 
7606 static void vmx_nmi_window_test(void)
7607 {
7608 	u64 nop_addr;
7609 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
7610 
7611 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
7612 		report_skip("CPU does not support the \"Virtual NMIs\" VM-execution control.");
7613 		return;
7614 	}
7615 
7616 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
7617 		report_skip("CPU does not support the \"NMI-window exiting\" VM-execution control.");
7618 		return;
7619 	}
7620 
7621 	vmx_window_test_db_count = 0;
7622 
7623 	report_prefix_push("NMI-window");
7624 	test_set_guest(vmx_nmi_window_test_guest);
7625 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
7626 	enter_guest();
7627 	skip_exit_vmcall();
7628 	nop_addr = vmcs_read(GUEST_RIP);
7629 
7630 	/*
7631 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
7632 	 * RIP will not advance.
7633 	 */
7634 	report_prefix_push("active, no blocking");
7635 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
7636 	enter_guest();
7637 	verify_nmi_window_exit(nop_addr);
7638 	report_prefix_pop();
7639 
7640 	/*
7641 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
7642 	 * a VM-exit on the next instruction after the nop. (The nop
7643 	 * is one byte.)
7644 	 */
7645 	report_prefix_push("active, blocking by MOV-SS");
7646 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
7647 	enter_guest();
7648 	verify_nmi_window_exit(nop_addr + 1);
7649 	report_prefix_pop();
7650 
7651 	/*
7652 	 * Ask for "NMI-window exiting" (with event injection), and
7653 	 * expect a VM-exit after the event is injected. (RIP should
7654 	 * be at the address specified in the IDT entry for #DB.)
7655 	 */
7656 	report_prefix_push("active, no blocking, injecting #DB");
7657 	vmcs_write(ENT_INTR_INFO,
7658 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
7659 	enter_guest();
7660 	verify_nmi_window_exit((u64)db_fault_addr);
7661 	report_prefix_pop();
7662 
7663 	/*
7664 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
7665 	 * a VM-exit after the next IRET (i.e. after the #DB handler
7666 	 * returns). So, RIP should be back at one byte past the nop.
7667 	 */
7668 	report_prefix_push("active, blocking by NMI");
7669 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
7670 	enter_guest();
7671 	verify_nmi_window_exit(nop_addr + 1);
7672 	report("#DB handler executed once (actual %d times)",
7673 	       vmx_window_test_db_count == 1,
7674 	       vmx_window_test_db_count);
7675 	report_prefix_pop();
7676 
7677 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
7678 		report_skip("CPU does not support activity state HLT.");
7679 	} else {
7680 		/*
7681 		 * Ask for "NMI-window exiting" when entering activity
7682 		 * state HLT, and expect an immediate VM-exit. RIP is
7683 		 * still one byte past the nop.
7684 		 */
7685 		report_prefix_push("halted, no blocking");
7686 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7687 		enter_guest();
7688 		verify_nmi_window_exit(nop_addr + 1);
7689 		report_prefix_pop();
7690 
7691 		/*
7692 		 * Ask for "NMI-window exiting" when entering activity
7693 		 * state HLT (with event injection), and expect a
7694 		 * VM-exit after the event is injected. (RIP should be
7695 		 * at the address specified in the IDT entry for #DB.)
7696 		 */
7697 		report_prefix_push("halted, no blocking, injecting #DB");
7698 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7699 		vmcs_write(ENT_INTR_INFO,
7700 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
7701 			   DB_VECTOR);
7702 		enter_guest();
7703 		verify_nmi_window_exit((u64)db_fault_addr);
7704 		report_prefix_pop();
7705 	}
7706 
7707 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
7708 	enter_guest();
7709 	report_prefix_pop();
7710 }
7711 
7712 static void vmx_intr_window_test_guest(void)
7713 {
7714 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
7715 
7716 	/*
7717 	 * The two consecutive STIs are to ensure that only the first
7718 	 * one has a shadow. Note that NOP and STI are one byte
7719 	 * instructions.
7720 	 */
7721 	asm volatile("vmcall\n\t"
7722 		     "nop\n\t"
7723 		     "sti\n\t"
7724 		     "sti\n\t");
7725 
7726 	handle_exception(DB_VECTOR, NULL);
7727 }
7728 
7729 static void verify_intr_window_exit(u64 rip)
7730 {
7731 	u32 exit_reason = vmcs_read(EXI_REASON);
7732 
7733 	report("Exit reason (%d) is 'interrupt window'",
7734 	       exit_reason == VMX_INTR_WINDOW, exit_reason);
7735 	report("RIP (%#lx) is %#lx", vmcs_read(GUEST_RIP) == rip,
7736 	       vmcs_read(GUEST_RIP), rip);
7737 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
7738 }
7739 
7740 static void vmx_intr_window_test(void)
7741 {
7742 	u64 vmcall_addr;
7743 	u64 nop_addr;
7744 	unsigned int orig_db_gate_type;
7745 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
7746 
7747 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
7748 		report_skip("CPU does not support the \"interrupt-window exiting\" VM-execution control.");
7749 		return;
7750 	}
7751 
7752 	/*
7753 	 * Change the IDT entry for #DB from interrupt gate to trap gate,
7754 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
7755 	 * be disabled after vectoring a #DB.
7756 	 */
7757 	orig_db_gate_type = boot_idt[DB_VECTOR].type;
7758 	boot_idt[DB_VECTOR].type = 15;
7759 
7760 	report_prefix_push("interrupt-window");
7761 	test_set_guest(vmx_intr_window_test_guest);
7762 	enter_guest();
7763 	assert_exit_reason(VMX_VMCALL);
7764 	vmcall_addr = vmcs_read(GUEST_RIP);
7765 
7766 	/*
7767 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
7768 	 * no blocking; expect an immediate VM-exit. Note that we have
7769 	 * not advanced past the vmcall instruction yet, so RIP should
7770 	 * point to the vmcall instruction.
7771 	 */
7772 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
7773 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
7774 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
7775 	enter_guest();
7776 	verify_intr_window_exit(vmcall_addr);
7777 	report_prefix_pop();
7778 
7779 	/*
7780 	 * Ask for "interrupt-window exiting" (with event injection)
7781 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
7782 	 * the event is injected. That is, RIP should should be at the
7783 	 * address specified in the IDT entry for #DB.
7784 	 */
7785 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #DB");
7786 	vmcs_write(ENT_INTR_INFO,
7787 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
7788 	vmcall_addr = vmcs_read(GUEST_RIP);
7789 	enter_guest();
7790 	verify_intr_window_exit((u64)db_fault_addr);
7791 	report_prefix_pop();
7792 
7793 	/*
7794 	 * Let the L2 guest run through the IRET, back to the VMCALL.
7795 	 * We have to clear the "interrupt-window exiting"
7796 	 * VM-execution control, or it would just keep causing
7797 	 * VM-exits. Then, advance past the VMCALL and set the
7798 	 * "interrupt-window exiting" VM-execution control again.
7799 	 */
7800 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
7801 	enter_guest();
7802 	skip_exit_vmcall();
7803 	nop_addr = vmcs_read(GUEST_RIP);
7804 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
7805 
7806 	/*
7807 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
7808 	 * RFLAGS.IF set, and expect a VM-exit on the next
7809 	 * instruction. (NOP is one byte.)
7810 	 */
7811 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
7812 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
7813 	enter_guest();
7814 	verify_intr_window_exit(nop_addr + 1);
7815 	report_prefix_pop();
7816 
7817 	/*
7818 	 * Back up to the NOP and ask for "interrupt-window exiting"
7819 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
7820 	 * on the next instruction. (NOP is one byte.)
7821 	 */
7822 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
7823 	vmcs_write(GUEST_RIP, nop_addr);
7824 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
7825 	enter_guest();
7826 	verify_intr_window_exit(nop_addr + 1);
7827 	report_prefix_pop();
7828 
7829 	/*
7830 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
7831 	 * and expect a VM-exit on the instruction following the STI
7832 	 * shadow. Only the first STI (which is one byte past the NOP)
7833 	 * should have a shadow. The second STI (which is two bytes
7834 	 * past the NOP) has no shadow. Therefore, the interrupt
7835 	 * window opens at three bytes past the NOP.
7836 	 */
7837 	report_prefix_push("active, RFLAGS.IF = 0");
7838 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
7839 	enter_guest();
7840 	verify_intr_window_exit(nop_addr + 3);
7841 	report_prefix_pop();
7842 
7843 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
7844 		report_skip("CPU does not support activity state HLT.");
7845 	} else {
7846 		/*
7847 		 * Ask for "interrupt-window exiting" when entering
7848 		 * activity state HLT, and expect an immediate
7849 		 * VM-exit. RIP is still three bytes past the nop.
7850 		 */
7851 		report_prefix_push("halted, no blocking");
7852 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7853 		enter_guest();
7854 		verify_intr_window_exit(nop_addr + 3);
7855 		report_prefix_pop();
7856 
7857 		/*
7858 		 * Ask for "interrupt-window exiting" when entering
7859 		 * activity state HLT (with event injection), and
7860 		 * expect a VM-exit after the event is injected. That
7861 		 * is, RIP should should be at the address specified
7862 		 * in the IDT entry for #DB.
7863 		 */
7864 		report_prefix_push("halted, no blocking, injecting #DB");
7865 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7866 		vmcs_write(ENT_INTR_INFO,
7867 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
7868 			   DB_VECTOR);
7869 		enter_guest();
7870 		verify_intr_window_exit((u64)db_fault_addr);
7871 		report_prefix_pop();
7872 	}
7873 
7874 	boot_idt[DB_VECTOR].type = orig_db_gate_type;
7875 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
7876 	enter_guest();
7877 	report_prefix_pop();
7878 }
7879 
7880 #define GUEST_TSC_OFFSET (1u << 30)
7881 
7882 static u64 guest_tsc;
7883 
7884 static void vmx_store_tsc_test_guest(void)
7885 {
7886 	guest_tsc = rdtsc();
7887 }
7888 
7889 /*
7890  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
7891  * list, the value saved is not subject to the TSC offset that is
7892  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
7893  */
7894 static void vmx_store_tsc_test(void)
7895 {
7896 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
7897 	u64 low, high;
7898 
7899 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
7900 		report_skip("'Use TSC offsetting' not supported");
7901 		return;
7902 	}
7903 
7904 	test_set_guest(vmx_store_tsc_test_guest);
7905 
7906 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
7907 	vmcs_write(EXI_MSR_ST_CNT, 1);
7908 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
7909 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
7910 
7911 	low = rdtsc();
7912 	enter_guest();
7913 	high = rdtsc();
7914 
7915 	report("RDTSC value in the guest (%lu) is in range [%lu, %lu]",
7916 	       low + GUEST_TSC_OFFSET <= guest_tsc &&
7917 	       guest_tsc <= high + GUEST_TSC_OFFSET,
7918 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
7919 	report("IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
7920 	       low <= msr_entry.value && msr_entry.value <= high,
7921 	       msr_entry.value, low, high);
7922 }
7923 
7924 static void vmx_db_test_guest(void)
7925 {
7926 	/*
7927 	 * For a hardware generated single-step #DB.
7928 	 */
7929 	asm volatile("vmcall;"
7930 		     "nop;"
7931 		     ".Lpost_nop:");
7932 	/*
7933 	 * ...in a MOVSS shadow, with pending debug exceptions.
7934 	 */
7935 	asm volatile("vmcall;"
7936 		     "nop;"
7937 		     ".Lpost_movss_nop:");
7938 	/*
7939 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
7940 	 * emulates it in software.)
7941 	 */
7942 	asm volatile("vmcall;"
7943 		     "wbinvd;"
7944 		     ".Lpost_wbinvd:");
7945 	/*
7946 	 * ...in a MOVSS shadow, with pending debug exceptions.
7947 	 */
7948 	asm volatile("vmcall;"
7949 		     "wbinvd;"
7950 		     ".Lpost_movss_wbinvd:");
7951 	/*
7952 	 * For a hardware generated single-step #DB in a transactional region.
7953 	 */
7954 	asm volatile("vmcall;"
7955 		     ".Lxbegin: xbegin .Lskip_rtm;"
7956 		     "xend;"
7957 		     ".Lskip_rtm:");
7958 }
7959 
7960 /*
7961  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
7962  * L2. No #DB is delivered and L2 continues to the next point of
7963  * interest.
7964  */
7965 static void dismiss_db(void)
7966 {
7967 	vmcs_write(GUEST_PENDING_DEBUG, 0);
7968 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
7969 	enter_guest();
7970 }
7971 
7972 /*
7973  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
7974  * Then throw away the #DB exception and resume L2.
7975  */
7976 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
7977 			  void *expected_rip, u64 expected_exit_qual,
7978 			  u64 expected_dr6)
7979 {
7980 	u32 reason = vmcs_read(EXI_REASON);
7981 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
7982 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
7983 	u64 guest_rip = vmcs_read(GUEST_RIP);
7984 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
7985 	u64 dr6 = read_dr6();
7986 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
7987 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
7988 
7989 	report("Expected #DB VM-exit",
7990 	       reason == VMX_EXC_NMI && intr_info == expected_intr_info);
7991 	report("Expected RIP %p (actual %lx)", (u64)expected_rip == guest_rip,
7992 	       expected_rip, guest_rip);
7993 	report_xfail("Expected pending debug exceptions 0 (actual %lx)",
7994 		     xfail_pdbg, 0 == guest_pending_dbg, guest_pending_dbg);
7995 	report_xfail("Expected exit qualification %lx (actual %lx)", xfail_qual,
7996 		     expected_exit_qual == exit_qual,
7997 		     expected_exit_qual, exit_qual);
7998 	report_xfail("Expected DR6 %lx (actual %lx)", xfail_dr6,
7999 		     expected_dr6 == dr6, expected_dr6, dr6);
8000 	dismiss_db();
8001 }
8002 
8003 /*
8004  * Assuming the guest has just exited on a VMCALL instruction, skip
8005  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
8006  * pending debug exceptions are non-zero, set the VMCS up as if the
8007  * previous instruction was a MOVSS that generated the indicated
8008  * pending debug exceptions. Then enter L2.
8009  */
8010 static void single_step_guest(const char *test_name, u64 starting_dr6,
8011 			      u64 pending_debug_exceptions)
8012 {
8013 	printf("\n%s\n", test_name);
8014 	skip_exit_vmcall();
8015 	write_dr6(starting_dr6);
8016 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
8017 	if (pending_debug_exceptions) {
8018 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
8019 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8020 	}
8021 	enter_guest();
8022 }
8023 
8024 /*
8025  * When L1 intercepts #DB, verify that a single-step trap clears
8026  * pending debug exceptions, populates the exit qualification field
8027  * properly, and that DR6 is not prematurely clobbered. In a
8028  * (simulated) MOVSS shadow, make sure that the pending debug
8029  * exception bits are properly accumulated into the exit qualification
8030  * field.
8031  */
8032 static void vmx_db_test(void)
8033 {
8034 	/*
8035 	 * We are going to set a few arbitrary bits in DR6 to verify that
8036 	 * (a) DR6 is not modified by an intercepted #DB, and
8037 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
8038          *     the exit qualification field for a subsequent #DB exception.
8039 	 */
8040 	const u64 starting_dr6 = DR6_RESERVED | BIT(13) | DR_TRAP3 | DR_TRAP1;
8041 	extern char post_nop asm(".Lpost_nop");
8042 	extern char post_movss_nop asm(".Lpost_movss_nop");
8043 	extern char post_wbinvd asm(".Lpost_wbinvd");
8044 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
8045 	extern char xbegin asm(".Lxbegin");
8046 	extern char skip_rtm asm(".Lskip_rtm");
8047 
8048 	/*
8049 	 * L1 wants to intercept #DB exceptions encountered in L2.
8050 	 */
8051 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
8052 
8053 	/*
8054 	 * Start L2 and run it up to the first point of interest.
8055 	 */
8056 	test_set_guest(vmx_db_test_guest);
8057 	enter_guest();
8058 
8059 	/*
8060 	 * Hardware-delivered #DB trap for single-step sets the
8061 	 * standard that L0 has to follow for emulated instructions.
8062 	 */
8063 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
8064 	check_db_exit(false, false, false, &post_nop, DR_STEP, starting_dr6);
8065 
8066 	/*
8067 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
8068 	 * also sets the standard that L0 has to follow for emulated
8069 	 * instructions. Here, we establish the VMCS pending debug
8070 	 * exceptions to indicate that the simulated MOVSS triggered a
8071 	 * data breakpoint as well as the single-step trap.
8072 	 */
8073 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
8074 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0 );
8075 	check_db_exit(false, false, false, &post_movss_nop, DR_STEP | DR_TRAP0,
8076 		      starting_dr6);
8077 
8078 	/*
8079 	 * L0 synthesized #DB trap for single-step is buggy, because
8080 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
8081 	 * reconstitute the exit qualification from the prematurely
8082 	 * modified DR6, but fails miserably.
8083 	 */
8084 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
8085 	check_db_exit(true, true, false, &post_wbinvd, DR_STEP, starting_dr6);
8086 
8087 	/*
8088 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
8089 	 * even worse, because L0 also leaves the pending debug
8090 	 * exceptions in the VMCS instead of accumulating them into
8091 	 * the exit qualification field for the #DB exception.
8092 	 */
8093 	single_step_guest("Software synthesized single-step in MOVSS shadow",
8094 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0);
8095 	check_db_exit(true, true, true, &post_movss_wbinvd, DR_STEP | DR_TRAP0,
8096 		      starting_dr6);
8097 
8098 	/*
8099 	 * Optional RTM test for hardware that supports RTM, to
8100 	 * demonstrate that the current volume 3 of the SDM
8101 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
8102 	 * qualification for debug exceptions is not reserved. It is
8103 	 * set to 1 if a debug exception (#DB) or a breakpoint
8104 	 * exception (#BP) occurs inside an RTM region while advanced
8105 	 * debugging of RTM transactional regions is enabled.
8106 	 */
8107 	if (this_cpu_has(X86_FEATURE_RTM)) {
8108 		vmcs_write(ENT_CONTROLS,
8109 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
8110 		/*
8111 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
8112 		 * in the guest to enable advanced debugging of RTM
8113 		 * transactional regions.
8114 		 */
8115 		vmcs_write(GUEST_DR7, BIT(11));
8116 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
8117 		single_step_guest("Hardware delivered single-step in "
8118 				  "transactional region", starting_dr6, 0);
8119 		check_db_exit(false, false, false, &xbegin, BIT(16),
8120 			      starting_dr6);
8121 	} else {
8122 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
8123 		enter_guest();
8124 	}
8125 }
8126 
8127 static void enable_vid(void)
8128 {
8129 	void *virtual_apic_page;
8130 
8131 	assert(cpu_has_apicv());
8132 
8133 	disable_intercept_for_x2apic_msrs();
8134 
8135 	virtual_apic_page = alloc_page();
8136 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
8137 
8138 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
8139 
8140 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
8141 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
8142 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
8143 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
8144 
8145 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
8146 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
8147 }
8148 
8149 static void trigger_ioapic_scan_thread(void *data)
8150 {
8151 	/* Wait until other CPU entered L2 */
8152 	while (vmx_get_test_stage() != 1)
8153 		;
8154 
8155 	/* Trigger ioapic scan */
8156 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
8157 	vmx_set_test_stage(2);
8158 }
8159 
8160 static void irq_79_handler_guest(isr_regs_t *regs)
8161 {
8162 	eoi();
8163 
8164 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
8165 	vmcall();
8166 }
8167 
8168 /*
8169  * Constant for num of busy-loop iterations after which
8170  * a timer interrupt should have happened in host
8171  */
8172 #define TIMER_INTERRUPT_DELAY 100000000
8173 
8174 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
8175 {
8176 	handle_irq(0x79, irq_79_handler_guest);
8177 	irq_enable();
8178 
8179 	/* Signal to L1 CPU to trigger ioapic scan */
8180 	vmx_set_test_stage(1);
8181 	/* Wait until L1 CPU to trigger ioapic scan */
8182 	while (vmx_get_test_stage() != 2)
8183 		;
8184 
8185 	/*
8186 	 * Wait for L0 timer interrupt to be raised while we run in L2
8187 	 * such that L0 will process the IOAPIC scan request before
8188 	 * resuming L2
8189 	 */
8190 	delay(TIMER_INTERRUPT_DELAY);
8191 
8192 	asm volatile ("int $0x79");
8193 }
8194 
8195 static void vmx_eoi_bitmap_ioapic_scan_test(void)
8196 {
8197 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
8198 		report_skip(__func__);
8199 		return;
8200 	}
8201 
8202 	enable_vid();
8203 
8204 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
8205 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
8206 
8207 	/*
8208 	 * Launch L2.
8209 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
8210 	 * In case the reason isn't VMX_VMCALL, the asserion inside
8211 	 * skip_exit_vmcall() will fail.
8212 	 */
8213 	enter_guest();
8214 	skip_exit_vmcall();
8215 
8216 	/* Let L2 finish */
8217 	enter_guest();
8218 	report(__func__, 1);
8219 }
8220 
8221 #define HLT_WITH_RVI_VECTOR		(0xf1)
8222 
8223 bool vmx_hlt_with_rvi_guest_isr_fired;
8224 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
8225 {
8226 	vmx_hlt_with_rvi_guest_isr_fired = true;
8227 	eoi();
8228 }
8229 
8230 static void vmx_hlt_with_rvi_guest(void)
8231 {
8232 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
8233 
8234 	irq_enable();
8235 	asm volatile ("nop");
8236 
8237 	vmcall();
8238 }
8239 
8240 static void vmx_hlt_with_rvi_test(void)
8241 {
8242 	if (!cpu_has_apicv()) {
8243 		report_skip(__func__);
8244 		return;
8245 	}
8246 
8247 	enable_vid();
8248 
8249 	vmx_hlt_with_rvi_guest_isr_fired = false;
8250 	test_set_guest(vmx_hlt_with_rvi_guest);
8251 
8252 	enter_guest();
8253 	skip_exit_vmcall();
8254 
8255 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8256 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
8257 	enter_guest();
8258 
8259 	report("Interrupt raised in guest", vmx_hlt_with_rvi_guest_isr_fired);
8260 }
8261 
8262 static void set_irq_line_thread(void *data)
8263 {
8264 	/* Wait until other CPU entered L2 */
8265 	while (vmx_get_test_stage() != 1)
8266 		;
8267 
8268 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
8269 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
8270 	vmx_set_test_stage(2);
8271 }
8272 
8273 static bool irq_78_handler_vmcall_before_eoi;
8274 static void irq_78_handler_guest(isr_regs_t *regs)
8275 {
8276 	set_irq_line(0xf, 0);
8277 	if (irq_78_handler_vmcall_before_eoi)
8278 		vmcall();
8279 	eoi();
8280 	vmcall();
8281 }
8282 
8283 static void vmx_apic_passthrough_guest(void)
8284 {
8285 	handle_irq(0x78, irq_78_handler_guest);
8286 	irq_enable();
8287 
8288 	/* If requested, wait for other CPU to trigger ioapic scan */
8289 	if (vmx_get_test_stage() < 1) {
8290 		vmx_set_test_stage(1);
8291 		while (vmx_get_test_stage() != 2)
8292 			;
8293 	}
8294 
8295 	set_irq_line(0xf, 1);
8296 }
8297 
8298 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
8299 {
8300 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
8301 		report_skip(__func__);
8302 		return;
8303 	}
8304 
8305 	/* Test device is required for generating IRQs */
8306 	if (!test_device_enabled()) {
8307 		report_skip(__func__);
8308 		return;
8309 	}
8310 	u64 cpu_ctrl_0 = CPU_SECONDARY;
8311 	u64 cpu_ctrl_1 = 0;
8312 
8313 	disable_intercept_for_x2apic_msrs();
8314 
8315 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
8316 
8317 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
8318 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
8319 
8320 	if (set_irq_line_from_thread) {
8321 		irq_78_handler_vmcall_before_eoi = false;
8322 		on_cpu_async(1, set_irq_line_thread, NULL);
8323 	} else {
8324 		irq_78_handler_vmcall_before_eoi = true;
8325 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
8326 		vmx_set_test_stage(2);
8327 	}
8328 	test_set_guest(vmx_apic_passthrough_guest);
8329 
8330 	if (irq_78_handler_vmcall_before_eoi) {
8331 		/* Before EOI remote_irr should still be set */
8332 		enter_guest();
8333 		skip_exit_vmcall();
8334 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
8335 			"IOAPIC pass-through: remote_irr=1 before EOI");
8336 	}
8337 
8338 	/* After EOI remote_irr should be cleared */
8339 	enter_guest();
8340 	skip_exit_vmcall();
8341 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
8342 		"IOAPIC pass-through: remote_irr=0 after EOI");
8343 
8344 	/* Let L2 finish */
8345 	enter_guest();
8346 	report(__func__, 1);
8347 }
8348 
8349 static void vmx_apic_passthrough_test(void)
8350 {
8351 	vmx_apic_passthrough(false);
8352 }
8353 
8354 static void vmx_apic_passthrough_thread_test(void)
8355 {
8356 	vmx_apic_passthrough(true);
8357 }
8358 
8359 static void vmx_apic_passthrough_tpr_threshold_guest(void)
8360 {
8361 	cli();
8362 	apic_set_tpr(0);
8363 }
8364 
8365 static bool vmx_apic_passthrough_tpr_threshold_ipi_isr_fired;
8366 static void vmx_apic_passthrough_tpr_threshold_ipi_isr(isr_regs_t *regs)
8367 {
8368 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = true;
8369 	eoi();
8370 }
8371 
8372 static void vmx_apic_passthrough_tpr_threshold_test(void)
8373 {
8374 	int ipi_vector = 0xe1;
8375 
8376 	disable_intercept_for_x2apic_msrs();
8377 	vmcs_clear_bits(PIN_CONTROLS, PIN_EXTINT);
8378 
8379 	/* Raise L0 TPR-threshold by queueing vector in LAPIC IRR */
8380 	cli();
8381 	apic_set_tpr((ipi_vector >> 4) + 1);
8382 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
8383 			APIC_DM_FIXED | ipi_vector,
8384 			0);
8385 
8386 	test_set_guest(vmx_apic_passthrough_tpr_threshold_guest);
8387 	enter_guest();
8388 
8389 	report("TPR was zero by guest", apic_get_tpr() == 0);
8390 
8391 	/* Clean pending self-IPI */
8392 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = false;
8393 	handle_irq(ipi_vector, vmx_apic_passthrough_tpr_threshold_ipi_isr);
8394 	sti();
8395 	asm volatile ("nop");
8396 	report("self-IPI fired", vmx_apic_passthrough_tpr_threshold_ipi_isr_fired);
8397 
8398 	report(__func__, 1);
8399 }
8400 
8401 static u64 init_signal_test_exit_reason;
8402 static bool init_signal_test_thread_continued;
8403 
8404 static void init_signal_test_thread(void *data)
8405 {
8406 	struct vmcs *test_vmcs = data;
8407 
8408 	/* Enter VMX operation (i.e. exec VMXON) */
8409 	u64 *ap_vmxon_region = alloc_page();
8410 	enable_vmx();
8411 	init_vmx(ap_vmxon_region);
8412 	_vmx_on(ap_vmxon_region);
8413 
8414 	/* Signal CPU have entered VMX operation */
8415 	vmx_set_test_stage(1);
8416 
8417 	/* Wait for BSP CPU to send INIT signal */
8418 	while (vmx_get_test_stage() != 2)
8419 		;
8420 
8421 	/*
8422 	 * Signal that we continue as usual as INIT signal
8423 	 * should be blocked while CPU is in VMX operation
8424 	 */
8425 	vmx_set_test_stage(3);
8426 
8427 	/* Wait for signal to enter VMX non-root mode */
8428 	while (vmx_get_test_stage() != 4)
8429 		;
8430 
8431 	/* Enter VMX non-root mode */
8432 	test_set_guest(v2_null_test_guest);
8433 	make_vmcs_current(test_vmcs);
8434 	enter_guest();
8435 	/* Save exit reason for BSP CPU to compare to expected result */
8436 	init_signal_test_exit_reason = vmcs_read(EXI_REASON);
8437 	/* VMCLEAR test-vmcs so it could be loaded by BSP CPU */
8438 	vmcs_clear(test_vmcs);
8439 	launched = false;
8440 	/* Signal that CPU exited to VMX root mode */
8441 	vmx_set_test_stage(5);
8442 
8443 	/* Wait for BSP CPU to signal to exit VMX operation */
8444 	while (vmx_get_test_stage() != 6)
8445 		;
8446 
8447 	/* Exit VMX operation (i.e. exec VMXOFF) */
8448 	vmx_off();
8449 
8450 	/*
8451 	 * Signal to BSP CPU that we continue as usual as INIT signal
8452 	 * should have been consumed by VMX_INIT exit from guest
8453 	 */
8454 	vmx_set_test_stage(7);
8455 
8456 	/* Wait for BSP CPU to signal to enter VMX operation */
8457 	while (vmx_get_test_stage() != 8)
8458 		;
8459 	/* Enter VMX operation (i.e. exec VMXON) */
8460 	_vmx_on(ap_vmxon_region);
8461 	/* Signal to BSP we are in VMX operation */
8462 	vmx_set_test_stage(9);
8463 
8464 	/* Wait for BSP CPU to send INIT signal */
8465 	while (vmx_get_test_stage() != 10)
8466 		;
8467 
8468 	/* Exit VMX operation (i.e. exec VMXOFF) */
8469 	vmx_off();
8470 
8471 	/*
8472 	 * Exiting VMX operation should result in latched
8473 	 * INIT signal being processed. Therefore, we should
8474 	 * never reach the below code. Thus, signal to BSP
8475 	 * CPU if we have reached here so it is able to
8476 	 * report an issue if it happens.
8477 	 */
8478 	init_signal_test_thread_continued = true;
8479 }
8480 
8481 #define INIT_SIGNAL_TEST_DELAY	100000000ULL
8482 
8483 static void vmx_init_signal_test(void)
8484 {
8485 	struct vmcs *test_vmcs;
8486 
8487 	if (cpu_count() < 2) {
8488 		report_skip(__func__);
8489 		return;
8490 	}
8491 
8492 	/* VMCLEAR test-vmcs so it could be loaded by other CPU */
8493 	vmcs_save(&test_vmcs);
8494 	vmcs_clear(test_vmcs);
8495 
8496 	vmx_set_test_stage(0);
8497 	on_cpu_async(1, init_signal_test_thread, test_vmcs);
8498 
8499 	/* Wait for other CPU to enter VMX operation */
8500 	while (vmx_get_test_stage() != 1)
8501 		;
8502 
8503 	/* Send INIT signal to other CPU */
8504 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
8505 				   id_map[1]);
8506 	/* Signal other CPU we have sent INIT signal */
8507 	vmx_set_test_stage(2);
8508 
8509 	/*
8510 	 * Wait reasonable amount of time for INIT signal to
8511 	 * be received on other CPU and verify that other CPU
8512 	 * have proceed as usual to next test stage as INIT
8513 	 * signal should be blocked while other CPU in
8514 	 * VMX operation
8515 	 */
8516 	delay(INIT_SIGNAL_TEST_DELAY);
8517 	report("INIT signal blocked when CPU in VMX operation",
8518 		   vmx_get_test_stage() == 3);
8519 	/* No point to continue if we failed at this point */
8520 	if (vmx_get_test_stage() != 3)
8521 		return;
8522 
8523 	/* Signal other CPU to enter VMX non-root mode */
8524 	init_signal_test_exit_reason = -1ull;
8525 	vmx_set_test_stage(4);
8526 	/*
8527 	 * Wait reasonable amont of time for other CPU
8528 	 * to exit to VMX root mode
8529 	 */
8530 	delay(INIT_SIGNAL_TEST_DELAY);
8531 	if (vmx_get_test_stage() != 5) {
8532 		report("Pending INIT signal didn't result in VMX exit", false);
8533 		return;
8534 	}
8535 	report("INIT signal during VMX non-root mode result in exit-reason %s (%lu)",
8536 			init_signal_test_exit_reason == VMX_INIT,
8537 			exit_reason_description(init_signal_test_exit_reason),
8538 			init_signal_test_exit_reason);
8539 
8540 	/* Run guest to completion */
8541 	make_vmcs_current(test_vmcs);
8542 	enter_guest();
8543 
8544 	/* Signal other CPU to exit VMX operation */
8545 	init_signal_test_thread_continued = false;
8546 	vmx_set_test_stage(6);
8547 
8548 	/* Wait reasonable amount of time for other CPU to exit VMX operation */
8549 	delay(INIT_SIGNAL_TEST_DELAY);
8550 	report("INIT signal consumed on VMX_INIT exit",
8551 		   vmx_get_test_stage() == 7);
8552 	/* No point to continue if we failed at this point */
8553 	if (vmx_get_test_stage() != 7)
8554 		return;
8555 
8556 	/* Signal other CPU to enter VMX operation */
8557 	vmx_set_test_stage(8);
8558 	/* Wait for other CPU to enter VMX operation */
8559 	while (vmx_get_test_stage() != 9)
8560 		;
8561 
8562 	/* Send INIT signal to other CPU */
8563 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
8564 				   id_map[1]);
8565 	/* Signal other CPU we have sent INIT signal */
8566 	vmx_set_test_stage(10);
8567 
8568 	/*
8569 	 * Wait reasonable amount of time for other CPU
8570 	 * to exit VMX operation and process INIT signal
8571 	 */
8572 	delay(INIT_SIGNAL_TEST_DELAY);
8573 	report("INIT signal processed after exit VMX operation",
8574 		   !init_signal_test_thread_continued);
8575 
8576 	/*
8577 	 * TODO: Send SIPI to other CPU to sipi_entry (See x86/cstart64.S)
8578 	 * to re-init it to kvm-unit-tests standard environment.
8579 	 * Somehow (?) verify that SIPI was indeed received.
8580 	 */
8581 }
8582 
8583 enum vmcs_access {
8584 	ACCESS_VMREAD,
8585 	ACCESS_VMWRITE,
8586 	ACCESS_NONE,
8587 };
8588 
8589 struct vmcs_shadow_test_common {
8590 	enum vmcs_access op;
8591 	enum Reason reason;
8592 	u64 field;
8593 	u64 value;
8594 	u64 flags;
8595 	u64 time;
8596 } l1_l2_common;
8597 
8598 static inline u64 vmread_flags(u64 field, u64 *val)
8599 {
8600 	u64 flags;
8601 
8602 	asm volatile ("vmread %2, %1; pushf; pop %0"
8603 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
8604 	return flags & X86_EFLAGS_ALU;
8605 }
8606 
8607 static inline u64 vmwrite_flags(u64 field, u64 val)
8608 {
8609 	u64 flags;
8610 
8611 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
8612 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
8613 	return flags & X86_EFLAGS_ALU;
8614 }
8615 
8616 static void vmx_vmcs_shadow_test_guest(void)
8617 {
8618 	struct vmcs_shadow_test_common *c = &l1_l2_common;
8619 	u64 start;
8620 
8621 	while (c->op != ACCESS_NONE) {
8622 		start = rdtsc();
8623 		switch (c->op) {
8624 		default:
8625 			c->flags = -1ull;
8626 			break;
8627 		case ACCESS_VMREAD:
8628 			c->flags = vmread_flags(c->field, &c->value);
8629 			break;
8630 		case ACCESS_VMWRITE:
8631 			c->flags = vmwrite_flags(c->field, 0);
8632 			break;
8633 		}
8634 		c->time = rdtsc() - start;
8635 		vmcall();
8636 	}
8637 }
8638 
8639 static u64 vmread_from_shadow(u64 field)
8640 {
8641 	struct vmcs *primary;
8642 	struct vmcs *shadow;
8643 	u64 value;
8644 
8645 	TEST_ASSERT(!vmcs_save(&primary));
8646 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
8647 	TEST_ASSERT(!make_vmcs_current(shadow));
8648 	value = vmcs_read(field);
8649 	TEST_ASSERT(!make_vmcs_current(primary));
8650 	return value;
8651 }
8652 
8653 static u64 vmwrite_to_shadow(u64 field, u64 value)
8654 {
8655 	struct vmcs *primary;
8656 	struct vmcs *shadow;
8657 
8658 	TEST_ASSERT(!vmcs_save(&primary));
8659 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
8660 	TEST_ASSERT(!make_vmcs_current(shadow));
8661 	vmcs_write(field, value);
8662 	value = vmcs_read(field);
8663 	TEST_ASSERT(!make_vmcs_current(primary));
8664 	return value;
8665 }
8666 
8667 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
8668 {
8669 	struct vmcs_shadow_test_common *c = &l1_l2_common;
8670 
8671 	c->op = access;
8672 	vmcs_write(VMX_INST_ERROR, 0);
8673 	enter_guest();
8674 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
8675 	if (c->reason != VMX_VMCALL) {
8676 		skip_exit_insn();
8677 		enter_guest();
8678 	}
8679 	skip_exit_vmcall();
8680 }
8681 
8682 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
8683 {
8684 	struct vmcs_shadow_test_common *c = &l1_l2_common;
8685 	struct vmcs *shadow;
8686 	u64 value;
8687 	uintptr_t flags[2];
8688 	bool good_shadow;
8689 	u32 vmx_inst_error;
8690 
8691 	report_prefix_pushf("field %lx", field);
8692 	c->field = field;
8693 
8694 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
8695 	if (shadow != (struct vmcs *)-1ull) {
8696 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
8697 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
8698 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
8699 	} else {
8700 		/*
8701 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
8702 		 * shadowed-fields should fail with setting RFLAGS.CF.
8703 		 */
8704 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
8705 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
8706 		good_shadow = false;
8707 	}
8708 
8709 	/* Intercept both VMREAD and VMWRITE. */
8710 	report_prefix_push("no VMREAD/VMWRITE permission");
8711 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
8712 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
8713 		set_bit(field, bitmap[ACCESS_VMREAD]);
8714 		set_bit(field, bitmap[ACCESS_VMWRITE]);
8715 	}
8716 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
8717 	report("not shadowed for VMWRITE", c->reason == VMX_VMWRITE);
8718 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
8719 	report("not shadowed for VMREAD", c->reason == VMX_VMREAD);
8720 	report_prefix_pop();
8721 
8722 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
8723 		goto out;
8724 
8725 	/* Permit shadowed VMREAD. */
8726 	report_prefix_push("VMREAD permission only");
8727 	clear_bit(field, bitmap[ACCESS_VMREAD]);
8728 	set_bit(field, bitmap[ACCESS_VMWRITE]);
8729 	if (good_shadow)
8730 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
8731 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
8732 	report("not shadowed for VMWRITE", c->reason == VMX_VMWRITE);
8733 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
8734 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
8735 	report("shadowed for VMREAD (in %ld cycles)", c->reason == VMX_VMCALL,
8736 	       c->time);
8737 	report("ALU flags after VMREAD (%lx) are as expected (%lx)",
8738 	       c->flags == flags[ACCESS_VMREAD],
8739 	       c->flags, flags[ACCESS_VMREAD]);
8740 	if (good_shadow)
8741 		report("value read from shadow (%lx) is as expected (%lx)",
8742 		       c->value == value, c->value, value);
8743 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
8744 		report("VMX_INST_ERROR (%d) is as expected (%d)",
8745 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
8746 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8747 	report_prefix_pop();
8748 
8749 	/* Permit shadowed VMWRITE. */
8750 	report_prefix_push("VMWRITE permission only");
8751 	set_bit(field, bitmap[ACCESS_VMREAD]);
8752 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
8753 	if (good_shadow)
8754 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
8755 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
8756 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
8757 	report("shadowed for VMWRITE (in %ld cycles)", c->reason == VMX_VMCALL,
8758 		c->time);
8759 	report("ALU flags after VMWRITE (%lx) are as expected (%lx)",
8760 	       c->flags == flags[ACCESS_VMREAD],
8761 	       c->flags, flags[ACCESS_VMREAD]);
8762 	if (good_shadow) {
8763 		value = vmread_from_shadow(field);
8764 		report("shadow VMCS value (%lx) is as expected (%lx)",
8765 		       value == 0, value, 0ul);
8766 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
8767 		report("VMX_INST_ERROR (%d) is as expected (%d)",
8768 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
8769 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8770 	}
8771 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
8772 	report("not shadowed for VMREAD", c->reason == VMX_VMREAD);
8773 	report_prefix_pop();
8774 
8775 	/* Permit shadowed VMREAD and VMWRITE. */
8776 	report_prefix_push("VMREAD and VMWRITE permission");
8777 	clear_bit(field, bitmap[ACCESS_VMREAD]);
8778 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
8779 	if (good_shadow)
8780 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
8781 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
8782 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
8783 	report("shadowed for VMWRITE (in %ld cycles)", c->reason == VMX_VMCALL,
8784 		c->time);
8785 	report("ALU flags after VMWRITE (%lx) are as expected (%lx)",
8786 	       c->flags == flags[ACCESS_VMREAD],
8787 	       c->flags, flags[ACCESS_VMREAD]);
8788 	if (good_shadow) {
8789 		value = vmread_from_shadow(field);
8790 		report("shadow VMCS value (%lx) is as expected (%lx)",
8791 		       value == 0, value, 0ul);
8792 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
8793 		report("VMX_INST_ERROR (%d) is as expected (%d)",
8794 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
8795 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8796 	}
8797 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
8798 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
8799 	report("shadowed for VMREAD (in %ld cycles)", c->reason == VMX_VMCALL,
8800 	       c->time);
8801 	report("ALU flags after VMREAD (%lx) are as expected (%lx)",
8802 	       c->flags == flags[ACCESS_VMREAD],
8803 	       c->flags, flags[ACCESS_VMREAD]);
8804 	if (good_shadow)
8805 		report("value read from shadow (%lx) is as expected (%lx)",
8806 		       c->value == 0, c->value, 0ul);
8807 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
8808 		report("VMX_INST_ERROR (%d) is as expected (%d)",
8809 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
8810 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8811 	report_prefix_pop();
8812 
8813 out:
8814 	report_prefix_pop();
8815 }
8816 
8817 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
8818 {
8819 	unsigned base;
8820 	unsigned index;
8821 	unsigned bit;
8822 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
8823 
8824 	/* Run test on all possible valid VMCS fields */
8825 	for (base = 0;
8826 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
8827 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
8828 		for (index = 0; index <= highest_index; index++)
8829 			vmcs_shadow_test_field(bitmap, base + index);
8830 
8831 	/*
8832 	 * Run tests on some invalid VMCS fields
8833 	 * (Have reserved bit set).
8834 	 */
8835 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
8836 		vmcs_shadow_test_field(bitmap, (1ull << bit));
8837 }
8838 
8839 static void vmx_vmcs_shadow_test(void)
8840 {
8841 	u8 *bitmap[2];
8842 	struct vmcs *shadow;
8843 
8844 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
8845 		printf("\t'Activate secondary controls' not supported.\n");
8846 		return;
8847 	}
8848 
8849 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
8850 		printf("\t'VMCS shadowing' not supported.\n");
8851 		return;
8852 	}
8853 
8854 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
8855 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
8856 		printf("\tVMWRITE can't modify VM-exit information fields.\n");
8857 		return;
8858 	}
8859 
8860 	test_set_guest(vmx_vmcs_shadow_test_guest);
8861 
8862 	bitmap[ACCESS_VMREAD] = alloc_page();
8863 	bitmap[ACCESS_VMWRITE] = alloc_page();
8864 
8865 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
8866 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
8867 
8868 	shadow = alloc_page();
8869 	shadow->hdr.revision_id = basic.revision;
8870 	shadow->hdr.shadow_vmcs = 1;
8871 	TEST_ASSERT(!vmcs_clear(shadow));
8872 
8873 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
8874 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
8875 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
8876 
8877 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
8878 	report_prefix_push("valid link pointer");
8879 	vmx_vmcs_shadow_test_body(bitmap);
8880 	report_prefix_pop();
8881 
8882 	vmcs_write(VMCS_LINK_PTR, -1ull);
8883 	report_prefix_push("invalid link pointer");
8884 	vmx_vmcs_shadow_test_body(bitmap);
8885 	report_prefix_pop();
8886 
8887 	l1_l2_common.op = ACCESS_NONE;
8888 	enter_guest();
8889 }
8890 
8891 
8892 
8893 static int invalid_msr_init(struct vmcs *vmcs)
8894 {
8895 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
8896 		printf("\tPreemption timer is not supported\n");
8897 		return VMX_TEST_EXIT;
8898 	}
8899 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
8900 	preempt_val = 10000000;
8901 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
8902 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
8903 
8904 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
8905 		printf("\tSave preemption value is not supported\n");
8906 
8907 	vmcs_write(ENT_MSR_LD_CNT, 1);
8908 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
8909 
8910 	return VMX_TEST_START;
8911 }
8912 
8913 
8914 static void invalid_msr_main(void)
8915 {
8916 	report("Invalid MSR load", 0);
8917 }
8918 
8919 static int invalid_msr_exit_handler(void)
8920 {
8921 	report("Invalid MSR load", 0);
8922 	print_vmexit_info();
8923 	return VMX_TEST_EXIT;
8924 }
8925 
8926 static int invalid_msr_entry_failure(struct vmentry_failure *failure)
8927 {
8928 	ulong reason;
8929 
8930 	reason = vmcs_read(EXI_REASON);
8931 	report("Invalid MSR load", reason == (0x80000000u | VMX_FAIL_MSR));
8932 	return VMX_TEST_VMEXIT;
8933 }
8934 
8935 /*
8936  * The max number of MSRs in an atomic switch MSR list is:
8937  * (111B + 1) * 512 = 4096
8938  *
8939  * Each list entry consumes:
8940  * 4-byte MSR index + 4 bytes reserved + 8-byte data = 16 bytes
8941  *
8942  * Allocate 128 kB to cover max_msr_list_size (i.e., 64 kB) and then some.
8943  */
8944 static const u32 msr_list_page_order = 5;
8945 
8946 static void atomic_switch_msr_limit_test_guest(void)
8947 {
8948 	vmcall();
8949 }
8950 
8951 static void populate_msr_list(struct vmx_msr_entry *msr_list,
8952 			      size_t byte_capacity, int count)
8953 {
8954 	int i;
8955 
8956 	for (i = 0; i < count; i++) {
8957 		msr_list[i].index = MSR_IA32_TSC;
8958 		msr_list[i].reserved = 0;
8959 		msr_list[i].value = 0x1234567890abcdef;
8960 	}
8961 
8962 	memset(msr_list + count, 0xff,
8963 	       byte_capacity - count * sizeof(*msr_list));
8964 }
8965 
8966 static int max_msr_list_size(void)
8967 {
8968 	u32 vmx_misc = rdmsr(MSR_IA32_VMX_MISC);
8969 	u32 factor = ((vmx_misc & GENMASK(27, 25)) >> 25) + 1;
8970 
8971 	return factor * 512;
8972 }
8973 
8974 static void atomic_switch_msrs_test(int count)
8975 {
8976 	struct vmx_msr_entry *vm_enter_load;
8977         struct vmx_msr_entry *vm_exit_load;
8978         struct vmx_msr_entry *vm_exit_store;
8979 	int max_allowed = max_msr_list_size();
8980 	int byte_capacity = 1ul << (msr_list_page_order + PAGE_SHIFT);
8981 	/* Exceeding the max MSR list size at exit trigers KVM to abort. */
8982 	int exit_count = count > max_allowed ? max_allowed : count;
8983 	int cleanup_count = count > max_allowed ? 2 : 1;
8984 	int i;
8985 
8986 	/*
8987 	 * Check for the IA32_TSC MSR,
8988 	 * available with the "TSC flag" and used to populate the MSR lists.
8989 	 */
8990 	if (!(cpuid(1).d & (1 << 4))) {
8991 		report_skip(__func__);
8992 		return;
8993 	}
8994 
8995 	/* Set L2 guest. */
8996 	test_set_guest(atomic_switch_msr_limit_test_guest);
8997 
8998 	/* Setup atomic MSR switch lists. */
8999 	vm_enter_load = alloc_pages(msr_list_page_order);
9000 	vm_exit_load = alloc_pages(msr_list_page_order);
9001 	vm_exit_store = alloc_pages(msr_list_page_order);
9002 
9003 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)vm_enter_load);
9004 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)vm_exit_load);
9005 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)vm_exit_store);
9006 
9007 	/*
9008 	 * VM-Enter should succeed up to the max number of MSRs per list, and
9009 	 * should not consume junk beyond the last entry.
9010 	 */
9011 	populate_msr_list(vm_enter_load, byte_capacity, count);
9012 	populate_msr_list(vm_exit_load, byte_capacity, exit_count);
9013 	populate_msr_list(vm_exit_store, byte_capacity, exit_count);
9014 
9015 	vmcs_write(ENT_MSR_LD_CNT, count);
9016 	vmcs_write(EXI_MSR_LD_CNT, exit_count);
9017 	vmcs_write(EXI_MSR_ST_CNT, exit_count);
9018 
9019 	if (count <= max_allowed) {
9020 		enter_guest();
9021 		assert_exit_reason(VMX_VMCALL);
9022 		skip_exit_vmcall();
9023 	} else {
9024 		u32 exit_reason;
9025 		u32 exit_reason_want;
9026 		u32 exit_qual;
9027 
9028 		enter_guest_with_invalid_guest_state();
9029 
9030 		exit_reason = vmcs_read(EXI_REASON);
9031 		exit_reason_want = VMX_FAIL_MSR | VMX_ENTRY_FAILURE;
9032 		report("exit_reason, %u, is %u.",
9033 		       exit_reason == exit_reason_want, exit_reason,
9034 		       exit_reason_want);
9035 
9036 		exit_qual = vmcs_read(EXI_QUALIFICATION);
9037 		report("exit_qual, %u, is %u.", exit_qual == max_allowed + 1,
9038 		       exit_qual, max_allowed + 1);
9039 	}
9040 
9041 	/* Cleanup. */
9042 	vmcs_write(ENT_MSR_LD_CNT, 0);
9043 	vmcs_write(EXI_MSR_LD_CNT, 0);
9044 	vmcs_write(EXI_MSR_ST_CNT, 0);
9045 	for (i = 0; i < cleanup_count; i++) {
9046 		enter_guest();
9047 		skip_exit_vmcall();
9048 	}
9049 	free_pages_by_order(vm_enter_load, msr_list_page_order);
9050 	free_pages_by_order(vm_exit_load, msr_list_page_order);
9051 	free_pages_by_order(vm_exit_store, msr_list_page_order);
9052 }
9053 
9054 static void atomic_switch_max_msrs_test(void)
9055 {
9056 	atomic_switch_msrs_test(max_msr_list_size());
9057 }
9058 
9059 static void atomic_switch_overflow_msrs_test(void)
9060 {
9061 	atomic_switch_msrs_test(max_msr_list_size() + 1);
9062 }
9063 
9064 #define TEST(name) { #name, .v2 = name }
9065 
9066 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
9067 struct vmx_test vmx_tests[] = {
9068 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
9069 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
9070 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
9071 		preemption_timer_exit_handler, NULL, {0} },
9072 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
9073 		test_ctrl_pat_exit_handler, NULL, {0} },
9074 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
9075 		test_ctrl_efer_exit_handler, NULL, {0} },
9076 	{ "CR shadowing", NULL, cr_shadowing_main,
9077 		cr_shadowing_exit_handler, NULL, {0} },
9078 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
9079 		NULL, {0} },
9080 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
9081 		insn_intercept_exit_handler, NULL, {0} },
9082 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
9083 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
9084 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
9085 	{ "VPID", vpid_init, vpid_main, vpid_exit_handler, NULL, {0} },
9086 	{ "interrupt", interrupt_init, interrupt_main,
9087 		interrupt_exit_handler, NULL, {0} },
9088 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
9089 		NULL, {0} },
9090 	{ "MSR switch", msr_switch_init, msr_switch_main,
9091 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
9092 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
9093 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
9094 		disable_rdtscp_exit_handler, NULL, {0} },
9095 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
9096 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
9097 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
9098 		exit_monitor_from_l2_handler, NULL, {0} },
9099 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
9100 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
9101 	/* Basic V2 tests. */
9102 	TEST(v2_null_test),
9103 	TEST(v2_multiple_entries_test),
9104 	TEST(fixture_test_case1),
9105 	TEST(fixture_test_case2),
9106 	/* Opcode tests. */
9107 	TEST(invvpid_test_v2),
9108 	/* VM-entry tests */
9109 	TEST(vmx_controls_test),
9110 	TEST(vmx_host_state_area_test),
9111 	TEST(vmx_guest_state_area_test),
9112 	TEST(vmentry_movss_shadow_test),
9113 	/* APICv tests */
9114 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
9115 	TEST(vmx_hlt_with_rvi_test),
9116 	TEST(apic_reg_virt_test),
9117 	TEST(virt_x2apic_mode_test),
9118 	/* APIC pass-through tests */
9119 	TEST(vmx_apic_passthrough_test),
9120 	TEST(vmx_apic_passthrough_thread_test),
9121 	TEST(vmx_apic_passthrough_tpr_threshold_test),
9122 	TEST(vmx_init_signal_test),
9123 	/* VMCS Shadowing tests */
9124 	TEST(vmx_vmcs_shadow_test),
9125 	/* Regression tests */
9126 	TEST(vmx_cr_load_test),
9127 	TEST(vmx_nm_test),
9128 	TEST(vmx_db_test),
9129 	TEST(vmx_nmi_window_test),
9130 	TEST(vmx_intr_window_test),
9131 	TEST(vmx_pending_event_test),
9132 	TEST(vmx_pending_event_hlt_test),
9133 	TEST(vmx_store_tsc_test),
9134 	/* EPT access tests. */
9135 	TEST(ept_access_test_not_present),
9136 	TEST(ept_access_test_read_only),
9137 	TEST(ept_access_test_write_only),
9138 	TEST(ept_access_test_read_write),
9139 	TEST(ept_access_test_execute_only),
9140 	TEST(ept_access_test_read_execute),
9141 	TEST(ept_access_test_write_execute),
9142 	TEST(ept_access_test_read_write_execute),
9143 	TEST(ept_access_test_reserved_bits),
9144 	TEST(ept_access_test_ignored_bits),
9145 	TEST(ept_access_test_paddr_not_present_ad_disabled),
9146 	TEST(ept_access_test_paddr_not_present_ad_enabled),
9147 	TEST(ept_access_test_paddr_read_only_ad_disabled),
9148 	TEST(ept_access_test_paddr_read_only_ad_enabled),
9149 	TEST(ept_access_test_paddr_read_write),
9150 	TEST(ept_access_test_paddr_read_write_execute),
9151 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
9152 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
9153 	TEST(ept_access_test_paddr_not_present_page_fault),
9154 	TEST(ept_access_test_force_2m_page),
9155 	/* Atomic MSR switch tests. */
9156 	TEST(atomic_switch_max_msrs_test),
9157 	TEST(atomic_switch_overflow_msrs_test),
9158 	{ NULL, NULL, NULL, NULL, NULL, {0} },
9159 };
9160