xref: /kvm-unit-tests/x86/vmx_tests.c (revision 4c8a99ca02252d4a2bee43f4558fe47ce5ab7ec0)
1 /*
2  * All test cases of nested virtualization should be in this file
3  *
4  * Author : Arthur Chunqi Li <yzt356@gmail.com>
5  */
6 
7 #include <asm/debugreg.h>
8 
9 #include "vmx.h"
10 #include "msr.h"
11 #include "processor.h"
12 #include "vm.h"
13 #include "pci.h"
14 #include "fwcfg.h"
15 #include "isr.h"
16 #include "desc.h"
17 #include "apic.h"
18 #include "types.h"
19 #include "vmalloc.h"
20 #include "alloc_page.h"
21 #include "smp.h"
22 #include "delay.h"
23 #include "access.h"
24 #include "x86/usermode.h"
25 
26 /*
27  * vmcs.GUEST_PENDING_DEBUG has the same format as DR6, although some bits that
28  * are legal in DR6 are reserved in vmcs.GUEST_PENDING_DEBUG.  And if any data
29  * or I/O breakpoint matches *and* was enabled, bit 12 is also set.
30  */
31 #define PENDING_DBG_TRAP	BIT(12)
32 
33 #define VPID_CAP_INVVPID_TYPES_SHIFT 40
34 
35 u64 ia32_pat;
36 u64 ia32_efer;
37 void *io_bitmap_a, *io_bitmap_b;
38 u16 ioport;
39 
40 unsigned long *pml4;
41 u64 eptp;
42 void *data_page1, *data_page2;
43 
44 phys_addr_t pci_physaddr;
45 
46 void *pml_log;
47 #define PML_INDEX 512
48 
49 static inline unsigned ffs(unsigned x)
50 {
51 	int pos = -1;
52 
53 	__asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0"
54 			     : "+r"(pos) : "rm"(x) : "eax");
55 	return pos + 1;
56 }
57 
58 static inline void vmcall(void)
59 {
60 	asm volatile("vmcall");
61 }
62 
63 static void basic_guest_main(void)
64 {
65 	report_pass("Basic VMX test");
66 }
67 
68 static int basic_exit_handler(union exit_reason exit_reason)
69 {
70 	report_fail("Basic VMX test");
71 	print_vmexit_info(exit_reason);
72 	return VMX_TEST_EXIT;
73 }
74 
75 static void vmenter_main(void)
76 {
77 	u64 rax;
78 	u64 rsp, resume_rsp;
79 
80 	report_pass("test vmlaunch");
81 
82 	asm volatile(
83 		"mov %%rsp, %0\n\t"
84 		"mov %3, %%rax\n\t"
85 		"vmcall\n\t"
86 		"mov %%rax, %1\n\t"
87 		"mov %%rsp, %2\n\t"
88 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
89 		: "g"(0xABCD));
90 	report((rax == 0xFFFF) && (rsp == resume_rsp), "test vmresume");
91 }
92 
93 static int vmenter_exit_handler(union exit_reason exit_reason)
94 {
95 	u64 guest_rip = vmcs_read(GUEST_RIP);
96 
97 	switch (exit_reason.basic) {
98 	case VMX_VMCALL:
99 		if (regs.rax != 0xABCD) {
100 			report_fail("test vmresume");
101 			return VMX_TEST_VMEXIT;
102 		}
103 		regs.rax = 0xFFFF;
104 		vmcs_write(GUEST_RIP, guest_rip + 3);
105 		return VMX_TEST_RESUME;
106 	default:
107 		report_fail("test vmresume");
108 		print_vmexit_info(exit_reason);
109 	}
110 	return VMX_TEST_VMEXIT;
111 }
112 
113 u32 preempt_scale;
114 volatile unsigned long long tsc_val;
115 volatile u32 preempt_val;
116 u64 saved_rip;
117 
118 static int preemption_timer_init(struct vmcs *vmcs)
119 {
120 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
121 		printf("\tPreemption timer is not supported\n");
122 		return VMX_TEST_EXIT;
123 	}
124 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
125 	preempt_val = 10000000;
126 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
127 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
128 
129 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
130 		printf("\tSave preemption value is not supported\n");
131 
132 	return VMX_TEST_START;
133 }
134 
135 static void preemption_timer_main(void)
136 {
137 	tsc_val = rdtsc();
138 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
139 		vmx_set_test_stage(0);
140 		vmcall();
141 		if (vmx_get_test_stage() == 1)
142 			vmcall();
143 	}
144 	vmx_set_test_stage(1);
145 	while (vmx_get_test_stage() == 1) {
146 		if (((rdtsc() - tsc_val) >> preempt_scale)
147 				> 10 * preempt_val) {
148 			vmx_set_test_stage(2);
149 			vmcall();
150 		}
151 	}
152 	tsc_val = rdtsc();
153 	asm volatile ("hlt");
154 	vmcall();
155 	vmx_set_test_stage(5);
156 	vmcall();
157 }
158 
159 static int preemption_timer_exit_handler(union exit_reason exit_reason)
160 {
161 	bool guest_halted;
162 	u64 guest_rip;
163 	u32 insn_len;
164 	u32 ctrl_exit;
165 
166 	guest_rip = vmcs_read(GUEST_RIP);
167 	insn_len = vmcs_read(EXI_INST_LEN);
168 	switch (exit_reason.basic) {
169 	case VMX_PREEMPT:
170 		switch (vmx_get_test_stage()) {
171 		case 1:
172 		case 2:
173 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val,
174 			       "busy-wait for preemption timer");
175 			vmx_set_test_stage(3);
176 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
177 			return VMX_TEST_RESUME;
178 		case 3:
179 			guest_halted =
180 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
181 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val
182 			        && guest_halted,
183 			       "preemption timer during hlt");
184 			vmx_set_test_stage(4);
185 			vmcs_write(PIN_CONTROLS,
186 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
187 			vmcs_write(EXI_CONTROLS,
188 				   vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT);
189 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
190 			return VMX_TEST_RESUME;
191 		case 4:
192 			report(saved_rip == guest_rip,
193 			       "preemption timer with 0 value");
194 			break;
195 		default:
196 			report_fail("Invalid stage.");
197 			print_vmexit_info(exit_reason);
198 			break;
199 		}
200 		break;
201 	case VMX_VMCALL:
202 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
203 		switch (vmx_get_test_stage()) {
204 		case 0:
205 			report(vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val,
206 			       "Keep preemption value");
207 			vmx_set_test_stage(1);
208 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
209 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
210 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
211 			vmcs_write(EXI_CONTROLS, ctrl_exit);
212 			return VMX_TEST_RESUME;
213 		case 1:
214 			report(vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val,
215 			       "Save preemption value");
216 			return VMX_TEST_RESUME;
217 		case 2:
218 			report_fail("busy-wait for preemption timer");
219 			vmx_set_test_stage(3);
220 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
221 			return VMX_TEST_RESUME;
222 		case 3:
223 			report_fail("preemption timer during hlt");
224 			vmx_set_test_stage(4);
225 			/* fall through */
226 		case 4:
227 			vmcs_write(PIN_CONTROLS,
228 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
229 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
230 			saved_rip = guest_rip + insn_len;
231 			return VMX_TEST_RESUME;
232 		case 5:
233 			report_fail("preemption timer with 0 value (vmcall stage 5)");
234 			break;
235 		default:
236 			// Should not reach here
237 			report_fail("unexpected stage, %d",
238 				    vmx_get_test_stage());
239 			print_vmexit_info(exit_reason);
240 			return VMX_TEST_VMEXIT;
241 		}
242 		break;
243 	default:
244 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
245 		print_vmexit_info(exit_reason);
246 	}
247 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
248 	return VMX_TEST_VMEXIT;
249 }
250 
251 static void msr_bmp_init(void)
252 {
253 	void *msr_bitmap;
254 	u32 ctrl_cpu0;
255 
256 	msr_bitmap = alloc_page();
257 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
258 	ctrl_cpu0 |= CPU_MSR_BITMAP;
259 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
260 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
261 }
262 
263 static void *get_msr_bitmap(void)
264 {
265 	void *msr_bitmap;
266 
267 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
268 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
269 	} else {
270 		msr_bitmap = alloc_page();
271 		memset(msr_bitmap, 0xff, PAGE_SIZE);
272 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
273 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
274 	}
275 
276 	return msr_bitmap;
277 }
278 
279 static void disable_intercept_for_x2apic_msrs(void)
280 {
281 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
282 	u32 msr;
283 
284 	for (msr = APIC_BASE_MSR;
285 		 msr < (APIC_BASE_MSR+0xff);
286 		 msr += BITS_PER_LONG) {
287 		unsigned int word = msr / BITS_PER_LONG;
288 
289 		msr_bitmap[word] = 0;
290 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
291 	}
292 }
293 
294 static int test_ctrl_pat_init(struct vmcs *vmcs)
295 {
296 	u64 ctrl_ent;
297 	u64 ctrl_exi;
298 
299 	msr_bmp_init();
300 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
301 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
302 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
303 		printf("\tSave/load PAT is not supported\n");
304 		return 1;
305 	}
306 
307 	ctrl_ent = vmcs_read(ENT_CONTROLS);
308 	ctrl_exi = vmcs_read(EXI_CONTROLS);
309 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
310 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
311 	vmcs_write(ENT_CONTROLS, ctrl_ent);
312 	vmcs_write(EXI_CONTROLS, ctrl_exi);
313 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
314 	vmcs_write(GUEST_PAT, 0x0);
315 	vmcs_write(HOST_PAT, ia32_pat);
316 	return VMX_TEST_START;
317 }
318 
319 static void test_ctrl_pat_main(void)
320 {
321 	u64 guest_ia32_pat;
322 
323 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
324 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
325 		printf("\tENT_LOAD_PAT is not supported.\n");
326 	else {
327 		if (guest_ia32_pat != 0) {
328 			report_fail("Entry load PAT");
329 			return;
330 		}
331 	}
332 	wrmsr(MSR_IA32_CR_PAT, 0x6);
333 	vmcall();
334 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
335 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
336 		report(guest_ia32_pat == ia32_pat, "Entry load PAT");
337 }
338 
339 static int test_ctrl_pat_exit_handler(union exit_reason exit_reason)
340 {
341 	u64 guest_rip;
342 	u64 guest_pat;
343 
344 	guest_rip = vmcs_read(GUEST_RIP);
345 	switch (exit_reason.basic) {
346 	case VMX_VMCALL:
347 		guest_pat = vmcs_read(GUEST_PAT);
348 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
349 			printf("\tEXI_SAVE_PAT is not supported\n");
350 			vmcs_write(GUEST_PAT, 0x6);
351 		} else {
352 			report(guest_pat == 0x6, "Exit save PAT");
353 		}
354 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
355 			printf("\tEXI_LOAD_PAT is not supported\n");
356 		else
357 			report(rdmsr(MSR_IA32_CR_PAT) == ia32_pat,
358 			       "Exit load PAT");
359 		vmcs_write(GUEST_PAT, ia32_pat);
360 		vmcs_write(GUEST_RIP, guest_rip + 3);
361 		return VMX_TEST_RESUME;
362 	default:
363 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
364 		break;
365 	}
366 	return VMX_TEST_VMEXIT;
367 }
368 
369 static int test_ctrl_efer_init(struct vmcs *vmcs)
370 {
371 	u64 ctrl_ent;
372 	u64 ctrl_exi;
373 
374 	msr_bmp_init();
375 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
376 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
377 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
378 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
379 	ia32_efer = rdmsr(MSR_EFER);
380 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
381 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
382 	return VMX_TEST_START;
383 }
384 
385 static void test_ctrl_efer_main(void)
386 {
387 	u64 guest_ia32_efer;
388 
389 	guest_ia32_efer = rdmsr(MSR_EFER);
390 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
391 		printf("\tENT_LOAD_EFER is not supported.\n");
392 	else {
393 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
394 			report_fail("Entry load EFER");
395 			return;
396 		}
397 	}
398 	wrmsr(MSR_EFER, ia32_efer);
399 	vmcall();
400 	guest_ia32_efer = rdmsr(MSR_EFER);
401 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
402 		report(guest_ia32_efer == ia32_efer, "Entry load EFER");
403 }
404 
405 static int test_ctrl_efer_exit_handler(union exit_reason exit_reason)
406 {
407 	u64 guest_rip;
408 	u64 guest_efer;
409 
410 	guest_rip = vmcs_read(GUEST_RIP);
411 	switch (exit_reason.basic) {
412 	case VMX_VMCALL:
413 		guest_efer = vmcs_read(GUEST_EFER);
414 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
415 			printf("\tEXI_SAVE_EFER is not supported\n");
416 			vmcs_write(GUEST_EFER, ia32_efer);
417 		} else {
418 			report(guest_efer == ia32_efer, "Exit save EFER");
419 		}
420 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
421 			printf("\tEXI_LOAD_EFER is not supported\n");
422 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
423 		} else {
424 			report(rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX),
425 			       "Exit load EFER");
426 		}
427 		vmcs_write(GUEST_PAT, ia32_efer);
428 		vmcs_write(GUEST_RIP, guest_rip + 3);
429 		return VMX_TEST_RESUME;
430 	default:
431 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
432 		break;
433 	}
434 	return VMX_TEST_VMEXIT;
435 }
436 
437 u32 guest_cr0, guest_cr4;
438 
439 static void cr_shadowing_main(void)
440 {
441 	u32 cr0, cr4, tmp;
442 
443 	// Test read through
444 	vmx_set_test_stage(0);
445 	guest_cr0 = read_cr0();
446 	if (vmx_get_test_stage() == 1)
447 		report_fail("Read through CR0");
448 	else
449 		vmcall();
450 	vmx_set_test_stage(1);
451 	guest_cr4 = read_cr4();
452 	if (vmx_get_test_stage() == 2)
453 		report_fail("Read through CR4");
454 	else
455 		vmcall();
456 	// Test write through
457 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
458 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
459 	vmx_set_test_stage(2);
460 	write_cr0(guest_cr0);
461 	if (vmx_get_test_stage() == 3)
462 		report_fail("Write through CR0");
463 	else
464 		vmcall();
465 	vmx_set_test_stage(3);
466 	write_cr4(guest_cr4);
467 	if (vmx_get_test_stage() == 4)
468 		report_fail("Write through CR4");
469 	else
470 		vmcall();
471 	// Test read shadow
472 	vmx_set_test_stage(4);
473 	vmcall();
474 	cr0 = read_cr0();
475 	if (vmx_get_test_stage() != 5)
476 		report(cr0 == guest_cr0, "Read shadowing CR0");
477 	vmx_set_test_stage(5);
478 	cr4 = read_cr4();
479 	if (vmx_get_test_stage() != 6)
480 		report(cr4 == guest_cr4, "Read shadowing CR4");
481 	// Test write shadow (same value with shadow)
482 	vmx_set_test_stage(6);
483 	write_cr0(guest_cr0);
484 	if (vmx_get_test_stage() == 7)
485 		report_fail("Write shadowing CR0 (same value with shadow)");
486 	else
487 		vmcall();
488 	vmx_set_test_stage(7);
489 	write_cr4(guest_cr4);
490 	if (vmx_get_test_stage() == 8)
491 		report_fail("Write shadowing CR4 (same value with shadow)");
492 	else
493 		vmcall();
494 	// Test write shadow (different value)
495 	vmx_set_test_stage(8);
496 	tmp = guest_cr0 ^ X86_CR0_TS;
497 	asm volatile("mov %0, %%rsi\n\t"
498 		"mov %%rsi, %%cr0\n\t"
499 		::"m"(tmp)
500 		:"rsi", "memory", "cc");
501 	report(vmx_get_test_stage() == 9,
502 	       "Write shadowing different X86_CR0_TS");
503 	vmx_set_test_stage(9);
504 	tmp = guest_cr0 ^ X86_CR0_MP;
505 	asm volatile("mov %0, %%rsi\n\t"
506 		"mov %%rsi, %%cr0\n\t"
507 		::"m"(tmp)
508 		:"rsi", "memory", "cc");
509 	report(vmx_get_test_stage() == 10,
510 	       "Write shadowing different X86_CR0_MP");
511 	vmx_set_test_stage(10);
512 	tmp = guest_cr4 ^ X86_CR4_TSD;
513 	asm volatile("mov %0, %%rsi\n\t"
514 		"mov %%rsi, %%cr4\n\t"
515 		::"m"(tmp)
516 		:"rsi", "memory", "cc");
517 	report(vmx_get_test_stage() == 11,
518 	       "Write shadowing different X86_CR4_TSD");
519 	vmx_set_test_stage(11);
520 	tmp = guest_cr4 ^ X86_CR4_DE;
521 	asm volatile("mov %0, %%rsi\n\t"
522 		"mov %%rsi, %%cr4\n\t"
523 		::"m"(tmp)
524 		:"rsi", "memory", "cc");
525 	report(vmx_get_test_stage() == 12,
526 	       "Write shadowing different X86_CR4_DE");
527 }
528 
529 static int cr_shadowing_exit_handler(union exit_reason exit_reason)
530 {
531 	u64 guest_rip;
532 	u32 insn_len;
533 	u32 exit_qual;
534 
535 	guest_rip = vmcs_read(GUEST_RIP);
536 	insn_len = vmcs_read(EXI_INST_LEN);
537 	exit_qual = vmcs_read(EXI_QUALIFICATION);
538 	switch (exit_reason.basic) {
539 	case VMX_VMCALL:
540 		switch (vmx_get_test_stage()) {
541 		case 0:
542 			report(guest_cr0 == vmcs_read(GUEST_CR0),
543 			       "Read through CR0");
544 			break;
545 		case 1:
546 			report(guest_cr4 == vmcs_read(GUEST_CR4),
547 			       "Read through CR4");
548 			break;
549 		case 2:
550 			report(guest_cr0 == vmcs_read(GUEST_CR0),
551 			       "Write through CR0");
552 			break;
553 		case 3:
554 			report(guest_cr4 == vmcs_read(GUEST_CR4),
555 			       "Write through CR4");
556 			break;
557 		case 4:
558 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
559 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
560 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
561 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
562 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
563 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
564 			break;
565 		case 6:
566 			report(guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)),
567 			       "Write shadowing CR0 (same value)");
568 			break;
569 		case 7:
570 			report(guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)),
571 			       "Write shadowing CR4 (same value)");
572 			break;
573 		default:
574 			// Should not reach here
575 			report_fail("unexpected stage, %d",
576 				    vmx_get_test_stage());
577 			print_vmexit_info(exit_reason);
578 			return VMX_TEST_VMEXIT;
579 		}
580 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
581 		return VMX_TEST_RESUME;
582 	case VMX_CR:
583 		switch (vmx_get_test_stage()) {
584 		case 4:
585 			report_fail("Read shadowing CR0");
586 			vmx_inc_test_stage();
587 			break;
588 		case 5:
589 			report_fail("Read shadowing CR4");
590 			vmx_inc_test_stage();
591 			break;
592 		case 6:
593 			report_fail("Write shadowing CR0 (same value)");
594 			vmx_inc_test_stage();
595 			break;
596 		case 7:
597 			report_fail("Write shadowing CR4 (same value)");
598 			vmx_inc_test_stage();
599 			break;
600 		case 8:
601 		case 9:
602 			// 0x600 encodes "mov %esi, %cr0"
603 			if (exit_qual == 0x600)
604 				vmx_inc_test_stage();
605 			break;
606 		case 10:
607 		case 11:
608 			// 0x604 encodes "mov %esi, %cr4"
609 			if (exit_qual == 0x604)
610 				vmx_inc_test_stage();
611 			break;
612 		default:
613 			// Should not reach here
614 			report_fail("unexpected stage, %d",
615 				    vmx_get_test_stage());
616 			print_vmexit_info(exit_reason);
617 			return VMX_TEST_VMEXIT;
618 		}
619 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
620 		return VMX_TEST_RESUME;
621 	default:
622 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
623 		print_vmexit_info(exit_reason);
624 	}
625 	return VMX_TEST_VMEXIT;
626 }
627 
628 static int iobmp_init(struct vmcs *vmcs)
629 {
630 	u32 ctrl_cpu0;
631 
632 	io_bitmap_a = alloc_page();
633 	io_bitmap_b = alloc_page();
634 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
635 	ctrl_cpu0 |= CPU_IO_BITMAP;
636 	ctrl_cpu0 &= (~CPU_IO);
637 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
638 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
639 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
640 	return VMX_TEST_START;
641 }
642 
643 static void iobmp_main(void)
644 {
645 	// stage 0, test IO pass
646 	vmx_set_test_stage(0);
647 	inb(0x5000);
648 	outb(0x0, 0x5000);
649 	report(vmx_get_test_stage() == 0, "I/O bitmap - I/O pass");
650 	// test IO width, in/out
651 	((u8 *)io_bitmap_a)[0] = 0xFF;
652 	vmx_set_test_stage(2);
653 	inb(0x0);
654 	report(vmx_get_test_stage() == 3, "I/O bitmap - trap in");
655 	vmx_set_test_stage(3);
656 	outw(0x0, 0x0);
657 	report(vmx_get_test_stage() == 4, "I/O bitmap - trap out");
658 	vmx_set_test_stage(4);
659 	inl(0x0);
660 	report(vmx_get_test_stage() == 5, "I/O bitmap - I/O width, long");
661 	// test low/high IO port
662 	vmx_set_test_stage(5);
663 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
664 	inb(0x5000);
665 	report(vmx_get_test_stage() == 6, "I/O bitmap - I/O port, low part");
666 	vmx_set_test_stage(6);
667 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
668 	inb(0x9000);
669 	report(vmx_get_test_stage() == 7, "I/O bitmap - I/O port, high part");
670 	// test partial pass
671 	vmx_set_test_stage(7);
672 	inl(0x4FFF);
673 	report(vmx_get_test_stage() == 8, "I/O bitmap - partial pass");
674 	// test overrun
675 	vmx_set_test_stage(8);
676 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
677 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
678 	inl(0xFFFF);
679 	report(vmx_get_test_stage() == 9, "I/O bitmap - overrun");
680 	vmx_set_test_stage(9);
681 	vmcall();
682 	outb(0x0, 0x0);
683 	report(vmx_get_test_stage() == 9,
684 	       "I/O bitmap - ignore unconditional exiting");
685 	vmx_set_test_stage(10);
686 	vmcall();
687 	outb(0x0, 0x0);
688 	report(vmx_get_test_stage() == 11,
689 	       "I/O bitmap - unconditional exiting");
690 }
691 
692 static int iobmp_exit_handler(union exit_reason exit_reason)
693 {
694 	u64 guest_rip;
695 	ulong exit_qual;
696 	u32 insn_len, ctrl_cpu0;
697 
698 	guest_rip = vmcs_read(GUEST_RIP);
699 	exit_qual = vmcs_read(EXI_QUALIFICATION);
700 	insn_len = vmcs_read(EXI_INST_LEN);
701 	switch (exit_reason.basic) {
702 	case VMX_IO:
703 		switch (vmx_get_test_stage()) {
704 		case 0:
705 		case 1:
706 			vmx_inc_test_stage();
707 			break;
708 		case 2:
709 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE,
710 			       "I/O bitmap - I/O width, byte");
711 			report(exit_qual & VMX_IO_IN,
712 			       "I/O bitmap - I/O direction, in");
713 			vmx_inc_test_stage();
714 			break;
715 		case 3:
716 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD,
717 			       "I/O bitmap - I/O width, word");
718 			report(!(exit_qual & VMX_IO_IN),
719 			       "I/O bitmap - I/O direction, out");
720 			vmx_inc_test_stage();
721 			break;
722 		case 4:
723 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG,
724 			       "I/O bitmap - I/O width, long");
725 			vmx_inc_test_stage();
726 			break;
727 		case 5:
728 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
729 				vmx_inc_test_stage();
730 			break;
731 		case 6:
732 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
733 				vmx_inc_test_stage();
734 			break;
735 		case 7:
736 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
737 				vmx_inc_test_stage();
738 			break;
739 		case 8:
740 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
741 				vmx_inc_test_stage();
742 			break;
743 		case 9:
744 		case 10:
745 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
746 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
747 			vmx_inc_test_stage();
748 			break;
749 		default:
750 			// Should not reach here
751 			report_fail("unexpected stage, %d",
752 				    vmx_get_test_stage());
753 			print_vmexit_info(exit_reason);
754 			return VMX_TEST_VMEXIT;
755 		}
756 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
757 		return VMX_TEST_RESUME;
758 	case VMX_VMCALL:
759 		switch (vmx_get_test_stage()) {
760 		case 9:
761 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
762 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
763 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
764 			break;
765 		case 10:
766 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
767 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
768 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
769 			break;
770 		default:
771 			// Should not reach here
772 			report_fail("unexpected stage, %d",
773 				    vmx_get_test_stage());
774 			print_vmexit_info(exit_reason);
775 			return VMX_TEST_VMEXIT;
776 		}
777 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
778 		return VMX_TEST_RESUME;
779 	default:
780 		printf("guest_rip = %#lx\n", guest_rip);
781 		printf("\tERROR : Unknown exit reason, 0x%x\n", exit_reason.full);
782 		break;
783 	}
784 	return VMX_TEST_VMEXIT;
785 }
786 
787 #define INSN_CPU0		0
788 #define INSN_CPU1		1
789 #define INSN_ALWAYS_TRAP	2
790 
791 #define FIELD_EXIT_QUAL		(1 << 0)
792 #define FIELD_INSN_INFO		(1 << 1)
793 
794 asm(
795 	"insn_hlt: hlt;ret\n\t"
796 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
797 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
798 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
799 	"insn_rdtsc: rdtsc;ret\n\t"
800 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
801 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
802 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
803 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
804 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
805 	"insn_pause: pause;ret\n\t"
806 	"insn_wbinvd: wbinvd;ret\n\t"
807 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
808 	"insn_invd: invd;ret\n\t"
809 	"insn_sgdt: sgdt gdt_descr;ret\n\t"
810 	"insn_lgdt: lgdt gdt_descr;ret\n\t"
811 	"insn_sidt: sidt idt_descr;ret\n\t"
812 	"insn_lidt: lidt idt_descr;ret\n\t"
813 	"insn_sldt: sldt %ax;ret\n\t"
814 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
815 	"insn_str: str %ax;ret\n\t"
816 	"insn_rdrand: rdrand %rax;ret\n\t"
817 	"insn_rdseed: rdseed %rax;ret\n\t"
818 );
819 extern void insn_hlt(void);
820 extern void insn_invlpg(void);
821 extern void insn_mwait(void);
822 extern void insn_rdpmc(void);
823 extern void insn_rdtsc(void);
824 extern void insn_cr3_load(void);
825 extern void insn_cr3_store(void);
826 extern void insn_cr8_load(void);
827 extern void insn_cr8_store(void);
828 extern void insn_monitor(void);
829 extern void insn_pause(void);
830 extern void insn_wbinvd(void);
831 extern void insn_sgdt(void);
832 extern void insn_lgdt(void);
833 extern void insn_sidt(void);
834 extern void insn_lidt(void);
835 extern void insn_sldt(void);
836 extern void insn_lldt(void);
837 extern void insn_str(void);
838 extern void insn_cpuid(void);
839 extern void insn_invd(void);
840 extern void insn_rdrand(void);
841 extern void insn_rdseed(void);
842 
843 u32 cur_insn;
844 u64 cr3;
845 
846 typedef bool (*supported_fn)(void);
847 
848 static bool this_cpu_has_mwait(void)
849 {
850 	return this_cpu_has(X86_FEATURE_MWAIT);
851 }
852 
853 struct insn_table {
854 	const char *name;
855 	u32 flag;
856 	void (*insn_func)(void);
857 	u32 type;
858 	u32 reason;
859 	ulong exit_qual;
860 	u32 insn_info;
861 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
862 	// which field need to be tested, reason is always tested
863 	u32 test_field;
864 	const supported_fn supported_fn;
865 	u8 disabled;
866 };
867 
868 /*
869  * Add more test cases of instruction intercept here. Elements in this
870  * table is:
871  *	name/control flag/insn function/type/exit reason/exit qulification/
872  *	instruction info/field to test
873  * The last field defines which fields (exit_qual and insn_info) need to be
874  * tested in exit handler. If set to 0, only "reason" is checked.
875  */
876 static struct insn_table insn_table[] = {
877 	// Flags for Primary Processor-Based VM-Execution Controls
878 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
879 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
880 		0x12345678, 0, FIELD_EXIT_QUAL},
881 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0, this_cpu_has_mwait},
882 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0, this_cpu_has_pmu},
883 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
884 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
885 		FIELD_EXIT_QUAL},
886 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
887 		FIELD_EXIT_QUAL},
888 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
889 		FIELD_EXIT_QUAL},
890 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
891 		FIELD_EXIT_QUAL},
892 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0, this_cpu_has_mwait},
893 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
894 	// Flags for Secondary Processor-Based VM-Execution Controls
895 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
896 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
897 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
898 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
899 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
900 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
901 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
902 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
903 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
904 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
905 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
906 	// Instructions always trap
907 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
908 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
909 	// Instructions never trap
910 	{NULL},
911 };
912 
913 static int insn_intercept_init(struct vmcs *vmcs)
914 {
915 	u32 ctrl_cpu, cur_insn;
916 
917 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
918 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
919 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
920 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
921 	cr3 = read_cr3();
922 
923 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
924 		if (insn_table[cur_insn].supported_fn == NULL)
925 			continue;
926 		insn_table[cur_insn].disabled = !insn_table[cur_insn].supported_fn();
927 	}
928 	return VMX_TEST_START;
929 }
930 
931 static void insn_intercept_main(void)
932 {
933 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
934 		vmx_set_test_stage(cur_insn * 2);
935 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
936 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
937 		    (insn_table[cur_insn].type == INSN_CPU1 &&
938 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
939 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
940 			       insn_table[cur_insn].type - INSN_CPU0,
941 			       insn_table[cur_insn].name);
942 			continue;
943 		}
944 
945 		if (insn_table[cur_insn].disabled) {
946 			printf("\tFeature required for %s is not supported.\n",
947 			       insn_table[cur_insn].name);
948 			continue;
949 		}
950 
951 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
952 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
953 		    (insn_table[cur_insn].type == INSN_CPU1 &&
954 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
955 			/* skip hlt, it stalls the guest and is tested below */
956 			if (insn_table[cur_insn].insn_func != insn_hlt)
957 				insn_table[cur_insn].insn_func();
958 			report(vmx_get_test_stage() == cur_insn * 2,
959 					"execute %s",
960 					insn_table[cur_insn].name);
961 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
962 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
963 			       insn_table[cur_insn].type - INSN_CPU0,
964 			       insn_table[cur_insn].name);
965 
966 		vmcall();
967 
968 		insn_table[cur_insn].insn_func();
969 		report(vmx_get_test_stage() == cur_insn * 2 + 1,
970 				"intercept %s",
971 				insn_table[cur_insn].name);
972 
973 		vmx_set_test_stage(cur_insn * 2 + 1);
974 		vmcall();
975 	}
976 }
977 
978 static int insn_intercept_exit_handler(union exit_reason exit_reason)
979 {
980 	u64 guest_rip;
981 	ulong exit_qual;
982 	u32 insn_len;
983 	u32 insn_info;
984 	bool pass;
985 
986 	guest_rip = vmcs_read(GUEST_RIP);
987 	exit_qual = vmcs_read(EXI_QUALIFICATION);
988 	insn_len = vmcs_read(EXI_INST_LEN);
989 	insn_info = vmcs_read(EXI_INST_INFO);
990 
991 	if (exit_reason.basic == VMX_VMCALL) {
992 		u32 val = 0;
993 
994 		if (insn_table[cur_insn].type == INSN_CPU0)
995 			val = vmcs_read(CPU_EXEC_CTRL0);
996 		else if (insn_table[cur_insn].type == INSN_CPU1)
997 			val = vmcs_read(CPU_EXEC_CTRL1);
998 
999 		if (vmx_get_test_stage() & 1)
1000 			val &= ~insn_table[cur_insn].flag;
1001 		else
1002 			val |= insn_table[cur_insn].flag;
1003 
1004 		if (insn_table[cur_insn].type == INSN_CPU0)
1005 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
1006 		else if (insn_table[cur_insn].type == INSN_CPU1)
1007 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
1008 	} else {
1009 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
1010 			insn_table[cur_insn].reason == exit_reason.full;
1011 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
1012 		    insn_table[cur_insn].exit_qual != exit_qual)
1013 			pass = false;
1014 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
1015 		    insn_table[cur_insn].insn_info != insn_info)
1016 			pass = false;
1017 		if (pass)
1018 			vmx_inc_test_stage();
1019 	}
1020 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1021 	return VMX_TEST_RESUME;
1022 }
1023 
1024 /**
1025  * __setup_ept - Setup the VMCS fields to enable Extended Page Tables (EPT)
1026  * @hpa:	Host physical address of the top-level, a.k.a. root, EPT table
1027  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1028  *
1029  * Returns 0 on success, 1 on failure.
1030  *
1031  * Note that @hpa doesn't need to point at actual memory if VM-Launch is
1032  * expected to fail, e.g. setup_dummy_ept() arbitrarily passes '0' to satisfy
1033  * the various EPTP consistency checks, but doesn't ensure backing for HPA '0'.
1034  */
1035 static int __setup_ept(u64 hpa, bool enable_ad)
1036 {
1037 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1038 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1039 		printf("\tEPT is not supported\n");
1040 		return 1;
1041 	}
1042 	if (!(ept_vpid.val & EPT_CAP_WB)) {
1043 		printf("\tWB memtype for EPT walks not supported\n");
1044 		return 1;
1045 	}
1046 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
1047 		printf("\tPWL4 is not supported\n");
1048 		return 1;
1049 	}
1050 
1051 	eptp = EPT_MEM_TYPE_WB;
1052 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1053 	eptp |= hpa;
1054 	if (enable_ad)
1055 		eptp |= EPTP_AD_FLAG;
1056 
1057 	vmcs_write(EPTP, eptp);
1058 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY);
1059 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT);
1060 
1061 	return 0;
1062 }
1063 
1064 /**
1065  * setup_ept - Enable Extended Page Tables (EPT) and setup an identity map
1066  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1067  *
1068  * Returns 0 on success, 1 on failure.
1069  *
1070  * This is the "real" function for setting up EPT tables, i.e. use this for
1071  * tests that need to run code in the guest with EPT enabled.
1072  */
1073 static int setup_ept(bool enable_ad)
1074 {
1075 	unsigned long end_of_memory;
1076 
1077 	pml4 = alloc_page();
1078 
1079 	if (__setup_ept(virt_to_phys(pml4), enable_ad))
1080 		return 1;
1081 
1082 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1083 	if (end_of_memory < (1ul << 32))
1084 		end_of_memory = (1ul << 32);
1085 	/* Cannot use large EPT pages if we need to track EPT
1086 	 * accessed/dirty bits at 4K granularity.
1087 	 */
1088 	setup_ept_range(pml4, 0, end_of_memory, 0,
1089 			!enable_ad && ept_2m_supported(),
1090 			EPT_WA | EPT_RA | EPT_EA);
1091 	return 0;
1092 }
1093 
1094 /**
1095  * setup_dummy_ept - Enable Extended Page Tables (EPT) with a dummy root HPA
1096  *
1097  * Setup EPT using a semi-arbitrary dummy root HPA.  This function is intended
1098  * for use by tests that need EPT enabled to verify dependent VMCS controls
1099  * but never expect to fully enter the guest, i.e. don't need setup the actual
1100  * EPT tables.
1101  */
1102 static void setup_dummy_ept(void)
1103 {
1104 	if (__setup_ept(0, false))
1105 		report_abort("EPT setup unexpectedly failed");
1106 }
1107 
1108 static int enable_unrestricted_guest(bool need_valid_ept)
1109 {
1110 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1111 	    !(ctrl_cpu_rev[1].clr & CPU_URG) ||
1112 	    !(ctrl_cpu_rev[1].clr & CPU_EPT))
1113 		return 1;
1114 
1115 	if (need_valid_ept)
1116 		setup_ept(false);
1117 	else
1118 		setup_dummy_ept();
1119 
1120 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
1121 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
1122 
1123 	return 0;
1124 }
1125 
1126 static void ept_enable_ad_bits(void)
1127 {
1128 	eptp |= EPTP_AD_FLAG;
1129 	vmcs_write(EPTP, eptp);
1130 }
1131 
1132 static void ept_disable_ad_bits(void)
1133 {
1134 	eptp &= ~EPTP_AD_FLAG;
1135 	vmcs_write(EPTP, eptp);
1136 }
1137 
1138 static int ept_ad_enabled(void)
1139 {
1140 	return eptp & EPTP_AD_FLAG;
1141 }
1142 
1143 static void ept_enable_ad_bits_or_skip_test(void)
1144 {
1145 	if (!ept_ad_bits_supported())
1146 		test_skip("EPT AD bits not supported.");
1147 	ept_enable_ad_bits();
1148 }
1149 
1150 static int apic_version;
1151 
1152 static int ept_init_common(bool have_ad)
1153 {
1154 	int ret;
1155 	struct pci_dev pcidev;
1156 
1157 	/* INVEPT is required by the EPT violation handler. */
1158 	if (!is_invept_type_supported(INVEPT_SINGLE))
1159 		return VMX_TEST_EXIT;
1160 
1161 	if (setup_ept(have_ad))
1162 		return VMX_TEST_EXIT;
1163 
1164 	data_page1 = alloc_page();
1165 	data_page2 = alloc_page();
1166 	*((u32 *)data_page1) = MAGIC_VAL_1;
1167 	*((u32 *)data_page2) = MAGIC_VAL_2;
1168 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1169 			EPT_RA | EPT_WA | EPT_EA);
1170 
1171 	apic_version = apic_read(APIC_LVR);
1172 
1173 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1174 	if (ret != PCIDEVADDR_INVALID) {
1175 		pci_dev_init(&pcidev, ret);
1176 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1177 	}
1178 
1179 	return VMX_TEST_START;
1180 }
1181 
1182 static int ept_init(struct vmcs *vmcs)
1183 {
1184 	return ept_init_common(false);
1185 }
1186 
1187 static void ept_common(void)
1188 {
1189 	vmx_set_test_stage(0);
1190 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1191 			*((u32 *)data_page1) != MAGIC_VAL_1)
1192 		report_fail("EPT basic framework - read");
1193 	else {
1194 		*((u32 *)data_page2) = MAGIC_VAL_3;
1195 		vmcall();
1196 		if (vmx_get_test_stage() == 1) {
1197 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1198 					*((u32 *)data_page2) == MAGIC_VAL_2)
1199 				report_pass("EPT basic framework");
1200 			else
1201 				report_pass("EPT basic framework - remap");
1202 		}
1203 	}
1204 	// Test EPT Misconfigurations
1205 	vmx_set_test_stage(1);
1206 	vmcall();
1207 	*((u32 *)data_page1) = MAGIC_VAL_1;
1208 	if (vmx_get_test_stage() != 2) {
1209 		report_fail("EPT misconfigurations");
1210 		goto t1;
1211 	}
1212 	vmx_set_test_stage(2);
1213 	vmcall();
1214 	*((u32 *)data_page1) = MAGIC_VAL_1;
1215 	report(vmx_get_test_stage() == 3, "EPT misconfigurations");
1216 t1:
1217 	// Test EPT violation
1218 	vmx_set_test_stage(3);
1219 	vmcall();
1220 	*((u32 *)data_page1) = MAGIC_VAL_1;
1221 	report(vmx_get_test_stage() == 4, "EPT violation - page permission");
1222 	// Violation caused by EPT paging structure
1223 	vmx_set_test_stage(4);
1224 	vmcall();
1225 	*((u32 *)data_page1) = MAGIC_VAL_2;
1226 	report(vmx_get_test_stage() == 5, "EPT violation - paging structure");
1227 
1228 	// MMIO Read/Write
1229 	vmx_set_test_stage(5);
1230 	vmcall();
1231 
1232 	*(u32 volatile *)pci_physaddr;
1233 	report(vmx_get_test_stage() == 6, "MMIO EPT violation - read");
1234 
1235 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1236 	report(vmx_get_test_stage() == 7, "MMIO EPT violation - write");
1237 }
1238 
1239 static void ept_main(void)
1240 {
1241 	ept_common();
1242 
1243 	// Test EPT access to L1 MMIO
1244 	vmx_set_test_stage(7);
1245 	report(*((u32 *)0xfee00030UL) == apic_version, "EPT - MMIO access");
1246 
1247 	// Test invalid operand for INVEPT
1248 	vmcall();
1249 	report(vmx_get_test_stage() == 8, "EPT - unsupported INVEPT");
1250 }
1251 
1252 static bool invept_test(int type, u64 eptp)
1253 {
1254 	bool ret, supported;
1255 
1256 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1257 	ret = __invept(type, eptp);
1258 
1259 	if (ret == !supported)
1260 		return false;
1261 
1262 	if (!supported)
1263 		printf("WARNING: unsupported invept passed!\n");
1264 	else
1265 		printf("WARNING: invept failed!\n");
1266 
1267 	return true;
1268 }
1269 
1270 static int pml_exit_handler(union exit_reason exit_reason)
1271 {
1272 	u16 index, count;
1273 	u64 *pmlbuf = pml_log;
1274 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1275 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1276 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1277 
1278 	switch (exit_reason.basic) {
1279 	case VMX_VMCALL:
1280 		switch (vmx_get_test_stage()) {
1281 		case 0:
1282 			index = vmcs_read(GUEST_PML_INDEX);
1283 			for (count = index + 1; count < PML_INDEX; count++) {
1284 				if (pmlbuf[count] == (u64)data_page2) {
1285 					vmx_inc_test_stage();
1286 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1287 					break;
1288 				}
1289 			}
1290 			break;
1291 		case 1:
1292 			index = vmcs_read(GUEST_PML_INDEX);
1293 			/* Keep clearing the dirty bit till a overflow */
1294 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1295 			break;
1296 		default:
1297 			report_fail("unexpected stage, %d.",
1298 			       vmx_get_test_stage());
1299 			print_vmexit_info(exit_reason);
1300 			return VMX_TEST_VMEXIT;
1301 		}
1302 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1303 		return VMX_TEST_RESUME;
1304 	case VMX_PML_FULL:
1305 		vmx_inc_test_stage();
1306 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1307 		return VMX_TEST_RESUME;
1308 	default:
1309 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1310 		print_vmexit_info(exit_reason);
1311 	}
1312 	return VMX_TEST_VMEXIT;
1313 }
1314 
1315 static int ept_exit_handler_common(union exit_reason exit_reason, bool have_ad)
1316 {
1317 	u64 guest_rip;
1318 	u64 guest_cr3;
1319 	u32 insn_len;
1320 	u32 exit_qual;
1321 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte,
1322 			     guest_pte_addr;
1323 
1324 	guest_rip = vmcs_read(GUEST_RIP);
1325 	guest_cr3 = vmcs_read(GUEST_CR3);
1326 	insn_len = vmcs_read(EXI_INST_LEN);
1327 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1328 	pteval_t *ptep;
1329 	switch (exit_reason.basic) {
1330 	case VMX_VMCALL:
1331 		switch (vmx_get_test_stage()) {
1332 		case 0:
1333 			check_ept_ad(pml4, guest_cr3,
1334 				     (unsigned long)data_page1,
1335 				     have_ad ? EPT_ACCESS_FLAG : 0,
1336 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1337 			check_ept_ad(pml4, guest_cr3,
1338 				     (unsigned long)data_page2,
1339 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1340 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1341 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1342 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1343 			if (have_ad)
1344 				invept(INVEPT_SINGLE, eptp);
1345 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1346 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1347 				vmx_inc_test_stage();
1348 				install_ept(pml4, (unsigned long)data_page2,
1349 						(unsigned long)data_page2,
1350 						EPT_RA | EPT_WA | EPT_EA);
1351 			} else
1352 				report_fail("EPT basic framework - write");
1353 			break;
1354 		case 1:
1355 			install_ept(pml4, (unsigned long)data_page1,
1356  				(unsigned long)data_page1, EPT_WA);
1357 			invept(INVEPT_SINGLE, eptp);
1358 			break;
1359 		case 2:
1360 			install_ept(pml4, (unsigned long)data_page1,
1361  				(unsigned long)data_page1,
1362  				EPT_RA | EPT_WA | EPT_EA |
1363  				(2 << EPT_MEM_TYPE_SHIFT));
1364 			invept(INVEPT_SINGLE, eptp);
1365 			break;
1366 		case 3:
1367 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1368 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1369 						1, &data_page1_pte));
1370 			set_ept_pte(pml4, (unsigned long)data_page1,
1371 				1, data_page1_pte & ~EPT_PRESENT);
1372 			invept(INVEPT_SINGLE, eptp);
1373 			break;
1374 		case 4:
1375 			ptep = get_pte_level((pgd_t *)guest_cr3, data_page1, /*level=*/2);
1376 			guest_pte_addr = virt_to_phys(ptep) & PAGE_MASK;
1377 
1378 			TEST_ASSERT(get_ept_pte(pml4, guest_pte_addr, 2, &data_page1_pte_pte));
1379 			set_ept_pte(pml4, guest_pte_addr, 2,
1380 				data_page1_pte_pte & ~EPT_PRESENT);
1381 			invept(INVEPT_SINGLE, eptp);
1382 			break;
1383 		case 5:
1384 			install_ept(pml4, (unsigned long)pci_physaddr,
1385 				(unsigned long)pci_physaddr, 0);
1386 			invept(INVEPT_SINGLE, eptp);
1387 			break;
1388 		case 7:
1389 			if (!invept_test(0, eptp))
1390 				vmx_inc_test_stage();
1391 			break;
1392 		// Should not reach here
1393 		default:
1394 			report_fail("ERROR - unexpected stage, %d.",
1395 			       vmx_get_test_stage());
1396 			print_vmexit_info(exit_reason);
1397 			return VMX_TEST_VMEXIT;
1398 		}
1399 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1400 		return VMX_TEST_RESUME;
1401 	case VMX_EPT_MISCONFIG:
1402 		switch (vmx_get_test_stage()) {
1403 		case 1:
1404 		case 2:
1405 			vmx_inc_test_stage();
1406 			install_ept(pml4, (unsigned long)data_page1,
1407  				(unsigned long)data_page1,
1408  				EPT_RA | EPT_WA | EPT_EA);
1409 			invept(INVEPT_SINGLE, eptp);
1410 			break;
1411 		// Should not reach here
1412 		default:
1413 			report_fail("ERROR - unexpected stage, %d.",
1414 			       vmx_get_test_stage());
1415 			print_vmexit_info(exit_reason);
1416 			return VMX_TEST_VMEXIT;
1417 		}
1418 		return VMX_TEST_RESUME;
1419 	case VMX_EPT_VIOLATION:
1420 		/*
1421 		 * Exit-qualifications are masked not to account for advanced
1422 		 * VM-exit information. Once KVM supports this feature, this
1423 		 * masking should be removed.
1424 		 */
1425 		exit_qual &= ~EPT_VLT_GUEST_MASK;
1426 
1427 		switch(vmx_get_test_stage()) {
1428 		case 3:
1429 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1430 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1431 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1432 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1433 					EPT_VLT_PADDR))
1434 				vmx_inc_test_stage();
1435 			set_ept_pte(pml4, (unsigned long)data_page1,
1436 				1, data_page1_pte | (EPT_PRESENT));
1437 			invept(INVEPT_SINGLE, eptp);
1438 			break;
1439 		case 4:
1440 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1441 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1442 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1443 			if (exit_qual == (EPT_VLT_RD |
1444 					  (have_ad ? EPT_VLT_WR : 0) |
1445 					  EPT_VLT_LADDR_VLD))
1446 				vmx_inc_test_stage();
1447 			set_ept_pte(pml4, guest_pte_addr, 2,
1448 				data_page1_pte_pte | (EPT_PRESENT));
1449 			invept(INVEPT_SINGLE, eptp);
1450 			break;
1451 		case 5:
1452 			if (exit_qual & EPT_VLT_RD)
1453 				vmx_inc_test_stage();
1454 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1455 						1, &memaddr_pte));
1456 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1457 			invept(INVEPT_SINGLE, eptp);
1458 			break;
1459 		case 6:
1460 			if (exit_qual & EPT_VLT_WR)
1461 				vmx_inc_test_stage();
1462 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1463 						1, &memaddr_pte));
1464 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1465 			invept(INVEPT_SINGLE, eptp);
1466 			break;
1467 		default:
1468 			// Should not reach here
1469 			report_fail("ERROR : unexpected stage, %d",
1470 			       vmx_get_test_stage());
1471 			print_vmexit_info(exit_reason);
1472 			return VMX_TEST_VMEXIT;
1473 		}
1474 		return VMX_TEST_RESUME;
1475 	default:
1476 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1477 		print_vmexit_info(exit_reason);
1478 	}
1479 	return VMX_TEST_VMEXIT;
1480 }
1481 
1482 static int ept_exit_handler(union exit_reason exit_reason)
1483 {
1484 	return ept_exit_handler_common(exit_reason, false);
1485 }
1486 
1487 static int eptad_init(struct vmcs *vmcs)
1488 {
1489 	int r = ept_init_common(true);
1490 
1491 	if (r == VMX_TEST_EXIT)
1492 		return r;
1493 
1494 	if (!ept_ad_bits_supported()) {
1495 		printf("\tEPT A/D bits are not supported");
1496 		return VMX_TEST_EXIT;
1497 	}
1498 
1499 	return r;
1500 }
1501 
1502 static int pml_init(struct vmcs *vmcs)
1503 {
1504 	u32 ctrl_cpu;
1505 	int r = eptad_init(vmcs);
1506 
1507 	if (r == VMX_TEST_EXIT)
1508 		return r;
1509 
1510 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1511 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1512 		printf("\tPML is not supported");
1513 		return VMX_TEST_EXIT;
1514 	}
1515 
1516 	pml_log = alloc_page();
1517 	vmcs_write(PMLADDR, (u64)pml_log);
1518 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1519 
1520 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1521 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1522 
1523 	return VMX_TEST_START;
1524 }
1525 
1526 static void pml_main(void)
1527 {
1528 	int count = 0;
1529 
1530 	vmx_set_test_stage(0);
1531 	*((u32 *)data_page2) = 0x1;
1532 	vmcall();
1533 	report(vmx_get_test_stage() == 1, "PML - Dirty GPA Logging");
1534 
1535 	while (vmx_get_test_stage() == 1) {
1536 		vmcall();
1537 		*((u32 *)data_page2) = 0x1;
1538 		if (count++ > PML_INDEX)
1539 			break;
1540 	}
1541 	report(vmx_get_test_stage() == 2, "PML Full Event");
1542 }
1543 
1544 static void eptad_main(void)
1545 {
1546 	ept_common();
1547 }
1548 
1549 static int eptad_exit_handler(union exit_reason exit_reason)
1550 {
1551 	return ept_exit_handler_common(exit_reason, true);
1552 }
1553 
1554 #define TIMER_VECTOR	222
1555 
1556 static volatile bool timer_fired;
1557 
1558 static void timer_isr(isr_regs_t *regs)
1559 {
1560 	timer_fired = true;
1561 	apic_write(APIC_EOI, 0);
1562 }
1563 
1564 static int interrupt_init(struct vmcs *vmcs)
1565 {
1566 	msr_bmp_init();
1567 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1568 	handle_irq(TIMER_VECTOR, timer_isr);
1569 	return VMX_TEST_START;
1570 }
1571 
1572 static void interrupt_main(void)
1573 {
1574 	long long start, loops;
1575 
1576 	vmx_set_test_stage(0);
1577 
1578 	apic_write(APIC_LVTT, TIMER_VECTOR);
1579 	irq_enable();
1580 
1581 	apic_write(APIC_TMICT, 1);
1582 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1583 		asm volatile ("nop");
1584 	report(timer_fired, "direct interrupt while running guest");
1585 
1586 	apic_write(APIC_TMICT, 0);
1587 	irq_disable();
1588 	vmcall();
1589 	timer_fired = false;
1590 	apic_write(APIC_TMICT, 1);
1591 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1592 		asm volatile ("nop");
1593 	report(timer_fired, "intercepted interrupt while running guest");
1594 
1595 	irq_enable();
1596 	apic_write(APIC_TMICT, 0);
1597 	irq_disable();
1598 	vmcall();
1599 	timer_fired = false;
1600 	start = rdtsc();
1601 	apic_write(APIC_TMICT, 1000000);
1602 
1603 	safe_halt();
1604 
1605 	report(rdtsc() - start > 1000000 && timer_fired,
1606 	       "direct interrupt + hlt");
1607 
1608 	apic_write(APIC_TMICT, 0);
1609 	irq_disable();
1610 	vmcall();
1611 	timer_fired = false;
1612 	start = rdtsc();
1613 	apic_write(APIC_TMICT, 1000000);
1614 
1615 	safe_halt();
1616 
1617 	report(rdtsc() - start > 10000 && timer_fired,
1618 	       "intercepted interrupt + hlt");
1619 
1620 	apic_write(APIC_TMICT, 0);
1621 	irq_disable();
1622 	vmcall();
1623 	timer_fired = false;
1624 	start = rdtsc();
1625 	apic_write(APIC_TMICT, 1000000);
1626 
1627 	irq_enable();
1628 	asm volatile ("nop");
1629 	vmcall();
1630 
1631 	report(rdtsc() - start > 10000 && timer_fired,
1632 	       "direct interrupt + activity state hlt");
1633 
1634 	apic_write(APIC_TMICT, 0);
1635 	irq_disable();
1636 	vmcall();
1637 	timer_fired = false;
1638 	start = rdtsc();
1639 	apic_write(APIC_TMICT, 1000000);
1640 
1641 	irq_enable();
1642 	asm volatile ("nop");
1643 	vmcall();
1644 
1645 	report(rdtsc() - start > 10000 && timer_fired,
1646 	       "intercepted interrupt + activity state hlt");
1647 
1648 	apic_write(APIC_TMICT, 0);
1649 	irq_disable();
1650 	vmx_set_test_stage(7);
1651 	vmcall();
1652 	timer_fired = false;
1653 	apic_write(APIC_TMICT, 1);
1654 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1655 		asm volatile ("nop");
1656 	report(timer_fired,
1657 	       "running a guest with interrupt acknowledgement set");
1658 
1659 	apic_write(APIC_TMICT, 0);
1660 	irq_enable();
1661 	timer_fired = false;
1662 	vmcall();
1663 	report(timer_fired, "Inject an event to a halted guest");
1664 }
1665 
1666 static int interrupt_exit_handler(union exit_reason exit_reason)
1667 {
1668 	u64 guest_rip = vmcs_read(GUEST_RIP);
1669 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1670 
1671 	switch (exit_reason.basic) {
1672 	case VMX_VMCALL:
1673 		switch (vmx_get_test_stage()) {
1674 		case 0:
1675 		case 2:
1676 		case 5:
1677 			vmcs_write(PIN_CONTROLS,
1678 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1679 			break;
1680 		case 7:
1681 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1682 			vmcs_write(PIN_CONTROLS,
1683 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1684 			break;
1685 		case 1:
1686 		case 3:
1687 			vmcs_write(PIN_CONTROLS,
1688 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1689 			break;
1690 		case 4:
1691 		case 6:
1692 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1693 			break;
1694 
1695 		case 8:
1696 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1697 			vmcs_write(ENT_INTR_INFO,
1698 				   TIMER_VECTOR |
1699 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1700 				   INTR_INFO_VALID_MASK);
1701 			break;
1702 		}
1703 		vmx_inc_test_stage();
1704 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1705 		return VMX_TEST_RESUME;
1706 	case VMX_EXTINT:
1707 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1708 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1709 			handle_external_interrupt(vector);
1710 		} else {
1711 			irq_enable();
1712 			asm volatile ("nop");
1713 			irq_disable();
1714 		}
1715 		if (vmx_get_test_stage() >= 2)
1716 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1717 		return VMX_TEST_RESUME;
1718 	default:
1719 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1720 		print_vmexit_info(exit_reason);
1721 	}
1722 
1723 	return VMX_TEST_VMEXIT;
1724 }
1725 
1726 
1727 static volatile int nmi_fired;
1728 
1729 #define NMI_DELAY 100000000ULL
1730 
1731 static void nmi_isr(isr_regs_t *regs)
1732 {
1733 	nmi_fired = true;
1734 }
1735 
1736 static int nmi_hlt_init(struct vmcs *vmcs)
1737 {
1738 	msr_bmp_init();
1739 	handle_irq(NMI_VECTOR, nmi_isr);
1740 	vmcs_write(PIN_CONTROLS,
1741 		   vmcs_read(PIN_CONTROLS) & ~PIN_NMI);
1742 	vmcs_write(PIN_CONTROLS,
1743 		   vmcs_read(PIN_CONTROLS) & ~PIN_VIRT_NMI);
1744 	return VMX_TEST_START;
1745 }
1746 
1747 static void nmi_message_thread(void *data)
1748 {
1749     while (vmx_get_test_stage() != 1)
1750         pause();
1751 
1752     delay(NMI_DELAY);
1753     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1754 
1755     while (vmx_get_test_stage() != 2)
1756         pause();
1757 
1758     delay(NMI_DELAY);
1759     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1760 }
1761 
1762 static void nmi_hlt_main(void)
1763 {
1764     long long start;
1765 
1766     if (cpu_count() < 2) {
1767         report_skip("%s : CPU count < 2", __func__);
1768         vmx_set_test_stage(-1);
1769         return;
1770     }
1771 
1772     vmx_set_test_stage(0);
1773     on_cpu_async(1, nmi_message_thread, NULL);
1774     start = rdtsc();
1775     vmx_set_test_stage(1);
1776     asm volatile ("hlt");
1777     report((rdtsc() - start > NMI_DELAY) && nmi_fired,
1778             "direct NMI + hlt");
1779     if (!nmi_fired)
1780         vmx_set_test_stage(-1);
1781     nmi_fired = false;
1782 
1783     vmcall();
1784 
1785     start = rdtsc();
1786     vmx_set_test_stage(2);
1787     asm volatile ("hlt");
1788     report((rdtsc() - start > NMI_DELAY) && !nmi_fired,
1789             "intercepted NMI + hlt");
1790     if (nmi_fired) {
1791         report(!nmi_fired, "intercepted NMI was dispatched");
1792         vmx_set_test_stage(-1);
1793         return;
1794     }
1795     vmx_set_test_stage(3);
1796 }
1797 
1798 static int nmi_hlt_exit_handler(union exit_reason exit_reason)
1799 {
1800     u64 guest_rip = vmcs_read(GUEST_RIP);
1801     u32 insn_len = vmcs_read(EXI_INST_LEN);
1802 
1803     switch (vmx_get_test_stage()) {
1804     case 1:
1805         if (exit_reason.basic != VMX_VMCALL) {
1806             report_fail("VMEXIT not due to vmcall. Exit reason 0x%x",
1807                         exit_reason.full);
1808             print_vmexit_info(exit_reason);
1809             return VMX_TEST_VMEXIT;
1810         }
1811 
1812         vmcs_write(PIN_CONTROLS,
1813                vmcs_read(PIN_CONTROLS) | PIN_NMI);
1814         vmcs_write(PIN_CONTROLS,
1815                vmcs_read(PIN_CONTROLS) | PIN_VIRT_NMI);
1816         vmcs_write(GUEST_RIP, guest_rip + insn_len);
1817         break;
1818 
1819     case 2:
1820         if (exit_reason.basic != VMX_EXC_NMI) {
1821             report_fail("VMEXIT not due to NMI intercept. Exit reason 0x%x",
1822                         exit_reason.full);
1823             print_vmexit_info(exit_reason);
1824             return VMX_TEST_VMEXIT;
1825         }
1826         report_pass("NMI intercept while running guest");
1827         vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1828         break;
1829 
1830     case 3:
1831         break;
1832 
1833     default:
1834         return VMX_TEST_VMEXIT;
1835     }
1836 
1837     if (vmx_get_test_stage() == 3)
1838         return VMX_TEST_VMEXIT;
1839 
1840     return VMX_TEST_RESUME;
1841 }
1842 
1843 
1844 static int dbgctls_init(struct vmcs *vmcs)
1845 {
1846 	u64 dr7 = 0x402;
1847 	u64 zero = 0;
1848 
1849 	msr_bmp_init();
1850 	asm volatile(
1851 		"mov %0,%%dr0\n\t"
1852 		"mov %0,%%dr1\n\t"
1853 		"mov %0,%%dr2\n\t"
1854 		"mov %1,%%dr7\n\t"
1855 		: : "r" (zero), "r" (dr7));
1856 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1857 	vmcs_write(GUEST_DR7, 0x404);
1858 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1859 
1860 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1861 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1862 
1863 	return VMX_TEST_START;
1864 }
1865 
1866 static void dbgctls_main(void)
1867 {
1868 	u64 dr7, debugctl;
1869 
1870 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1871 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1872 	/* Commented out: KVM does not support DEBUGCTL so far */
1873 	(void)debugctl;
1874 	report(dr7 == 0x404, "Load debug controls" /* && debugctl == 0x2 */);
1875 
1876 	dr7 = 0x408;
1877 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1878 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1879 
1880 	vmx_set_test_stage(0);
1881 	vmcall();
1882 	report(vmx_get_test_stage() == 1, "Save debug controls");
1883 
1884 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1885 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1886 		printf("\tDebug controls are always loaded/saved\n");
1887 		return;
1888 	}
1889 	vmx_set_test_stage(2);
1890 	vmcall();
1891 
1892 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1893 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1894 	/* Commented out: KVM does not support DEBUGCTL so far */
1895 	(void)debugctl;
1896 	report(dr7 == 0x402,
1897 	       "Guest=host debug controls" /* && debugctl == 0x1 */);
1898 
1899 	dr7 = 0x408;
1900 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1901 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1902 
1903 	vmx_set_test_stage(3);
1904 	vmcall();
1905 	report(vmx_get_test_stage() == 4, "Don't save debug controls");
1906 }
1907 
1908 static int dbgctls_exit_handler(union exit_reason exit_reason)
1909 {
1910 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1911 	u64 guest_rip = vmcs_read(GUEST_RIP);
1912 	u64 dr7, debugctl;
1913 
1914 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1915 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1916 
1917 	switch (exit_reason.basic) {
1918 	case VMX_VMCALL:
1919 		switch (vmx_get_test_stage()) {
1920 		case 0:
1921 			if (dr7 == 0x400 && debugctl == 0 &&
1922 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1923 			    Commented out: KVM does not support DEBUGCTL so far
1924 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1925 				vmx_inc_test_stage();
1926 			break;
1927 		case 2:
1928 			dr7 = 0x402;
1929 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1930 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1931 			vmcs_write(GUEST_DR7, 0x404);
1932 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1933 
1934 			vmcs_write(ENT_CONTROLS,
1935 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1936 			vmcs_write(EXI_CONTROLS,
1937 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1938 			break;
1939 		case 3:
1940 			if (dr7 == 0x400 && debugctl == 0 &&
1941 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1942 			    Commented out: KVM does not support DEBUGCTL so far
1943 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1944 				vmx_inc_test_stage();
1945 			break;
1946 		}
1947 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1948 		return VMX_TEST_RESUME;
1949 	default:
1950 		report_fail("Unknown exit reason, %d", exit_reason.full);
1951 		print_vmexit_info(exit_reason);
1952 	}
1953 	return VMX_TEST_VMEXIT;
1954 }
1955 
1956 struct vmx_msr_entry {
1957 	u32 index;
1958 	u32 reserved;
1959 	u64 value;
1960 } __attribute__((packed));
1961 
1962 #define MSR_MAGIC 0x31415926
1963 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1964 
1965 static int msr_switch_init(struct vmcs *vmcs)
1966 {
1967 	msr_bmp_init();
1968 	exit_msr_store = alloc_page();
1969 	exit_msr_load = alloc_page();
1970 	entry_msr_load = alloc_page();
1971 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1972 	entry_msr_load[0].value = MSR_MAGIC;
1973 
1974 	vmx_set_test_stage(1);
1975 	vmcs_write(ENT_MSR_LD_CNT, 1);
1976 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1977 	vmcs_write(EXI_MSR_ST_CNT, 1);
1978 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1979 	vmcs_write(EXI_MSR_LD_CNT, 1);
1980 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1981 	return VMX_TEST_START;
1982 }
1983 
1984 static void msr_switch_main(void)
1985 {
1986 	if (vmx_get_test_stage() == 1) {
1987 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC,
1988 		       "VM entry MSR load");
1989 		vmx_set_test_stage(2);
1990 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
1991 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
1992 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
1993 		exit_msr_load[0].value = MSR_MAGIC + 2;
1994 	}
1995 	vmcall();
1996 }
1997 
1998 static int msr_switch_exit_handler(union exit_reason exit_reason)
1999 {
2000 	if (exit_reason.basic == VMX_VMCALL && vmx_get_test_stage() == 2) {
2001 		report(exit_msr_store[0].value == MSR_MAGIC + 1,
2002 		       "VM exit MSR store");
2003 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2,
2004 		       "VM exit MSR load");
2005 		vmx_set_test_stage(3);
2006 		entry_msr_load[0].index = MSR_FS_BASE;
2007 		return VMX_TEST_RESUME;
2008 	}
2009 	printf("ERROR %s: unexpected stage=%u or reason=0x%x\n",
2010 		__func__, vmx_get_test_stage(), exit_reason.full);
2011 	return VMX_TEST_EXIT;
2012 }
2013 
2014 static int msr_switch_entry_failure(struct vmentry_result *result)
2015 {
2016 	if (result->vm_fail) {
2017 		printf("ERROR %s: VM-Fail on %s\n", __func__, result->instr);
2018 		return VMX_TEST_EXIT;
2019 	}
2020 
2021 	if (result->exit_reason.failed_vmentry &&
2022 	    result->exit_reason.basic == VMX_FAIL_MSR &&
2023 	    vmx_get_test_stage() == 3) {
2024 		report(vmcs_read(EXI_QUALIFICATION) == 1,
2025 		       "VM entry MSR load: try to load FS_BASE");
2026 		return VMX_TEST_VMEXIT;
2027 	}
2028 	printf("ERROR %s: unexpected stage=%u or reason=%x\n",
2029 		__func__, vmx_get_test_stage(), result->exit_reason.full);
2030 	return VMX_TEST_EXIT;
2031 }
2032 
2033 static int vmmcall_init(struct vmcs *vmcs)
2034 {
2035 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
2036 	return VMX_TEST_START;
2037 }
2038 
2039 static void vmmcall_main(void)
2040 {
2041 	asm volatile(
2042 		"mov $0xABCD, %%rax\n\t"
2043 		"vmmcall\n\t"
2044 		::: "rax");
2045 
2046 	report_fail("VMMCALL");
2047 }
2048 
2049 static int vmmcall_exit_handler(union exit_reason exit_reason)
2050 {
2051 	switch (exit_reason.basic) {
2052 	case VMX_VMCALL:
2053 		printf("here\n");
2054 		report_fail("VMMCALL triggers #UD");
2055 		break;
2056 	case VMX_EXC_NMI:
2057 		report((vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR,
2058 		       "VMMCALL triggers #UD");
2059 		break;
2060 	default:
2061 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2062 		print_vmexit_info(exit_reason);
2063 	}
2064 
2065 	return VMX_TEST_VMEXIT;
2066 }
2067 
2068 static int disable_rdtscp_init(struct vmcs *vmcs)
2069 {
2070 	u32 ctrl_cpu1;
2071 
2072 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
2073 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
2074 		ctrl_cpu1 &= ~CPU_RDTSCP;
2075 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
2076 	}
2077 
2078 	return VMX_TEST_START;
2079 }
2080 
2081 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
2082 {
2083 	switch (vmx_get_test_stage()) {
2084 	case 0:
2085 		report_pass("RDTSCP triggers #UD");
2086 		vmx_inc_test_stage();
2087 		regs->rip += 3;
2088 		break;
2089 	case 2:
2090 		report_pass("RDPID triggers #UD");
2091 		vmx_inc_test_stage();
2092 		regs->rip += 4;
2093 		break;
2094 	}
2095 	return;
2096 
2097 }
2098 
2099 static void disable_rdtscp_main(void)
2100 {
2101 	/* Test that #UD is properly injected in L2.  */
2102 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2103 
2104 	vmx_set_test_stage(0);
2105 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2106 	vmcall();
2107 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2108 
2109 	handle_exception(UD_VECTOR, 0);
2110 	vmcall();
2111 }
2112 
2113 static int disable_rdtscp_exit_handler(union exit_reason exit_reason)
2114 {
2115 	switch (exit_reason.basic) {
2116 	case VMX_VMCALL:
2117 		switch (vmx_get_test_stage()) {
2118 		case 0:
2119 			report_fail("RDTSCP triggers #UD");
2120 			vmx_inc_test_stage();
2121 			/* fallthrough */
2122 		case 1:
2123 			vmx_inc_test_stage();
2124 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2125 			return VMX_TEST_RESUME;
2126 		case 2:
2127 			report_fail("RDPID triggers #UD");
2128 			break;
2129 		}
2130 		break;
2131 
2132 	default:
2133 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2134 		print_vmexit_info(exit_reason);
2135 	}
2136 	return VMX_TEST_VMEXIT;
2137 }
2138 
2139 static int int3_init(struct vmcs *vmcs)
2140 {
2141 	vmcs_write(EXC_BITMAP, ~0u);
2142 	return VMX_TEST_START;
2143 }
2144 
2145 static void int3_guest_main(void)
2146 {
2147 	asm volatile ("int3");
2148 }
2149 
2150 static int int3_exit_handler(union exit_reason exit_reason)
2151 {
2152 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2153 
2154 	report(exit_reason.basic == VMX_EXC_NMI &&
2155 	       (intr_info & INTR_INFO_VALID_MASK) &&
2156 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2157 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2158 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2159 	       "L1 intercepts #BP");
2160 
2161 	return VMX_TEST_VMEXIT;
2162 }
2163 
2164 static int into_init(struct vmcs *vmcs)
2165 {
2166 	vmcs_write(EXC_BITMAP, ~0u);
2167 	return VMX_TEST_START;
2168 }
2169 
2170 static void into_guest_main(void)
2171 {
2172 	struct far_pointer32 fp = {
2173 		.offset = (uintptr_t)&&into,
2174 		.selector = KERNEL_CS32,
2175 	};
2176 	uintptr_t rsp;
2177 
2178 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
2179 
2180 	if (fp.offset != (uintptr_t)&&into) {
2181 		printf("Code address too high.\n");
2182 		return;
2183 	}
2184 	if ((u32)rsp != rsp) {
2185 		printf("Stack address too high.\n");
2186 		return;
2187 	}
2188 
2189 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2190 	return;
2191 into:
2192 	asm volatile (".code32;"
2193 		      "movl $0x7fffffff, %eax;"
2194 		      "addl %eax, %eax;"
2195 		      "into;"
2196 		      "lret;"
2197 		      ".code64");
2198 	__builtin_unreachable();
2199 }
2200 
2201 static int into_exit_handler(union exit_reason exit_reason)
2202 {
2203 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2204 
2205 	report(exit_reason.basic == VMX_EXC_NMI &&
2206 	       (intr_info & INTR_INFO_VALID_MASK) &&
2207 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2208 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2209 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2210 	       "L1 intercepts #OF");
2211 
2212 	return VMX_TEST_VMEXIT;
2213 }
2214 
2215 static void exit_monitor_from_l2_main(void)
2216 {
2217 	printf("Calling exit(0) from l2...\n");
2218 	exit(0);
2219 }
2220 
2221 static int exit_monitor_from_l2_handler(union exit_reason exit_reason)
2222 {
2223 	report_fail("The guest should have killed the VMM");
2224 	return VMX_TEST_EXIT;
2225 }
2226 
2227 static void assert_exit_reason(u64 expected)
2228 {
2229 	u64 actual = vmcs_read(EXI_REASON);
2230 
2231 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2232 			   exit_reason_description(expected),
2233 			   exit_reason_description(actual));
2234 }
2235 
2236 static void skip_exit_insn(void)
2237 {
2238 	u64 guest_rip = vmcs_read(GUEST_RIP);
2239 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2240 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2241 }
2242 
2243 static void skip_exit_vmcall(void)
2244 {
2245 	assert_exit_reason(VMX_VMCALL);
2246 	skip_exit_insn();
2247 }
2248 
2249 static void v2_null_test_guest(void)
2250 {
2251 }
2252 
2253 static void v2_null_test(void)
2254 {
2255 	test_set_guest(v2_null_test_guest);
2256 	enter_guest();
2257 	report_pass(__func__);
2258 }
2259 
2260 static void v2_multiple_entries_test_guest(void)
2261 {
2262 	vmx_set_test_stage(1);
2263 	vmcall();
2264 	vmx_set_test_stage(2);
2265 }
2266 
2267 static void v2_multiple_entries_test(void)
2268 {
2269 	test_set_guest(v2_multiple_entries_test_guest);
2270 	enter_guest();
2271 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2272 	skip_exit_vmcall();
2273 	enter_guest();
2274 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2275 	report_pass(__func__);
2276 }
2277 
2278 static int fixture_test_data = 1;
2279 
2280 static void fixture_test_teardown(void *data)
2281 {
2282 	*((int *) data) = 1;
2283 }
2284 
2285 static void fixture_test_guest(void)
2286 {
2287 	fixture_test_data++;
2288 }
2289 
2290 
2291 static void fixture_test_setup(void)
2292 {
2293 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2294 			   "fixture_test_teardown didn't run?!");
2295 	fixture_test_data = 2;
2296 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2297 	test_set_guest(fixture_test_guest);
2298 }
2299 
2300 static void fixture_test_case1(void)
2301 {
2302 	fixture_test_setup();
2303 	TEST_ASSERT_EQ(2, fixture_test_data);
2304 	enter_guest();
2305 	TEST_ASSERT_EQ(3, fixture_test_data);
2306 	report_pass(__func__);
2307 }
2308 
2309 static void fixture_test_case2(void)
2310 {
2311 	fixture_test_setup();
2312 	TEST_ASSERT_EQ(2, fixture_test_data);
2313 	enter_guest();
2314 	TEST_ASSERT_EQ(3, fixture_test_data);
2315 	report_pass(__func__);
2316 }
2317 
2318 enum ept_access_op {
2319 	OP_READ,
2320 	OP_WRITE,
2321 	OP_EXEC,
2322 	OP_FLUSH_TLB,
2323 	OP_EXIT,
2324 };
2325 
2326 static struct ept_access_test_data {
2327 	unsigned long gpa;
2328 	unsigned long *gva;
2329 	unsigned long hpa;
2330 	unsigned long *hva;
2331 	enum ept_access_op op;
2332 } ept_access_test_data;
2333 
2334 extern unsigned char ret42_start;
2335 extern unsigned char ret42_end;
2336 
2337 /* Returns 42. */
2338 asm(
2339 	".align 64\n"
2340 	"ret42_start:\n"
2341 	"mov $42, %eax\n"
2342 	"ret\n"
2343 	"ret42_end:\n"
2344 );
2345 
2346 static void
2347 diagnose_ept_violation_qual(u64 expected, u64 actual)
2348 {
2349 
2350 #define DIAGNOSE(flag)							\
2351 do {									\
2352 	if ((expected & flag) != (actual & flag))			\
2353 		printf(#flag " %sexpected\n",				\
2354 		       (expected & flag) ? "" : "un");			\
2355 } while (0)
2356 
2357 	DIAGNOSE(EPT_VLT_RD);
2358 	DIAGNOSE(EPT_VLT_WR);
2359 	DIAGNOSE(EPT_VLT_FETCH);
2360 	DIAGNOSE(EPT_VLT_PERM_RD);
2361 	DIAGNOSE(EPT_VLT_PERM_WR);
2362 	DIAGNOSE(EPT_VLT_PERM_EX);
2363 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2364 	DIAGNOSE(EPT_VLT_PADDR);
2365 
2366 #undef DIAGNOSE
2367 }
2368 
2369 static void do_ept_access_op(enum ept_access_op op)
2370 {
2371 	ept_access_test_data.op = op;
2372 	enter_guest();
2373 }
2374 
2375 /*
2376  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2377  * needed by tests that modify guest PTEs.
2378  */
2379 static void ept_access_test_guest_flush_tlb(void)
2380 {
2381 	do_ept_access_op(OP_FLUSH_TLB);
2382 	skip_exit_vmcall();
2383 }
2384 
2385 /*
2386  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2387  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2388  * a huge page.
2389  */
2390 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2391 				 unsigned long clear, unsigned long set)
2392 {
2393 	struct ept_access_test_data *data = &ept_access_test_data;
2394 	unsigned long orig_pte;
2395 	unsigned long pte;
2396 
2397 	/* Screw with the mapping at the requested level. */
2398 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2399 	pte = orig_pte;
2400 	if (mkhuge)
2401 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2402 	else
2403 		pte = orig_pte;
2404 	pte = (pte & ~clear) | set;
2405 	set_ept_pte(pml4, gpa, level, pte);
2406 	invept(INVEPT_SINGLE, eptp);
2407 
2408 	return orig_pte;
2409 }
2410 
2411 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2412 {
2413 	set_ept_pte(pml4, gpa, level, orig_pte);
2414 	invept(INVEPT_SINGLE, eptp);
2415 }
2416 
2417 static void do_ept_violation(bool leaf, enum ept_access_op op,
2418 			     u64 expected_qual, u64 expected_paddr)
2419 {
2420 	u64 qual;
2421 
2422 	/* Try the access and observe the violation. */
2423 	do_ept_access_op(op);
2424 
2425 	assert_exit_reason(VMX_EPT_VIOLATION);
2426 
2427 	qual = vmcs_read(EXI_QUALIFICATION);
2428 
2429 	/* Mask undefined bits (which may later be defined in certain cases). */
2430 	qual &= ~(EPT_VLT_GUEST_USER | EPT_VLT_GUEST_RW | EPT_VLT_GUEST_EX |
2431 		 EPT_VLT_PERM_USER_EX);
2432 
2433 	diagnose_ept_violation_qual(expected_qual, qual);
2434 	TEST_EXPECT_EQ(expected_qual, qual);
2435 
2436 	#if 0
2437 	/* Disable for now otherwise every test will fail */
2438 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2439 		       (unsigned long) (
2440 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2441 	#endif
2442 	/*
2443 	 * TODO: tests that probe expected_paddr in pages other than the one at
2444 	 * the beginning of the 1g region.
2445 	 */
2446 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2447 }
2448 
2449 static void
2450 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2451 			      unsigned long set, enum ept_access_op op,
2452 			      u64 expected_qual)
2453 {
2454 	struct ept_access_test_data *data = &ept_access_test_data;
2455 	unsigned long orig_pte;
2456 
2457 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2458 
2459 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2460 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2461 					 data->gpa);
2462 
2463 	/* Fix the violation and resume the op loop. */
2464 	ept_untwiddle(data->gpa, level, orig_pte);
2465 	enter_guest();
2466 	skip_exit_vmcall();
2467 }
2468 
2469 static void
2470 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2471 		       enum ept_access_op op, u64 expected_qual)
2472 {
2473 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2474 				      expected_qual);
2475 	if (ept_huge_pages_supported(level))
2476 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2477 					      expected_qual);
2478 }
2479 
2480 static void ept_violation(unsigned long clear, unsigned long set,
2481 			  enum ept_access_op op, u64 expected_qual)
2482 {
2483 	ept_violation_at_level(1, clear, set, op, expected_qual);
2484 	ept_violation_at_level(2, clear, set, op, expected_qual);
2485 	ept_violation_at_level(3, clear, set, op, expected_qual);
2486 	ept_violation_at_level(4, clear, set, op, expected_qual);
2487 }
2488 
2489 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2490 				       u64 expected_qual)
2491 {
2492 	ept_violation(EPT_PRESENT, access, op,
2493 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2494 }
2495 
2496 /*
2497  * For translations that don't involve a GVA, that is physical address (paddr)
2498  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2499  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2500  * translations don't involve GVAs, such as when the hardware does the guest
2501  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2502  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2503  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2504  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2505  *
2506  * Note that paddr violations can also be triggered by loading PAE page tables
2507  * with wonky addresses. We don't test that yet.
2508  *
2509  * This function modifies the EPT entry that maps the GPA that the guest page
2510  * table entry mapping ept_access_test_data.gva resides on.
2511  *
2512  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2513  *
2514  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2515  *
2516  *	@op		Guest operation to perform with
2517  *			ept_access_test_data.gva.
2518  *
2519  *	@expect_violation
2520  *			Is a violation expected during the paddr access?
2521  *
2522  *	@expected_qual	Expected qualification for the EPT violation.
2523  *			EPT_VLT_PADDR should be clear.
2524  */
2525 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2526 			     enum ept_access_op op, bool expect_violation,
2527 			     u64 expected_qual)
2528 {
2529 	struct ept_access_test_data *data = &ept_access_test_data;
2530 	unsigned long *ptep;
2531 	unsigned long gpa;
2532 	unsigned long orig_epte;
2533 	unsigned long epte;
2534 	int i;
2535 
2536 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2537 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2538 	TEST_ASSERT(ptep);
2539 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2540 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2541 	ept_access_test_guest_flush_tlb();
2542 
2543 	/*
2544 	 * Now modify the access bits on the EPT entry for the GPA that the
2545 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2546 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2547 	 * constructed our test such that those other 511 PTEs aren't used by
2548 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2549 	 * PTE we're modifying is at the beginning of a 4K page and the
2550 	 * following 511 entries are also under our control (and not touched by
2551 	 * the guest).
2552 	 */
2553 	gpa = virt_to_phys(ptep);
2554 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2555 	/*
2556 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2557 	 * otherwise our level=1 twiddling below will fail. We use the
2558 	 * identity map (gpa = gpa) since page tables are shared with the host.
2559 	 */
2560 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2561 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2562 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2563 
2564 	if (expect_violation) {
2565 		do_ept_violation(/*leaf=*/true, op,
2566 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2567 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2568 		do_ept_access_op(op);
2569 	} else {
2570 		do_ept_access_op(op);
2571 		if (ept_ad_enabled()) {
2572 			for (i = EPT_PAGE_LEVEL; i > 0; i--) {
2573 				TEST_ASSERT(get_ept_pte(pml4, gpa, i, &epte));
2574 				TEST_ASSERT(epte & EPT_ACCESS_FLAG);
2575 				if (i == 1)
2576 					TEST_ASSERT(epte & EPT_DIRTY_FLAG);
2577 				else
2578 					TEST_ASSERT_EQ(epte & EPT_DIRTY_FLAG, 0);
2579 			}
2580 		}
2581 
2582 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2583 	}
2584 
2585 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2586 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2587 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2588 
2589 	skip_exit_vmcall();
2590 }
2591 
2592 static void ept_access_allowed_paddr(unsigned long ept_access,
2593 				     unsigned long pte_ad,
2594 				     enum ept_access_op op)
2595 {
2596 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2597 			 /*expected_qual=*/-1);
2598 }
2599 
2600 static void ept_access_violation_paddr(unsigned long ept_access,
2601 				       unsigned long pte_ad,
2602 				       enum ept_access_op op,
2603 				       u64 expected_qual)
2604 {
2605 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2606 			 expected_qual);
2607 }
2608 
2609 
2610 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2611 					unsigned long clear,
2612 					unsigned long set,
2613 					enum ept_access_op op)
2614 {
2615 	struct ept_access_test_data *data = &ept_access_test_data;
2616 	unsigned long orig_pte;
2617 
2618 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2619 
2620 	/* No violation. Should proceed to vmcall. */
2621 	do_ept_access_op(op);
2622 	skip_exit_vmcall();
2623 
2624 	ept_untwiddle(data->gpa, level, orig_pte);
2625 }
2626 
2627 static void ept_allowed_at_level(int level, unsigned long clear,
2628 				 unsigned long set, enum ept_access_op op)
2629 {
2630 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2631 	if (ept_huge_pages_supported(level))
2632 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2633 }
2634 
2635 static void ept_allowed(unsigned long clear, unsigned long set,
2636 			enum ept_access_op op)
2637 {
2638 	ept_allowed_at_level(1, clear, set, op);
2639 	ept_allowed_at_level(2, clear, set, op);
2640 	ept_allowed_at_level(3, clear, set, op);
2641 	ept_allowed_at_level(4, clear, set, op);
2642 }
2643 
2644 static void ept_ignored_bit(int bit)
2645 {
2646 	/* Set the bit. */
2647 	ept_allowed(0, 1ul << bit, OP_READ);
2648 	ept_allowed(0, 1ul << bit, OP_WRITE);
2649 	ept_allowed(0, 1ul << bit, OP_EXEC);
2650 
2651 	/* Clear the bit. */
2652 	ept_allowed(1ul << bit, 0, OP_READ);
2653 	ept_allowed(1ul << bit, 0, OP_WRITE);
2654 	ept_allowed(1ul << bit, 0, OP_EXEC);
2655 }
2656 
2657 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2658 {
2659 	ept_allowed(EPT_PRESENT, access, op);
2660 }
2661 
2662 
2663 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2664 					     unsigned long clear,
2665 					     unsigned long set,
2666 					     enum ept_access_op op)
2667 {
2668 	struct ept_access_test_data *data = &ept_access_test_data;
2669 	unsigned long orig_pte;
2670 
2671 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2672 
2673 	do_ept_access_op(op);
2674 	assert_exit_reason(VMX_EPT_MISCONFIG);
2675 
2676 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2677 	#if 0
2678 	/* broken: */
2679 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2680 	#endif
2681 	#if 0
2682 	/*
2683 	 * broken:
2684 	 * According to description of exit qual for EPT violation,
2685 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2686 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2687 	 * to be set for msiconfig.
2688 	 */
2689 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2690 		       (unsigned long) (
2691 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2692 	#endif
2693 
2694 	/* Fix the violation and resume the op loop. */
2695 	ept_untwiddle(data->gpa, level, orig_pte);
2696 	enter_guest();
2697 	skip_exit_vmcall();
2698 }
2699 
2700 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2701 					  unsigned long clear,
2702 					  unsigned long set)
2703 {
2704 	/* The op shouldn't matter (read, write, exec), so try them all! */
2705 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2706 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2707 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2708 }
2709 
2710 static void ept_misconfig_at_level(int level, unsigned long clear,
2711 				   unsigned long set)
2712 {
2713 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2714 	if (ept_huge_pages_supported(level))
2715 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2716 }
2717 
2718 static void ept_misconfig(unsigned long clear, unsigned long set)
2719 {
2720 	ept_misconfig_at_level(1, clear, set);
2721 	ept_misconfig_at_level(2, clear, set);
2722 	ept_misconfig_at_level(3, clear, set);
2723 	ept_misconfig_at_level(4, clear, set);
2724 }
2725 
2726 static void ept_access_misconfig(unsigned long access)
2727 {
2728 	ept_misconfig(EPT_PRESENT, access);
2729 }
2730 
2731 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2732 {
2733 	/* Setting the bit causes a misconfig. */
2734 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2735 
2736 	/* Making the entry non-present turns reserved bits into ignored. */
2737 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2738 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2739 }
2740 
2741 static void ept_reserved_bit_at_level_huge(int level, int bit)
2742 {
2743 	/* Setting the bit causes a misconfig. */
2744 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2745 
2746 	/* Making the entry non-present turns reserved bits into ignored. */
2747 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2748 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2749 }
2750 
2751 static void ept_reserved_bit_at_level(int level, int bit)
2752 {
2753 	/* Setting the bit causes a misconfig. */
2754 	ept_misconfig_at_level(level, 0, 1ul << bit);
2755 
2756 	/* Making the entry non-present turns reserved bits into ignored. */
2757 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2758 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2759 }
2760 
2761 static void ept_reserved_bit(int bit)
2762 {
2763 	ept_reserved_bit_at_level(1, bit);
2764 	ept_reserved_bit_at_level(2, bit);
2765 	ept_reserved_bit_at_level(3, bit);
2766 	ept_reserved_bit_at_level(4, bit);
2767 }
2768 
2769 #define PAGE_2M_ORDER 9
2770 #define PAGE_1G_ORDER 18
2771 
2772 static void *get_1g_page(void)
2773 {
2774 	static void *alloc;
2775 
2776 	if (!alloc)
2777 		alloc = alloc_pages(PAGE_1G_ORDER);
2778 	return alloc;
2779 }
2780 
2781 static void ept_access_test_teardown(void *unused)
2782 {
2783 	/* Exit the guest cleanly. */
2784 	do_ept_access_op(OP_EXIT);
2785 }
2786 
2787 static void ept_access_test_guest(void)
2788 {
2789 	struct ept_access_test_data *data = &ept_access_test_data;
2790 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2791 
2792 	while (true) {
2793 		switch (data->op) {
2794 		case OP_READ:
2795 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2796 			break;
2797 		case OP_WRITE:
2798 			*data->gva = MAGIC_VAL_2;
2799 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2800 			*data->gva = MAGIC_VAL_1;
2801 			break;
2802 		case OP_EXEC:
2803 			TEST_ASSERT_EQ(42, code());
2804 			break;
2805 		case OP_FLUSH_TLB:
2806 			write_cr3(read_cr3());
2807 			break;
2808 		case OP_EXIT:
2809 			return;
2810 		default:
2811 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2812 		}
2813 		vmcall();
2814 	}
2815 }
2816 
2817 static void ept_access_test_setup(void)
2818 {
2819 	struct ept_access_test_data *data = &ept_access_test_data;
2820 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2821 	unsigned long size = npages * PAGE_SIZE;
2822 	unsigned long *page_table = current_page_table();
2823 	unsigned long pte;
2824 
2825 	if (setup_ept(false))
2826 		test_skip("EPT not supported");
2827 
2828 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2829 	if (cpuid_maxphyaddr() < 40)
2830 		test_skip("Test needs MAXPHYADDR >= 40");
2831 
2832 	test_set_guest(ept_access_test_guest);
2833 	test_add_teardown(ept_access_test_teardown, NULL);
2834 
2835 	data->hva = get_1g_page();
2836 	TEST_ASSERT(data->hva);
2837 	data->hpa = virt_to_phys(data->hva);
2838 
2839 	data->gpa = 1ul << 39;
2840 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2841 				   size);
2842 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2843 	install_pages(page_table, data->gpa, size, data->gva);
2844 
2845 	/*
2846 	 * Make sure nothing's mapped here so the tests that screw with the
2847 	 * pml4 entry don't inadvertently break something.
2848 	 */
2849 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2850 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2851 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2852 
2853 	data->hva[0] = MAGIC_VAL_1;
2854 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2855 }
2856 
2857 static void ept_access_test_not_present(void)
2858 {
2859 	ept_access_test_setup();
2860 	/* --- */
2861 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2862 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2863 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2864 }
2865 
2866 static void ept_access_test_read_only(void)
2867 {
2868 	ept_access_test_setup();
2869 
2870 	/* r-- */
2871 	ept_access_allowed(EPT_RA, OP_READ);
2872 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2873 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2874 }
2875 
2876 static void ept_access_test_write_only(void)
2877 {
2878 	ept_access_test_setup();
2879 	/* -w- */
2880 	ept_access_misconfig(EPT_WA);
2881 }
2882 
2883 static void ept_access_test_read_write(void)
2884 {
2885 	ept_access_test_setup();
2886 	/* rw- */
2887 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2888 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2889 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2890 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2891 }
2892 
2893 
2894 static void ept_access_test_execute_only(void)
2895 {
2896 	ept_access_test_setup();
2897 	/* --x */
2898 	if (ept_execute_only_supported()) {
2899 		ept_access_violation(EPT_EA, OP_READ,
2900 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2901 		ept_access_violation(EPT_EA, OP_WRITE,
2902 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2903 		ept_access_allowed(EPT_EA, OP_EXEC);
2904 	} else {
2905 		ept_access_misconfig(EPT_EA);
2906 	}
2907 }
2908 
2909 static void ept_access_test_read_execute(void)
2910 {
2911 	ept_access_test_setup();
2912 	/* r-x */
2913 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2914 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2915 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2916 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2917 }
2918 
2919 static void ept_access_test_write_execute(void)
2920 {
2921 	ept_access_test_setup();
2922 	/* -wx */
2923 	ept_access_misconfig(EPT_WA | EPT_EA);
2924 }
2925 
2926 static void ept_access_test_read_write_execute(void)
2927 {
2928 	ept_access_test_setup();
2929 	/* rwx */
2930 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2931 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2932 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2933 }
2934 
2935 static void ept_access_test_reserved_bits(void)
2936 {
2937 	int i;
2938 	int maxphyaddr;
2939 
2940 	ept_access_test_setup();
2941 
2942 	/* Reserved bits above maxphyaddr. */
2943 	maxphyaddr = cpuid_maxphyaddr();
2944 	for (i = maxphyaddr; i <= 51; i++) {
2945 		report_prefix_pushf("reserved_bit=%d", i);
2946 		ept_reserved_bit(i);
2947 		report_prefix_pop();
2948 	}
2949 
2950 	/* Level-specific reserved bits. */
2951 	ept_reserved_bit_at_level_nohuge(2, 3);
2952 	ept_reserved_bit_at_level_nohuge(2, 4);
2953 	ept_reserved_bit_at_level_nohuge(2, 5);
2954 	ept_reserved_bit_at_level_nohuge(2, 6);
2955 	/* 2M alignment. */
2956 	for (i = 12; i < 20; i++) {
2957 		report_prefix_pushf("reserved_bit=%d", i);
2958 		ept_reserved_bit_at_level_huge(2, i);
2959 		report_prefix_pop();
2960 	}
2961 	ept_reserved_bit_at_level_nohuge(3, 3);
2962 	ept_reserved_bit_at_level_nohuge(3, 4);
2963 	ept_reserved_bit_at_level_nohuge(3, 5);
2964 	ept_reserved_bit_at_level_nohuge(3, 6);
2965 	/* 1G alignment. */
2966 	for (i = 12; i < 29; i++) {
2967 		report_prefix_pushf("reserved_bit=%d", i);
2968 		ept_reserved_bit_at_level_huge(3, i);
2969 		report_prefix_pop();
2970 	}
2971 	ept_reserved_bit_at_level(4, 3);
2972 	ept_reserved_bit_at_level(4, 4);
2973 	ept_reserved_bit_at_level(4, 5);
2974 	ept_reserved_bit_at_level(4, 6);
2975 	ept_reserved_bit_at_level(4, 7);
2976 }
2977 
2978 static void ept_access_test_ignored_bits(void)
2979 {
2980 	ept_access_test_setup();
2981 	/*
2982 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2983 	 * far as translation is concerned even if AD bits are enabled in the
2984 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2985 	 * control is 0.
2986 	 */
2987 	ept_ignored_bit(8);
2988 	ept_ignored_bit(9);
2989 	ept_ignored_bit(10);
2990 	ept_ignored_bit(11);
2991 	ept_ignored_bit(52);
2992 	ept_ignored_bit(53);
2993 	ept_ignored_bit(54);
2994 	ept_ignored_bit(55);
2995 	ept_ignored_bit(56);
2996 	ept_ignored_bit(57);
2997 	ept_ignored_bit(58);
2998 	ept_ignored_bit(59);
2999 	ept_ignored_bit(60);
3000 	ept_ignored_bit(61);
3001 	ept_ignored_bit(62);
3002 	ept_ignored_bit(63);
3003 }
3004 
3005 static void ept_access_test_paddr_not_present_ad_disabled(void)
3006 {
3007 	ept_access_test_setup();
3008 	ept_disable_ad_bits();
3009 
3010 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
3011 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
3012 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
3013 }
3014 
3015 static void ept_access_test_paddr_not_present_ad_enabled(void)
3016 {
3017 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
3018 
3019 	ept_access_test_setup();
3020 	ept_enable_ad_bits_or_skip_test();
3021 
3022 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
3023 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
3024 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
3025 }
3026 
3027 static void ept_access_test_paddr_read_only_ad_disabled(void)
3028 {
3029 	/*
3030 	 * When EPT AD bits are disabled, all accesses to guest paging
3031 	 * structures are reported separately as a read and (after
3032 	 * translation of the GPA to host physical address) a read+write
3033 	 * if the A/D bits have to be set.
3034 	 */
3035 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3036 
3037 	ept_access_test_setup();
3038 	ept_disable_ad_bits();
3039 
3040 	/* Can't update A bit, so all accesses fail. */
3041 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3042 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3043 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3044 	/* AD bits disabled, so only writes try to update the D bit. */
3045 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
3046 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3047 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
3048 	/* Both A and D already set, so read-only is OK. */
3049 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
3050 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
3051 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
3052 }
3053 
3054 static void ept_access_test_paddr_read_only_ad_enabled(void)
3055 {
3056 	/*
3057 	 * When EPT AD bits are enabled, all accesses to guest paging
3058 	 * structures are considered writes as far as EPT translation
3059 	 * is concerned.
3060 	 */
3061 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3062 
3063 	ept_access_test_setup();
3064 	ept_enable_ad_bits_or_skip_test();
3065 
3066 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3067 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3068 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3069 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
3070 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3071 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
3072 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
3073 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
3074 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
3075 }
3076 
3077 static void ept_access_test_paddr_read_write(void)
3078 {
3079 	ept_access_test_setup();
3080 	/* Read-write access to paging structure. */
3081 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
3082 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
3083 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
3084 }
3085 
3086 static void ept_access_test_paddr_read_write_execute(void)
3087 {
3088 	ept_access_test_setup();
3089 	/* RWX access to paging structure. */
3090 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
3091 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
3092 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
3093 }
3094 
3095 static void ept_access_test_paddr_read_execute_ad_disabled(void)
3096 {
3097   	/*
3098 	 * When EPT AD bits are disabled, all accesses to guest paging
3099 	 * structures are reported separately as a read and (after
3100 	 * translation of the GPA to host physical address) a read+write
3101 	 * if the A/D bits have to be set.
3102 	 */
3103 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3104 
3105 	ept_access_test_setup();
3106 	ept_disable_ad_bits();
3107 
3108 	/* Can't update A bit, so all accesses fail. */
3109 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3110 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3111 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3112 	/* AD bits disabled, so only writes try to update the D bit. */
3113 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
3114 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3115 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3116 	/* Both A and D already set, so read-only is OK. */
3117 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3118 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3119 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3120 }
3121 
3122 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3123 {
3124 	/*
3125 	 * When EPT AD bits are enabled, all accesses to guest paging
3126 	 * structures are considered writes as far as EPT translation
3127 	 * is concerned.
3128 	 */
3129 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3130 
3131 	ept_access_test_setup();
3132 	ept_enable_ad_bits_or_skip_test();
3133 
3134 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3135 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3136 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3137 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3138 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3139 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3140 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3141 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3142 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3143 }
3144 
3145 static void ept_access_test_paddr_not_present_page_fault(void)
3146 {
3147 	ept_access_test_setup();
3148 	/*
3149 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3150 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3151 	 * Thus guest page fault before host takes EPT violation for trying to
3152 	 * update A bit.
3153 	 */
3154 }
3155 
3156 static void ept_access_test_force_2m_page(void)
3157 {
3158 	ept_access_test_setup();
3159 
3160 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3161 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3162 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3163 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3164 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3165 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3166 }
3167 
3168 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3169 {
3170 	if (!is_invvpid_type_supported(type))
3171 		return false;
3172 
3173 	if (vpid >> 16)
3174 		return false;
3175 
3176 	if (type != INVVPID_ALL && !vpid)
3177 		return false;
3178 
3179 	if (type == INVVPID_ADDR && !is_canonical(gla))
3180 		return false;
3181 
3182 	return true;
3183 }
3184 
3185 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3186 {
3187 	int rc;
3188 	bool valid = invvpid_valid(type, vpid, gla);
3189 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3190 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3191 	/*
3192 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3193 	 * that we can tell if it is updated by INVVPID.
3194 	 */
3195 	vmcs_read(~0);
3196 	rc = __invvpid(type, vpid, gla);
3197 	report(!rc == valid, "INVVPID type %ld VPID %lx GLA %lx %s", type,
3198 	       vpid, gla,
3199 	       valid ? "passes" : "fails");
3200 	report(vmcs_read(VMX_INST_ERROR) == expected,
3201 	       "After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3202 	       rc ? "failed" : "successful",
3203 	       expected, vmcs_read(VMX_INST_ERROR));
3204 }
3205 
3206 static inline unsigned long get_first_supported_invvpid_type(void)
3207 {
3208 	u64 type = ffs(ept_vpid.val >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3209 
3210 	__TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3211 	return type;
3212 }
3213 
3214 static void ds_invvpid(void *data)
3215 {
3216 	asm volatile("invvpid %0, %1"
3217 		     :
3218 		     : "m"(*(struct invvpid_operand *)data),
3219 		       "r"(get_first_supported_invvpid_type()));
3220 }
3221 
3222 /*
3223  * The SS override is ignored in 64-bit mode, so we use an addressing
3224  * mode with %rsp as the base register to generate an implicit SS
3225  * reference.
3226  */
3227 static void ss_invvpid(void *data)
3228 {
3229 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3230 		     : "+r"(data)
3231 		     : "r"(get_first_supported_invvpid_type()));
3232 }
3233 
3234 static void invvpid_test_gp(void)
3235 {
3236 	bool fault;
3237 
3238 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3239 				   (void *)NONCANONICAL);
3240 	report(fault, "INVVPID with non-canonical DS operand raises #GP");
3241 }
3242 
3243 static void invvpid_test_ss(void)
3244 {
3245 	bool fault;
3246 
3247 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3248 				   (void *)NONCANONICAL);
3249 	report(fault, "INVVPID with non-canonical SS operand raises #SS");
3250 }
3251 
3252 static void invvpid_test_pf(void)
3253 {
3254 	void *vpage = alloc_vpage();
3255 	bool fault;
3256 
3257 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3258 	report(fault, "INVVPID with unmapped operand raises #PF");
3259 }
3260 
3261 static void try_compat_invvpid(void *unused)
3262 {
3263 	struct far_pointer32 fp = {
3264 		.offset = (uintptr_t)&&invvpid,
3265 		.selector = KERNEL_CS32,
3266 	};
3267 	uintptr_t rsp;
3268 
3269 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
3270 
3271 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3272 			"Code address too high.");
3273 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3274 
3275 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3276 	return;
3277 invvpid:
3278 	asm volatile (".code32;"
3279 		      "invvpid (%eax), %eax;"
3280 		      "lret;"
3281 		      ".code64");
3282 	__builtin_unreachable();
3283 }
3284 
3285 static void invvpid_test_compatibility_mode(void)
3286 {
3287 	bool fault;
3288 
3289 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3290 	report(fault, "Compatibility mode INVVPID raises #UD");
3291 }
3292 
3293 static void invvpid_test_not_in_vmx_operation(void)
3294 {
3295 	bool fault;
3296 
3297 	TEST_ASSERT(!vmx_off());
3298 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3299 	report(fault, "INVVPID outside of VMX operation raises #UD");
3300 	TEST_ASSERT(!vmx_on());
3301 }
3302 
3303 /*
3304  * This does not test real-address mode, virtual-8086 mode, protected mode,
3305  * or CPL > 0.
3306  */
3307 static void invvpid_test(void)
3308 {
3309 	int i;
3310 	unsigned types = 0;
3311 	unsigned type;
3312 
3313 	if (!is_vpid_supported())
3314 		test_skip("VPID not supported");
3315 
3316 	if (!is_invvpid_supported())
3317 		test_skip("INVVPID not supported.\n");
3318 
3319 	if (is_invvpid_type_supported(INVVPID_ADDR))
3320 		types |= 1u << INVVPID_ADDR;
3321 	if (is_invvpid_type_supported(INVVPID_CONTEXT_GLOBAL))
3322 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3323 	if (is_invvpid_type_supported(INVVPID_ALL))
3324 		types |= 1u << INVVPID_ALL;
3325 	if (is_invvpid_type_supported(INVVPID_CONTEXT_LOCAL))
3326 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3327 
3328 	if (!types)
3329 		test_skip("No INVVPID types supported.\n");
3330 
3331 	for (i = -127; i < 128; i++)
3332 		try_invvpid(i, 0xffff, 0);
3333 
3334 	/*
3335 	 * VPID must not be more than 16 bits.
3336 	 */
3337 	for (i = 0; i < 64; i++)
3338 		for (type = 0; type < 4; type++)
3339 			if (types & (1u << type))
3340 				try_invvpid(type, 1ul << i, 0);
3341 
3342 	/*
3343 	 * VPID must not be zero, except for "all contexts."
3344 	 */
3345 	for (type = 0; type < 4; type++)
3346 		if (types & (1u << type))
3347 			try_invvpid(type, 0, 0);
3348 
3349 	/*
3350 	 * The gla operand is only validated for single-address INVVPID.
3351 	 */
3352 	if (types & (1u << INVVPID_ADDR))
3353 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3354 
3355 	invvpid_test_gp();
3356 	invvpid_test_ss();
3357 	invvpid_test_pf();
3358 	invvpid_test_compatibility_mode();
3359 	invvpid_test_not_in_vmx_operation();
3360 }
3361 
3362 /*
3363  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3364  * at least as far as the guest-state checks. Returns false if the
3365  * VMLAUNCH fails early and execution falls through to the next
3366  * instruction.
3367  */
3368 static bool vmlaunch_succeeds(void)
3369 {
3370 	u32 exit_reason;
3371 
3372 	/*
3373 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3374 	 * unsupported VMCS component"). The caller can then check
3375 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3376 	 */
3377 	vmcs_write(~0u, 0);
3378 
3379 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3380 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3381 				   :
3382 				   : "r" ((u64)HOST_RSP)
3383 				   : "cc", "memory"
3384 				   : success);
3385 	return false;
3386 success:
3387 	exit_reason = vmcs_read(EXI_REASON);
3388 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3389 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3390 	return true;
3391 }
3392 
3393 /*
3394  * Try to launch the current VMCS.
3395  */
3396 static void test_vmx_vmlaunch(u32 xerror)
3397 {
3398 	bool success = vmlaunch_succeeds();
3399 	u32 vmx_inst_err;
3400 
3401 	report(success == !xerror, "vmlaunch %s",
3402 	       !xerror ? "succeeds" : "fails");
3403 	if (!success && xerror) {
3404 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3405 		report(vmx_inst_err == xerror,
3406 		       "VMX inst error is %d (actual %d)", xerror,
3407 		       vmx_inst_err);
3408 	}
3409 }
3410 
3411 /*
3412  * Try to launch the current VMCS, and expect one of two possible
3413  * errors (or success) codes.
3414  */
3415 static void test_vmx_vmlaunch2(u32 xerror1, u32 xerror2)
3416 {
3417 	bool success = vmlaunch_succeeds();
3418 	u32 vmx_inst_err;
3419 
3420 	if (!xerror1 == !xerror2)
3421 		report(success == !xerror1, "vmlaunch %s",
3422 		       !xerror1 ? "succeeds" : "fails");
3423 
3424 	if (!success && (xerror1 || xerror2)) {
3425 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3426 		report(vmx_inst_err == xerror1 || vmx_inst_err == xerror2,
3427 		       "VMX inst error is %d or %d (actual %d)", xerror1,
3428 		       xerror2, vmx_inst_err);
3429 	}
3430 }
3431 
3432 static void test_vmx_invalid_controls(void)
3433 {
3434 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3435 }
3436 
3437 static void test_vmx_valid_controls(void)
3438 {
3439 	test_vmx_vmlaunch(0);
3440 }
3441 
3442 /*
3443  * Test a particular value of a VM-execution control bit, if the value
3444  * is required or if the value is zero.
3445  */
3446 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3447 				    enum Encoding encoding, unsigned bit,
3448 				    unsigned val)
3449 {
3450 	u32 mask = 1u << bit;
3451 	bool expected;
3452 	u32 controls;
3453 
3454 	if (msr.set & mask)
3455 		TEST_ASSERT(msr.clr & mask);
3456 
3457 	/*
3458 	 * We can't arbitrarily turn on a control bit, because it may
3459 	 * introduce dependencies on other VMCS fields. So, we only
3460 	 * test turning on bits that have a required setting.
3461 	 */
3462 	if (val && (msr.clr & mask) && !(msr.set & mask))
3463 		return;
3464 
3465 	report_prefix_pushf("%s %s bit %d",
3466 			    val ? "Set" : "Clear", name, bit);
3467 
3468 	controls = vmcs_read(encoding);
3469 	if (val) {
3470 		vmcs_write(encoding, msr.set | mask);
3471 		expected = (msr.clr & mask);
3472 	} else {
3473 		vmcs_write(encoding, msr.set & ~mask);
3474 		expected = !(msr.set & mask);
3475 	}
3476 	if (expected)
3477 		test_vmx_valid_controls();
3478 	else
3479 		test_vmx_invalid_controls();
3480 	vmcs_write(encoding, controls);
3481 	report_prefix_pop();
3482 }
3483 
3484 /*
3485  * Test reserved values of a VM-execution control bit, based on the
3486  * allowed bit settings from the corresponding VMX capability MSR.
3487  */
3488 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3489 			      enum Encoding encoding, unsigned bit)
3490 {
3491 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3492 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3493 }
3494 
3495 /*
3496  * Reserved bits in the pin-based VM-execution controls must be set
3497  * properly. Software may consult the VMX capability MSRs to determine
3498  * the proper settings.
3499  * [Intel SDM]
3500  */
3501 static void test_pin_based_ctls(void)
3502 {
3503 	unsigned bit;
3504 
3505 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3506 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3507 	for (bit = 0; bit < 32; bit++)
3508 		test_rsvd_ctl_bit("pin-based controls",
3509 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3510 }
3511 
3512 /*
3513  * Reserved bits in the primary processor-based VM-execution controls
3514  * must be set properly. Software may consult the VMX capability MSRs
3515  * to determine the proper settings.
3516  * [Intel SDM]
3517  */
3518 static void test_primary_processor_based_ctls(void)
3519 {
3520 	unsigned bit;
3521 
3522 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3523 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3524 	for (bit = 0; bit < 32; bit++)
3525 		test_rsvd_ctl_bit("primary processor-based controls",
3526 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3527 }
3528 
3529 /*
3530  * If the "activate secondary controls" primary processor-based
3531  * VM-execution control is 1, reserved bits in the secondary
3532  * processor-based VM-execution controls must be cleared. Software may
3533  * consult the VMX capability MSRs to determine which bits are
3534  * reserved.
3535  * If the "activate secondary controls" primary processor-based
3536  * VM-execution control is 0 (or if the processor does not support the
3537  * 1-setting of that control), no checks are performed on the
3538  * secondary processor-based VM-execution controls.
3539  * [Intel SDM]
3540  */
3541 static void test_secondary_processor_based_ctls(void)
3542 {
3543 	u32 primary;
3544 	u32 secondary;
3545 	unsigned bit;
3546 
3547 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3548 		return;
3549 
3550 	primary = vmcs_read(CPU_EXEC_CTRL0);
3551 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3552 
3553 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3554 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3555 	for (bit = 0; bit < 32; bit++)
3556 		test_rsvd_ctl_bit("secondary processor-based controls",
3557 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3558 
3559 	/*
3560 	 * When the "activate secondary controls" VM-execution control
3561 	 * is clear, there are no checks on the secondary controls.
3562 	 */
3563 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3564 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3565 	report(vmlaunch_succeeds(),
3566 	       "Secondary processor-based controls ignored");
3567 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3568 	vmcs_write(CPU_EXEC_CTRL0, primary);
3569 }
3570 
3571 static void try_cr3_target_count(unsigned i, unsigned max)
3572 {
3573 	report_prefix_pushf("CR3 target count 0x%x", i);
3574 	vmcs_write(CR3_TARGET_COUNT, i);
3575 	if (i <= max)
3576 		test_vmx_valid_controls();
3577 	else
3578 		test_vmx_invalid_controls();
3579 	report_prefix_pop();
3580 }
3581 
3582 /*
3583  * The CR3-target count must not be greater than 4. Future processors
3584  * may support a different number of CR3-target values. Software
3585  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3586  * number of values supported.
3587  * [Intel SDM]
3588  */
3589 static void test_cr3_targets(void)
3590 {
3591 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3592 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3593 	unsigned i;
3594 
3595 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3596 	TEST_ASSERT(supported_targets <= 256);
3597 
3598 	try_cr3_target_count(-1u, supported_targets);
3599 	try_cr3_target_count(0x80000000, supported_targets);
3600 	try_cr3_target_count(0x7fffffff, supported_targets);
3601 	for (i = 0; i <= supported_targets + 1; i++)
3602 		try_cr3_target_count(i, supported_targets);
3603 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3604 
3605 	/* VMWRITE to nonexistent target fields should fail. */
3606 	for (i = supported_targets; i < 256; i++)
3607 		TEST_ASSERT(vmcs_write(CR3_TARGET_0 + i*2, 0));
3608 }
3609 
3610 /*
3611  * Test a particular address setting in the VMCS
3612  */
3613 static void test_vmcs_addr(const char *name,
3614 			   enum Encoding encoding,
3615 			   u64 align,
3616 			   bool ignored,
3617 			   bool skip_beyond_mapped_ram,
3618 			   u64 addr)
3619 {
3620 	report_prefix_pushf("%s = %lx", name, addr);
3621 	vmcs_write(encoding, addr);
3622 	if (skip_beyond_mapped_ram &&
3623 	    addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3624 	    addr < (1ul << cpuid_maxphyaddr()))
3625 		printf("Skipping physical address beyond mapped RAM\n");
3626 	else if (ignored || (IS_ALIGNED(addr, align) &&
3627 	    addr < (1ul << cpuid_maxphyaddr())))
3628 		test_vmx_valid_controls();
3629 	else
3630 		test_vmx_invalid_controls();
3631 	report_prefix_pop();
3632 }
3633 
3634 /*
3635  * Test interesting values for a VMCS address
3636  */
3637 static void test_vmcs_addr_values(const char *name,
3638 				  enum Encoding encoding,
3639 				  u64 align,
3640 				  bool ignored,
3641 				  bool skip_beyond_mapped_ram,
3642 				  u32 bit_start, u32 bit_end)
3643 {
3644 	unsigned i;
3645 	u64 orig_val = vmcs_read(encoding);
3646 
3647 	for (i = bit_start; i <= bit_end; i++)
3648 		test_vmcs_addr(name, encoding, align, ignored,
3649 			       skip_beyond_mapped_ram, 1ul << i);
3650 
3651 	test_vmcs_addr(name, encoding, align, ignored,
3652 		       skip_beyond_mapped_ram, PAGE_SIZE - 1);
3653 	test_vmcs_addr(name, encoding, align, ignored,
3654 		       skip_beyond_mapped_ram, PAGE_SIZE);
3655 	test_vmcs_addr(name, encoding, align, ignored,
3656 		       skip_beyond_mapped_ram,
3657 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3658 	test_vmcs_addr(name, encoding, align, ignored,
3659 		       skip_beyond_mapped_ram, -1ul);
3660 
3661 	vmcs_write(encoding, orig_val);
3662 }
3663 
3664 /*
3665  * Test a physical address reference in the VMCS, when the corresponding
3666  * feature is enabled and when the corresponding feature is disabled.
3667  */
3668 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3669 				     const char *field_name,
3670 				     const char *control_name, u64 align,
3671 				     bool skip_beyond_mapped_ram,
3672 				     bool control_primary)
3673 {
3674 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3675 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3676 	u64 page_addr;
3677 
3678 	if (control_primary) {
3679 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3680 			return;
3681 	} else {
3682 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3683 			return;
3684 	}
3685 
3686 	page_addr = vmcs_read(field);
3687 
3688 	report_prefix_pushf("%s enabled", control_name);
3689 	if (control_primary) {
3690 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3691 	} else {
3692 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3693 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3694 	}
3695 
3696 	test_vmcs_addr_values(field_name, field, align, false,
3697 			      skip_beyond_mapped_ram, 0, 63);
3698 	report_prefix_pop();
3699 
3700 	report_prefix_pushf("%s disabled", control_name);
3701 	if (control_primary) {
3702 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3703 	} else {
3704 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3705 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3706 	}
3707 
3708 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3709 	report_prefix_pop();
3710 
3711 	vmcs_write(field, page_addr);
3712 	vmcs_write(CPU_EXEC_CTRL0, primary);
3713 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3714 }
3715 
3716 /*
3717  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3718  * each I/O-bitmap address must be 0. Neither address should set any
3719  * bits beyond the processor's physical-address width.
3720  * [Intel SDM]
3721  */
3722 static void test_io_bitmaps(void)
3723 {
3724 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3725 				 "I/O bitmap A", "Use I/O bitmaps",
3726 				 PAGE_SIZE, false, true);
3727 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3728 				 "I/O bitmap B", "Use I/O bitmaps",
3729 				 PAGE_SIZE, false, true);
3730 }
3731 
3732 /*
3733  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3734  * the MSR-bitmap address must be 0. The address should not set any
3735  * bits beyond the processor's physical-address width.
3736  * [Intel SDM]
3737  */
3738 static void test_msr_bitmap(void)
3739 {
3740 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3741 				 "MSR bitmap", "Use MSR bitmaps",
3742 				 PAGE_SIZE, false, true);
3743 }
3744 
3745 /*
3746  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3747  * address must satisfy the following checks:
3748  * - Bits 11:0 of the address must be 0.
3749  * - The address should not set any bits beyond the processor's
3750  *   physical-address width.
3751  * [Intel SDM]
3752  */
3753 static void test_apic_virt_addr(void)
3754 {
3755 	/*
3756 	 * Ensure the processor will never use the virtual-APIC page, since
3757 	 * we will point it to invalid RAM.  Otherwise KVM is puzzled about
3758 	 * what we're trying to achieve and fails vmentry.
3759 	 */
3760 	u32 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
3761 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0 | CPU_CR8_LOAD | CPU_CR8_STORE);
3762 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3763 				 "virtual-APIC address", "Use TPR shadow",
3764 				 PAGE_SIZE, false, true);
3765 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
3766 }
3767 
3768 /*
3769  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3770  * APIC-access address must satisfy the following checks:
3771  *  - Bits 11:0 of the address must be 0.
3772  *  - The address should not set any bits beyond the processor's
3773  *    physical-address width.
3774  * [Intel SDM]
3775  */
3776 static void test_apic_access_addr(void)
3777 {
3778 	void *apic_access_page = alloc_page();
3779 
3780 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3781 
3782 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3783 				 "APIC-access address",
3784 				 "virtualize APIC-accesses", PAGE_SIZE,
3785 				 true, false);
3786 }
3787 
3788 static bool set_bit_pattern(u8 mask, u32 *secondary)
3789 {
3790 	u8 i;
3791 	bool flag = false;
3792 	u32 test_bits[3] = {
3793 		CPU_VIRT_X2APIC,
3794 		CPU_APIC_REG_VIRT,
3795 		CPU_VINTD
3796 	};
3797 
3798         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3799 		if ((mask & (1u << i)) &&
3800 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3801 			*secondary |= test_bits[i];
3802 			flag = true;
3803 		}
3804 	}
3805 
3806 	return (flag);
3807 }
3808 
3809 /*
3810  * If the "use TPR shadow" VM-execution control is 0, the following
3811  * VM-execution controls must also be 0:
3812  * 	- virtualize x2APIC mode
3813  *	- APIC-register virtualization
3814  *	- virtual-interrupt delivery
3815  *    [Intel SDM]
3816  *
3817  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3818  *    "virtualize APIC accesses" VM-execution control must be 0.
3819  *    [Intel SDM]
3820  */
3821 static void test_apic_virtual_ctls(void)
3822 {
3823 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3824 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3825 	u32 primary = saved_primary;
3826 	u32 secondary = saved_secondary;
3827 	bool is_ctrl_valid = false;
3828 	char str[10] = "disabled";
3829 	u8 i = 0, j;
3830 
3831 	/*
3832 	 * First test
3833 	 */
3834 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3835 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3836 		return;
3837 
3838 	primary |= CPU_SECONDARY;
3839 	primary &= ~CPU_TPR_SHADOW;
3840 	vmcs_write(CPU_EXEC_CTRL0, primary);
3841 
3842 	while (1) {
3843 		for (j = 1; j < 8; j++) {
3844 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3845 			if (primary & CPU_TPR_SHADOW) {
3846 				is_ctrl_valid = true;
3847 			} else {
3848 				if (! set_bit_pattern(j, &secondary))
3849 					is_ctrl_valid = true;
3850 				else
3851 					is_ctrl_valid = false;
3852 			}
3853 
3854 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3855 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3856 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3857 			if (is_ctrl_valid)
3858 				test_vmx_valid_controls();
3859 			else
3860 				test_vmx_invalid_controls();
3861 			report_prefix_pop();
3862 		}
3863 
3864 		if (i == 1)
3865 			break;
3866 		i++;
3867 
3868 		primary |= CPU_TPR_SHADOW;
3869 		vmcs_write(CPU_EXEC_CTRL0, primary);
3870 		strcpy(str, "enabled");
3871 	}
3872 
3873 	/*
3874 	 * Second test
3875 	 */
3876 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3877 
3878 	primary = saved_primary;
3879 	secondary = saved_secondary;
3880 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3881 		return;
3882 
3883 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3884 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3885 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3886 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3887 	test_vmx_valid_controls();
3888 	report_prefix_pop();
3889 
3890 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3891 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3892 	test_vmx_valid_controls();
3893 	report_prefix_pop();
3894 
3895 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3896 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3897 	test_vmx_invalid_controls();
3898 	report_prefix_pop();
3899 
3900 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3901 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3902 	test_vmx_valid_controls();
3903 	report_prefix_pop();
3904 
3905 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3906 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3907 }
3908 
3909 /*
3910  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3911  * "external-interrupt exiting" VM-execution control must be 1.
3912  * [Intel SDM]
3913  */
3914 static void test_virtual_intr_ctls(void)
3915 {
3916 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3917 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3918 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3919 	u32 primary = saved_primary;
3920 	u32 secondary = saved_secondary;
3921 	u32 pin = saved_pin;
3922 
3923 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3924 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
3925 		return;
3926 
3927 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3928 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
3929 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3930 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
3931 	test_vmx_valid_controls();
3932 	report_prefix_pop();
3933 
3934 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
3935 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3936 	test_vmx_invalid_controls();
3937 	report_prefix_pop();
3938 
3939 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
3940 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
3941 	test_vmx_valid_controls();
3942 	report_prefix_pop();
3943 
3944 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3945 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3946 	test_vmx_invalid_controls();
3947 	report_prefix_pop();
3948 
3949 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3950 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3951 	vmcs_write(PIN_CONTROLS, saved_pin);
3952 }
3953 
3954 static void test_pi_desc_addr(u64 addr, bool is_ctrl_valid)
3955 {
3956 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
3957 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
3958 	if (is_ctrl_valid)
3959 		test_vmx_valid_controls();
3960 	else
3961 		test_vmx_invalid_controls();
3962 	report_prefix_pop();
3963 }
3964 
3965 /*
3966  * If the "process posted interrupts" VM-execution control is 1, the
3967  * following must be true:
3968  *
3969  *	- The "virtual-interrupt delivery" VM-execution control is 1.
3970  *	- The "acknowledge interrupt on exit" VM-exit control is 1.
3971  *	- The posted-interrupt notification vector has a value in the
3972  *	- range 0 - 255 (bits 15:8 are all 0).
3973  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
3974  *	- The posted-interrupt descriptor address does not set any bits
3975  *	  beyond the processor's physical-address width.
3976  * [Intel SDM]
3977  */
3978 static void test_posted_intr(void)
3979 {
3980 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3981 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3982 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3983 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
3984 	u32 primary = saved_primary;
3985 	u32 secondary = saved_secondary;
3986 	u32 pin = saved_pin;
3987 	u32 exit_ctl = exit_ctl_saved;
3988 	u16 vec;
3989 	int i;
3990 
3991 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
3992 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3993 	    (ctrl_exit_rev.clr & EXI_INTA)))
3994 		return;
3995 
3996 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3997 
3998 	/*
3999 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
4000 	 */
4001 	pin |= PIN_POST_INTR;
4002 	vmcs_write(PIN_CONTROLS, pin);
4003 	secondary &= ~CPU_VINTD;
4004 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4005 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
4006 	test_vmx_invalid_controls();
4007 	report_prefix_pop();
4008 
4009 	secondary |= CPU_VINTD;
4010 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4011 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
4012 	test_vmx_invalid_controls();
4013 	report_prefix_pop();
4014 
4015 	exit_ctl &= ~EXI_INTA;
4016 	vmcs_write(EXI_CONTROLS, exit_ctl);
4017 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
4018 	test_vmx_invalid_controls();
4019 	report_prefix_pop();
4020 
4021 	exit_ctl |= EXI_INTA;
4022 	vmcs_write(EXI_CONTROLS, exit_ctl);
4023 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4024 	test_vmx_valid_controls();
4025 	report_prefix_pop();
4026 
4027 	secondary &= ~CPU_VINTD;
4028 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4029 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
4030 	test_vmx_invalid_controls();
4031 	report_prefix_pop();
4032 
4033 	secondary |= CPU_VINTD;
4034 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4035 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4036 	test_vmx_valid_controls();
4037 	report_prefix_pop();
4038 
4039 	/*
4040 	 * Test posted-interrupt notification vector
4041 	 */
4042 	for (i = 0; i < 8; i++) {
4043 		vec = (1ul << i);
4044 		vmcs_write(PINV, vec);
4045 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4046 		test_vmx_valid_controls();
4047 		report_prefix_pop();
4048 	}
4049 	for (i = 8; i < 16; i++) {
4050 		vec = (1ul << i);
4051 		vmcs_write(PINV, vec);
4052 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4053 		test_vmx_invalid_controls();
4054 		report_prefix_pop();
4055 	}
4056 
4057 	vec &= ~(0xff << 8);
4058 	vmcs_write(PINV, vec);
4059 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4060 	test_vmx_valid_controls();
4061 	report_prefix_pop();
4062 
4063 	/*
4064 	 * Test posted-interrupt descriptor address
4065 	 */
4066 	for (i = 0; i < 6; i++) {
4067 		test_pi_desc_addr(1ul << i, false);
4068 	}
4069 
4070 	test_pi_desc_addr(0xf0, false);
4071 	test_pi_desc_addr(0xff, false);
4072 	test_pi_desc_addr(0x0f, false);
4073 	test_pi_desc_addr(0x8000, true);
4074 	test_pi_desc_addr(0x00, true);
4075 	test_pi_desc_addr(0xc000, true);
4076 
4077 	test_vmcs_addr_values("process-posted interrupts",
4078 			       POSTED_INTR_DESC_ADDR, 64,
4079 			       false, false, 0, 63);
4080 
4081 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4082 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4083 	vmcs_write(PIN_CONTROLS, saved_pin);
4084 }
4085 
4086 static void test_apic_ctls(void)
4087 {
4088 	test_apic_virt_addr();
4089 	test_apic_access_addr();
4090 	test_apic_virtual_ctls();
4091 	test_virtual_intr_ctls();
4092 	test_posted_intr();
4093 }
4094 
4095 /*
4096  * If the "enable VPID" VM-execution control is 1, the value of the
4097  * of the VPID VM-execution control field must not be 0000H.
4098  * [Intel SDM]
4099  */
4100 static void test_vpid(void)
4101 {
4102 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4103 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4104 	u16 vpid = 0x0000;
4105 	int i;
4106 
4107 	if (!is_vpid_supported()) {
4108 		report_skip("%s : Secondary controls and/or VPID not supported", __func__);
4109 		return;
4110 	}
4111 
4112 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
4113 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
4114 	vmcs_write(VPID, vpid);
4115 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
4116 	test_vmx_valid_controls();
4117 	report_prefix_pop();
4118 
4119 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
4120 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4121 	test_vmx_invalid_controls();
4122 	report_prefix_pop();
4123 
4124 	for (i = 0; i < 16; i++) {
4125 		vpid = (short)1 << i;;
4126 		vmcs_write(VPID, vpid);
4127 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4128 		test_vmx_valid_controls();
4129 		report_prefix_pop();
4130 	}
4131 
4132 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4133 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4134 }
4135 
4136 static void set_vtpr(unsigned vtpr)
4137 {
4138 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
4139 }
4140 
4141 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
4142 {
4143 	bool valid = true;
4144 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4145 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4146 
4147 	if ((primary & CPU_TPR_SHADOW) &&
4148 	    (!(primary & CPU_SECONDARY) ||
4149 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
4150 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
4151 
4152 	set_vtpr(vtpr);
4153 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
4154 	    threshold, (vtpr >> 4) & 0xf);
4155 	if (valid)
4156 		test_vmx_valid_controls();
4157 	else
4158 		test_vmx_invalid_controls();
4159 	report_prefix_pop();
4160 }
4161 
4162 static void test_invalid_event_injection(void)
4163 {
4164 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
4165 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
4166 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
4167 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
4168 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
4169 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4170 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4171 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4172 	u32 cnt;
4173 
4174 	/* Setup */
4175 	report_prefix_push("invalid event injection");
4176 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4177 	vmcs_write(ENT_INST_LEN, 0x00000001);
4178 
4179 	/* The field's interruption type is not set to a reserved value. */
4180 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4181 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4182 			    "RESERVED interruption type invalid [-]",
4183 			    ent_intr_info);
4184 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4185 	test_vmx_invalid_controls();
4186 	report_prefix_pop();
4187 
4188 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4189 			DE_VECTOR;
4190 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4191 			    "RESERVED interruption type invalid [+]",
4192 			    ent_intr_info);
4193 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4194 	test_vmx_valid_controls();
4195 	report_prefix_pop();
4196 
4197 	/* If the interruption type is other event, the vector is 0. */
4198 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4199 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4200 			    "(OTHER EVENT && vector != 0) invalid [-]",
4201 			    ent_intr_info);
4202 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4203 	test_vmx_invalid_controls();
4204 	report_prefix_pop();
4205 
4206 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4207 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4208 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4209 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4210 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4211 	test_vmx_invalid_controls();
4212 	report_prefix_pop();
4213 
4214 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4215 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4216 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4217 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4218 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4219 	test_vmx_valid_controls();
4220 	report_prefix_pop();
4221 
4222 	/*
4223 	 * If the interruption type
4224 	 * is HW exception, the vector is at most 31.
4225 	 */
4226 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4227 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4228 			    "(HW exception && vector > 31) invalid [-]",
4229 			    ent_intr_info);
4230 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4231 	test_vmx_invalid_controls();
4232 	report_prefix_pop();
4233 
4234 	/*
4235 	 * deliver-error-code is 1 iff either
4236 	 * (a) the "unrestricted guest" VM-execution control is 0
4237 	 * (b) CR0.PE is set.
4238 	 */
4239 
4240 	/* Assert that unrestricted guest is disabled or unsupported */
4241 	assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
4242 	       !(secondary_save & CPU_URG));
4243 
4244 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4245 			GP_VECTOR;
4246 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4247 			    "error code <-> (!URG || prot_mode) [-]",
4248 			    ent_intr_info);
4249 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4250 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4251 	test_vmx_invalid_controls();
4252 	report_prefix_pop();
4253 
4254 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4255 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4256 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4257 			    "error code <-> (!URG || prot_mode) [+]",
4258 			    ent_intr_info);
4259 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4260 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4261 	test_vmx_valid_controls();
4262 	report_prefix_pop();
4263 
4264 	if (enable_unrestricted_guest(false))
4265 		goto skip_unrestricted_guest;
4266 
4267 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4268 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4269 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4270 			    "error code <-> (!URG || prot_mode) [-]",
4271 			    ent_intr_info);
4272 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4273 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4274 	test_vmx_invalid_controls();
4275 	report_prefix_pop();
4276 
4277 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4278 			GP_VECTOR;
4279 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4280 			    "error code <-> (!URG || prot_mode) [-]",
4281 			    ent_intr_info);
4282 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4283 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4284 	test_vmx_invalid_controls();
4285 	report_prefix_pop();
4286 
4287 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4288 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4289 
4290 skip_unrestricted_guest:
4291 	vmcs_write(GUEST_CR0, guest_cr0_save);
4292 
4293 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4294 	report_prefix_push("error code <-> HW exception");
4295 	for (cnt = 0; cnt < 8; cnt++) {
4296 		u32 exception_type_mask = cnt << 8;
4297 		u32 deliver_error_code_mask =
4298 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4299 			INTR_INFO_DELIVER_CODE_MASK : 0;
4300 
4301 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4302 				exception_type_mask | GP_VECTOR;
4303 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4304 				    ent_intr_info);
4305 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4306 		test_vmx_invalid_controls();
4307 		report_prefix_pop();
4308 	}
4309 	report_prefix_pop();
4310 
4311 	/*
4312 	 * deliver-error-code is 1 iff the the vector
4313 	 * indicates an exception that would normally deliver an error code
4314 	 */
4315 	report_prefix_push("error code <-> vector delivers error code");
4316 	for (cnt = 0; cnt < 32; cnt++) {
4317 		bool has_error_code = false;
4318 		u32 deliver_error_code_mask;
4319 
4320 		switch (cnt) {
4321 		case DF_VECTOR:
4322 		case TS_VECTOR:
4323 		case NP_VECTOR:
4324 		case SS_VECTOR:
4325 		case GP_VECTOR:
4326 		case PF_VECTOR:
4327 		case AC_VECTOR:
4328 			has_error_code = true;
4329 		case CP_VECTOR:
4330 			/* Some CPUs have error code and some do not, skip */
4331 			continue;
4332 		}
4333 
4334 		/* Negative case */
4335 		deliver_error_code_mask = has_error_code ?
4336 						0 :
4337 						INTR_INFO_DELIVER_CODE_MASK;
4338 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4339 				INTR_TYPE_HARD_EXCEPTION | cnt;
4340 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4341 				    ent_intr_info);
4342 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4343 		test_vmx_invalid_controls();
4344 		report_prefix_pop();
4345 
4346 		/* Positive case */
4347 		deliver_error_code_mask = has_error_code ?
4348 						INTR_INFO_DELIVER_CODE_MASK :
4349 						0;
4350 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4351 				INTR_TYPE_HARD_EXCEPTION | cnt;
4352 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4353 				    ent_intr_info);
4354 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4355 		test_vmx_valid_controls();
4356 		report_prefix_pop();
4357 	}
4358 	report_prefix_pop();
4359 
4360 	/* Reserved bits in the field (30:12) are 0. */
4361 	report_prefix_push("reserved bits clear");
4362 	for (cnt = 12; cnt <= 30; cnt++) {
4363 		ent_intr_info = ent_intr_info_base |
4364 				INTR_INFO_DELIVER_CODE_MASK |
4365 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4366 				(1U << cnt);
4367 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4368 				    ent_intr_info);
4369 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4370 		test_vmx_invalid_controls();
4371 		report_prefix_pop();
4372 	}
4373 	report_prefix_pop();
4374 
4375 	/*
4376 	 * If deliver-error-code is 1
4377 	 * bits 31:16 of the VM-entry exception error-code field are 0.
4378 	 */
4379 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4380 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4381 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4382 			    "VM-entry exception error code[31:16] clear",
4383 			    ent_intr_info);
4384 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4385 	for (cnt = 16; cnt <= 31; cnt++) {
4386 		ent_intr_err = 1U << cnt;
4387 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4388 				    ent_intr_err);
4389 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4390 		test_vmx_invalid_controls();
4391 		report_prefix_pop();
4392 	}
4393 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4394 	report_prefix_pop();
4395 
4396 	/*
4397 	 * If the interruption type is software interrupt, software exception,
4398 	 * or privileged software exception, the VM-entry instruction-length
4399 	 * field is in the range 0 - 15.
4400 	 */
4401 
4402 	for (cnt = 0; cnt < 3; cnt++) {
4403 		switch (cnt) {
4404 		case 0:
4405 			ent_intr_info = ent_intr_info_base |
4406 					INTR_TYPE_SOFT_INTR;
4407 			break;
4408 		case 1:
4409 			ent_intr_info = ent_intr_info_base |
4410 					INTR_TYPE_SOFT_EXCEPTION;
4411 			break;
4412 		case 2:
4413 			ent_intr_info = ent_intr_info_base |
4414 					INTR_TYPE_PRIV_SW_EXCEPTION;
4415 			break;
4416 		}
4417 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4418 				    "VM-entry instruction-length check",
4419 				    ent_intr_info);
4420 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4421 
4422 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4423 		ent_intr_len = -1;
4424 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4425 				    ent_intr_len);
4426 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4427 		test_vmx_invalid_controls();
4428 		report_prefix_pop();
4429 
4430 		/* Instruction length set to 16 should fail */
4431 		ent_intr_len = 0x00000010;
4432 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4433 				    ent_intr_len);
4434 		vmcs_write(ENT_INST_LEN, 0x00000010);
4435 		test_vmx_invalid_controls();
4436 		report_prefix_pop();
4437 
4438 		report_prefix_pop();
4439 	}
4440 
4441 	/* Cleanup */
4442 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4443 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4444 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4445 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4446 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4447 	vmcs_write(GUEST_CR0, guest_cr0_save);
4448 	report_prefix_pop();
4449 }
4450 
4451 /*
4452  * Test interesting vTPR values for a given TPR threshold.
4453  */
4454 static void test_vtpr_values(unsigned threshold)
4455 {
4456 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4457 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4458 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4459 }
4460 
4461 static void try_tpr_threshold(unsigned threshold)
4462 {
4463 	bool valid = true;
4464 
4465 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4466 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4467 
4468 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4469 	    (secondary & CPU_VINTD)))
4470 		valid = !(threshold >> 4);
4471 
4472 	set_vtpr(-1);
4473 	vmcs_write(TPR_THRESHOLD, threshold);
4474 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4475 	if (valid)
4476 		test_vmx_valid_controls();
4477 	else
4478 		test_vmx_invalid_controls();
4479 	report_prefix_pop();
4480 
4481 	if (valid)
4482 		test_vtpr_values(threshold);
4483 }
4484 
4485 /*
4486  * Test interesting TPR threshold values.
4487  */
4488 static void test_tpr_threshold_values(void)
4489 {
4490 	unsigned i;
4491 
4492 	for (i = 0; i < 0x10; i++)
4493 		try_tpr_threshold(i);
4494 	for (i = 4; i < 32; i++)
4495 		try_tpr_threshold(1u << i);
4496 	try_tpr_threshold(-1u);
4497 	try_tpr_threshold(0x7fffffff);
4498 }
4499 
4500 /*
4501  * This test covers the following two VM entry checks:
4502  *
4503  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4504  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4505  *         31:4 of the TPR threshold VM-execution control field must
4506 	   be 0.
4507  *         [Intel SDM]
4508  *
4509  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4510  *          "virtual-interrupt delivery" VM-execution control is 0
4511  *          and the "virtualize APIC accesses" VM-execution control
4512  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4513  *          control field must not be greater than the value of bits
4514  *          7:4 of VTPR.
4515  *          [Intel SDM]
4516  */
4517 static void test_tpr_threshold(void)
4518 {
4519 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4520 	u64 apic_virt_addr = vmcs_read(APIC_VIRT_ADDR);
4521 	u64 threshold = vmcs_read(TPR_THRESHOLD);
4522 	void *virtual_apic_page;
4523 
4524 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4525 		return;
4526 
4527 	virtual_apic_page = alloc_page();
4528 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4529 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4530 
4531 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4532 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4533 	test_tpr_threshold_values();
4534 	report_prefix_pop();
4535 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4536 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4537 	test_tpr_threshold_values();
4538 	report_prefix_pop();
4539 
4540 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4541 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES))))
4542 		goto out;
4543 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4544 
4545 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4546 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4547 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4548 		test_tpr_threshold_values();
4549 		report_prefix_pop();
4550 
4551 		vmcs_write(CPU_EXEC_CTRL0,
4552 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4553 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4554 		test_tpr_threshold_values();
4555 		report_prefix_pop();
4556 	}
4557 
4558 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4559 		vmcs_write(CPU_EXEC_CTRL0,
4560 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4561 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4562 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4563 		test_tpr_threshold_values();
4564 		report_prefix_pop();
4565 
4566 		vmcs_write(CPU_EXEC_CTRL0,
4567 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4568 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4569 		test_tpr_threshold_values();
4570 		report_prefix_pop();
4571 	}
4572 
4573 	if ((ctrl_cpu_rev[1].clr &
4574 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4575 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4576 		vmcs_write(CPU_EXEC_CTRL0,
4577 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4578 		vmcs_write(CPU_EXEC_CTRL1,
4579 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4580 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4581 		test_tpr_threshold_values();
4582 		report_prefix_pop();
4583 
4584 		vmcs_write(CPU_EXEC_CTRL0,
4585 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4586 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4587 		test_tpr_threshold_values();
4588 		report_prefix_pop();
4589 	}
4590 
4591 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4592 out:
4593 	vmcs_write(TPR_THRESHOLD, threshold);
4594 	vmcs_write(APIC_VIRT_ADDR, apic_virt_addr);
4595 	vmcs_write(CPU_EXEC_CTRL0, primary);
4596 }
4597 
4598 /*
4599  * This test verifies the following two vmentry checks:
4600  *
4601  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4602  *  VM-execution control must be 0.
4603  *  [Intel SDM]
4604  *
4605  *  If the "virtual NMIs" VM-execution control is 0, the "NMI-window
4606  *  exiting" VM-execution control must be 0.
4607  *  [Intel SDM]
4608  */
4609 static void test_nmi_ctrls(void)
4610 {
4611 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4612 
4613 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4614 	    (PIN_NMI | PIN_VIRT_NMI)) {
4615 		report_skip("%s : NMI exiting and/or Virtual NMIs not supported", __func__);
4616 		return;
4617 	}
4618 
4619 	/* Save the controls so that we can restore them after our tests */
4620 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4621 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4622 
4623 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4624 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4625 
4626 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4627 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4628 	test_vmx_valid_controls();
4629 	report_prefix_pop();
4630 
4631 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4632 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4633 	test_vmx_invalid_controls();
4634 	report_prefix_pop();
4635 
4636 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4637 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4638 	test_vmx_valid_controls();
4639 	report_prefix_pop();
4640 
4641 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4642 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4643 	test_vmx_valid_controls();
4644 	report_prefix_pop();
4645 
4646 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4647 		report_info("NMI-window exiting is not supported, skipping...");
4648 		goto done;
4649 	}
4650 
4651 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4652 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4653 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4654 	test_vmx_invalid_controls();
4655 	report_prefix_pop();
4656 
4657 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4658 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4659 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4660 	test_vmx_valid_controls();
4661 	report_prefix_pop();
4662 
4663 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4664 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4665 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4666 	test_vmx_valid_controls();
4667 	report_prefix_pop();
4668 
4669 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4670 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4671 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4672 	test_vmx_valid_controls();
4673 	report_prefix_pop();
4674 
4675 	/* Restore the controls to their original values */
4676 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4677 done:
4678 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4679 }
4680 
4681 static void test_eptp_ad_bit(u64 eptp, bool is_ctrl_valid)
4682 {
4683 	vmcs_write(EPTP, eptp);
4684 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4685 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4686 	if (is_ctrl_valid)
4687 		test_vmx_valid_controls();
4688 	else
4689 		test_vmx_invalid_controls();
4690 	report_prefix_pop();
4691 
4692 }
4693 
4694 /*
4695  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4696  *    control field must satisfy the following checks:
4697  *
4698  *     - The EPT memory type (bits 2:0) must be a value supported by the
4699  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4700  *     - Bits 5:3 (1 less than the EPT page-walk length) must indicate a
4701  *	 supported EPT page-walk length.
4702  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4703  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4704  *	 indicating that the processor does not support accessed and dirty
4705  *	 dirty flags for EPT.
4706  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4707  *	 physical-address width) must all be 0.
4708  *
4709  * 2. If the "unrestricted guest" VM-execution control is 1, the
4710  *    "enable EPT" VM-execution control must also be 1.
4711  */
4712 static void test_ept_eptp(void)
4713 {
4714 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4715 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4716 	u64 eptp_saved = vmcs_read(EPTP);
4717 	u32 primary = primary_saved;
4718 	u32 secondary = secondary_saved;
4719 	u64 eptp = eptp_saved;
4720 	u32 i, maxphysaddr;
4721 	u64 j, resv_bits_mask = 0;
4722 
4723 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4724 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4725 		report_skip("%s : \"CPU secondary\" and/or \"enable EPT\" exec control not supported", __func__);
4726 		return;
4727 	}
4728 
4729 	/* Support for 4-level EPT is mandatory. */
4730 	report(is_4_level_ept_supported(), "4-level EPT support check");
4731 
4732 	primary |= CPU_SECONDARY;
4733 	vmcs_write(CPU_EXEC_CTRL0, primary);
4734 	secondary |= CPU_EPT;
4735 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4736 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4737 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4738 	vmcs_write(EPTP, eptp);
4739 
4740 	for (i = 0; i < 8; i++) {
4741 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4742 		vmcs_write(EPTP, eptp);
4743 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4744 		    eptp & EPT_MEM_TYPE_MASK);
4745 		if (is_ept_memtype_supported(i))
4746 			test_vmx_valid_controls();
4747 		else
4748 			test_vmx_invalid_controls();
4749 		report_prefix_pop();
4750 	}
4751 
4752 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4753 
4754 	/*
4755 	 * Page walk length (bits 5:3).  Note, the value in VMCS.EPTP "is 1
4756 	 * less than the EPT page-walk length".
4757 	 */
4758 	for (i = 0; i < 8; i++) {
4759 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4760 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4761 
4762 		vmcs_write(EPTP, eptp);
4763 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4764 		    eptp & EPTP_PG_WALK_LEN_MASK);
4765 		if (i == 3 || (i == 4 && is_5_level_ept_supported()))
4766 			test_vmx_valid_controls();
4767 		else
4768 			test_vmx_invalid_controls();
4769 		report_prefix_pop();
4770 	}
4771 
4772 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4773 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4774 
4775 	/*
4776 	 * Accessed and dirty flag (bit 6)
4777 	 */
4778 	if (ept_ad_bits_supported()) {
4779 		report_info("Processor supports accessed and dirty flag");
4780 		eptp &= ~EPTP_AD_FLAG;
4781 		test_eptp_ad_bit(eptp, true);
4782 
4783 		eptp |= EPTP_AD_FLAG;
4784 		test_eptp_ad_bit(eptp, true);
4785 	} else {
4786 		report_info("Processor does not supports accessed and dirty flag");
4787 		eptp &= ~EPTP_AD_FLAG;
4788 		test_eptp_ad_bit(eptp, true);
4789 
4790 		eptp |= EPTP_AD_FLAG;
4791 		test_eptp_ad_bit(eptp, false);
4792 	}
4793 
4794 	/*
4795 	 * Reserved bits [11:7] and [63:N]
4796 	 */
4797 	for (i = 0; i < 32; i++) {
4798 		eptp = (eptp &
4799 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4800 		    (i << EPTP_RESERV_BITS_SHIFT);
4801 		vmcs_write(EPTP, eptp);
4802 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4803 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4804 		    EPTP_RESERV_BITS_MASK);
4805 		if (i == 0)
4806 			test_vmx_valid_controls();
4807 		else
4808 			test_vmx_invalid_controls();
4809 		report_prefix_pop();
4810 	}
4811 
4812 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4813 
4814 	maxphysaddr = cpuid_maxphyaddr();
4815 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4816 		resv_bits_mask |= 1ul << i;
4817 	}
4818 
4819 	for (j = maxphysaddr - 1; j <= 63; j++) {
4820 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4821 		    (j < maxphysaddr ? 0 : 1ul << j);
4822 		vmcs_write(EPTP, eptp);
4823 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4824 		    (eptp >> maxphysaddr) & resv_bits_mask);
4825 		if (j < maxphysaddr)
4826 			test_vmx_valid_controls();
4827 		else
4828 			test_vmx_invalid_controls();
4829 		report_prefix_pop();
4830 	}
4831 
4832 	secondary &= ~(CPU_EPT | CPU_URG);
4833 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4834 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4835 	test_vmx_valid_controls();
4836 	report_prefix_pop();
4837 
4838 	if (!(ctrl_cpu_rev[1].clr & CPU_URG))
4839 		goto skip_unrestricted_guest;
4840 
4841 	secondary |= CPU_URG;
4842 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4843 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4844 	test_vmx_invalid_controls();
4845 	report_prefix_pop();
4846 
4847 	secondary |= CPU_EPT;
4848 	setup_dummy_ept();
4849 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4850 	test_vmx_valid_controls();
4851 	report_prefix_pop();
4852 
4853 skip_unrestricted_guest:
4854 	secondary &= ~CPU_URG;
4855 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4856 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4857 	test_vmx_valid_controls();
4858 	report_prefix_pop();
4859 
4860 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4861 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4862 	vmcs_write(EPTP, eptp_saved);
4863 }
4864 
4865 /*
4866  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4867  * VM-execution control must also be 1. In addition, the PML address
4868  * must satisfy the following checks:
4869  *
4870  *    * Bits 11:0 of the address must be 0.
4871  *    * The address should not set any bits beyond the processor's
4872  *	physical-address width.
4873  *
4874  *  [Intel SDM]
4875  */
4876 static void test_pml(void)
4877 {
4878 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4879 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4880 	u32 primary = primary_saved;
4881 	u32 secondary = secondary_saved;
4882 
4883 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4884 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
4885 		report_skip("%s : \"Secondary execution\" or \"enable EPT\" or \"enable PML\" control not supported", __func__);
4886 		return;
4887 	}
4888 
4889 	primary |= CPU_SECONDARY;
4890 	vmcs_write(CPU_EXEC_CTRL0, primary);
4891 	secondary &= ~(CPU_PML | CPU_EPT);
4892 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4893 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
4894 	test_vmx_valid_controls();
4895 	report_prefix_pop();
4896 
4897 	secondary |= CPU_PML;
4898 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4899 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
4900 	test_vmx_invalid_controls();
4901 	report_prefix_pop();
4902 
4903 	secondary |= CPU_EPT;
4904 	setup_dummy_ept();
4905 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
4906 	test_vmx_valid_controls();
4907 	report_prefix_pop();
4908 
4909 	secondary &= ~CPU_PML;
4910 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4911 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
4912 	test_vmx_valid_controls();
4913 	report_prefix_pop();
4914 
4915 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
4916 				 PAGE_SIZE, false, false);
4917 
4918 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4919 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4920 }
4921 
4922  /*
4923  * If the "activate VMX-preemption timer" VM-execution control is 0, the
4924  * the "save VMX-preemption timer value" VM-exit control must also be 0.
4925  *
4926  *  [Intel SDM]
4927  */
4928 static void test_vmx_preemption_timer(void)
4929 {
4930 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4931 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
4932 	u32 pin = saved_pin;
4933 	u32 exit = saved_exit;
4934 
4935 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
4936 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
4937 		report_skip("%s : \"Save-VMX-preemption-timer\" and/or \"Enable-VMX-preemption-timer\" control not supported", __func__);
4938 		return;
4939 	}
4940 
4941 	pin |= PIN_PREEMPT;
4942 	vmcs_write(PIN_CONTROLS, pin);
4943 	exit &= ~EXI_SAVE_PREEMPT;
4944 	vmcs_write(EXI_CONTROLS, exit);
4945 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
4946 	test_vmx_valid_controls();
4947 	report_prefix_pop();
4948 
4949 	exit |= EXI_SAVE_PREEMPT;
4950 	vmcs_write(EXI_CONTROLS, exit);
4951 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
4952 	test_vmx_valid_controls();
4953 	report_prefix_pop();
4954 
4955 	pin &= ~PIN_PREEMPT;
4956 	vmcs_write(PIN_CONTROLS, pin);
4957 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
4958 	test_vmx_invalid_controls();
4959 	report_prefix_pop();
4960 
4961 	exit &= ~EXI_SAVE_PREEMPT;
4962 	vmcs_write(EXI_CONTROLS, exit);
4963 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
4964 	test_vmx_valid_controls();
4965 	report_prefix_pop();
4966 
4967 	vmcs_write(PIN_CONTROLS, saved_pin);
4968 	vmcs_write(EXI_CONTROLS, saved_exit);
4969 }
4970 
4971 extern unsigned char test_mtf1;
4972 extern unsigned char test_mtf2;
4973 extern unsigned char test_mtf3;
4974 extern unsigned char test_mtf4;
4975 
4976 static void test_mtf_guest(void)
4977 {
4978 	asm ("vmcall;\n\t"
4979 	     "out %al, $0x80;\n\t"
4980 	     "test_mtf1:\n\t"
4981 	     "vmcall;\n\t"
4982 	     "out %al, $0x80;\n\t"
4983 	     "test_mtf2:\n\t"
4984 	     /*
4985 	      * Prepare for the 'MOV CR3' test. Attempt to induce a
4986 	      * general-protection fault by moving a non-canonical address into
4987 	      * CR3. The 'MOV CR3' instruction does not take an imm64 operand,
4988 	      * so we must MOV the desired value into a register first.
4989 	      *
4990 	      * MOV RAX is done before the VMCALL such that MTF is only enabled
4991 	      * for the instruction under test.
4992 	      */
4993 	     "mov $0xaaaaaaaaaaaaaaaa, %rax;\n\t"
4994 	     "vmcall;\n\t"
4995 	     "mov %rax, %cr3;\n\t"
4996 	     "test_mtf3:\n\t"
4997 	     "vmcall;\n\t"
4998 	     /*
4999 	      * ICEBP/INT1 instruction. Though the instruction is now
5000 	      * documented, don't rely on assemblers enumerating the
5001 	      * instruction. Resort to hand assembly.
5002 	      */
5003 	     ".byte 0xf1;\n\t"
5004 	     "vmcall;\n\t"
5005 	     "test_mtf4:\n\t"
5006 	     "mov $0, %eax;\n\t");
5007 }
5008 
5009 static void test_mtf_gp_handler(struct ex_regs *regs)
5010 {
5011 	regs->rip = (unsigned long) &test_mtf3;
5012 }
5013 
5014 static void test_mtf_db_handler(struct ex_regs *regs)
5015 {
5016 }
5017 
5018 static void enable_mtf(void)
5019 {
5020 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5021 
5022 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 | CPU_MTF);
5023 }
5024 
5025 static void disable_mtf(void)
5026 {
5027 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5028 
5029 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 & ~CPU_MTF);
5030 }
5031 
5032 static void enable_tf(void)
5033 {
5034 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5035 
5036 	vmcs_write(GUEST_RFLAGS, rflags | X86_EFLAGS_TF);
5037 }
5038 
5039 static void disable_tf(void)
5040 {
5041 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5042 
5043 	vmcs_write(GUEST_RFLAGS, rflags & ~X86_EFLAGS_TF);
5044 }
5045 
5046 static void report_mtf(const char *insn_name, unsigned long exp_rip)
5047 {
5048 	unsigned long rip = vmcs_read(GUEST_RIP);
5049 
5050 	assert_exit_reason(VMX_MTF);
5051 	report(rip == exp_rip, "MTF VM-exit after %s. RIP: 0x%lx (expected 0x%lx)",
5052 	       insn_name, rip, exp_rip);
5053 }
5054 
5055 static void vmx_mtf_test(void)
5056 {
5057 	unsigned long pending_dbg;
5058 	handler old_gp, old_db;
5059 
5060 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5061 		report_skip("%s : \"Monitor trap flag\" exec control not supported", __func__);
5062 		return;
5063 	}
5064 
5065 	test_set_guest(test_mtf_guest);
5066 
5067 	/* Expect an MTF VM-exit after OUT instruction */
5068 	enter_guest();
5069 	skip_exit_vmcall();
5070 
5071 	enable_mtf();
5072 	enter_guest();
5073 	report_mtf("OUT", (unsigned long) &test_mtf1);
5074 	disable_mtf();
5075 
5076 	/*
5077 	 * Concurrent #DB trap and MTF on instruction boundary. Expect MTF
5078 	 * VM-exit with populated 'pending debug exceptions' VMCS field.
5079 	 */
5080 	enter_guest();
5081 	skip_exit_vmcall();
5082 
5083 	enable_mtf();
5084 	enable_tf();
5085 
5086 	enter_guest();
5087 	report_mtf("OUT", (unsigned long) &test_mtf2);
5088 	pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
5089 	report(pending_dbg & DR6_BS,
5090 	       "'pending debug exceptions' field after MTF VM-exit: 0x%lx (expected 0x%lx)",
5091 	       pending_dbg, (unsigned long) DR6_BS);
5092 
5093 	disable_mtf();
5094 	disable_tf();
5095 	vmcs_write(GUEST_PENDING_DEBUG, 0);
5096 
5097 	/*
5098 	 * #GP exception takes priority over MTF. Expect MTF VM-exit with RIP
5099 	 * advanced to first instruction of #GP handler.
5100 	 */
5101 	enter_guest();
5102 	skip_exit_vmcall();
5103 
5104 	old_gp = handle_exception(GP_VECTOR, test_mtf_gp_handler);
5105 
5106 	enable_mtf();
5107 	enter_guest();
5108 	report_mtf("MOV CR3", (unsigned long) get_idt_addr(&boot_idt[GP_VECTOR]));
5109 	disable_mtf();
5110 
5111 	/*
5112 	 * Concurrent MTF and privileged software exception (i.e. ICEBP/INT1).
5113 	 * MTF should follow the delivery of #DB trap, though the SDM doesn't
5114 	 * provide clear indication of the relative priority.
5115 	 */
5116 	enter_guest();
5117 	skip_exit_vmcall();
5118 
5119 	handle_exception(GP_VECTOR, old_gp);
5120 	old_db = handle_exception(DB_VECTOR, test_mtf_db_handler);
5121 
5122 	enable_mtf();
5123 	enter_guest();
5124 	report_mtf("INT1", (unsigned long) get_idt_addr(&boot_idt[DB_VECTOR]));
5125 	disable_mtf();
5126 
5127 	enter_guest();
5128 	skip_exit_vmcall();
5129 	handle_exception(DB_VECTOR, old_db);
5130 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK | INTR_TYPE_OTHER_EVENT);
5131 	enter_guest();
5132 	report_mtf("injected MTF", (unsigned long) &test_mtf4);
5133 	enter_guest();
5134 }
5135 
5136 extern char vmx_mtf_pdpte_guest_begin;
5137 extern char vmx_mtf_pdpte_guest_end;
5138 
5139 asm("vmx_mtf_pdpte_guest_begin:\n\t"
5140     "mov %cr0, %rax\n\t"    /* save CR0 with PG=1                 */
5141     "vmcall\n\t"            /* on return from this CR0.PG=0       */
5142     "mov %rax, %cr0\n\t"    /* restore CR0.PG=1 to enter PAE mode */
5143     "vmcall\n\t"
5144     "retq\n\t"
5145     "vmx_mtf_pdpte_guest_end:");
5146 
5147 static void vmx_mtf_pdpte_test(void)
5148 {
5149 	void *test_mtf_pdpte_guest;
5150 	pteval_t *pdpt;
5151 	u32 guest_ar_cs;
5152 	u64 guest_efer;
5153 	pteval_t *pte;
5154 	u64 guest_cr0;
5155 	u64 guest_cr3;
5156 	u64 guest_cr4;
5157 	u64 ent_ctls;
5158 	int i;
5159 
5160 	if (setup_ept(false))
5161 		return;
5162 
5163 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5164 		report_skip("%s : \"Monitor trap flag\" exec control not supported", __func__);
5165 		return;
5166 	}
5167 
5168 	if (!(ctrl_cpu_rev[1].clr & CPU_URG)) {
5169 		report_skip("%s : \"Unrestricted guest\" exec control not supported", __func__);
5170 		return;
5171 	}
5172 
5173 	vmcs_write(EXC_BITMAP, ~0);
5174 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
5175 
5176 	/*
5177 	 * Copy the guest code to an identity-mapped page.
5178 	 */
5179 	test_mtf_pdpte_guest = alloc_page();
5180 	memcpy(test_mtf_pdpte_guest, &vmx_mtf_pdpte_guest_begin,
5181 	       &vmx_mtf_pdpte_guest_end - &vmx_mtf_pdpte_guest_begin);
5182 
5183 	test_set_guest(test_mtf_pdpte_guest);
5184 
5185 	enter_guest();
5186 	skip_exit_vmcall();
5187 
5188 	/*
5189 	 * Put the guest in non-paged 32-bit protected mode, ready to enter
5190 	 * PAE mode when CR0.PG is set. CR4.PAE will already have been set
5191 	 * when the guest started out in long mode.
5192 	 */
5193 	ent_ctls = vmcs_read(ENT_CONTROLS);
5194 	vmcs_write(ENT_CONTROLS, ent_ctls & ~ENT_GUEST_64);
5195 
5196 	guest_efer = vmcs_read(GUEST_EFER);
5197 	vmcs_write(GUEST_EFER, guest_efer & ~(EFER_LMA | EFER_LME));
5198 
5199 	/*
5200 	 * Set CS access rights bits for 32-bit protected mode:
5201 	 * 3:0    B execute/read/accessed
5202 	 * 4      1 code or data
5203 	 * 6:5    0 descriptor privilege level
5204 	 * 7      1 present
5205 	 * 11:8   0 reserved
5206 	 * 12     0 available for use by system software
5207 	 * 13     0 64 bit mode not active
5208 	 * 14     1 default operation size 32-bit segment
5209 	 * 15     1 page granularity: segment limit in 4K units
5210 	 * 16     0 segment usable
5211 	 * 31:17  0 reserved
5212 	 */
5213 	guest_ar_cs = vmcs_read(GUEST_AR_CS);
5214 	vmcs_write(GUEST_AR_CS, 0xc09b);
5215 
5216 	guest_cr0 = vmcs_read(GUEST_CR0);
5217 	vmcs_write(GUEST_CR0, guest_cr0 & ~X86_CR0_PG);
5218 
5219 	guest_cr4 = vmcs_read(GUEST_CR4);
5220 	vmcs_write(GUEST_CR4, guest_cr4 & ~X86_CR4_PCIDE);
5221 
5222 	guest_cr3 = vmcs_read(GUEST_CR3);
5223 
5224 	/*
5225 	 * Turn the 4-level page table into a PAE page table by following the 0th
5226 	 * PML4 entry to a PDPT page, and grab the first four PDPTEs from that
5227 	 * page.
5228 	 *
5229 	 * Why does this work?
5230 	 *
5231 	 * PAE uses 32-bit addressing which implies:
5232 	 * Bits 11:0   page offset
5233 	 * Bits 20:12  entry into 512-entry page table
5234 	 * Bits 29:21  entry into a 512-entry directory table
5235 	 * Bits 31:30  entry into the page directory pointer table.
5236 	 * Bits 63:32  zero
5237 	 *
5238 	 * As only 2 bits are needed to select the PDPTEs for the entire
5239 	 * 32-bit address space, take the first 4 PDPTEs in the level 3 page
5240 	 * directory pointer table. It doesn't matter which of these PDPTEs
5241 	 * are present because they must cover the guest code given that it
5242 	 * has already run successfully.
5243 	 *
5244 	 * Get a pointer to PTE for GVA=0 in the page directory pointer table
5245 	 */
5246 	pte = get_pte_level(
5247             (pgd_t *)phys_to_virt(guest_cr3 & ~X86_CR3_PCID_MASK), 0,
5248             PDPT_LEVEL);
5249 
5250 	/*
5251 	 * Need some memory for the 4-entry PAE page directory pointer
5252 	 * table. Use the end of the identity-mapped page where the guest code
5253 	 * is stored. There is definitely space as the guest code is only a
5254 	 * few bytes.
5255 	 */
5256 	pdpt = test_mtf_pdpte_guest + PAGE_SIZE - 4 * sizeof(pteval_t);
5257 
5258 	/*
5259 	 * Copy the first four PDPTEs into the PAE page table with reserved
5260 	 * bits cleared. Note that permission bits from the PML4E and PDPTE
5261 	 * are not propagated.
5262 	 */
5263 	for (i = 0; i < 4; i++) {
5264 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_RSVD_MASK),
5265 				   "PDPTE has invalid reserved bits");
5266 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_PAGE_SIZE_MASK),
5267 				   "Cannot use 1GB super pages for PAE");
5268 		pdpt[i] = pte[i] & ~(PAE_PDPTE_RSVD_MASK);
5269 	}
5270 	vmcs_write(GUEST_CR3, virt_to_phys(pdpt));
5271 
5272 	enable_mtf();
5273 	enter_guest();
5274 	assert_exit_reason(VMX_MTF);
5275 	disable_mtf();
5276 
5277 	/*
5278 	 * The four PDPTEs should have been loaded into the VMCS when
5279 	 * the guest set CR0.PG to enter PAE mode.
5280 	 */
5281 	for (i = 0; i < 4; i++) {
5282 		u64 pdpte = vmcs_read(GUEST_PDPTE + 2 * i);
5283 
5284 		report(pdpte == pdpt[i], "PDPTE%d is 0x%lx (expected 0x%lx)",
5285 		       i, pdpte, pdpt[i]);
5286 	}
5287 
5288 	/*
5289 	 * Now, try to enter the guest in PAE mode. If the PDPTEs in the
5290 	 * vmcs are wrong, this will fail.
5291 	 */
5292 	enter_guest();
5293 	skip_exit_vmcall();
5294 
5295 	/*
5296 	 * Return guest to 64-bit mode and wrap up.
5297 	 */
5298 	vmcs_write(ENT_CONTROLS, ent_ctls);
5299 	vmcs_write(GUEST_EFER, guest_efer);
5300 	vmcs_write(GUEST_AR_CS, guest_ar_cs);
5301 	vmcs_write(GUEST_CR0, guest_cr0);
5302 	vmcs_write(GUEST_CR4, guest_cr4);
5303 	vmcs_write(GUEST_CR3, guest_cr3);
5304 
5305 	enter_guest();
5306 }
5307 
5308 /*
5309  * Tests for VM-execution control fields
5310  */
5311 static void test_vm_execution_ctls(void)
5312 {
5313 	test_pin_based_ctls();
5314 	test_primary_processor_based_ctls();
5315 	test_secondary_processor_based_ctls();
5316 	test_cr3_targets();
5317 	test_io_bitmaps();
5318 	test_msr_bitmap();
5319 	test_apic_ctls();
5320 	test_tpr_threshold();
5321 	test_nmi_ctrls();
5322 	test_pml();
5323 	test_vpid();
5324 	test_ept_eptp();
5325 	test_vmx_preemption_timer();
5326 }
5327 
5328  /*
5329   * The following checks are performed for the VM-entry MSR-load address if
5330   * the VM-entry MSR-load count field is non-zero:
5331   *
5332   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
5333   *      The address should not set any bits beyond the processor's
5334   *      physical-address width.
5335   *
5336   *    - The address of the last byte in the VM-entry MSR-load area
5337   *      should not set any bits beyond the processor's physical-address
5338   *      width. The address of this last byte is VM-entry MSR-load address
5339   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5340   *      uses more bits than the processor's physical-address width.)
5341   *
5342   *
5343   *  [Intel SDM]
5344   */
5345 static void test_entry_msr_load(void)
5346 {
5347 	entry_msr_load = alloc_page();
5348 	u64 tmp;
5349 	u32 entry_msr_ld_cnt = 1;
5350 	int i;
5351 	u32 addr_len = 64;
5352 
5353 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
5354 
5355 	/* Check first 4 bits of VM-entry MSR-load address */
5356 	for (i = 0; i < 4; i++) {
5357 		tmp = (u64)entry_msr_load | 1ull << i;
5358 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
5359 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
5360 				    tmp & 0xf);
5361 		test_vmx_invalid_controls();
5362 		report_prefix_pop();
5363 	}
5364 
5365 	if (basic.val & (1ul << 48))
5366 		addr_len = 32;
5367 
5368 	test_vmcs_addr_values("VM-entry-MSR-load address",
5369 				ENTER_MSR_LD_ADDR, 16, false, false,
5370 				4, addr_len - 1);
5371 
5372 	/*
5373 	 * Check last byte of VM-entry MSR-load address
5374 	 */
5375 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
5376 
5377 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5378 							i < 64; i++) {
5379 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
5380 			1ul << i;
5381 		vmcs_write(ENTER_MSR_LD_ADDR,
5382 			   tmp - (entry_msr_ld_cnt * 16 - 1));
5383 		test_vmx_invalid_controls();
5384 	}
5385 
5386 	vmcs_write(ENT_MSR_LD_CNT, 2);
5387 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5388 	test_vmx_invalid_controls();
5389 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5390 	test_vmx_valid_controls();
5391 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5392 	test_vmx_valid_controls();
5393 }
5394 
5395 static struct vmx_state_area_test_data {
5396 	u32 msr;
5397 	u64 exp;
5398 	bool enabled;
5399 } vmx_state_area_test_data;
5400 
5401 static void guest_state_test_main(void)
5402 {
5403 	u64 obs;
5404 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
5405 
5406 	while (1) {
5407 		if (vmx_get_test_stage() == 2)
5408 			break;
5409 
5410 		if (data->enabled) {
5411 			obs = rdmsr(data->msr);
5412 			report(data->exp == obs,
5413 			       "Guest state is 0x%lx (expected 0x%lx)",
5414 			       obs, data->exp);
5415 		}
5416 
5417 		vmcall();
5418 	}
5419 
5420 	asm volatile("fnop");
5421 }
5422 
5423 static void test_guest_state(const char *test, bool xfail, u64 field,
5424 			     const char * field_name)
5425 {
5426 	struct vmentry_result result;
5427 	u8 abort_flags;
5428 
5429 	abort_flags = ABORT_ON_EARLY_VMENTRY_FAIL;
5430 	if (!xfail)
5431 		abort_flags = ABORT_ON_INVALID_GUEST_STATE;
5432 
5433 	__enter_guest(abort_flags, &result);
5434 
5435 	report(result.exit_reason.failed_vmentry == xfail &&
5436 	       ((xfail && result.exit_reason.basic == VMX_FAIL_STATE) ||
5437 	        (!xfail && result.exit_reason.basic == VMX_VMCALL)) &&
5438 		(!xfail || vmcs_read(EXI_QUALIFICATION) == ENTRY_FAIL_DEFAULT),
5439 	        "%s, %s = %lx", test, field_name, field);
5440 
5441 	if (!result.exit_reason.failed_vmentry)
5442 		skip_exit_insn();
5443 }
5444 
5445 /*
5446  * Tests for VM-entry control fields
5447  */
5448 static void test_vm_entry_ctls(void)
5449 {
5450 	test_invalid_event_injection();
5451 	test_entry_msr_load();
5452 }
5453 
5454 /*
5455  * The following checks are performed for the VM-exit MSR-store address if
5456  * the VM-exit MSR-store count field is non-zero:
5457  *
5458  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
5459  *      The address should not set any bits beyond the processor's
5460  *      physical-address width.
5461  *
5462  *    - The address of the last byte in the VM-exit MSR-store area
5463  *      should not set any bits beyond the processor's physical-address
5464  *      width. The address of this last byte is VM-exit MSR-store address
5465  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5466  *      uses more bits than the processor's physical-address width.)
5467  *
5468  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
5469  * in the range 63:32.
5470  *
5471  *  [Intel SDM]
5472  */
5473 static void test_exit_msr_store(void)
5474 {
5475 	exit_msr_store = alloc_page();
5476 	u64 tmp;
5477 	u32 exit_msr_st_cnt = 1;
5478 	int i;
5479 	u32 addr_len = 64;
5480 
5481 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
5482 
5483 	/* Check first 4 bits of VM-exit MSR-store address */
5484 	for (i = 0; i < 4; i++) {
5485 		tmp = (u64)exit_msr_store | 1ull << i;
5486 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
5487 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
5488 				    tmp & 0xf);
5489 		test_vmx_invalid_controls();
5490 		report_prefix_pop();
5491 	}
5492 
5493 	if (basic.val & (1ul << 48))
5494 		addr_len = 32;
5495 
5496 	test_vmcs_addr_values("VM-exit-MSR-store address",
5497 				EXIT_MSR_ST_ADDR, 16, false, false,
5498 				4, addr_len - 1);
5499 
5500 	/*
5501 	 * Check last byte of VM-exit MSR-store address
5502 	 */
5503 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
5504 
5505 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5506 							i < 64; i++) {
5507 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
5508 			1ul << i;
5509 		vmcs_write(EXIT_MSR_ST_ADDR,
5510 			   tmp - (exit_msr_st_cnt * 16 - 1));
5511 		test_vmx_invalid_controls();
5512 	}
5513 
5514 	vmcs_write(EXI_MSR_ST_CNT, 2);
5515 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5516 	test_vmx_invalid_controls();
5517 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5518 	test_vmx_valid_controls();
5519 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5520 	test_vmx_valid_controls();
5521 }
5522 
5523 /*
5524  * Tests for VM-exit controls
5525  */
5526 static void test_vm_exit_ctls(void)
5527 {
5528 	test_exit_msr_store();
5529 }
5530 
5531 /*
5532  * Check that the virtual CPU checks all of the VMX controls as
5533  * documented in the Intel SDM.
5534  */
5535 static void vmx_controls_test(void)
5536 {
5537 	/*
5538 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
5539 	 * fail due to invalid guest state, should we make it that
5540 	 * far.
5541 	 */
5542 	vmcs_write(GUEST_RFLAGS, 0);
5543 
5544 	test_vm_execution_ctls();
5545 	test_vm_exit_ctls();
5546 	test_vm_entry_ctls();
5547 }
5548 
5549 struct apic_reg_virt_config {
5550 	bool apic_register_virtualization;
5551 	bool use_tpr_shadow;
5552 	bool virtualize_apic_accesses;
5553 	bool virtualize_x2apic_mode;
5554 	bool activate_secondary_controls;
5555 };
5556 
5557 struct apic_reg_test {
5558 	const char *name;
5559 	struct apic_reg_virt_config apic_reg_virt_config;
5560 };
5561 
5562 struct apic_reg_virt_expectation {
5563 	enum Reason rd_exit_reason;
5564 	enum Reason wr_exit_reason;
5565 	u32 val;
5566 	u32 (*virt_fn)(u32);
5567 
5568 	/*
5569 	 * If false, accessing the APIC access address from L2 is treated as a
5570 	 * normal memory operation, rather than triggering virtualization.
5571 	 */
5572 	bool virtualize_apic_accesses;
5573 };
5574 
5575 static u32 apic_virt_identity(u32 val)
5576 {
5577 	return val;
5578 }
5579 
5580 static u32 apic_virt_nibble1(u32 val)
5581 {
5582 	return val & 0xf0;
5583 }
5584 
5585 static u32 apic_virt_byte3(u32 val)
5586 {
5587 	return val & (0xff << 24);
5588 }
5589 
5590 static bool apic_reg_virt_exit_expectation(
5591 	u32 reg, struct apic_reg_virt_config *config,
5592 	struct apic_reg_virt_expectation *expectation)
5593 {
5594 	/* Good configs, where some L2 APIC accesses are virtualized. */
5595 	bool virtualize_apic_accesses_only =
5596 		config->virtualize_apic_accesses &&
5597 		!config->use_tpr_shadow &&
5598 		!config->apic_register_virtualization &&
5599 		!config->virtualize_x2apic_mode &&
5600 		config->activate_secondary_controls;
5601 	bool virtualize_apic_accesses_and_use_tpr_shadow =
5602 		config->virtualize_apic_accesses &&
5603 		config->use_tpr_shadow &&
5604 		!config->apic_register_virtualization &&
5605 		!config->virtualize_x2apic_mode &&
5606 		config->activate_secondary_controls;
5607 	bool apic_register_virtualization =
5608 		config->virtualize_apic_accesses &&
5609 		config->use_tpr_shadow &&
5610 		config->apic_register_virtualization &&
5611 		!config->virtualize_x2apic_mode &&
5612 		config->activate_secondary_controls;
5613 
5614 	expectation->val = MAGIC_VAL_1;
5615 	expectation->virt_fn = apic_virt_identity;
5616 	expectation->virtualize_apic_accesses =
5617 		config->virtualize_apic_accesses &&
5618 		config->activate_secondary_controls;
5619 	if (virtualize_apic_accesses_only) {
5620 		expectation->rd_exit_reason = VMX_APIC_ACCESS;
5621 		expectation->wr_exit_reason = VMX_APIC_ACCESS;
5622 	} else if (virtualize_apic_accesses_and_use_tpr_shadow) {
5623 		switch (reg) {
5624 		case APIC_TASKPRI:
5625 			expectation->rd_exit_reason = VMX_VMCALL;
5626 			expectation->wr_exit_reason = VMX_VMCALL;
5627 			expectation->virt_fn = apic_virt_nibble1;
5628 			break;
5629 		default:
5630 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5631 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5632 		}
5633 	} else if (apic_register_virtualization) {
5634 		expectation->rd_exit_reason = VMX_VMCALL;
5635 
5636 		switch (reg) {
5637 		case APIC_ID:
5638 		case APIC_EOI:
5639 		case APIC_LDR:
5640 		case APIC_DFR:
5641 		case APIC_SPIV:
5642 		case APIC_ESR:
5643 		case APIC_ICR:
5644 		case APIC_LVTT:
5645 		case APIC_LVTTHMR:
5646 		case APIC_LVTPC:
5647 		case APIC_LVT0:
5648 		case APIC_LVT1:
5649 		case APIC_LVTERR:
5650 		case APIC_TMICT:
5651 		case APIC_TDCR:
5652 			expectation->wr_exit_reason = VMX_APIC_WRITE;
5653 			break;
5654 		case APIC_LVR:
5655 		case APIC_ISR ... APIC_ISR + 0x70:
5656 		case APIC_TMR ... APIC_TMR + 0x70:
5657 		case APIC_IRR ... APIC_IRR + 0x70:
5658 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5659 			break;
5660 		case APIC_TASKPRI:
5661 			expectation->wr_exit_reason = VMX_VMCALL;
5662 			expectation->virt_fn = apic_virt_nibble1;
5663 			break;
5664 		case APIC_ICR2:
5665 			expectation->wr_exit_reason = VMX_VMCALL;
5666 			expectation->virt_fn = apic_virt_byte3;
5667 			break;
5668 		default:
5669 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5670 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5671 		}
5672 	} else if (!expectation->virtualize_apic_accesses) {
5673 		/*
5674 		 * No APIC registers are directly virtualized. This includes
5675 		 * VTPR, which can be virtualized through MOV to/from CR8 via
5676 		 * the use TPR shadow control, but not through directly
5677 		 * accessing VTPR.
5678 		 */
5679 		expectation->rd_exit_reason = VMX_VMCALL;
5680 		expectation->wr_exit_reason = VMX_VMCALL;
5681 	} else {
5682 		printf("Cannot parse APIC register virtualization config:\n"
5683 		       "\tvirtualize_apic_accesses: %d\n"
5684 		       "\tuse_tpr_shadow: %d\n"
5685 		       "\tapic_register_virtualization: %d\n"
5686 		       "\tvirtualize_x2apic_mode: %d\n"
5687 		       "\tactivate_secondary_controls: %d\n",
5688 		       config->virtualize_apic_accesses,
5689 		       config->use_tpr_shadow,
5690 		       config->apic_register_virtualization,
5691 		       config->virtualize_x2apic_mode,
5692 		       config->activate_secondary_controls);
5693 
5694 		return false;
5695 	}
5696 
5697 	return true;
5698 }
5699 
5700 struct apic_reg_test apic_reg_tests[] = {
5701 	/* Good configs, where some L2 APIC accesses are virtualized. */
5702 	{
5703 		.name = "Virtualize APIC accesses",
5704 		.apic_reg_virt_config = {
5705 			.virtualize_apic_accesses = true,
5706 			.use_tpr_shadow = false,
5707 			.apic_register_virtualization = false,
5708 			.virtualize_x2apic_mode = false,
5709 			.activate_secondary_controls = true,
5710 		},
5711 	},
5712 	{
5713 		.name = "Virtualize APIC accesses + Use TPR shadow",
5714 		.apic_reg_virt_config = {
5715 			.virtualize_apic_accesses = true,
5716 			.use_tpr_shadow = true,
5717 			.apic_register_virtualization = false,
5718 			.virtualize_x2apic_mode = false,
5719 			.activate_secondary_controls = true,
5720 		},
5721 	},
5722 	{
5723 		.name = "APIC-register virtualization",
5724 		.apic_reg_virt_config = {
5725 			.virtualize_apic_accesses = true,
5726 			.use_tpr_shadow = true,
5727 			.apic_register_virtualization = true,
5728 			.virtualize_x2apic_mode = false,
5729 			.activate_secondary_controls = true,
5730 		},
5731 	},
5732 
5733 	/*
5734 	 * Test that the secondary processor-based VM-execution controls are
5735 	 * correctly ignored when "activate secondary controls" is disabled.
5736 	 */
5737 	{
5738 		.name = "Activate secondary controls off",
5739 		.apic_reg_virt_config = {
5740 			.virtualize_apic_accesses = true,
5741 			.use_tpr_shadow = false,
5742 			.apic_register_virtualization = true,
5743 			.virtualize_x2apic_mode = true,
5744 			.activate_secondary_controls = false,
5745 		},
5746 	},
5747 	{
5748 		.name = "Activate secondary controls off + Use TPR shadow",
5749 		.apic_reg_virt_config = {
5750 			.virtualize_apic_accesses = true,
5751 			.use_tpr_shadow = true,
5752 			.apic_register_virtualization = true,
5753 			.virtualize_x2apic_mode = true,
5754 			.activate_secondary_controls = false,
5755 		},
5756 	},
5757 
5758 	/*
5759 	 * Test that the APIC access address is treated like an arbitrary memory
5760 	 * address when "virtualize APIC accesses" is disabled.
5761 	 */
5762 	{
5763 		.name = "Virtualize APIC accesses off + Use TPR shadow",
5764 		.apic_reg_virt_config = {
5765 			.virtualize_apic_accesses = false,
5766 			.use_tpr_shadow = true,
5767 			.apic_register_virtualization = true,
5768 			.virtualize_x2apic_mode = true,
5769 			.activate_secondary_controls = true,
5770 		},
5771 	},
5772 
5773 	/*
5774 	 * Test that VM entry fails due to invalid controls when
5775 	 * "APIC-register virtualization" is enabled while "use TPR shadow" is
5776 	 * disabled.
5777 	 */
5778 	{
5779 		.name = "APIC-register virtualization + Use TPR shadow off",
5780 		.apic_reg_virt_config = {
5781 			.virtualize_apic_accesses = true,
5782 			.use_tpr_shadow = false,
5783 			.apic_register_virtualization = true,
5784 			.virtualize_x2apic_mode = false,
5785 			.activate_secondary_controls = true,
5786 		},
5787 	},
5788 
5789 	/*
5790 	 * Test that VM entry fails due to invalid controls when
5791 	 * "Virtualize x2APIC mode" is enabled while "use TPR shadow" is
5792 	 * disabled.
5793 	 */
5794 	{
5795 		.name = "Virtualize x2APIC mode + Use TPR shadow off",
5796 		.apic_reg_virt_config = {
5797 			.virtualize_apic_accesses = false,
5798 			.use_tpr_shadow = false,
5799 			.apic_register_virtualization = false,
5800 			.virtualize_x2apic_mode = true,
5801 			.activate_secondary_controls = true,
5802 		},
5803 	},
5804 	{
5805 		.name = "Virtualize x2APIC mode + Use TPR shadow off v2",
5806 		.apic_reg_virt_config = {
5807 			.virtualize_apic_accesses = false,
5808 			.use_tpr_shadow = false,
5809 			.apic_register_virtualization = true,
5810 			.virtualize_x2apic_mode = true,
5811 			.activate_secondary_controls = true,
5812 		},
5813 	},
5814 
5815 	/*
5816 	 * Test that VM entry fails due to invalid controls when
5817 	 * "virtualize x2APIC mode" is enabled while "virtualize APIC accesses"
5818 	 * is enabled.
5819 	 */
5820 	{
5821 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses",
5822 		.apic_reg_virt_config = {
5823 			.virtualize_apic_accesses = true,
5824 			.use_tpr_shadow = true,
5825 			.apic_register_virtualization = false,
5826 			.virtualize_x2apic_mode = true,
5827 			.activate_secondary_controls = true,
5828 		},
5829 	},
5830 	{
5831 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses v2",
5832 		.apic_reg_virt_config = {
5833 			.virtualize_apic_accesses = true,
5834 			.use_tpr_shadow = true,
5835 			.apic_register_virtualization = true,
5836 			.virtualize_x2apic_mode = true,
5837 			.activate_secondary_controls = true,
5838 		},
5839 	},
5840 };
5841 
5842 enum Apic_op {
5843 	APIC_OP_XAPIC_RD,
5844 	APIC_OP_XAPIC_WR,
5845 	TERMINATE,
5846 };
5847 
5848 static u32 vmx_xapic_read(u32 *apic_access_address, u32 reg)
5849 {
5850 	return *(volatile u32 *)((uintptr_t)apic_access_address + reg);
5851 }
5852 
5853 static void vmx_xapic_write(u32 *apic_access_address, u32 reg, u32 val)
5854 {
5855 	*(volatile u32 *)((uintptr_t)apic_access_address + reg) = val;
5856 }
5857 
5858 struct apic_reg_virt_guest_args {
5859 	enum Apic_op op;
5860 	u32 *apic_access_address;
5861 	u32 reg;
5862 	u32 val;
5863 	bool check_rd;
5864 	u32 (*virt_fn)(u32);
5865 } apic_reg_virt_guest_args;
5866 
5867 static void apic_reg_virt_guest(void)
5868 {
5869 	volatile struct apic_reg_virt_guest_args *args =
5870 		&apic_reg_virt_guest_args;
5871 
5872 	for (;;) {
5873 		enum Apic_op op = args->op;
5874 		u32 *apic_access_address = args->apic_access_address;
5875 		u32 reg = args->reg;
5876 		u32 val = args->val;
5877 		bool check_rd = args->check_rd;
5878 		u32 (*virt_fn)(u32) = args->virt_fn;
5879 
5880 		if (op == TERMINATE)
5881 			break;
5882 
5883 		if (op == APIC_OP_XAPIC_RD) {
5884 			u32 ret = vmx_xapic_read(apic_access_address, reg);
5885 
5886 			if (check_rd) {
5887 				u32 want = virt_fn(val);
5888 				u32 got = virt_fn(ret);
5889 
5890 				report(got == want,
5891 				       "read 0x%x, expected 0x%x.", got, want);
5892 			}
5893 		} else if (op == APIC_OP_XAPIC_WR) {
5894 			vmx_xapic_write(apic_access_address, reg, val);
5895 		}
5896 
5897 		/*
5898 		 * The L1 should always execute a vmcall after it's done testing
5899 		 * an individual APIC operation. This helps to validate that the
5900 		 * L1 and L2 are in sync with each other, as expected.
5901 		 */
5902 		vmcall();
5903 	}
5904 }
5905 
5906 static void test_xapic_rd(
5907 	u32 reg, struct apic_reg_virt_expectation *expectation,
5908 	u32 *apic_access_address, u32 *virtual_apic_page)
5909 {
5910 	u32 val = expectation->val;
5911 	u32 exit_reason_want = expectation->rd_exit_reason;
5912 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5913 
5914 	report_prefix_pushf("xapic - reading 0x%03x", reg);
5915 
5916 	/* Configure guest to do an xapic read */
5917 	args->op = APIC_OP_XAPIC_RD;
5918 	args->apic_access_address = apic_access_address;
5919 	args->reg = reg;
5920 	args->val = val;
5921 	args->check_rd = exit_reason_want == VMX_VMCALL;
5922 	args->virt_fn = expectation->virt_fn;
5923 
5924 	/* Setup virtual APIC page */
5925 	if (!expectation->virtualize_apic_accesses) {
5926 		apic_access_address[apic_reg_index(reg)] = val;
5927 		virtual_apic_page[apic_reg_index(reg)] = 0;
5928 	} else if (exit_reason_want == VMX_VMCALL) {
5929 		apic_access_address[apic_reg_index(reg)] = 0;
5930 		virtual_apic_page[apic_reg_index(reg)] = val;
5931 	}
5932 
5933 	/* Enter guest */
5934 	enter_guest();
5935 
5936 	/*
5937 	 * Validate the behavior and
5938 	 * pass a magic value back to the guest.
5939 	 */
5940 	if (exit_reason_want == VMX_APIC_ACCESS) {
5941 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5942 
5943 		assert_exit_reason(exit_reason_want);
5944 		report(apic_page_offset == reg,
5945 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
5946 		       apic_page_offset, reg);
5947 		skip_exit_insn();
5948 
5949 		/* Reenter guest so it can consume/check rcx and exit again. */
5950 		enter_guest();
5951 	} else if (exit_reason_want != VMX_VMCALL) {
5952 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
5953 	}
5954 
5955 	skip_exit_vmcall();
5956 	report_prefix_pop();
5957 }
5958 
5959 static void test_xapic_wr(
5960 	u32 reg, struct apic_reg_virt_expectation *expectation,
5961 	u32 *apic_access_address, u32 *virtual_apic_page)
5962 {
5963 	u32 val = expectation->val;
5964 	u32 exit_reason_want = expectation->wr_exit_reason;
5965 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5966 	bool virtualized =
5967 		expectation->virtualize_apic_accesses &&
5968 		(exit_reason_want == VMX_APIC_WRITE ||
5969 		 exit_reason_want == VMX_VMCALL);
5970 	bool checked = false;
5971 
5972 	report_prefix_pushf("xapic - writing 0x%x to 0x%03x", val, reg);
5973 
5974 	/* Configure guest to do an xapic read */
5975 	args->op = APIC_OP_XAPIC_WR;
5976 	args->apic_access_address = apic_access_address;
5977 	args->reg = reg;
5978 	args->val = val;
5979 
5980 	/* Setup virtual APIC page */
5981 	if (virtualized || !expectation->virtualize_apic_accesses) {
5982 		apic_access_address[apic_reg_index(reg)] = 0;
5983 		virtual_apic_page[apic_reg_index(reg)] = 0;
5984 	}
5985 
5986 	/* Enter guest */
5987 	enter_guest();
5988 
5989 	/*
5990 	 * Validate the behavior and
5991 	 * pass a magic value back to the guest.
5992 	 */
5993 	if (exit_reason_want == VMX_APIC_ACCESS) {
5994 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5995 
5996 		assert_exit_reason(exit_reason_want);
5997 		report(apic_page_offset == reg,
5998 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
5999 		       apic_page_offset, reg);
6000 		skip_exit_insn();
6001 
6002 		/* Reenter guest so it can consume/check rcx and exit again. */
6003 		enter_guest();
6004 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6005 		assert_exit_reason(exit_reason_want);
6006 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6007 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%x",
6008 		       apic_reg_index(reg),
6009 		       virtual_apic_page[apic_reg_index(reg)], val);
6010 		checked = true;
6011 
6012 		/* Reenter guest so it can consume/check rcx and exit again. */
6013 		enter_guest();
6014 	} else if (exit_reason_want != VMX_VMCALL) {
6015 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6016 	}
6017 
6018 	assert_exit_reason(VMX_VMCALL);
6019 	if (virtualized && !checked) {
6020 		u32 want = expectation->virt_fn(val);
6021 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6022 		got = expectation->virt_fn(got);
6023 
6024 		report(got == want, "exitless write; val is 0x%x, want 0x%x",
6025 		       got, want);
6026 	} else if (!expectation->virtualize_apic_accesses && !checked) {
6027 		u32 got = apic_access_address[apic_reg_index(reg)];
6028 
6029 		report(got == val,
6030 		       "non-virtualized write; val is 0x%x, want 0x%x", got,
6031 		       val);
6032 	} else if (!expectation->virtualize_apic_accesses && checked) {
6033 		report_fail("Non-virtualized write was prematurely checked!");
6034 	}
6035 
6036 	skip_exit_vmcall();
6037 	report_prefix_pop();
6038 }
6039 
6040 enum Config_type {
6041 	CONFIG_TYPE_GOOD,
6042 	CONFIG_TYPE_UNSUPPORTED,
6043 	CONFIG_TYPE_VMENTRY_FAILS_EARLY,
6044 };
6045 
6046 static enum Config_type configure_apic_reg_virt_test(
6047 	struct apic_reg_virt_config *apic_reg_virt_config)
6048 {
6049 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6050 	u32 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6051 	/* Configs where L2 entry fails early, due to invalid controls. */
6052 	bool use_tpr_shadow_incorrectly_off =
6053 		!apic_reg_virt_config->use_tpr_shadow &&
6054 		(apic_reg_virt_config->apic_register_virtualization ||
6055 		 apic_reg_virt_config->virtualize_x2apic_mode) &&
6056 		apic_reg_virt_config->activate_secondary_controls;
6057 	bool virtualize_apic_accesses_incorrectly_on =
6058 		apic_reg_virt_config->virtualize_apic_accesses &&
6059 		apic_reg_virt_config->virtualize_x2apic_mode &&
6060 		apic_reg_virt_config->activate_secondary_controls;
6061 	bool vmentry_fails_early =
6062 		use_tpr_shadow_incorrectly_off ||
6063 		virtualize_apic_accesses_incorrectly_on;
6064 
6065 	if (apic_reg_virt_config->activate_secondary_controls) {
6066 		if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
6067 			printf("VM-execution control \"activate secondary controls\" NOT supported.\n");
6068 			return CONFIG_TYPE_UNSUPPORTED;
6069 		}
6070 		cpu_exec_ctrl0 |= CPU_SECONDARY;
6071 	} else {
6072 		cpu_exec_ctrl0 &= ~CPU_SECONDARY;
6073 	}
6074 
6075 	if (apic_reg_virt_config->virtualize_apic_accesses) {
6076 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES)) {
6077 			printf("VM-execution control \"virtualize APIC accesses\" NOT supported.\n");
6078 			return CONFIG_TYPE_UNSUPPORTED;
6079 		}
6080 		cpu_exec_ctrl1 |= CPU_VIRT_APIC_ACCESSES;
6081 	} else {
6082 		cpu_exec_ctrl1 &= ~CPU_VIRT_APIC_ACCESSES;
6083 	}
6084 
6085 	if (apic_reg_virt_config->use_tpr_shadow) {
6086 		if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6087 			printf("VM-execution control \"use TPR shadow\" NOT supported.\n");
6088 			return CONFIG_TYPE_UNSUPPORTED;
6089 		}
6090 		cpu_exec_ctrl0 |= CPU_TPR_SHADOW;
6091 	} else {
6092 		cpu_exec_ctrl0 &= ~CPU_TPR_SHADOW;
6093 	}
6094 
6095 	if (apic_reg_virt_config->apic_register_virtualization) {
6096 		if (!(ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT)) {
6097 			printf("VM-execution control \"APIC-register virtualization\" NOT supported.\n");
6098 			return CONFIG_TYPE_UNSUPPORTED;
6099 		}
6100 		cpu_exec_ctrl1 |= CPU_APIC_REG_VIRT;
6101 	} else {
6102 		cpu_exec_ctrl1 &= ~CPU_APIC_REG_VIRT;
6103 	}
6104 
6105 	if (apic_reg_virt_config->virtualize_x2apic_mode) {
6106 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_X2APIC)) {
6107 			printf("VM-execution control \"virtualize x2APIC mode\" NOT supported.\n");
6108 			return CONFIG_TYPE_UNSUPPORTED;
6109 		}
6110 		cpu_exec_ctrl1 |= CPU_VIRT_X2APIC;
6111 	} else {
6112 		cpu_exec_ctrl1 &= ~CPU_VIRT_X2APIC;
6113 	}
6114 
6115 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6116 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6117 
6118 	if (vmentry_fails_early)
6119 		return CONFIG_TYPE_VMENTRY_FAILS_EARLY;
6120 
6121 	return CONFIG_TYPE_GOOD;
6122 }
6123 
6124 static bool cpu_has_apicv(void)
6125 {
6126 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
6127 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
6128 		(ctrl_pin_rev.clr & PIN_POST_INTR));
6129 }
6130 
6131 /* Validates APIC register access across valid virtualization configurations. */
6132 static void apic_reg_virt_test(void)
6133 {
6134 	u32 *apic_access_address;
6135 	u32 *virtual_apic_page;
6136 	u64 control;
6137 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6138 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6139 	int i;
6140 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
6141 
6142 	if (!cpu_has_apicv()) {
6143 		report_skip("%s : Not all required APICv bits supported", __func__);
6144 		return;
6145 	}
6146 
6147 	control = cpu_exec_ctrl1;
6148 	control &= ~CPU_VINTD;
6149 	vmcs_write(CPU_EXEC_CTRL1, control);
6150 
6151 	test_set_guest(apic_reg_virt_guest);
6152 
6153 	/*
6154 	 * From the SDM: The 1-setting of the "virtualize APIC accesses"
6155 	 * VM-execution is guaranteed to apply only if translations to the
6156 	 * APIC-access address use a 4-KByte page.
6157 	 */
6158 	apic_access_address = alloc_page();
6159 	force_4k_page(apic_access_address);
6160 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_address));
6161 
6162 	virtual_apic_page = alloc_page();
6163 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6164 
6165 	for (i = 0; i < ARRAY_SIZE(apic_reg_tests); i++) {
6166 		struct apic_reg_test *apic_reg_test = &apic_reg_tests[i];
6167 		struct apic_reg_virt_config *apic_reg_virt_config =
6168 				&apic_reg_test->apic_reg_virt_config;
6169 		enum Config_type config_type;
6170 		u32 reg;
6171 
6172 		printf("--- %s test ---\n", apic_reg_test->name);
6173 		config_type =
6174 			configure_apic_reg_virt_test(apic_reg_virt_config);
6175 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6176 			printf("Skip because of missing features.\n");
6177 			continue;
6178 		}
6179 
6180 		if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6181 			enter_guest_with_bad_controls();
6182 			continue;
6183 		}
6184 
6185 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6186 			struct apic_reg_virt_expectation expectation = {};
6187 			bool ok;
6188 
6189 			ok = apic_reg_virt_exit_expectation(
6190 				reg, apic_reg_virt_config, &expectation);
6191 			if (!ok) {
6192 				report_fail("Malformed test.");
6193 				break;
6194 			}
6195 
6196 			test_xapic_rd(reg, &expectation, apic_access_address,
6197 				      virtual_apic_page);
6198 			test_xapic_wr(reg, &expectation, apic_access_address,
6199 				      virtual_apic_page);
6200 		}
6201 	}
6202 
6203 	/* Terminate the guest */
6204 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6205 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6206 	args->op = TERMINATE;
6207 	enter_guest();
6208 	assert_exit_reason(VMX_VMCALL);
6209 }
6210 
6211 struct virt_x2apic_mode_config {
6212 	struct apic_reg_virt_config apic_reg_virt_config;
6213 	bool virtual_interrupt_delivery;
6214 	bool use_msr_bitmaps;
6215 	bool disable_x2apic_msr_intercepts;
6216 	bool disable_x2apic;
6217 };
6218 
6219 struct virt_x2apic_mode_test_case {
6220 	const char *name;
6221 	struct virt_x2apic_mode_config virt_x2apic_mode_config;
6222 };
6223 
6224 enum Virt_x2apic_mode_behavior_type {
6225 	X2APIC_ACCESS_VIRTUALIZED,
6226 	X2APIC_ACCESS_PASSED_THROUGH,
6227 	X2APIC_ACCESS_TRIGGERS_GP,
6228 };
6229 
6230 struct virt_x2apic_mode_expectation {
6231 	enum Reason rd_exit_reason;
6232 	enum Reason wr_exit_reason;
6233 
6234 	/*
6235 	 * RDMSR and WRMSR handle 64-bit values. However, except for ICR, all of
6236 	 * the x2APIC registers are 32 bits. Notice:
6237 	 *   1. vmx_x2apic_read() clears the upper 32 bits for 32-bit registers.
6238 	 *   2. vmx_x2apic_write() expects the val arg to be well-formed.
6239 	 */
6240 	u64 rd_val;
6241 	u64 wr_val;
6242 
6243 	/*
6244 	 * Compares input to virtualized output;
6245 	 * 1st arg is pointer to return expected virtualization output.
6246 	 */
6247 	u64 (*virt_fn)(u64);
6248 
6249 	enum Virt_x2apic_mode_behavior_type rd_behavior;
6250 	enum Virt_x2apic_mode_behavior_type wr_behavior;
6251 	bool wr_only;
6252 };
6253 
6254 static u64 virt_x2apic_mode_identity(u64 val)
6255 {
6256 	return val;
6257 }
6258 
6259 static u64 virt_x2apic_mode_nibble1(u64 val)
6260 {
6261 	return val & 0xf0;
6262 }
6263 
6264 static void virt_x2apic_mode_rd_expectation(
6265 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6266 	bool apic_register_virtualization, bool virtual_interrupt_delivery,
6267 	struct virt_x2apic_mode_expectation *expectation)
6268 {
6269 	bool readable =
6270 		!x2apic_reg_reserved(reg) &&
6271 		reg != APIC_EOI;
6272 
6273 	expectation->rd_exit_reason = VMX_VMCALL;
6274 	expectation->virt_fn = virt_x2apic_mode_identity;
6275 	if (virt_x2apic_mode_on && apic_register_virtualization) {
6276 		expectation->rd_val = MAGIC_VAL_1;
6277 		if (reg == APIC_PROCPRI && virtual_interrupt_delivery)
6278 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6279 		else if (reg == APIC_TASKPRI)
6280 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6281 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6282 	} else if (virt_x2apic_mode_on && !apic_register_virtualization &&
6283 		   reg == APIC_TASKPRI) {
6284 		expectation->rd_val = MAGIC_VAL_1;
6285 		expectation->virt_fn = virt_x2apic_mode_nibble1;
6286 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6287 	} else if (!disable_x2apic && readable) {
6288 		expectation->rd_val = apic_read(reg);
6289 		expectation->rd_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6290 	} else {
6291 		expectation->rd_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6292 	}
6293 }
6294 
6295 /*
6296  * get_x2apic_wr_val() creates an innocuous write value for an x2APIC register.
6297  *
6298  * For writable registers, get_x2apic_wr_val() deposits the write value into the
6299  * val pointer arg and returns true. For non-writable registers, val is not
6300  * modified and get_x2apic_wr_val() returns false.
6301  */
6302 static bool get_x2apic_wr_val(u32 reg, u64 *val)
6303 {
6304 	switch (reg) {
6305 	case APIC_TASKPRI:
6306 		/* Bits 31:8 are reserved. */
6307 		*val &= 0xff;
6308 		break;
6309 	case APIC_EOI:
6310 	case APIC_ESR:
6311 	case APIC_TMICT:
6312 		/*
6313 		 * EOI, ESR: WRMSR of a non-zero value causes #GP(0).
6314 		 * TMICT: A write of 0 to the initial-count register effectively
6315 		 *        stops the local APIC timer, in both one-shot and
6316 		 *        periodic mode.
6317 		 */
6318 		*val = 0;
6319 		break;
6320 	case APIC_SPIV:
6321 	case APIC_LVTT:
6322 	case APIC_LVTTHMR:
6323 	case APIC_LVTPC:
6324 	case APIC_LVT0:
6325 	case APIC_LVT1:
6326 	case APIC_LVTERR:
6327 	case APIC_TDCR:
6328 		/*
6329 		 * To avoid writing a 1 to a reserved bit or causing some other
6330 		 * unintended side effect, read the current value and use it as
6331 		 * the write value.
6332 		 */
6333 		*val = apic_read(reg);
6334 		break;
6335 	case APIC_CMCI:
6336 		if (!apic_lvt_entry_supported(6))
6337 			return false;
6338 		*val = apic_read(reg);
6339 		break;
6340 	case APIC_ICR:
6341 		*val = 0x40000 | 0xf1;
6342 		break;
6343 	case APIC_SELF_IPI:
6344 		/*
6345 		 * With special processing (i.e., virtualize x2APIC mode +
6346 		 * virtual interrupt delivery), writing zero causes an
6347 		 * APIC-write VM exit. We plan to add a test for enabling
6348 		 * "virtual-interrupt delivery" in VMCS12, and that's where we
6349 		 * will test a self IPI with special processing.
6350 		 */
6351 		*val = 0x0;
6352 		break;
6353 	default:
6354 		return false;
6355 	}
6356 
6357 	return true;
6358 }
6359 
6360 static bool special_processing_applies(u32 reg, u64 *val,
6361 				       bool virt_int_delivery)
6362 {
6363 	bool special_processing =
6364 		(reg == APIC_TASKPRI) ||
6365 		(virt_int_delivery &&
6366 		 (reg == APIC_EOI || reg == APIC_SELF_IPI));
6367 
6368 	if (special_processing) {
6369 		TEST_ASSERT(get_x2apic_wr_val(reg, val));
6370 		return true;
6371 	}
6372 
6373 	return false;
6374 }
6375 
6376 static void virt_x2apic_mode_wr_expectation(
6377 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6378 	bool virt_int_delivery,
6379 	struct virt_x2apic_mode_expectation *expectation)
6380 {
6381 	expectation->wr_exit_reason = VMX_VMCALL;
6382 	expectation->wr_val = MAGIC_VAL_1;
6383 	expectation->wr_only = false;
6384 
6385 	if (virt_x2apic_mode_on &&
6386 	    special_processing_applies(reg, &expectation->wr_val,
6387 				       virt_int_delivery)) {
6388 		expectation->wr_behavior = X2APIC_ACCESS_VIRTUALIZED;
6389 		if (reg == APIC_SELF_IPI)
6390 			expectation->wr_exit_reason = VMX_APIC_WRITE;
6391 	} else if (!disable_x2apic &&
6392 		   get_x2apic_wr_val(reg, &expectation->wr_val)) {
6393 		expectation->wr_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6394 		if (reg == APIC_EOI || reg == APIC_SELF_IPI)
6395 			expectation->wr_only = true;
6396 		if (reg == APIC_ICR)
6397 			expectation->wr_exit_reason = VMX_EXTINT;
6398 	} else {
6399 		expectation->wr_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6400 		/*
6401 		 * Writing 1 to a reserved bit triggers a #GP.
6402 		 * Thus, set the write value to 0, which seems
6403 		 * the most likely to detect a missed #GP.
6404 		 */
6405 		expectation->wr_val = 0;
6406 	}
6407 }
6408 
6409 static void virt_x2apic_mode_exit_expectation(
6410 	u32 reg, struct virt_x2apic_mode_config *config,
6411 	struct virt_x2apic_mode_expectation *expectation)
6412 {
6413 	struct apic_reg_virt_config *base_config =
6414 		&config->apic_reg_virt_config;
6415 	bool virt_x2apic_mode_on =
6416 		base_config->virtualize_x2apic_mode &&
6417 		config->use_msr_bitmaps &&
6418 		config->disable_x2apic_msr_intercepts &&
6419 		base_config->activate_secondary_controls;
6420 
6421 	virt_x2apic_mode_wr_expectation(
6422 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6423 		config->virtual_interrupt_delivery, expectation);
6424 	virt_x2apic_mode_rd_expectation(
6425 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6426 		base_config->apic_register_virtualization,
6427 		config->virtual_interrupt_delivery, expectation);
6428 }
6429 
6430 struct virt_x2apic_mode_test_case virt_x2apic_mode_tests[] = {
6431 	/*
6432 	 * Baseline "virtualize x2APIC mode" configuration:
6433 	 *   - virtualize x2APIC mode
6434 	 *   - virtual-interrupt delivery
6435 	 *   - APIC-register virtualization
6436 	 *   - x2APIC MSR intercepts disabled
6437 	 *
6438 	 * Reads come from virtual APIC page, special processing applies to
6439 	 * VTPR, EOI, and SELF IPI, and all other writes pass through to L1
6440 	 * APIC.
6441 	 */
6442 	{
6443 		.name = "Baseline",
6444 		.virt_x2apic_mode_config = {
6445 			.virtual_interrupt_delivery = true,
6446 			.use_msr_bitmaps = true,
6447 			.disable_x2apic_msr_intercepts = true,
6448 			.disable_x2apic = false,
6449 			.apic_reg_virt_config = {
6450 				.apic_register_virtualization = true,
6451 				.use_tpr_shadow = true,
6452 				.virtualize_apic_accesses = false,
6453 				.virtualize_x2apic_mode = true,
6454 				.activate_secondary_controls = true,
6455 			},
6456 		},
6457 	},
6458 	{
6459 		.name = "Baseline w/ x2apic disabled",
6460 		.virt_x2apic_mode_config = {
6461 			.virtual_interrupt_delivery = true,
6462 			.use_msr_bitmaps = true,
6463 			.disable_x2apic_msr_intercepts = true,
6464 			.disable_x2apic = true,
6465 			.apic_reg_virt_config = {
6466 				.apic_register_virtualization = true,
6467 				.use_tpr_shadow = true,
6468 				.virtualize_apic_accesses = false,
6469 				.virtualize_x2apic_mode = true,
6470 				.activate_secondary_controls = true,
6471 			},
6472 		},
6473 	},
6474 
6475 	/*
6476 	 * Baseline, minus virtual-interrupt delivery. Reads come from virtual
6477 	 * APIC page, special processing applies to VTPR, and all other writes
6478 	 * pass through to L1 APIC.
6479 	 */
6480 	{
6481 		.name = "Baseline - virtual interrupt delivery",
6482 		.virt_x2apic_mode_config = {
6483 			.virtual_interrupt_delivery = false,
6484 			.use_msr_bitmaps = true,
6485 			.disable_x2apic_msr_intercepts = true,
6486 			.disable_x2apic = false,
6487 			.apic_reg_virt_config = {
6488 				.apic_register_virtualization = true,
6489 				.use_tpr_shadow = true,
6490 				.virtualize_apic_accesses = false,
6491 				.virtualize_x2apic_mode = true,
6492 				.activate_secondary_controls = true,
6493 			},
6494 		},
6495 	},
6496 
6497 	/*
6498 	 * Baseline, minus APIC-register virtualization. x2APIC reads pass
6499 	 * through to L1's APIC, unless reading VTPR
6500 	 */
6501 	{
6502 		.name = "Virtualize x2APIC mode, no APIC reg virt",
6503 		.virt_x2apic_mode_config = {
6504 			.virtual_interrupt_delivery = true,
6505 			.use_msr_bitmaps = true,
6506 			.disable_x2apic_msr_intercepts = true,
6507 			.disable_x2apic = false,
6508 			.apic_reg_virt_config = {
6509 				.apic_register_virtualization = false,
6510 				.use_tpr_shadow = true,
6511 				.virtualize_apic_accesses = false,
6512 				.virtualize_x2apic_mode = true,
6513 				.activate_secondary_controls = true,
6514 			},
6515 		},
6516 	},
6517 	{
6518 		.name = "Virtualize x2APIC mode, no APIC reg virt, x2APIC off",
6519 		.virt_x2apic_mode_config = {
6520 			.virtual_interrupt_delivery = true,
6521 			.use_msr_bitmaps = true,
6522 			.disable_x2apic_msr_intercepts = true,
6523 			.disable_x2apic = true,
6524 			.apic_reg_virt_config = {
6525 				.apic_register_virtualization = false,
6526 				.use_tpr_shadow = true,
6527 				.virtualize_apic_accesses = false,
6528 				.virtualize_x2apic_mode = true,
6529 				.activate_secondary_controls = true,
6530 			},
6531 		},
6532 	},
6533 
6534 	/*
6535 	 * Enable "virtualize x2APIC mode" and "APIC-register virtualization",
6536 	 * and disable intercepts for the x2APIC MSRs, but fail to enable
6537 	 * "activate secondary controls" (i.e. L2 gets access to L1's x2APIC
6538 	 * MSRs).
6539 	 */
6540 	{
6541 		.name = "Fail to enable activate secondary controls",
6542 		.virt_x2apic_mode_config = {
6543 			.virtual_interrupt_delivery = true,
6544 			.use_msr_bitmaps = true,
6545 			.disable_x2apic_msr_intercepts = true,
6546 			.disable_x2apic = false,
6547 			.apic_reg_virt_config = {
6548 				.apic_register_virtualization = true,
6549 				.use_tpr_shadow = true,
6550 				.virtualize_apic_accesses = false,
6551 				.virtualize_x2apic_mode = true,
6552 				.activate_secondary_controls = false,
6553 			},
6554 		},
6555 	},
6556 
6557 	/*
6558 	 * Enable "APIC-register virtualization" and enable "activate secondary
6559 	 * controls" and disable intercepts for the x2APIC MSRs, but do not
6560 	 * enable the "virtualize x2APIC mode" VM-execution control (i.e. L2
6561 	 * gets access to L1's x2APIC MSRs).
6562 	 */
6563 	{
6564 		.name = "Fail to enable virtualize x2APIC mode",
6565 		.virt_x2apic_mode_config = {
6566 			.virtual_interrupt_delivery = true,
6567 			.use_msr_bitmaps = true,
6568 			.disable_x2apic_msr_intercepts = true,
6569 			.disable_x2apic = false,
6570 			.apic_reg_virt_config = {
6571 				.apic_register_virtualization = true,
6572 				.use_tpr_shadow = true,
6573 				.virtualize_apic_accesses = false,
6574 				.virtualize_x2apic_mode = false,
6575 				.activate_secondary_controls = true,
6576 			},
6577 		},
6578 	},
6579 
6580 	/*
6581 	 * Disable "Virtualize x2APIC mode", disable x2APIC MSR intercepts, and
6582 	 * enable "APIC-register virtualization" --> L2 gets L1's x2APIC MSRs.
6583 	 */
6584 	{
6585 		.name = "Baseline",
6586 		.virt_x2apic_mode_config = {
6587 			.virtual_interrupt_delivery = true,
6588 			.use_msr_bitmaps = true,
6589 			.disable_x2apic_msr_intercepts = true,
6590 			.disable_x2apic = false,
6591 			.apic_reg_virt_config = {
6592 				.apic_register_virtualization = true,
6593 				.use_tpr_shadow = true,
6594 				.virtualize_apic_accesses = false,
6595 				.virtualize_x2apic_mode = false,
6596 				.activate_secondary_controls = true,
6597 			},
6598 		},
6599 	},
6600 };
6601 
6602 enum X2apic_op {
6603 	X2APIC_OP_RD,
6604 	X2APIC_OP_WR,
6605 	X2APIC_TERMINATE,
6606 };
6607 
6608 static u64 vmx_x2apic_read(u32 reg)
6609 {
6610 	u32 msr_addr = x2apic_msr(reg);
6611 	u64 val;
6612 
6613 	val = rdmsr(msr_addr);
6614 
6615 	return val;
6616 }
6617 
6618 static void vmx_x2apic_write(u32 reg, u64 val)
6619 {
6620 	u32 msr_addr = x2apic_msr(reg);
6621 
6622 	wrmsr(msr_addr, val);
6623 }
6624 
6625 struct virt_x2apic_mode_guest_args {
6626 	enum X2apic_op op;
6627 	u32 reg;
6628 	u64 val;
6629 	bool should_gp;
6630 	u64 (*virt_fn)(u64);
6631 } virt_x2apic_mode_guest_args;
6632 
6633 static volatile bool handle_x2apic_gp_ran;
6634 static volatile u32 handle_x2apic_gp_insn_len;
6635 static void handle_x2apic_gp(struct ex_regs *regs)
6636 {
6637 	handle_x2apic_gp_ran = true;
6638 	regs->rip += handle_x2apic_gp_insn_len;
6639 }
6640 
6641 static handler setup_x2apic_gp_handler(void)
6642 {
6643 	handler old_handler;
6644 
6645 	old_handler = handle_exception(GP_VECTOR, handle_x2apic_gp);
6646 	/* RDMSR and WRMSR are both 2 bytes, assuming no prefixes. */
6647 	handle_x2apic_gp_insn_len = 2;
6648 
6649 	return old_handler;
6650 }
6651 
6652 static void teardown_x2apic_gp_handler(handler old_handler)
6653 {
6654 	handle_exception(GP_VECTOR, old_handler);
6655 
6656 	/*
6657 	 * Defensively reset instruction length, so that if the handler is
6658 	 * incorrectly used, it will loop infinitely, rather than run off into
6659 	 * la la land.
6660 	 */
6661 	handle_x2apic_gp_insn_len = 0;
6662 	handle_x2apic_gp_ran = false;
6663 }
6664 
6665 static void virt_x2apic_mode_guest(void)
6666 {
6667 	volatile struct virt_x2apic_mode_guest_args *args =
6668 		&virt_x2apic_mode_guest_args;
6669 
6670 	for (;;) {
6671 		enum X2apic_op op = args->op;
6672 		u32 reg = args->reg;
6673 		u64 val = args->val;
6674 		bool should_gp = args->should_gp;
6675 		u64 (*virt_fn)(u64) = args->virt_fn;
6676 		handler old_handler;
6677 
6678 		if (op == X2APIC_TERMINATE)
6679 			break;
6680 
6681 		if (should_gp) {
6682 			TEST_ASSERT(!handle_x2apic_gp_ran);
6683 			old_handler = setup_x2apic_gp_handler();
6684 		}
6685 
6686 		if (op == X2APIC_OP_RD) {
6687 			u64 ret = vmx_x2apic_read(reg);
6688 
6689 			if (!should_gp) {
6690 				u64 want = virt_fn(val);
6691 				u64 got = virt_fn(ret);
6692 
6693 				report(got == want,
6694 				       "APIC read; got 0x%lx, want 0x%lx.",
6695 				       got, want);
6696 			}
6697 		} else if (op == X2APIC_OP_WR) {
6698 			vmx_x2apic_write(reg, val);
6699 		}
6700 
6701 		if (should_gp) {
6702 			report(handle_x2apic_gp_ran,
6703 			       "x2APIC op triggered GP.");
6704 			teardown_x2apic_gp_handler(old_handler);
6705 		}
6706 
6707 		/*
6708 		 * The L1 should always execute a vmcall after it's done testing
6709 		 * an individual APIC operation. This helps to validate that the
6710 		 * L1 and L2 are in sync with each other, as expected.
6711 		 */
6712 		vmcall();
6713 	}
6714 }
6715 
6716 static void test_x2apic_rd(
6717 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6718 	u32 *virtual_apic_page)
6719 {
6720 	u64 val = expectation->rd_val;
6721 	u32 exit_reason_want = expectation->rd_exit_reason;
6722 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6723 
6724 	report_prefix_pushf("x2apic - reading 0x%03x", reg);
6725 
6726 	/* Configure guest to do an x2apic read */
6727 	args->op = X2APIC_OP_RD;
6728 	args->reg = reg;
6729 	args->val = val;
6730 	args->should_gp = expectation->rd_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6731 	args->virt_fn = expectation->virt_fn;
6732 
6733 	/* Setup virtual APIC page */
6734 	if (expectation->rd_behavior == X2APIC_ACCESS_VIRTUALIZED)
6735 		virtual_apic_page[apic_reg_index(reg)] = (u32)val;
6736 
6737 	/* Enter guest */
6738 	enter_guest();
6739 
6740 	if (exit_reason_want != VMX_VMCALL) {
6741 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6742 	}
6743 
6744 	skip_exit_vmcall();
6745 	report_prefix_pop();
6746 }
6747 
6748 static volatile bool handle_x2apic_ipi_ran;
6749 static void handle_x2apic_ipi(isr_regs_t *regs)
6750 {
6751 	handle_x2apic_ipi_ran = true;
6752 	eoi();
6753 }
6754 
6755 static void test_x2apic_wr(
6756 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6757 	u32 *virtual_apic_page)
6758 {
6759 	u64 val = expectation->wr_val;
6760 	u32 exit_reason_want = expectation->wr_exit_reason;
6761 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6762 	int ipi_vector = 0xf1;
6763 	u32 restore_val = 0;
6764 
6765 	report_prefix_pushf("x2apic - writing 0x%lx to 0x%03x", val, reg);
6766 
6767 	/* Configure guest to do an x2apic read */
6768 	args->op = X2APIC_OP_WR;
6769 	args->reg = reg;
6770 	args->val = val;
6771 	args->should_gp = expectation->wr_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6772 
6773 	/* Setup virtual APIC page */
6774 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED)
6775 		virtual_apic_page[apic_reg_index(reg)] = 0;
6776 	if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH && !expectation->wr_only)
6777 		restore_val = apic_read(reg);
6778 
6779 	/* Setup IPI handler */
6780 	handle_x2apic_ipi_ran = false;
6781 	handle_irq(ipi_vector, handle_x2apic_ipi);
6782 
6783 	/* Enter guest */
6784 	enter_guest();
6785 
6786 	/*
6787 	 * Validate the behavior and
6788 	 * pass a magic value back to the guest.
6789 	 */
6790 	if (exit_reason_want == VMX_EXTINT) {
6791 		assert_exit_reason(exit_reason_want);
6792 
6793 		/* Clear the external interrupt. */
6794 		irq_enable();
6795 		asm volatile ("nop");
6796 		irq_disable();
6797 		report(handle_x2apic_ipi_ran,
6798 		       "Got pending interrupt after IRQ enabled.");
6799 
6800 		enter_guest();
6801 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6802 		assert_exit_reason(exit_reason_want);
6803 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6804 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%lx",
6805 		       apic_reg_index(reg),
6806 		       virtual_apic_page[apic_reg_index(reg)], val);
6807 
6808 		/* Reenter guest so it can consume/check rcx and exit again. */
6809 		enter_guest();
6810 	} else if (exit_reason_want != VMX_VMCALL) {
6811 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6812 	}
6813 
6814 	assert_exit_reason(VMX_VMCALL);
6815 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) {
6816 		u64 want = val;
6817 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6818 
6819 		report(got == want, "x2APIC write; got 0x%x, want 0x%lx", got,
6820 		       want);
6821 	} else if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH) {
6822 		if (!expectation->wr_only) {
6823 			u32 got = apic_read(reg);
6824 			bool ok;
6825 
6826 			/*
6827 			 * When L1's TPR is passed through to L2, the lower
6828 			 * nibble can be lost. For example, if L2 executes
6829 			 * WRMSR(0x808, 0x78), then, L1 might read 0x70.
6830 			 *
6831 			 * Here's how the lower nibble can get lost:
6832 			 *   1. L2 executes WRMSR(0x808, 0x78).
6833 			 *   2. L2 exits to L0 with a WRMSR exit.
6834 			 *   3. L0 emulates WRMSR, by writing L1's TPR.
6835 			 *   4. L0 re-enters L2.
6836 			 *   5. L2 exits to L0 (reason doesn't matter).
6837 			 *   6. L0 reflects L2's exit to L1.
6838 			 *   7. Before entering L1, L0 exits to user-space
6839 			 *      (e.g., to satisfy TPR access reporting).
6840 			 *   8. User-space executes KVM_SET_REGS ioctl, which
6841 			 *      clears the lower nibble of L1's TPR.
6842 			 */
6843 			if (reg == APIC_TASKPRI) {
6844 				got = apic_virt_nibble1(got);
6845 				val = apic_virt_nibble1(val);
6846 			}
6847 
6848 			ok = got == val;
6849 			report(ok,
6850 			       "non-virtualized write; val is 0x%x, want 0x%lx",
6851 			       got, val);
6852 			apic_write(reg, restore_val);
6853 		} else {
6854 			report_pass("non-virtualized and write-only OK");
6855 		}
6856 	}
6857 	skip_exit_insn();
6858 
6859 	report_prefix_pop();
6860 }
6861 
6862 static enum Config_type configure_virt_x2apic_mode_test(
6863 	struct virt_x2apic_mode_config *virt_x2apic_mode_config,
6864 	u8 *msr_bitmap_page)
6865 {
6866 	int msr;
6867 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6868 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6869 
6870 	/* x2apic-specific VMCS config */
6871 	if (virt_x2apic_mode_config->use_msr_bitmaps) {
6872 		/* virt_x2apic_mode_test() checks for MSR bitmaps support */
6873 		cpu_exec_ctrl0 |= CPU_MSR_BITMAP;
6874 	} else {
6875 		cpu_exec_ctrl0 &= ~CPU_MSR_BITMAP;
6876 	}
6877 
6878 	if (virt_x2apic_mode_config->virtual_interrupt_delivery) {
6879 		if (!(ctrl_cpu_rev[1].clr & CPU_VINTD)) {
6880 			report_skip("%s : \"virtual-interrupt delivery\" exec control not supported", __func__);
6881 			return CONFIG_TYPE_UNSUPPORTED;
6882 		}
6883 		cpu_exec_ctrl1 |= CPU_VINTD;
6884 	} else {
6885 		cpu_exec_ctrl1 &= ~CPU_VINTD;
6886 	}
6887 
6888 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6889 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6890 
6891 	/* x2APIC MSR intercepts are usually off for "Virtualize x2APIC mode" */
6892 	for (msr = 0x800; msr <= 0x8ff; msr++) {
6893 		if (virt_x2apic_mode_config->disable_x2apic_msr_intercepts) {
6894 			clear_bit(msr, msr_bitmap_page + 0x000);
6895 			clear_bit(msr, msr_bitmap_page + 0x800);
6896 		} else {
6897 			set_bit(msr, msr_bitmap_page + 0x000);
6898 			set_bit(msr, msr_bitmap_page + 0x800);
6899 		}
6900 	}
6901 
6902 	/* x2APIC mode can impact virtualization */
6903 	reset_apic();
6904 	if (!virt_x2apic_mode_config->disable_x2apic)
6905 		enable_x2apic();
6906 
6907 	return configure_apic_reg_virt_test(
6908 		&virt_x2apic_mode_config->apic_reg_virt_config);
6909 }
6910 
6911 static void virt_x2apic_mode_test(void)
6912 {
6913 	u32 *virtual_apic_page;
6914 	u8 *msr_bitmap_page;
6915 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6916 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6917 	int i;
6918 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6919 
6920 	if (!cpu_has_apicv()) {
6921 		report_skip("%s : Not all required APICv bits supported", __func__);
6922 		return;
6923 	}
6924 
6925 	/*
6926 	 * This is to exercise an issue in KVM's logic to merge L0's and L1's
6927 	 * MSR bitmaps. Previously, an L1 could get at L0's x2APIC MSRs by
6928 	 * writing the IA32_SPEC_CTRL MSR or the IA32_PRED_CMD MSRs. KVM would
6929 	 * then proceed to manipulate the MSR bitmaps, as if VMCS12 had the
6930 	 * "Virtualize x2APIC mod" control set, even when it didn't.
6931 	 */
6932 	if (this_cpu_has(X86_FEATURE_SPEC_CTRL))
6933 		wrmsr(MSR_IA32_SPEC_CTRL, 1);
6934 
6935 	/*
6936 	 * Check that VMCS12 supports:
6937 	 *   - "Virtual-APIC address", indicated by "use TPR shadow"
6938 	 *   - "MSR-bitmap address", indicated by "use MSR bitmaps"
6939 	 */
6940 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6941 		report_skip("%s : \"Use TPR shadow\" exec control not supported", __func__);
6942 		return;
6943 	} else if (!(ctrl_cpu_rev[0].clr & CPU_MSR_BITMAP)) {
6944 		report_skip("%s : \"Use MSR bitmaps\" exec control not supported", __func__);
6945 		return;
6946 	}
6947 
6948 	test_set_guest(virt_x2apic_mode_guest);
6949 
6950 	virtual_apic_page = alloc_page();
6951 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6952 
6953 	msr_bitmap_page = alloc_page();
6954 	memset(msr_bitmap_page, 0xff, PAGE_SIZE);
6955 	vmcs_write(MSR_BITMAP, virt_to_phys(msr_bitmap_page));
6956 
6957 	for (i = 0; i < ARRAY_SIZE(virt_x2apic_mode_tests); i++) {
6958 		struct virt_x2apic_mode_test_case *virt_x2apic_mode_test_case =
6959 			&virt_x2apic_mode_tests[i];
6960 		struct virt_x2apic_mode_config *virt_x2apic_mode_config =
6961 			&virt_x2apic_mode_test_case->virt_x2apic_mode_config;
6962 		enum Config_type config_type;
6963 		u32 reg;
6964 
6965 		printf("--- %s test ---\n", virt_x2apic_mode_test_case->name);
6966 		config_type =
6967 			configure_virt_x2apic_mode_test(virt_x2apic_mode_config,
6968 							msr_bitmap_page);
6969 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6970 			report_skip("Skip because of missing features.");
6971 			continue;
6972 		} else if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6973 			enter_guest_with_bad_controls();
6974 			continue;
6975 		}
6976 
6977 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6978 			struct virt_x2apic_mode_expectation expectation;
6979 
6980 			virt_x2apic_mode_exit_expectation(
6981 				reg, virt_x2apic_mode_config, &expectation);
6982 
6983 			test_x2apic_rd(reg, &expectation, virtual_apic_page);
6984 			test_x2apic_wr(reg, &expectation, virtual_apic_page);
6985 		}
6986 	}
6987 
6988 
6989 	/* Terminate the guest */
6990 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6991 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6992 	args->op = X2APIC_TERMINATE;
6993 	enter_guest();
6994 	assert_exit_reason(VMX_VMCALL);
6995 }
6996 
6997 static void test_ctl_reg(const char *cr_name, u64 cr, u64 fixed0, u64 fixed1)
6998 {
6999 	u64 val;
7000 	u64 cr_saved = vmcs_read(cr);
7001 	int i;
7002 
7003 	val = fixed0 & fixed1;
7004 	if (cr == HOST_CR4)
7005 		vmcs_write(cr, val | X86_CR4_PAE);
7006 	else
7007 		vmcs_write(cr, val);
7008 	report_prefix_pushf("%s %lx", cr_name, val);
7009 	if (val == fixed0)
7010 		test_vmx_vmlaunch(0);
7011 	else
7012 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7013 	report_prefix_pop();
7014 
7015 	for (i = 0; i < 64; i++) {
7016 
7017 		/* Set a bit when the corresponding bit in fixed1 is 0 */
7018 		if ((fixed1 & (1ull << i)) == 0) {
7019 			if (cr == HOST_CR4 && ((1ull << i) & X86_CR4_SMEP ||
7020 					       (1ull << i) & X86_CR4_SMAP))
7021 				continue;
7022 
7023 			vmcs_write(cr, cr_saved | (1ull << i));
7024 			report_prefix_pushf("%s %llx", cr_name,
7025 						cr_saved | (1ull << i));
7026 			test_vmx_vmlaunch(
7027 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7028 			report_prefix_pop();
7029 		}
7030 
7031 		/* Unset a bit when the corresponding bit in fixed0 is 1 */
7032 		if (fixed0 & (1ull << i)) {
7033 			vmcs_write(cr, cr_saved & ~(1ull << i));
7034 			report_prefix_pushf("%s %llx", cr_name,
7035 						cr_saved & ~(1ull << i));
7036 			test_vmx_vmlaunch(
7037 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7038 			report_prefix_pop();
7039 		}
7040 	}
7041 
7042 	vmcs_write(cr, cr_saved);
7043 }
7044 
7045 /*
7046  * 1. The CR0 field must not set any bit to a value not supported in VMX
7047  *    operation.
7048  * 2. The CR4 field must not set any bit to a value not supported in VMX
7049  *    operation.
7050  * 3. On processors that support Intel 64 architecture, the CR3 field must
7051  *    be such that bits 63:52 and bits in the range 51:32 beyond the
7052  *    processor's physical-address width must be 0.
7053  *
7054  *  [Intel SDM]
7055  */
7056 static void test_host_ctl_regs(void)
7057 {
7058 	u64 fixed0, fixed1, cr3, cr3_saved;
7059 	int i;
7060 
7061 	/* Test CR0 */
7062 	fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0);
7063 	fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1);
7064 	test_ctl_reg("HOST_CR0", HOST_CR0, fixed0, fixed1);
7065 
7066 	/* Test CR4 */
7067 	fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0);
7068 	fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1) &
7069 		 ~(X86_CR4_SMEP | X86_CR4_SMAP);
7070 	test_ctl_reg("HOST_CR4", HOST_CR4, fixed0, fixed1);
7071 
7072 	/* Test CR3 */
7073 	cr3_saved = vmcs_read(HOST_CR3);
7074 	for (i = cpuid_maxphyaddr(); i < 64; i++) {
7075 		cr3 = cr3_saved | (1ul << i);
7076 		vmcs_write(HOST_CR3, cr3);
7077 		report_prefix_pushf("HOST_CR3 %lx", cr3);
7078 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7079 		report_prefix_pop();
7080 	}
7081 
7082 	vmcs_write(HOST_CR3, cr3_saved);
7083 }
7084 
7085 static void test_efer_vmlaunch(u32 fld, bool ok)
7086 {
7087 	if (fld == HOST_EFER) {
7088 		if (ok)
7089 			test_vmx_vmlaunch(0);
7090 		else
7091 			test_vmx_vmlaunch2(VMXERR_ENTRY_INVALID_CONTROL_FIELD,
7092 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7093 	} else {
7094 		test_guest_state("EFER test", !ok, GUEST_EFER, "GUEST_EFER");
7095 	}
7096 }
7097 
7098 static void test_efer_one(u32 fld, const char * fld_name, u64 efer,
7099 			  u32 ctrl_fld, u64 ctrl,
7100 			  int i, const char *efer_bit_name)
7101 {
7102 	bool ok;
7103 
7104 	ok = true;
7105 	if (ctrl_fld == EXI_CONTROLS && (ctrl & EXI_LOAD_EFER)) {
7106 		if (!!(efer & EFER_LMA) != !!(ctrl & EXI_HOST_64))
7107 			ok = false;
7108 		if (!!(efer & EFER_LME) != !!(ctrl & EXI_HOST_64))
7109 			ok = false;
7110 	}
7111 	if (ctrl_fld == ENT_CONTROLS && (ctrl & ENT_LOAD_EFER)) {
7112 		/* Check LMA too since CR0.PG is set.  */
7113 		if (!!(efer & EFER_LMA) != !!(ctrl & ENT_GUEST_64))
7114 			ok = false;
7115 		if (!!(efer & EFER_LME) != !!(ctrl & ENT_GUEST_64))
7116 			ok = false;
7117 	}
7118 
7119 	/*
7120 	 * Skip the test if it would enter the guest in 32-bit mode.
7121 	 * Perhaps write the test in assembly and make sure it
7122 	 * can be run in either mode?
7123 	 */
7124 	if (fld == GUEST_EFER && ok && !(ctrl & ENT_GUEST_64))
7125 		return;
7126 
7127 	vmcs_write(ctrl_fld, ctrl);
7128 	vmcs_write(fld, efer);
7129 	report_prefix_pushf("%s %s bit turned %s, controls %s",
7130 			    fld_name, efer_bit_name,
7131 			    (i & 1) ? "on" : "off",
7132 			    (i & 2) ? "on" : "off");
7133 
7134 	test_efer_vmlaunch(fld, ok);
7135 	report_prefix_pop();
7136 }
7137 
7138 static void test_efer_bit(u32 fld, const char * fld_name,
7139 			  u32 ctrl_fld, u64 ctrl_bit, u64 efer_bit,
7140 			  const char *efer_bit_name)
7141 {
7142 	u64 efer_saved = vmcs_read(fld);
7143 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7144 	int i;
7145 
7146 	for (i = 0; i < 4; i++) {
7147 		u64 efer = efer_saved & ~efer_bit;
7148 		u64 ctrl = ctrl_saved & ~ctrl_bit;
7149 
7150 		if (i & 1)
7151 			efer |= efer_bit;
7152 		if (i & 2)
7153 			ctrl |= ctrl_bit;
7154 
7155 		test_efer_one(fld, fld_name, efer, ctrl_fld, ctrl,
7156 			      i, efer_bit_name);
7157 	}
7158 
7159 	vmcs_write(ctrl_fld, ctrl_saved);
7160 	vmcs_write(fld, efer_saved);
7161 }
7162 
7163 static void test_efer(u32 fld, const char * fld_name, u32 ctrl_fld,
7164 		      u64 ctrl_bit1, u64 ctrl_bit2)
7165 {
7166 	u64 efer_saved = vmcs_read(fld);
7167 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7168 	u64 efer_reserved_bits =  ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
7169 	u64 i;
7170 	u64 efer;
7171 
7172 	if (this_cpu_has(X86_FEATURE_NX))
7173 		efer_reserved_bits &= ~EFER_NX;
7174 
7175 	if (!ctrl_bit1) {
7176 		report_skip("%s : \"Load-IA32-EFER\" exit control not supported", __func__);
7177 		goto test_entry_exit_mode;
7178 	}
7179 
7180 	report_prefix_pushf("%s %lx", fld_name, efer_saved);
7181 	test_efer_vmlaunch(fld, true);
7182 	report_prefix_pop();
7183 
7184 	/*
7185 	 * Check reserved bits
7186 	 */
7187 	vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit1);
7188 	for (i = 0; i < 64; i++) {
7189 		if ((1ull << i) & efer_reserved_bits) {
7190 			efer = efer_saved | (1ull << i);
7191 			vmcs_write(fld, efer);
7192 			report_prefix_pushf("%s %lx", fld_name, efer);
7193 			test_efer_vmlaunch(fld, true);
7194 			report_prefix_pop();
7195 		}
7196 	}
7197 
7198 	vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit1);
7199 	for (i = 0; i < 64; i++) {
7200 		if ((1ull << i) & efer_reserved_bits) {
7201 			efer = efer_saved | (1ull << i);
7202 			vmcs_write(fld, efer);
7203 			report_prefix_pushf("%s %lx", fld_name, efer);
7204 			test_efer_vmlaunch(fld, false);
7205 			report_prefix_pop();
7206 		}
7207 	}
7208 
7209 	vmcs_write(ctrl_fld, ctrl_saved);
7210 	vmcs_write(fld, efer_saved);
7211 
7212 	/*
7213 	 * Check LMA and LME bits
7214 	 */
7215 	test_efer_bit(fld, fld_name,
7216 		      ctrl_fld, ctrl_bit1,
7217 		      EFER_LMA,
7218 		      "EFER_LMA");
7219 	test_efer_bit(fld, fld_name,
7220 		      ctrl_fld, ctrl_bit1,
7221 		      EFER_LME,
7222 		      "EFER_LME");
7223 
7224 test_entry_exit_mode:
7225 	test_efer_bit(fld, fld_name,
7226 		      ctrl_fld, ctrl_bit2,
7227 		      EFER_LMA,
7228 		      "EFER_LMA");
7229 	test_efer_bit(fld, fld_name,
7230 		      ctrl_fld, ctrl_bit2,
7231 		      EFER_LME,
7232 		      "EFER_LME");
7233 }
7234 
7235 /*
7236  * If the 'load IA32_EFER' VM-exit control is 1, bits reserved in the
7237  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7238  * the values of the LMA and LME bits in the field must each be that of
7239  * the 'host address-space size' VM-exit control.
7240  *
7241  *  [Intel SDM]
7242  */
7243 static void test_host_efer(void)
7244 {
7245 	test_efer(HOST_EFER, "HOST_EFER", EXI_CONTROLS,
7246 		  ctrl_exit_rev.clr & EXI_LOAD_EFER,
7247 		  EXI_HOST_64);
7248 }
7249 
7250 /*
7251  * If the 'load IA32_EFER' VM-enter control is 1, bits reserved in the
7252  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7253  * the values of the LMA and LME bits in the field must each be that of
7254  * the 'IA32e-mode guest' VM-exit control.
7255  */
7256 static void test_guest_efer(void)
7257 {
7258 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER)) {
7259 		report_skip("%s : \"Load-IA32-EFER\" entry control not supported", __func__);
7260 		return;
7261 	}
7262 
7263 	vmcs_write(GUEST_EFER, rdmsr(MSR_EFER));
7264 	test_efer(GUEST_EFER, "GUEST_EFER", ENT_CONTROLS,
7265 		  ctrl_enter_rev.clr & ENT_LOAD_EFER,
7266 		  ENT_GUEST_64);
7267 }
7268 
7269 /*
7270  * PAT values higher than 8 are uninteresting since they're likely lumped
7271  * in with "8". We only test values above 8 one bit at a time,
7272  * in order to reduce the number of VM-Entries and keep the runtime reasonable.
7273  */
7274 #define	PAT_VAL_LIMIT	8
7275 
7276 static void test_pat(u32 field, const char * field_name, u32 ctrl_field,
7277 		     u64 ctrl_bit)
7278 {
7279 	u32 ctrl_saved = vmcs_read(ctrl_field);
7280 	u64 pat_saved = vmcs_read(field);
7281 	u64 i, val;
7282 	u32 j;
7283 	int error;
7284 
7285 	vmcs_clear_bits(ctrl_field, ctrl_bit);
7286 
7287 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7288 		/* Test PAT0..PAT7 fields */
7289 		for (j = 0; j < (i ? 8 : 1); j++) {
7290 			val = i << j * 8;
7291 			vmcs_write(field, val);
7292 			if (field == HOST_PAT) {
7293 				report_prefix_pushf("%s %lx", field_name, val);
7294 				test_vmx_vmlaunch(0);
7295 				report_prefix_pop();
7296 
7297 			} else {	// GUEST_PAT
7298 				test_guest_state("ENT_LOAD_PAT enabled", false,
7299 						 val, "GUEST_PAT");
7300 			}
7301 		}
7302 	}
7303 
7304 	vmcs_set_bits(ctrl_field, ctrl_bit);
7305 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7306 		/* Test PAT0..PAT7 fields */
7307 		for (j = 0; j < (i ? 8 : 1); j++) {
7308 			val = i << j * 8;
7309 			vmcs_write(field, val);
7310 
7311 			if (field == HOST_PAT) {
7312 				report_prefix_pushf("%s %lx", field_name, val);
7313 				if (i == 0x2 || i == 0x3 || i >= 0x8)
7314 					error =
7315 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
7316 				else
7317 					error = 0;
7318 
7319 				test_vmx_vmlaunch(error);
7320 				report_prefix_pop();
7321 
7322 			} else {	// GUEST_PAT
7323 				error = (i == 0x2 || i == 0x3 || i >= 0x8);
7324 				test_guest_state("ENT_LOAD_PAT enabled", !!error,
7325 						 val, "GUEST_PAT");
7326 			}
7327 
7328 		}
7329 	}
7330 
7331 	vmcs_write(ctrl_field, ctrl_saved);
7332 	vmcs_write(field, pat_saved);
7333 }
7334 
7335 /*
7336  *  If the "load IA32_PAT" VM-exit control is 1, the value of the field
7337  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7338  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7339  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7340  *  6 (WB), or 7 (UC-).
7341  *
7342  *  [Intel SDM]
7343  */
7344 static void test_load_host_pat(void)
7345 {
7346 	/*
7347 	 * "load IA32_PAT" VM-exit control
7348 	 */
7349 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) {
7350 		report_skip("%s : \"Load-IA32-PAT\" exit control not supported", __func__);
7351 		return;
7352 	}
7353 
7354 	test_pat(HOST_PAT, "HOST_PAT", EXI_CONTROLS, EXI_LOAD_PAT);
7355 }
7356 
7357 union cpuidA_eax {
7358 	struct {
7359 		unsigned int version_id:8;
7360 		unsigned int num_counters_gp:8;
7361 		unsigned int bit_width:8;
7362 		unsigned int mask_length:8;
7363 	} split;
7364 	unsigned int full;
7365 };
7366 
7367 union cpuidA_edx {
7368 	struct {
7369 		unsigned int num_counters_fixed:5;
7370 		unsigned int bit_width_fixed:8;
7371 		unsigned int reserved:9;
7372 	} split;
7373 	unsigned int full;
7374 };
7375 
7376 static bool valid_pgc(u64 val)
7377 {
7378 	struct cpuid id;
7379 	union cpuidA_eax eax;
7380 	union cpuidA_edx edx;
7381 	u64 mask;
7382 
7383 	id = cpuid(0xA);
7384 	eax.full = id.a;
7385 	edx.full = id.d;
7386 	mask = ~(((1ull << eax.split.num_counters_gp) - 1) |
7387 		 (((1ull << edx.split.num_counters_fixed) - 1) << 32));
7388 
7389 	return !(val & mask);
7390 }
7391 
7392 static void test_pgc_vmlaunch(u32 xerror, u32 xreason, bool xfail, bool host)
7393 {
7394 	u32 inst_err;
7395 	u64 obs;
7396 	bool success;
7397 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7398 
7399 	if (host) {
7400 		success = vmlaunch_succeeds();
7401 		obs = rdmsr(data->msr);
7402 		if (!success) {
7403 			inst_err = vmcs_read(VMX_INST_ERROR);
7404 			report(xerror == inst_err, "vmlaunch failed, "
7405 			       "VMX Inst Error is %d (expected %d)",
7406 			       inst_err, xerror);
7407 		} else {
7408 			report(!data->enabled || data->exp == obs,
7409 			       "Host state is 0x%lx (expected 0x%lx)",
7410 			       obs, data->exp);
7411 			report(success != xfail, "vmlaunch succeeded");
7412 		}
7413 	} else {
7414 		test_guest_state("load GUEST_PERF_GLOBAL_CTRL", xfail,
7415 				 GUEST_PERF_GLOBAL_CTRL,
7416 				 "GUEST_PERF_GLOBAL_CTRL");
7417 	}
7418 }
7419 
7420 /*
7421  * test_load_perf_global_ctrl is a generic function for testing the
7422  * "load IA32_PERF_GLOBAL_CTRL" VM-{Entry,Exit} controls. This test function
7423  * tests the provided ctrl_val when disabled and enabled.
7424  *
7425  * @nr: VMCS field number corresponding to the host/guest state field
7426  * @name: Name of the above VMCS field for printing in test report
7427  * @ctrl_nr: VMCS field number corresponding to the VM-{Entry,Exit} control
7428  * @ctrl_val: Bit to set on the ctrl_field
7429  */
7430 static void test_perf_global_ctrl(u32 nr, const char *name, u32 ctrl_nr,
7431 				  const char *ctrl_name, u64 ctrl_val)
7432 {
7433 	u64 ctrl_saved = vmcs_read(ctrl_nr);
7434 	u64 pgc_saved = vmcs_read(nr);
7435 	u64 i, val;
7436 	bool host = nr == HOST_PERF_GLOBAL_CTRL;
7437 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7438 
7439 	data->msr = MSR_CORE_PERF_GLOBAL_CTRL;
7440 	msr_bmp_init();
7441 	vmcs_write(ctrl_nr, ctrl_saved & ~ctrl_val);
7442 	data->enabled = false;
7443 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=0 on %s",
7444 			    ctrl_name);
7445 
7446 	for (i = 0; i < 64; i++) {
7447 		val = 1ull << i;
7448 		vmcs_write(nr, val);
7449 		report_prefix_pushf("%s = 0x%lx", name, val);
7450 		test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7451 		report_prefix_pop();
7452 	}
7453 	report_prefix_pop();
7454 
7455 	vmcs_write(ctrl_nr, ctrl_saved | ctrl_val);
7456 	data->enabled = true;
7457 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=1 on %s",
7458 			    ctrl_name);
7459 	for (i = 0; i < 64; i++) {
7460 		val = 1ull << i;
7461 		data->exp = val;
7462 		vmcs_write(nr, val);
7463 		report_prefix_pushf("%s = 0x%lx", name, val);
7464 		if (valid_pgc(val)) {
7465 			test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7466 		} else {
7467 			if (host)
7468 				test_pgc_vmlaunch(
7469 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
7470 					0,
7471 					true,
7472 					host);
7473 			else
7474 				test_pgc_vmlaunch(
7475 					0,
7476 					VMX_ENTRY_FAILURE | VMX_FAIL_STATE,
7477 					true,
7478 					host);
7479 		}
7480 		report_prefix_pop();
7481 	}
7482 
7483 	data->enabled = false;
7484 	report_prefix_pop();
7485 	vmcs_write(ctrl_nr, ctrl_saved);
7486 	vmcs_write(nr, pgc_saved);
7487 }
7488 
7489 static void test_load_host_perf_global_ctrl(void)
7490 {
7491 	if (!this_cpu_has_perf_global_ctrl()) {
7492 		report_skip("%s : \"IA32_PERF_GLOBAL_CTRL\" MSR not supported", __func__);
7493 		return;
7494 	}
7495 
7496 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PERF)) {
7497 		report_skip("%s : \"Load IA32_PERF_GLOBAL_CTRL\" exit control not supported", __func__);
7498 		return;
7499 	}
7500 
7501 	test_perf_global_ctrl(HOST_PERF_GLOBAL_CTRL, "HOST_PERF_GLOBAL_CTRL",
7502 				   EXI_CONTROLS, "EXI_CONTROLS", EXI_LOAD_PERF);
7503 }
7504 
7505 
7506 static void test_load_guest_perf_global_ctrl(void)
7507 {
7508 	if (!this_cpu_has_perf_global_ctrl()) {
7509 		report_skip("%s : \"IA32_PERF_GLOBAL_CTRL\" MSR not supported", __func__);
7510 		return;
7511 	}
7512 
7513 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PERF)) {
7514 		report_skip("%s : \"Load IA32_PERF_GLOBAL_CTRL\" entry control not supported", __func__);
7515 		return;
7516 	}
7517 
7518 	test_perf_global_ctrl(GUEST_PERF_GLOBAL_CTRL, "GUEST_PERF_GLOBAL_CTRL",
7519 				   ENT_CONTROLS, "ENT_CONTROLS", ENT_LOAD_PERF);
7520 }
7521 
7522 
7523 /*
7524  * test_vmcs_field - test a value for the given VMCS field
7525  * @field: VMCS field
7526  * @field_name: string name of VMCS field
7527  * @bit_start: starting bit
7528  * @bit_end: ending bit
7529  * @val: value that the bit range must or must not contain
7530  * @valid_val: whether value given in 'val' must be valid or not
7531  * @error: expected VMCS error when vmentry fails for an invalid value
7532  */
7533 static void test_vmcs_field(u64 field, const char *field_name, u32 bit_start,
7534 			    u32 bit_end, u64 val, bool valid_val, u32 error)
7535 {
7536 	u64 field_saved = vmcs_read(field);
7537 	u32 i;
7538 	u64 tmp;
7539 	u32 bit_on;
7540 	u64 mask = ~0ull;
7541 
7542 	mask = (mask >> bit_end) << bit_end;
7543 	mask = mask | ((1 << bit_start) - 1);
7544 	tmp = (field_saved & mask) | (val << bit_start);
7545 
7546 	vmcs_write(field, tmp);
7547 	report_prefix_pushf("%s %lx", field_name, tmp);
7548 	if (valid_val)
7549 		test_vmx_vmlaunch(0);
7550 	else
7551 		test_vmx_vmlaunch(error);
7552 	report_prefix_pop();
7553 
7554 	for (i = bit_start; i <= bit_end; i = i + 2) {
7555 		bit_on = ((1ull < i) & (val << bit_start)) ? 0 : 1;
7556 		if (bit_on)
7557 			tmp = field_saved | (1ull << i);
7558 		else
7559 			tmp = field_saved & ~(1ull << i);
7560 		vmcs_write(field, tmp);
7561 		report_prefix_pushf("%s %lx", field_name, tmp);
7562 		if (valid_val)
7563 			test_vmx_vmlaunch(error);
7564 		else
7565 			test_vmx_vmlaunch(0);
7566 		report_prefix_pop();
7567 	}
7568 
7569 	vmcs_write(field, field_saved);
7570 }
7571 
7572 static void test_canonical(u64 field, const char * field_name, bool host)
7573 {
7574 	u64 addr_saved = vmcs_read(field);
7575 
7576 	/*
7577 	 * Use the existing value if possible.  Writing a random canonical
7578 	 * value is not an option as doing so would corrupt the field being
7579 	 * tested and likely hose the test.
7580 	 */
7581 	if (is_canonical(addr_saved)) {
7582 		if (host) {
7583 			report_prefix_pushf("%s %lx", field_name, addr_saved);
7584 			test_vmx_vmlaunch(0);
7585 			report_prefix_pop();
7586 		} else {
7587 			test_guest_state("Test canonical address", false,
7588 					 addr_saved, field_name);
7589 		}
7590 	}
7591 
7592 	vmcs_write(field, NONCANONICAL);
7593 
7594 	if (host) {
7595 		report_prefix_pushf("%s %llx", field_name, NONCANONICAL);
7596 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7597 		report_prefix_pop();
7598 	} else {
7599 		test_guest_state("Test non-canonical address", true,
7600 				 NONCANONICAL, field_name);
7601 	}
7602 
7603 	vmcs_write(field, addr_saved);
7604 }
7605 
7606 #define TEST_RPL_TI_FLAGS(reg, name)				\
7607 	test_vmcs_field(reg, name, 0, 2, 0x0, true,		\
7608 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7609 
7610 #define TEST_CS_TR_FLAGS(reg, name)				\
7611 	test_vmcs_field(reg, name, 3, 15, 0x0000, false,	\
7612 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7613 
7614 /*
7615  * 1. In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the
7616  *    RPL (bits 1:0) and the TI flag (bit 2) must be 0.
7617  * 2. The selector fields for CS and TR cannot be 0000H.
7618  * 3. The selector field for SS cannot be 0000H if the "host address-space
7619  *    size" VM-exit control is 0.
7620  * 4. On processors that support Intel 64 architecture, the base-address
7621  *    fields for FS, GS and TR must contain canonical addresses.
7622  */
7623 static void test_host_segment_regs(void)
7624 {
7625 	u16 selector_saved;
7626 
7627 	/*
7628 	 * Test RPL and TI flags
7629 	 */
7630 	TEST_RPL_TI_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7631 	TEST_RPL_TI_FLAGS(HOST_SEL_SS, "HOST_SEL_SS");
7632 	TEST_RPL_TI_FLAGS(HOST_SEL_DS, "HOST_SEL_DS");
7633 	TEST_RPL_TI_FLAGS(HOST_SEL_ES, "HOST_SEL_ES");
7634 	TEST_RPL_TI_FLAGS(HOST_SEL_FS, "HOST_SEL_FS");
7635 	TEST_RPL_TI_FLAGS(HOST_SEL_GS, "HOST_SEL_GS");
7636 	TEST_RPL_TI_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7637 
7638 	/*
7639 	 * Test that CS and TR fields can not be 0x0000
7640 	 */
7641 	TEST_CS_TR_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7642 	TEST_CS_TR_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7643 
7644 	/*
7645 	 * SS field can not be 0x0000 if "host address-space size" VM-exit
7646 	 * control is 0
7647 	 */
7648 	selector_saved = vmcs_read(HOST_SEL_SS);
7649 	vmcs_write(HOST_SEL_SS, 0);
7650 	report_prefix_pushf("HOST_SEL_SS 0");
7651 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7652 		test_vmx_vmlaunch(0);
7653 	} else {
7654 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7655 	}
7656 	report_prefix_pop();
7657 
7658 	vmcs_write(HOST_SEL_SS, selector_saved);
7659 
7660 	/*
7661 	 * Base address for FS, GS and TR must be canonical
7662 	 */
7663 	test_canonical(HOST_BASE_FS, "HOST_BASE_FS", true);
7664 	test_canonical(HOST_BASE_GS, "HOST_BASE_GS", true);
7665 	test_canonical(HOST_BASE_TR, "HOST_BASE_TR", true);
7666 }
7667 
7668 /*
7669  *  On processors that support Intel 64 architecture, the base-address
7670  *  fields for GDTR and IDTR must contain canonical addresses.
7671  */
7672 static void test_host_desc_tables(void)
7673 {
7674 	test_canonical(HOST_BASE_GDTR, "HOST_BASE_GDTR", true);
7675 	test_canonical(HOST_BASE_IDTR, "HOST_BASE_IDTR", true);
7676 }
7677 
7678 /*
7679  * If the "host address-space size" VM-exit control is 0, the following must
7680  * hold:
7681  *    - The "IA-32e mode guest" VM-entry control is 0.
7682  *    - Bit 17 of the CR4 field (corresponding to CR4.PCIDE) is 0.
7683  *    - Bits 63:32 in the RIP field are 0.
7684  *
7685  * If the "host address-space size" VM-exit control is 1, the following must
7686  * hold:
7687  *    - Bit 5 of the CR4 field (corresponding to CR4.PAE) is 1.
7688  *    - The RIP field contains a canonical address.
7689  *
7690  */
7691 static void test_host_addr_size(void)
7692 {
7693 	u64 cr4_saved = vmcs_read(HOST_CR4);
7694 	u64 rip_saved = vmcs_read(HOST_RIP);
7695 	u64 entry_ctrl_saved = vmcs_read(ENT_CONTROLS);
7696 	int i;
7697 	u64 tmp;
7698 
7699 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7700 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7701 		report_prefix_pushf("\"IA-32e mode guest\" enabled");
7702 		test_vmx_vmlaunch(0);
7703 		report_prefix_pop();
7704 
7705 		vmcs_write(HOST_CR4, cr4_saved | X86_CR4_PCIDE);
7706 		report_prefix_pushf("\"CR4.PCIDE\" set");
7707 		test_vmx_vmlaunch(0);
7708 		report_prefix_pop();
7709 
7710 		for (i = 32; i <= 63; i = i + 4) {
7711 			tmp = rip_saved | 1ull << i;
7712 			vmcs_write(HOST_RIP, tmp);
7713 			report_prefix_pushf("HOST_RIP %lx", tmp);
7714 			test_vmx_vmlaunch(0);
7715 			report_prefix_pop();
7716 		}
7717 
7718 		if (cr4_saved & X86_CR4_PAE) {
7719 			vmcs_write(HOST_CR4, cr4_saved  & ~X86_CR4_PAE);
7720 			report_prefix_pushf("\"CR4.PAE\" unset");
7721 			test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7722 		} else {
7723 			report_prefix_pushf("\"CR4.PAE\" set");
7724 			test_vmx_vmlaunch(0);
7725 		}
7726 		report_prefix_pop();
7727 
7728 		vmcs_write(HOST_RIP, NONCANONICAL);
7729 		report_prefix_pushf("HOST_RIP %llx", NONCANONICAL);
7730 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7731 		report_prefix_pop();
7732 
7733 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7734 		vmcs_write(HOST_RIP, rip_saved);
7735 		vmcs_write(HOST_CR4, cr4_saved);
7736 
7737 		/* Restore host's active RIP and CR4 values. */
7738 		report_prefix_pushf("restore host state");
7739 		test_vmx_vmlaunch(0);
7740 		report_prefix_pop();
7741 	}
7742 }
7743 
7744 /*
7745  * Check that the virtual CPU checks the VMX Host State Area as
7746  * documented in the Intel SDM.
7747  */
7748 static void vmx_host_state_area_test(void)
7749 {
7750 	/*
7751 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
7752 	 * fail due to invalid guest state, should we make it that
7753 	 * far.
7754 	 */
7755 	vmcs_write(GUEST_RFLAGS, 0);
7756 
7757 	test_host_ctl_regs();
7758 
7759 	test_canonical(HOST_SYSENTER_ESP, "HOST_SYSENTER_ESP", true);
7760 	test_canonical(HOST_SYSENTER_EIP, "HOST_SYSENTER_EIP", true);
7761 
7762 	test_host_efer();
7763 	test_load_host_pat();
7764 	test_host_segment_regs();
7765 	test_host_desc_tables();
7766 	test_host_addr_size();
7767 	test_load_host_perf_global_ctrl();
7768 }
7769 
7770 /*
7771  * If the "load debug controls" VM-entry control is 1, bits 63:32 in
7772  * the DR7 field must be 0.
7773  *
7774  * [Intel SDM]
7775  */
7776 static void test_guest_dr7(void)
7777 {
7778 	u32 ent_saved = vmcs_read(ENT_CONTROLS);
7779 	u64 dr7_saved = vmcs_read(GUEST_DR7);
7780 	u64 val;
7781 	int i;
7782 
7783 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS) {
7784 		vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7785 		for (i = 0; i < 64; i++) {
7786 			val = 1ull << i;
7787 			vmcs_write(GUEST_DR7, val);
7788 			test_guest_state("ENT_LOAD_DBGCTLS disabled", false,
7789 					 val, "GUEST_DR7");
7790 		}
7791 	}
7792 	if (ctrl_enter_rev.clr & ENT_LOAD_DBGCTLS) {
7793 		vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7794 		for (i = 0; i < 64; i++) {
7795 			val = 1ull << i;
7796 			vmcs_write(GUEST_DR7, val);
7797 			test_guest_state("ENT_LOAD_DBGCTLS enabled", i >= 32,
7798 					 val, "GUEST_DR7");
7799 		}
7800 	}
7801 	vmcs_write(GUEST_DR7, dr7_saved);
7802 	vmcs_write(ENT_CONTROLS, ent_saved);
7803 }
7804 
7805 /*
7806  *  If the "load IA32_PAT" VM-entry control is 1, the value of the field
7807  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7808  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7809  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7810  *  6 (WB), or 7 (UC-).
7811  *
7812  *  [Intel SDM]
7813  */
7814 static void test_load_guest_pat(void)
7815 {
7816 	/*
7817 	 * "load IA32_PAT" VM-entry control
7818 	 */
7819 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
7820 		report_skip("%s : \"Load-IA32-PAT\" entry control not supported", __func__);
7821 		return;
7822 	}
7823 
7824 	test_pat(GUEST_PAT, "GUEST_PAT", ENT_CONTROLS, ENT_LOAD_PAT);
7825 }
7826 
7827 #define MSR_IA32_BNDCFGS_RSVD_MASK	0x00000ffc
7828 
7829 /*
7830  * If the "load IA32_BNDCFGS" VM-entry control is 1, the following
7831  * checks are performed on the field for the IA32_BNDCFGS MSR:
7832  *
7833  *   - Bits reserved in the IA32_BNDCFGS MSR must be 0.
7834  *   - The linear address in bits 63:12 must be canonical.
7835  *
7836  *  [Intel SDM]
7837  */
7838 static void test_load_guest_bndcfgs(void)
7839 {
7840 	u64 bndcfgs_saved = vmcs_read(GUEST_BNDCFGS);
7841 	u64 bndcfgs;
7842 
7843 	if (!(ctrl_enter_rev.clr & ENT_LOAD_BNDCFGS)) {
7844 		report_skip("%s : \"Load-IA32-BNDCFGS\" entry control not supported", __func__);
7845 		return;
7846 	}
7847 
7848 	vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7849 
7850 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7851 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7852 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7853 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7854 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7855 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7856 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7857 
7858 	vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7859 
7860 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7861 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7862 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7863 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7864 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7865 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7866 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7867 
7868 	vmcs_write(GUEST_BNDCFGS, bndcfgs_saved);
7869 }
7870 
7871 #define	GUEST_SEG_UNUSABLE_MASK	(1u << 16)
7872 #define	GUEST_SEG_SEL_TI_MASK	(1u << 2)
7873 
7874 
7875 #define	TEST_SEGMENT_SEL(test, xfail, sel, val)				\
7876 do {									\
7877 	vmcs_write(sel, val);						\
7878 	test_guest_state(test " segment", xfail, val, xstr(sel));	\
7879 } while (0)
7880 
7881 #define	TEST_INVALID_SEG_SEL(sel, val) \
7882 	TEST_SEGMENT_SEL("Invalid: " xstr(val), true, sel, val);
7883 
7884 #define	TEST_VALID_SEG_SEL(sel, val) \
7885 	TEST_SEGMENT_SEL("Valid: " xstr(val), false, sel, val);
7886 
7887 /*
7888  * The following checks are done on the Selector field of the Guest Segment
7889  * Registers:
7890  *    - TR. The TI flag (bit 2) must be 0.
7891  *    - LDTR. If LDTR is usable, the TI flag (bit 2) must be 0.
7892  *    - SS. If the guest will not be virtual-8086 and the "unrestricted
7893  *	guest" VM-execution control is 0, the RPL (bits 1:0) must equal
7894  *	the RPL of the selector field for CS.
7895  *
7896  *  [Intel SDM]
7897  */
7898 static void test_guest_segment_sel_fields(void)
7899 {
7900 	u16 sel_saved;
7901 	u32 ar_saved;
7902 	u32 cpu_ctrl0_saved;
7903 	u32 cpu_ctrl1_saved;
7904 	u16 cs_rpl_bits;
7905 
7906 	/*
7907 	 * Test for GUEST_SEL_TR
7908 	 */
7909 	sel_saved = vmcs_read(GUEST_SEL_TR);
7910 	TEST_INVALID_SEG_SEL(GUEST_SEL_TR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7911 	vmcs_write(GUEST_SEL_TR, sel_saved);
7912 
7913 	/*
7914 	 * Test for GUEST_SEL_LDTR
7915 	 */
7916 	sel_saved = vmcs_read(GUEST_SEL_LDTR);
7917 	ar_saved = vmcs_read(GUEST_AR_LDTR);
7918 	/* LDTR is set unusable */
7919 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7920 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7921 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7922 	/* LDTR is set usable */
7923 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7924 	TEST_INVALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7925 
7926 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7927 
7928 	vmcs_write(GUEST_AR_LDTR, ar_saved);
7929 	vmcs_write(GUEST_SEL_LDTR, sel_saved);
7930 
7931 	/*
7932 	 * Test for GUEST_SEL_SS
7933 	 */
7934 	cpu_ctrl0_saved = vmcs_read(CPU_EXEC_CTRL0);
7935 	cpu_ctrl1_saved = vmcs_read(CPU_EXEC_CTRL1);
7936 	ar_saved = vmcs_read(GUEST_AR_SS);
7937 	/* Turn off "unrestricted guest" vm-execution control */
7938 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved & ~CPU_URG);
7939 	cs_rpl_bits = vmcs_read(GUEST_SEL_CS) & 0x3;
7940 	sel_saved = vmcs_read(GUEST_SEL_SS);
7941 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7942 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7943 	/* Make SS usable if it's unusable or vice-versa */
7944 	if (ar_saved & GUEST_SEG_UNUSABLE_MASK)
7945 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7946 	else
7947 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7948 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7949 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7950 
7951 	/* Need a valid EPTP as the passing case fully enters the guest. */
7952 	if (enable_unrestricted_guest(true))
7953 		goto skip_ss_tests;
7954 
7955 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7956 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7957 
7958 	/* Make SS usable if it's unusable or vice-versa */
7959 	if (vmcs_read(GUEST_AR_SS) & GUEST_SEG_UNUSABLE_MASK)
7960 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7961 	else
7962 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7963 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7964 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7965 skip_ss_tests:
7966 
7967 	vmcs_write(GUEST_AR_SS, ar_saved);
7968 	vmcs_write(GUEST_SEL_SS, sel_saved);
7969 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrl0_saved);
7970 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved);
7971 }
7972 
7973 #define	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(xfail, seg_base)			\
7974 do {										\
7975 	addr_saved = vmcs_read(seg_base);					\
7976 	for (i = 32; i < 63; i = i + 4) {					\
7977 		addr = addr_saved | 1ull << i;					\
7978 		vmcs_write(seg_base, addr);					\
7979 		test_guest_state("seg.BASE[63:32] != 0, usable = " xstr(xfail),	\
7980 				 xfail, addr, xstr(seg_base));			\
7981 	}									\
7982 	vmcs_write(seg_base, addr_saved);					\
7983 } while (0)
7984 
7985 #define	TEST_SEGMENT_BASE_ADDR_CANONICAL(xfail, seg_base)		  \
7986 do {									  \
7987 	addr_saved = vmcs_read(seg_base);				  \
7988 	vmcs_write(seg_base, NONCANONICAL);				  \
7989 	test_guest_state("seg.BASE non-canonical, usable = " xstr(xfail), \
7990 			 xfail, NONCANONICAL, xstr(seg_base));		  \
7991 	vmcs_write(seg_base, addr_saved);				  \
7992 } while (0)
7993 
7994 /*
7995  * The following checks are done on the Base Address field of the Guest
7996  * Segment Registers on processors that support Intel 64 architecture:
7997  *    - TR, FS, GS : The address must be canonical.
7998  *    - LDTR : If LDTR is usable, the address must be canonical.
7999  *    - CS : Bits 63:32 of the address must be zero.
8000  *    - SS, DS, ES : If the register is usable, bits 63:32 of the address
8001  *	must be zero.
8002  *
8003  *  [Intel SDM]
8004  */
8005 static void test_guest_segment_base_addr_fields(void)
8006 {
8007 	u64 addr_saved;
8008 	u64 addr;
8009 	u32 ar_saved;
8010 	int i;
8011 
8012 	/*
8013 	 * The address of TR, FS, GS and LDTR must be canonical.
8014 	 */
8015 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_TR);
8016 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_FS);
8017 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_GS);
8018 	ar_saved = vmcs_read(GUEST_AR_LDTR);
8019 	/* Make LDTR unusable */
8020 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8021 	TEST_SEGMENT_BASE_ADDR_CANONICAL(false, GUEST_BASE_LDTR);
8022 	/* Make LDTR usable */
8023 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8024 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_LDTR);
8025 
8026 	vmcs_write(GUEST_AR_LDTR, ar_saved);
8027 
8028 	/*
8029 	 * Bits 63:32 in CS, SS, DS and ES base address must be zero
8030 	 */
8031 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_CS);
8032 	ar_saved = vmcs_read(GUEST_AR_SS);
8033 	/* Make SS unusable */
8034 	vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8035 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_SS);
8036 	/* Make SS usable */
8037 	vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8038 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_SS);
8039 	vmcs_write(GUEST_AR_SS, ar_saved);
8040 
8041 	ar_saved = vmcs_read(GUEST_AR_DS);
8042 	/* Make DS unusable */
8043 	vmcs_write(GUEST_AR_DS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8044 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_DS);
8045 	/* Make DS usable */
8046 	vmcs_write(GUEST_AR_DS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8047 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_DS);
8048 	vmcs_write(GUEST_AR_DS, ar_saved);
8049 
8050 	ar_saved = vmcs_read(GUEST_AR_ES);
8051 	/* Make ES unusable */
8052 	vmcs_write(GUEST_AR_ES, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8053 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_ES);
8054 	/* Make ES usable */
8055 	vmcs_write(GUEST_AR_ES, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8056 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_ES);
8057 	vmcs_write(GUEST_AR_ES, ar_saved);
8058 }
8059 
8060 /*
8061  * Check that the virtual CPU checks the VMX Guest State Area as
8062  * documented in the Intel SDM.
8063  */
8064 static void vmx_guest_state_area_test(void)
8065 {
8066 	vmx_set_test_stage(1);
8067 	test_set_guest(guest_state_test_main);
8068 
8069 	/*
8070 	 * The IA32_SYSENTER_ESP field and the IA32_SYSENTER_EIP field
8071 	 * must each contain a canonical address.
8072 	 */
8073 	test_canonical(GUEST_SYSENTER_ESP, "GUEST_SYSENTER_ESP", false);
8074 	test_canonical(GUEST_SYSENTER_EIP, "GUEST_SYSENTER_EIP", false);
8075 
8076 	test_guest_dr7();
8077 	test_load_guest_pat();
8078 	test_guest_efer();
8079 	test_load_guest_perf_global_ctrl();
8080 	test_load_guest_bndcfgs();
8081 
8082 	test_guest_segment_sel_fields();
8083 	test_guest_segment_base_addr_fields();
8084 
8085 	test_canonical(GUEST_BASE_GDTR, "GUEST_BASE_GDTR", false);
8086 	test_canonical(GUEST_BASE_IDTR, "GUEST_BASE_IDTR", false);
8087 
8088 	u32 guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_GDTR);
8089 	int i;
8090 	for (i = 16; i <= 31; i++) {
8091 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8092 		vmcs_write(GUEST_LIMIT_GDTR, tmp);
8093 		test_guest_state("GDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_GDTR");
8094 	}
8095 	vmcs_write(GUEST_LIMIT_GDTR, guest_desc_limit_saved);
8096 
8097 	guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_IDTR);
8098 	for (i = 16; i <= 31; i++) {
8099 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8100 		vmcs_write(GUEST_LIMIT_IDTR, tmp);
8101 		test_guest_state("IDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_IDTR");
8102 	}
8103 	vmcs_write(GUEST_LIMIT_IDTR, guest_desc_limit_saved);
8104 
8105 	/*
8106 	 * Let the guest finish execution
8107 	 */
8108 	vmx_set_test_stage(2);
8109 	enter_guest();
8110 }
8111 
8112 extern void unrestricted_guest_main(void);
8113 asm (".code32\n"
8114 	"unrestricted_guest_main:\n"
8115 	"vmcall\n"
8116 	"nop\n"
8117 	"mov $1, %edi\n"
8118 	"call hypercall\n"
8119 	".code64\n");
8120 
8121 static void setup_unrestricted_guest(void)
8122 {
8123 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) & ~(X86_CR0_PG));
8124 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) & ~ENT_GUEST_64);
8125 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) & ~EFER_LMA);
8126 	vmcs_write(GUEST_RIP, virt_to_phys(unrestricted_guest_main));
8127 }
8128 
8129 static void unsetup_unrestricted_guest(void)
8130 {
8131 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) | X86_CR0_PG);
8132 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_GUEST_64);
8133 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) | EFER_LMA);
8134 	vmcs_write(GUEST_RIP, (u64) phys_to_virt(vmcs_read(GUEST_RIP)));
8135 	vmcs_write(GUEST_RSP, (u64) phys_to_virt(vmcs_read(GUEST_RSP)));
8136 }
8137 
8138 /*
8139  * If "unrestricted guest" secondary VM-execution control is set, guests
8140  * can run in unpaged protected mode.
8141  */
8142 static void vmentry_unrestricted_guest_test(void)
8143 {
8144 	if (enable_unrestricted_guest(true)) {
8145 		report_skip("%s: \"Unrestricted guest\" exec control not supported", __func__);
8146 		return;
8147 	}
8148 
8149 	test_set_guest(unrestricted_guest_main);
8150 	setup_unrestricted_guest();
8151 	test_guest_state("Unrestricted guest test", false, CPU_URG, "CPU_URG");
8152 
8153 	/*
8154 	 * Let the guest finish execution as a regular guest
8155 	 */
8156 	unsetup_unrestricted_guest();
8157 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) & ~CPU_URG);
8158 	enter_guest();
8159 }
8160 
8161 static bool valid_vmcs_for_vmentry(void)
8162 {
8163 	struct vmcs *current_vmcs = NULL;
8164 
8165 	if (vmcs_save(&current_vmcs))
8166 		return false;
8167 
8168 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
8169 }
8170 
8171 static void try_vmentry_in_movss_shadow(void)
8172 {
8173 	u32 vm_inst_err;
8174 	u32 flags;
8175 	bool early_failure = false;
8176 	u32 expected_flags = X86_EFLAGS_FIXED;
8177 	bool valid_vmcs = valid_vmcs_for_vmentry();
8178 
8179 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
8180 
8181 	/*
8182 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
8183 	 * unsupported VMCS component").
8184 	 */
8185 	vmcs_write(~0u, 0);
8186 
8187 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
8188 			      "vmwrite %%rsp, %%rdx;"
8189 			      "mov 0f, %%rax;"
8190 			      "mov %[host_rip], %%edx;"
8191 			      "vmwrite %%rax, %%rdx;"
8192 			      "mov $-1, %%ah;"
8193 			      "sahf;"
8194 			      "mov %%ss, %%ax;"
8195 			      "mov %%ax, %%ss;"
8196 			      "vmlaunch;"
8197 			      "mov $1, %[early_failure];"
8198 			      "0: lahf;"
8199 			      "movzbl %%ah, %[flags]"
8200 			      : [early_failure] "+r" (early_failure),
8201 				[flags] "=&a" (flags)
8202 			      : [host_rsp] "i" (HOST_RSP),
8203 				[host_rip] "i" (HOST_RIP)
8204 			      : "rdx", "cc", "memory");
8205 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
8206 
8207 	report(early_failure, "Early VM-entry failure");
8208 	report(flags == expected_flags, "RFLAGS[8:0] is %x (actual %x)",
8209 	       expected_flags, flags);
8210 	if (valid_vmcs)
8211 		report(vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
8212 		       "VM-instruction error is %d (actual %d)",
8213 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
8214 }
8215 
8216 static void vmentry_movss_shadow_test(void)
8217 {
8218 	struct vmcs *orig_vmcs;
8219 
8220 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
8221 
8222 	/*
8223 	 * Set the launched flag on the current VMCS to verify the correct
8224 	 * error priority, below.
8225 	 */
8226 	test_set_guest(v2_null_test_guest);
8227 	enter_guest();
8228 
8229 	/*
8230 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
8231 	 * fail due to invalid guest state (if we make it that far).
8232 	 */
8233 	vmcs_write(GUEST_RFLAGS, 0);
8234 
8235 	/*
8236 	 * "VM entry with events blocked by MOV SS" takes precedence over
8237 	 * "VMLAUNCH with non-clear VMCS."
8238 	 */
8239 	report_prefix_push("valid current-VMCS");
8240 	try_vmentry_in_movss_shadow();
8241 	report_prefix_pop();
8242 
8243 	/*
8244 	 * VMfailInvalid takes precedence over "VM entry with events
8245 	 * blocked by MOV SS."
8246 	 */
8247 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
8248 	report_prefix_push("no current-VMCS");
8249 	try_vmentry_in_movss_shadow();
8250 	report_prefix_pop();
8251 
8252 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
8253 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8254 }
8255 
8256 static void vmx_ldtr_test_guest(void)
8257 {
8258 	u16 ldtr = sldt();
8259 
8260 	report(ldtr == NP_SEL, "Expected %x for L2 LDTR selector (got %x)",
8261 	       NP_SEL, ldtr);
8262 }
8263 
8264 /*
8265  * Ensure that the L1 LDTR is set to 0 on VM-exit.
8266  */
8267 static void vmx_ldtr_test(void)
8268 {
8269 	const u8 ldt_ar = 0x82; /* Present LDT */
8270 	u16 sel = FIRST_SPARE_SEL;
8271 
8272 	/* Set up a non-zero L1 LDTR prior to VM-entry. */
8273 	set_gdt_entry(sel, 0, 0, ldt_ar, 0);
8274 	lldt(sel);
8275 
8276 	test_set_guest(vmx_ldtr_test_guest);
8277 	/*
8278 	 * Set up a different LDTR for L2. The actual GDT contents are
8279 	 * irrelevant, since we stuff the hidden descriptor state
8280 	 * straight into the VMCS rather than reading it from the GDT.
8281 	 */
8282 	vmcs_write(GUEST_SEL_LDTR, NP_SEL);
8283 	vmcs_write(GUEST_AR_LDTR, ldt_ar);
8284 	enter_guest();
8285 
8286 	/*
8287 	 * VM-exit should clear LDTR (and make it unusable, but we
8288 	 * won't verify that here).
8289 	 */
8290 	sel = sldt();
8291 	report(!sel, "Expected 0 for L1 LDTR selector (got %x)", sel);
8292 }
8293 
8294 static void vmx_single_vmcall_guest(void)
8295 {
8296 	vmcall();
8297 }
8298 
8299 static void vmx_cr_load_test(void)
8300 {
8301 	unsigned long cr3, cr4, orig_cr3, orig_cr4;
8302 	u32 ctrls[2] = {0};
8303 	pgd_t *pml5;
8304 
8305 	orig_cr4 = read_cr4();
8306 	orig_cr3 = read_cr3();
8307 
8308 	if (!this_cpu_has(X86_FEATURE_PCID)) {
8309 		report_skip("%s : PCID not detected", __func__);
8310 		return;
8311 	}
8312 	if (!this_cpu_has(X86_FEATURE_MCE)) {
8313 		report_skip("%s : MCE not detected", __func__);
8314 		return;
8315 	}
8316 
8317 	TEST_ASSERT(!(orig_cr3 & X86_CR3_PCID_MASK));
8318 
8319 	/* Enable PCID for L1. */
8320 	cr4 = orig_cr4 | X86_CR4_PCIDE;
8321 	cr3 = orig_cr3 | 0x1;
8322 	TEST_ASSERT(!write_cr4_safe(cr4));
8323 	write_cr3(cr3);
8324 
8325 	test_set_guest(vmx_single_vmcall_guest);
8326 	vmcs_write(HOST_CR4, cr4);
8327 	vmcs_write(HOST_CR3, cr3);
8328 	enter_guest();
8329 
8330 	/*
8331 	 * No exception is expected.
8332 	 *
8333 	 * NB. KVM loads the last guest write to CR4 into CR4 read
8334 	 *     shadow. In order to trigger an exit to KVM, we can toggle a
8335 	 *     bit that is owned by KVM. We use CR4.MCE, which shall
8336 	 *     have no side effect because normally no guest MCE (e.g., as the
8337 	 *     result of bad memory) would happen during this test.
8338 	 */
8339 	TEST_ASSERT(!write_cr4_safe(cr4 ^ X86_CR4_MCE));
8340 
8341 	/* Cleanup L1 state. */
8342 	write_cr3(orig_cr3);
8343 	TEST_ASSERT(!write_cr4_safe(orig_cr4));
8344 
8345 	if (!this_cpu_has(X86_FEATURE_LA57))
8346 		goto done;
8347 
8348 	/*
8349 	 * Allocate a full page for PML5 to guarantee alignment, though only
8350 	 * the first entry needs to be filled (the test's virtual addresses
8351 	 * most definitely do not have any of bits 56:48 set).
8352 	 */
8353 	pml5 = alloc_page();
8354 	*pml5 = orig_cr3 | PT_PRESENT_MASK | PT_WRITABLE_MASK;
8355 
8356 	/*
8357 	 * Transition to/from 5-level paging in the host via VM-Exit.  CR4.LA57
8358 	 * can't be toggled while long is active via MOV CR4, but there are no
8359 	 * such restrictions on VM-Exit.
8360 	 */
8361 lol_5level:
8362 	vmcs_write(HOST_CR4, orig_cr4 | X86_CR4_LA57);
8363 	vmcs_write(HOST_CR3, virt_to_phys(pml5));
8364 	enter_guest();
8365 
8366 	/*
8367 	 * VMREAD with a memory operand to verify KVM detects the LA57 change,
8368 	 * e.g. uses the correct guest root level in gva_to_gpa().
8369 	 */
8370 	TEST_ASSERT(vmcs_readm(HOST_CR3) == virt_to_phys(pml5));
8371 	TEST_ASSERT(vmcs_readm(HOST_CR4) == (orig_cr4 | X86_CR4_LA57));
8372 
8373 	vmcs_write(HOST_CR4, orig_cr4);
8374 	vmcs_write(HOST_CR3, orig_cr3);
8375 	enter_guest();
8376 
8377 	TEST_ASSERT(vmcs_readm(HOST_CR3) == orig_cr3);
8378 	TEST_ASSERT(vmcs_readm(HOST_CR4) == orig_cr4);
8379 
8380 	/*
8381 	 * And now do the same LA57 shenanigans with EPT enabled.  KVM uses
8382 	 * two separate MMUs when L1 uses TDP, whereas the above shadow paging
8383 	 * version shares an MMU between L1 and L2.
8384 	 *
8385 	 * If the saved execution controls are non-zero then the EPT version
8386 	 * has already run.  In that case, restore the old controls.  If EPT
8387 	 * setup fails, e.g. EPT isn't supported, fall through and finish up.
8388 	 */
8389 	if (ctrls[0]) {
8390 		vmcs_write(CPU_EXEC_CTRL0, ctrls[0]);
8391 		vmcs_write(CPU_EXEC_CTRL1, ctrls[1]);
8392 	} else if (!setup_ept(false)) {
8393 		ctrls[0] = vmcs_read(CPU_EXEC_CTRL0);
8394 		ctrls[1]  = vmcs_read(CPU_EXEC_CTRL1);
8395 		goto lol_5level;
8396 	}
8397 
8398 	free_page(pml5);
8399 
8400 done:
8401 	skip_exit_vmcall();
8402 	enter_guest();
8403 }
8404 
8405 static void vmx_cr4_osxsave_test_guest(void)
8406 {
8407 	write_cr4(read_cr4() & ~X86_CR4_OSXSAVE);
8408 }
8409 
8410 /*
8411  * Ensure that kvm recalculates the L1 guest's CPUID.01H:ECX.OSXSAVE
8412  * after VM-exit from an L2 guest that sets CR4.OSXSAVE to a different
8413  * value than in L1.
8414  */
8415 static void vmx_cr4_osxsave_test(void)
8416 {
8417 	if (!this_cpu_has(X86_FEATURE_XSAVE)) {
8418 		report_skip("%s : XSAVE not detected", __func__);
8419 		return;
8420 	}
8421 
8422 	if (!(read_cr4() & X86_CR4_OSXSAVE)) {
8423 		unsigned long cr4 = read_cr4() | X86_CR4_OSXSAVE;
8424 
8425 		write_cr4(cr4);
8426 		vmcs_write(GUEST_CR4, cr4);
8427 		vmcs_write(HOST_CR4, cr4);
8428 	}
8429 
8430 	TEST_ASSERT(this_cpu_has(X86_FEATURE_OSXSAVE));
8431 
8432 	test_set_guest(vmx_cr4_osxsave_test_guest);
8433 	enter_guest();
8434 
8435 	TEST_ASSERT(this_cpu_has(X86_FEATURE_OSXSAVE));
8436 }
8437 
8438 static void vmx_nm_test_guest(void)
8439 {
8440 	write_cr0(read_cr0() | X86_CR0_TS);
8441 	asm volatile("fnop");
8442 }
8443 
8444 static void check_nm_exit(const char *test)
8445 {
8446 	u32 reason = vmcs_read(EXI_REASON);
8447 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
8448 	const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8449 		NM_VECTOR;
8450 
8451 	report(reason == VMX_EXC_NMI && intr_info == expected, "%s", test);
8452 }
8453 
8454 /*
8455  * This test checks that:
8456  *
8457  * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then
8458  *     a subsequent #NM VM-exit is reflected to L1.
8459  *
8460  * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a
8461  *     subsequent #NM VM-exit is reflected to L1.
8462  */
8463 static void vmx_nm_test(void)
8464 {
8465 	unsigned long cr0 = read_cr0();
8466 
8467 	test_set_guest(vmx_nm_test_guest);
8468 
8469 	/*
8470 	 * L1 wants to intercept #NM exceptions encountered in L2.
8471 	 */
8472 	vmcs_write(EXC_BITMAP, 1 << NM_VECTOR);
8473 
8474 	/*
8475 	 * Launch L2 with CR0.TS clear, but don't claim host ownership of
8476 	 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop,
8477 	 * which will raise #NM. L0 should reflect the #NM VM-exit to L1.
8478 	 */
8479 	vmcs_write(CR0_MASK, 0);
8480 	vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS);
8481 	enter_guest();
8482 	check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1");
8483 
8484 	/*
8485 	 * Re-enter L2 at the fnop instruction, with CR0.TS clear but
8486 	 * CR0.EM set. The fnop will still raise #NM, and L0 should
8487 	 * reflect the #NM VM-exit to L1.
8488 	 */
8489 	vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM);
8490 	enter_guest();
8491 	check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1");
8492 
8493 	/*
8494 	 * Re-enter L2 at the fnop instruction, with both CR0.TS and
8495 	 * CR0.EM clear. There will be no #NM, and the L2 guest should
8496 	 * exit normally.
8497 	 */
8498 	vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM));
8499 	enter_guest();
8500 }
8501 
8502 bool vmx_pending_event_ipi_fired;
8503 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
8504 {
8505 	vmx_pending_event_ipi_fired = true;
8506 	eoi();
8507 }
8508 
8509 bool vmx_pending_event_guest_run;
8510 static void vmx_pending_event_guest(void)
8511 {
8512 	vmcall();
8513 	vmx_pending_event_guest_run = true;
8514 }
8515 
8516 static void vmx_pending_event_test_core(bool guest_hlt)
8517 {
8518 	int ipi_vector = 0xf1;
8519 
8520 	vmx_pending_event_ipi_fired = false;
8521 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
8522 
8523 	vmx_pending_event_guest_run = false;
8524 	test_set_guest(vmx_pending_event_guest);
8525 
8526 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
8527 
8528 	enter_guest();
8529 	skip_exit_vmcall();
8530 
8531 	if (guest_hlt)
8532 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8533 
8534 	irq_disable();
8535 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
8536 				   APIC_DM_FIXED | ipi_vector,
8537 				   0);
8538 
8539 	enter_guest();
8540 
8541 	assert_exit_reason(VMX_EXTINT);
8542 	report(!vmx_pending_event_guest_run,
8543 	       "Guest did not run before host received IPI");
8544 
8545 	irq_enable();
8546 	asm volatile ("nop");
8547 	irq_disable();
8548 	report(vmx_pending_event_ipi_fired,
8549 	       "Got pending interrupt after IRQ enabled");
8550 
8551 	if (guest_hlt)
8552 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8553 
8554 	enter_guest();
8555 	report(vmx_pending_event_guest_run,
8556 	       "Guest finished running when no interrupt");
8557 }
8558 
8559 static void vmx_pending_event_test(void)
8560 {
8561 	vmx_pending_event_test_core(false);
8562 }
8563 
8564 static void vmx_pending_event_hlt_test(void)
8565 {
8566 	vmx_pending_event_test_core(true);
8567 }
8568 
8569 static int vmx_window_test_db_count;
8570 
8571 static void vmx_window_test_db_handler(struct ex_regs *regs)
8572 {
8573 	vmx_window_test_db_count++;
8574 }
8575 
8576 static void vmx_nmi_window_test_guest(void)
8577 {
8578 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8579 
8580 	asm volatile("vmcall\n\t"
8581 		     "nop\n\t");
8582 
8583 	handle_exception(DB_VECTOR, NULL);
8584 }
8585 
8586 static void verify_nmi_window_exit(u64 rip)
8587 {
8588 	u32 exit_reason = vmcs_read(EXI_REASON);
8589 
8590 	report(exit_reason == VMX_NMI_WINDOW,
8591 	       "Exit reason (%d) is 'NMI window'", exit_reason);
8592 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8593 	       vmcs_read(GUEST_RIP), rip);
8594 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8595 }
8596 
8597 static void vmx_nmi_window_test(void)
8598 {
8599 	u64 nop_addr;
8600 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8601 
8602 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
8603 		report_skip("%s : \"Virtual NMIs\" exec control not supported", __func__);
8604 		return;
8605 	}
8606 
8607 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
8608 		report_skip("%s : \"NMI-window exiting\" exec control not supported", __func__);
8609 		return;
8610 	}
8611 
8612 	vmx_window_test_db_count = 0;
8613 
8614 	report_prefix_push("NMI-window");
8615 	test_set_guest(vmx_nmi_window_test_guest);
8616 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
8617 	enter_guest();
8618 	skip_exit_vmcall();
8619 	nop_addr = vmcs_read(GUEST_RIP);
8620 
8621 	/*
8622 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
8623 	 * RIP will not advance.
8624 	 */
8625 	report_prefix_push("active, no blocking");
8626 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8627 	enter_guest();
8628 	verify_nmi_window_exit(nop_addr);
8629 	report_prefix_pop();
8630 
8631 	/*
8632 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
8633 	 * a VM-exit on the next instruction after the nop. (The nop
8634 	 * is one byte.)
8635 	 */
8636 	report_prefix_push("active, blocking by MOV-SS");
8637 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8638 	enter_guest();
8639 	verify_nmi_window_exit(nop_addr + 1);
8640 	report_prefix_pop();
8641 
8642 	/*
8643 	 * Ask for "NMI-window exiting" (with event injection), and
8644 	 * expect a VM-exit after the event is injected. (RIP should
8645 	 * be at the address specified in the IDT entry for #DB.)
8646 	 */
8647 	report_prefix_push("active, no blocking, injecting #DB");
8648 	vmcs_write(ENT_INTR_INFO,
8649 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8650 	enter_guest();
8651 	verify_nmi_window_exit((u64)db_fault_addr);
8652 	report_prefix_pop();
8653 
8654 	/*
8655 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
8656 	 * a VM-exit after the next IRET (i.e. after the #DB handler
8657 	 * returns). So, RIP should be back at one byte past the nop.
8658 	 */
8659 	report_prefix_push("active, blocking by NMI");
8660 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
8661 	enter_guest();
8662 	verify_nmi_window_exit(nop_addr + 1);
8663 	report(vmx_window_test_db_count == 1,
8664 	       "#DB handler executed once (actual %d times)",
8665 	       vmx_window_test_db_count);
8666 	report_prefix_pop();
8667 
8668 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8669 		report_skip("CPU does not support activity state HLT.");
8670 	} else {
8671 		/*
8672 		 * Ask for "NMI-window exiting" when entering activity
8673 		 * state HLT, and expect an immediate VM-exit. RIP is
8674 		 * still one byte past the nop.
8675 		 */
8676 		report_prefix_push("halted, no blocking");
8677 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8678 		enter_guest();
8679 		verify_nmi_window_exit(nop_addr + 1);
8680 		report_prefix_pop();
8681 
8682 		/*
8683 		 * Ask for "NMI-window exiting" when entering activity
8684 		 * state HLT (with event injection), and expect a
8685 		 * VM-exit after the event is injected. (RIP should be
8686 		 * at the address specified in the IDT entry for #DB.)
8687 		 */
8688 		report_prefix_push("halted, no blocking, injecting #DB");
8689 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8690 		vmcs_write(ENT_INTR_INFO,
8691 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8692 			   DB_VECTOR);
8693 		enter_guest();
8694 		verify_nmi_window_exit((u64)db_fault_addr);
8695 		report_prefix_pop();
8696 	}
8697 
8698 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8699 	enter_guest();
8700 	report_prefix_pop();
8701 }
8702 
8703 static void vmx_intr_window_test_guest(void)
8704 {
8705 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8706 
8707 	/*
8708 	 * The two consecutive STIs are to ensure that only the first
8709 	 * one has a shadow. Note that NOP and STI are one byte
8710 	 * instructions.
8711 	 */
8712 	asm volatile("vmcall\n\t"
8713 		     "nop\n\t"
8714 		     "sti\n\t"
8715 		     "sti\n\t");
8716 
8717 	handle_exception(DB_VECTOR, NULL);
8718 }
8719 
8720 static void verify_intr_window_exit(u64 rip)
8721 {
8722 	u32 exit_reason = vmcs_read(EXI_REASON);
8723 
8724 	report(exit_reason == VMX_INTR_WINDOW,
8725 	       "Exit reason (%d) is 'interrupt window'", exit_reason);
8726 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8727 	       vmcs_read(GUEST_RIP), rip);
8728 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8729 }
8730 
8731 static void vmx_intr_window_test(void)
8732 {
8733 	u64 vmcall_addr;
8734 	u64 nop_addr;
8735 	unsigned int orig_db_gate_type;
8736 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8737 
8738 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
8739 		report_skip("%s : \"Interrupt-window exiting\" exec control not supported", __func__);
8740 		return;
8741 	}
8742 
8743 	/*
8744 	 * Change the IDT entry for #DB from interrupt gate to trap gate,
8745 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
8746 	 * be disabled after vectoring a #DB.
8747 	 */
8748 	orig_db_gate_type = boot_idt[DB_VECTOR].type;
8749 	boot_idt[DB_VECTOR].type = 15;
8750 
8751 	report_prefix_push("interrupt-window");
8752 	test_set_guest(vmx_intr_window_test_guest);
8753 	enter_guest();
8754 	assert_exit_reason(VMX_VMCALL);
8755 	vmcall_addr = vmcs_read(GUEST_RIP);
8756 
8757 	/*
8758 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
8759 	 * no blocking; expect an immediate VM-exit. Note that we have
8760 	 * not advanced past the vmcall instruction yet, so RIP should
8761 	 * point to the vmcall instruction.
8762 	 */
8763 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
8764 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8765 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
8766 	enter_guest();
8767 	verify_intr_window_exit(vmcall_addr);
8768 	report_prefix_pop();
8769 
8770 	/*
8771 	 * Ask for "interrupt-window exiting" (with event injection)
8772 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
8773 	 * the event is injected. That is, RIP should should be at the
8774 	 * address specified in the IDT entry for #DB.
8775 	 */
8776 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #DB");
8777 	vmcs_write(ENT_INTR_INFO,
8778 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8779 	vmcall_addr = vmcs_read(GUEST_RIP);
8780 	enter_guest();
8781 	verify_intr_window_exit((u64)db_fault_addr);
8782 	report_prefix_pop();
8783 
8784 	/*
8785 	 * Let the L2 guest run through the IRET, back to the VMCALL.
8786 	 * We have to clear the "interrupt-window exiting"
8787 	 * VM-execution control, or it would just keep causing
8788 	 * VM-exits. Then, advance past the VMCALL and set the
8789 	 * "interrupt-window exiting" VM-execution control again.
8790 	 */
8791 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8792 	enter_guest();
8793 	skip_exit_vmcall();
8794 	nop_addr = vmcs_read(GUEST_RIP);
8795 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8796 
8797 	/*
8798 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
8799 	 * RFLAGS.IF set, and expect a VM-exit on the next
8800 	 * instruction. (NOP is one byte.)
8801 	 */
8802 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
8803 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8804 	enter_guest();
8805 	verify_intr_window_exit(nop_addr + 1);
8806 	report_prefix_pop();
8807 
8808 	/*
8809 	 * Back up to the NOP and ask for "interrupt-window exiting"
8810 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
8811 	 * on the next instruction. (NOP is one byte.)
8812 	 */
8813 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
8814 	vmcs_write(GUEST_RIP, nop_addr);
8815 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
8816 	enter_guest();
8817 	verify_intr_window_exit(nop_addr + 1);
8818 	report_prefix_pop();
8819 
8820 	/*
8821 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
8822 	 * and expect a VM-exit on the instruction following the STI
8823 	 * shadow. Only the first STI (which is one byte past the NOP)
8824 	 * should have a shadow. The second STI (which is two bytes
8825 	 * past the NOP) has no shadow. Therefore, the interrupt
8826 	 * window opens at three bytes past the NOP.
8827 	 */
8828 	report_prefix_push("active, RFLAGS.IF = 0");
8829 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8830 	enter_guest();
8831 	verify_intr_window_exit(nop_addr + 3);
8832 	report_prefix_pop();
8833 
8834 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8835 		report_skip("CPU does not support activity state HLT.");
8836 	} else {
8837 		/*
8838 		 * Ask for "interrupt-window exiting" when entering
8839 		 * activity state HLT, and expect an immediate
8840 		 * VM-exit. RIP is still three bytes past the nop.
8841 		 */
8842 		report_prefix_push("halted, no blocking");
8843 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8844 		enter_guest();
8845 		verify_intr_window_exit(nop_addr + 3);
8846 		report_prefix_pop();
8847 
8848 		/*
8849 		 * Ask for "interrupt-window exiting" when entering
8850 		 * activity state HLT (with event injection), and
8851 		 * expect a VM-exit after the event is injected. That
8852 		 * is, RIP should should be at the address specified
8853 		 * in the IDT entry for #DB.
8854 		 */
8855 		report_prefix_push("halted, no blocking, injecting #DB");
8856 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8857 		vmcs_write(ENT_INTR_INFO,
8858 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8859 			   DB_VECTOR);
8860 		enter_guest();
8861 		verify_intr_window_exit((u64)db_fault_addr);
8862 		report_prefix_pop();
8863 	}
8864 
8865 	boot_idt[DB_VECTOR].type = orig_db_gate_type;
8866 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8867 	enter_guest();
8868 	report_prefix_pop();
8869 }
8870 
8871 #define GUEST_TSC_OFFSET (1u << 30)
8872 
8873 static u64 guest_tsc;
8874 
8875 static void vmx_store_tsc_test_guest(void)
8876 {
8877 	guest_tsc = rdtsc();
8878 }
8879 
8880 /*
8881  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
8882  * list, the value saved is not subject to the TSC offset that is
8883  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
8884  */
8885 static void vmx_store_tsc_test(void)
8886 {
8887 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
8888 	u64 low, high;
8889 
8890 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
8891 		report_skip("%s : \"Use TSC offsetting\" exec control not supported", __func__);
8892 		return;
8893 	}
8894 
8895 	test_set_guest(vmx_store_tsc_test_guest);
8896 
8897 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
8898 	vmcs_write(EXI_MSR_ST_CNT, 1);
8899 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
8900 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
8901 
8902 	low = rdtsc();
8903 	enter_guest();
8904 	high = rdtsc();
8905 
8906 	report(low + GUEST_TSC_OFFSET <= guest_tsc &&
8907 	       guest_tsc <= high + GUEST_TSC_OFFSET,
8908 	       "RDTSC value in the guest (%lu) is in range [%lu, %lu]",
8909 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
8910 	report(low <= msr_entry.value && msr_entry.value <= high,
8911 	       "IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
8912 	       msr_entry.value, low, high);
8913 }
8914 
8915 static void vmx_preemption_timer_zero_test_db_handler(struct ex_regs *regs)
8916 {
8917 }
8918 
8919 static void vmx_preemption_timer_zero_test_guest(void)
8920 {
8921 	while (vmx_get_test_stage() < 3)
8922 		vmcall();
8923 }
8924 
8925 static void vmx_preemption_timer_zero_activate_preemption_timer(void)
8926 {
8927 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
8928 	vmcs_write(PREEMPT_TIMER_VALUE, 0);
8929 }
8930 
8931 static void vmx_preemption_timer_zero_advance_past_vmcall(void)
8932 {
8933 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
8934 	enter_guest();
8935 	skip_exit_vmcall();
8936 }
8937 
8938 static void vmx_preemption_timer_zero_inject_db(bool intercept_db)
8939 {
8940 	vmx_preemption_timer_zero_activate_preemption_timer();
8941 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK |
8942 		   INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8943 	vmcs_write(EXC_BITMAP, intercept_db ? 1 << DB_VECTOR : 0);
8944 	enter_guest();
8945 }
8946 
8947 static void vmx_preemption_timer_zero_set_pending_dbg(u32 exception_bitmap)
8948 {
8949 	vmx_preemption_timer_zero_activate_preemption_timer();
8950 	vmcs_write(GUEST_PENDING_DEBUG, PENDING_DBG_TRAP | DR6_TRAP1);
8951 	vmcs_write(EXC_BITMAP, exception_bitmap);
8952 	enter_guest();
8953 }
8954 
8955 static void vmx_preemption_timer_zero_expect_preempt_at_rip(u64 expected_rip)
8956 {
8957 	u32 reason = (u32)vmcs_read(EXI_REASON);
8958 	u64 guest_rip = vmcs_read(GUEST_RIP);
8959 
8960 	report(reason == VMX_PREEMPT && guest_rip == expected_rip,
8961 	       "Exit reason is 0x%x (expected 0x%x) and guest RIP is %lx (0x%lx expected).",
8962 	       reason, VMX_PREEMPT, guest_rip, expected_rip);
8963 }
8964 
8965 /*
8966  * This test ensures that when the VMX preemption timer is zero at
8967  * VM-entry, a VM-exit occurs after any event injection and after any
8968  * pending debug exceptions are raised, but before execution of any
8969  * guest instructions.
8970  */
8971 static void vmx_preemption_timer_zero_test(void)
8972 {
8973 	u64 db_fault_address = (u64)get_idt_addr(&boot_idt[DB_VECTOR]);
8974 	handler old_db;
8975 	u32 reason;
8976 
8977 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
8978 		report_skip("%s : \"Activate VMX-preemption timer\" pin control not supported", __func__);
8979 		return;
8980 	}
8981 
8982 	/*
8983 	 * Install a custom #DB handler that doesn't abort.
8984 	 */
8985 	old_db = handle_exception(DB_VECTOR,
8986 				  vmx_preemption_timer_zero_test_db_handler);
8987 
8988 	test_set_guest(vmx_preemption_timer_zero_test_guest);
8989 
8990 	/*
8991 	 * VMX-preemption timer should fire after event injection.
8992 	 */
8993 	vmx_set_test_stage(0);
8994 	vmx_preemption_timer_zero_inject_db(0);
8995 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8996 	vmx_preemption_timer_zero_advance_past_vmcall();
8997 
8998 	/*
8999 	 * VMX-preemption timer should fire after event injection.
9000 	 * Exception bitmap is irrelevant, since you can't intercept
9001 	 * an event that you injected.
9002 	 */
9003 	vmx_set_test_stage(1);
9004 	vmx_preemption_timer_zero_inject_db(true);
9005 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
9006 	vmx_preemption_timer_zero_advance_past_vmcall();
9007 
9008 	/*
9009 	 * VMX-preemption timer should fire after pending debug exceptions
9010 	 * have delivered a #DB trap.
9011 	 */
9012 	vmx_set_test_stage(2);
9013 	vmx_preemption_timer_zero_set_pending_dbg(0);
9014 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
9015 	vmx_preemption_timer_zero_advance_past_vmcall();
9016 
9017 	/*
9018 	 * VMX-preemption timer would fire after pending debug exceptions
9019 	 * have delivered a #DB trap, but in this case, the #DB trap is
9020 	 * intercepted.
9021 	 */
9022 	vmx_set_test_stage(3);
9023 	vmx_preemption_timer_zero_set_pending_dbg(1 << DB_VECTOR);
9024 	reason = (u32)vmcs_read(EXI_REASON);
9025 	report(reason == VMX_EXC_NMI, "Exit reason is 0x%x (expected 0x%x)",
9026 	       reason, VMX_EXC_NMI);
9027 
9028 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9029 	enter_guest();
9030 
9031 	handle_exception(DB_VECTOR, old_db);
9032 }
9033 
9034 static u64 vmx_preemption_timer_tf_test_prev_rip;
9035 
9036 static void vmx_preemption_timer_tf_test_db_handler(struct ex_regs *regs)
9037 {
9038 	extern char vmx_preemption_timer_tf_test_endloop;
9039 
9040 	if (vmx_get_test_stage() == 2) {
9041 		/*
9042 		 * Stage 2 means that we're done, one way or another.
9043 		 * Arrange for the iret to drop us out of the wbinvd
9044 		 * loop and stop single-stepping.
9045 		 */
9046 		regs->rip = (u64)&vmx_preemption_timer_tf_test_endloop;
9047 		regs->rflags &= ~X86_EFLAGS_TF;
9048 	} else if (regs->rip == vmx_preemption_timer_tf_test_prev_rip) {
9049 		/*
9050 		 * The RIP should alternate between the wbinvd and the
9051 		 * jmp instruction in the code below. If we ever see
9052 		 * the same instruction twice in a row, that means a
9053 		 * single-step trap has been dropped. Let the
9054 		 * hypervisor know about the failure by executing a
9055 		 * VMCALL.
9056 		 */
9057 		vmcall();
9058 	}
9059 	vmx_preemption_timer_tf_test_prev_rip = regs->rip;
9060 }
9061 
9062 static void vmx_preemption_timer_tf_test_guest(void)
9063 {
9064 	/*
9065 	 * The hypervisor doesn't intercept WBINVD, so the loop below
9066 	 * shouldn't be a problem--it's just two instructions
9067 	 * executing in VMX non-root mode. However, when the
9068 	 * hypervisor is running in a virtual environment, the parent
9069 	 * hypervisor might intercept WBINVD and emulate it. If the
9070 	 * parent hypervisor is broken, the single-step trap after the
9071 	 * WBINVD might be lost.
9072 	 */
9073 	asm volatile("vmcall\n\t"
9074 		     "0: wbinvd\n\t"
9075 		     "1: jmp 0b\n\t"
9076 		     "vmx_preemption_timer_tf_test_endloop:");
9077 }
9078 
9079 /*
9080  * Ensure that the delivery of a "VMX-preemption timer expired"
9081  * VM-exit doesn't disrupt single-stepping in the guest. Note that
9082  * passing this test doesn't ensure correctness, because the test will
9083  * only fail if the VMX-preemtion timer fires at the right time (or
9084  * the wrong time, as it were).
9085  */
9086 static void vmx_preemption_timer_tf_test(void)
9087 {
9088 	handler old_db;
9089 	u32 reason;
9090 	int i;
9091 
9092 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9093 		report_skip("%s : \"Activate VMX-preemption timer\" pin control not supported", __func__);
9094 		return;
9095 	}
9096 
9097 	old_db = handle_exception(DB_VECTOR,
9098 				  vmx_preemption_timer_tf_test_db_handler);
9099 
9100 	test_set_guest(vmx_preemption_timer_tf_test_guest);
9101 
9102 	enter_guest();
9103 	skip_exit_vmcall();
9104 
9105 	vmx_set_test_stage(1);
9106 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9107 	vmcs_write(PREEMPT_TIMER_VALUE, 50000);
9108 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9109 
9110 	/*
9111 	 * The only exit we should see is "VMX-preemption timer
9112 	 * expired."  If we get a VMCALL exit, that means the #DB
9113 	 * handler has detected a missing single-step trap. It doesn't
9114 	 * matter where the guest RIP is when the VMX-preemption timer
9115 	 * expires (whether it's in the WBINVD loop or in the #DB
9116 	 * handler)--a single-step trap should never be discarded.
9117 	 */
9118 	for (i = 0; i < 10000; i++) {
9119 		enter_guest();
9120 		reason = (u32)vmcs_read(EXI_REASON);
9121 		if (reason == VMX_PREEMPT)
9122 			continue;
9123 		TEST_ASSERT(reason == VMX_VMCALL);
9124 		skip_exit_insn();
9125 		break;
9126 	}
9127 
9128 	report(reason == VMX_PREEMPT, "No single-step traps skipped");
9129 
9130 	vmx_set_test_stage(2);
9131 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9132 	enter_guest();
9133 
9134 	handle_exception(DB_VECTOR, old_db);
9135 }
9136 
9137 #define VMX_PREEMPTION_TIMER_EXPIRY_CYCLES 1000000
9138 
9139 static u64 vmx_preemption_timer_expiry_start;
9140 static u64 vmx_preemption_timer_expiry_finish;
9141 
9142 static void vmx_preemption_timer_expiry_test_guest(void)
9143 {
9144 	vmcall();
9145 	vmx_preemption_timer_expiry_start = fenced_rdtsc();
9146 
9147 	while (vmx_get_test_stage() == 0)
9148 		vmx_preemption_timer_expiry_finish = fenced_rdtsc();
9149 }
9150 
9151 /*
9152  * Test that the VMX-preemption timer is not excessively delayed.
9153  *
9154  * Per the SDM, volume 3, VM-entry starts the VMX-preemption timer
9155  * with the unsigned value in the VMX-preemption timer-value field,
9156  * and the VMX-preemption timer counts down by 1 every time bit X in
9157  * the TSC changes due to a TSC increment (where X is
9158  * IA32_VMX_MISC[4:0]). If the timer counts down to zero in any state
9159  * other than the wait-for-SIPI state, the logical processor
9160  * transitions to the C0 C-state and causes a VM-exit.
9161  *
9162  * The guest code above reads the starting TSC after VM-entry. At this
9163  * point, the VMX-preemption timer has already been activated. Next,
9164  * the guest code reads the current TSC in a loop, storing the value
9165  * read to memory.
9166  *
9167  * If the RDTSC in the loop reads a value past the VMX-preemption
9168  * timer deadline, then the VMX-preemption timer VM-exit must be
9169  * delivered before the next instruction retires. Even if a higher
9170  * priority SMI is delivered first, the VMX-preemption timer VM-exit
9171  * must be delivered before the next instruction retires. Hence, a TSC
9172  * value past the VMX-preemption timer deadline might be read, but it
9173  * cannot be stored. If a TSC value past the deadline *is* stored,
9174  * then the architectural specification has been violated.
9175  */
9176 static void vmx_preemption_timer_expiry_test(void)
9177 {
9178 	u32 preemption_timer_value;
9179 	union vmx_misc misc;
9180 	u64 tsc_deadline;
9181 	u32 reason;
9182 
9183 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9184 		report_skip("%s : \"Activate VMX-preemption timer\" pin control not supported", __func__);
9185 		return;
9186 	}
9187 
9188 	test_set_guest(vmx_preemption_timer_expiry_test_guest);
9189 
9190 	enter_guest();
9191 	skip_exit_vmcall();
9192 
9193 	misc.val = rdmsr(MSR_IA32_VMX_MISC);
9194 	preemption_timer_value =
9195 		VMX_PREEMPTION_TIMER_EXPIRY_CYCLES >> misc.pt_bit;
9196 
9197 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9198 	vmcs_write(PREEMPT_TIMER_VALUE, preemption_timer_value);
9199 	vmx_set_test_stage(0);
9200 
9201 	enter_guest();
9202 	reason = (u32)vmcs_read(EXI_REASON);
9203 	TEST_ASSERT(reason == VMX_PREEMPT);
9204 
9205 	tsc_deadline = ((vmx_preemption_timer_expiry_start >> misc.pt_bit) <<
9206 			misc.pt_bit) + (preemption_timer_value << misc.pt_bit);
9207 
9208 	report(vmx_preemption_timer_expiry_finish < tsc_deadline,
9209 	       "Last stored guest TSC (%lu) < TSC deadline (%lu)",
9210 	       vmx_preemption_timer_expiry_finish, tsc_deadline);
9211 
9212 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9213 	vmx_set_test_stage(1);
9214 	enter_guest();
9215 }
9216 
9217 static void vmx_db_test_guest(void)
9218 {
9219 	/*
9220 	 * For a hardware generated single-step #DB.
9221 	 */
9222 	asm volatile("vmcall;"
9223 		     "nop;"
9224 		     ".Lpost_nop:");
9225 	/*
9226 	 * ...in a MOVSS shadow, with pending debug exceptions.
9227 	 */
9228 	asm volatile("vmcall;"
9229 		     "nop;"
9230 		     ".Lpost_movss_nop:");
9231 	/*
9232 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
9233 	 * emulates it in software.)
9234 	 */
9235 	asm volatile("vmcall;"
9236 		     "wbinvd;"
9237 		     ".Lpost_wbinvd:");
9238 	/*
9239 	 * ...in a MOVSS shadow, with pending debug exceptions.
9240 	 */
9241 	asm volatile("vmcall;"
9242 		     "wbinvd;"
9243 		     ".Lpost_movss_wbinvd:");
9244 	/*
9245 	 * For a hardware generated single-step #DB in a transactional region.
9246 	 */
9247 	asm volatile("vmcall;"
9248 		     ".Lxbegin: xbegin .Lskip_rtm;"
9249 		     "xend;"
9250 		     ".Lskip_rtm:");
9251 }
9252 
9253 /*
9254  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
9255  * L2. No #DB is delivered and L2 continues to the next point of
9256  * interest.
9257  */
9258 static void dismiss_db(void)
9259 {
9260 	vmcs_write(GUEST_PENDING_DEBUG, 0);
9261 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
9262 	enter_guest();
9263 }
9264 
9265 /*
9266  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
9267  * Then throw away the #DB exception and resume L2.
9268  */
9269 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
9270 			  void *expected_rip, u64 expected_exit_qual,
9271 			  u64 expected_dr6)
9272 {
9273 	u32 reason = vmcs_read(EXI_REASON);
9274 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
9275 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
9276 	u64 guest_rip = vmcs_read(GUEST_RIP);
9277 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
9278 	u64 dr6 = read_dr6();
9279 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
9280 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
9281 
9282 	report(reason == VMX_EXC_NMI && intr_info == expected_intr_info,
9283 	       "Expected #DB VM-exit");
9284 	report((u64)expected_rip == guest_rip, "Expected RIP %p (actual %lx)",
9285 	       expected_rip, guest_rip);
9286 	report_xfail(xfail_pdbg, 0 == guest_pending_dbg,
9287 		     "Expected pending debug exceptions 0 (actual %lx)",
9288 		     guest_pending_dbg);
9289 	report_xfail(xfail_qual, expected_exit_qual == exit_qual,
9290 		     "Expected exit qualification %lx (actual %lx)",
9291 		     expected_exit_qual, exit_qual);
9292 	report_xfail(xfail_dr6, expected_dr6 == dr6,
9293 		     "Expected DR6 %lx (actual %lx)", expected_dr6, dr6);
9294 	dismiss_db();
9295 }
9296 
9297 /*
9298  * Assuming the guest has just exited on a VMCALL instruction, skip
9299  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
9300  * pending debug exceptions are non-zero, set the VMCS up as if the
9301  * previous instruction was a MOVSS that generated the indicated
9302  * pending debug exceptions. Then enter L2.
9303  */
9304 static void single_step_guest(const char *test_name, u64 starting_dr6,
9305 			      u64 pending_debug_exceptions)
9306 {
9307 	printf("\n%s\n", test_name);
9308 	skip_exit_vmcall();
9309 	write_dr6(starting_dr6);
9310 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9311 	if (pending_debug_exceptions) {
9312 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
9313 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
9314 	}
9315 	enter_guest();
9316 }
9317 
9318 /*
9319  * When L1 intercepts #DB, verify that a single-step trap clears
9320  * pending debug exceptions, populates the exit qualification field
9321  * properly, and that DR6 is not prematurely clobbered. In a
9322  * (simulated) MOVSS shadow, make sure that the pending debug
9323  * exception bits are properly accumulated into the exit qualification
9324  * field.
9325  */
9326 static void vmx_db_test(void)
9327 {
9328 	/*
9329 	 * We are going to set a few arbitrary bits in DR6 to verify that
9330 	 * (a) DR6 is not modified by an intercepted #DB, and
9331 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
9332          *     the exit qualification field for a subsequent #DB exception.
9333 	 */
9334 	const u64 starting_dr6 = DR6_ACTIVE_LOW | DR6_BS | DR6_TRAP3 | DR6_TRAP1;
9335 	extern char post_nop asm(".Lpost_nop");
9336 	extern char post_movss_nop asm(".Lpost_movss_nop");
9337 	extern char post_wbinvd asm(".Lpost_wbinvd");
9338 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
9339 	extern char xbegin asm(".Lxbegin");
9340 	extern char skip_rtm asm(".Lskip_rtm");
9341 
9342 	/*
9343 	 * L1 wants to intercept #DB exceptions encountered in L2.
9344 	 */
9345 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
9346 
9347 	/*
9348 	 * Start L2 and run it up to the first point of interest.
9349 	 */
9350 	test_set_guest(vmx_db_test_guest);
9351 	enter_guest();
9352 
9353 	/*
9354 	 * Hardware-delivered #DB trap for single-step sets the
9355 	 * standard that L0 has to follow for emulated instructions.
9356 	 */
9357 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
9358 	check_db_exit(false, false, false, &post_nop, DR6_BS, starting_dr6);
9359 
9360 	/*
9361 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
9362 	 * also sets the standard that L0 has to follow for emulated
9363 	 * instructions. Here, we establish the VMCS pending debug
9364 	 * exceptions to indicate that the simulated MOVSS triggered a
9365 	 * data breakpoint as well as the single-step trap.
9366 	 */
9367 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
9368 			  starting_dr6, DR6_BS | PENDING_DBG_TRAP | DR6_TRAP0);
9369 	check_db_exit(false, false, false, &post_movss_nop, DR6_BS | DR6_TRAP0,
9370 		      starting_dr6);
9371 
9372 	/*
9373 	 * L0 synthesized #DB trap for single-step is buggy, because
9374 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
9375 	 * reconstitute the exit qualification from the prematurely
9376 	 * modified DR6, but fails miserably.
9377 	 */
9378 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
9379 	check_db_exit(false, false, false, &post_wbinvd, DR6_BS, starting_dr6);
9380 
9381 	/*
9382 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
9383 	 * even worse, because L0 also leaves the pending debug
9384 	 * exceptions in the VMCS instead of accumulating them into
9385 	 * the exit qualification field for the #DB exception.
9386 	 */
9387 	single_step_guest("Software synthesized single-step in MOVSS shadow",
9388 			  starting_dr6, DR6_BS | PENDING_DBG_TRAP | DR6_TRAP0);
9389 	check_db_exit(true, false, true, &post_movss_wbinvd, DR6_BS | DR6_TRAP0,
9390 		      starting_dr6);
9391 
9392 	/*
9393 	 * Optional RTM test for hardware that supports RTM, to
9394 	 * demonstrate that the current volume 3 of the SDM
9395 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
9396 	 * qualification for debug exceptions is not reserved. It is
9397 	 * set to 1 if a debug exception (#DB) or a breakpoint
9398 	 * exception (#BP) occurs inside an RTM region while advanced
9399 	 * debugging of RTM transactional regions is enabled.
9400 	 */
9401 	if (this_cpu_has(X86_FEATURE_RTM)) {
9402 		vmcs_write(ENT_CONTROLS,
9403 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
9404 		/*
9405 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
9406 		 * in the guest to enable advanced debugging of RTM
9407 		 * transactional regions.
9408 		 */
9409 		vmcs_write(GUEST_DR7, BIT(11));
9410 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
9411 		single_step_guest("Hardware delivered single-step in "
9412 				  "transactional region", starting_dr6, 0);
9413 		check_db_exit(false, false, false, &xbegin, BIT(16),
9414 			      starting_dr6);
9415 	} else {
9416 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
9417 		enter_guest();
9418 	}
9419 }
9420 
9421 static void enable_vid(void)
9422 {
9423 	void *virtual_apic_page;
9424 
9425 	assert(cpu_has_apicv());
9426 
9427 	disable_intercept_for_x2apic_msrs();
9428 
9429 	virtual_apic_page = alloc_page();
9430 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
9431 
9432 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
9433 
9434 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
9435 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
9436 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
9437 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
9438 
9439 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
9440 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
9441 }
9442 
9443 static void trigger_ioapic_scan_thread(void *data)
9444 {
9445 	/* Wait until other CPU entered L2 */
9446 	while (vmx_get_test_stage() != 1)
9447 		;
9448 
9449 	/* Trigger ioapic scan */
9450 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
9451 	vmx_set_test_stage(2);
9452 }
9453 
9454 static void irq_79_handler_guest(isr_regs_t *regs)
9455 {
9456 	eoi();
9457 
9458 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
9459 	vmcall();
9460 }
9461 
9462 /*
9463  * Constant for num of busy-loop iterations after which
9464  * a timer interrupt should have happened in host
9465  */
9466 #define TIMER_INTERRUPT_DELAY 100000000
9467 
9468 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
9469 {
9470 	handle_irq(0x79, irq_79_handler_guest);
9471 	irq_enable();
9472 
9473 	/* Signal to L1 CPU to trigger ioapic scan */
9474 	vmx_set_test_stage(1);
9475 	/* Wait until L1 CPU to trigger ioapic scan */
9476 	while (vmx_get_test_stage() != 2)
9477 		;
9478 
9479 	/*
9480 	 * Wait for L0 timer interrupt to be raised while we run in L2
9481 	 * such that L0 will process the IOAPIC scan request before
9482 	 * resuming L2
9483 	 */
9484 	delay(TIMER_INTERRUPT_DELAY);
9485 
9486 	asm volatile ("int $0x79");
9487 }
9488 
9489 static void vmx_eoi_bitmap_ioapic_scan_test(void)
9490 {
9491 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
9492 		report_skip("%s : Not all required APICv bits supported or CPU count < 2", __func__);
9493 		return;
9494 	}
9495 
9496 	enable_vid();
9497 
9498 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
9499 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
9500 
9501 	/*
9502 	 * Launch L2.
9503 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
9504 	 * In case the reason isn't VMX_VMCALL, the asserion inside
9505 	 * skip_exit_vmcall() will fail.
9506 	 */
9507 	enter_guest();
9508 	skip_exit_vmcall();
9509 
9510 	/* Let L2 finish */
9511 	enter_guest();
9512 	report_pass(__func__);
9513 }
9514 
9515 #define HLT_WITH_RVI_VECTOR		(0xf1)
9516 
9517 bool vmx_hlt_with_rvi_guest_isr_fired;
9518 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
9519 {
9520 	vmx_hlt_with_rvi_guest_isr_fired = true;
9521 	eoi();
9522 }
9523 
9524 static void vmx_hlt_with_rvi_guest(void)
9525 {
9526 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
9527 
9528 	irq_enable();
9529 	asm volatile ("nop");
9530 
9531 	vmcall();
9532 }
9533 
9534 static void vmx_hlt_with_rvi_test(void)
9535 {
9536 	if (!cpu_has_apicv()) {
9537 		report_skip("%s : Not all required APICv bits supported", __func__);
9538 		return;
9539 	}
9540 
9541 	enable_vid();
9542 
9543 	vmx_hlt_with_rvi_guest_isr_fired = false;
9544 	test_set_guest(vmx_hlt_with_rvi_guest);
9545 
9546 	enter_guest();
9547 	skip_exit_vmcall();
9548 
9549 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
9550 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
9551 	enter_guest();
9552 
9553 	report(vmx_hlt_with_rvi_guest_isr_fired, "Interrupt raised in guest");
9554 }
9555 
9556 static void set_irq_line_thread(void *data)
9557 {
9558 	/* Wait until other CPU entered L2 */
9559 	while (vmx_get_test_stage() != 1)
9560 		;
9561 
9562 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
9563 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9564 	vmx_set_test_stage(2);
9565 }
9566 
9567 static bool irq_78_handler_vmcall_before_eoi;
9568 static void irq_78_handler_guest(isr_regs_t *regs)
9569 {
9570 	set_irq_line(0xf, 0);
9571 	if (irq_78_handler_vmcall_before_eoi)
9572 		vmcall();
9573 	eoi();
9574 	vmcall();
9575 }
9576 
9577 static void vmx_apic_passthrough_guest(void)
9578 {
9579 	handle_irq(0x78, irq_78_handler_guest);
9580 	irq_enable();
9581 
9582 	/* If requested, wait for other CPU to trigger ioapic scan */
9583 	if (vmx_get_test_stage() < 1) {
9584 		vmx_set_test_stage(1);
9585 		while (vmx_get_test_stage() != 2)
9586 			;
9587 	}
9588 
9589 	set_irq_line(0xf, 1);
9590 }
9591 
9592 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
9593 {
9594 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
9595 		report_skip("%s : CPU count < 2", __func__);
9596 		return;
9597 	}
9598 
9599 	/* Test device is required for generating IRQs */
9600 	if (!test_device_enabled()) {
9601 		report_skip("%s : No test device enabled", __func__);
9602 		return;
9603 	}
9604 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9605 	u64 cpu_ctrl_1 = 0;
9606 
9607 	disable_intercept_for_x2apic_msrs();
9608 
9609 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9610 
9611 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9612 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9613 
9614 	if (set_irq_line_from_thread) {
9615 		irq_78_handler_vmcall_before_eoi = false;
9616 		on_cpu_async(1, set_irq_line_thread, NULL);
9617 	} else {
9618 		irq_78_handler_vmcall_before_eoi = true;
9619 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9620 		vmx_set_test_stage(2);
9621 	}
9622 	test_set_guest(vmx_apic_passthrough_guest);
9623 
9624 	if (irq_78_handler_vmcall_before_eoi) {
9625 		/* Before EOI remote_irr should still be set */
9626 		enter_guest();
9627 		skip_exit_vmcall();
9628 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
9629 			"IOAPIC pass-through: remote_irr=1 before EOI");
9630 	}
9631 
9632 	/* After EOI remote_irr should be cleared */
9633 	enter_guest();
9634 	skip_exit_vmcall();
9635 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
9636 		"IOAPIC pass-through: remote_irr=0 after EOI");
9637 
9638 	/* Let L2 finish */
9639 	enter_guest();
9640 	report_pass(__func__);
9641 }
9642 
9643 static void vmx_apic_passthrough_test(void)
9644 {
9645 	vmx_apic_passthrough(false);
9646 }
9647 
9648 static void vmx_apic_passthrough_thread_test(void)
9649 {
9650 	vmx_apic_passthrough(true);
9651 }
9652 
9653 static void vmx_apic_passthrough_tpr_threshold_guest(void)
9654 {
9655 	cli();
9656 	apic_set_tpr(0);
9657 }
9658 
9659 static bool vmx_apic_passthrough_tpr_threshold_ipi_isr_fired;
9660 static void vmx_apic_passthrough_tpr_threshold_ipi_isr(isr_regs_t *regs)
9661 {
9662 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = true;
9663 	eoi();
9664 }
9665 
9666 static void vmx_apic_passthrough_tpr_threshold_test(void)
9667 {
9668 	int ipi_vector = 0xe1;
9669 
9670 	disable_intercept_for_x2apic_msrs();
9671 	vmcs_clear_bits(PIN_CONTROLS, PIN_EXTINT);
9672 
9673 	/* Raise L0 TPR-threshold by queueing vector in LAPIC IRR */
9674 	cli();
9675 	apic_set_tpr((ipi_vector >> 4) + 1);
9676 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
9677 			APIC_DM_FIXED | ipi_vector,
9678 			0);
9679 
9680 	test_set_guest(vmx_apic_passthrough_tpr_threshold_guest);
9681 	enter_guest();
9682 
9683 	report(apic_get_tpr() == 0, "TPR was zero by guest");
9684 
9685 	/* Clean pending self-IPI */
9686 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = false;
9687 	handle_irq(ipi_vector, vmx_apic_passthrough_tpr_threshold_ipi_isr);
9688 	sti();
9689 	asm volatile ("nop");
9690 	report(vmx_apic_passthrough_tpr_threshold_ipi_isr_fired, "self-IPI fired");
9691 
9692 	report_pass(__func__);
9693 }
9694 
9695 static u64 init_signal_test_exit_reason;
9696 static bool init_signal_test_thread_continued;
9697 
9698 static void init_signal_test_thread(void *data)
9699 {
9700 	struct vmcs *test_vmcs = data;
9701 
9702 	/* Enter VMX operation (i.e. exec VMXON) */
9703 	u64 *ap_vmxon_region = alloc_page();
9704 	enable_vmx();
9705 	init_vmx(ap_vmxon_region);
9706 	TEST_ASSERT(!__vmxon_safe(ap_vmxon_region));
9707 
9708 	/* Signal CPU have entered VMX operation */
9709 	vmx_set_test_stage(1);
9710 
9711 	/* Wait for BSP CPU to send INIT signal */
9712 	while (vmx_get_test_stage() != 2)
9713 		;
9714 
9715 	/*
9716 	 * Signal that we continue as usual as INIT signal
9717 	 * should be blocked while CPU is in VMX operation
9718 	 */
9719 	vmx_set_test_stage(3);
9720 
9721 	/* Wait for signal to enter VMX non-root mode */
9722 	while (vmx_get_test_stage() != 4)
9723 		;
9724 
9725 	/* Enter VMX non-root mode */
9726 	test_set_guest(v2_null_test_guest);
9727 	make_vmcs_current(test_vmcs);
9728 	enter_guest();
9729 	/* Save exit reason for BSP CPU to compare to expected result */
9730 	init_signal_test_exit_reason = vmcs_read(EXI_REASON);
9731 	/* VMCLEAR test-vmcs so it could be loaded by BSP CPU */
9732 	vmcs_clear(test_vmcs);
9733 	launched = false;
9734 	/* Signal that CPU exited to VMX root mode */
9735 	vmx_set_test_stage(5);
9736 
9737 	/* Wait for BSP CPU to signal to exit VMX operation */
9738 	while (vmx_get_test_stage() != 6)
9739 		;
9740 
9741 	/* Exit VMX operation (i.e. exec VMXOFF) */
9742 	vmx_off();
9743 
9744 	/*
9745 	 * Signal to BSP CPU that we continue as usual as INIT signal
9746 	 * should have been consumed by VMX_INIT exit from guest
9747 	 */
9748 	vmx_set_test_stage(7);
9749 
9750 	/* Wait for BSP CPU to signal to enter VMX operation */
9751 	while (vmx_get_test_stage() != 8)
9752 		;
9753 	/* Enter VMX operation (i.e. exec VMXON) */
9754 	TEST_ASSERT(!__vmxon_safe(ap_vmxon_region));
9755 	/* Signal to BSP we are in VMX operation */
9756 	vmx_set_test_stage(9);
9757 
9758 	/* Wait for BSP CPU to send INIT signal */
9759 	while (vmx_get_test_stage() != 10)
9760 		;
9761 
9762 	/* Exit VMX operation (i.e. exec VMXOFF) */
9763 	vmx_off();
9764 
9765 	/*
9766 	 * Exiting VMX operation should result in latched
9767 	 * INIT signal being processed. Therefore, we should
9768 	 * never reach the below code. Thus, signal to BSP
9769 	 * CPU if we have reached here so it is able to
9770 	 * report an issue if it happens.
9771 	 */
9772 	init_signal_test_thread_continued = true;
9773 }
9774 
9775 #define INIT_SIGNAL_TEST_DELAY	100000000ULL
9776 
9777 static void vmx_init_signal_test(void)
9778 {
9779 	struct vmcs *test_vmcs;
9780 
9781 	if (cpu_count() < 2) {
9782 		report_skip("%s : CPU count < 2", __func__);
9783 		return;
9784 	}
9785 
9786 	/* VMCLEAR test-vmcs so it could be loaded by other CPU */
9787 	vmcs_save(&test_vmcs);
9788 	vmcs_clear(test_vmcs);
9789 
9790 	vmx_set_test_stage(0);
9791 	on_cpu_async(1, init_signal_test_thread, test_vmcs);
9792 
9793 	/* Wait for other CPU to enter VMX operation */
9794 	while (vmx_get_test_stage() != 1)
9795 		;
9796 
9797 	/* Send INIT signal to other CPU */
9798 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9799 				   id_map[1]);
9800 	/* Signal other CPU we have sent INIT signal */
9801 	vmx_set_test_stage(2);
9802 
9803 	/*
9804 	 * Wait reasonable amount of time for INIT signal to
9805 	 * be received on other CPU and verify that other CPU
9806 	 * have proceed as usual to next test stage as INIT
9807 	 * signal should be blocked while other CPU in
9808 	 * VMX operation
9809 	 */
9810 	delay(INIT_SIGNAL_TEST_DELAY);
9811 	report(vmx_get_test_stage() == 3,
9812 	       "INIT signal blocked when CPU in VMX operation");
9813 	/* No point to continue if we failed at this point */
9814 	if (vmx_get_test_stage() != 3)
9815 		return;
9816 
9817 	/* Signal other CPU to enter VMX non-root mode */
9818 	init_signal_test_exit_reason = -1ull;
9819 	vmx_set_test_stage(4);
9820 	/*
9821 	 * Wait reasonable amont of time for other CPU
9822 	 * to exit to VMX root mode
9823 	 */
9824 	delay(INIT_SIGNAL_TEST_DELAY);
9825 	if (vmx_get_test_stage() != 5) {
9826 		report_fail("Pending INIT signal didn't result in VMX exit");
9827 		return;
9828 	}
9829 	report(init_signal_test_exit_reason == VMX_INIT,
9830 			"INIT signal during VMX non-root mode result in exit-reason %s (%lu)",
9831 			exit_reason_description(init_signal_test_exit_reason),
9832 			init_signal_test_exit_reason);
9833 
9834 	/* Run guest to completion */
9835 	make_vmcs_current(test_vmcs);
9836 	enter_guest();
9837 
9838 	/* Signal other CPU to exit VMX operation */
9839 	init_signal_test_thread_continued = false;
9840 	vmx_set_test_stage(6);
9841 
9842 	/* Wait reasonable amount of time for other CPU to exit VMX operation */
9843 	delay(INIT_SIGNAL_TEST_DELAY);
9844 	report(vmx_get_test_stage() == 7,
9845 	       "INIT signal consumed on VMX_INIT exit");
9846 	/* No point to continue if we failed at this point */
9847 	if (vmx_get_test_stage() != 7)
9848 		return;
9849 
9850 	/* Signal other CPU to enter VMX operation */
9851 	vmx_set_test_stage(8);
9852 	/* Wait for other CPU to enter VMX operation */
9853 	while (vmx_get_test_stage() != 9)
9854 		;
9855 
9856 	/* Send INIT signal to other CPU */
9857 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9858 				   id_map[1]);
9859 	/* Signal other CPU we have sent INIT signal */
9860 	vmx_set_test_stage(10);
9861 
9862 	/*
9863 	 * Wait reasonable amount of time for other CPU
9864 	 * to exit VMX operation and process INIT signal
9865 	 */
9866 	delay(INIT_SIGNAL_TEST_DELAY);
9867 	report(!init_signal_test_thread_continued,
9868 	       "INIT signal processed after exit VMX operation");
9869 
9870 	/*
9871 	 * TODO: Send SIPI to other CPU to sipi_entry (See x86/cstart64.S)
9872 	 * to re-init it to kvm-unit-tests standard environment.
9873 	 * Somehow (?) verify that SIPI was indeed received.
9874 	 */
9875 }
9876 
9877 #define SIPI_SIGNAL_TEST_DELAY	100000000ULL
9878 
9879 static void vmx_sipi_test_guest(void)
9880 {
9881 	if (apic_id() == 0) {
9882 		/* wait AP enter guest with activity=WAIT_SIPI */
9883 		while (vmx_get_test_stage() != 1)
9884 			;
9885 		delay(SIPI_SIGNAL_TEST_DELAY);
9886 
9887 		/* First SIPI signal */
9888 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9889 		report_pass("BSP(L2): Send first SIPI to cpu[%d]", id_map[1]);
9890 
9891 		/* wait AP enter guest */
9892 		while (vmx_get_test_stage() != 2)
9893 			;
9894 		delay(SIPI_SIGNAL_TEST_DELAY);
9895 
9896 		/* Second SIPI signal should be ignored since AP is not in WAIT_SIPI state */
9897 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9898 		report_pass("BSP(L2): Send second SIPI to cpu[%d]", id_map[1]);
9899 
9900 		/* Delay a while to check whether second SIPI would cause VMExit */
9901 		delay(SIPI_SIGNAL_TEST_DELAY);
9902 
9903 		/* Test is done, notify AP to exit test */
9904 		vmx_set_test_stage(3);
9905 
9906 		/* wait AP exit non-root mode */
9907 		while (vmx_get_test_stage() != 5)
9908 			;
9909 	} else {
9910 		/* wait BSP notify test is done */
9911 		while (vmx_get_test_stage() != 3)
9912 			;
9913 
9914 		/* AP exit guest */
9915 		vmx_set_test_stage(4);
9916 	}
9917 }
9918 
9919 static void sipi_test_ap_thread(void *data)
9920 {
9921 	struct vmcs *ap_vmcs;
9922 	u64 *ap_vmxon_region;
9923 	void *ap_stack, *ap_syscall_stack;
9924 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9925 	u64 cpu_ctrl_1 = 0;
9926 
9927 	/* Enter VMX operation (i.e. exec VMXON) */
9928 	ap_vmxon_region = alloc_page();
9929 	enable_vmx();
9930 	init_vmx(ap_vmxon_region);
9931 	TEST_ASSERT(!__vmxon_safe(ap_vmxon_region));
9932 	init_vmcs(&ap_vmcs);
9933 	make_vmcs_current(ap_vmcs);
9934 
9935 	/* Set stack for AP */
9936 	ap_stack = alloc_page();
9937 	ap_syscall_stack = alloc_page();
9938 	vmcs_write(GUEST_RSP, (u64)(ap_stack + PAGE_SIZE - 1));
9939 	vmcs_write(GUEST_SYSENTER_ESP, (u64)(ap_syscall_stack + PAGE_SIZE - 1));
9940 
9941 	/* passthrough lapic to L2 */
9942 	disable_intercept_for_x2apic_msrs();
9943 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9944 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9945 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9946 
9947 	/* Set guest activity state to wait-for-SIPI state */
9948 	vmcs_write(GUEST_ACTV_STATE, ACTV_WAIT_SIPI);
9949 
9950 	vmx_set_test_stage(1);
9951 
9952 	/* AP enter guest */
9953 	enter_guest();
9954 
9955 	if (vmcs_read(EXI_REASON) == VMX_SIPI) {
9956 		report_pass("AP: Handle SIPI VMExit");
9957 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
9958 		vmx_set_test_stage(2);
9959 	} else {
9960 		report_fail("AP: Unexpected VMExit, reason=%ld", vmcs_read(EXI_REASON));
9961 		vmx_off();
9962 		return;
9963 	}
9964 
9965 	/* AP enter guest */
9966 	enter_guest();
9967 
9968 	report(vmcs_read(EXI_REASON) != VMX_SIPI,
9969 		"AP: should no SIPI VMExit since activity is not in WAIT_SIPI state");
9970 
9971 	/* notify BSP that AP is already exit from non-root mode */
9972 	vmx_set_test_stage(5);
9973 
9974 	/* Leave VMX operation */
9975 	vmx_off();
9976 }
9977 
9978 static void vmx_sipi_signal_test(void)
9979 {
9980 	if (!(rdmsr(MSR_IA32_VMX_MISC) & MSR_IA32_VMX_MISC_ACTIVITY_WAIT_SIPI)) {
9981 		report_skip("%s : \"ACTIVITY_WAIT_SIPI state\" not supported", __func__);
9982 		return;
9983 	}
9984 
9985 	if (cpu_count() < 2) {
9986 		report_skip("%s : CPU count < 2", __func__);
9987 		return;
9988 	}
9989 
9990 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9991 	u64 cpu_ctrl_1 = 0;
9992 
9993 	/* passthrough lapic to L2 */
9994 	disable_intercept_for_x2apic_msrs();
9995 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9996 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9997 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9998 
9999 	test_set_guest(vmx_sipi_test_guest);
10000 
10001 	/* update CR3 on AP */
10002 	on_cpu(1, update_cr3, (void *)read_cr3());
10003 
10004 	/* start AP */
10005 	on_cpu_async(1, sipi_test_ap_thread, NULL);
10006 
10007 	vmx_set_test_stage(0);
10008 
10009 	/* BSP enter guest */
10010 	enter_guest();
10011 }
10012 
10013 
10014 enum vmcs_access {
10015 	ACCESS_VMREAD,
10016 	ACCESS_VMWRITE,
10017 	ACCESS_NONE,
10018 };
10019 
10020 struct vmcs_shadow_test_common {
10021 	enum vmcs_access op;
10022 	enum Reason reason;
10023 	u64 field;
10024 	u64 value;
10025 	u64 flags;
10026 	u64 time;
10027 } l1_l2_common;
10028 
10029 static inline u64 vmread_flags(u64 field, u64 *val)
10030 {
10031 	u64 flags;
10032 
10033 	asm volatile ("vmread %2, %1; pushf; pop %0"
10034 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
10035 	return flags & X86_EFLAGS_ALU;
10036 }
10037 
10038 static inline u64 vmwrite_flags(u64 field, u64 val)
10039 {
10040 	u64 flags;
10041 
10042 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
10043 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
10044 	return flags & X86_EFLAGS_ALU;
10045 }
10046 
10047 static void vmx_vmcs_shadow_test_guest(void)
10048 {
10049 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10050 	u64 start;
10051 
10052 	while (c->op != ACCESS_NONE) {
10053 		start = rdtsc();
10054 		switch (c->op) {
10055 		default:
10056 			c->flags = -1ull;
10057 			break;
10058 		case ACCESS_VMREAD:
10059 			c->flags = vmread_flags(c->field, &c->value);
10060 			break;
10061 		case ACCESS_VMWRITE:
10062 			c->flags = vmwrite_flags(c->field, 0);
10063 			break;
10064 		}
10065 		c->time = rdtsc() - start;
10066 		vmcall();
10067 	}
10068 }
10069 
10070 static u64 vmread_from_shadow(u64 field)
10071 {
10072 	struct vmcs *primary;
10073 	struct vmcs *shadow;
10074 	u64 value;
10075 
10076 	TEST_ASSERT(!vmcs_save(&primary));
10077 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10078 	TEST_ASSERT(!make_vmcs_current(shadow));
10079 	value = vmcs_read(field);
10080 	TEST_ASSERT(!make_vmcs_current(primary));
10081 	return value;
10082 }
10083 
10084 static u64 vmwrite_to_shadow(u64 field, u64 value)
10085 {
10086 	struct vmcs *primary;
10087 	struct vmcs *shadow;
10088 
10089 	TEST_ASSERT(!vmcs_save(&primary));
10090 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10091 	TEST_ASSERT(!make_vmcs_current(shadow));
10092 	vmcs_write(field, value);
10093 	value = vmcs_read(field);
10094 	TEST_ASSERT(!make_vmcs_current(primary));
10095 	return value;
10096 }
10097 
10098 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
10099 {
10100 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10101 
10102 	c->op = access;
10103 	vmcs_write(VMX_INST_ERROR, 0);
10104 	enter_guest();
10105 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
10106 	if (c->reason != VMX_VMCALL) {
10107 		skip_exit_insn();
10108 		enter_guest();
10109 	}
10110 	skip_exit_vmcall();
10111 }
10112 
10113 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
10114 {
10115 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10116 	struct vmcs *shadow;
10117 	u64 value;
10118 	uintptr_t flags[2];
10119 	bool good_shadow;
10120 	u32 vmx_inst_error;
10121 
10122 	report_prefix_pushf("field %lx", field);
10123 	c->field = field;
10124 
10125 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10126 	if (shadow != (struct vmcs *)-1ull) {
10127 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
10128 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
10129 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
10130 	} else {
10131 		/*
10132 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
10133 		 * shadowed-fields should fail with setting RFLAGS.CF.
10134 		 */
10135 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
10136 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
10137 		good_shadow = false;
10138 	}
10139 
10140 	/* Intercept both VMREAD and VMWRITE. */
10141 	report_prefix_push("no VMREAD/VMWRITE permission");
10142 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
10143 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
10144 		set_bit(field, bitmap[ACCESS_VMREAD]);
10145 		set_bit(field, bitmap[ACCESS_VMWRITE]);
10146 	}
10147 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10148 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10149 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10150 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10151 	report_prefix_pop();
10152 
10153 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
10154 		goto out;
10155 
10156 	/* Permit shadowed VMREAD. */
10157 	report_prefix_push("VMREAD permission only");
10158 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10159 	set_bit(field, bitmap[ACCESS_VMWRITE]);
10160 	if (good_shadow)
10161 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10162 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10163 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10164 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10165 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10166 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10167 	       c->time);
10168 	report(c->flags == flags[ACCESS_VMREAD],
10169 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10170 	       c->flags, flags[ACCESS_VMREAD]);
10171 	if (good_shadow)
10172 		report(c->value == value,
10173 		       "value read from shadow (%lx) is as expected (%lx)",
10174 		       c->value, value);
10175 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10176 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10177 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10178 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10179 	report_prefix_pop();
10180 
10181 	/* Permit shadowed VMWRITE. */
10182 	report_prefix_push("VMWRITE permission only");
10183 	set_bit(field, bitmap[ACCESS_VMREAD]);
10184 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10185 	if (good_shadow)
10186 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10187 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10188 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10189 	report(c->reason == VMX_VMCALL,
10190 		"shadowed for VMWRITE (in %ld cycles)",
10191 		c->time);
10192 	report(c->flags == flags[ACCESS_VMREAD],
10193 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10194 	       c->flags, flags[ACCESS_VMREAD]);
10195 	if (good_shadow) {
10196 		value = vmread_from_shadow(field);
10197 		report(value == 0,
10198 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10199 		       0ul);
10200 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10201 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10202 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10203 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10204 	}
10205 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10206 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10207 	report_prefix_pop();
10208 
10209 	/* Permit shadowed VMREAD and VMWRITE. */
10210 	report_prefix_push("VMREAD and VMWRITE permission");
10211 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10212 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10213 	if (good_shadow)
10214 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10215 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10216 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10217 	report(c->reason == VMX_VMCALL,
10218 		"shadowed for VMWRITE (in %ld cycles)",
10219 		c->time);
10220 	report(c->flags == flags[ACCESS_VMREAD],
10221 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10222 	       c->flags, flags[ACCESS_VMREAD]);
10223 	if (good_shadow) {
10224 		value = vmread_from_shadow(field);
10225 		report(value == 0,
10226 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10227 		       0ul);
10228 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10229 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10230 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10231 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10232 	}
10233 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10234 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10235 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10236 	       c->time);
10237 	report(c->flags == flags[ACCESS_VMREAD],
10238 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10239 	       c->flags, flags[ACCESS_VMREAD]);
10240 	if (good_shadow)
10241 		report(c->value == 0,
10242 		       "value read from shadow (%lx) is as expected (%lx)",
10243 		       c->value, 0ul);
10244 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10245 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10246 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10247 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10248 	report_prefix_pop();
10249 
10250 out:
10251 	report_prefix_pop();
10252 }
10253 
10254 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
10255 {
10256 	unsigned base;
10257 	unsigned index;
10258 	unsigned bit;
10259 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
10260 
10261 	/* Run test on all possible valid VMCS fields */
10262 	for (base = 0;
10263 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
10264 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
10265 		for (index = 0; index <= highest_index; index++)
10266 			vmcs_shadow_test_field(bitmap, base + index);
10267 
10268 	/*
10269 	 * Run tests on some invalid VMCS fields
10270 	 * (Have reserved bit set).
10271 	 */
10272 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
10273 		vmcs_shadow_test_field(bitmap, (1ull << bit));
10274 }
10275 
10276 static void vmx_vmcs_shadow_test(void)
10277 {
10278 	u8 *bitmap[2];
10279 	struct vmcs *shadow;
10280 
10281 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
10282 		report_skip("%s : \"Activate secondary controls\" not supported", __func__);
10283 		return;
10284 	}
10285 
10286 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
10287 		report_skip("%s : \"VMCS shadowing\" not supported", __func__);
10288 		return;
10289 	}
10290 
10291 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
10292 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
10293 		report_skip("%s : VMWRITE can't modify VM-exit information fields.", __func__);
10294 		return;
10295 	}
10296 
10297 	test_set_guest(vmx_vmcs_shadow_test_guest);
10298 
10299 	bitmap[ACCESS_VMREAD] = alloc_page();
10300 	bitmap[ACCESS_VMWRITE] = alloc_page();
10301 
10302 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
10303 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
10304 
10305 	shadow = alloc_page();
10306 	shadow->hdr.revision_id = basic.revision;
10307 	shadow->hdr.shadow_vmcs = 1;
10308 	TEST_ASSERT(!vmcs_clear(shadow));
10309 
10310 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
10311 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
10312 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
10313 
10314 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
10315 	report_prefix_push("valid link pointer");
10316 	vmx_vmcs_shadow_test_body(bitmap);
10317 	report_prefix_pop();
10318 
10319 	vmcs_write(VMCS_LINK_PTR, -1ull);
10320 	report_prefix_push("invalid link pointer");
10321 	vmx_vmcs_shadow_test_body(bitmap);
10322 	report_prefix_pop();
10323 
10324 	l1_l2_common.op = ACCESS_NONE;
10325 	enter_guest();
10326 }
10327 
10328 /*
10329  * This test monitors the difference between a guest RDTSC instruction
10330  * and the IA32_TIME_STAMP_COUNTER MSR value stored in the VMCS12
10331  * VM-exit MSR-store list when taking a VM-exit on the instruction
10332  * following RDTSC.
10333  */
10334 #define RDTSC_DIFF_ITERS 100000
10335 #define RDTSC_DIFF_FAILS 100
10336 #define HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD 750
10337 
10338 /*
10339  * Set 'use TSC offsetting' and set the guest offset to the
10340  * inverse of the host's current TSC value, so that the guest starts running
10341  * with an effective TSC value of 0.
10342  */
10343 static void reset_guest_tsc_to_zero(void)
10344 {
10345 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
10346 	vmcs_write(TSC_OFFSET, -rdtsc());
10347 }
10348 
10349 static void rdtsc_vmexit_diff_test_guest(void)
10350 {
10351 	int i;
10352 
10353 	for (i = 0; i < RDTSC_DIFF_ITERS; i++)
10354 		/* Ensure rdtsc is the last instruction before the vmcall. */
10355 		asm volatile("rdtsc; vmcall" : : : "eax", "edx");
10356 }
10357 
10358 /*
10359  * This function only considers the "use TSC offsetting" VM-execution
10360  * control.  It does not handle "use TSC scaling" (because the latter
10361  * isn't available to the host today.)
10362  */
10363 static unsigned long long host_time_to_guest_time(unsigned long long t)
10364 {
10365 	TEST_ASSERT(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
10366 		    !(vmcs_read(CPU_EXEC_CTRL1) & CPU_USE_TSC_SCALING));
10367 
10368 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_USE_TSC_OFFSET)
10369 		t += vmcs_read(TSC_OFFSET);
10370 
10371 	return t;
10372 }
10373 
10374 static unsigned long long rdtsc_vmexit_diff_test_iteration(void)
10375 {
10376 	unsigned long long guest_tsc, host_to_guest_tsc;
10377 
10378 	enter_guest();
10379 	skip_exit_vmcall();
10380 	guest_tsc = (u32) regs.rax + (regs.rdx << 32);
10381 	host_to_guest_tsc = host_time_to_guest_time(exit_msr_store[0].value);
10382 
10383 	return host_to_guest_tsc - guest_tsc;
10384 }
10385 
10386 static void rdtsc_vmexit_diff_test(void)
10387 {
10388 	unsigned long long delta;
10389 	int fail = 0;
10390 	int i;
10391 
10392 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET))
10393 		test_skip("CPU doesn't support the 'use TSC offsetting' processor-based VM-execution control.\n");
10394 
10395 	test_set_guest(rdtsc_vmexit_diff_test_guest);
10396 
10397 	reset_guest_tsc_to_zero();
10398 
10399 	/*
10400 	 * Set up the VMCS12 VM-exit MSR-store list to store just one
10401 	 * MSR: IA32_TIME_STAMP_COUNTER. Note that the value stored is
10402 	 * in the host time domain (i.e., it is not adjusted according
10403 	 * to the TSC multiplier and TSC offset fields in the VMCS12,
10404 	 * as a guest RDTSC would be.)
10405 	 */
10406 	exit_msr_store = alloc_page();
10407 	exit_msr_store[0].index = MSR_IA32_TSC;
10408 	vmcs_write(EXI_MSR_ST_CNT, 1);
10409 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(exit_msr_store));
10410 
10411 	for (i = 0; i < RDTSC_DIFF_ITERS && fail < RDTSC_DIFF_FAILS; i++) {
10412 		delta = rdtsc_vmexit_diff_test_iteration();
10413 		if (delta >= HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD)
10414 			fail++;
10415 	}
10416 
10417 	enter_guest();
10418 
10419 	report(fail < RDTSC_DIFF_FAILS,
10420 	       "RDTSC to VM-exit delta too high in %d of %d iterations, last = %llu",
10421 	       fail, i, delta);
10422 }
10423 
10424 static int invalid_msr_init(struct vmcs *vmcs)
10425 {
10426 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
10427 		printf("\tPreemption timer is not supported\n");
10428 		return VMX_TEST_EXIT;
10429 	}
10430 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
10431 	preempt_val = 10000000;
10432 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
10433 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
10434 
10435 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
10436 		printf("\tSave preemption value is not supported\n");
10437 
10438 	vmcs_write(ENT_MSR_LD_CNT, 1);
10439 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
10440 
10441 	return VMX_TEST_START;
10442 }
10443 
10444 
10445 static void invalid_msr_main(void)
10446 {
10447 	report_fail("Invalid MSR load");
10448 }
10449 
10450 static int invalid_msr_exit_handler(union exit_reason exit_reason)
10451 {
10452 	report_fail("Invalid MSR load");
10453 	print_vmexit_info(exit_reason);
10454 	return VMX_TEST_EXIT;
10455 }
10456 
10457 static int invalid_msr_entry_failure(struct vmentry_result *result)
10458 {
10459 	report(result->exit_reason.failed_vmentry &&
10460 	       result->exit_reason.basic == VMX_FAIL_MSR, "Invalid MSR load");
10461 	return VMX_TEST_VMEXIT;
10462 }
10463 
10464 /*
10465  * The max number of MSRs in an atomic switch MSR list is:
10466  * (111B + 1) * 512 = 4096
10467  *
10468  * Each list entry consumes:
10469  * 4-byte MSR index + 4 bytes reserved + 8-byte data = 16 bytes
10470  *
10471  * Allocate 128 kB to cover max_msr_list_size (i.e., 64 kB) and then some.
10472  */
10473 static const u32 msr_list_page_order = 5;
10474 
10475 static void atomic_switch_msr_limit_test_guest(void)
10476 {
10477 	vmcall();
10478 }
10479 
10480 static void populate_msr_list(struct vmx_msr_entry *msr_list,
10481 			      size_t byte_capacity, int count)
10482 {
10483 	int i;
10484 
10485 	for (i = 0; i < count; i++) {
10486 		msr_list[i].index = MSR_IA32_TSC;
10487 		msr_list[i].reserved = 0;
10488 		msr_list[i].value = 0x1234567890abcdef;
10489 	}
10490 
10491 	memset(msr_list + count, 0xff,
10492 	       byte_capacity - count * sizeof(*msr_list));
10493 }
10494 
10495 static int max_msr_list_size(void)
10496 {
10497 	u32 vmx_misc = rdmsr(MSR_IA32_VMX_MISC);
10498 	u32 factor = ((vmx_misc & GENMASK(27, 25)) >> 25) + 1;
10499 
10500 	return factor * 512;
10501 }
10502 
10503 static void atomic_switch_msrs_test(int count)
10504 {
10505 	struct vmx_msr_entry *vm_enter_load;
10506         struct vmx_msr_entry *vm_exit_load;
10507         struct vmx_msr_entry *vm_exit_store;
10508 	int max_allowed = max_msr_list_size();
10509 	int byte_capacity = 1ul << (msr_list_page_order + PAGE_SHIFT);
10510 	/* Exceeding the max MSR list size at exit triggers KVM to abort. */
10511 	int exit_count = count > max_allowed ? max_allowed : count;
10512 	int cleanup_count = count > max_allowed ? 2 : 1;
10513 	int i;
10514 
10515 	/*
10516 	 * Check for the IA32_TSC MSR,
10517 	 * available with the "TSC flag" and used to populate the MSR lists.
10518 	 */
10519 	if (!(cpuid(1).d & (1 << 4))) {
10520 		report_skip("%s : \"Time Stamp Counter\" not supported", __func__);
10521 		return;
10522 	}
10523 
10524 	/* Set L2 guest. */
10525 	test_set_guest(atomic_switch_msr_limit_test_guest);
10526 
10527 	/* Setup atomic MSR switch lists. */
10528 	vm_enter_load = alloc_pages(msr_list_page_order);
10529 	vm_exit_load = alloc_pages(msr_list_page_order);
10530 	vm_exit_store = alloc_pages(msr_list_page_order);
10531 
10532 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)vm_enter_load);
10533 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)vm_exit_load);
10534 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)vm_exit_store);
10535 
10536 	/*
10537 	 * VM-Enter should succeed up to the max number of MSRs per list, and
10538 	 * should not consume junk beyond the last entry.
10539 	 */
10540 	populate_msr_list(vm_enter_load, byte_capacity, count);
10541 	populate_msr_list(vm_exit_load, byte_capacity, exit_count);
10542 	populate_msr_list(vm_exit_store, byte_capacity, exit_count);
10543 
10544 	vmcs_write(ENT_MSR_LD_CNT, count);
10545 	vmcs_write(EXI_MSR_LD_CNT, exit_count);
10546 	vmcs_write(EXI_MSR_ST_CNT, exit_count);
10547 
10548 	if (count <= max_allowed) {
10549 		enter_guest();
10550 		assert_exit_reason(VMX_VMCALL);
10551 		skip_exit_vmcall();
10552 	} else {
10553 		u32 exit_qual;
10554 
10555 		test_guest_state("Invalid MSR Load Count", true, count,
10556 				 "ENT_MSR_LD_CNT");
10557 
10558 		exit_qual = vmcs_read(EXI_QUALIFICATION);
10559 		report(exit_qual == max_allowed + 1, "exit_qual, %u, is %u.",
10560 		       exit_qual, max_allowed + 1);
10561 	}
10562 
10563 	/* Cleanup. */
10564 	vmcs_write(ENT_MSR_LD_CNT, 0);
10565 	vmcs_write(EXI_MSR_LD_CNT, 0);
10566 	vmcs_write(EXI_MSR_ST_CNT, 0);
10567 	for (i = 0; i < cleanup_count; i++) {
10568 		enter_guest();
10569 		skip_exit_vmcall();
10570 	}
10571 	free_pages_by_order(vm_enter_load, msr_list_page_order);
10572 	free_pages_by_order(vm_exit_load, msr_list_page_order);
10573 	free_pages_by_order(vm_exit_store, msr_list_page_order);
10574 }
10575 
10576 static void atomic_switch_max_msrs_test(void)
10577 {
10578 	atomic_switch_msrs_test(max_msr_list_size());
10579 }
10580 
10581 static void atomic_switch_overflow_msrs_test(void)
10582 {
10583 	if (test_device_enabled())
10584 		atomic_switch_msrs_test(max_msr_list_size() + 1);
10585 	else
10586 		test_skip("Test is only supported on KVM");
10587 }
10588 
10589 static void vmx_pf_exception_test_guest(void)
10590 {
10591 	ac_test_run(PT_LEVEL_PML4);
10592 }
10593 
10594 typedef void (*invalidate_tlb_t)(void *data);
10595 
10596 static void __vmx_pf_exception_test(invalidate_tlb_t inv_fn, void *data)
10597 {
10598 	u64 efer;
10599 	struct cpuid cpuid;
10600 
10601 	test_set_guest(vmx_pf_exception_test_guest);
10602 
10603 	/* Intercept INVLPG when to perform TLB invalidation from L1 (this). */
10604 	if (inv_fn)
10605 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INVLPG);
10606 	else
10607 		vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INVLPG);
10608 
10609 	enter_guest();
10610 
10611 	while (vmcs_read(EXI_REASON) != VMX_VMCALL) {
10612 		switch (vmcs_read(EXI_REASON)) {
10613 		case VMX_RDMSR:
10614 			assert(regs.rcx == MSR_EFER);
10615 			efer = vmcs_read(GUEST_EFER);
10616 			regs.rdx = efer >> 32;
10617 			regs.rax = efer & 0xffffffff;
10618 			break;
10619 		case VMX_WRMSR:
10620 			assert(regs.rcx == MSR_EFER);
10621 			efer = regs.rdx << 32 | (regs.rax & 0xffffffff);
10622 			vmcs_write(GUEST_EFER, efer);
10623 			break;
10624 		case VMX_CPUID:
10625 			cpuid = (struct cpuid) {0, 0, 0, 0};
10626 			cpuid = raw_cpuid(regs.rax, regs.rcx);
10627 			regs.rax = cpuid.a;
10628 			regs.rbx = cpuid.b;
10629 			regs.rcx = cpuid.c;
10630 			regs.rdx = cpuid.d;
10631 			break;
10632 		case VMX_INVLPG:
10633 			inv_fn(data);
10634 			break;
10635 		default:
10636 			assert_msg(false,
10637 				"Unexpected exit to L1, exit_reason: %s (0x%lx)",
10638 				exit_reason_description(vmcs_read(EXI_REASON)),
10639 				vmcs_read(EXI_REASON));
10640 		}
10641 		skip_exit_insn();
10642 		enter_guest();
10643 	}
10644 
10645 	assert_exit_reason(VMX_VMCALL);
10646 }
10647 
10648 static void vmx_pf_exception_test(void)
10649 {
10650 	__vmx_pf_exception_test(NULL, NULL);
10651 }
10652 
10653 static void invalidate_tlb_no_vpid(void *data)
10654 {
10655 	/* If VPID is disabled, the TLB is flushed on VM-Enter and VM-Exit. */
10656 }
10657 
10658 static void vmx_pf_no_vpid_test(void)
10659 {
10660 	if (is_vpid_supported())
10661 		vmcs_clear_bits(CPU_EXEC_CTRL1, CPU_VPID);
10662 
10663 	__vmx_pf_exception_test(invalidate_tlb_no_vpid, NULL);
10664 }
10665 
10666 static void invalidate_tlb_invvpid_addr(void *data)
10667 {
10668 	invvpid(INVVPID_ALL, *(u16 *)data, vmcs_read(EXI_QUALIFICATION));
10669 }
10670 
10671 static void invalidate_tlb_new_vpid(void *data)
10672 {
10673 	u16 *vpid = data;
10674 
10675 	/*
10676 	 * Bump VPID to effectively flush L2's TLB from L0's perspective.
10677 	 * Invalidate all VPIDs when the VPID wraps to zero as hardware/KVM is
10678 	 * architecturally allowed to keep TLB entries indefinitely.
10679 	 */
10680 	++(*vpid);
10681 	if (*vpid == 0) {
10682 		++(*vpid);
10683 		invvpid(INVVPID_ALL, 0, 0);
10684 	}
10685 	vmcs_write(VPID, *vpid);
10686 }
10687 
10688 static void __vmx_pf_vpid_test(invalidate_tlb_t inv_fn, u16 vpid)
10689 {
10690 	if (!is_vpid_supported())
10691 		test_skip("VPID unsupported");
10692 
10693 	if (!is_invvpid_supported())
10694 		test_skip("INVVPID unsupported");
10695 
10696 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
10697 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VPID);
10698 	vmcs_write(VPID, vpid);
10699 
10700 	__vmx_pf_exception_test(inv_fn, &vpid);
10701 }
10702 
10703 static void vmx_pf_invvpid_test(void)
10704 {
10705 	if (!is_invvpid_type_supported(INVVPID_ADDR))
10706 		test_skip("INVVPID ADDR unsupported");
10707 
10708 	__vmx_pf_vpid_test(invalidate_tlb_invvpid_addr, 0xaaaa);
10709 }
10710 
10711 static void vmx_pf_vpid_test(void)
10712 {
10713 	/* Need INVVPID(ALL) to flush VPIDs upon wrap/reuse. */
10714 	if (!is_invvpid_type_supported(INVVPID_ALL))
10715 		test_skip("INVVPID ALL unsupported");
10716 
10717 	__vmx_pf_vpid_test(invalidate_tlb_new_vpid, 1);
10718 }
10719 
10720 static void vmx_l2_gp_test(void)
10721 {
10722 	*(volatile u64 *)NONCANONICAL = 0;
10723 }
10724 
10725 static void vmx_l2_ud_test(void)
10726 {
10727 	asm volatile ("ud2");
10728 }
10729 
10730 static void vmx_l2_de_test(void)
10731 {
10732 	asm volatile (
10733 		"xor %%eax, %%eax\n\t"
10734 		"xor %%ebx, %%ebx\n\t"
10735 		"xor %%edx, %%edx\n\t"
10736 		"idiv %%ebx\n\t"
10737 		::: "eax", "ebx", "edx");
10738 }
10739 
10740 static void vmx_l2_bp_test(void)
10741 {
10742 	asm volatile ("int3");
10743 }
10744 
10745 static void vmx_l2_db_test(void)
10746 {
10747 	write_rflags(read_rflags() | X86_EFLAGS_TF);
10748 }
10749 
10750 static uint64_t usermode_callback(void)
10751 {
10752 	/*
10753 	 * Trigger an #AC by writing 8 bytes to a 4-byte aligned address.
10754 	 * Disclaimer: It is assumed that the stack pointer is aligned
10755 	 * on a 16-byte boundary as x86_64 stacks should be.
10756 	 */
10757 	asm volatile("movq $0, -0x4(%rsp)");
10758 
10759 	return 0;
10760 }
10761 
10762 static void vmx_l2_ac_test(void)
10763 {
10764 	bool hit_ac = false;
10765 
10766 	write_cr0(read_cr0() | X86_CR0_AM);
10767 	write_rflags(read_rflags() | X86_EFLAGS_AC);
10768 
10769 	run_in_user(usermode_callback, AC_VECTOR, 0, 0, 0, 0, &hit_ac);
10770 	report(hit_ac, "Usermode #AC handled in L2");
10771 	vmcall();
10772 }
10773 
10774 struct vmx_exception_test {
10775 	u8 vector;
10776 	void (*guest_code)(void);
10777 };
10778 
10779 struct vmx_exception_test vmx_exception_tests[] = {
10780 	{ GP_VECTOR, vmx_l2_gp_test },
10781 	{ UD_VECTOR, vmx_l2_ud_test },
10782 	{ DE_VECTOR, vmx_l2_de_test },
10783 	{ DB_VECTOR, vmx_l2_db_test },
10784 	{ BP_VECTOR, vmx_l2_bp_test },
10785 	{ AC_VECTOR, vmx_l2_ac_test },
10786 };
10787 
10788 static u8 vmx_exception_test_vector;
10789 
10790 static void vmx_exception_handler(struct ex_regs *regs)
10791 {
10792 	report(regs->vector == vmx_exception_test_vector,
10793 	       "Handling %s in L2's exception handler",
10794 	       exception_mnemonic(vmx_exception_test_vector));
10795 	vmcall();
10796 }
10797 
10798 static void handle_exception_in_l2(u8 vector)
10799 {
10800 	handler old_handler = handle_exception(vector, vmx_exception_handler);
10801 
10802 	vmx_exception_test_vector = vector;
10803 
10804 	enter_guest();
10805 	report(vmcs_read(EXI_REASON) == VMX_VMCALL,
10806 	       "%s handled by L2", exception_mnemonic(vector));
10807 
10808 	handle_exception(vector, old_handler);
10809 }
10810 
10811 static void handle_exception_in_l1(u32 vector)
10812 {
10813 	u32 old_eb = vmcs_read(EXC_BITMAP);
10814 
10815 	vmcs_write(EXC_BITMAP, old_eb | (1u << vector));
10816 
10817 	enter_guest();
10818 
10819 	report((vmcs_read(EXI_REASON) == VMX_EXC_NMI) &&
10820 	       ((vmcs_read(EXI_INTR_INFO) & 0xff) == vector),
10821 	       "%s handled by L1", exception_mnemonic(vector));
10822 
10823 	vmcs_write(EXC_BITMAP, old_eb);
10824 }
10825 
10826 static void vmx_exception_test(void)
10827 {
10828 	struct vmx_exception_test *t;
10829 	int i;
10830 
10831 	for (i = 0; i < ARRAY_SIZE(vmx_exception_tests); i++) {
10832 		t = &vmx_exception_tests[i];
10833 
10834 		/*
10835 		 * Override the guest code before each run even though it's the
10836 		 * same code, the VMCS guest state needs to be reinitialized.
10837 		 */
10838 		test_override_guest(t->guest_code);
10839 		handle_exception_in_l2(t->vector);
10840 
10841 		test_override_guest(t->guest_code);
10842 		handle_exception_in_l1(t->vector);
10843 	}
10844 
10845 	test_set_guest_finished();
10846 }
10847 
10848 #define TEST(name) { #name, .v2 = name }
10849 
10850 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
10851 struct vmx_test vmx_tests[] = {
10852 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
10853 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
10854 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
10855 		preemption_timer_exit_handler, NULL, {0} },
10856 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
10857 		test_ctrl_pat_exit_handler, NULL, {0} },
10858 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
10859 		test_ctrl_efer_exit_handler, NULL, {0} },
10860 	{ "CR shadowing", NULL, cr_shadowing_main,
10861 		cr_shadowing_exit_handler, NULL, {0} },
10862 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
10863 		NULL, {0} },
10864 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
10865 		insn_intercept_exit_handler, NULL, {0} },
10866 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
10867 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
10868 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
10869 	{ "interrupt", interrupt_init, interrupt_main,
10870 		interrupt_exit_handler, NULL, {0} },
10871 	{ "nmi_hlt", nmi_hlt_init, nmi_hlt_main,
10872 		nmi_hlt_exit_handler, NULL, {0} },
10873 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
10874 		NULL, {0} },
10875 	{ "MSR switch", msr_switch_init, msr_switch_main,
10876 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
10877 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
10878 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
10879 		disable_rdtscp_exit_handler, NULL, {0} },
10880 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
10881 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
10882 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
10883 		exit_monitor_from_l2_handler, NULL, {0} },
10884 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
10885 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
10886 	/* Basic V2 tests. */
10887 	TEST(v2_null_test),
10888 	TEST(v2_multiple_entries_test),
10889 	TEST(fixture_test_case1),
10890 	TEST(fixture_test_case2),
10891 	/* Opcode tests. */
10892 	TEST(invvpid_test),
10893 	/* VM-entry tests */
10894 	TEST(vmx_controls_test),
10895 	TEST(vmx_host_state_area_test),
10896 	TEST(vmx_guest_state_area_test),
10897 	TEST(vmentry_movss_shadow_test),
10898 	TEST(vmentry_unrestricted_guest_test),
10899 	/* APICv tests */
10900 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
10901 	TEST(vmx_hlt_with_rvi_test),
10902 	TEST(apic_reg_virt_test),
10903 	TEST(virt_x2apic_mode_test),
10904 	/* APIC pass-through tests */
10905 	TEST(vmx_apic_passthrough_test),
10906 	TEST(vmx_apic_passthrough_thread_test),
10907 	TEST(vmx_apic_passthrough_tpr_threshold_test),
10908 	TEST(vmx_init_signal_test),
10909 	TEST(vmx_sipi_signal_test),
10910 	/* VMCS Shadowing tests */
10911 	TEST(vmx_vmcs_shadow_test),
10912 	/* Regression tests */
10913 	TEST(vmx_ldtr_test),
10914 	TEST(vmx_cr_load_test),
10915 	TEST(vmx_cr4_osxsave_test),
10916 	TEST(vmx_nm_test),
10917 	TEST(vmx_db_test),
10918 	TEST(vmx_nmi_window_test),
10919 	TEST(vmx_intr_window_test),
10920 	TEST(vmx_pending_event_test),
10921 	TEST(vmx_pending_event_hlt_test),
10922 	TEST(vmx_store_tsc_test),
10923 	TEST(vmx_preemption_timer_zero_test),
10924 	TEST(vmx_preemption_timer_tf_test),
10925 	TEST(vmx_preemption_timer_expiry_test),
10926 	/* EPT access tests. */
10927 	TEST(ept_access_test_not_present),
10928 	TEST(ept_access_test_read_only),
10929 	TEST(ept_access_test_write_only),
10930 	TEST(ept_access_test_read_write),
10931 	TEST(ept_access_test_execute_only),
10932 	TEST(ept_access_test_read_execute),
10933 	TEST(ept_access_test_write_execute),
10934 	TEST(ept_access_test_read_write_execute),
10935 	TEST(ept_access_test_reserved_bits),
10936 	TEST(ept_access_test_ignored_bits),
10937 	TEST(ept_access_test_paddr_not_present_ad_disabled),
10938 	TEST(ept_access_test_paddr_not_present_ad_enabled),
10939 	TEST(ept_access_test_paddr_read_only_ad_disabled),
10940 	TEST(ept_access_test_paddr_read_only_ad_enabled),
10941 	TEST(ept_access_test_paddr_read_write),
10942 	TEST(ept_access_test_paddr_read_write_execute),
10943 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
10944 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
10945 	TEST(ept_access_test_paddr_not_present_page_fault),
10946 	TEST(ept_access_test_force_2m_page),
10947 	/* Atomic MSR switch tests. */
10948 	TEST(atomic_switch_max_msrs_test),
10949 	TEST(atomic_switch_overflow_msrs_test),
10950 	TEST(rdtsc_vmexit_diff_test),
10951 	TEST(vmx_mtf_test),
10952 	TEST(vmx_mtf_pdpte_test),
10953 	TEST(vmx_pf_exception_test),
10954 	TEST(vmx_pf_no_vpid_test),
10955 	TEST(vmx_pf_invvpid_test),
10956 	TEST(vmx_pf_vpid_test),
10957 	TEST(vmx_exception_test),
10958 	{ NULL, NULL, NULL, NULL, NULL, {0} },
10959 };
10960