xref: /kvm-unit-tests/x86/vmx_tests.c (revision 3b6c56b3e69223b5c4cbf294f4ddd537dc30524b)
1 /*
2  * All test cases of nested virtualization should be in this file
3  *
4  * Author : Arthur Chunqi Li <yzt356@gmail.com>
5  */
6 
7 #include <asm/debugreg.h>
8 
9 #include "vmx.h"
10 #include "msr.h"
11 #include "processor.h"
12 #include "vm.h"
13 #include "pci.h"
14 #include "fwcfg.h"
15 #include "isr.h"
16 #include "desc.h"
17 #include "apic.h"
18 #include "types.h"
19 #include "vmalloc.h"
20 #include "alloc_page.h"
21 #include "smp.h"
22 #include "delay.h"
23 #include "access.h"
24 
25 #define VPID_CAP_INVVPID_TYPES_SHIFT 40
26 
27 u64 ia32_pat;
28 u64 ia32_efer;
29 void *io_bitmap_a, *io_bitmap_b;
30 u16 ioport;
31 
32 unsigned long *pml4;
33 u64 eptp;
34 void *data_page1, *data_page2;
35 
36 phys_addr_t pci_physaddr;
37 
38 void *pml_log;
39 #define PML_INDEX 512
40 
41 static inline unsigned ffs(unsigned x)
42 {
43 	int pos = -1;
44 
45 	__asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0"
46 			     : "+r"(pos) : "rm"(x) : "eax");
47 	return pos + 1;
48 }
49 
50 static inline void vmcall(void)
51 {
52 	asm volatile("vmcall");
53 }
54 
55 static void basic_guest_main(void)
56 {
57 	report_pass("Basic VMX test");
58 }
59 
60 static int basic_exit_handler(union exit_reason exit_reason)
61 {
62 	report_fail("Basic VMX test");
63 	print_vmexit_info(exit_reason);
64 	return VMX_TEST_EXIT;
65 }
66 
67 static void vmenter_main(void)
68 {
69 	u64 rax;
70 	u64 rsp, resume_rsp;
71 
72 	report_pass("test vmlaunch");
73 
74 	asm volatile(
75 		"mov %%rsp, %0\n\t"
76 		"mov %3, %%rax\n\t"
77 		"vmcall\n\t"
78 		"mov %%rax, %1\n\t"
79 		"mov %%rsp, %2\n\t"
80 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
81 		: "g"(0xABCD));
82 	report((rax == 0xFFFF) && (rsp == resume_rsp), "test vmresume");
83 }
84 
85 static int vmenter_exit_handler(union exit_reason exit_reason)
86 {
87 	u64 guest_rip = vmcs_read(GUEST_RIP);
88 
89 	switch (exit_reason.basic) {
90 	case VMX_VMCALL:
91 		if (regs.rax != 0xABCD) {
92 			report_fail("test vmresume");
93 			return VMX_TEST_VMEXIT;
94 		}
95 		regs.rax = 0xFFFF;
96 		vmcs_write(GUEST_RIP, guest_rip + 3);
97 		return VMX_TEST_RESUME;
98 	default:
99 		report_fail("test vmresume");
100 		print_vmexit_info(exit_reason);
101 	}
102 	return VMX_TEST_VMEXIT;
103 }
104 
105 u32 preempt_scale;
106 volatile unsigned long long tsc_val;
107 volatile u32 preempt_val;
108 u64 saved_rip;
109 
110 static int preemption_timer_init(struct vmcs *vmcs)
111 {
112 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
113 		printf("\tPreemption timer is not supported\n");
114 		return VMX_TEST_EXIT;
115 	}
116 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
117 	preempt_val = 10000000;
118 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
119 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
120 
121 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
122 		printf("\tSave preemption value is not supported\n");
123 
124 	return VMX_TEST_START;
125 }
126 
127 static void preemption_timer_main(void)
128 {
129 	tsc_val = rdtsc();
130 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
131 		vmx_set_test_stage(0);
132 		vmcall();
133 		if (vmx_get_test_stage() == 1)
134 			vmcall();
135 	}
136 	vmx_set_test_stage(1);
137 	while (vmx_get_test_stage() == 1) {
138 		if (((rdtsc() - tsc_val) >> preempt_scale)
139 				> 10 * preempt_val) {
140 			vmx_set_test_stage(2);
141 			vmcall();
142 		}
143 	}
144 	tsc_val = rdtsc();
145 	asm volatile ("hlt");
146 	vmcall();
147 	vmx_set_test_stage(5);
148 	vmcall();
149 }
150 
151 static int preemption_timer_exit_handler(union exit_reason exit_reason)
152 {
153 	bool guest_halted;
154 	u64 guest_rip;
155 	u32 insn_len;
156 	u32 ctrl_exit;
157 
158 	guest_rip = vmcs_read(GUEST_RIP);
159 	insn_len = vmcs_read(EXI_INST_LEN);
160 	switch (exit_reason.basic) {
161 	case VMX_PREEMPT:
162 		switch (vmx_get_test_stage()) {
163 		case 1:
164 		case 2:
165 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val,
166 			       "busy-wait for preemption timer");
167 			vmx_set_test_stage(3);
168 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
169 			return VMX_TEST_RESUME;
170 		case 3:
171 			guest_halted =
172 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
173 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val
174 			        && guest_halted,
175 			       "preemption timer during hlt");
176 			vmx_set_test_stage(4);
177 			vmcs_write(PIN_CONTROLS,
178 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
179 			vmcs_write(EXI_CONTROLS,
180 				   vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT);
181 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
182 			return VMX_TEST_RESUME;
183 		case 4:
184 			report(saved_rip == guest_rip,
185 			       "preemption timer with 0 value");
186 			break;
187 		default:
188 			report_fail("Invalid stage.");
189 			print_vmexit_info(exit_reason);
190 			break;
191 		}
192 		break;
193 	case VMX_VMCALL:
194 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
195 		switch (vmx_get_test_stage()) {
196 		case 0:
197 			report(vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val,
198 			       "Keep preemption value");
199 			vmx_set_test_stage(1);
200 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
201 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
202 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
203 			vmcs_write(EXI_CONTROLS, ctrl_exit);
204 			return VMX_TEST_RESUME;
205 		case 1:
206 			report(vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val,
207 			       "Save preemption value");
208 			return VMX_TEST_RESUME;
209 		case 2:
210 			report_fail("busy-wait for preemption timer");
211 			vmx_set_test_stage(3);
212 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
213 			return VMX_TEST_RESUME;
214 		case 3:
215 			report_fail("preemption timer during hlt");
216 			vmx_set_test_stage(4);
217 			/* fall through */
218 		case 4:
219 			vmcs_write(PIN_CONTROLS,
220 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
221 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
222 			saved_rip = guest_rip + insn_len;
223 			return VMX_TEST_RESUME;
224 		case 5:
225 			report_fail("preemption timer with 0 value (vmcall stage 5)");
226 			break;
227 		default:
228 			// Should not reach here
229 			report_fail("unexpected stage, %d",
230 				    vmx_get_test_stage());
231 			print_vmexit_info(exit_reason);
232 			return VMX_TEST_VMEXIT;
233 		}
234 		break;
235 	default:
236 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
237 		print_vmexit_info(exit_reason);
238 	}
239 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
240 	return VMX_TEST_VMEXIT;
241 }
242 
243 static void msr_bmp_init(void)
244 {
245 	void *msr_bitmap;
246 	u32 ctrl_cpu0;
247 
248 	msr_bitmap = alloc_page();
249 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
250 	ctrl_cpu0 |= CPU_MSR_BITMAP;
251 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
252 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
253 }
254 
255 static void *get_msr_bitmap(void)
256 {
257 	void *msr_bitmap;
258 
259 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
260 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
261 	} else {
262 		msr_bitmap = alloc_page();
263 		memset(msr_bitmap, 0xff, PAGE_SIZE);
264 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
265 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
266 	}
267 
268 	return msr_bitmap;
269 }
270 
271 static void disable_intercept_for_x2apic_msrs(void)
272 {
273 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
274 	u32 msr;
275 
276 	for (msr = APIC_BASE_MSR;
277 		 msr < (APIC_BASE_MSR+0xff);
278 		 msr += BITS_PER_LONG) {
279 		unsigned int word = msr / BITS_PER_LONG;
280 
281 		msr_bitmap[word] = 0;
282 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
283 	}
284 }
285 
286 static int test_ctrl_pat_init(struct vmcs *vmcs)
287 {
288 	u64 ctrl_ent;
289 	u64 ctrl_exi;
290 
291 	msr_bmp_init();
292 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
293 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
294 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
295 		printf("\tSave/load PAT is not supported\n");
296 		return 1;
297 	}
298 
299 	ctrl_ent = vmcs_read(ENT_CONTROLS);
300 	ctrl_exi = vmcs_read(EXI_CONTROLS);
301 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
302 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
303 	vmcs_write(ENT_CONTROLS, ctrl_ent);
304 	vmcs_write(EXI_CONTROLS, ctrl_exi);
305 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
306 	vmcs_write(GUEST_PAT, 0x0);
307 	vmcs_write(HOST_PAT, ia32_pat);
308 	return VMX_TEST_START;
309 }
310 
311 static void test_ctrl_pat_main(void)
312 {
313 	u64 guest_ia32_pat;
314 
315 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
316 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
317 		printf("\tENT_LOAD_PAT is not supported.\n");
318 	else {
319 		if (guest_ia32_pat != 0) {
320 			report_fail("Entry load PAT");
321 			return;
322 		}
323 	}
324 	wrmsr(MSR_IA32_CR_PAT, 0x6);
325 	vmcall();
326 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
327 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
328 		report(guest_ia32_pat == ia32_pat, "Entry load PAT");
329 }
330 
331 static int test_ctrl_pat_exit_handler(union exit_reason exit_reason)
332 {
333 	u64 guest_rip;
334 	u64 guest_pat;
335 
336 	guest_rip = vmcs_read(GUEST_RIP);
337 	switch (exit_reason.basic) {
338 	case VMX_VMCALL:
339 		guest_pat = vmcs_read(GUEST_PAT);
340 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
341 			printf("\tEXI_SAVE_PAT is not supported\n");
342 			vmcs_write(GUEST_PAT, 0x6);
343 		} else {
344 			report(guest_pat == 0x6, "Exit save PAT");
345 		}
346 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
347 			printf("\tEXI_LOAD_PAT is not supported\n");
348 		else
349 			report(rdmsr(MSR_IA32_CR_PAT) == ia32_pat,
350 			       "Exit load PAT");
351 		vmcs_write(GUEST_PAT, ia32_pat);
352 		vmcs_write(GUEST_RIP, guest_rip + 3);
353 		return VMX_TEST_RESUME;
354 	default:
355 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
356 		break;
357 	}
358 	return VMX_TEST_VMEXIT;
359 }
360 
361 static int test_ctrl_efer_init(struct vmcs *vmcs)
362 {
363 	u64 ctrl_ent;
364 	u64 ctrl_exi;
365 
366 	msr_bmp_init();
367 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
368 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
369 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
370 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
371 	ia32_efer = rdmsr(MSR_EFER);
372 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
373 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
374 	return VMX_TEST_START;
375 }
376 
377 static void test_ctrl_efer_main(void)
378 {
379 	u64 guest_ia32_efer;
380 
381 	guest_ia32_efer = rdmsr(MSR_EFER);
382 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
383 		printf("\tENT_LOAD_EFER is not supported.\n");
384 	else {
385 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
386 			report_fail("Entry load EFER");
387 			return;
388 		}
389 	}
390 	wrmsr(MSR_EFER, ia32_efer);
391 	vmcall();
392 	guest_ia32_efer = rdmsr(MSR_EFER);
393 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
394 		report(guest_ia32_efer == ia32_efer, "Entry load EFER");
395 }
396 
397 static int test_ctrl_efer_exit_handler(union exit_reason exit_reason)
398 {
399 	u64 guest_rip;
400 	u64 guest_efer;
401 
402 	guest_rip = vmcs_read(GUEST_RIP);
403 	switch (exit_reason.basic) {
404 	case VMX_VMCALL:
405 		guest_efer = vmcs_read(GUEST_EFER);
406 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
407 			printf("\tEXI_SAVE_EFER is not supported\n");
408 			vmcs_write(GUEST_EFER, ia32_efer);
409 		} else {
410 			report(guest_efer == ia32_efer, "Exit save EFER");
411 		}
412 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
413 			printf("\tEXI_LOAD_EFER is not supported\n");
414 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
415 		} else {
416 			report(rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX),
417 			       "Exit load EFER");
418 		}
419 		vmcs_write(GUEST_PAT, ia32_efer);
420 		vmcs_write(GUEST_RIP, guest_rip + 3);
421 		return VMX_TEST_RESUME;
422 	default:
423 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
424 		break;
425 	}
426 	return VMX_TEST_VMEXIT;
427 }
428 
429 u32 guest_cr0, guest_cr4;
430 
431 static void cr_shadowing_main(void)
432 {
433 	u32 cr0, cr4, tmp;
434 
435 	// Test read through
436 	vmx_set_test_stage(0);
437 	guest_cr0 = read_cr0();
438 	if (vmx_get_test_stage() == 1)
439 		report_fail("Read through CR0");
440 	else
441 		vmcall();
442 	vmx_set_test_stage(1);
443 	guest_cr4 = read_cr4();
444 	if (vmx_get_test_stage() == 2)
445 		report_fail("Read through CR4");
446 	else
447 		vmcall();
448 	// Test write through
449 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
450 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
451 	vmx_set_test_stage(2);
452 	write_cr0(guest_cr0);
453 	if (vmx_get_test_stage() == 3)
454 		report_fail("Write throuth CR0");
455 	else
456 		vmcall();
457 	vmx_set_test_stage(3);
458 	write_cr4(guest_cr4);
459 	if (vmx_get_test_stage() == 4)
460 		report_fail("Write through CR4");
461 	else
462 		vmcall();
463 	// Test read shadow
464 	vmx_set_test_stage(4);
465 	vmcall();
466 	cr0 = read_cr0();
467 	if (vmx_get_test_stage() != 5)
468 		report(cr0 == guest_cr0, "Read shadowing CR0");
469 	vmx_set_test_stage(5);
470 	cr4 = read_cr4();
471 	if (vmx_get_test_stage() != 6)
472 		report(cr4 == guest_cr4, "Read shadowing CR4");
473 	// Test write shadow (same value with shadow)
474 	vmx_set_test_stage(6);
475 	write_cr0(guest_cr0);
476 	if (vmx_get_test_stage() == 7)
477 		report_fail("Write shadowing CR0 (same value with shadow)");
478 	else
479 		vmcall();
480 	vmx_set_test_stage(7);
481 	write_cr4(guest_cr4);
482 	if (vmx_get_test_stage() == 8)
483 		report_fail("Write shadowing CR4 (same value with shadow)");
484 	else
485 		vmcall();
486 	// Test write shadow (different value)
487 	vmx_set_test_stage(8);
488 	tmp = guest_cr0 ^ X86_CR0_TS;
489 	asm volatile("mov %0, %%rsi\n\t"
490 		"mov %%rsi, %%cr0\n\t"
491 		::"m"(tmp)
492 		:"rsi", "memory", "cc");
493 	report(vmx_get_test_stage() == 9,
494 	       "Write shadowing different X86_CR0_TS");
495 	vmx_set_test_stage(9);
496 	tmp = guest_cr0 ^ X86_CR0_MP;
497 	asm volatile("mov %0, %%rsi\n\t"
498 		"mov %%rsi, %%cr0\n\t"
499 		::"m"(tmp)
500 		:"rsi", "memory", "cc");
501 	report(vmx_get_test_stage() == 10,
502 	       "Write shadowing different X86_CR0_MP");
503 	vmx_set_test_stage(10);
504 	tmp = guest_cr4 ^ X86_CR4_TSD;
505 	asm volatile("mov %0, %%rsi\n\t"
506 		"mov %%rsi, %%cr4\n\t"
507 		::"m"(tmp)
508 		:"rsi", "memory", "cc");
509 	report(vmx_get_test_stage() == 11,
510 	       "Write shadowing different X86_CR4_TSD");
511 	vmx_set_test_stage(11);
512 	tmp = guest_cr4 ^ X86_CR4_DE;
513 	asm volatile("mov %0, %%rsi\n\t"
514 		"mov %%rsi, %%cr4\n\t"
515 		::"m"(tmp)
516 		:"rsi", "memory", "cc");
517 	report(vmx_get_test_stage() == 12,
518 	       "Write shadowing different X86_CR4_DE");
519 }
520 
521 static int cr_shadowing_exit_handler(union exit_reason exit_reason)
522 {
523 	u64 guest_rip;
524 	u32 insn_len;
525 	u32 exit_qual;
526 
527 	guest_rip = vmcs_read(GUEST_RIP);
528 	insn_len = vmcs_read(EXI_INST_LEN);
529 	exit_qual = vmcs_read(EXI_QUALIFICATION);
530 	switch (exit_reason.basic) {
531 	case VMX_VMCALL:
532 		switch (vmx_get_test_stage()) {
533 		case 0:
534 			report(guest_cr0 == vmcs_read(GUEST_CR0),
535 			       "Read through CR0");
536 			break;
537 		case 1:
538 			report(guest_cr4 == vmcs_read(GUEST_CR4),
539 			       "Read through CR4");
540 			break;
541 		case 2:
542 			report(guest_cr0 == vmcs_read(GUEST_CR0),
543 			       "Write through CR0");
544 			break;
545 		case 3:
546 			report(guest_cr4 == vmcs_read(GUEST_CR4),
547 			       "Write through CR4");
548 			break;
549 		case 4:
550 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
551 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
552 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
553 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
554 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
555 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
556 			break;
557 		case 6:
558 			report(guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)),
559 			       "Write shadowing CR0 (same value)");
560 			break;
561 		case 7:
562 			report(guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)),
563 			       "Write shadowing CR4 (same value)");
564 			break;
565 		default:
566 			// Should not reach here
567 			report_fail("unexpected stage, %d",
568 				    vmx_get_test_stage());
569 			print_vmexit_info(exit_reason);
570 			return VMX_TEST_VMEXIT;
571 		}
572 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
573 		return VMX_TEST_RESUME;
574 	case VMX_CR:
575 		switch (vmx_get_test_stage()) {
576 		case 4:
577 			report_fail("Read shadowing CR0");
578 			vmx_inc_test_stage();
579 			break;
580 		case 5:
581 			report_fail("Read shadowing CR4");
582 			vmx_inc_test_stage();
583 			break;
584 		case 6:
585 			report_fail("Write shadowing CR0 (same value)");
586 			vmx_inc_test_stage();
587 			break;
588 		case 7:
589 			report_fail("Write shadowing CR4 (same value)");
590 			vmx_inc_test_stage();
591 			break;
592 		case 8:
593 		case 9:
594 			// 0x600 encodes "mov %esi, %cr0"
595 			if (exit_qual == 0x600)
596 				vmx_inc_test_stage();
597 			break;
598 		case 10:
599 		case 11:
600 			// 0x604 encodes "mov %esi, %cr4"
601 			if (exit_qual == 0x604)
602 				vmx_inc_test_stage();
603 			break;
604 		default:
605 			// Should not reach here
606 			report_fail("unexpected stage, %d",
607 				    vmx_get_test_stage());
608 			print_vmexit_info(exit_reason);
609 			return VMX_TEST_VMEXIT;
610 		}
611 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
612 		return VMX_TEST_RESUME;
613 	default:
614 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
615 		print_vmexit_info(exit_reason);
616 	}
617 	return VMX_TEST_VMEXIT;
618 }
619 
620 static int iobmp_init(struct vmcs *vmcs)
621 {
622 	u32 ctrl_cpu0;
623 
624 	io_bitmap_a = alloc_page();
625 	io_bitmap_b = alloc_page();
626 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
627 	ctrl_cpu0 |= CPU_IO_BITMAP;
628 	ctrl_cpu0 &= (~CPU_IO);
629 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
630 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
631 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
632 	return VMX_TEST_START;
633 }
634 
635 static void iobmp_main(void)
636 {
637 	// stage 0, test IO pass
638 	vmx_set_test_stage(0);
639 	inb(0x5000);
640 	outb(0x0, 0x5000);
641 	report(vmx_get_test_stage() == 0, "I/O bitmap - I/O pass");
642 	// test IO width, in/out
643 	((u8 *)io_bitmap_a)[0] = 0xFF;
644 	vmx_set_test_stage(2);
645 	inb(0x0);
646 	report(vmx_get_test_stage() == 3, "I/O bitmap - trap in");
647 	vmx_set_test_stage(3);
648 	outw(0x0, 0x0);
649 	report(vmx_get_test_stage() == 4, "I/O bitmap - trap out");
650 	vmx_set_test_stage(4);
651 	inl(0x0);
652 	report(vmx_get_test_stage() == 5, "I/O bitmap - I/O width, long");
653 	// test low/high IO port
654 	vmx_set_test_stage(5);
655 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
656 	inb(0x5000);
657 	report(vmx_get_test_stage() == 6, "I/O bitmap - I/O port, low part");
658 	vmx_set_test_stage(6);
659 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
660 	inb(0x9000);
661 	report(vmx_get_test_stage() == 7, "I/O bitmap - I/O port, high part");
662 	// test partial pass
663 	vmx_set_test_stage(7);
664 	inl(0x4FFF);
665 	report(vmx_get_test_stage() == 8, "I/O bitmap - partial pass");
666 	// test overrun
667 	vmx_set_test_stage(8);
668 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
669 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
670 	inl(0xFFFF);
671 	report(vmx_get_test_stage() == 9, "I/O bitmap - overrun");
672 	vmx_set_test_stage(9);
673 	vmcall();
674 	outb(0x0, 0x0);
675 	report(vmx_get_test_stage() == 9,
676 	       "I/O bitmap - ignore unconditional exiting");
677 	vmx_set_test_stage(10);
678 	vmcall();
679 	outb(0x0, 0x0);
680 	report(vmx_get_test_stage() == 11,
681 	       "I/O bitmap - unconditional exiting");
682 }
683 
684 static int iobmp_exit_handler(union exit_reason exit_reason)
685 {
686 	u64 guest_rip;
687 	ulong exit_qual;
688 	u32 insn_len, ctrl_cpu0;
689 
690 	guest_rip = vmcs_read(GUEST_RIP);
691 	exit_qual = vmcs_read(EXI_QUALIFICATION);
692 	insn_len = vmcs_read(EXI_INST_LEN);
693 	switch (exit_reason.basic) {
694 	case VMX_IO:
695 		switch (vmx_get_test_stage()) {
696 		case 0:
697 		case 1:
698 			vmx_inc_test_stage();
699 			break;
700 		case 2:
701 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE,
702 			       "I/O bitmap - I/O width, byte");
703 			report(exit_qual & VMX_IO_IN,
704 			       "I/O bitmap - I/O direction, in");
705 			vmx_inc_test_stage();
706 			break;
707 		case 3:
708 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD,
709 			       "I/O bitmap - I/O width, word");
710 			report(!(exit_qual & VMX_IO_IN),
711 			       "I/O bitmap - I/O direction, out");
712 			vmx_inc_test_stage();
713 			break;
714 		case 4:
715 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG,
716 			       "I/O bitmap - I/O width, long");
717 			vmx_inc_test_stage();
718 			break;
719 		case 5:
720 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
721 				vmx_inc_test_stage();
722 			break;
723 		case 6:
724 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
725 				vmx_inc_test_stage();
726 			break;
727 		case 7:
728 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
729 				vmx_inc_test_stage();
730 			break;
731 		case 8:
732 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
733 				vmx_inc_test_stage();
734 			break;
735 		case 9:
736 		case 10:
737 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
738 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
739 			vmx_inc_test_stage();
740 			break;
741 		default:
742 			// Should not reach here
743 			report_fail("unexpected stage, %d",
744 				    vmx_get_test_stage());
745 			print_vmexit_info(exit_reason);
746 			return VMX_TEST_VMEXIT;
747 		}
748 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
749 		return VMX_TEST_RESUME;
750 	case VMX_VMCALL:
751 		switch (vmx_get_test_stage()) {
752 		case 9:
753 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
754 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
755 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
756 			break;
757 		case 10:
758 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
759 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
760 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
761 			break;
762 		default:
763 			// Should not reach here
764 			report_fail("unexpected stage, %d",
765 				    vmx_get_test_stage());
766 			print_vmexit_info(exit_reason);
767 			return VMX_TEST_VMEXIT;
768 		}
769 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
770 		return VMX_TEST_RESUME;
771 	default:
772 		printf("guest_rip = %#lx\n", guest_rip);
773 		printf("\tERROR : Unknown exit reason, 0x%x\n", exit_reason.full);
774 		break;
775 	}
776 	return VMX_TEST_VMEXIT;
777 }
778 
779 #define INSN_CPU0		0
780 #define INSN_CPU1		1
781 #define INSN_ALWAYS_TRAP	2
782 
783 #define FIELD_EXIT_QUAL		(1 << 0)
784 #define FIELD_INSN_INFO		(1 << 1)
785 
786 asm(
787 	"insn_hlt: hlt;ret\n\t"
788 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
789 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
790 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
791 	"insn_rdtsc: rdtsc;ret\n\t"
792 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
793 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
794 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
795 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
796 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
797 	"insn_pause: pause;ret\n\t"
798 	"insn_wbinvd: wbinvd;ret\n\t"
799 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
800 	"insn_invd: invd;ret\n\t"
801 	"insn_sgdt: sgdt gdt_descr;ret\n\t"
802 	"insn_lgdt: lgdt gdt_descr;ret\n\t"
803 	"insn_sidt: sidt idt_descr;ret\n\t"
804 	"insn_lidt: lidt idt_descr;ret\n\t"
805 	"insn_sldt: sldt %ax;ret\n\t"
806 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
807 	"insn_str: str %ax;ret\n\t"
808 	"insn_rdrand: rdrand %rax;ret\n\t"
809 	"insn_rdseed: rdseed %rax;ret\n\t"
810 );
811 extern void insn_hlt(void);
812 extern void insn_invlpg(void);
813 extern void insn_mwait(void);
814 extern void insn_rdpmc(void);
815 extern void insn_rdtsc(void);
816 extern void insn_cr3_load(void);
817 extern void insn_cr3_store(void);
818 extern void insn_cr8_load(void);
819 extern void insn_cr8_store(void);
820 extern void insn_monitor(void);
821 extern void insn_pause(void);
822 extern void insn_wbinvd(void);
823 extern void insn_sgdt(void);
824 extern void insn_lgdt(void);
825 extern void insn_sidt(void);
826 extern void insn_lidt(void);
827 extern void insn_sldt(void);
828 extern void insn_lldt(void);
829 extern void insn_str(void);
830 extern void insn_cpuid(void);
831 extern void insn_invd(void);
832 extern void insn_rdrand(void);
833 extern void insn_rdseed(void);
834 
835 u32 cur_insn;
836 u64 cr3;
837 
838 #define X86_FEATURE_MONITOR	(1 << 3)
839 
840 typedef bool (*supported_fn)(void);
841 
842 static bool monitor_supported(void)
843 {
844 	return this_cpu_has(X86_FEATURE_MWAIT);
845 }
846 
847 struct insn_table {
848 	const char *name;
849 	u32 flag;
850 	void (*insn_func)(void);
851 	u32 type;
852 	u32 reason;
853 	ulong exit_qual;
854 	u32 insn_info;
855 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
856 	// which field need to be tested, reason is always tested
857 	u32 test_field;
858 	const supported_fn supported_fn;
859 	u8 disabled;
860 };
861 
862 /*
863  * Add more test cases of instruction intercept here. Elements in this
864  * table is:
865  *	name/control flag/insn function/type/exit reason/exit qulification/
866  *	instruction info/field to test
867  * The last field defines which fields (exit_qual and insn_info) need to be
868  * tested in exit handler. If set to 0, only "reason" is checked.
869  */
870 static struct insn_table insn_table[] = {
871 	// Flags for Primary Processor-Based VM-Execution Controls
872 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
873 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
874 		0x12345678, 0, FIELD_EXIT_QUAL},
875 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0, &monitor_supported},
876 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0},
877 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
878 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
879 		FIELD_EXIT_QUAL},
880 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
881 		FIELD_EXIT_QUAL},
882 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
883 		FIELD_EXIT_QUAL},
884 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
885 		FIELD_EXIT_QUAL},
886 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0, &monitor_supported},
887 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
888 	// Flags for Secondary Processor-Based VM-Execution Controls
889 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
890 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
891 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
892 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
893 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
894 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
895 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
896 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
897 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
898 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
899 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
900 	// Instructions always trap
901 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
902 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
903 	// Instructions never trap
904 	{NULL},
905 };
906 
907 static int insn_intercept_init(struct vmcs *vmcs)
908 {
909 	u32 ctrl_cpu, cur_insn;
910 
911 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
912 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
913 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
914 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
915 	cr3 = read_cr3();
916 
917 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
918 		if (insn_table[cur_insn].supported_fn == NULL)
919 			continue;
920 		insn_table[cur_insn].disabled = !insn_table[cur_insn].supported_fn();
921 	}
922 	return VMX_TEST_START;
923 }
924 
925 static void insn_intercept_main(void)
926 {
927 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
928 		vmx_set_test_stage(cur_insn * 2);
929 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
930 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
931 		    (insn_table[cur_insn].type == INSN_CPU1 &&
932 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
933 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
934 			       insn_table[cur_insn].type - INSN_CPU0,
935 			       insn_table[cur_insn].name);
936 			continue;
937 		}
938 
939 		if (insn_table[cur_insn].disabled) {
940 			printf("\tFeature required for %s is not supported.\n",
941 			       insn_table[cur_insn].name);
942 			continue;
943 		}
944 
945 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
946 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
947 		    (insn_table[cur_insn].type == INSN_CPU1 &&
948 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
949 			/* skip hlt, it stalls the guest and is tested below */
950 			if (insn_table[cur_insn].insn_func != insn_hlt)
951 				insn_table[cur_insn].insn_func();
952 			report(vmx_get_test_stage() == cur_insn * 2,
953 					"execute %s",
954 					insn_table[cur_insn].name);
955 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
956 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
957 			       insn_table[cur_insn].type - INSN_CPU0,
958 			       insn_table[cur_insn].name);
959 
960 		vmcall();
961 
962 		insn_table[cur_insn].insn_func();
963 		report(vmx_get_test_stage() == cur_insn * 2 + 1,
964 				"intercept %s",
965 				insn_table[cur_insn].name);
966 
967 		vmx_set_test_stage(cur_insn * 2 + 1);
968 		vmcall();
969 	}
970 }
971 
972 static int insn_intercept_exit_handler(union exit_reason exit_reason)
973 {
974 	u64 guest_rip;
975 	ulong exit_qual;
976 	u32 insn_len;
977 	u32 insn_info;
978 	bool pass;
979 
980 	guest_rip = vmcs_read(GUEST_RIP);
981 	exit_qual = vmcs_read(EXI_QUALIFICATION);
982 	insn_len = vmcs_read(EXI_INST_LEN);
983 	insn_info = vmcs_read(EXI_INST_INFO);
984 
985 	if (exit_reason.basic == VMX_VMCALL) {
986 		u32 val = 0;
987 
988 		if (insn_table[cur_insn].type == INSN_CPU0)
989 			val = vmcs_read(CPU_EXEC_CTRL0);
990 		else if (insn_table[cur_insn].type == INSN_CPU1)
991 			val = vmcs_read(CPU_EXEC_CTRL1);
992 
993 		if (vmx_get_test_stage() & 1)
994 			val &= ~insn_table[cur_insn].flag;
995 		else
996 			val |= insn_table[cur_insn].flag;
997 
998 		if (insn_table[cur_insn].type == INSN_CPU0)
999 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
1000 		else if (insn_table[cur_insn].type == INSN_CPU1)
1001 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
1002 	} else {
1003 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
1004 			insn_table[cur_insn].reason == exit_reason.full;
1005 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
1006 		    insn_table[cur_insn].exit_qual != exit_qual)
1007 			pass = false;
1008 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
1009 		    insn_table[cur_insn].insn_info != insn_info)
1010 			pass = false;
1011 		if (pass)
1012 			vmx_inc_test_stage();
1013 	}
1014 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1015 	return VMX_TEST_RESUME;
1016 }
1017 
1018 /**
1019  * __setup_ept - Setup the VMCS fields to enable Extended Page Tables (EPT)
1020  * @hpa:	Host physical address of the top-level, a.k.a. root, EPT table
1021  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1022  *
1023  * Returns 0 on success, 1 on failure.
1024  *
1025  * Note that @hpa doesn't need to point at actual memory if VM-Launch is
1026  * expected to fail, e.g. setup_dummy_ept() arbitrarily passes '0' to satisfy
1027  * the various EPTP consistency checks, but doesn't ensure backing for HPA '0'.
1028  */
1029 static int __setup_ept(u64 hpa, bool enable_ad)
1030 {
1031 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1032 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1033 		printf("\tEPT is not supported\n");
1034 		return 1;
1035 	}
1036 	if (!(ept_vpid.val & EPT_CAP_WB)) {
1037 		printf("\tWB memtype for EPT walks not supported\n");
1038 		return 1;
1039 	}
1040 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
1041 		printf("\tPWL4 is not supported\n");
1042 		return 1;
1043 	}
1044 
1045 	eptp = EPT_MEM_TYPE_WB;
1046 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1047 	eptp |= hpa;
1048 	if (enable_ad)
1049 		eptp |= EPTP_AD_FLAG;
1050 
1051 	vmcs_write(EPTP, eptp);
1052 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY);
1053 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT);
1054 
1055 	return 0;
1056 }
1057 
1058 /**
1059  * setup_ept - Enable Extended Page Tables (EPT) and setup an identity map
1060  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1061  *
1062  * Returns 0 on success, 1 on failure.
1063  *
1064  * This is the "real" function for setting up EPT tables, i.e. use this for
1065  * tests that need to run code in the guest with EPT enabled.
1066  */
1067 static int setup_ept(bool enable_ad)
1068 {
1069 	unsigned long end_of_memory;
1070 
1071 	pml4 = alloc_page();
1072 
1073 	if (__setup_ept(virt_to_phys(pml4), enable_ad))
1074 		return 1;
1075 
1076 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1077 	if (end_of_memory < (1ul << 32))
1078 		end_of_memory = (1ul << 32);
1079 	/* Cannot use large EPT pages if we need to track EPT
1080 	 * accessed/dirty bits at 4K granularity.
1081 	 */
1082 	setup_ept_range(pml4, 0, end_of_memory, 0,
1083 			!enable_ad && ept_2m_supported(),
1084 			EPT_WA | EPT_RA | EPT_EA);
1085 	return 0;
1086 }
1087 
1088 /**
1089  * setup_dummy_ept - Enable Extended Page Tables (EPT) with a dummy root HPA
1090  *
1091  * Setup EPT using a semi-arbitrary dummy root HPA.  This function is intended
1092  * for use by tests that need EPT enabled to verify dependent VMCS controls
1093  * but never expect to fully enter the guest, i.e. don't need setup the actual
1094  * EPT tables.
1095  */
1096 static void setup_dummy_ept(void)
1097 {
1098 	if (__setup_ept(0, false))
1099 		report_abort("EPT setup unexpectedly failed");
1100 }
1101 
1102 static int enable_unrestricted_guest(bool need_valid_ept)
1103 {
1104 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1105 	    !(ctrl_cpu_rev[1].clr & CPU_URG) ||
1106 	    !(ctrl_cpu_rev[1].clr & CPU_EPT))
1107 		return 1;
1108 
1109 	if (need_valid_ept)
1110 		setup_ept(false);
1111 	else
1112 		setup_dummy_ept();
1113 
1114 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
1115 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
1116 
1117 	return 0;
1118 }
1119 
1120 static void ept_enable_ad_bits(void)
1121 {
1122 	eptp |= EPTP_AD_FLAG;
1123 	vmcs_write(EPTP, eptp);
1124 }
1125 
1126 static void ept_disable_ad_bits(void)
1127 {
1128 	eptp &= ~EPTP_AD_FLAG;
1129 	vmcs_write(EPTP, eptp);
1130 }
1131 
1132 static int ept_ad_enabled(void)
1133 {
1134 	return eptp & EPTP_AD_FLAG;
1135 }
1136 
1137 static void ept_enable_ad_bits_or_skip_test(void)
1138 {
1139 	if (!ept_ad_bits_supported())
1140 		test_skip("EPT AD bits not supported.");
1141 	ept_enable_ad_bits();
1142 }
1143 
1144 static int apic_version;
1145 
1146 static int ept_init_common(bool have_ad)
1147 {
1148 	int ret;
1149 	struct pci_dev pcidev;
1150 
1151 	/* INVEPT is required by the EPT violation handler. */
1152 	if (!is_invept_type_supported(INVEPT_SINGLE))
1153 		return VMX_TEST_EXIT;
1154 
1155 	if (setup_ept(have_ad))
1156 		return VMX_TEST_EXIT;
1157 
1158 	data_page1 = alloc_page();
1159 	data_page2 = alloc_page();
1160 	*((u32 *)data_page1) = MAGIC_VAL_1;
1161 	*((u32 *)data_page2) = MAGIC_VAL_2;
1162 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1163 			EPT_RA | EPT_WA | EPT_EA);
1164 
1165 	apic_version = apic_read(APIC_LVR);
1166 
1167 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1168 	if (ret != PCIDEVADDR_INVALID) {
1169 		pci_dev_init(&pcidev, ret);
1170 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1171 	}
1172 
1173 	return VMX_TEST_START;
1174 }
1175 
1176 static int ept_init(struct vmcs *vmcs)
1177 {
1178 	return ept_init_common(false);
1179 }
1180 
1181 static void ept_common(void)
1182 {
1183 	vmx_set_test_stage(0);
1184 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1185 			*((u32 *)data_page1) != MAGIC_VAL_1)
1186 		report_fail("EPT basic framework - read");
1187 	else {
1188 		*((u32 *)data_page2) = MAGIC_VAL_3;
1189 		vmcall();
1190 		if (vmx_get_test_stage() == 1) {
1191 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1192 					*((u32 *)data_page2) == MAGIC_VAL_2)
1193 				report_pass("EPT basic framework");
1194 			else
1195 				report_pass("EPT basic framework - remap");
1196 		}
1197 	}
1198 	// Test EPT Misconfigurations
1199 	vmx_set_test_stage(1);
1200 	vmcall();
1201 	*((u32 *)data_page1) = MAGIC_VAL_1;
1202 	if (vmx_get_test_stage() != 2) {
1203 		report_fail("EPT misconfigurations");
1204 		goto t1;
1205 	}
1206 	vmx_set_test_stage(2);
1207 	vmcall();
1208 	*((u32 *)data_page1) = MAGIC_VAL_1;
1209 	report(vmx_get_test_stage() == 3, "EPT misconfigurations");
1210 t1:
1211 	// Test EPT violation
1212 	vmx_set_test_stage(3);
1213 	vmcall();
1214 	*((u32 *)data_page1) = MAGIC_VAL_1;
1215 	report(vmx_get_test_stage() == 4, "EPT violation - page permission");
1216 	// Violation caused by EPT paging structure
1217 	vmx_set_test_stage(4);
1218 	vmcall();
1219 	*((u32 *)data_page1) = MAGIC_VAL_2;
1220 	report(vmx_get_test_stage() == 5, "EPT violation - paging structure");
1221 
1222 	// MMIO Read/Write
1223 	vmx_set_test_stage(5);
1224 	vmcall();
1225 
1226 	*(u32 volatile *)pci_physaddr;
1227 	report(vmx_get_test_stage() == 6, "MMIO EPT violation - read");
1228 
1229 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1230 	report(vmx_get_test_stage() == 7, "MMIO EPT violation - write");
1231 }
1232 
1233 static void ept_main(void)
1234 {
1235 	ept_common();
1236 
1237 	// Test EPT access to L1 MMIO
1238 	vmx_set_test_stage(7);
1239 	report(*((u32 *)0xfee00030UL) == apic_version, "EPT - MMIO access");
1240 
1241 	// Test invalid operand for INVEPT
1242 	vmcall();
1243 	report(vmx_get_test_stage() == 8, "EPT - unsupported INVEPT");
1244 }
1245 
1246 static bool invept_test(int type, u64 eptp)
1247 {
1248 	bool ret, supported;
1249 
1250 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1251 	ret = __invept(type, eptp);
1252 
1253 	if (ret == !supported)
1254 		return false;
1255 
1256 	if (!supported)
1257 		printf("WARNING: unsupported invept passed!\n");
1258 	else
1259 		printf("WARNING: invept failed!\n");
1260 
1261 	return true;
1262 }
1263 
1264 static int pml_exit_handler(union exit_reason exit_reason)
1265 {
1266 	u16 index, count;
1267 	u64 *pmlbuf = pml_log;
1268 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1269 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1270 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1271 
1272 	switch (exit_reason.basic) {
1273 	case VMX_VMCALL:
1274 		switch (vmx_get_test_stage()) {
1275 		case 0:
1276 			index = vmcs_read(GUEST_PML_INDEX);
1277 			for (count = index + 1; count < PML_INDEX; count++) {
1278 				if (pmlbuf[count] == (u64)data_page2) {
1279 					vmx_inc_test_stage();
1280 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1281 					break;
1282 				}
1283 			}
1284 			break;
1285 		case 1:
1286 			index = vmcs_read(GUEST_PML_INDEX);
1287 			/* Keep clearing the dirty bit till a overflow */
1288 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1289 			break;
1290 		default:
1291 			report_fail("unexpected stage, %d.",
1292 			       vmx_get_test_stage());
1293 			print_vmexit_info(exit_reason);
1294 			return VMX_TEST_VMEXIT;
1295 		}
1296 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1297 		return VMX_TEST_RESUME;
1298 	case VMX_PML_FULL:
1299 		vmx_inc_test_stage();
1300 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1301 		return VMX_TEST_RESUME;
1302 	default:
1303 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1304 		print_vmexit_info(exit_reason);
1305 	}
1306 	return VMX_TEST_VMEXIT;
1307 }
1308 
1309 static int ept_exit_handler_common(union exit_reason exit_reason, bool have_ad)
1310 {
1311 	u64 guest_rip;
1312 	u64 guest_cr3;
1313 	u32 insn_len;
1314 	u32 exit_qual;
1315 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte,
1316 			     guest_pte_addr;
1317 
1318 	guest_rip = vmcs_read(GUEST_RIP);
1319 	guest_cr3 = vmcs_read(GUEST_CR3);
1320 	insn_len = vmcs_read(EXI_INST_LEN);
1321 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1322 	pteval_t *ptep;
1323 	switch (exit_reason.basic) {
1324 	case VMX_VMCALL:
1325 		switch (vmx_get_test_stage()) {
1326 		case 0:
1327 			check_ept_ad(pml4, guest_cr3,
1328 				     (unsigned long)data_page1,
1329 				     have_ad ? EPT_ACCESS_FLAG : 0,
1330 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1331 			check_ept_ad(pml4, guest_cr3,
1332 				     (unsigned long)data_page2,
1333 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1334 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1335 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1336 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1337 			if (have_ad)
1338 				invept(INVEPT_SINGLE, eptp);
1339 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1340 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1341 				vmx_inc_test_stage();
1342 				install_ept(pml4, (unsigned long)data_page2,
1343 						(unsigned long)data_page2,
1344 						EPT_RA | EPT_WA | EPT_EA);
1345 			} else
1346 				report_fail("EPT basic framework - write");
1347 			break;
1348 		case 1:
1349 			install_ept(pml4, (unsigned long)data_page1,
1350  				(unsigned long)data_page1, EPT_WA);
1351 			invept(INVEPT_SINGLE, eptp);
1352 			break;
1353 		case 2:
1354 			install_ept(pml4, (unsigned long)data_page1,
1355  				(unsigned long)data_page1,
1356  				EPT_RA | EPT_WA | EPT_EA |
1357  				(2 << EPT_MEM_TYPE_SHIFT));
1358 			invept(INVEPT_SINGLE, eptp);
1359 			break;
1360 		case 3:
1361 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1362 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1363 						1, &data_page1_pte));
1364 			set_ept_pte(pml4, (unsigned long)data_page1,
1365 				1, data_page1_pte & ~EPT_PRESENT);
1366 			invept(INVEPT_SINGLE, eptp);
1367 			break;
1368 		case 4:
1369 			ptep = get_pte_level((pgd_t *)guest_cr3, data_page1, /*level=*/2);
1370 			guest_pte_addr = virt_to_phys(ptep) & PAGE_MASK;
1371 
1372 			TEST_ASSERT(get_ept_pte(pml4, guest_pte_addr, 2, &data_page1_pte_pte));
1373 			set_ept_pte(pml4, guest_pte_addr, 2,
1374 				data_page1_pte_pte & ~EPT_PRESENT);
1375 			invept(INVEPT_SINGLE, eptp);
1376 			break;
1377 		case 5:
1378 			install_ept(pml4, (unsigned long)pci_physaddr,
1379 				(unsigned long)pci_physaddr, 0);
1380 			invept(INVEPT_SINGLE, eptp);
1381 			break;
1382 		case 7:
1383 			if (!invept_test(0, eptp))
1384 				vmx_inc_test_stage();
1385 			break;
1386 		// Should not reach here
1387 		default:
1388 			report_fail("ERROR - unexpected stage, %d.",
1389 			       vmx_get_test_stage());
1390 			print_vmexit_info(exit_reason);
1391 			return VMX_TEST_VMEXIT;
1392 		}
1393 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1394 		return VMX_TEST_RESUME;
1395 	case VMX_EPT_MISCONFIG:
1396 		switch (vmx_get_test_stage()) {
1397 		case 1:
1398 		case 2:
1399 			vmx_inc_test_stage();
1400 			install_ept(pml4, (unsigned long)data_page1,
1401  				(unsigned long)data_page1,
1402  				EPT_RA | EPT_WA | EPT_EA);
1403 			invept(INVEPT_SINGLE, eptp);
1404 			break;
1405 		// Should not reach here
1406 		default:
1407 			report_fail("ERROR - unexpected stage, %d.",
1408 			       vmx_get_test_stage());
1409 			print_vmexit_info(exit_reason);
1410 			return VMX_TEST_VMEXIT;
1411 		}
1412 		return VMX_TEST_RESUME;
1413 	case VMX_EPT_VIOLATION:
1414 		/*
1415 		 * Exit-qualifications are masked not to account for advanced
1416 		 * VM-exit information. Once KVM supports this feature, this
1417 		 * masking should be removed.
1418 		 */
1419 		exit_qual &= ~EPT_VLT_GUEST_MASK;
1420 
1421 		switch(vmx_get_test_stage()) {
1422 		case 3:
1423 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1424 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1425 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1426 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1427 					EPT_VLT_PADDR))
1428 				vmx_inc_test_stage();
1429 			set_ept_pte(pml4, (unsigned long)data_page1,
1430 				1, data_page1_pte | (EPT_PRESENT));
1431 			invept(INVEPT_SINGLE, eptp);
1432 			break;
1433 		case 4:
1434 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1435 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1436 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1437 			if (exit_qual == (EPT_VLT_RD |
1438 					  (have_ad ? EPT_VLT_WR : 0) |
1439 					  EPT_VLT_LADDR_VLD))
1440 				vmx_inc_test_stage();
1441 			set_ept_pte(pml4, guest_pte_addr, 2,
1442 				data_page1_pte_pte | (EPT_PRESENT));
1443 			invept(INVEPT_SINGLE, eptp);
1444 			break;
1445 		case 5:
1446 			if (exit_qual & EPT_VLT_RD)
1447 				vmx_inc_test_stage();
1448 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1449 						1, &memaddr_pte));
1450 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1451 			invept(INVEPT_SINGLE, eptp);
1452 			break;
1453 		case 6:
1454 			if (exit_qual & EPT_VLT_WR)
1455 				vmx_inc_test_stage();
1456 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1457 						1, &memaddr_pte));
1458 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1459 			invept(INVEPT_SINGLE, eptp);
1460 			break;
1461 		default:
1462 			// Should not reach here
1463 			report_fail("ERROR : unexpected stage, %d",
1464 			       vmx_get_test_stage());
1465 			print_vmexit_info(exit_reason);
1466 			return VMX_TEST_VMEXIT;
1467 		}
1468 		return VMX_TEST_RESUME;
1469 	default:
1470 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1471 		print_vmexit_info(exit_reason);
1472 	}
1473 	return VMX_TEST_VMEXIT;
1474 }
1475 
1476 static int ept_exit_handler(union exit_reason exit_reason)
1477 {
1478 	return ept_exit_handler_common(exit_reason, false);
1479 }
1480 
1481 static int eptad_init(struct vmcs *vmcs)
1482 {
1483 	int r = ept_init_common(true);
1484 
1485 	if (r == VMX_TEST_EXIT)
1486 		return r;
1487 
1488 	if ((rdmsr(MSR_IA32_VMX_EPT_VPID_CAP) & EPT_CAP_AD_FLAG) == 0) {
1489 		printf("\tEPT A/D bits are not supported");
1490 		return VMX_TEST_EXIT;
1491 	}
1492 
1493 	return r;
1494 }
1495 
1496 static int pml_init(struct vmcs *vmcs)
1497 {
1498 	u32 ctrl_cpu;
1499 	int r = eptad_init(vmcs);
1500 
1501 	if (r == VMX_TEST_EXIT)
1502 		return r;
1503 
1504 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1505 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1506 		printf("\tPML is not supported");
1507 		return VMX_TEST_EXIT;
1508 	}
1509 
1510 	pml_log = alloc_page();
1511 	vmcs_write(PMLADDR, (u64)pml_log);
1512 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1513 
1514 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1515 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1516 
1517 	return VMX_TEST_START;
1518 }
1519 
1520 static void pml_main(void)
1521 {
1522 	int count = 0;
1523 
1524 	vmx_set_test_stage(0);
1525 	*((u32 *)data_page2) = 0x1;
1526 	vmcall();
1527 	report(vmx_get_test_stage() == 1, "PML - Dirty GPA Logging");
1528 
1529 	while (vmx_get_test_stage() == 1) {
1530 		vmcall();
1531 		*((u32 *)data_page2) = 0x1;
1532 		if (count++ > PML_INDEX)
1533 			break;
1534 	}
1535 	report(vmx_get_test_stage() == 2, "PML Full Event");
1536 }
1537 
1538 static void eptad_main(void)
1539 {
1540 	ept_common();
1541 }
1542 
1543 static int eptad_exit_handler(union exit_reason exit_reason)
1544 {
1545 	return ept_exit_handler_common(exit_reason, true);
1546 }
1547 
1548 #define TIMER_VECTOR	222
1549 
1550 static volatile bool timer_fired;
1551 
1552 static void timer_isr(isr_regs_t *regs)
1553 {
1554 	timer_fired = true;
1555 	apic_write(APIC_EOI, 0);
1556 }
1557 
1558 static int interrupt_init(struct vmcs *vmcs)
1559 {
1560 	msr_bmp_init();
1561 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1562 	handle_irq(TIMER_VECTOR, timer_isr);
1563 	return VMX_TEST_START;
1564 }
1565 
1566 static void interrupt_main(void)
1567 {
1568 	long long start, loops;
1569 
1570 	vmx_set_test_stage(0);
1571 
1572 	apic_write(APIC_LVTT, TIMER_VECTOR);
1573 	irq_enable();
1574 
1575 	apic_write(APIC_TMICT, 1);
1576 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1577 		asm volatile ("nop");
1578 	report(timer_fired, "direct interrupt while running guest");
1579 
1580 	apic_write(APIC_TMICT, 0);
1581 	irq_disable();
1582 	vmcall();
1583 	timer_fired = false;
1584 	apic_write(APIC_TMICT, 1);
1585 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1586 		asm volatile ("nop");
1587 	report(timer_fired, "intercepted interrupt while running guest");
1588 
1589 	irq_enable();
1590 	apic_write(APIC_TMICT, 0);
1591 	irq_disable();
1592 	vmcall();
1593 	timer_fired = false;
1594 	start = rdtsc();
1595 	apic_write(APIC_TMICT, 1000000);
1596 
1597 	safe_halt();
1598 
1599 	report(rdtsc() - start > 1000000 && timer_fired,
1600 	       "direct interrupt + hlt");
1601 
1602 	apic_write(APIC_TMICT, 0);
1603 	irq_disable();
1604 	vmcall();
1605 	timer_fired = false;
1606 	start = rdtsc();
1607 	apic_write(APIC_TMICT, 1000000);
1608 
1609 	safe_halt();
1610 
1611 	report(rdtsc() - start > 10000 && timer_fired,
1612 	       "intercepted interrupt + hlt");
1613 
1614 	apic_write(APIC_TMICT, 0);
1615 	irq_disable();
1616 	vmcall();
1617 	timer_fired = false;
1618 	start = rdtsc();
1619 	apic_write(APIC_TMICT, 1000000);
1620 
1621 	irq_enable();
1622 	asm volatile ("nop");
1623 	vmcall();
1624 
1625 	report(rdtsc() - start > 10000 && timer_fired,
1626 	       "direct interrupt + activity state hlt");
1627 
1628 	apic_write(APIC_TMICT, 0);
1629 	irq_disable();
1630 	vmcall();
1631 	timer_fired = false;
1632 	start = rdtsc();
1633 	apic_write(APIC_TMICT, 1000000);
1634 
1635 	irq_enable();
1636 	asm volatile ("nop");
1637 	vmcall();
1638 
1639 	report(rdtsc() - start > 10000 && timer_fired,
1640 	       "intercepted interrupt + activity state hlt");
1641 
1642 	apic_write(APIC_TMICT, 0);
1643 	irq_disable();
1644 	vmx_set_test_stage(7);
1645 	vmcall();
1646 	timer_fired = false;
1647 	apic_write(APIC_TMICT, 1);
1648 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1649 		asm volatile ("nop");
1650 	report(timer_fired,
1651 	       "running a guest with interrupt acknowledgement set");
1652 
1653 	apic_write(APIC_TMICT, 0);
1654 	irq_enable();
1655 	timer_fired = false;
1656 	vmcall();
1657 	report(timer_fired, "Inject an event to a halted guest");
1658 }
1659 
1660 static int interrupt_exit_handler(union exit_reason exit_reason)
1661 {
1662 	u64 guest_rip = vmcs_read(GUEST_RIP);
1663 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1664 
1665 	switch (exit_reason.basic) {
1666 	case VMX_VMCALL:
1667 		switch (vmx_get_test_stage()) {
1668 		case 0:
1669 		case 2:
1670 		case 5:
1671 			vmcs_write(PIN_CONTROLS,
1672 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1673 			break;
1674 		case 7:
1675 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1676 			vmcs_write(PIN_CONTROLS,
1677 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1678 			break;
1679 		case 1:
1680 		case 3:
1681 			vmcs_write(PIN_CONTROLS,
1682 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1683 			break;
1684 		case 4:
1685 		case 6:
1686 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1687 			break;
1688 
1689 		case 8:
1690 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1691 			vmcs_write(ENT_INTR_INFO,
1692 				   TIMER_VECTOR |
1693 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1694 				   INTR_INFO_VALID_MASK);
1695 			break;
1696 		}
1697 		vmx_inc_test_stage();
1698 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1699 		return VMX_TEST_RESUME;
1700 	case VMX_EXTINT:
1701 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1702 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1703 			handle_external_interrupt(vector);
1704 		} else {
1705 			irq_enable();
1706 			asm volatile ("nop");
1707 			irq_disable();
1708 		}
1709 		if (vmx_get_test_stage() >= 2)
1710 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1711 		return VMX_TEST_RESUME;
1712 	default:
1713 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1714 		print_vmexit_info(exit_reason);
1715 	}
1716 
1717 	return VMX_TEST_VMEXIT;
1718 }
1719 
1720 
1721 static volatile int nmi_fired;
1722 
1723 #define NMI_DELAY 100000000ULL
1724 
1725 static void nmi_isr(isr_regs_t *regs)
1726 {
1727 	nmi_fired = true;
1728 }
1729 
1730 static int nmi_hlt_init(struct vmcs *vmcs)
1731 {
1732 	msr_bmp_init();
1733 	handle_irq(NMI_VECTOR, nmi_isr);
1734 	vmcs_write(PIN_CONTROLS,
1735 		   vmcs_read(PIN_CONTROLS) & ~PIN_NMI);
1736 	vmcs_write(PIN_CONTROLS,
1737 		   vmcs_read(PIN_CONTROLS) & ~PIN_VIRT_NMI);
1738 	return VMX_TEST_START;
1739 }
1740 
1741 static void nmi_message_thread(void *data)
1742 {
1743     while (vmx_get_test_stage() != 1)
1744         pause();
1745 
1746     delay(NMI_DELAY);
1747     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1748 
1749     while (vmx_get_test_stage() != 2)
1750         pause();
1751 
1752     delay(NMI_DELAY);
1753     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1754 }
1755 
1756 static void nmi_hlt_main(void)
1757 {
1758     long long start;
1759 
1760     if (cpu_count() < 2) {
1761         report_skip(__func__);
1762         vmx_set_test_stage(-1);
1763         return;
1764     }
1765 
1766     vmx_set_test_stage(0);
1767     on_cpu_async(1, nmi_message_thread, NULL);
1768     start = rdtsc();
1769     vmx_set_test_stage(1);
1770     asm volatile ("hlt");
1771     report((rdtsc() - start > NMI_DELAY) && nmi_fired,
1772             "direct NMI + hlt");
1773     if (!nmi_fired)
1774         vmx_set_test_stage(-1);
1775     nmi_fired = false;
1776 
1777     vmcall();
1778 
1779     start = rdtsc();
1780     vmx_set_test_stage(2);
1781     asm volatile ("hlt");
1782     report((rdtsc() - start > NMI_DELAY) && !nmi_fired,
1783             "intercepted NMI + hlt");
1784     if (nmi_fired) {
1785         report(!nmi_fired, "intercepted NMI was dispatched");
1786         vmx_set_test_stage(-1);
1787         return;
1788     }
1789     vmx_set_test_stage(3);
1790 }
1791 
1792 static int nmi_hlt_exit_handler(union exit_reason exit_reason)
1793 {
1794     u64 guest_rip = vmcs_read(GUEST_RIP);
1795     u32 insn_len = vmcs_read(EXI_INST_LEN);
1796 
1797     switch (vmx_get_test_stage()) {
1798     case 1:
1799         if (exit_reason.basic != VMX_VMCALL) {
1800             report_fail("VMEXIT not due to vmcall. Exit reason 0x%x",
1801                         exit_reason.full);
1802             print_vmexit_info(exit_reason);
1803             return VMX_TEST_VMEXIT;
1804         }
1805 
1806         vmcs_write(PIN_CONTROLS,
1807                vmcs_read(PIN_CONTROLS) | PIN_NMI);
1808         vmcs_write(PIN_CONTROLS,
1809                vmcs_read(PIN_CONTROLS) | PIN_VIRT_NMI);
1810         vmcs_write(GUEST_RIP, guest_rip + insn_len);
1811         break;
1812 
1813     case 2:
1814         if (exit_reason.basic != VMX_EXC_NMI) {
1815             report_fail("VMEXIT not due to NMI intercept. Exit reason 0x%x",
1816                         exit_reason.full);
1817             print_vmexit_info(exit_reason);
1818             return VMX_TEST_VMEXIT;
1819         }
1820         report_pass("NMI intercept while running guest");
1821         vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1822         break;
1823 
1824     case 3:
1825         break;
1826 
1827     default:
1828         return VMX_TEST_VMEXIT;
1829     }
1830 
1831     if (vmx_get_test_stage() == 3)
1832         return VMX_TEST_VMEXIT;
1833 
1834     return VMX_TEST_RESUME;
1835 }
1836 
1837 
1838 static int dbgctls_init(struct vmcs *vmcs)
1839 {
1840 	u64 dr7 = 0x402;
1841 	u64 zero = 0;
1842 
1843 	msr_bmp_init();
1844 	asm volatile(
1845 		"mov %0,%%dr0\n\t"
1846 		"mov %0,%%dr1\n\t"
1847 		"mov %0,%%dr2\n\t"
1848 		"mov %1,%%dr7\n\t"
1849 		: : "r" (zero), "r" (dr7));
1850 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1851 	vmcs_write(GUEST_DR7, 0x404);
1852 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1853 
1854 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1855 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1856 
1857 	return VMX_TEST_START;
1858 }
1859 
1860 static void dbgctls_main(void)
1861 {
1862 	u64 dr7, debugctl;
1863 
1864 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1865 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1866 	/* Commented out: KVM does not support DEBUGCTL so far */
1867 	(void)debugctl;
1868 	report(dr7 == 0x404, "Load debug controls" /* && debugctl == 0x2 */);
1869 
1870 	dr7 = 0x408;
1871 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1872 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1873 
1874 	vmx_set_test_stage(0);
1875 	vmcall();
1876 	report(vmx_get_test_stage() == 1, "Save debug controls");
1877 
1878 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1879 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1880 		printf("\tDebug controls are always loaded/saved\n");
1881 		return;
1882 	}
1883 	vmx_set_test_stage(2);
1884 	vmcall();
1885 
1886 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1887 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1888 	/* Commented out: KVM does not support DEBUGCTL so far */
1889 	(void)debugctl;
1890 	report(dr7 == 0x402,
1891 	       "Guest=host debug controls" /* && debugctl == 0x1 */);
1892 
1893 	dr7 = 0x408;
1894 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1895 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1896 
1897 	vmx_set_test_stage(3);
1898 	vmcall();
1899 	report(vmx_get_test_stage() == 4, "Don't save debug controls");
1900 }
1901 
1902 static int dbgctls_exit_handler(union exit_reason exit_reason)
1903 {
1904 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1905 	u64 guest_rip = vmcs_read(GUEST_RIP);
1906 	u64 dr7, debugctl;
1907 
1908 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1909 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1910 
1911 	switch (exit_reason.basic) {
1912 	case VMX_VMCALL:
1913 		switch (vmx_get_test_stage()) {
1914 		case 0:
1915 			if (dr7 == 0x400 && debugctl == 0 &&
1916 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1917 			    Commented out: KVM does not support DEBUGCTL so far
1918 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1919 				vmx_inc_test_stage();
1920 			break;
1921 		case 2:
1922 			dr7 = 0x402;
1923 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1924 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1925 			vmcs_write(GUEST_DR7, 0x404);
1926 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1927 
1928 			vmcs_write(ENT_CONTROLS,
1929 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1930 			vmcs_write(EXI_CONTROLS,
1931 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1932 			break;
1933 		case 3:
1934 			if (dr7 == 0x400 && debugctl == 0 &&
1935 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1936 			    Commented out: KVM does not support DEBUGCTL so far
1937 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1938 				vmx_inc_test_stage();
1939 			break;
1940 		}
1941 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1942 		return VMX_TEST_RESUME;
1943 	default:
1944 		report_fail("Unknown exit reason, %d", exit_reason.full);
1945 		print_vmexit_info(exit_reason);
1946 	}
1947 	return VMX_TEST_VMEXIT;
1948 }
1949 
1950 struct vmx_msr_entry {
1951 	u32 index;
1952 	u32 reserved;
1953 	u64 value;
1954 } __attribute__((packed));
1955 
1956 #define MSR_MAGIC 0x31415926
1957 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1958 
1959 static int msr_switch_init(struct vmcs *vmcs)
1960 {
1961 	msr_bmp_init();
1962 	exit_msr_store = alloc_page();
1963 	exit_msr_load = alloc_page();
1964 	entry_msr_load = alloc_page();
1965 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1966 	entry_msr_load[0].value = MSR_MAGIC;
1967 
1968 	vmx_set_test_stage(1);
1969 	vmcs_write(ENT_MSR_LD_CNT, 1);
1970 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1971 	vmcs_write(EXI_MSR_ST_CNT, 1);
1972 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1973 	vmcs_write(EXI_MSR_LD_CNT, 1);
1974 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1975 	return VMX_TEST_START;
1976 }
1977 
1978 static void msr_switch_main(void)
1979 {
1980 	if (vmx_get_test_stage() == 1) {
1981 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC,
1982 		       "VM entry MSR load");
1983 		vmx_set_test_stage(2);
1984 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
1985 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
1986 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
1987 		exit_msr_load[0].value = MSR_MAGIC + 2;
1988 	}
1989 	vmcall();
1990 }
1991 
1992 static int msr_switch_exit_handler(union exit_reason exit_reason)
1993 {
1994 	if (exit_reason.basic == VMX_VMCALL && vmx_get_test_stage() == 2) {
1995 		report(exit_msr_store[0].value == MSR_MAGIC + 1,
1996 		       "VM exit MSR store");
1997 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2,
1998 		       "VM exit MSR load");
1999 		vmx_set_test_stage(3);
2000 		entry_msr_load[0].index = MSR_FS_BASE;
2001 		return VMX_TEST_RESUME;
2002 	}
2003 	printf("ERROR %s: unexpected stage=%u or reason=0x%x\n",
2004 		__func__, vmx_get_test_stage(), exit_reason.full);
2005 	return VMX_TEST_EXIT;
2006 }
2007 
2008 static int msr_switch_entry_failure(struct vmentry_result *result)
2009 {
2010 	if (result->vm_fail) {
2011 		printf("ERROR %s: VM-Fail on %s\n", __func__, result->instr);
2012 		return VMX_TEST_EXIT;
2013 	}
2014 
2015 	if (result->exit_reason.failed_vmentry &&
2016 	    result->exit_reason.basic == VMX_FAIL_MSR &&
2017 	    vmx_get_test_stage() == 3) {
2018 		report(vmcs_read(EXI_QUALIFICATION) == 1,
2019 		       "VM entry MSR load: try to load FS_BASE");
2020 		return VMX_TEST_VMEXIT;
2021 	}
2022 	printf("ERROR %s: unexpected stage=%u or reason=%x\n",
2023 		__func__, vmx_get_test_stage(), result->exit_reason.full);
2024 	return VMX_TEST_EXIT;
2025 }
2026 
2027 static int vmmcall_init(struct vmcs *vmcs)
2028 {
2029 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
2030 	return VMX_TEST_START;
2031 }
2032 
2033 static void vmmcall_main(void)
2034 {
2035 	asm volatile(
2036 		"mov $0xABCD, %%rax\n\t"
2037 		"vmmcall\n\t"
2038 		::: "rax");
2039 
2040 	report_fail("VMMCALL");
2041 }
2042 
2043 static int vmmcall_exit_handler(union exit_reason exit_reason)
2044 {
2045 	switch (exit_reason.basic) {
2046 	case VMX_VMCALL:
2047 		printf("here\n");
2048 		report_fail("VMMCALL triggers #UD");
2049 		break;
2050 	case VMX_EXC_NMI:
2051 		report((vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR,
2052 		       "VMMCALL triggers #UD");
2053 		break;
2054 	default:
2055 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2056 		print_vmexit_info(exit_reason);
2057 	}
2058 
2059 	return VMX_TEST_VMEXIT;
2060 }
2061 
2062 static int disable_rdtscp_init(struct vmcs *vmcs)
2063 {
2064 	u32 ctrl_cpu1;
2065 
2066 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
2067 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
2068 		ctrl_cpu1 &= ~CPU_RDTSCP;
2069 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
2070 	}
2071 
2072 	return VMX_TEST_START;
2073 }
2074 
2075 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
2076 {
2077 	switch (vmx_get_test_stage()) {
2078 	case 0:
2079 		report_pass("RDTSCP triggers #UD");
2080 		vmx_inc_test_stage();
2081 		regs->rip += 3;
2082 		break;
2083 	case 2:
2084 		report_pass("RDPID triggers #UD");
2085 		vmx_inc_test_stage();
2086 		regs->rip += 4;
2087 		break;
2088 	}
2089 	return;
2090 
2091 }
2092 
2093 static void disable_rdtscp_main(void)
2094 {
2095 	/* Test that #UD is properly injected in L2.  */
2096 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2097 
2098 	vmx_set_test_stage(0);
2099 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2100 	vmcall();
2101 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2102 
2103 	handle_exception(UD_VECTOR, 0);
2104 	vmcall();
2105 }
2106 
2107 static int disable_rdtscp_exit_handler(union exit_reason exit_reason)
2108 {
2109 	switch (exit_reason.basic) {
2110 	case VMX_VMCALL:
2111 		switch (vmx_get_test_stage()) {
2112 		case 0:
2113 			report_fail("RDTSCP triggers #UD");
2114 			vmx_inc_test_stage();
2115 			/* fallthrough */
2116 		case 1:
2117 			vmx_inc_test_stage();
2118 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2119 			return VMX_TEST_RESUME;
2120 		case 2:
2121 			report_fail("RDPID triggers #UD");
2122 			break;
2123 		}
2124 		break;
2125 
2126 	default:
2127 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2128 		print_vmexit_info(exit_reason);
2129 	}
2130 	return VMX_TEST_VMEXIT;
2131 }
2132 
2133 static int int3_init(struct vmcs *vmcs)
2134 {
2135 	vmcs_write(EXC_BITMAP, ~0u);
2136 	return VMX_TEST_START;
2137 }
2138 
2139 static void int3_guest_main(void)
2140 {
2141 	asm volatile ("int3");
2142 }
2143 
2144 static int int3_exit_handler(union exit_reason exit_reason)
2145 {
2146 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2147 
2148 	report(exit_reason.basic == VMX_EXC_NMI &&
2149 	       (intr_info & INTR_INFO_VALID_MASK) &&
2150 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2151 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2152 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2153 	       "L1 intercepts #BP");
2154 
2155 	return VMX_TEST_VMEXIT;
2156 }
2157 
2158 static int into_init(struct vmcs *vmcs)
2159 {
2160 	vmcs_write(EXC_BITMAP, ~0u);
2161 	return VMX_TEST_START;
2162 }
2163 
2164 static void into_guest_main(void)
2165 {
2166 	struct far_pointer32 fp = {
2167 		.offset = (uintptr_t)&&into,
2168 		.selector = KERNEL_CS32,
2169 	};
2170 	uintptr_t rsp;
2171 
2172 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
2173 
2174 	if (fp.offset != (uintptr_t)&&into) {
2175 		printf("Code address too high.\n");
2176 		return;
2177 	}
2178 	if ((u32)rsp != rsp) {
2179 		printf("Stack address too high.\n");
2180 		return;
2181 	}
2182 
2183 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2184 	return;
2185 into:
2186 	asm volatile (".code32;"
2187 		      "movl $0x7fffffff, %eax;"
2188 		      "addl %eax, %eax;"
2189 		      "into;"
2190 		      "lret;"
2191 		      ".code64");
2192 	__builtin_unreachable();
2193 }
2194 
2195 static int into_exit_handler(union exit_reason exit_reason)
2196 {
2197 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2198 
2199 	report(exit_reason.basic == VMX_EXC_NMI &&
2200 	       (intr_info & INTR_INFO_VALID_MASK) &&
2201 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2202 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2203 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2204 	       "L1 intercepts #OF");
2205 
2206 	return VMX_TEST_VMEXIT;
2207 }
2208 
2209 static void exit_monitor_from_l2_main(void)
2210 {
2211 	printf("Calling exit(0) from l2...\n");
2212 	exit(0);
2213 }
2214 
2215 static int exit_monitor_from_l2_handler(union exit_reason exit_reason)
2216 {
2217 	report_fail("The guest should have killed the VMM");
2218 	return VMX_TEST_EXIT;
2219 }
2220 
2221 static void assert_exit_reason(u64 expected)
2222 {
2223 	u64 actual = vmcs_read(EXI_REASON);
2224 
2225 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2226 			   exit_reason_description(expected),
2227 			   exit_reason_description(actual));
2228 }
2229 
2230 static void skip_exit_insn(void)
2231 {
2232 	u64 guest_rip = vmcs_read(GUEST_RIP);
2233 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2234 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2235 }
2236 
2237 static void skip_exit_vmcall(void)
2238 {
2239 	assert_exit_reason(VMX_VMCALL);
2240 	skip_exit_insn();
2241 }
2242 
2243 static void v2_null_test_guest(void)
2244 {
2245 }
2246 
2247 static void v2_null_test(void)
2248 {
2249 	test_set_guest(v2_null_test_guest);
2250 	enter_guest();
2251 	report_pass(__func__);
2252 }
2253 
2254 static void v2_multiple_entries_test_guest(void)
2255 {
2256 	vmx_set_test_stage(1);
2257 	vmcall();
2258 	vmx_set_test_stage(2);
2259 }
2260 
2261 static void v2_multiple_entries_test(void)
2262 {
2263 	test_set_guest(v2_multiple_entries_test_guest);
2264 	enter_guest();
2265 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2266 	skip_exit_vmcall();
2267 	enter_guest();
2268 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2269 	report_pass(__func__);
2270 }
2271 
2272 static int fixture_test_data = 1;
2273 
2274 static void fixture_test_teardown(void *data)
2275 {
2276 	*((int *) data) = 1;
2277 }
2278 
2279 static void fixture_test_guest(void)
2280 {
2281 	fixture_test_data++;
2282 }
2283 
2284 
2285 static void fixture_test_setup(void)
2286 {
2287 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2288 			   "fixture_test_teardown didn't run?!");
2289 	fixture_test_data = 2;
2290 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2291 	test_set_guest(fixture_test_guest);
2292 }
2293 
2294 static void fixture_test_case1(void)
2295 {
2296 	fixture_test_setup();
2297 	TEST_ASSERT_EQ(2, fixture_test_data);
2298 	enter_guest();
2299 	TEST_ASSERT_EQ(3, fixture_test_data);
2300 	report_pass(__func__);
2301 }
2302 
2303 static void fixture_test_case2(void)
2304 {
2305 	fixture_test_setup();
2306 	TEST_ASSERT_EQ(2, fixture_test_data);
2307 	enter_guest();
2308 	TEST_ASSERT_EQ(3, fixture_test_data);
2309 	report_pass(__func__);
2310 }
2311 
2312 enum ept_access_op {
2313 	OP_READ,
2314 	OP_WRITE,
2315 	OP_EXEC,
2316 	OP_FLUSH_TLB,
2317 	OP_EXIT,
2318 };
2319 
2320 static struct ept_access_test_data {
2321 	unsigned long gpa;
2322 	unsigned long *gva;
2323 	unsigned long hpa;
2324 	unsigned long *hva;
2325 	enum ept_access_op op;
2326 } ept_access_test_data;
2327 
2328 extern unsigned char ret42_start;
2329 extern unsigned char ret42_end;
2330 
2331 /* Returns 42. */
2332 asm(
2333 	".align 64\n"
2334 	"ret42_start:\n"
2335 	"mov $42, %eax\n"
2336 	"ret\n"
2337 	"ret42_end:\n"
2338 );
2339 
2340 static void
2341 diagnose_ept_violation_qual(u64 expected, u64 actual)
2342 {
2343 
2344 #define DIAGNOSE(flag)							\
2345 do {									\
2346 	if ((expected & flag) != (actual & flag))			\
2347 		printf(#flag " %sexpected\n",				\
2348 		       (expected & flag) ? "" : "un");			\
2349 } while (0)
2350 
2351 	DIAGNOSE(EPT_VLT_RD);
2352 	DIAGNOSE(EPT_VLT_WR);
2353 	DIAGNOSE(EPT_VLT_FETCH);
2354 	DIAGNOSE(EPT_VLT_PERM_RD);
2355 	DIAGNOSE(EPT_VLT_PERM_WR);
2356 	DIAGNOSE(EPT_VLT_PERM_EX);
2357 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2358 	DIAGNOSE(EPT_VLT_PADDR);
2359 
2360 #undef DIAGNOSE
2361 }
2362 
2363 static void do_ept_access_op(enum ept_access_op op)
2364 {
2365 	ept_access_test_data.op = op;
2366 	enter_guest();
2367 }
2368 
2369 /*
2370  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2371  * needed by tests that modify guest PTEs.
2372  */
2373 static void ept_access_test_guest_flush_tlb(void)
2374 {
2375 	do_ept_access_op(OP_FLUSH_TLB);
2376 	skip_exit_vmcall();
2377 }
2378 
2379 /*
2380  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2381  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2382  * a huge page.
2383  */
2384 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2385 				 unsigned long clear, unsigned long set)
2386 {
2387 	struct ept_access_test_data *data = &ept_access_test_data;
2388 	unsigned long orig_pte;
2389 	unsigned long pte;
2390 
2391 	/* Screw with the mapping at the requested level. */
2392 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2393 	pte = orig_pte;
2394 	if (mkhuge)
2395 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2396 	else
2397 		pte = orig_pte;
2398 	pte = (pte & ~clear) | set;
2399 	set_ept_pte(pml4, gpa, level, pte);
2400 	invept(INVEPT_SINGLE, eptp);
2401 
2402 	return orig_pte;
2403 }
2404 
2405 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2406 {
2407 	set_ept_pte(pml4, gpa, level, orig_pte);
2408 	invept(INVEPT_SINGLE, eptp);
2409 }
2410 
2411 static void do_ept_violation(bool leaf, enum ept_access_op op,
2412 			     u64 expected_qual, u64 expected_paddr)
2413 {
2414 	u64 qual;
2415 
2416 	/* Try the access and observe the violation. */
2417 	do_ept_access_op(op);
2418 
2419 	assert_exit_reason(VMX_EPT_VIOLATION);
2420 
2421 	qual = vmcs_read(EXI_QUALIFICATION);
2422 
2423 	/* Mask undefined bits (which may later be defined in certain cases). */
2424 	qual &= ~(EPT_VLT_GUEST_USER | EPT_VLT_GUEST_RW | EPT_VLT_GUEST_EX |
2425 		 EPT_VLT_PERM_USER_EX);
2426 
2427 	diagnose_ept_violation_qual(expected_qual, qual);
2428 	TEST_EXPECT_EQ(expected_qual, qual);
2429 
2430 	#if 0
2431 	/* Disable for now otherwise every test will fail */
2432 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2433 		       (unsigned long) (
2434 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2435 	#endif
2436 	/*
2437 	 * TODO: tests that probe expected_paddr in pages other than the one at
2438 	 * the beginning of the 1g region.
2439 	 */
2440 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2441 }
2442 
2443 static void
2444 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2445 			      unsigned long set, enum ept_access_op op,
2446 			      u64 expected_qual)
2447 {
2448 	struct ept_access_test_data *data = &ept_access_test_data;
2449 	unsigned long orig_pte;
2450 
2451 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2452 
2453 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2454 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2455 					 data->gpa);
2456 
2457 	/* Fix the violation and resume the op loop. */
2458 	ept_untwiddle(data->gpa, level, orig_pte);
2459 	enter_guest();
2460 	skip_exit_vmcall();
2461 }
2462 
2463 static void
2464 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2465 		       enum ept_access_op op, u64 expected_qual)
2466 {
2467 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2468 				      expected_qual);
2469 	if (ept_huge_pages_supported(level))
2470 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2471 					      expected_qual);
2472 }
2473 
2474 static void ept_violation(unsigned long clear, unsigned long set,
2475 			  enum ept_access_op op, u64 expected_qual)
2476 {
2477 	ept_violation_at_level(1, clear, set, op, expected_qual);
2478 	ept_violation_at_level(2, clear, set, op, expected_qual);
2479 	ept_violation_at_level(3, clear, set, op, expected_qual);
2480 	ept_violation_at_level(4, clear, set, op, expected_qual);
2481 }
2482 
2483 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2484 				       u64 expected_qual)
2485 {
2486 	ept_violation(EPT_PRESENT, access, op,
2487 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2488 }
2489 
2490 /*
2491  * For translations that don't involve a GVA, that is physical address (paddr)
2492  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2493  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2494  * translations don't involve GVAs, such as when the hardware does the guest
2495  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2496  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2497  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2498  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2499  *
2500  * Note that paddr violations can also be triggered by loading PAE page tables
2501  * with wonky addresses. We don't test that yet.
2502  *
2503  * This function modifies the EPT entry that maps the GPA that the guest page
2504  * table entry mapping ept_access_test_data.gva resides on.
2505  *
2506  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2507  *
2508  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2509  *
2510  *	@op		Guest operation to perform with
2511  *			ept_access_test_data.gva.
2512  *
2513  *	@expect_violation
2514  *			Is a violation expected during the paddr access?
2515  *
2516  *	@expected_qual	Expected qualification for the EPT violation.
2517  *			EPT_VLT_PADDR should be clear.
2518  */
2519 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2520 			     enum ept_access_op op, bool expect_violation,
2521 			     u64 expected_qual)
2522 {
2523 	struct ept_access_test_data *data = &ept_access_test_data;
2524 	unsigned long *ptep;
2525 	unsigned long gpa;
2526 	unsigned long orig_epte;
2527 	unsigned long epte;
2528 	int i;
2529 
2530 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2531 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2532 	TEST_ASSERT(ptep);
2533 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2534 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2535 	ept_access_test_guest_flush_tlb();
2536 
2537 	/*
2538 	 * Now modify the access bits on the EPT entry for the GPA that the
2539 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2540 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2541 	 * constructed our test such that those other 511 PTEs aren't used by
2542 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2543 	 * PTE we're modifying is at the beginning of a 4K page and the
2544 	 * following 511 entires are also under our control (and not touched by
2545 	 * the guest).
2546 	 */
2547 	gpa = virt_to_phys(ptep);
2548 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2549 	/*
2550 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2551 	 * otherwise our level=1 twiddling below will fail. We use the
2552 	 * identity map (gpa = gpa) since page tables are shared with the host.
2553 	 */
2554 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2555 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2556 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2557 
2558 	if (expect_violation) {
2559 		do_ept_violation(/*leaf=*/true, op,
2560 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2561 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2562 		do_ept_access_op(op);
2563 	} else {
2564 		do_ept_access_op(op);
2565 		if (ept_ad_enabled()) {
2566 			for (i = EPT_PAGE_LEVEL; i > 0; i--) {
2567 				TEST_ASSERT(get_ept_pte(pml4, gpa, i, &epte));
2568 				TEST_ASSERT(epte & EPT_ACCESS_FLAG);
2569 				if (i == 1)
2570 					TEST_ASSERT(epte & EPT_DIRTY_FLAG);
2571 				else
2572 					TEST_ASSERT_EQ(epte & EPT_DIRTY_FLAG, 0);
2573 			}
2574 		}
2575 
2576 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2577 	}
2578 
2579 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2580 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2581 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2582 
2583 	skip_exit_vmcall();
2584 }
2585 
2586 static void ept_access_allowed_paddr(unsigned long ept_access,
2587 				     unsigned long pte_ad,
2588 				     enum ept_access_op op)
2589 {
2590 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2591 			 /*expected_qual=*/-1);
2592 }
2593 
2594 static void ept_access_violation_paddr(unsigned long ept_access,
2595 				       unsigned long pte_ad,
2596 				       enum ept_access_op op,
2597 				       u64 expected_qual)
2598 {
2599 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2600 			 expected_qual);
2601 }
2602 
2603 
2604 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2605 					unsigned long clear,
2606 					unsigned long set,
2607 					enum ept_access_op op)
2608 {
2609 	struct ept_access_test_data *data = &ept_access_test_data;
2610 	unsigned long orig_pte;
2611 
2612 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2613 
2614 	/* No violation. Should proceed to vmcall. */
2615 	do_ept_access_op(op);
2616 	skip_exit_vmcall();
2617 
2618 	ept_untwiddle(data->gpa, level, orig_pte);
2619 }
2620 
2621 static void ept_allowed_at_level(int level, unsigned long clear,
2622 				 unsigned long set, enum ept_access_op op)
2623 {
2624 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2625 	if (ept_huge_pages_supported(level))
2626 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2627 }
2628 
2629 static void ept_allowed(unsigned long clear, unsigned long set,
2630 			enum ept_access_op op)
2631 {
2632 	ept_allowed_at_level(1, clear, set, op);
2633 	ept_allowed_at_level(2, clear, set, op);
2634 	ept_allowed_at_level(3, clear, set, op);
2635 	ept_allowed_at_level(4, clear, set, op);
2636 }
2637 
2638 static void ept_ignored_bit(int bit)
2639 {
2640 	/* Set the bit. */
2641 	ept_allowed(0, 1ul << bit, OP_READ);
2642 	ept_allowed(0, 1ul << bit, OP_WRITE);
2643 	ept_allowed(0, 1ul << bit, OP_EXEC);
2644 
2645 	/* Clear the bit. */
2646 	ept_allowed(1ul << bit, 0, OP_READ);
2647 	ept_allowed(1ul << bit, 0, OP_WRITE);
2648 	ept_allowed(1ul << bit, 0, OP_EXEC);
2649 }
2650 
2651 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2652 {
2653 	ept_allowed(EPT_PRESENT, access, op);
2654 }
2655 
2656 
2657 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2658 					     unsigned long clear,
2659 					     unsigned long set,
2660 					     enum ept_access_op op)
2661 {
2662 	struct ept_access_test_data *data = &ept_access_test_data;
2663 	unsigned long orig_pte;
2664 
2665 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2666 
2667 	do_ept_access_op(op);
2668 	assert_exit_reason(VMX_EPT_MISCONFIG);
2669 
2670 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2671 	#if 0
2672 	/* broken: */
2673 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2674 	#endif
2675 	#if 0
2676 	/*
2677 	 * broken:
2678 	 * According to description of exit qual for EPT violation,
2679 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2680 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2681 	 * to be set for msiconfig.
2682 	 */
2683 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2684 		       (unsigned long) (
2685 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2686 	#endif
2687 
2688 	/* Fix the violation and resume the op loop. */
2689 	ept_untwiddle(data->gpa, level, orig_pte);
2690 	enter_guest();
2691 	skip_exit_vmcall();
2692 }
2693 
2694 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2695 					  unsigned long clear,
2696 					  unsigned long set)
2697 {
2698 	/* The op shouldn't matter (read, write, exec), so try them all! */
2699 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2700 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2701 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2702 }
2703 
2704 static void ept_misconfig_at_level(int level, unsigned long clear,
2705 				   unsigned long set)
2706 {
2707 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2708 	if (ept_huge_pages_supported(level))
2709 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2710 }
2711 
2712 static void ept_misconfig(unsigned long clear, unsigned long set)
2713 {
2714 	ept_misconfig_at_level(1, clear, set);
2715 	ept_misconfig_at_level(2, clear, set);
2716 	ept_misconfig_at_level(3, clear, set);
2717 	ept_misconfig_at_level(4, clear, set);
2718 }
2719 
2720 static void ept_access_misconfig(unsigned long access)
2721 {
2722 	ept_misconfig(EPT_PRESENT, access);
2723 }
2724 
2725 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2726 {
2727 	/* Setting the bit causes a misconfig. */
2728 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2729 
2730 	/* Making the entry non-present turns reserved bits into ignored. */
2731 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2732 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2733 }
2734 
2735 static void ept_reserved_bit_at_level_huge(int level, int bit)
2736 {
2737 	/* Setting the bit causes a misconfig. */
2738 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2739 
2740 	/* Making the entry non-present turns reserved bits into ignored. */
2741 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2742 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2743 }
2744 
2745 static void ept_reserved_bit_at_level(int level, int bit)
2746 {
2747 	/* Setting the bit causes a misconfig. */
2748 	ept_misconfig_at_level(level, 0, 1ul << bit);
2749 
2750 	/* Making the entry non-present turns reserved bits into ignored. */
2751 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2752 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2753 }
2754 
2755 static void ept_reserved_bit(int bit)
2756 {
2757 	ept_reserved_bit_at_level(1, bit);
2758 	ept_reserved_bit_at_level(2, bit);
2759 	ept_reserved_bit_at_level(3, bit);
2760 	ept_reserved_bit_at_level(4, bit);
2761 }
2762 
2763 #define PAGE_2M_ORDER 9
2764 #define PAGE_1G_ORDER 18
2765 
2766 static void *get_1g_page(void)
2767 {
2768 	static void *alloc;
2769 
2770 	if (!alloc)
2771 		alloc = alloc_pages(PAGE_1G_ORDER);
2772 	return alloc;
2773 }
2774 
2775 static void ept_access_test_teardown(void *unused)
2776 {
2777 	/* Exit the guest cleanly. */
2778 	do_ept_access_op(OP_EXIT);
2779 }
2780 
2781 static void ept_access_test_guest(void)
2782 {
2783 	struct ept_access_test_data *data = &ept_access_test_data;
2784 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2785 
2786 	while (true) {
2787 		switch (data->op) {
2788 		case OP_READ:
2789 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2790 			break;
2791 		case OP_WRITE:
2792 			*data->gva = MAGIC_VAL_2;
2793 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2794 			*data->gva = MAGIC_VAL_1;
2795 			break;
2796 		case OP_EXEC:
2797 			TEST_ASSERT_EQ(42, code());
2798 			break;
2799 		case OP_FLUSH_TLB:
2800 			write_cr3(read_cr3());
2801 			break;
2802 		case OP_EXIT:
2803 			return;
2804 		default:
2805 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2806 		}
2807 		vmcall();
2808 	}
2809 }
2810 
2811 static void ept_access_test_setup(void)
2812 {
2813 	struct ept_access_test_data *data = &ept_access_test_data;
2814 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2815 	unsigned long size = npages * PAGE_SIZE;
2816 	unsigned long *page_table = current_page_table();
2817 	unsigned long pte;
2818 
2819 	if (setup_ept(false))
2820 		test_skip("EPT not supported");
2821 
2822 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2823 	if (cpuid_maxphyaddr() < 40)
2824 		test_skip("Test needs MAXPHYADDR >= 40");
2825 
2826 	test_set_guest(ept_access_test_guest);
2827 	test_add_teardown(ept_access_test_teardown, NULL);
2828 
2829 	data->hva = get_1g_page();
2830 	TEST_ASSERT(data->hva);
2831 	data->hpa = virt_to_phys(data->hva);
2832 
2833 	data->gpa = 1ul << 39;
2834 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2835 				   size);
2836 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2837 	install_pages(page_table, data->gpa, size, data->gva);
2838 
2839 	/*
2840 	 * Make sure nothing's mapped here so the tests that screw with the
2841 	 * pml4 entry don't inadvertently break something.
2842 	 */
2843 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2844 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2845 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2846 
2847 	data->hva[0] = MAGIC_VAL_1;
2848 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2849 }
2850 
2851 static void ept_access_test_not_present(void)
2852 {
2853 	ept_access_test_setup();
2854 	/* --- */
2855 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2856 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2857 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2858 }
2859 
2860 static void ept_access_test_read_only(void)
2861 {
2862 	ept_access_test_setup();
2863 
2864 	/* r-- */
2865 	ept_access_allowed(EPT_RA, OP_READ);
2866 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2867 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2868 }
2869 
2870 static void ept_access_test_write_only(void)
2871 {
2872 	ept_access_test_setup();
2873 	/* -w- */
2874 	ept_access_misconfig(EPT_WA);
2875 }
2876 
2877 static void ept_access_test_read_write(void)
2878 {
2879 	ept_access_test_setup();
2880 	/* rw- */
2881 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2882 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2883 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2884 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2885 }
2886 
2887 
2888 static void ept_access_test_execute_only(void)
2889 {
2890 	ept_access_test_setup();
2891 	/* --x */
2892 	if (ept_execute_only_supported()) {
2893 		ept_access_violation(EPT_EA, OP_READ,
2894 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2895 		ept_access_violation(EPT_EA, OP_WRITE,
2896 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2897 		ept_access_allowed(EPT_EA, OP_EXEC);
2898 	} else {
2899 		ept_access_misconfig(EPT_EA);
2900 	}
2901 }
2902 
2903 static void ept_access_test_read_execute(void)
2904 {
2905 	ept_access_test_setup();
2906 	/* r-x */
2907 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2908 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2909 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2910 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2911 }
2912 
2913 static void ept_access_test_write_execute(void)
2914 {
2915 	ept_access_test_setup();
2916 	/* -wx */
2917 	ept_access_misconfig(EPT_WA | EPT_EA);
2918 }
2919 
2920 static void ept_access_test_read_write_execute(void)
2921 {
2922 	ept_access_test_setup();
2923 	/* rwx */
2924 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2925 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2926 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2927 }
2928 
2929 static void ept_access_test_reserved_bits(void)
2930 {
2931 	int i;
2932 	int maxphyaddr;
2933 
2934 	ept_access_test_setup();
2935 
2936 	/* Reserved bits above maxphyaddr. */
2937 	maxphyaddr = cpuid_maxphyaddr();
2938 	for (i = maxphyaddr; i <= 51; i++) {
2939 		report_prefix_pushf("reserved_bit=%d", i);
2940 		ept_reserved_bit(i);
2941 		report_prefix_pop();
2942 	}
2943 
2944 	/* Level-specific reserved bits. */
2945 	ept_reserved_bit_at_level_nohuge(2, 3);
2946 	ept_reserved_bit_at_level_nohuge(2, 4);
2947 	ept_reserved_bit_at_level_nohuge(2, 5);
2948 	ept_reserved_bit_at_level_nohuge(2, 6);
2949 	/* 2M alignment. */
2950 	for (i = 12; i < 20; i++) {
2951 		report_prefix_pushf("reserved_bit=%d", i);
2952 		ept_reserved_bit_at_level_huge(2, i);
2953 		report_prefix_pop();
2954 	}
2955 	ept_reserved_bit_at_level_nohuge(3, 3);
2956 	ept_reserved_bit_at_level_nohuge(3, 4);
2957 	ept_reserved_bit_at_level_nohuge(3, 5);
2958 	ept_reserved_bit_at_level_nohuge(3, 6);
2959 	/* 1G alignment. */
2960 	for (i = 12; i < 29; i++) {
2961 		report_prefix_pushf("reserved_bit=%d", i);
2962 		ept_reserved_bit_at_level_huge(3, i);
2963 		report_prefix_pop();
2964 	}
2965 	ept_reserved_bit_at_level(4, 3);
2966 	ept_reserved_bit_at_level(4, 4);
2967 	ept_reserved_bit_at_level(4, 5);
2968 	ept_reserved_bit_at_level(4, 6);
2969 	ept_reserved_bit_at_level(4, 7);
2970 }
2971 
2972 static void ept_access_test_ignored_bits(void)
2973 {
2974 	ept_access_test_setup();
2975 	/*
2976 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2977 	 * far as translation is concerned even if AD bits are enabled in the
2978 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2979 	 * control is 0.
2980 	 */
2981 	ept_ignored_bit(8);
2982 	ept_ignored_bit(9);
2983 	ept_ignored_bit(10);
2984 	ept_ignored_bit(11);
2985 	ept_ignored_bit(52);
2986 	ept_ignored_bit(53);
2987 	ept_ignored_bit(54);
2988 	ept_ignored_bit(55);
2989 	ept_ignored_bit(56);
2990 	ept_ignored_bit(57);
2991 	ept_ignored_bit(58);
2992 	ept_ignored_bit(59);
2993 	ept_ignored_bit(60);
2994 	ept_ignored_bit(61);
2995 	ept_ignored_bit(62);
2996 	ept_ignored_bit(63);
2997 }
2998 
2999 static void ept_access_test_paddr_not_present_ad_disabled(void)
3000 {
3001 	ept_access_test_setup();
3002 	ept_disable_ad_bits();
3003 
3004 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
3005 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
3006 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
3007 }
3008 
3009 static void ept_access_test_paddr_not_present_ad_enabled(void)
3010 {
3011 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
3012 
3013 	ept_access_test_setup();
3014 	ept_enable_ad_bits_or_skip_test();
3015 
3016 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
3017 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
3018 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
3019 }
3020 
3021 static void ept_access_test_paddr_read_only_ad_disabled(void)
3022 {
3023 	/*
3024 	 * When EPT AD bits are disabled, all accesses to guest paging
3025 	 * structures are reported separately as a read and (after
3026 	 * translation of the GPA to host physical address) a read+write
3027 	 * if the A/D bits have to be set.
3028 	 */
3029 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3030 
3031 	ept_access_test_setup();
3032 	ept_disable_ad_bits();
3033 
3034 	/* Can't update A bit, so all accesses fail. */
3035 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3036 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3037 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3038 	/* AD bits disabled, so only writes try to update the D bit. */
3039 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
3040 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3041 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
3042 	/* Both A and D already set, so read-only is OK. */
3043 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
3044 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
3045 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
3046 }
3047 
3048 static void ept_access_test_paddr_read_only_ad_enabled(void)
3049 {
3050 	/*
3051 	 * When EPT AD bits are enabled, all accesses to guest paging
3052 	 * structures are considered writes as far as EPT translation
3053 	 * is concerned.
3054 	 */
3055 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3056 
3057 	ept_access_test_setup();
3058 	ept_enable_ad_bits_or_skip_test();
3059 
3060 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3061 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3062 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3063 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
3064 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3065 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
3066 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
3067 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
3068 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
3069 }
3070 
3071 static void ept_access_test_paddr_read_write(void)
3072 {
3073 	ept_access_test_setup();
3074 	/* Read-write access to paging structure. */
3075 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
3076 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
3077 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
3078 }
3079 
3080 static void ept_access_test_paddr_read_write_execute(void)
3081 {
3082 	ept_access_test_setup();
3083 	/* RWX access to paging structure. */
3084 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
3085 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
3086 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
3087 }
3088 
3089 static void ept_access_test_paddr_read_execute_ad_disabled(void)
3090 {
3091   	/*
3092 	 * When EPT AD bits are disabled, all accesses to guest paging
3093 	 * structures are reported separately as a read and (after
3094 	 * translation of the GPA to host physical address) a read+write
3095 	 * if the A/D bits have to be set.
3096 	 */
3097 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3098 
3099 	ept_access_test_setup();
3100 	ept_disable_ad_bits();
3101 
3102 	/* Can't update A bit, so all accesses fail. */
3103 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3104 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3105 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3106 	/* AD bits disabled, so only writes try to update the D bit. */
3107 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
3108 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3109 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3110 	/* Both A and D already set, so read-only is OK. */
3111 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3112 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3113 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3114 }
3115 
3116 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3117 {
3118 	/*
3119 	 * When EPT AD bits are enabled, all accesses to guest paging
3120 	 * structures are considered writes as far as EPT translation
3121 	 * is concerned.
3122 	 */
3123 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3124 
3125 	ept_access_test_setup();
3126 	ept_enable_ad_bits_or_skip_test();
3127 
3128 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3129 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3130 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3131 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3132 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3133 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3134 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3135 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3136 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3137 }
3138 
3139 static void ept_access_test_paddr_not_present_page_fault(void)
3140 {
3141 	ept_access_test_setup();
3142 	/*
3143 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3144 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3145 	 * Thus guest page fault before host takes EPT violation for trying to
3146 	 * update A bit.
3147 	 */
3148 }
3149 
3150 static void ept_access_test_force_2m_page(void)
3151 {
3152 	ept_access_test_setup();
3153 
3154 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3155 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3156 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3157 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3158 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3159 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3160 }
3161 
3162 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3163 {
3164 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3165 
3166 	TEST_ASSERT(msr & VPID_CAP_INVVPID);
3167 
3168 	if (type < INVVPID_ADDR || type > INVVPID_CONTEXT_LOCAL)
3169 		return false;
3170 
3171 	if (!(msr & (1ull << (type + VPID_CAP_INVVPID_TYPES_SHIFT))))
3172 		return false;
3173 
3174 	if (vpid >> 16)
3175 		return false;
3176 
3177 	if (type != INVVPID_ALL && !vpid)
3178 		return false;
3179 
3180 	if (type == INVVPID_ADDR && !is_canonical(gla))
3181 		return false;
3182 
3183 	return true;
3184 }
3185 
3186 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3187 {
3188 	int rc;
3189 	bool valid = invvpid_valid(type, vpid, gla);
3190 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3191 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3192 	/*
3193 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3194 	 * that we can tell if it is updated by INVVPID.
3195 	 */
3196 	vmcs_read(~0);
3197 	rc = __invvpid(type, vpid, gla);
3198 	report(!rc == valid, "INVVPID type %ld VPID %lx GLA %lx %s", type,
3199 	       vpid, gla,
3200 	       valid ? "passes" : "fails");
3201 	report(vmcs_read(VMX_INST_ERROR) == expected,
3202 	       "After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3203 	       rc ? "failed" : "successful",
3204 	       expected, vmcs_read(VMX_INST_ERROR));
3205 }
3206 
3207 static void ds_invvpid(void *data)
3208 {
3209 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3210 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3211 
3212 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3213 	asm volatile("invvpid %0, %1"
3214 		     :
3215 		     : "m"(*(struct invvpid_operand *)data),
3216 		       "r"(type));
3217 }
3218 
3219 /*
3220  * The SS override is ignored in 64-bit mode, so we use an addressing
3221  * mode with %rsp as the base register to generate an implicit SS
3222  * reference.
3223  */
3224 static void ss_invvpid(void *data)
3225 {
3226 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3227 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3228 
3229 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3230 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3231 		     : "+r"(data)
3232 		     : "r"(type));
3233 }
3234 
3235 static void invvpid_test_gp(void)
3236 {
3237 	bool fault;
3238 
3239 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3240 				   (void *)NONCANONICAL);
3241 	report(fault, "INVVPID with non-canonical DS operand raises #GP");
3242 }
3243 
3244 static void invvpid_test_ss(void)
3245 {
3246 	bool fault;
3247 
3248 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3249 				   (void *)NONCANONICAL);
3250 	report(fault, "INVVPID with non-canonical SS operand raises #SS");
3251 }
3252 
3253 static void invvpid_test_pf(void)
3254 {
3255 	void *vpage = alloc_vpage();
3256 	bool fault;
3257 
3258 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3259 	report(fault, "INVVPID with unmapped operand raises #PF");
3260 }
3261 
3262 static void try_compat_invvpid(void *unused)
3263 {
3264 	struct far_pointer32 fp = {
3265 		.offset = (uintptr_t)&&invvpid,
3266 		.selector = KERNEL_CS32,
3267 	};
3268 	uintptr_t rsp;
3269 
3270 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
3271 
3272 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3273 			"Code address too high.");
3274 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3275 
3276 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3277 	return;
3278 invvpid:
3279 	asm volatile (".code32;"
3280 		      "invvpid (%eax), %eax;"
3281 		      "lret;"
3282 		      ".code64");
3283 	__builtin_unreachable();
3284 }
3285 
3286 static void invvpid_test_compatibility_mode(void)
3287 {
3288 	bool fault;
3289 
3290 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3291 	report(fault, "Compatibility mode INVVPID raises #UD");
3292 }
3293 
3294 static void invvpid_test_not_in_vmx_operation(void)
3295 {
3296 	bool fault;
3297 
3298 	TEST_ASSERT(!vmx_off());
3299 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3300 	report(fault, "INVVPID outside of VMX operation raises #UD");
3301 	TEST_ASSERT(!vmx_on());
3302 }
3303 
3304 /*
3305  * This does not test real-address mode, virtual-8086 mode, protected mode,
3306  * or CPL > 0.
3307  */
3308 static void invvpid_test(void)
3309 {
3310 	u64 msr;
3311 	int i;
3312 	unsigned types = 0;
3313 	unsigned type;
3314 
3315 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
3316 	    !(ctrl_cpu_rev[1].clr & CPU_VPID))
3317 		test_skip("VPID not supported");
3318 
3319 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3320 
3321 	if (!(msr & VPID_CAP_INVVPID))
3322 		test_skip("INVVPID not supported.\n");
3323 
3324 	if (msr & VPID_CAP_INVVPID_ADDR)
3325 		types |= 1u << INVVPID_ADDR;
3326 	if (msr & VPID_CAP_INVVPID_CXTGLB)
3327 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3328 	if (msr & VPID_CAP_INVVPID_ALL)
3329 		types |= 1u << INVVPID_ALL;
3330 	if (msr & VPID_CAP_INVVPID_CXTLOC)
3331 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3332 
3333 	if (!types)
3334 		test_skip("No INVVPID types supported.\n");
3335 
3336 	for (i = -127; i < 128; i++)
3337 		try_invvpid(i, 0xffff, 0);
3338 
3339 	/*
3340 	 * VPID must not be more than 16 bits.
3341 	 */
3342 	for (i = 0; i < 64; i++)
3343 		for (type = 0; type < 4; type++)
3344 			if (types & (1u << type))
3345 				try_invvpid(type, 1ul << i, 0);
3346 
3347 	/*
3348 	 * VPID must not be zero, except for "all contexts."
3349 	 */
3350 	for (type = 0; type < 4; type++)
3351 		if (types & (1u << type))
3352 			try_invvpid(type, 0, 0);
3353 
3354 	/*
3355 	 * The gla operand is only validated for single-address INVVPID.
3356 	 */
3357 	if (types & (1u << INVVPID_ADDR))
3358 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3359 
3360 	invvpid_test_gp();
3361 	invvpid_test_ss();
3362 	invvpid_test_pf();
3363 	invvpid_test_compatibility_mode();
3364 	invvpid_test_not_in_vmx_operation();
3365 }
3366 
3367 /*
3368  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3369  * at least as far as the guest-state checks. Returns false if the
3370  * VMLAUNCH fails early and execution falls through to the next
3371  * instruction.
3372  */
3373 static bool vmlaunch_succeeds(void)
3374 {
3375 	u32 exit_reason;
3376 
3377 	/*
3378 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3379 	 * unsupported VMCS component"). The caller can then check
3380 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3381 	 */
3382 	vmcs_write(~0u, 0);
3383 
3384 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3385 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3386 				   :
3387 				   : "r" ((u64)HOST_RSP)
3388 				   : "cc", "memory"
3389 				   : success);
3390 	return false;
3391 success:
3392 	exit_reason = vmcs_read(EXI_REASON);
3393 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3394 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3395 	return true;
3396 }
3397 
3398 /*
3399  * Try to launch the current VMCS.
3400  */
3401 static void test_vmx_vmlaunch(u32 xerror)
3402 {
3403 	bool success = vmlaunch_succeeds();
3404 	u32 vmx_inst_err;
3405 
3406 	report(success == !xerror, "vmlaunch %s",
3407 	       !xerror ? "succeeds" : "fails");
3408 	if (!success && xerror) {
3409 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3410 		report(vmx_inst_err == xerror,
3411 		       "VMX inst error is %d (actual %d)", xerror,
3412 		       vmx_inst_err);
3413 	}
3414 }
3415 
3416 /*
3417  * Try to launch the current VMCS, and expect one of two possible
3418  * errors (or success) codes.
3419  */
3420 static void test_vmx_vmlaunch2(u32 xerror1, u32 xerror2)
3421 {
3422 	bool success = vmlaunch_succeeds();
3423 	u32 vmx_inst_err;
3424 
3425 	if (!xerror1 == !xerror2)
3426 		report(success == !xerror1, "vmlaunch %s",
3427 		       !xerror1 ? "succeeds" : "fails");
3428 
3429 	if (!success && (xerror1 || xerror2)) {
3430 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3431 		report(vmx_inst_err == xerror1 || vmx_inst_err == xerror2,
3432 		       "VMX inst error is %d or %d (actual %d)", xerror1,
3433 		       xerror2, vmx_inst_err);
3434 	}
3435 }
3436 
3437 static void test_vmx_invalid_controls(void)
3438 {
3439 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3440 }
3441 
3442 static void test_vmx_valid_controls(void)
3443 {
3444 	test_vmx_vmlaunch(0);
3445 }
3446 
3447 /*
3448  * Test a particular value of a VM-execution control bit, if the value
3449  * is required or if the value is zero.
3450  */
3451 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3452 				    enum Encoding encoding, unsigned bit,
3453 				    unsigned val)
3454 {
3455 	u32 mask = 1u << bit;
3456 	bool expected;
3457 	u32 controls;
3458 
3459 	if (msr.set & mask)
3460 		TEST_ASSERT(msr.clr & mask);
3461 
3462 	/*
3463 	 * We can't arbitrarily turn on a control bit, because it may
3464 	 * introduce dependencies on other VMCS fields. So, we only
3465 	 * test turning on bits that have a required setting.
3466 	 */
3467 	if (val && (msr.clr & mask) && !(msr.set & mask))
3468 		return;
3469 
3470 	report_prefix_pushf("%s %s bit %d",
3471 			    val ? "Set" : "Clear", name, bit);
3472 
3473 	controls = vmcs_read(encoding);
3474 	if (val) {
3475 		vmcs_write(encoding, msr.set | mask);
3476 		expected = (msr.clr & mask);
3477 	} else {
3478 		vmcs_write(encoding, msr.set & ~mask);
3479 		expected = !(msr.set & mask);
3480 	}
3481 	if (expected)
3482 		test_vmx_valid_controls();
3483 	else
3484 		test_vmx_invalid_controls();
3485 	vmcs_write(encoding, controls);
3486 	report_prefix_pop();
3487 }
3488 
3489 /*
3490  * Test reserved values of a VM-execution control bit, based on the
3491  * allowed bit settings from the corresponding VMX capability MSR.
3492  */
3493 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3494 			      enum Encoding encoding, unsigned bit)
3495 {
3496 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3497 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3498 }
3499 
3500 /*
3501  * Reserved bits in the pin-based VM-execution controls must be set
3502  * properly. Software may consult the VMX capability MSRs to determine
3503  * the proper settings.
3504  * [Intel SDM]
3505  */
3506 static void test_pin_based_ctls(void)
3507 {
3508 	unsigned bit;
3509 
3510 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3511 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3512 	for (bit = 0; bit < 32; bit++)
3513 		test_rsvd_ctl_bit("pin-based controls",
3514 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3515 }
3516 
3517 /*
3518  * Reserved bits in the primary processor-based VM-execution controls
3519  * must be set properly. Software may consult the VMX capability MSRs
3520  * to determine the proper settings.
3521  * [Intel SDM]
3522  */
3523 static void test_primary_processor_based_ctls(void)
3524 {
3525 	unsigned bit;
3526 
3527 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3528 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3529 	for (bit = 0; bit < 32; bit++)
3530 		test_rsvd_ctl_bit("primary processor-based controls",
3531 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3532 }
3533 
3534 /*
3535  * If the "activate secondary controls" primary processor-based
3536  * VM-execution control is 1, reserved bits in the secondary
3537  * processor-based VM-execution controls must be cleared. Software may
3538  * consult the VMX capability MSRs to determine which bits are
3539  * reserved.
3540  * If the "activate secondary controls" primary processor-based
3541  * VM-execution control is 0 (or if the processor does not support the
3542  * 1-setting of that control), no checks are performed on the
3543  * secondary processor-based VM-execution controls.
3544  * [Intel SDM]
3545  */
3546 static void test_secondary_processor_based_ctls(void)
3547 {
3548 	u32 primary;
3549 	u32 secondary;
3550 	unsigned bit;
3551 
3552 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3553 		return;
3554 
3555 	primary = vmcs_read(CPU_EXEC_CTRL0);
3556 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3557 
3558 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3559 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3560 	for (bit = 0; bit < 32; bit++)
3561 		test_rsvd_ctl_bit("secondary processor-based controls",
3562 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3563 
3564 	/*
3565 	 * When the "activate secondary controls" VM-execution control
3566 	 * is clear, there are no checks on the secondary controls.
3567 	 */
3568 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3569 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3570 	report(vmlaunch_succeeds(),
3571 	       "Secondary processor-based controls ignored");
3572 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3573 	vmcs_write(CPU_EXEC_CTRL0, primary);
3574 }
3575 
3576 static void try_cr3_target_count(unsigned i, unsigned max)
3577 {
3578 	report_prefix_pushf("CR3 target count 0x%x", i);
3579 	vmcs_write(CR3_TARGET_COUNT, i);
3580 	if (i <= max)
3581 		test_vmx_valid_controls();
3582 	else
3583 		test_vmx_invalid_controls();
3584 	report_prefix_pop();
3585 }
3586 
3587 /*
3588  * The CR3-target count must not be greater than 4. Future processors
3589  * may support a different number of CR3-target values. Software
3590  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3591  * number of values supported.
3592  * [Intel SDM]
3593  */
3594 static void test_cr3_targets(void)
3595 {
3596 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3597 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3598 	unsigned i;
3599 
3600 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3601 	TEST_ASSERT(supported_targets <= 256);
3602 
3603 	try_cr3_target_count(-1u, supported_targets);
3604 	try_cr3_target_count(0x80000000, supported_targets);
3605 	try_cr3_target_count(0x7fffffff, supported_targets);
3606 	for (i = 0; i <= supported_targets + 1; i++)
3607 		try_cr3_target_count(i, supported_targets);
3608 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3609 
3610 	/* VMWRITE to nonexistent target fields should fail. */
3611 	for (i = supported_targets; i < 256; i++)
3612 		TEST_ASSERT(vmcs_write(CR3_TARGET_0 + i*2, 0));
3613 }
3614 
3615 /*
3616  * Test a particular address setting in the VMCS
3617  */
3618 static void test_vmcs_addr(const char *name,
3619 			   enum Encoding encoding,
3620 			   u64 align,
3621 			   bool ignored,
3622 			   bool skip_beyond_mapped_ram,
3623 			   u64 addr)
3624 {
3625 	report_prefix_pushf("%s = %lx", name, addr);
3626 	vmcs_write(encoding, addr);
3627 	if (skip_beyond_mapped_ram &&
3628 	    addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3629 	    addr < (1ul << cpuid_maxphyaddr()))
3630 		printf("Skipping physical address beyond mapped RAM\n");
3631 	else if (ignored || (IS_ALIGNED(addr, align) &&
3632 	    addr < (1ul << cpuid_maxphyaddr())))
3633 		test_vmx_valid_controls();
3634 	else
3635 		test_vmx_invalid_controls();
3636 	report_prefix_pop();
3637 }
3638 
3639 /*
3640  * Test interesting values for a VMCS address
3641  */
3642 static void test_vmcs_addr_values(const char *name,
3643 				  enum Encoding encoding,
3644 				  u64 align,
3645 				  bool ignored,
3646 				  bool skip_beyond_mapped_ram,
3647 				  u32 bit_start, u32 bit_end)
3648 {
3649 	unsigned i;
3650 	u64 orig_val = vmcs_read(encoding);
3651 
3652 	for (i = bit_start; i <= bit_end; i++)
3653 		test_vmcs_addr(name, encoding, align, ignored,
3654 			       skip_beyond_mapped_ram, 1ul << i);
3655 
3656 	test_vmcs_addr(name, encoding, align, ignored,
3657 		       skip_beyond_mapped_ram, PAGE_SIZE - 1);
3658 	test_vmcs_addr(name, encoding, align, ignored,
3659 		       skip_beyond_mapped_ram, PAGE_SIZE);
3660 	test_vmcs_addr(name, encoding, align, ignored,
3661 		       skip_beyond_mapped_ram,
3662 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3663 	test_vmcs_addr(name, encoding, align, ignored,
3664 		       skip_beyond_mapped_ram, -1ul);
3665 
3666 	vmcs_write(encoding, orig_val);
3667 }
3668 
3669 /*
3670  * Test a physical address reference in the VMCS, when the corresponding
3671  * feature is enabled and when the corresponding feature is disabled.
3672  */
3673 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3674 				     const char *field_name,
3675 				     const char *control_name, u64 align,
3676 				     bool skip_beyond_mapped_ram,
3677 				     bool control_primary)
3678 {
3679 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3680 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3681 	u64 page_addr;
3682 
3683 	if (control_primary) {
3684 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3685 			return;
3686 	} else {
3687 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3688 			return;
3689 	}
3690 
3691 	page_addr = vmcs_read(field);
3692 
3693 	report_prefix_pushf("%s enabled", control_name);
3694 	if (control_primary) {
3695 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3696 	} else {
3697 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3698 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3699 	}
3700 
3701 	test_vmcs_addr_values(field_name, field, align, false,
3702 			      skip_beyond_mapped_ram, 0, 63);
3703 	report_prefix_pop();
3704 
3705 	report_prefix_pushf("%s disabled", control_name);
3706 	if (control_primary) {
3707 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3708 	} else {
3709 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3710 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3711 	}
3712 
3713 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3714 	report_prefix_pop();
3715 
3716 	vmcs_write(field, page_addr);
3717 	vmcs_write(CPU_EXEC_CTRL0, primary);
3718 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3719 }
3720 
3721 /*
3722  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3723  * each I/O-bitmap address must be 0. Neither address should set any
3724  * bits beyond the processor's physical-address width.
3725  * [Intel SDM]
3726  */
3727 static void test_io_bitmaps(void)
3728 {
3729 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3730 				 "I/O bitmap A", "Use I/O bitmaps",
3731 				 PAGE_SIZE, false, true);
3732 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3733 				 "I/O bitmap B", "Use I/O bitmaps",
3734 				 PAGE_SIZE, false, true);
3735 }
3736 
3737 /*
3738  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3739  * the MSR-bitmap address must be 0. The address should not set any
3740  * bits beyond the processor's physical-address width.
3741  * [Intel SDM]
3742  */
3743 static void test_msr_bitmap(void)
3744 {
3745 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3746 				 "MSR bitmap", "Use MSR bitmaps",
3747 				 PAGE_SIZE, false, true);
3748 }
3749 
3750 /*
3751  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3752  * address must satisfy the following checks:
3753  * - Bits 11:0 of the address must be 0.
3754  * - The address should not set any bits beyond the processor's
3755  *   physical-address width.
3756  * [Intel SDM]
3757  */
3758 static void test_apic_virt_addr(void)
3759 {
3760 	/*
3761 	 * Ensure the processor will never use the virtual-APIC page, since
3762 	 * we will point it to invalid RAM.  Otherwise KVM is puzzled about
3763 	 * what we're trying to achieve and fails vmentry.
3764 	 */
3765 	u32 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
3766 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0 | CPU_CR8_LOAD | CPU_CR8_STORE);
3767 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3768 				 "virtual-APIC address", "Use TPR shadow",
3769 				 PAGE_SIZE, false, true);
3770 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
3771 }
3772 
3773 /*
3774  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3775  * APIC-access address must satisfy the following checks:
3776  *  - Bits 11:0 of the address must be 0.
3777  *  - The address should not set any bits beyond the processor's
3778  *    physical-address width.
3779  * [Intel SDM]
3780  */
3781 static void test_apic_access_addr(void)
3782 {
3783 	void *apic_access_page = alloc_page();
3784 
3785 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3786 
3787 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3788 				 "APIC-access address",
3789 				 "virtualize APIC-accesses", PAGE_SIZE,
3790 				 true, false);
3791 }
3792 
3793 static bool set_bit_pattern(u8 mask, u32 *secondary)
3794 {
3795 	u8 i;
3796 	bool flag = false;
3797 	u32 test_bits[3] = {
3798 		CPU_VIRT_X2APIC,
3799 		CPU_APIC_REG_VIRT,
3800 		CPU_VINTD
3801 	};
3802 
3803         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3804 		if ((mask & (1u << i)) &&
3805 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3806 			*secondary |= test_bits[i];
3807 			flag = true;
3808 		}
3809 	}
3810 
3811 	return (flag);
3812 }
3813 
3814 /*
3815  * If the "use TPR shadow" VM-execution control is 0, the following
3816  * VM-execution controls must also be 0:
3817  * 	- virtualize x2APIC mode
3818  *	- APIC-register virtualization
3819  *	- virtual-interrupt delivery
3820  *    [Intel SDM]
3821  *
3822  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3823  *    "virtualize APIC accesses" VM-execution control must be 0.
3824  *    [Intel SDM]
3825  */
3826 static void test_apic_virtual_ctls(void)
3827 {
3828 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3829 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3830 	u32 primary = saved_primary;
3831 	u32 secondary = saved_secondary;
3832 	bool ctrl = false;
3833 	char str[10] = "disabled";
3834 	u8 i = 0, j;
3835 
3836 	/*
3837 	 * First test
3838 	 */
3839 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3840 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3841 		return;
3842 
3843 	primary |= CPU_SECONDARY;
3844 	primary &= ~CPU_TPR_SHADOW;
3845 	vmcs_write(CPU_EXEC_CTRL0, primary);
3846 
3847 	while (1) {
3848 		for (j = 1; j < 8; j++) {
3849 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3850 			if (primary & CPU_TPR_SHADOW) {
3851 				ctrl = true;
3852 			} else {
3853 				if (! set_bit_pattern(j, &secondary))
3854 					ctrl = true;
3855 				else
3856 					ctrl = false;
3857 			}
3858 
3859 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3860 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3861 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3862 			if (ctrl)
3863 				test_vmx_valid_controls();
3864 			else
3865 				test_vmx_invalid_controls();
3866 			report_prefix_pop();
3867 		}
3868 
3869 		if (i == 1)
3870 			break;
3871 		i++;
3872 
3873 		primary |= CPU_TPR_SHADOW;
3874 		vmcs_write(CPU_EXEC_CTRL0, primary);
3875 		strcpy(str, "enabled");
3876 	}
3877 
3878 	/*
3879 	 * Second test
3880 	 */
3881 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3882 
3883 	primary = saved_primary;
3884 	secondary = saved_secondary;
3885 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3886 		return;
3887 
3888 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3889 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3890 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3891 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3892 	test_vmx_valid_controls();
3893 	report_prefix_pop();
3894 
3895 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3896 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3897 	test_vmx_valid_controls();
3898 	report_prefix_pop();
3899 
3900 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3901 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3902 	test_vmx_invalid_controls();
3903 	report_prefix_pop();
3904 
3905 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3906 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3907 	test_vmx_valid_controls();
3908 	report_prefix_pop();
3909 
3910 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3911 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3912 }
3913 
3914 /*
3915  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3916  * "external-interrupt exiting" VM-execution control must be 1.
3917  * [Intel SDM]
3918  */
3919 static void test_virtual_intr_ctls(void)
3920 {
3921 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3922 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3923 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3924 	u32 primary = saved_primary;
3925 	u32 secondary = saved_secondary;
3926 	u32 pin = saved_pin;
3927 
3928 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3929 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
3930 		return;
3931 
3932 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3933 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
3934 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3935 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
3936 	test_vmx_valid_controls();
3937 	report_prefix_pop();
3938 
3939 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
3940 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3941 	test_vmx_invalid_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 enabled");
3946 	test_vmx_valid_controls();
3947 	report_prefix_pop();
3948 
3949 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3950 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3951 	test_vmx_invalid_controls();
3952 	report_prefix_pop();
3953 
3954 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3955 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3956 	vmcs_write(PIN_CONTROLS, saved_pin);
3957 }
3958 
3959 static void test_pi_desc_addr(u64 addr, bool ctrl)
3960 {
3961 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
3962 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
3963 	if (ctrl)
3964 		test_vmx_valid_controls();
3965 	else
3966 		test_vmx_invalid_controls();
3967 	report_prefix_pop();
3968 }
3969 
3970 /*
3971  * If the "process posted interrupts" VM-execution control is 1, the
3972  * following must be true:
3973  *
3974  *	- The "virtual-interrupt delivery" VM-execution control is 1.
3975  *	- The "acknowledge interrupt on exit" VM-exit control is 1.
3976  *	- The posted-interrupt notification vector has a value in the
3977  *	- range 0 - 255 (bits 15:8 are all 0).
3978  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
3979  *	- The posted-interrupt descriptor address does not set any bits
3980  *	  beyond the processor's physical-address width.
3981  * [Intel SDM]
3982  */
3983 static void test_posted_intr(void)
3984 {
3985 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3986 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3987 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3988 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
3989 	u32 primary = saved_primary;
3990 	u32 secondary = saved_secondary;
3991 	u32 pin = saved_pin;
3992 	u32 exit_ctl = exit_ctl_saved;
3993 	u16 vec;
3994 	int i;
3995 
3996 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
3997 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3998 	    (ctrl_exit_rev.clr & EXI_INTA)))
3999 		return;
4000 
4001 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
4002 
4003 	/*
4004 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
4005 	 */
4006 	pin |= PIN_POST_INTR;
4007 	vmcs_write(PIN_CONTROLS, pin);
4008 	secondary &= ~CPU_VINTD;
4009 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4010 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
4011 	test_vmx_invalid_controls();
4012 	report_prefix_pop();
4013 
4014 	secondary |= CPU_VINTD;
4015 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4016 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
4017 	test_vmx_invalid_controls();
4018 	report_prefix_pop();
4019 
4020 	exit_ctl &= ~EXI_INTA;
4021 	vmcs_write(EXI_CONTROLS, exit_ctl);
4022 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
4023 	test_vmx_invalid_controls();
4024 	report_prefix_pop();
4025 
4026 	exit_ctl |= EXI_INTA;
4027 	vmcs_write(EXI_CONTROLS, exit_ctl);
4028 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4029 	test_vmx_valid_controls();
4030 	report_prefix_pop();
4031 
4032 	secondary &= ~CPU_VINTD;
4033 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4034 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
4035 	test_vmx_invalid_controls();
4036 	report_prefix_pop();
4037 
4038 	secondary |= CPU_VINTD;
4039 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4040 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4041 	test_vmx_valid_controls();
4042 	report_prefix_pop();
4043 
4044 	/*
4045 	 * Test posted-interrupt notification vector
4046 	 */
4047 	for (i = 0; i < 8; i++) {
4048 		vec = (1ul << i);
4049 		vmcs_write(PINV, vec);
4050 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4051 		test_vmx_valid_controls();
4052 		report_prefix_pop();
4053 	}
4054 	for (i = 8; i < 16; i++) {
4055 		vec = (1ul << i);
4056 		vmcs_write(PINV, vec);
4057 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4058 		test_vmx_invalid_controls();
4059 		report_prefix_pop();
4060 	}
4061 
4062 	vec &= ~(0xff << 8);
4063 	vmcs_write(PINV, vec);
4064 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4065 	test_vmx_valid_controls();
4066 	report_prefix_pop();
4067 
4068 	/*
4069 	 * Test posted-interrupt descriptor addresss
4070 	 */
4071 	for (i = 0; i < 6; i++) {
4072 		test_pi_desc_addr(1ul << i, false);
4073 	}
4074 
4075 	test_pi_desc_addr(0xf0, false);
4076 	test_pi_desc_addr(0xff, false);
4077 	test_pi_desc_addr(0x0f, false);
4078 	test_pi_desc_addr(0x8000, true);
4079 	test_pi_desc_addr(0x00, true);
4080 	test_pi_desc_addr(0xc000, true);
4081 
4082 	test_vmcs_addr_values("process-posted interrupts",
4083 			       POSTED_INTR_DESC_ADDR, 64,
4084 			       false, false, 0, 63);
4085 
4086 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4087 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4088 	vmcs_write(PIN_CONTROLS, saved_pin);
4089 }
4090 
4091 static void test_apic_ctls(void)
4092 {
4093 	test_apic_virt_addr();
4094 	test_apic_access_addr();
4095 	test_apic_virtual_ctls();
4096 	test_virtual_intr_ctls();
4097 	test_posted_intr();
4098 }
4099 
4100 /*
4101  * If the "enable VPID" VM-execution control is 1, the value of the
4102  * of the VPID VM-execution control field must not be 0000H.
4103  * [Intel SDM]
4104  */
4105 static void test_vpid(void)
4106 {
4107 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4108 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4109 	u16 vpid = 0x0000;
4110 	int i;
4111 
4112 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4113 	    (ctrl_cpu_rev[1].clr & CPU_VPID))) {
4114 		printf("Secondary controls and/or VPID not supported\n");
4115 		return;
4116 	}
4117 
4118 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
4119 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
4120 	vmcs_write(VPID, vpid);
4121 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
4122 	test_vmx_valid_controls();
4123 	report_prefix_pop();
4124 
4125 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
4126 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4127 	test_vmx_invalid_controls();
4128 	report_prefix_pop();
4129 
4130 	for (i = 0; i < 16; i++) {
4131 		vpid = (short)1 << i;;
4132 		vmcs_write(VPID, vpid);
4133 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4134 		test_vmx_valid_controls();
4135 		report_prefix_pop();
4136 	}
4137 
4138 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4139 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4140 }
4141 
4142 static void set_vtpr(unsigned vtpr)
4143 {
4144 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
4145 }
4146 
4147 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
4148 {
4149 	bool valid = true;
4150 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4151 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4152 
4153 	if ((primary & CPU_TPR_SHADOW) &&
4154 	    (!(primary & CPU_SECONDARY) ||
4155 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
4156 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
4157 
4158 	set_vtpr(vtpr);
4159 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
4160 	    threshold, (vtpr >> 4) & 0xf);
4161 	if (valid)
4162 		test_vmx_valid_controls();
4163 	else
4164 		test_vmx_invalid_controls();
4165 	report_prefix_pop();
4166 }
4167 
4168 static void test_invalid_event_injection(void)
4169 {
4170 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
4171 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
4172 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
4173 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
4174 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
4175 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4176 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4177 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4178 	u32 cnt;
4179 
4180 	/* Setup */
4181 	report_prefix_push("invalid event injection");
4182 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4183 	vmcs_write(ENT_INST_LEN, 0x00000001);
4184 
4185 	/* The field's interruption type is not set to a reserved value. */
4186 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4187 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4188 			    "RESERVED interruption type invalid [-]",
4189 			    ent_intr_info);
4190 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4191 	test_vmx_invalid_controls();
4192 	report_prefix_pop();
4193 
4194 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4195 			DE_VECTOR;
4196 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4197 			    "RESERVED interruption type invalid [+]",
4198 			    ent_intr_info);
4199 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4200 	test_vmx_valid_controls();
4201 	report_prefix_pop();
4202 
4203 	/* If the interruption type is other event, the vector is 0. */
4204 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4205 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4206 			    "(OTHER EVENT && vector != 0) invalid [-]",
4207 			    ent_intr_info);
4208 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4209 	test_vmx_invalid_controls();
4210 	report_prefix_pop();
4211 
4212 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4213 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4214 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4215 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4216 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4217 	test_vmx_invalid_controls();
4218 	report_prefix_pop();
4219 
4220 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4221 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4222 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4223 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4224 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4225 	test_vmx_valid_controls();
4226 	report_prefix_pop();
4227 
4228 	/*
4229 	 * If the interruption type
4230 	 * is HW exception, the vector is at most 31.
4231 	 */
4232 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4233 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4234 			    "(HW exception && vector > 31) invalid [-]",
4235 			    ent_intr_info);
4236 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4237 	test_vmx_invalid_controls();
4238 	report_prefix_pop();
4239 
4240 	/*
4241 	 * deliver-error-code is 1 iff either
4242 	 * (a) the "unrestricted guest" VM-execution control is 0
4243 	 * (b) CR0.PE is set.
4244 	 */
4245 
4246 	/* Assert that unrestricted guest is disabled or unsupported */
4247 	assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
4248 	       !(secondary_save & CPU_URG));
4249 
4250 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4251 			GP_VECTOR;
4252 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4253 			    "error code <-> (!URG || prot_mode) [-]",
4254 			    ent_intr_info);
4255 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4256 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4257 	test_vmx_invalid_controls();
4258 	report_prefix_pop();
4259 
4260 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4261 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4262 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4263 			    "error code <-> (!URG || prot_mode) [+]",
4264 			    ent_intr_info);
4265 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4266 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4267 	test_vmx_valid_controls();
4268 	report_prefix_pop();
4269 
4270 	if (enable_unrestricted_guest(false))
4271 		goto skip_unrestricted_guest;
4272 
4273 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4274 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4275 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4276 			    "error code <-> (!URG || prot_mode) [-]",
4277 			    ent_intr_info);
4278 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4279 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4280 	test_vmx_invalid_controls();
4281 	report_prefix_pop();
4282 
4283 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4284 			GP_VECTOR;
4285 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4286 			    "error code <-> (!URG || prot_mode) [-]",
4287 			    ent_intr_info);
4288 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4289 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4290 	test_vmx_invalid_controls();
4291 	report_prefix_pop();
4292 
4293 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4294 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4295 
4296 skip_unrestricted_guest:
4297 	vmcs_write(GUEST_CR0, guest_cr0_save);
4298 
4299 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4300 	report_prefix_push("error code <-> HW exception");
4301 	for (cnt = 0; cnt < 8; cnt++) {
4302 		u32 exception_type_mask = cnt << 8;
4303 		u32 deliver_error_code_mask =
4304 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4305 			INTR_INFO_DELIVER_CODE_MASK : 0;
4306 
4307 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4308 				exception_type_mask | GP_VECTOR;
4309 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4310 				    ent_intr_info);
4311 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4312 		test_vmx_invalid_controls();
4313 		report_prefix_pop();
4314 	}
4315 	report_prefix_pop();
4316 
4317 	/*
4318 	 * deliver-error-code is 1 iff the the vector
4319 	 * indicates an exception that would normally deliver an error code
4320 	 */
4321 	report_prefix_push("error code <-> vector delivers error code");
4322 	for (cnt = 0; cnt < 32; cnt++) {
4323 		bool has_error_code = false;
4324 		u32 deliver_error_code_mask;
4325 
4326 		switch (cnt) {
4327 		case DF_VECTOR:
4328 		case TS_VECTOR:
4329 		case NP_VECTOR:
4330 		case SS_VECTOR:
4331 		case GP_VECTOR:
4332 		case PF_VECTOR:
4333 		case AC_VECTOR:
4334 			has_error_code = true;
4335 		case CP_VECTOR:
4336 			/* Some CPUs have error code and some do not, skip */
4337 			continue;
4338 		}
4339 
4340 		/* Negative case */
4341 		deliver_error_code_mask = has_error_code ?
4342 						0 :
4343 						INTR_INFO_DELIVER_CODE_MASK;
4344 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4345 				INTR_TYPE_HARD_EXCEPTION | cnt;
4346 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4347 				    ent_intr_info);
4348 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4349 		test_vmx_invalid_controls();
4350 		report_prefix_pop();
4351 
4352 		/* Positive case */
4353 		deliver_error_code_mask = has_error_code ?
4354 						INTR_INFO_DELIVER_CODE_MASK :
4355 						0;
4356 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4357 				INTR_TYPE_HARD_EXCEPTION | cnt;
4358 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4359 				    ent_intr_info);
4360 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4361 		test_vmx_valid_controls();
4362 		report_prefix_pop();
4363 	}
4364 	report_prefix_pop();
4365 
4366 	/* Reserved bits in the field (30:12) are 0. */
4367 	report_prefix_push("reserved bits clear");
4368 	for (cnt = 12; cnt <= 30; cnt++) {
4369 		ent_intr_info = ent_intr_info_base |
4370 				INTR_INFO_DELIVER_CODE_MASK |
4371 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4372 				(1U << cnt);
4373 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4374 				    ent_intr_info);
4375 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4376 		test_vmx_invalid_controls();
4377 		report_prefix_pop();
4378 	}
4379 	report_prefix_pop();
4380 
4381 	/*
4382 	 * If deliver-error-code is 1
4383 	 * bits 31:16 of the VM-entry exception error-code field are 0.
4384 	 */
4385 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4386 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4387 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4388 			    "VM-entry exception error code[31:16] clear",
4389 			    ent_intr_info);
4390 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4391 	for (cnt = 16; cnt <= 31; cnt++) {
4392 		ent_intr_err = 1U << cnt;
4393 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4394 				    ent_intr_err);
4395 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4396 		test_vmx_invalid_controls();
4397 		report_prefix_pop();
4398 	}
4399 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4400 	report_prefix_pop();
4401 
4402 	/*
4403 	 * If the interruption type is software interrupt, software exception,
4404 	 * or privileged software exception, the VM-entry instruction-length
4405 	 * field is in the range 0 - 15.
4406 	 */
4407 
4408 	for (cnt = 0; cnt < 3; cnt++) {
4409 		switch (cnt) {
4410 		case 0:
4411 			ent_intr_info = ent_intr_info_base |
4412 					INTR_TYPE_SOFT_INTR;
4413 			break;
4414 		case 1:
4415 			ent_intr_info = ent_intr_info_base |
4416 					INTR_TYPE_SOFT_EXCEPTION;
4417 			break;
4418 		case 2:
4419 			ent_intr_info = ent_intr_info_base |
4420 					INTR_TYPE_PRIV_SW_EXCEPTION;
4421 			break;
4422 		}
4423 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4424 				    "VM-entry instruction-length check",
4425 				    ent_intr_info);
4426 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4427 
4428 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4429 		ent_intr_len = -1;
4430 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4431 				    ent_intr_len);
4432 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4433 		test_vmx_invalid_controls();
4434 		report_prefix_pop();
4435 
4436 		/* Instruction length set to 16 should fail */
4437 		ent_intr_len = 0x00000010;
4438 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4439 				    ent_intr_len);
4440 		vmcs_write(ENT_INST_LEN, 0x00000010);
4441 		test_vmx_invalid_controls();
4442 		report_prefix_pop();
4443 
4444 		report_prefix_pop();
4445 	}
4446 
4447 	/* Cleanup */
4448 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4449 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4450 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4451 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4452 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4453 	vmcs_write(GUEST_CR0, guest_cr0_save);
4454 	report_prefix_pop();
4455 }
4456 
4457 /*
4458  * Test interesting vTPR values for a given TPR threshold.
4459  */
4460 static void test_vtpr_values(unsigned threshold)
4461 {
4462 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4463 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4464 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4465 }
4466 
4467 static void try_tpr_threshold(unsigned threshold)
4468 {
4469 	bool valid = true;
4470 
4471 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4472 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4473 
4474 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4475 	    (secondary & CPU_VINTD)))
4476 		valid = !(threshold >> 4);
4477 
4478 	set_vtpr(-1);
4479 	vmcs_write(TPR_THRESHOLD, threshold);
4480 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4481 	if (valid)
4482 		test_vmx_valid_controls();
4483 	else
4484 		test_vmx_invalid_controls();
4485 	report_prefix_pop();
4486 
4487 	if (valid)
4488 		test_vtpr_values(threshold);
4489 }
4490 
4491 /*
4492  * Test interesting TPR threshold values.
4493  */
4494 static void test_tpr_threshold_values(void)
4495 {
4496 	unsigned i;
4497 
4498 	for (i = 0; i < 0x10; i++)
4499 		try_tpr_threshold(i);
4500 	for (i = 4; i < 32; i++)
4501 		try_tpr_threshold(1u << i);
4502 	try_tpr_threshold(-1u);
4503 	try_tpr_threshold(0x7fffffff);
4504 }
4505 
4506 /*
4507  * This test covers the following two VM entry checks:
4508  *
4509  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4510  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4511  *         31:4 of the TPR threshold VM-execution control field must
4512 	   be 0.
4513  *         [Intel SDM]
4514  *
4515  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4516  *          "virtual-interrupt delivery" VM-execution control is 0
4517  *          and the "virtualize APIC accesses" VM-execution control
4518  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4519  *          control field must not be greater than the value of bits
4520  *          7:4 of VTPR.
4521  *          [Intel SDM]
4522  */
4523 static void test_tpr_threshold(void)
4524 {
4525 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4526 	u64 apic_virt_addr = vmcs_read(APIC_VIRT_ADDR);
4527 	u64 threshold = vmcs_read(TPR_THRESHOLD);
4528 	void *virtual_apic_page;
4529 
4530 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4531 		return;
4532 
4533 	virtual_apic_page = alloc_page();
4534 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4535 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4536 
4537 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4538 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4539 	test_tpr_threshold_values();
4540 	report_prefix_pop();
4541 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4542 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4543 	test_tpr_threshold_values();
4544 	report_prefix_pop();
4545 
4546 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4547 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES))))
4548 		goto out;
4549 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4550 
4551 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4552 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4553 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4554 		test_tpr_threshold_values();
4555 		report_prefix_pop();
4556 
4557 		vmcs_write(CPU_EXEC_CTRL0,
4558 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4559 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4560 		test_tpr_threshold_values();
4561 		report_prefix_pop();
4562 	}
4563 
4564 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4565 		vmcs_write(CPU_EXEC_CTRL0,
4566 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4567 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4568 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4569 		test_tpr_threshold_values();
4570 		report_prefix_pop();
4571 
4572 		vmcs_write(CPU_EXEC_CTRL0,
4573 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4574 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4575 		test_tpr_threshold_values();
4576 		report_prefix_pop();
4577 	}
4578 
4579 	if ((ctrl_cpu_rev[1].clr &
4580 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4581 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4582 		vmcs_write(CPU_EXEC_CTRL0,
4583 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4584 		vmcs_write(CPU_EXEC_CTRL1,
4585 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4586 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4587 		test_tpr_threshold_values();
4588 		report_prefix_pop();
4589 
4590 		vmcs_write(CPU_EXEC_CTRL0,
4591 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4592 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4593 		test_tpr_threshold_values();
4594 		report_prefix_pop();
4595 	}
4596 
4597 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4598 out:
4599 	vmcs_write(TPR_THRESHOLD, threshold);
4600 	vmcs_write(APIC_VIRT_ADDR, apic_virt_addr);
4601 	vmcs_write(CPU_EXEC_CTRL0, primary);
4602 }
4603 
4604 /*
4605  * This test verifies the following two vmentry checks:
4606  *
4607  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4608  *  VM-execution control must be 0.
4609  *  [Intel SDM]
4610  *
4611  *  If the "virtual NMIs" VM-execution control is 0, the "NMI-window
4612  *  exiting" VM-execution control must be 0.
4613  *  [Intel SDM]
4614  */
4615 static void test_nmi_ctrls(void)
4616 {
4617 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4618 
4619 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4620 	    (PIN_NMI | PIN_VIRT_NMI)) {
4621 		printf("NMI exiting and Virtual NMIs are not supported !\n");
4622 		return;
4623 	}
4624 
4625 	/* Save the controls so that we can restore them after our tests */
4626 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4627 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4628 
4629 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4630 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4631 
4632 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4633 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4634 	test_vmx_valid_controls();
4635 	report_prefix_pop();
4636 
4637 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4638 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4639 	test_vmx_invalid_controls();
4640 	report_prefix_pop();
4641 
4642 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4643 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4644 	test_vmx_valid_controls();
4645 	report_prefix_pop();
4646 
4647 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4648 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4649 	test_vmx_valid_controls();
4650 	report_prefix_pop();
4651 
4652 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4653 		report_info("NMI-window exiting is not supported, skipping...");
4654 		goto done;
4655 	}
4656 
4657 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4658 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4659 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4660 	test_vmx_invalid_controls();
4661 	report_prefix_pop();
4662 
4663 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4664 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4665 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
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 | CPU_NMI_WINDOW);
4671 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4672 	test_vmx_valid_controls();
4673 	report_prefix_pop();
4674 
4675 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4676 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4677 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4678 	test_vmx_valid_controls();
4679 	report_prefix_pop();
4680 
4681 	/* Restore the controls to their original values */
4682 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4683 done:
4684 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4685 }
4686 
4687 static void test_eptp_ad_bit(u64 eptp, bool ctrl)
4688 {
4689 	vmcs_write(EPTP, eptp);
4690 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4691 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4692 	if (ctrl)
4693 		test_vmx_valid_controls();
4694 	else
4695 		test_vmx_invalid_controls();
4696 	report_prefix_pop();
4697 
4698 }
4699 
4700 /*
4701  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4702  *    control field must satisfy the following checks:
4703  *
4704  *     - The EPT memory type (bits 2:0) must be a value supported by the
4705  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4706  *     - Bits 5:3 (1 less than the EPT page-walk length) must indicate a
4707  *	 supported EPT page-walk length.
4708  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4709  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4710  *	 indicating that the processor does not support accessed and dirty
4711  *	 dirty flags for EPT.
4712  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4713  *	 physical-address width) must all be 0.
4714  *
4715  * 2. If the "unrestricted guest" VM-execution control is 1, the
4716  *    "enable EPT" VM-execution control must also be 1.
4717  */
4718 static void test_ept_eptp(void)
4719 {
4720 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4721 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4722 	u64 eptp_saved = vmcs_read(EPTP);
4723 	u32 primary = primary_saved;
4724 	u32 secondary = secondary_saved;
4725 	u64 msr, eptp = eptp_saved;
4726 	bool un_cache = false;
4727 	bool wr_bk = false;
4728 	bool ctrl;
4729 	u32 i, maxphysaddr;
4730 	u64 j, resv_bits_mask = 0;
4731 
4732 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4733 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4734 		printf("\"CPU secondary\" and/or \"enable EPT\" execution controls are not supported !\n");
4735 		return;
4736 	}
4737 
4738 	/*
4739 	 * Memory type (bits 2:0)
4740 	 */
4741 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
4742 	if (msr & EPT_CAP_UC)
4743 		un_cache = true;
4744 	if (msr & EPT_CAP_WB)
4745 		wr_bk = true;
4746 
4747 	/* Support for 4-level EPT is mandatory. */
4748 	report(msr & EPT_CAP_PWL4, "4-level EPT support check");
4749 
4750 	primary |= CPU_SECONDARY;
4751 	vmcs_write(CPU_EXEC_CTRL0, primary);
4752 	secondary |= CPU_EPT;
4753 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4754 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4755 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4756 	vmcs_write(EPTP, eptp);
4757 
4758 	for (i = 0; i < 8; i++) {
4759 		if (i == 0) {
4760 			if (un_cache) {
4761 				report_info("EPT paging structure memory-type is Un-cacheable\n");
4762 				ctrl = true;
4763 			} else {
4764 				ctrl = false;
4765 			}
4766 		} else if (i == 6) {
4767 			if (wr_bk) {
4768 				report_info("EPT paging structure memory-type is Write-back\n");
4769 				ctrl = true;
4770 			} else {
4771 				ctrl = false;
4772 			}
4773 		} else {
4774 			ctrl = false;
4775 		}
4776 
4777 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4778 		vmcs_write(EPTP, eptp);
4779 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4780 		    eptp & EPT_MEM_TYPE_MASK);
4781 		if (ctrl)
4782 			test_vmx_valid_controls();
4783 		else
4784 			test_vmx_invalid_controls();
4785 		report_prefix_pop();
4786 	}
4787 
4788 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4789 
4790 	/*
4791 	 * Page walk length (bits 5:3).  Note, the value in VMCS.EPTP "is 1
4792 	 * less than the EPT page-walk length".
4793 	 */
4794 	for (i = 0; i < 8; i++) {
4795 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4796 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4797 		if (i == 3 || (i == 4 && (msr & EPT_CAP_PWL5)))
4798 			ctrl = true;
4799 		else
4800 			ctrl = false;
4801 
4802 		vmcs_write(EPTP, eptp);
4803 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4804 		    eptp & EPTP_PG_WALK_LEN_MASK);
4805 		if (ctrl)
4806 			test_vmx_valid_controls();
4807 		else
4808 			test_vmx_invalid_controls();
4809 		report_prefix_pop();
4810 	}
4811 
4812 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4813 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4814 
4815 	/*
4816 	 * Accessed and dirty flag (bit 6)
4817 	 */
4818 	if (msr & EPT_CAP_AD_FLAG) {
4819 		report_info("Processor supports accessed and dirty flag");
4820 		eptp &= ~EPTP_AD_FLAG;
4821 		test_eptp_ad_bit(eptp, true);
4822 
4823 		eptp |= EPTP_AD_FLAG;
4824 		test_eptp_ad_bit(eptp, true);
4825 	} else {
4826 		report_info("Processor does not supports accessed and dirty flag");
4827 		eptp &= ~EPTP_AD_FLAG;
4828 		test_eptp_ad_bit(eptp, true);
4829 
4830 		eptp |= EPTP_AD_FLAG;
4831 		test_eptp_ad_bit(eptp, false);
4832 	}
4833 
4834 	/*
4835 	 * Reserved bits [11:7] and [63:N]
4836 	 */
4837 	for (i = 0; i < 32; i++) {
4838 		eptp = (eptp &
4839 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4840 		    (i << EPTP_RESERV_BITS_SHIFT);
4841 		vmcs_write(EPTP, eptp);
4842 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4843 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4844 		    EPTP_RESERV_BITS_MASK);
4845 		if (i == 0)
4846 			test_vmx_valid_controls();
4847 		else
4848 			test_vmx_invalid_controls();
4849 		report_prefix_pop();
4850 	}
4851 
4852 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4853 
4854 	maxphysaddr = cpuid_maxphyaddr();
4855 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4856 		resv_bits_mask |= 1ul << i;
4857 	}
4858 
4859 	for (j = maxphysaddr - 1; j <= 63; j++) {
4860 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4861 		    (j < maxphysaddr ? 0 : 1ul << j);
4862 		vmcs_write(EPTP, eptp);
4863 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4864 		    (eptp >> maxphysaddr) & resv_bits_mask);
4865 		if (j < maxphysaddr)
4866 			test_vmx_valid_controls();
4867 		else
4868 			test_vmx_invalid_controls();
4869 		report_prefix_pop();
4870 	}
4871 
4872 	secondary &= ~(CPU_EPT | CPU_URG);
4873 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4874 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4875 	test_vmx_valid_controls();
4876 	report_prefix_pop();
4877 
4878 	if (!(ctrl_cpu_rev[1].clr & CPU_URG))
4879 		goto skip_unrestricted_guest;
4880 
4881 	secondary |= CPU_URG;
4882 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4883 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4884 	test_vmx_invalid_controls();
4885 	report_prefix_pop();
4886 
4887 	secondary |= CPU_EPT;
4888 	setup_dummy_ept();
4889 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4890 	test_vmx_valid_controls();
4891 	report_prefix_pop();
4892 
4893 skip_unrestricted_guest:
4894 	secondary &= ~CPU_URG;
4895 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4896 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4897 	test_vmx_valid_controls();
4898 	report_prefix_pop();
4899 
4900 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4901 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4902 	vmcs_write(EPTP, eptp_saved);
4903 }
4904 
4905 /*
4906  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4907  * VM-execution control must also be 1. In addition, the PML address
4908  * must satisfy the following checks:
4909  *
4910  *    * Bits 11:0 of the address must be 0.
4911  *    * The address should not set any bits beyond the processor's
4912  *	physical-address width.
4913  *
4914  *  [Intel SDM]
4915  */
4916 static void test_pml(void)
4917 {
4918 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4919 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4920 	u32 primary = primary_saved;
4921 	u32 secondary = secondary_saved;
4922 
4923 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4924 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
4925 		printf("\"Secondary execution\" control or \"enable EPT\" control or \"enable PML\" control is not supported !\n");
4926 		return;
4927 	}
4928 
4929 	primary |= CPU_SECONDARY;
4930 	vmcs_write(CPU_EXEC_CTRL0, primary);
4931 	secondary &= ~(CPU_PML | CPU_EPT);
4932 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4933 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
4934 	test_vmx_valid_controls();
4935 	report_prefix_pop();
4936 
4937 	secondary |= CPU_PML;
4938 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4939 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
4940 	test_vmx_invalid_controls();
4941 	report_prefix_pop();
4942 
4943 	secondary |= CPU_EPT;
4944 	setup_dummy_ept();
4945 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
4946 	test_vmx_valid_controls();
4947 	report_prefix_pop();
4948 
4949 	secondary &= ~CPU_PML;
4950 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4951 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
4952 	test_vmx_valid_controls();
4953 	report_prefix_pop();
4954 
4955 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
4956 				 PAGE_SIZE, false, false);
4957 
4958 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4959 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4960 }
4961 
4962  /*
4963  * If the "activate VMX-preemption timer" VM-execution control is 0, the
4964  * the "save VMX-preemption timer value" VM-exit control must also be 0.
4965  *
4966  *  [Intel SDM]
4967  */
4968 static void test_vmx_preemption_timer(void)
4969 {
4970 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4971 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
4972 	u32 pin = saved_pin;
4973 	u32 exit = saved_exit;
4974 
4975 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
4976 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
4977 		printf("\"Save-VMX-preemption-timer\" control and/or \"Enable-VMX-preemption-timer\" control is not supported\n");
4978 		return;
4979 	}
4980 
4981 	pin |= PIN_PREEMPT;
4982 	vmcs_write(PIN_CONTROLS, pin);
4983 	exit &= ~EXI_SAVE_PREEMPT;
4984 	vmcs_write(EXI_CONTROLS, exit);
4985 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
4986 	test_vmx_valid_controls();
4987 	report_prefix_pop();
4988 
4989 	exit |= EXI_SAVE_PREEMPT;
4990 	vmcs_write(EXI_CONTROLS, exit);
4991 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
4992 	test_vmx_valid_controls();
4993 	report_prefix_pop();
4994 
4995 	pin &= ~PIN_PREEMPT;
4996 	vmcs_write(PIN_CONTROLS, pin);
4997 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
4998 	test_vmx_invalid_controls();
4999 	report_prefix_pop();
5000 
5001 	exit &= ~EXI_SAVE_PREEMPT;
5002 	vmcs_write(EXI_CONTROLS, exit);
5003 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
5004 	test_vmx_valid_controls();
5005 	report_prefix_pop();
5006 
5007 	vmcs_write(PIN_CONTROLS, saved_pin);
5008 	vmcs_write(EXI_CONTROLS, saved_exit);
5009 }
5010 
5011 extern unsigned char test_mtf1;
5012 extern unsigned char test_mtf2;
5013 extern unsigned char test_mtf3;
5014 extern unsigned char test_mtf4;
5015 
5016 static void test_mtf_guest(void)
5017 {
5018 	asm ("vmcall;\n\t"
5019 	     "out %al, $0x80;\n\t"
5020 	     "test_mtf1:\n\t"
5021 	     "vmcall;\n\t"
5022 	     "out %al, $0x80;\n\t"
5023 	     "test_mtf2:\n\t"
5024 	     /*
5025 	      * Prepare for the 'MOV CR3' test. Attempt to induce a
5026 	      * general-protection fault by moving a non-canonical address into
5027 	      * CR3. The 'MOV CR3' instruction does not take an imm64 operand,
5028 	      * so we must MOV the desired value into a register first.
5029 	      *
5030 	      * MOV RAX is done before the VMCALL such that MTF is only enabled
5031 	      * for the instruction under test.
5032 	      */
5033 	     "mov $0xaaaaaaaaaaaaaaaa, %rax;\n\t"
5034 	     "vmcall;\n\t"
5035 	     "mov %rax, %cr3;\n\t"
5036 	     "test_mtf3:\n\t"
5037 	     "vmcall;\n\t"
5038 	     /*
5039 	      * ICEBP/INT1 instruction. Though the instruction is now
5040 	      * documented, don't rely on assemblers enumerating the
5041 	      * instruction. Resort to hand assembly.
5042 	      */
5043 	     ".byte 0xf1;\n\t"
5044 	     "vmcall;\n\t"
5045 	     "test_mtf4:\n\t"
5046 	     "mov $0, %eax;\n\t");
5047 }
5048 
5049 static void test_mtf_gp_handler(struct ex_regs *regs)
5050 {
5051 	regs->rip = (unsigned long) &test_mtf3;
5052 }
5053 
5054 static void test_mtf_db_handler(struct ex_regs *regs)
5055 {
5056 }
5057 
5058 static void enable_mtf(void)
5059 {
5060 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5061 
5062 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 | CPU_MTF);
5063 }
5064 
5065 static void disable_mtf(void)
5066 {
5067 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5068 
5069 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 & ~CPU_MTF);
5070 }
5071 
5072 static void enable_tf(void)
5073 {
5074 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5075 
5076 	vmcs_write(GUEST_RFLAGS, rflags | X86_EFLAGS_TF);
5077 }
5078 
5079 static void disable_tf(void)
5080 {
5081 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5082 
5083 	vmcs_write(GUEST_RFLAGS, rflags & ~X86_EFLAGS_TF);
5084 }
5085 
5086 static void report_mtf(const char *insn_name, unsigned long exp_rip)
5087 {
5088 	unsigned long rip = vmcs_read(GUEST_RIP);
5089 
5090 	assert_exit_reason(VMX_MTF);
5091 	report(rip == exp_rip, "MTF VM-exit after %s. RIP: 0x%lx (expected 0x%lx)",
5092 	       insn_name, rip, exp_rip);
5093 }
5094 
5095 static void vmx_mtf_test(void)
5096 {
5097 	unsigned long pending_dbg;
5098 	handler old_gp, old_db;
5099 
5100 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5101 		printf("CPU does not support the 'monitor trap flag' processor-based VM-execution control.\n");
5102 		return;
5103 	}
5104 
5105 	test_set_guest(test_mtf_guest);
5106 
5107 	/* Expect an MTF VM-exit after OUT instruction */
5108 	enter_guest();
5109 	skip_exit_vmcall();
5110 
5111 	enable_mtf();
5112 	enter_guest();
5113 	report_mtf("OUT", (unsigned long) &test_mtf1);
5114 	disable_mtf();
5115 
5116 	/*
5117 	 * Concurrent #DB trap and MTF on instruction boundary. Expect MTF
5118 	 * VM-exit with populated 'pending debug exceptions' VMCS field.
5119 	 */
5120 	enter_guest();
5121 	skip_exit_vmcall();
5122 
5123 	enable_mtf();
5124 	enable_tf();
5125 
5126 	enter_guest();
5127 	report_mtf("OUT", (unsigned long) &test_mtf2);
5128 	pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
5129 	report(pending_dbg & DR_STEP,
5130 	       "'pending debug exceptions' field after MTF VM-exit: 0x%lx (expected 0x%lx)",
5131 	       pending_dbg, (unsigned long) DR_STEP);
5132 
5133 	disable_mtf();
5134 	disable_tf();
5135 	vmcs_write(GUEST_PENDING_DEBUG, 0);
5136 
5137 	/*
5138 	 * #GP exception takes priority over MTF. Expect MTF VM-exit with RIP
5139 	 * advanced to first instruction of #GP handler.
5140 	 */
5141 	enter_guest();
5142 	skip_exit_vmcall();
5143 
5144 	old_gp = handle_exception(GP_VECTOR, test_mtf_gp_handler);
5145 
5146 	enable_mtf();
5147 	enter_guest();
5148 	report_mtf("MOV CR3", (unsigned long) get_idt_addr(&boot_idt[GP_VECTOR]));
5149 	disable_mtf();
5150 
5151 	/*
5152 	 * Concurrent MTF and privileged software exception (i.e. ICEBP/INT1).
5153 	 * MTF should follow the delivery of #DB trap, though the SDM doesn't
5154 	 * provide clear indication of the relative priority.
5155 	 */
5156 	enter_guest();
5157 	skip_exit_vmcall();
5158 
5159 	handle_exception(GP_VECTOR, old_gp);
5160 	old_db = handle_exception(DB_VECTOR, test_mtf_db_handler);
5161 
5162 	enable_mtf();
5163 	enter_guest();
5164 	report_mtf("INT1", (unsigned long) get_idt_addr(&boot_idt[DB_VECTOR]));
5165 	disable_mtf();
5166 
5167 	enter_guest();
5168 	skip_exit_vmcall();
5169 	handle_exception(DB_VECTOR, old_db);
5170 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK | INTR_TYPE_OTHER_EVENT);
5171 	enter_guest();
5172 	report_mtf("injected MTF", (unsigned long) &test_mtf4);
5173 	enter_guest();
5174 }
5175 
5176 extern char vmx_mtf_pdpte_guest_begin;
5177 extern char vmx_mtf_pdpte_guest_end;
5178 
5179 asm("vmx_mtf_pdpte_guest_begin:\n\t"
5180     "mov %cr0, %rax\n\t"    /* save CR0 with PG=1                 */
5181     "vmcall\n\t"            /* on return from this CR0.PG=0       */
5182     "mov %rax, %cr0\n\t"    /* restore CR0.PG=1 to enter PAE mode */
5183     "vmcall\n\t"
5184     "retq\n\t"
5185     "vmx_mtf_pdpte_guest_end:");
5186 
5187 static void vmx_mtf_pdpte_test(void)
5188 {
5189 	void *test_mtf_pdpte_guest;
5190 	pteval_t *pdpt;
5191 	u32 guest_ar_cs;
5192 	u64 guest_efer;
5193 	pteval_t *pte;
5194 	u64 guest_cr0;
5195 	u64 guest_cr3;
5196 	u64 guest_cr4;
5197 	u64 ent_ctls;
5198 	int i;
5199 
5200 	if (setup_ept(false))
5201 		return;
5202 
5203 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5204 		printf("CPU does not support 'monitor trap flag.'\n");
5205 		return;
5206 	}
5207 
5208 	if (!(ctrl_cpu_rev[1].clr & CPU_URG)) {
5209 		printf("CPU does not support 'unrestricted guest.'\n");
5210 		return;
5211 	}
5212 
5213 	vmcs_write(EXC_BITMAP, ~0);
5214 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
5215 
5216 	/*
5217 	 * Copy the guest code to an identity-mapped page.
5218 	 */
5219 	test_mtf_pdpte_guest = alloc_page();
5220 	memcpy(test_mtf_pdpte_guest, &vmx_mtf_pdpte_guest_begin,
5221 	       &vmx_mtf_pdpte_guest_end - &vmx_mtf_pdpte_guest_begin);
5222 
5223 	test_set_guest(test_mtf_pdpte_guest);
5224 
5225 	enter_guest();
5226 	skip_exit_vmcall();
5227 
5228 	/*
5229 	 * Put the guest in non-paged 32-bit protected mode, ready to enter
5230 	 * PAE mode when CR0.PG is set. CR4.PAE will already have been set
5231 	 * when the guest started out in long mode.
5232 	 */
5233 	ent_ctls = vmcs_read(ENT_CONTROLS);
5234 	vmcs_write(ENT_CONTROLS, ent_ctls & ~ENT_GUEST_64);
5235 
5236 	guest_efer = vmcs_read(GUEST_EFER);
5237 	vmcs_write(GUEST_EFER, guest_efer & ~(EFER_LMA | EFER_LME));
5238 
5239 	/*
5240 	 * Set CS access rights bits for 32-bit protected mode:
5241 	 * 3:0    B execute/read/accessed
5242 	 * 4      1 code or data
5243 	 * 6:5    0 descriptor privilege level
5244 	 * 7      1 present
5245 	 * 11:8   0 reserved
5246 	 * 12     0 available for use by system software
5247 	 * 13     0 64 bit mode not active
5248 	 * 14     1 default operation size 32-bit segment
5249 	 * 15     1 page granularity: segment limit in 4K units
5250 	 * 16     0 segment usable
5251 	 * 31:17  0 reserved
5252 	 */
5253 	guest_ar_cs = vmcs_read(GUEST_AR_CS);
5254 	vmcs_write(GUEST_AR_CS, 0xc09b);
5255 
5256 	guest_cr0 = vmcs_read(GUEST_CR0);
5257 	vmcs_write(GUEST_CR0, guest_cr0 & ~X86_CR0_PG);
5258 
5259 	guest_cr4 = vmcs_read(GUEST_CR4);
5260 	vmcs_write(GUEST_CR4, guest_cr4 & ~X86_CR4_PCIDE);
5261 
5262 	guest_cr3 = vmcs_read(GUEST_CR3);
5263 
5264 	/*
5265 	 * Turn the 4-level page table into a PAE page table by following the 0th
5266 	 * PML4 entry to a PDPT page, and grab the first four PDPTEs from that
5267 	 * page.
5268 	 *
5269 	 * Why does this work?
5270 	 *
5271 	 * PAE uses 32-bit addressing which implies:
5272 	 * Bits 11:0   page offset
5273 	 * Bits 20:12  entry into 512-entry page table
5274 	 * Bits 29:21  entry into a 512-entry directory table
5275 	 * Bits 31:30  entry into the page directory pointer table.
5276 	 * Bits 63:32  zero
5277 	 *
5278 	 * As only 2 bits are needed to select the PDPTEs for the entire
5279 	 * 32-bit address space, take the first 4 PDPTEs in the level 3 page
5280 	 * directory pointer table. It doesn't matter which of these PDPTEs
5281 	 * are present because they must cover the guest code given that it
5282 	 * has already run successfully.
5283 	 *
5284 	 * Get a pointer to PTE for GVA=0 in the page directory pointer table
5285 	 */
5286 	pte = get_pte_level(
5287             (pgd_t *)phys_to_virt(guest_cr3 & ~X86_CR3_PCID_MASK), 0,
5288             PDPT_LEVEL);
5289 
5290 	/*
5291 	 * Need some memory for the 4-entry PAE page directory pointer
5292 	 * table. Use the end of the identity-mapped page where the guest code
5293 	 * is stored. There is definitely space as the guest code is only a
5294 	 * few bytes.
5295 	 */
5296 	pdpt = test_mtf_pdpte_guest + PAGE_SIZE - 4 * sizeof(pteval_t);
5297 
5298 	/*
5299 	 * Copy the first four PDPTEs into the PAE page table with reserved
5300 	 * bits cleared. Note that permission bits from the PML4E and PDPTE
5301 	 * are not propagated.
5302 	 */
5303 	for (i = 0; i < 4; i++) {
5304 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_RSVD_MASK),
5305 				   "PDPTE has invalid reserved bits");
5306 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_PAGE_SIZE_MASK),
5307 				   "Cannot use 1GB super pages for PAE");
5308 		pdpt[i] = pte[i] & ~(PAE_PDPTE_RSVD_MASK);
5309 	}
5310 	vmcs_write(GUEST_CR3, virt_to_phys(pdpt));
5311 
5312 	enable_mtf();
5313 	enter_guest();
5314 	assert_exit_reason(VMX_MTF);
5315 	disable_mtf();
5316 
5317 	/*
5318 	 * The four PDPTEs should have been loaded into the VMCS when
5319 	 * the guest set CR0.PG to enter PAE mode.
5320 	 */
5321 	for (i = 0; i < 4; i++) {
5322 		u64 pdpte = vmcs_read(GUEST_PDPTE + 2 * i);
5323 
5324 		report(pdpte == pdpt[i], "PDPTE%d is 0x%lx (expected 0x%lx)",
5325 		       i, pdpte, pdpt[i]);
5326 	}
5327 
5328 	/*
5329 	 * Now, try to enter the guest in PAE mode. If the PDPTEs in the
5330 	 * vmcs are wrong, this will fail.
5331 	 */
5332 	enter_guest();
5333 	skip_exit_vmcall();
5334 
5335 	/*
5336 	 * Return guest to 64-bit mode and wrap up.
5337 	 */
5338 	vmcs_write(ENT_CONTROLS, ent_ctls);
5339 	vmcs_write(GUEST_EFER, guest_efer);
5340 	vmcs_write(GUEST_AR_CS, guest_ar_cs);
5341 	vmcs_write(GUEST_CR0, guest_cr0);
5342 	vmcs_write(GUEST_CR4, guest_cr4);
5343 	vmcs_write(GUEST_CR3, guest_cr3);
5344 
5345 	enter_guest();
5346 }
5347 
5348 /*
5349  * Tests for VM-execution control fields
5350  */
5351 static void test_vm_execution_ctls(void)
5352 {
5353 	test_pin_based_ctls();
5354 	test_primary_processor_based_ctls();
5355 	test_secondary_processor_based_ctls();
5356 	test_cr3_targets();
5357 	test_io_bitmaps();
5358 	test_msr_bitmap();
5359 	test_apic_ctls();
5360 	test_tpr_threshold();
5361 	test_nmi_ctrls();
5362 	test_pml();
5363 	test_vpid();
5364 	test_ept_eptp();
5365 	test_vmx_preemption_timer();
5366 }
5367 
5368  /*
5369   * The following checks are performed for the VM-entry MSR-load address if
5370   * the VM-entry MSR-load count field is non-zero:
5371   *
5372   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
5373   *      The address should not set any bits beyond the processor's
5374   *      physical-address width.
5375   *
5376   *    - The address of the last byte in the VM-entry MSR-load area
5377   *      should not set any bits beyond the processor's physical-address
5378   *      width. The address of this last byte is VM-entry MSR-load address
5379   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5380   *      uses more bits than the processor's physical-address width.)
5381   *
5382   *
5383   *  [Intel SDM]
5384   */
5385 static void test_entry_msr_load(void)
5386 {
5387 	entry_msr_load = alloc_page();
5388 	u64 tmp;
5389 	u32 entry_msr_ld_cnt = 1;
5390 	int i;
5391 	u32 addr_len = 64;
5392 
5393 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
5394 
5395 	/* Check first 4 bits of VM-entry MSR-load address */
5396 	for (i = 0; i < 4; i++) {
5397 		tmp = (u64)entry_msr_load | 1ull << i;
5398 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
5399 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
5400 				    tmp & 0xf);
5401 		test_vmx_invalid_controls();
5402 		report_prefix_pop();
5403 	}
5404 
5405 	if (basic.val & (1ul << 48))
5406 		addr_len = 32;
5407 
5408 	test_vmcs_addr_values("VM-entry-MSR-load address",
5409 				ENTER_MSR_LD_ADDR, 16, false, false,
5410 				4, addr_len - 1);
5411 
5412 	/*
5413 	 * Check last byte of VM-entry MSR-load address
5414 	 */
5415 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
5416 
5417 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5418 							i < 64; i++) {
5419 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
5420 			1ul << i;
5421 		vmcs_write(ENTER_MSR_LD_ADDR,
5422 			   tmp - (entry_msr_ld_cnt * 16 - 1));
5423 		test_vmx_invalid_controls();
5424 	}
5425 
5426 	vmcs_write(ENT_MSR_LD_CNT, 2);
5427 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5428 	test_vmx_invalid_controls();
5429 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5430 	test_vmx_valid_controls();
5431 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5432 	test_vmx_valid_controls();
5433 }
5434 
5435 static struct vmx_state_area_test_data {
5436 	u32 msr;
5437 	u64 exp;
5438 	bool enabled;
5439 } vmx_state_area_test_data;
5440 
5441 static void guest_state_test_main(void)
5442 {
5443 	u64 obs;
5444 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
5445 
5446 	while (1) {
5447 		if (vmx_get_test_stage() == 2)
5448 			break;
5449 
5450 		if (data->enabled) {
5451 			obs = rdmsr(data->msr);
5452 			report(data->exp == obs,
5453 			       "Guest state is 0x%lx (expected 0x%lx)",
5454 			       obs, data->exp);
5455 		}
5456 
5457 		vmcall();
5458 	}
5459 
5460 	asm volatile("fnop");
5461 }
5462 
5463 static void test_guest_state(const char *test, bool xfail, u64 field,
5464 			     const char * field_name)
5465 {
5466 	struct vmentry_result result;
5467 	u8 abort_flags;
5468 
5469 	abort_flags = ABORT_ON_EARLY_VMENTRY_FAIL;
5470 	if (!xfail)
5471 		abort_flags = ABORT_ON_INVALID_GUEST_STATE;
5472 
5473 	__enter_guest(abort_flags, &result);
5474 
5475 	report(result.exit_reason.failed_vmentry == xfail &&
5476 	       ((xfail && result.exit_reason.basic == VMX_FAIL_STATE) ||
5477 	        (!xfail && result.exit_reason.basic == VMX_VMCALL)) &&
5478 		(!xfail || vmcs_read(EXI_QUALIFICATION) == ENTRY_FAIL_DEFAULT),
5479 	        "%s, %s = %lx", test, field_name, field);
5480 
5481 	if (!result.exit_reason.failed_vmentry)
5482 		skip_exit_insn();
5483 }
5484 
5485 /*
5486  * Tests for VM-entry control fields
5487  */
5488 static void test_vm_entry_ctls(void)
5489 {
5490 	test_invalid_event_injection();
5491 	test_entry_msr_load();
5492 }
5493 
5494 /*
5495  * The following checks are performed for the VM-exit MSR-store address if
5496  * the VM-exit MSR-store count field is non-zero:
5497  *
5498  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
5499  *      The address should not set any bits beyond the processor's
5500  *      physical-address width.
5501  *
5502  *    - The address of the last byte in the VM-exit MSR-store area
5503  *      should not set any bits beyond the processor's physical-address
5504  *      width. The address of this last byte is VM-exit MSR-store address
5505  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5506  *      uses more bits than the processor's physical-address width.)
5507  *
5508  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
5509  * in the range 63:32.
5510  *
5511  *  [Intel SDM]
5512  */
5513 static void test_exit_msr_store(void)
5514 {
5515 	exit_msr_store = alloc_page();
5516 	u64 tmp;
5517 	u32 exit_msr_st_cnt = 1;
5518 	int i;
5519 	u32 addr_len = 64;
5520 
5521 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
5522 
5523 	/* Check first 4 bits of VM-exit MSR-store address */
5524 	for (i = 0; i < 4; i++) {
5525 		tmp = (u64)exit_msr_store | 1ull << i;
5526 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
5527 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
5528 				    tmp & 0xf);
5529 		test_vmx_invalid_controls();
5530 		report_prefix_pop();
5531 	}
5532 
5533 	if (basic.val & (1ul << 48))
5534 		addr_len = 32;
5535 
5536 	test_vmcs_addr_values("VM-exit-MSR-store address",
5537 				EXIT_MSR_ST_ADDR, 16, false, false,
5538 				4, addr_len - 1);
5539 
5540 	/*
5541 	 * Check last byte of VM-exit MSR-store address
5542 	 */
5543 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
5544 
5545 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5546 							i < 64; i++) {
5547 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
5548 			1ul << i;
5549 		vmcs_write(EXIT_MSR_ST_ADDR,
5550 			   tmp - (exit_msr_st_cnt * 16 - 1));
5551 		test_vmx_invalid_controls();
5552 	}
5553 
5554 	vmcs_write(EXI_MSR_ST_CNT, 2);
5555 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5556 	test_vmx_invalid_controls();
5557 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5558 	test_vmx_valid_controls();
5559 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5560 	test_vmx_valid_controls();
5561 }
5562 
5563 /*
5564  * Tests for VM-exit controls
5565  */
5566 static void test_vm_exit_ctls(void)
5567 {
5568 	test_exit_msr_store();
5569 }
5570 
5571 /*
5572  * Check that the virtual CPU checks all of the VMX controls as
5573  * documented in the Intel SDM.
5574  */
5575 static void vmx_controls_test(void)
5576 {
5577 	/*
5578 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
5579 	 * fail due to invalid guest state, should we make it that
5580 	 * far.
5581 	 */
5582 	vmcs_write(GUEST_RFLAGS, 0);
5583 
5584 	test_vm_execution_ctls();
5585 	test_vm_exit_ctls();
5586 	test_vm_entry_ctls();
5587 }
5588 
5589 struct apic_reg_virt_config {
5590 	bool apic_register_virtualization;
5591 	bool use_tpr_shadow;
5592 	bool virtualize_apic_accesses;
5593 	bool virtualize_x2apic_mode;
5594 	bool activate_secondary_controls;
5595 };
5596 
5597 struct apic_reg_test {
5598 	const char *name;
5599 	struct apic_reg_virt_config apic_reg_virt_config;
5600 };
5601 
5602 struct apic_reg_virt_expectation {
5603 	enum Reason rd_exit_reason;
5604 	enum Reason wr_exit_reason;
5605 	u32 val;
5606 	u32 (*virt_fn)(u32);
5607 
5608 	/*
5609 	 * If false, accessing the APIC access address from L2 is treated as a
5610 	 * normal memory operation, rather than triggering virtualization.
5611 	 */
5612 	bool virtualize_apic_accesses;
5613 };
5614 
5615 static u32 apic_virt_identity(u32 val)
5616 {
5617 	return val;
5618 }
5619 
5620 static u32 apic_virt_nibble1(u32 val)
5621 {
5622 	return val & 0xf0;
5623 }
5624 
5625 static u32 apic_virt_byte3(u32 val)
5626 {
5627 	return val & (0xff << 24);
5628 }
5629 
5630 static bool apic_reg_virt_exit_expectation(
5631 	u32 reg, struct apic_reg_virt_config *config,
5632 	struct apic_reg_virt_expectation *expectation)
5633 {
5634 	/* Good configs, where some L2 APIC accesses are virtualized. */
5635 	bool virtualize_apic_accesses_only =
5636 		config->virtualize_apic_accesses &&
5637 		!config->use_tpr_shadow &&
5638 		!config->apic_register_virtualization &&
5639 		!config->virtualize_x2apic_mode &&
5640 		config->activate_secondary_controls;
5641 	bool virtualize_apic_accesses_and_use_tpr_shadow =
5642 		config->virtualize_apic_accesses &&
5643 		config->use_tpr_shadow &&
5644 		!config->apic_register_virtualization &&
5645 		!config->virtualize_x2apic_mode &&
5646 		config->activate_secondary_controls;
5647 	bool apic_register_virtualization =
5648 		config->virtualize_apic_accesses &&
5649 		config->use_tpr_shadow &&
5650 		config->apic_register_virtualization &&
5651 		!config->virtualize_x2apic_mode &&
5652 		config->activate_secondary_controls;
5653 
5654 	expectation->val = MAGIC_VAL_1;
5655 	expectation->virt_fn = apic_virt_identity;
5656 	expectation->virtualize_apic_accesses =
5657 		config->virtualize_apic_accesses &&
5658 		config->activate_secondary_controls;
5659 	if (virtualize_apic_accesses_only) {
5660 		expectation->rd_exit_reason = VMX_APIC_ACCESS;
5661 		expectation->wr_exit_reason = VMX_APIC_ACCESS;
5662 	} else if (virtualize_apic_accesses_and_use_tpr_shadow) {
5663 		switch (reg) {
5664 		case APIC_TASKPRI:
5665 			expectation->rd_exit_reason = VMX_VMCALL;
5666 			expectation->wr_exit_reason = VMX_VMCALL;
5667 			expectation->virt_fn = apic_virt_nibble1;
5668 			break;
5669 		default:
5670 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5671 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5672 		}
5673 	} else if (apic_register_virtualization) {
5674 		expectation->rd_exit_reason = VMX_VMCALL;
5675 
5676 		switch (reg) {
5677 		case APIC_ID:
5678 		case APIC_EOI:
5679 		case APIC_LDR:
5680 		case APIC_DFR:
5681 		case APIC_SPIV:
5682 		case APIC_ESR:
5683 		case APIC_ICR:
5684 		case APIC_LVTT:
5685 		case APIC_LVTTHMR:
5686 		case APIC_LVTPC:
5687 		case APIC_LVT0:
5688 		case APIC_LVT1:
5689 		case APIC_LVTERR:
5690 		case APIC_TMICT:
5691 		case APIC_TDCR:
5692 			expectation->wr_exit_reason = VMX_APIC_WRITE;
5693 			break;
5694 		case APIC_LVR:
5695 		case APIC_ISR ... APIC_ISR + 0x70:
5696 		case APIC_TMR ... APIC_TMR + 0x70:
5697 		case APIC_IRR ... APIC_IRR + 0x70:
5698 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5699 			break;
5700 		case APIC_TASKPRI:
5701 			expectation->wr_exit_reason = VMX_VMCALL;
5702 			expectation->virt_fn = apic_virt_nibble1;
5703 			break;
5704 		case APIC_ICR2:
5705 			expectation->wr_exit_reason = VMX_VMCALL;
5706 			expectation->virt_fn = apic_virt_byte3;
5707 			break;
5708 		default:
5709 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5710 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5711 		}
5712 	} else if (!expectation->virtualize_apic_accesses) {
5713 		/*
5714 		 * No APIC registers are directly virtualized. This includes
5715 		 * VTPR, which can be virtualized through MOV to/from CR8 via
5716 		 * the use TPR shadow control, but not through directly
5717 		 * accessing VTPR.
5718 		 */
5719 		expectation->rd_exit_reason = VMX_VMCALL;
5720 		expectation->wr_exit_reason = VMX_VMCALL;
5721 	} else {
5722 		printf("Cannot parse APIC register virtualization config:\n"
5723 		       "\tvirtualize_apic_accesses: %d\n"
5724 		       "\tuse_tpr_shadow: %d\n"
5725 		       "\tapic_register_virtualization: %d\n"
5726 		       "\tvirtualize_x2apic_mode: %d\n"
5727 		       "\tactivate_secondary_controls: %d\n",
5728 		       config->virtualize_apic_accesses,
5729 		       config->use_tpr_shadow,
5730 		       config->apic_register_virtualization,
5731 		       config->virtualize_x2apic_mode,
5732 		       config->activate_secondary_controls);
5733 
5734 		return false;
5735 	}
5736 
5737 	return true;
5738 }
5739 
5740 struct apic_reg_test apic_reg_tests[] = {
5741 	/* Good configs, where some L2 APIC accesses are virtualized. */
5742 	{
5743 		.name = "Virtualize APIC accesses",
5744 		.apic_reg_virt_config = {
5745 			.virtualize_apic_accesses = true,
5746 			.use_tpr_shadow = false,
5747 			.apic_register_virtualization = false,
5748 			.virtualize_x2apic_mode = false,
5749 			.activate_secondary_controls = true,
5750 		},
5751 	},
5752 	{
5753 		.name = "Virtualize APIC accesses + Use TPR shadow",
5754 		.apic_reg_virt_config = {
5755 			.virtualize_apic_accesses = true,
5756 			.use_tpr_shadow = true,
5757 			.apic_register_virtualization = false,
5758 			.virtualize_x2apic_mode = false,
5759 			.activate_secondary_controls = true,
5760 		},
5761 	},
5762 	{
5763 		.name = "APIC-register virtualization",
5764 		.apic_reg_virt_config = {
5765 			.virtualize_apic_accesses = true,
5766 			.use_tpr_shadow = true,
5767 			.apic_register_virtualization = true,
5768 			.virtualize_x2apic_mode = false,
5769 			.activate_secondary_controls = true,
5770 		},
5771 	},
5772 
5773 	/*
5774 	 * Test that the secondary processor-based VM-execution controls are
5775 	 * correctly ignored when "activate secondary controls" is disabled.
5776 	 */
5777 	{
5778 		.name = "Activate secondary controls off",
5779 		.apic_reg_virt_config = {
5780 			.virtualize_apic_accesses = true,
5781 			.use_tpr_shadow = false,
5782 			.apic_register_virtualization = true,
5783 			.virtualize_x2apic_mode = true,
5784 			.activate_secondary_controls = false,
5785 		},
5786 	},
5787 	{
5788 		.name = "Activate secondary controls off + Use TPR shadow",
5789 		.apic_reg_virt_config = {
5790 			.virtualize_apic_accesses = true,
5791 			.use_tpr_shadow = true,
5792 			.apic_register_virtualization = true,
5793 			.virtualize_x2apic_mode = true,
5794 			.activate_secondary_controls = false,
5795 		},
5796 	},
5797 
5798 	/*
5799 	 * Test that the APIC access address is treated like an arbitrary memory
5800 	 * address when "virtualize APIC accesses" is disabled.
5801 	 */
5802 	{
5803 		.name = "Virtualize APIC accesses off + Use TPR shadow",
5804 		.apic_reg_virt_config = {
5805 			.virtualize_apic_accesses = false,
5806 			.use_tpr_shadow = true,
5807 			.apic_register_virtualization = true,
5808 			.virtualize_x2apic_mode = true,
5809 			.activate_secondary_controls = true,
5810 		},
5811 	},
5812 
5813 	/*
5814 	 * Test that VM entry fails due to invalid controls when
5815 	 * "APIC-register virtualization" is enabled while "use TPR shadow" is
5816 	 * disabled.
5817 	 */
5818 	{
5819 		.name = "APIC-register virtualization + Use TPR shadow off",
5820 		.apic_reg_virt_config = {
5821 			.virtualize_apic_accesses = true,
5822 			.use_tpr_shadow = false,
5823 			.apic_register_virtualization = true,
5824 			.virtualize_x2apic_mode = false,
5825 			.activate_secondary_controls = true,
5826 		},
5827 	},
5828 
5829 	/*
5830 	 * Test that VM entry fails due to invalid controls when
5831 	 * "Virtualize x2APIC mode" is enabled while "use TPR shadow" is
5832 	 * disabled.
5833 	 */
5834 	{
5835 		.name = "Virtualize x2APIC mode + Use TPR shadow off",
5836 		.apic_reg_virt_config = {
5837 			.virtualize_apic_accesses = false,
5838 			.use_tpr_shadow = false,
5839 			.apic_register_virtualization = false,
5840 			.virtualize_x2apic_mode = true,
5841 			.activate_secondary_controls = true,
5842 		},
5843 	},
5844 	{
5845 		.name = "Virtualize x2APIC mode + Use TPR shadow off v2",
5846 		.apic_reg_virt_config = {
5847 			.virtualize_apic_accesses = false,
5848 			.use_tpr_shadow = false,
5849 			.apic_register_virtualization = true,
5850 			.virtualize_x2apic_mode = true,
5851 			.activate_secondary_controls = true,
5852 		},
5853 	},
5854 
5855 	/*
5856 	 * Test that VM entry fails due to invalid controls when
5857 	 * "virtualize x2APIC mode" is enabled while "virtualize APIC accesses"
5858 	 * is enabled.
5859 	 */
5860 	{
5861 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses",
5862 		.apic_reg_virt_config = {
5863 			.virtualize_apic_accesses = true,
5864 			.use_tpr_shadow = true,
5865 			.apic_register_virtualization = false,
5866 			.virtualize_x2apic_mode = true,
5867 			.activate_secondary_controls = true,
5868 		},
5869 	},
5870 	{
5871 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses v2",
5872 		.apic_reg_virt_config = {
5873 			.virtualize_apic_accesses = true,
5874 			.use_tpr_shadow = true,
5875 			.apic_register_virtualization = true,
5876 			.virtualize_x2apic_mode = true,
5877 			.activate_secondary_controls = true,
5878 		},
5879 	},
5880 };
5881 
5882 enum Apic_op {
5883 	APIC_OP_XAPIC_RD,
5884 	APIC_OP_XAPIC_WR,
5885 	TERMINATE,
5886 };
5887 
5888 static u32 vmx_xapic_read(u32 *apic_access_address, u32 reg)
5889 {
5890 	return *(volatile u32 *)((uintptr_t)apic_access_address + reg);
5891 }
5892 
5893 static void vmx_xapic_write(u32 *apic_access_address, u32 reg, u32 val)
5894 {
5895 	*(volatile u32 *)((uintptr_t)apic_access_address + reg) = val;
5896 }
5897 
5898 struct apic_reg_virt_guest_args {
5899 	enum Apic_op op;
5900 	u32 *apic_access_address;
5901 	u32 reg;
5902 	u32 val;
5903 	bool check_rd;
5904 	u32 (*virt_fn)(u32);
5905 } apic_reg_virt_guest_args;
5906 
5907 static void apic_reg_virt_guest(void)
5908 {
5909 	volatile struct apic_reg_virt_guest_args *args =
5910 		&apic_reg_virt_guest_args;
5911 
5912 	for (;;) {
5913 		enum Apic_op op = args->op;
5914 		u32 *apic_access_address = args->apic_access_address;
5915 		u32 reg = args->reg;
5916 		u32 val = args->val;
5917 		bool check_rd = args->check_rd;
5918 		u32 (*virt_fn)(u32) = args->virt_fn;
5919 
5920 		if (op == TERMINATE)
5921 			break;
5922 
5923 		if (op == APIC_OP_XAPIC_RD) {
5924 			u32 ret = vmx_xapic_read(apic_access_address, reg);
5925 
5926 			if (check_rd) {
5927 				u32 want = virt_fn(val);
5928 				u32 got = virt_fn(ret);
5929 
5930 				report(got == want,
5931 				       "read 0x%x, expected 0x%x.", got, want);
5932 			}
5933 		} else if (op == APIC_OP_XAPIC_WR) {
5934 			vmx_xapic_write(apic_access_address, reg, val);
5935 		}
5936 
5937 		/*
5938 		 * The L1 should always execute a vmcall after it's done testing
5939 		 * an individual APIC operation. This helps to validate that the
5940 		 * L1 and L2 are in sync with each other, as expected.
5941 		 */
5942 		vmcall();
5943 	}
5944 }
5945 
5946 static void test_xapic_rd(
5947 	u32 reg, struct apic_reg_virt_expectation *expectation,
5948 	u32 *apic_access_address, u32 *virtual_apic_page)
5949 {
5950 	u32 val = expectation->val;
5951 	u32 exit_reason_want = expectation->rd_exit_reason;
5952 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5953 
5954 	report_prefix_pushf("xapic - reading 0x%03x", reg);
5955 
5956 	/* Configure guest to do an xapic read */
5957 	args->op = APIC_OP_XAPIC_RD;
5958 	args->apic_access_address = apic_access_address;
5959 	args->reg = reg;
5960 	args->val = val;
5961 	args->check_rd = exit_reason_want == VMX_VMCALL;
5962 	args->virt_fn = expectation->virt_fn;
5963 
5964 	/* Setup virtual APIC page */
5965 	if (!expectation->virtualize_apic_accesses) {
5966 		apic_access_address[apic_reg_index(reg)] = val;
5967 		virtual_apic_page[apic_reg_index(reg)] = 0;
5968 	} else if (exit_reason_want == VMX_VMCALL) {
5969 		apic_access_address[apic_reg_index(reg)] = 0;
5970 		virtual_apic_page[apic_reg_index(reg)] = val;
5971 	}
5972 
5973 	/* Enter guest */
5974 	enter_guest();
5975 
5976 	/*
5977 	 * Validate the behavior and
5978 	 * pass a magic value back to the guest.
5979 	 */
5980 	if (exit_reason_want == VMX_APIC_ACCESS) {
5981 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5982 
5983 		assert_exit_reason(exit_reason_want);
5984 		report(apic_page_offset == reg,
5985 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
5986 		       apic_page_offset, reg);
5987 		skip_exit_insn();
5988 
5989 		/* Reenter guest so it can consume/check rcx and exit again. */
5990 		enter_guest();
5991 	} else if (exit_reason_want != VMX_VMCALL) {
5992 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
5993 	}
5994 
5995 	skip_exit_vmcall();
5996 	report_prefix_pop();
5997 }
5998 
5999 static void test_xapic_wr(
6000 	u32 reg, struct apic_reg_virt_expectation *expectation,
6001 	u32 *apic_access_address, u32 *virtual_apic_page)
6002 {
6003 	u32 val = expectation->val;
6004 	u32 exit_reason_want = expectation->wr_exit_reason;
6005 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
6006 	bool virtualized =
6007 		expectation->virtualize_apic_accesses &&
6008 		(exit_reason_want == VMX_APIC_WRITE ||
6009 		 exit_reason_want == VMX_VMCALL);
6010 	bool checked = false;
6011 
6012 	report_prefix_pushf("xapic - writing 0x%x to 0x%03x", val, reg);
6013 
6014 	/* Configure guest to do an xapic read */
6015 	args->op = APIC_OP_XAPIC_WR;
6016 	args->apic_access_address = apic_access_address;
6017 	args->reg = reg;
6018 	args->val = val;
6019 
6020 	/* Setup virtual APIC page */
6021 	if (virtualized || !expectation->virtualize_apic_accesses) {
6022 		apic_access_address[apic_reg_index(reg)] = 0;
6023 		virtual_apic_page[apic_reg_index(reg)] = 0;
6024 	}
6025 
6026 	/* Enter guest */
6027 	enter_guest();
6028 
6029 	/*
6030 	 * Validate the behavior and
6031 	 * pass a magic value back to the guest.
6032 	 */
6033 	if (exit_reason_want == VMX_APIC_ACCESS) {
6034 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
6035 
6036 		assert_exit_reason(exit_reason_want);
6037 		report(apic_page_offset == reg,
6038 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
6039 		       apic_page_offset, reg);
6040 		skip_exit_insn();
6041 
6042 		/* Reenter guest so it can consume/check rcx and exit again. */
6043 		enter_guest();
6044 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6045 		assert_exit_reason(exit_reason_want);
6046 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6047 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%x",
6048 		       apic_reg_index(reg),
6049 		       virtual_apic_page[apic_reg_index(reg)], val);
6050 		checked = true;
6051 
6052 		/* Reenter guest so it can consume/check rcx and exit again. */
6053 		enter_guest();
6054 	} else if (exit_reason_want != VMX_VMCALL) {
6055 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6056 	}
6057 
6058 	assert_exit_reason(VMX_VMCALL);
6059 	if (virtualized && !checked) {
6060 		u32 want = expectation->virt_fn(val);
6061 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6062 		got = expectation->virt_fn(got);
6063 
6064 		report(got == want, "exitless write; val is 0x%x, want 0x%x",
6065 		       got, want);
6066 	} else if (!expectation->virtualize_apic_accesses && !checked) {
6067 		u32 got = apic_access_address[apic_reg_index(reg)];
6068 
6069 		report(got == val,
6070 		       "non-virtualized write; val is 0x%x, want 0x%x", got,
6071 		       val);
6072 	} else if (!expectation->virtualize_apic_accesses && checked) {
6073 		report_fail("Non-virtualized write was prematurely checked!");
6074 	}
6075 
6076 	skip_exit_vmcall();
6077 	report_prefix_pop();
6078 }
6079 
6080 enum Config_type {
6081 	CONFIG_TYPE_GOOD,
6082 	CONFIG_TYPE_UNSUPPORTED,
6083 	CONFIG_TYPE_VMENTRY_FAILS_EARLY,
6084 };
6085 
6086 static enum Config_type configure_apic_reg_virt_test(
6087 	struct apic_reg_virt_config *apic_reg_virt_config)
6088 {
6089 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6090 	u32 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6091 	/* Configs where L2 entry fails early, due to invalid controls. */
6092 	bool use_tpr_shadow_incorrectly_off =
6093 		!apic_reg_virt_config->use_tpr_shadow &&
6094 		(apic_reg_virt_config->apic_register_virtualization ||
6095 		 apic_reg_virt_config->virtualize_x2apic_mode) &&
6096 		apic_reg_virt_config->activate_secondary_controls;
6097 	bool virtualize_apic_accesses_incorrectly_on =
6098 		apic_reg_virt_config->virtualize_apic_accesses &&
6099 		apic_reg_virt_config->virtualize_x2apic_mode &&
6100 		apic_reg_virt_config->activate_secondary_controls;
6101 	bool vmentry_fails_early =
6102 		use_tpr_shadow_incorrectly_off ||
6103 		virtualize_apic_accesses_incorrectly_on;
6104 
6105 	if (apic_reg_virt_config->activate_secondary_controls) {
6106 		if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
6107 			printf("VM-execution control \"activate secondary controls\" NOT supported.\n");
6108 			return CONFIG_TYPE_UNSUPPORTED;
6109 		}
6110 		cpu_exec_ctrl0 |= CPU_SECONDARY;
6111 	} else {
6112 		cpu_exec_ctrl0 &= ~CPU_SECONDARY;
6113 	}
6114 
6115 	if (apic_reg_virt_config->virtualize_apic_accesses) {
6116 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES)) {
6117 			printf("VM-execution control \"virtualize APIC accesses\" NOT supported.\n");
6118 			return CONFIG_TYPE_UNSUPPORTED;
6119 		}
6120 		cpu_exec_ctrl1 |= CPU_VIRT_APIC_ACCESSES;
6121 	} else {
6122 		cpu_exec_ctrl1 &= ~CPU_VIRT_APIC_ACCESSES;
6123 	}
6124 
6125 	if (apic_reg_virt_config->use_tpr_shadow) {
6126 		if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6127 			printf("VM-execution control \"use TPR shadow\" NOT supported.\n");
6128 			return CONFIG_TYPE_UNSUPPORTED;
6129 		}
6130 		cpu_exec_ctrl0 |= CPU_TPR_SHADOW;
6131 	} else {
6132 		cpu_exec_ctrl0 &= ~CPU_TPR_SHADOW;
6133 	}
6134 
6135 	if (apic_reg_virt_config->apic_register_virtualization) {
6136 		if (!(ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT)) {
6137 			printf("VM-execution control \"APIC-register virtualization\" NOT supported.\n");
6138 			return CONFIG_TYPE_UNSUPPORTED;
6139 		}
6140 		cpu_exec_ctrl1 |= CPU_APIC_REG_VIRT;
6141 	} else {
6142 		cpu_exec_ctrl1 &= ~CPU_APIC_REG_VIRT;
6143 	}
6144 
6145 	if (apic_reg_virt_config->virtualize_x2apic_mode) {
6146 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_X2APIC)) {
6147 			printf("VM-execution control \"virtualize x2APIC mode\" NOT supported.\n");
6148 			return CONFIG_TYPE_UNSUPPORTED;
6149 		}
6150 		cpu_exec_ctrl1 |= CPU_VIRT_X2APIC;
6151 	} else {
6152 		cpu_exec_ctrl1 &= ~CPU_VIRT_X2APIC;
6153 	}
6154 
6155 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6156 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6157 
6158 	if (vmentry_fails_early)
6159 		return CONFIG_TYPE_VMENTRY_FAILS_EARLY;
6160 
6161 	return CONFIG_TYPE_GOOD;
6162 }
6163 
6164 static bool cpu_has_apicv(void)
6165 {
6166 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
6167 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
6168 		(ctrl_pin_rev.clr & PIN_POST_INTR));
6169 }
6170 
6171 /* Validates APIC register access across valid virtualization configurations. */
6172 static void apic_reg_virt_test(void)
6173 {
6174 	u32 *apic_access_address;
6175 	u32 *virtual_apic_page;
6176 	u64 control;
6177 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6178 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6179 	int i;
6180 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
6181 
6182 	if (!cpu_has_apicv()) {
6183 		report_skip(__func__);
6184 		return;
6185 	}
6186 
6187 	control = cpu_exec_ctrl1;
6188 	control &= ~CPU_VINTD;
6189 	vmcs_write(CPU_EXEC_CTRL1, control);
6190 
6191 	test_set_guest(apic_reg_virt_guest);
6192 
6193 	/*
6194 	 * From the SDM: The 1-setting of the "virtualize APIC accesses"
6195 	 * VM-execution is guaranteed to apply only if translations to the
6196 	 * APIC-access address use a 4-KByte page.
6197 	 */
6198 	apic_access_address = alloc_page();
6199 	force_4k_page(apic_access_address);
6200 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_address));
6201 
6202 	virtual_apic_page = alloc_page();
6203 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6204 
6205 	for (i = 0; i < ARRAY_SIZE(apic_reg_tests); i++) {
6206 		struct apic_reg_test *apic_reg_test = &apic_reg_tests[i];
6207 		struct apic_reg_virt_config *apic_reg_virt_config =
6208 				&apic_reg_test->apic_reg_virt_config;
6209 		enum Config_type config_type;
6210 		u32 reg;
6211 
6212 		printf("--- %s test ---\n", apic_reg_test->name);
6213 		config_type =
6214 			configure_apic_reg_virt_test(apic_reg_virt_config);
6215 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6216 			printf("Skip because of missing features.\n");
6217 			continue;
6218 		}
6219 
6220 		if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6221 			enter_guest_with_bad_controls();
6222 			continue;
6223 		}
6224 
6225 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6226 			struct apic_reg_virt_expectation expectation = {};
6227 			bool ok;
6228 
6229 			ok = apic_reg_virt_exit_expectation(
6230 				reg, apic_reg_virt_config, &expectation);
6231 			if (!ok) {
6232 				report_fail("Malformed test.");
6233 				break;
6234 			}
6235 
6236 			test_xapic_rd(reg, &expectation, apic_access_address,
6237 				      virtual_apic_page);
6238 			test_xapic_wr(reg, &expectation, apic_access_address,
6239 				      virtual_apic_page);
6240 		}
6241 	}
6242 
6243 	/* Terminate the guest */
6244 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6245 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6246 	args->op = TERMINATE;
6247 	enter_guest();
6248 	assert_exit_reason(VMX_VMCALL);
6249 }
6250 
6251 struct virt_x2apic_mode_config {
6252 	struct apic_reg_virt_config apic_reg_virt_config;
6253 	bool virtual_interrupt_delivery;
6254 	bool use_msr_bitmaps;
6255 	bool disable_x2apic_msr_intercepts;
6256 	bool disable_x2apic;
6257 };
6258 
6259 struct virt_x2apic_mode_test_case {
6260 	const char *name;
6261 	struct virt_x2apic_mode_config virt_x2apic_mode_config;
6262 };
6263 
6264 enum Virt_x2apic_mode_behavior_type {
6265 	X2APIC_ACCESS_VIRTUALIZED,
6266 	X2APIC_ACCESS_PASSED_THROUGH,
6267 	X2APIC_ACCESS_TRIGGERS_GP,
6268 };
6269 
6270 struct virt_x2apic_mode_expectation {
6271 	enum Reason rd_exit_reason;
6272 	enum Reason wr_exit_reason;
6273 
6274 	/*
6275 	 * RDMSR and WRMSR handle 64-bit values. However, except for ICR, all of
6276 	 * the x2APIC registers are 32 bits. Notice:
6277 	 *   1. vmx_x2apic_read() clears the upper 32 bits for 32-bit registers.
6278 	 *   2. vmx_x2apic_write() expects the val arg to be well-formed.
6279 	 */
6280 	u64 rd_val;
6281 	u64 wr_val;
6282 
6283 	/*
6284 	 * Compares input to virtualized output;
6285 	 * 1st arg is pointer to return expected virtualization output.
6286 	 */
6287 	u64 (*virt_fn)(u64);
6288 
6289 	enum Virt_x2apic_mode_behavior_type rd_behavior;
6290 	enum Virt_x2apic_mode_behavior_type wr_behavior;
6291 	bool wr_only;
6292 };
6293 
6294 static u64 virt_x2apic_mode_identity(u64 val)
6295 {
6296 	return val;
6297 }
6298 
6299 static u64 virt_x2apic_mode_nibble1(u64 val)
6300 {
6301 	return val & 0xf0;
6302 }
6303 
6304 static void virt_x2apic_mode_rd_expectation(
6305 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6306 	bool apic_register_virtualization, bool virtual_interrupt_delivery,
6307 	struct virt_x2apic_mode_expectation *expectation)
6308 {
6309 	bool readable =
6310 		!x2apic_reg_reserved(reg) &&
6311 		reg != APIC_EOI;
6312 
6313 	expectation->rd_exit_reason = VMX_VMCALL;
6314 	expectation->virt_fn = virt_x2apic_mode_identity;
6315 	if (virt_x2apic_mode_on && apic_register_virtualization) {
6316 		expectation->rd_val = MAGIC_VAL_1;
6317 		if (reg == APIC_PROCPRI && virtual_interrupt_delivery)
6318 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6319 		else if (reg == APIC_TASKPRI)
6320 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6321 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6322 	} else if (virt_x2apic_mode_on && !apic_register_virtualization &&
6323 		   reg == APIC_TASKPRI) {
6324 		expectation->rd_val = MAGIC_VAL_1;
6325 		expectation->virt_fn = virt_x2apic_mode_nibble1;
6326 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6327 	} else if (!disable_x2apic && readable) {
6328 		expectation->rd_val = apic_read(reg);
6329 		expectation->rd_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6330 	} else {
6331 		expectation->rd_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6332 	}
6333 }
6334 
6335 /*
6336  * get_x2apic_wr_val() creates an innocuous write value for an x2APIC register.
6337  *
6338  * For writable registers, get_x2apic_wr_val() deposits the write value into the
6339  * val pointer arg and returns true. For non-writable registers, val is not
6340  * modified and get_x2apic_wr_val() returns false.
6341  */
6342 static bool get_x2apic_wr_val(u32 reg, u64 *val)
6343 {
6344 	switch (reg) {
6345 	case APIC_TASKPRI:
6346 		/* Bits 31:8 are reserved. */
6347 		*val &= 0xff;
6348 		break;
6349 	case APIC_EOI:
6350 	case APIC_ESR:
6351 	case APIC_TMICT:
6352 		/*
6353 		 * EOI, ESR: WRMSR of a non-zero value causes #GP(0).
6354 		 * TMICT: A write of 0 to the initial-count register effectively
6355 		 *        stops the local APIC timer, in both one-shot and
6356 		 *        periodic mode.
6357 		 */
6358 		*val = 0;
6359 		break;
6360 	case APIC_SPIV:
6361 	case APIC_LVTT:
6362 	case APIC_LVTTHMR:
6363 	case APIC_LVTPC:
6364 	case APIC_LVT0:
6365 	case APIC_LVT1:
6366 	case APIC_LVTERR:
6367 	case APIC_TDCR:
6368 		/*
6369 		 * To avoid writing a 1 to a reserved bit or causing some other
6370 		 * unintended side effect, read the current value and use it as
6371 		 * the write value.
6372 		 */
6373 		*val = apic_read(reg);
6374 		break;
6375 	case APIC_CMCI:
6376 		if (!apic_lvt_entry_supported(6))
6377 			return false;
6378 		*val = apic_read(reg);
6379 		break;
6380 	case APIC_ICR:
6381 		*val = 0x40000 | 0xf1;
6382 		break;
6383 	case APIC_SELF_IPI:
6384 		/*
6385 		 * With special processing (i.e., virtualize x2APIC mode +
6386 		 * virtual interrupt delivery), writing zero causes an
6387 		 * APIC-write VM exit. We plan to add a test for enabling
6388 		 * "virtual-interrupt delivery" in VMCS12, and that's where we
6389 		 * will test a self IPI with special processing.
6390 		 */
6391 		*val = 0x0;
6392 		break;
6393 	default:
6394 		return false;
6395 	}
6396 
6397 	return true;
6398 }
6399 
6400 static bool special_processing_applies(u32 reg, u64 *val,
6401 				       bool virt_int_delivery)
6402 {
6403 	bool special_processing =
6404 		(reg == APIC_TASKPRI) ||
6405 		(virt_int_delivery &&
6406 		 (reg == APIC_EOI || reg == APIC_SELF_IPI));
6407 
6408 	if (special_processing) {
6409 		TEST_ASSERT(get_x2apic_wr_val(reg, val));
6410 		return true;
6411 	}
6412 
6413 	return false;
6414 }
6415 
6416 static void virt_x2apic_mode_wr_expectation(
6417 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6418 	bool virt_int_delivery,
6419 	struct virt_x2apic_mode_expectation *expectation)
6420 {
6421 	expectation->wr_exit_reason = VMX_VMCALL;
6422 	expectation->wr_val = MAGIC_VAL_1;
6423 	expectation->wr_only = false;
6424 
6425 	if (virt_x2apic_mode_on &&
6426 	    special_processing_applies(reg, &expectation->wr_val,
6427 				       virt_int_delivery)) {
6428 		expectation->wr_behavior = X2APIC_ACCESS_VIRTUALIZED;
6429 		if (reg == APIC_SELF_IPI)
6430 			expectation->wr_exit_reason = VMX_APIC_WRITE;
6431 	} else if (!disable_x2apic &&
6432 		   get_x2apic_wr_val(reg, &expectation->wr_val)) {
6433 		expectation->wr_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6434 		if (reg == APIC_EOI || reg == APIC_SELF_IPI)
6435 			expectation->wr_only = true;
6436 		if (reg == APIC_ICR)
6437 			expectation->wr_exit_reason = VMX_EXTINT;
6438 	} else {
6439 		expectation->wr_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6440 		/*
6441 		 * Writing 1 to a reserved bit triggers a #GP.
6442 		 * Thus, set the write value to 0, which seems
6443 		 * the most likely to detect a missed #GP.
6444 		 */
6445 		expectation->wr_val = 0;
6446 	}
6447 }
6448 
6449 static void virt_x2apic_mode_exit_expectation(
6450 	u32 reg, struct virt_x2apic_mode_config *config,
6451 	struct virt_x2apic_mode_expectation *expectation)
6452 {
6453 	struct apic_reg_virt_config *base_config =
6454 		&config->apic_reg_virt_config;
6455 	bool virt_x2apic_mode_on =
6456 		base_config->virtualize_x2apic_mode &&
6457 		config->use_msr_bitmaps &&
6458 		config->disable_x2apic_msr_intercepts &&
6459 		base_config->activate_secondary_controls;
6460 
6461 	virt_x2apic_mode_wr_expectation(
6462 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6463 		config->virtual_interrupt_delivery, expectation);
6464 	virt_x2apic_mode_rd_expectation(
6465 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6466 		base_config->apic_register_virtualization,
6467 		config->virtual_interrupt_delivery, expectation);
6468 }
6469 
6470 struct virt_x2apic_mode_test_case virt_x2apic_mode_tests[] = {
6471 	/*
6472 	 * Baseline "virtualize x2APIC mode" configuration:
6473 	 *   - virtualize x2APIC mode
6474 	 *   - virtual-interrupt delivery
6475 	 *   - APIC-register virtualization
6476 	 *   - x2APIC MSR intercepts disabled
6477 	 *
6478 	 * Reads come from virtual APIC page, special processing applies to
6479 	 * VTPR, EOI, and SELF IPI, and all other writes pass through to L1
6480 	 * APIC.
6481 	 */
6482 	{
6483 		.name = "Baseline",
6484 		.virt_x2apic_mode_config = {
6485 			.virtual_interrupt_delivery = true,
6486 			.use_msr_bitmaps = true,
6487 			.disable_x2apic_msr_intercepts = true,
6488 			.disable_x2apic = false,
6489 			.apic_reg_virt_config = {
6490 				.apic_register_virtualization = true,
6491 				.use_tpr_shadow = true,
6492 				.virtualize_apic_accesses = false,
6493 				.virtualize_x2apic_mode = true,
6494 				.activate_secondary_controls = true,
6495 			},
6496 		},
6497 	},
6498 	{
6499 		.name = "Baseline w/ x2apic disabled",
6500 		.virt_x2apic_mode_config = {
6501 			.virtual_interrupt_delivery = true,
6502 			.use_msr_bitmaps = true,
6503 			.disable_x2apic_msr_intercepts = true,
6504 			.disable_x2apic = true,
6505 			.apic_reg_virt_config = {
6506 				.apic_register_virtualization = true,
6507 				.use_tpr_shadow = true,
6508 				.virtualize_apic_accesses = false,
6509 				.virtualize_x2apic_mode = true,
6510 				.activate_secondary_controls = true,
6511 			},
6512 		},
6513 	},
6514 
6515 	/*
6516 	 * Baseline, minus virtual-interrupt delivery. Reads come from virtual
6517 	 * APIC page, special processing applies to VTPR, and all other writes
6518 	 * pass through to L1 APIC.
6519 	 */
6520 	{
6521 		.name = "Baseline - virtual interrupt delivery",
6522 		.virt_x2apic_mode_config = {
6523 			.virtual_interrupt_delivery = false,
6524 			.use_msr_bitmaps = true,
6525 			.disable_x2apic_msr_intercepts = true,
6526 			.disable_x2apic = false,
6527 			.apic_reg_virt_config = {
6528 				.apic_register_virtualization = true,
6529 				.use_tpr_shadow = true,
6530 				.virtualize_apic_accesses = false,
6531 				.virtualize_x2apic_mode = true,
6532 				.activate_secondary_controls = true,
6533 			},
6534 		},
6535 	},
6536 
6537 	/*
6538 	 * Baseline, minus APIC-register virtualization. x2APIC reads pass
6539 	 * through to L1's APIC, unless reading VTPR
6540 	 */
6541 	{
6542 		.name = "Virtualize x2APIC mode, no APIC reg virt",
6543 		.virt_x2apic_mode_config = {
6544 			.virtual_interrupt_delivery = true,
6545 			.use_msr_bitmaps = true,
6546 			.disable_x2apic_msr_intercepts = true,
6547 			.disable_x2apic = false,
6548 			.apic_reg_virt_config = {
6549 				.apic_register_virtualization = false,
6550 				.use_tpr_shadow = true,
6551 				.virtualize_apic_accesses = false,
6552 				.virtualize_x2apic_mode = true,
6553 				.activate_secondary_controls = true,
6554 			},
6555 		},
6556 	},
6557 	{
6558 		.name = "Virtualize x2APIC mode, no APIC reg virt, x2APIC off",
6559 		.virt_x2apic_mode_config = {
6560 			.virtual_interrupt_delivery = true,
6561 			.use_msr_bitmaps = true,
6562 			.disable_x2apic_msr_intercepts = true,
6563 			.disable_x2apic = true,
6564 			.apic_reg_virt_config = {
6565 				.apic_register_virtualization = false,
6566 				.use_tpr_shadow = true,
6567 				.virtualize_apic_accesses = false,
6568 				.virtualize_x2apic_mode = true,
6569 				.activate_secondary_controls = true,
6570 			},
6571 		},
6572 	},
6573 
6574 	/*
6575 	 * Enable "virtualize x2APIC mode" and "APIC-register virtualization",
6576 	 * and disable intercepts for the x2APIC MSRs, but fail to enable
6577 	 * "activate secondary controls" (i.e. L2 gets access to L1's x2APIC
6578 	 * MSRs).
6579 	 */
6580 	{
6581 		.name = "Fail to enable activate secondary controls",
6582 		.virt_x2apic_mode_config = {
6583 			.virtual_interrupt_delivery = true,
6584 			.use_msr_bitmaps = true,
6585 			.disable_x2apic_msr_intercepts = true,
6586 			.disable_x2apic = false,
6587 			.apic_reg_virt_config = {
6588 				.apic_register_virtualization = true,
6589 				.use_tpr_shadow = true,
6590 				.virtualize_apic_accesses = false,
6591 				.virtualize_x2apic_mode = true,
6592 				.activate_secondary_controls = false,
6593 			},
6594 		},
6595 	},
6596 
6597 	/*
6598 	 * Enable "APIC-register virtualization" and enable "activate secondary
6599 	 * controls" and disable intercepts for the x2APIC MSRs, but do not
6600 	 * enable the "virtualize x2APIC mode" VM-execution control (i.e. L2
6601 	 * gets access to L1's x2APIC MSRs).
6602 	 */
6603 	{
6604 		.name = "Fail to enable virtualize x2APIC mode",
6605 		.virt_x2apic_mode_config = {
6606 			.virtual_interrupt_delivery = true,
6607 			.use_msr_bitmaps = true,
6608 			.disable_x2apic_msr_intercepts = true,
6609 			.disable_x2apic = false,
6610 			.apic_reg_virt_config = {
6611 				.apic_register_virtualization = true,
6612 				.use_tpr_shadow = true,
6613 				.virtualize_apic_accesses = false,
6614 				.virtualize_x2apic_mode = false,
6615 				.activate_secondary_controls = true,
6616 			},
6617 		},
6618 	},
6619 
6620 	/*
6621 	 * Disable "Virtualize x2APIC mode", disable x2APIC MSR intercepts, and
6622 	 * enable "APIC-register virtualization" --> L2 gets L1's x2APIC MSRs.
6623 	 */
6624 	{
6625 		.name = "Baseline",
6626 		.virt_x2apic_mode_config = {
6627 			.virtual_interrupt_delivery = true,
6628 			.use_msr_bitmaps = true,
6629 			.disable_x2apic_msr_intercepts = true,
6630 			.disable_x2apic = false,
6631 			.apic_reg_virt_config = {
6632 				.apic_register_virtualization = true,
6633 				.use_tpr_shadow = true,
6634 				.virtualize_apic_accesses = false,
6635 				.virtualize_x2apic_mode = false,
6636 				.activate_secondary_controls = true,
6637 			},
6638 		},
6639 	},
6640 };
6641 
6642 enum X2apic_op {
6643 	X2APIC_OP_RD,
6644 	X2APIC_OP_WR,
6645 	X2APIC_TERMINATE,
6646 };
6647 
6648 static u64 vmx_x2apic_read(u32 reg)
6649 {
6650 	u32 msr_addr = x2apic_msr(reg);
6651 	u64 val;
6652 
6653 	val = rdmsr(msr_addr);
6654 
6655 	return val;
6656 }
6657 
6658 static void vmx_x2apic_write(u32 reg, u64 val)
6659 {
6660 	u32 msr_addr = x2apic_msr(reg);
6661 
6662 	wrmsr(msr_addr, val);
6663 }
6664 
6665 struct virt_x2apic_mode_guest_args {
6666 	enum X2apic_op op;
6667 	u32 reg;
6668 	u64 val;
6669 	bool should_gp;
6670 	u64 (*virt_fn)(u64);
6671 } virt_x2apic_mode_guest_args;
6672 
6673 static volatile bool handle_x2apic_gp_ran;
6674 static volatile u32 handle_x2apic_gp_insn_len;
6675 static void handle_x2apic_gp(struct ex_regs *regs)
6676 {
6677 	handle_x2apic_gp_ran = true;
6678 	regs->rip += handle_x2apic_gp_insn_len;
6679 }
6680 
6681 static handler setup_x2apic_gp_handler(void)
6682 {
6683 	handler old_handler;
6684 
6685 	old_handler = handle_exception(GP_VECTOR, handle_x2apic_gp);
6686 	/* RDMSR and WRMSR are both 2 bytes, assuming no prefixes. */
6687 	handle_x2apic_gp_insn_len = 2;
6688 
6689 	return old_handler;
6690 }
6691 
6692 static void teardown_x2apic_gp_handler(handler old_handler)
6693 {
6694 	handle_exception(GP_VECTOR, old_handler);
6695 
6696 	/*
6697 	 * Defensively reset instruction length, so that if the handler is
6698 	 * incorrectly used, it will loop infinitely, rather than run off into
6699 	 * la la land.
6700 	 */
6701 	handle_x2apic_gp_insn_len = 0;
6702 	handle_x2apic_gp_ran = false;
6703 }
6704 
6705 static void virt_x2apic_mode_guest(void)
6706 {
6707 	volatile struct virt_x2apic_mode_guest_args *args =
6708 		&virt_x2apic_mode_guest_args;
6709 
6710 	for (;;) {
6711 		enum X2apic_op op = args->op;
6712 		u32 reg = args->reg;
6713 		u64 val = args->val;
6714 		bool should_gp = args->should_gp;
6715 		u64 (*virt_fn)(u64) = args->virt_fn;
6716 		handler old_handler;
6717 
6718 		if (op == X2APIC_TERMINATE)
6719 			break;
6720 
6721 		if (should_gp) {
6722 			TEST_ASSERT(!handle_x2apic_gp_ran);
6723 			old_handler = setup_x2apic_gp_handler();
6724 		}
6725 
6726 		if (op == X2APIC_OP_RD) {
6727 			u64 ret = vmx_x2apic_read(reg);
6728 
6729 			if (!should_gp) {
6730 				u64 want = virt_fn(val);
6731 				u64 got = virt_fn(ret);
6732 
6733 				report(got == want,
6734 				       "APIC read; got 0x%lx, want 0x%lx.",
6735 				       got, want);
6736 			}
6737 		} else if (op == X2APIC_OP_WR) {
6738 			vmx_x2apic_write(reg, val);
6739 		}
6740 
6741 		if (should_gp) {
6742 			report(handle_x2apic_gp_ran,
6743 			       "x2APIC op triggered GP.");
6744 			teardown_x2apic_gp_handler(old_handler);
6745 		}
6746 
6747 		/*
6748 		 * The L1 should always execute a vmcall after it's done testing
6749 		 * an individual APIC operation. This helps to validate that the
6750 		 * L1 and L2 are in sync with each other, as expected.
6751 		 */
6752 		vmcall();
6753 	}
6754 }
6755 
6756 static void test_x2apic_rd(
6757 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6758 	u32 *virtual_apic_page)
6759 {
6760 	u64 val = expectation->rd_val;
6761 	u32 exit_reason_want = expectation->rd_exit_reason;
6762 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6763 
6764 	report_prefix_pushf("x2apic - reading 0x%03x", reg);
6765 
6766 	/* Configure guest to do an x2apic read */
6767 	args->op = X2APIC_OP_RD;
6768 	args->reg = reg;
6769 	args->val = val;
6770 	args->should_gp = expectation->rd_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6771 	args->virt_fn = expectation->virt_fn;
6772 
6773 	/* Setup virtual APIC page */
6774 	if (expectation->rd_behavior == X2APIC_ACCESS_VIRTUALIZED)
6775 		virtual_apic_page[apic_reg_index(reg)] = (u32)val;
6776 
6777 	/* Enter guest */
6778 	enter_guest();
6779 
6780 	if (exit_reason_want != VMX_VMCALL) {
6781 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6782 	}
6783 
6784 	skip_exit_vmcall();
6785 	report_prefix_pop();
6786 }
6787 
6788 static volatile bool handle_x2apic_ipi_ran;
6789 static void handle_x2apic_ipi(isr_regs_t *regs)
6790 {
6791 	handle_x2apic_ipi_ran = true;
6792 	eoi();
6793 }
6794 
6795 static void test_x2apic_wr(
6796 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6797 	u32 *virtual_apic_page)
6798 {
6799 	u64 val = expectation->wr_val;
6800 	u32 exit_reason_want = expectation->wr_exit_reason;
6801 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6802 	int ipi_vector = 0xf1;
6803 	u32 restore_val = 0;
6804 
6805 	report_prefix_pushf("x2apic - writing 0x%lx to 0x%03x", val, reg);
6806 
6807 	/* Configure guest to do an x2apic read */
6808 	args->op = X2APIC_OP_WR;
6809 	args->reg = reg;
6810 	args->val = val;
6811 	args->should_gp = expectation->wr_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6812 
6813 	/* Setup virtual APIC page */
6814 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED)
6815 		virtual_apic_page[apic_reg_index(reg)] = 0;
6816 	if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH && !expectation->wr_only)
6817 		restore_val = apic_read(reg);
6818 
6819 	/* Setup IPI handler */
6820 	handle_x2apic_ipi_ran = false;
6821 	handle_irq(ipi_vector, handle_x2apic_ipi);
6822 
6823 	/* Enter guest */
6824 	enter_guest();
6825 
6826 	/*
6827 	 * Validate the behavior and
6828 	 * pass a magic value back to the guest.
6829 	 */
6830 	if (exit_reason_want == VMX_EXTINT) {
6831 		assert_exit_reason(exit_reason_want);
6832 
6833 		/* Clear the external interrupt. */
6834 		irq_enable();
6835 		asm volatile ("nop");
6836 		irq_disable();
6837 		report(handle_x2apic_ipi_ran,
6838 		       "Got pending interrupt after IRQ enabled.");
6839 
6840 		enter_guest();
6841 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6842 		assert_exit_reason(exit_reason_want);
6843 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6844 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%lx",
6845 		       apic_reg_index(reg),
6846 		       virtual_apic_page[apic_reg_index(reg)], val);
6847 
6848 		/* Reenter guest so it can consume/check rcx and exit again. */
6849 		enter_guest();
6850 	} else if (exit_reason_want != VMX_VMCALL) {
6851 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6852 	}
6853 
6854 	assert_exit_reason(VMX_VMCALL);
6855 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) {
6856 		u64 want = val;
6857 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6858 
6859 		report(got == want, "x2APIC write; got 0x%x, want 0x%lx", got,
6860 		       want);
6861 	} else if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH) {
6862 		if (!expectation->wr_only) {
6863 			u32 got = apic_read(reg);
6864 			bool ok;
6865 
6866 			/*
6867 			 * When L1's TPR is passed through to L2, the lower
6868 			 * nibble can be lost. For example, if L2 executes
6869 			 * WRMSR(0x808, 0x78), then, L1 might read 0x70.
6870 			 *
6871 			 * Here's how the lower nibble can get lost:
6872 			 *   1. L2 executes WRMSR(0x808, 0x78).
6873 			 *   2. L2 exits to L0 with a WRMSR exit.
6874 			 *   3. L0 emulates WRMSR, by writing L1's TPR.
6875 			 *   4. L0 re-enters L2.
6876 			 *   5. L2 exits to L0 (reason doesn't matter).
6877 			 *   6. L0 reflects L2's exit to L1.
6878 			 *   7. Before entering L1, L0 exits to user-space
6879 			 *      (e.g., to satisfy TPR access reporting).
6880 			 *   8. User-space executes KVM_SET_REGS ioctl, which
6881 			 *      clears the lower nibble of L1's TPR.
6882 			 */
6883 			if (reg == APIC_TASKPRI) {
6884 				got = apic_virt_nibble1(got);
6885 				val = apic_virt_nibble1(val);
6886 			}
6887 
6888 			ok = got == val;
6889 			report(ok,
6890 			       "non-virtualized write; val is 0x%x, want 0x%lx",
6891 			       got, val);
6892 			apic_write(reg, restore_val);
6893 		} else {
6894 			report_pass("non-virtualized and write-only OK");
6895 		}
6896 	}
6897 	skip_exit_insn();
6898 
6899 	report_prefix_pop();
6900 }
6901 
6902 static enum Config_type configure_virt_x2apic_mode_test(
6903 	struct virt_x2apic_mode_config *virt_x2apic_mode_config,
6904 	u8 *msr_bitmap_page)
6905 {
6906 	int msr;
6907 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6908 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6909 
6910 	/* x2apic-specific VMCS config */
6911 	if (virt_x2apic_mode_config->use_msr_bitmaps) {
6912 		/* virt_x2apic_mode_test() checks for MSR bitmaps support */
6913 		cpu_exec_ctrl0 |= CPU_MSR_BITMAP;
6914 	} else {
6915 		cpu_exec_ctrl0 &= ~CPU_MSR_BITMAP;
6916 	}
6917 
6918 	if (virt_x2apic_mode_config->virtual_interrupt_delivery) {
6919 		if (!(ctrl_cpu_rev[1].clr & CPU_VINTD)) {
6920 			report_skip("VM-execution control \"virtual-interrupt delivery\" NOT supported.\n");
6921 			return CONFIG_TYPE_UNSUPPORTED;
6922 		}
6923 		cpu_exec_ctrl1 |= CPU_VINTD;
6924 	} else {
6925 		cpu_exec_ctrl1 &= ~CPU_VINTD;
6926 	}
6927 
6928 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6929 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6930 
6931 	/* x2APIC MSR intercepts are usually off for "Virtualize x2APIC mode" */
6932 	for (msr = 0x800; msr <= 0x8ff; msr++) {
6933 		if (virt_x2apic_mode_config->disable_x2apic_msr_intercepts) {
6934 			clear_bit(msr, msr_bitmap_page + 0x000);
6935 			clear_bit(msr, msr_bitmap_page + 0x800);
6936 		} else {
6937 			set_bit(msr, msr_bitmap_page + 0x000);
6938 			set_bit(msr, msr_bitmap_page + 0x800);
6939 		}
6940 	}
6941 
6942 	/* x2APIC mode can impact virtualization */
6943 	reset_apic();
6944 	if (!virt_x2apic_mode_config->disable_x2apic)
6945 		enable_x2apic();
6946 
6947 	return configure_apic_reg_virt_test(
6948 		&virt_x2apic_mode_config->apic_reg_virt_config);
6949 }
6950 
6951 static void virt_x2apic_mode_test(void)
6952 {
6953 	u32 *virtual_apic_page;
6954 	u8 *msr_bitmap_page;
6955 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6956 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6957 	int i;
6958 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6959 
6960 	if (!cpu_has_apicv()) {
6961 		report_skip(__func__);
6962 		return;
6963 	}
6964 
6965 	/*
6966 	 * This is to exercise an issue in KVM's logic to merge L0's and L1's
6967 	 * MSR bitmaps. Previously, an L1 could get at L0's x2APIC MSRs by
6968 	 * writing the IA32_SPEC_CTRL MSR or the IA32_PRED_CMD MSRs. KVM would
6969 	 * then proceed to manipulate the MSR bitmaps, as if VMCS12 had the
6970 	 * "Virtualize x2APIC mod" control set, even when it didn't.
6971 	 */
6972 	if (has_spec_ctrl())
6973 		wrmsr(MSR_IA32_SPEC_CTRL, 1);
6974 
6975 	/*
6976 	 * Check that VMCS12 supports:
6977 	 *   - "Virtual-APIC address", indicated by "use TPR shadow"
6978 	 *   - "MSR-bitmap address", indicated by "use MSR bitmaps"
6979 	 */
6980 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6981 		report_skip("VM-execution control \"use TPR shadow\" NOT supported.\n");
6982 		return;
6983 	} else if (!(ctrl_cpu_rev[0].clr & CPU_MSR_BITMAP)) {
6984 		report_skip("VM-execution control \"use MSR bitmaps\" NOT supported.\n");
6985 		return;
6986 	}
6987 
6988 	test_set_guest(virt_x2apic_mode_guest);
6989 
6990 	virtual_apic_page = alloc_page();
6991 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6992 
6993 	msr_bitmap_page = alloc_page();
6994 	memset(msr_bitmap_page, 0xff, PAGE_SIZE);
6995 	vmcs_write(MSR_BITMAP, virt_to_phys(msr_bitmap_page));
6996 
6997 	for (i = 0; i < ARRAY_SIZE(virt_x2apic_mode_tests); i++) {
6998 		struct virt_x2apic_mode_test_case *virt_x2apic_mode_test_case =
6999 			&virt_x2apic_mode_tests[i];
7000 		struct virt_x2apic_mode_config *virt_x2apic_mode_config =
7001 			&virt_x2apic_mode_test_case->virt_x2apic_mode_config;
7002 		enum Config_type config_type;
7003 		u32 reg;
7004 
7005 		printf("--- %s test ---\n", virt_x2apic_mode_test_case->name);
7006 		config_type =
7007 			configure_virt_x2apic_mode_test(virt_x2apic_mode_config,
7008 							msr_bitmap_page);
7009 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
7010 			report_skip("Skip because of missing features.\n");
7011 			continue;
7012 		} else if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
7013 			enter_guest_with_bad_controls();
7014 			continue;
7015 		}
7016 
7017 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
7018 			struct virt_x2apic_mode_expectation expectation;
7019 
7020 			virt_x2apic_mode_exit_expectation(
7021 				reg, virt_x2apic_mode_config, &expectation);
7022 
7023 			test_x2apic_rd(reg, &expectation, virtual_apic_page);
7024 			test_x2apic_wr(reg, &expectation, virtual_apic_page);
7025 		}
7026 	}
7027 
7028 
7029 	/* Terminate the guest */
7030 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
7031 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
7032 	args->op = X2APIC_TERMINATE;
7033 	enter_guest();
7034 	assert_exit_reason(VMX_VMCALL);
7035 }
7036 
7037 static void test_ctl_reg(const char *cr_name, u64 cr, u64 fixed0, u64 fixed1)
7038 {
7039 	u64 val;
7040 	u64 cr_saved = vmcs_read(cr);
7041 	int i;
7042 
7043 	val = fixed0 & fixed1;
7044 	if (cr == HOST_CR4)
7045 		vmcs_write(cr, val | X86_CR4_PAE);
7046 	else
7047 		vmcs_write(cr, val);
7048 	report_prefix_pushf("%s %lx", cr_name, val);
7049 	if (val == fixed0)
7050 		test_vmx_vmlaunch(0);
7051 	else
7052 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7053 	report_prefix_pop();
7054 
7055 	for (i = 0; i < 64; i++) {
7056 
7057 		/* Set a bit when the corresponding bit in fixed1 is 0 */
7058 		if ((fixed1 & (1ull << i)) == 0) {
7059 			if (cr == HOST_CR4 && ((1ull << i) & X86_CR4_SMEP ||
7060 					       (1ull << i) & X86_CR4_SMAP))
7061 				continue;
7062 
7063 			vmcs_write(cr, cr_saved | (1ull << i));
7064 			report_prefix_pushf("%s %llx", cr_name,
7065 						cr_saved | (1ull << i));
7066 			test_vmx_vmlaunch(
7067 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7068 			report_prefix_pop();
7069 		}
7070 
7071 		/* Unset a bit when the corresponding bit in fixed0 is 1 */
7072 		if (fixed0 & (1ull << i)) {
7073 			vmcs_write(cr, cr_saved & ~(1ull << i));
7074 			report_prefix_pushf("%s %llx", cr_name,
7075 						cr_saved & ~(1ull << i));
7076 			test_vmx_vmlaunch(
7077 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7078 			report_prefix_pop();
7079 		}
7080 	}
7081 
7082 	vmcs_write(cr, cr_saved);
7083 }
7084 
7085 /*
7086  * 1. The CR0 field must not set any bit to a value not supported in VMX
7087  *    operation.
7088  * 2. The CR4 field must not set any bit to a value not supported in VMX
7089  *    operation.
7090  * 3. On processors that support Intel 64 architecture, the CR3 field must
7091  *    be such that bits 63:52 and bits in the range 51:32 beyond the
7092  *    processor's physical-address width must be 0.
7093  *
7094  *  [Intel SDM]
7095  */
7096 static void test_host_ctl_regs(void)
7097 {
7098 	u64 fixed0, fixed1, cr3, cr3_saved;
7099 	int i;
7100 
7101 	/* Test CR0 */
7102 	fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0);
7103 	fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1);
7104 	test_ctl_reg("HOST_CR0", HOST_CR0, fixed0, fixed1);
7105 
7106 	/* Test CR4 */
7107 	fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0);
7108 	fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1) &
7109 		 ~(X86_CR4_SMEP | X86_CR4_SMAP);
7110 	test_ctl_reg("HOST_CR4", HOST_CR4, fixed0, fixed1);
7111 
7112 	/* Test CR3 */
7113 	cr3_saved = vmcs_read(HOST_CR3);
7114 	for (i = cpuid_maxphyaddr(); i < 64; i++) {
7115 		cr3 = cr3_saved | (1ul << i);
7116 		vmcs_write(HOST_CR3, cr3);
7117 		report_prefix_pushf("HOST_CR3 %lx", cr3);
7118 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7119 		report_prefix_pop();
7120 	}
7121 
7122 	vmcs_write(HOST_CR3, cr3_saved);
7123 }
7124 
7125 static void test_efer_vmlaunch(u32 fld, bool ok)
7126 {
7127 	if (fld == HOST_EFER) {
7128 		if (ok)
7129 			test_vmx_vmlaunch(0);
7130 		else
7131 			test_vmx_vmlaunch2(VMXERR_ENTRY_INVALID_CONTROL_FIELD,
7132 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7133 	} else {
7134 		test_guest_state("EFER test", !ok, GUEST_EFER, "GUEST_EFER");
7135 	}
7136 }
7137 
7138 static void test_efer_one(u32 fld, const char * fld_name, u64 efer,
7139 			  u32 ctrl_fld, u64 ctrl,
7140 			  int i, const char *efer_bit_name)
7141 {
7142 	bool ok;
7143 
7144 	ok = true;
7145 	if (ctrl_fld == EXI_CONTROLS && (ctrl & EXI_LOAD_EFER)) {
7146 		if (!!(efer & EFER_LMA) != !!(ctrl & EXI_HOST_64))
7147 			ok = false;
7148 		if (!!(efer & EFER_LME) != !!(ctrl & EXI_HOST_64))
7149 			ok = false;
7150 	}
7151 	if (ctrl_fld == ENT_CONTROLS && (ctrl & ENT_LOAD_EFER)) {
7152 		/* Check LMA too since CR0.PG is set.  */
7153 		if (!!(efer & EFER_LMA) != !!(ctrl & ENT_GUEST_64))
7154 			ok = false;
7155 		if (!!(efer & EFER_LME) != !!(ctrl & ENT_GUEST_64))
7156 			ok = false;
7157 	}
7158 
7159 	/*
7160 	 * Skip the test if it would enter the guest in 32-bit mode.
7161 	 * Perhaps write the test in assembly and make sure it
7162 	 * can be run in either mode?
7163 	 */
7164 	if (fld == GUEST_EFER && ok && !(ctrl & ENT_GUEST_64))
7165 		return;
7166 
7167 	vmcs_write(ctrl_fld, ctrl);
7168 	vmcs_write(fld, efer);
7169 	report_prefix_pushf("%s %s bit turned %s, controls %s",
7170 			    fld_name, efer_bit_name,
7171 			    (i & 1) ? "on" : "off",
7172 			    (i & 2) ? "on" : "off");
7173 
7174 	test_efer_vmlaunch(fld, ok);
7175 	report_prefix_pop();
7176 }
7177 
7178 static void test_efer_bit(u32 fld, const char * fld_name,
7179 			  u32 ctrl_fld, u64 ctrl_bit, u64 efer_bit,
7180 			  const char *efer_bit_name)
7181 {
7182 	u64 efer_saved = vmcs_read(fld);
7183 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7184 	int i;
7185 
7186 	for (i = 0; i < 4; i++) {
7187 		u64 efer = efer_saved & ~efer_bit;
7188 		u64 ctrl = ctrl_saved & ~ctrl_bit;
7189 
7190 		if (i & 1)
7191 			efer |= efer_bit;
7192 		if (i & 2)
7193 			ctrl |= ctrl_bit;
7194 
7195 		test_efer_one(fld, fld_name, efer, ctrl_fld, ctrl,
7196 			      i, efer_bit_name);
7197 	}
7198 
7199 	vmcs_write(ctrl_fld, ctrl_saved);
7200 	vmcs_write(fld, efer_saved);
7201 }
7202 
7203 static void test_efer(u32 fld, const char * fld_name, u32 ctrl_fld,
7204 		      u64 ctrl_bit1, u64 ctrl_bit2)
7205 {
7206 	u64 efer_saved = vmcs_read(fld);
7207 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7208 	u64 efer_reserved_bits =  ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
7209 	u64 i;
7210 	u64 efer;
7211 
7212 	if (cpu_has_efer_nx())
7213 		efer_reserved_bits &= ~EFER_NX;
7214 
7215 	if (!ctrl_bit1) {
7216 		printf("\"Load-IA32-EFER\" exit control not supported\n");
7217 		goto test_entry_exit_mode;
7218 	}
7219 
7220 	report_prefix_pushf("%s %lx", fld_name, efer_saved);
7221 	test_efer_vmlaunch(fld, true);
7222 	report_prefix_pop();
7223 
7224 	/*
7225 	 * Check reserved bits
7226 	 */
7227 	vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit1);
7228 	for (i = 0; i < 64; i++) {
7229 		if ((1ull << i) & efer_reserved_bits) {
7230 			efer = efer_saved | (1ull << i);
7231 			vmcs_write(fld, efer);
7232 			report_prefix_pushf("%s %lx", fld_name, efer);
7233 			test_efer_vmlaunch(fld, true);
7234 			report_prefix_pop();
7235 		}
7236 	}
7237 
7238 	vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit1);
7239 	for (i = 0; i < 64; i++) {
7240 		if ((1ull << i) & efer_reserved_bits) {
7241 			efer = efer_saved | (1ull << i);
7242 			vmcs_write(fld, efer);
7243 			report_prefix_pushf("%s %lx", fld_name, efer);
7244 			test_efer_vmlaunch(fld, false);
7245 			report_prefix_pop();
7246 		}
7247 	}
7248 
7249 	vmcs_write(ctrl_fld, ctrl_saved);
7250 	vmcs_write(fld, efer_saved);
7251 
7252 	/*
7253 	 * Check LMA and LME bits
7254 	 */
7255 	test_efer_bit(fld, fld_name,
7256 		      ctrl_fld, ctrl_bit1,
7257 		      EFER_LMA,
7258 		      "EFER_LMA");
7259 	test_efer_bit(fld, fld_name,
7260 		      ctrl_fld, ctrl_bit1,
7261 		      EFER_LME,
7262 		      "EFER_LME");
7263 
7264 test_entry_exit_mode:
7265 	test_efer_bit(fld, fld_name,
7266 		      ctrl_fld, ctrl_bit2,
7267 		      EFER_LMA,
7268 		      "EFER_LMA");
7269 	test_efer_bit(fld, fld_name,
7270 		      ctrl_fld, ctrl_bit2,
7271 		      EFER_LME,
7272 		      "EFER_LME");
7273 }
7274 
7275 /*
7276  * If the 'load IA32_EFER' VM-exit control is 1, bits reserved in the
7277  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7278  * the values of the LMA and LME bits in the field must each be that of
7279  * the 'host address-space size' VM-exit control.
7280  *
7281  *  [Intel SDM]
7282  */
7283 static void test_host_efer(void)
7284 {
7285 	test_efer(HOST_EFER, "HOST_EFER", EXI_CONTROLS,
7286 		  ctrl_exit_rev.clr & EXI_LOAD_EFER,
7287 		  EXI_HOST_64);
7288 }
7289 
7290 /*
7291  * If the 'load IA32_EFER' VM-enter control is 1, bits reserved in the
7292  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7293  * the values of the LMA and LME bits in the field must each be that of
7294  * the 'IA32e-mode guest' VM-exit control.
7295  */
7296 static void test_guest_efer(void)
7297 {
7298 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER)) {
7299 		printf("\"Load-IA32-EFER\" entry control not supported\n");
7300 		return;
7301 	}
7302 
7303 	vmcs_write(GUEST_EFER, rdmsr(MSR_EFER));
7304 	test_efer(GUEST_EFER, "GUEST_EFER", ENT_CONTROLS,
7305 		  ctrl_enter_rev.clr & ENT_LOAD_EFER,
7306 		  ENT_GUEST_64);
7307 }
7308 
7309 /*
7310  * PAT values higher than 8 are uninteresting since they're likely lumped
7311  * in with "8". We only test values above 8 one bit at a time,
7312  * in order to reduce the number of VM-Entries and keep the runtime reasonable.
7313  */
7314 #define	PAT_VAL_LIMIT	8
7315 
7316 static void test_pat(u32 field, const char * field_name, u32 ctrl_field,
7317 		     u64 ctrl_bit)
7318 {
7319 	u32 ctrl_saved = vmcs_read(ctrl_field);
7320 	u64 pat_saved = vmcs_read(field);
7321 	u64 i, val;
7322 	u32 j;
7323 	int error;
7324 
7325 	vmcs_clear_bits(ctrl_field, ctrl_bit);
7326 
7327 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7328 		/* Test PAT0..PAT7 fields */
7329 		for (j = 0; j < (i ? 8 : 1); j++) {
7330 			val = i << j * 8;
7331 			vmcs_write(field, val);
7332 			if (field == HOST_PAT) {
7333 				report_prefix_pushf("%s %lx", field_name, val);
7334 				test_vmx_vmlaunch(0);
7335 				report_prefix_pop();
7336 
7337 			} else {	// GUEST_PAT
7338 				test_guest_state("ENT_LOAD_PAT enabled", false,
7339 						 val, "GUEST_PAT");
7340 			}
7341 		}
7342 	}
7343 
7344 	vmcs_set_bits(ctrl_field, ctrl_bit);
7345 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7346 		/* Test PAT0..PAT7 fields */
7347 		for (j = 0; j < (i ? 8 : 1); j++) {
7348 			val = i << j * 8;
7349 			vmcs_write(field, val);
7350 
7351 			if (field == HOST_PAT) {
7352 				report_prefix_pushf("%s %lx", field_name, val);
7353 				if (i == 0x2 || i == 0x3 || i >= 0x8)
7354 					error =
7355 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
7356 				else
7357 					error = 0;
7358 
7359 				test_vmx_vmlaunch(error);
7360 				report_prefix_pop();
7361 
7362 			} else {	// GUEST_PAT
7363 				error = (i == 0x2 || i == 0x3 || i >= 0x8);
7364 				test_guest_state("ENT_LOAD_PAT enabled", !!error,
7365 						 val, "GUEST_PAT");
7366 			}
7367 
7368 		}
7369 	}
7370 
7371 	vmcs_write(ctrl_field, ctrl_saved);
7372 	vmcs_write(field, pat_saved);
7373 }
7374 
7375 /*
7376  *  If the "load IA32_PAT" VM-exit control is 1, the value of the field
7377  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7378  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7379  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7380  *  6 (WB), or 7 (UC-).
7381  *
7382  *  [Intel SDM]
7383  */
7384 static void test_load_host_pat(void)
7385 {
7386 	/*
7387 	 * "load IA32_PAT" VM-exit control
7388 	 */
7389 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) {
7390 		printf("\"Load-IA32-PAT\" exit control not supported\n");
7391 		return;
7392 	}
7393 
7394 	test_pat(HOST_PAT, "HOST_PAT", EXI_CONTROLS, EXI_LOAD_PAT);
7395 }
7396 
7397 union cpuidA_eax {
7398 	struct {
7399 		unsigned int version_id:8;
7400 		unsigned int num_counters_gp:8;
7401 		unsigned int bit_width:8;
7402 		unsigned int mask_length:8;
7403 	} split;
7404 	unsigned int full;
7405 };
7406 
7407 union cpuidA_edx {
7408 	struct {
7409 		unsigned int num_counters_fixed:5;
7410 		unsigned int bit_width_fixed:8;
7411 		unsigned int reserved:9;
7412 	} split;
7413 	unsigned int full;
7414 };
7415 
7416 static bool valid_pgc(u64 val)
7417 {
7418 	struct cpuid id;
7419 	union cpuidA_eax eax;
7420 	union cpuidA_edx edx;
7421 	u64 mask;
7422 
7423 	id = cpuid(0xA);
7424 	eax.full = id.a;
7425 	edx.full = id.d;
7426 	mask = ~(((1ull << eax.split.num_counters_gp) - 1) |
7427 		 (((1ull << edx.split.num_counters_fixed) - 1) << 32));
7428 
7429 	return !(val & mask);
7430 }
7431 
7432 static void test_pgc_vmlaunch(u32 xerror, u32 xreason, bool xfail, bool host)
7433 {
7434 	u32 inst_err;
7435 	u64 obs;
7436 	bool success;
7437 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7438 
7439 	if (host) {
7440 		success = vmlaunch_succeeds();
7441 		obs = rdmsr(data->msr);
7442 		if (!success) {
7443 			inst_err = vmcs_read(VMX_INST_ERROR);
7444 			report(xerror == inst_err, "vmlaunch failed, "
7445 			       "VMX Inst Error is %d (expected %d)",
7446 			       inst_err, xerror);
7447 		} else {
7448 			report(!data->enabled || data->exp == obs,
7449 			       "Host state is 0x%lx (expected 0x%lx)",
7450 			       obs, data->exp);
7451 			report(success != xfail, "vmlaunch succeeded");
7452 		}
7453 	} else {
7454 		test_guest_state("load GUEST_PERF_GLOBAL_CTRL", xfail,
7455 				 GUEST_PERF_GLOBAL_CTRL,
7456 				 "GUEST_PERF_GLOBAL_CTRL");
7457 	}
7458 }
7459 
7460 /*
7461  * test_load_perf_global_ctrl is a generic function for testing the
7462  * "load IA32_PERF_GLOBAL_CTRL" VM-{Entry,Exit} controls. This test function
7463  * tests the provided ctrl_val when disabled and enabled.
7464  *
7465  * @nr: VMCS field number corresponding to the host/guest state field
7466  * @name: Name of the above VMCS field for printing in test report
7467  * @ctrl_nr: VMCS field number corresponding to the VM-{Entry,Exit} control
7468  * @ctrl_val: Bit to set on the ctrl_field
7469  */
7470 static void test_perf_global_ctrl(u32 nr, const char *name, u32 ctrl_nr,
7471 				  const char *ctrl_name, u64 ctrl_val)
7472 {
7473 	u64 ctrl_saved = vmcs_read(ctrl_nr);
7474 	u64 pgc_saved = vmcs_read(nr);
7475 	u64 i, val;
7476 	bool host = nr == HOST_PERF_GLOBAL_CTRL;
7477 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7478 
7479 	data->msr = MSR_CORE_PERF_GLOBAL_CTRL;
7480 	msr_bmp_init();
7481 	vmcs_write(ctrl_nr, ctrl_saved & ~ctrl_val);
7482 	data->enabled = false;
7483 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=0 on %s",
7484 			    ctrl_name);
7485 
7486 	for (i = 0; i < 64; i++) {
7487 		val = 1ull << i;
7488 		vmcs_write(nr, val);
7489 		report_prefix_pushf("%s = 0x%lx", name, val);
7490 		test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7491 		report_prefix_pop();
7492 	}
7493 	report_prefix_pop();
7494 
7495 	vmcs_write(ctrl_nr, ctrl_saved | ctrl_val);
7496 	data->enabled = true;
7497 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=1 on %s",
7498 			    ctrl_name);
7499 	for (i = 0; i < 64; i++) {
7500 		val = 1ull << i;
7501 		data->exp = val;
7502 		vmcs_write(nr, val);
7503 		report_prefix_pushf("%s = 0x%lx", name, val);
7504 		if (valid_pgc(val)) {
7505 			test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7506 		} else {
7507 			if (host)
7508 				test_pgc_vmlaunch(
7509 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
7510 					0,
7511 					true,
7512 					host);
7513 			else
7514 				test_pgc_vmlaunch(
7515 					0,
7516 					VMX_ENTRY_FAILURE | VMX_FAIL_STATE,
7517 					true,
7518 					host);
7519 		}
7520 		report_prefix_pop();
7521 	}
7522 
7523 	data->enabled = false;
7524 	report_prefix_pop();
7525 	vmcs_write(ctrl_nr, ctrl_saved);
7526 	vmcs_write(nr, pgc_saved);
7527 }
7528 
7529 static void test_load_host_perf_global_ctrl(void)
7530 {
7531 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PERF)) {
7532 		printf("\"load IA32_PERF_GLOBAL_CTRL\" exit control not supported\n");
7533 		return;
7534 	}
7535 
7536 	test_perf_global_ctrl(HOST_PERF_GLOBAL_CTRL, "HOST_PERF_GLOBAL_CTRL",
7537 				   EXI_CONTROLS, "EXI_CONTROLS", EXI_LOAD_PERF);
7538 }
7539 
7540 
7541 static void test_load_guest_perf_global_ctrl(void)
7542 {
7543 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PERF)) {
7544 		printf("\"load IA32_PERF_GLOBAL_CTRL\" entry control not supported\n");
7545 		return;
7546 	}
7547 
7548 	test_perf_global_ctrl(GUEST_PERF_GLOBAL_CTRL, "GUEST_PERF_GLOBAL_CTRL",
7549 				   ENT_CONTROLS, "ENT_CONTROLS", ENT_LOAD_PERF);
7550 }
7551 
7552 
7553 /*
7554  * test_vmcs_field - test a value for the given VMCS field
7555  * @field: VMCS field
7556  * @field_name: string name of VMCS field
7557  * @bit_start: starting bit
7558  * @bit_end: ending bit
7559  * @val: value that the bit range must or must not contain
7560  * @valid_val: whether value given in 'val' must be valid or not
7561  * @error: expected VMCS error when vmentry fails for an invalid value
7562  */
7563 static void test_vmcs_field(u64 field, const char *field_name, u32 bit_start,
7564 			    u32 bit_end, u64 val, bool valid_val, u32 error)
7565 {
7566 	u64 field_saved = vmcs_read(field);
7567 	u32 i;
7568 	u64 tmp;
7569 	u32 bit_on;
7570 	u64 mask = ~0ull;
7571 
7572 	mask = (mask >> bit_end) << bit_end;
7573 	mask = mask | ((1 << bit_start) - 1);
7574 	tmp = (field_saved & mask) | (val << bit_start);
7575 
7576 	vmcs_write(field, tmp);
7577 	report_prefix_pushf("%s %lx", field_name, tmp);
7578 	if (valid_val)
7579 		test_vmx_vmlaunch(0);
7580 	else
7581 		test_vmx_vmlaunch(error);
7582 	report_prefix_pop();
7583 
7584 	for (i = bit_start; i <= bit_end; i = i + 2) {
7585 		bit_on = ((1ull < i) & (val << bit_start)) ? 0 : 1;
7586 		if (bit_on)
7587 			tmp = field_saved | (1ull << i);
7588 		else
7589 			tmp = field_saved & ~(1ull << i);
7590 		vmcs_write(field, tmp);
7591 		report_prefix_pushf("%s %lx", field_name, tmp);
7592 		if (valid_val)
7593 			test_vmx_vmlaunch(error);
7594 		else
7595 			test_vmx_vmlaunch(0);
7596 		report_prefix_pop();
7597 	}
7598 
7599 	vmcs_write(field, field_saved);
7600 }
7601 
7602 static void test_canonical(u64 field, const char * field_name, bool host)
7603 {
7604 	u64 addr_saved = vmcs_read(field);
7605 
7606 	/*
7607 	 * Use the existing value if possible.  Writing a random canonical
7608 	 * value is not an option as doing so would corrupt the field being
7609 	 * tested and likely hose the test.
7610 	 */
7611 	if (is_canonical(addr_saved)) {
7612 		if (host) {
7613 			report_prefix_pushf("%s %lx", field_name, addr_saved);
7614 			test_vmx_vmlaunch(0);
7615 			report_prefix_pop();
7616 		} else {
7617 			test_guest_state("Test canonical address", false,
7618 					 addr_saved, field_name);
7619 		}
7620 	}
7621 
7622 	vmcs_write(field, NONCANONICAL);
7623 
7624 	if (host) {
7625 		report_prefix_pushf("%s %llx", field_name, NONCANONICAL);
7626 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7627 		report_prefix_pop();
7628 	} else {
7629 		test_guest_state("Test non-canonical address", true,
7630 				 NONCANONICAL, field_name);
7631 	}
7632 
7633 	vmcs_write(field, addr_saved);
7634 }
7635 
7636 #define TEST_RPL_TI_FLAGS(reg, name)				\
7637 	test_vmcs_field(reg, name, 0, 2, 0x0, true,		\
7638 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7639 
7640 #define TEST_CS_TR_FLAGS(reg, name)				\
7641 	test_vmcs_field(reg, name, 3, 15, 0x0000, false,	\
7642 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7643 
7644 /*
7645  * 1. In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the
7646  *    RPL (bits 1:0) and the TI flag (bit 2) must be 0.
7647  * 2. The selector fields for CS and TR cannot be 0000H.
7648  * 3. The selector field for SS cannot be 0000H if the "host address-space
7649  *    size" VM-exit control is 0.
7650  * 4. On processors that support Intel 64 architecture, the base-address
7651  *    fields for FS, GS and TR must contain canonical addresses.
7652  */
7653 static void test_host_segment_regs(void)
7654 {
7655 	u16 selector_saved;
7656 
7657 	/*
7658 	 * Test RPL and TI flags
7659 	 */
7660 	TEST_RPL_TI_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7661 	TEST_RPL_TI_FLAGS(HOST_SEL_SS, "HOST_SEL_SS");
7662 	TEST_RPL_TI_FLAGS(HOST_SEL_DS, "HOST_SEL_DS");
7663 	TEST_RPL_TI_FLAGS(HOST_SEL_ES, "HOST_SEL_ES");
7664 	TEST_RPL_TI_FLAGS(HOST_SEL_FS, "HOST_SEL_FS");
7665 	TEST_RPL_TI_FLAGS(HOST_SEL_GS, "HOST_SEL_GS");
7666 	TEST_RPL_TI_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7667 
7668 	/*
7669 	 * Test that CS and TR fields can not be 0x0000
7670 	 */
7671 	TEST_CS_TR_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7672 	TEST_CS_TR_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7673 
7674 	/*
7675 	 * SS field can not be 0x0000 if "host address-space size" VM-exit
7676 	 * control is 0
7677 	 */
7678 	selector_saved = vmcs_read(HOST_SEL_SS);
7679 	vmcs_write(HOST_SEL_SS, 0);
7680 	report_prefix_pushf("HOST_SEL_SS 0");
7681 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7682 		test_vmx_vmlaunch(0);
7683 	} else {
7684 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7685 	}
7686 	report_prefix_pop();
7687 
7688 	vmcs_write(HOST_SEL_SS, selector_saved);
7689 
7690 	/*
7691 	 * Base address for FS, GS and TR must be canonical
7692 	 */
7693 	test_canonical(HOST_BASE_FS, "HOST_BASE_FS", true);
7694 	test_canonical(HOST_BASE_GS, "HOST_BASE_GS", true);
7695 	test_canonical(HOST_BASE_TR, "HOST_BASE_TR", true);
7696 }
7697 
7698 /*
7699  *  On processors that support Intel 64 architecture, the base-address
7700  *  fields for GDTR and IDTR must contain canonical addresses.
7701  */
7702 static void test_host_desc_tables(void)
7703 {
7704 	test_canonical(HOST_BASE_GDTR, "HOST_BASE_GDTR", true);
7705 	test_canonical(HOST_BASE_IDTR, "HOST_BASE_IDTR", true);
7706 }
7707 
7708 /*
7709  * If the "host address-space size" VM-exit control is 0, the following must
7710  * hold:
7711  *    - The "IA-32e mode guest" VM-entry control is 0.
7712  *    - Bit 17 of the CR4 field (corresponding to CR4.PCIDE) is 0.
7713  *    - Bits 63:32 in the RIP field are 0.
7714  *
7715  * If the "host address-space size" VM-exit control is 1, the following must
7716  * hold:
7717  *    - Bit 5 of the CR4 field (corresponding to CR4.PAE) is 1.
7718  *    - The RIP field contains a canonical address.
7719  *
7720  */
7721 static void test_host_addr_size(void)
7722 {
7723 	u64 cr4_saved = vmcs_read(HOST_CR4);
7724 	u64 rip_saved = vmcs_read(HOST_RIP);
7725 	u64 entry_ctrl_saved = vmcs_read(ENT_CONTROLS);
7726 	int i;
7727 	u64 tmp;
7728 
7729 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7730 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7731 		report_prefix_pushf("\"IA-32e mode guest\" enabled");
7732 		test_vmx_vmlaunch(0);
7733 		report_prefix_pop();
7734 
7735 		vmcs_write(HOST_CR4, cr4_saved | X86_CR4_PCIDE);
7736 		report_prefix_pushf("\"CR4.PCIDE\" set");
7737 		test_vmx_vmlaunch(0);
7738 		report_prefix_pop();
7739 
7740 		for (i = 32; i <= 63; i = i + 4) {
7741 			tmp = rip_saved | 1ull << i;
7742 			vmcs_write(HOST_RIP, tmp);
7743 			report_prefix_pushf("HOST_RIP %lx", tmp);
7744 			test_vmx_vmlaunch(0);
7745 			report_prefix_pop();
7746 		}
7747 
7748 		if (cr4_saved & X86_CR4_PAE) {
7749 			vmcs_write(HOST_CR4, cr4_saved  & ~X86_CR4_PAE);
7750 			report_prefix_pushf("\"CR4.PAE\" unset");
7751 			test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7752 		} else {
7753 			report_prefix_pushf("\"CR4.PAE\" set");
7754 			test_vmx_vmlaunch(0);
7755 		}
7756 		report_prefix_pop();
7757 
7758 		vmcs_write(HOST_RIP, NONCANONICAL);
7759 		report_prefix_pushf("HOST_RIP %llx", NONCANONICAL);
7760 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7761 		report_prefix_pop();
7762 
7763 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7764 		vmcs_write(HOST_RIP, rip_saved);
7765 		vmcs_write(HOST_CR4, cr4_saved);
7766 
7767 		/* Restore host's active RIP and CR4 values. */
7768 		report_prefix_pushf("restore host state");
7769 		test_vmx_vmlaunch(0);
7770 		report_prefix_pop();
7771 	}
7772 }
7773 
7774 /*
7775  * Check that the virtual CPU checks the VMX Host State Area as
7776  * documented in the Intel SDM.
7777  */
7778 static void vmx_host_state_area_test(void)
7779 {
7780 	/*
7781 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
7782 	 * fail due to invalid guest state, should we make it that
7783 	 * far.
7784 	 */
7785 	vmcs_write(GUEST_RFLAGS, 0);
7786 
7787 	test_host_ctl_regs();
7788 
7789 	test_canonical(HOST_SYSENTER_ESP, "HOST_SYSENTER_ESP", true);
7790 	test_canonical(HOST_SYSENTER_EIP, "HOST_SYSENTER_EIP", true);
7791 
7792 	test_host_efer();
7793 	test_load_host_pat();
7794 	test_host_segment_regs();
7795 	test_host_desc_tables();
7796 	test_host_addr_size();
7797 	test_load_host_perf_global_ctrl();
7798 }
7799 
7800 /*
7801  * If the "load debug controls" VM-entry control is 1, bits 63:32 in
7802  * the DR7 field must be 0.
7803  *
7804  * [Intel SDM]
7805  */
7806 static void test_guest_dr7(void)
7807 {
7808 	u32 ent_saved = vmcs_read(ENT_CONTROLS);
7809 	u64 dr7_saved = vmcs_read(GUEST_DR7);
7810 	u64 val;
7811 	int i;
7812 
7813 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS) {
7814 		vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7815 		for (i = 0; i < 64; i++) {
7816 			val = 1ull << i;
7817 			vmcs_write(GUEST_DR7, val);
7818 			test_guest_state("ENT_LOAD_DBGCTLS disabled", false,
7819 					 val, "GUEST_DR7");
7820 		}
7821 	}
7822 	if (ctrl_enter_rev.clr & ENT_LOAD_DBGCTLS) {
7823 		vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7824 		for (i = 0; i < 64; i++) {
7825 			val = 1ull << i;
7826 			vmcs_write(GUEST_DR7, val);
7827 			test_guest_state("ENT_LOAD_DBGCTLS enabled", i >= 32,
7828 					 val, "GUEST_DR7");
7829 		}
7830 	}
7831 	vmcs_write(GUEST_DR7, dr7_saved);
7832 	vmcs_write(ENT_CONTROLS, ent_saved);
7833 }
7834 
7835 /*
7836  *  If the "load IA32_PAT" VM-entry control is 1, the value of the field
7837  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7838  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7839  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7840  *  6 (WB), or 7 (UC-).
7841  *
7842  *  [Intel SDM]
7843  */
7844 static void test_load_guest_pat(void)
7845 {
7846 	/*
7847 	 * "load IA32_PAT" VM-entry control
7848 	 */
7849 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
7850 		printf("\"Load-IA32-PAT\" entry control not supported\n");
7851 		return;
7852 	}
7853 
7854 	test_pat(GUEST_PAT, "GUEST_PAT", ENT_CONTROLS, ENT_LOAD_PAT);
7855 }
7856 
7857 #define MSR_IA32_BNDCFGS_RSVD_MASK	0x00000ffc
7858 
7859 /*
7860  * If the "load IA32_BNDCFGS" VM-entry control is 1, the following
7861  * checks are performed on the field for the IA32_BNDCFGS MSR:
7862  *
7863  *   - Bits reserved in the IA32_BNDCFGS MSR must be 0.
7864  *   - The linear address in bits 63:12 must be canonical.
7865  *
7866  *  [Intel SDM]
7867  */
7868 static void test_load_guest_bndcfgs(void)
7869 {
7870 	u64 bndcfgs_saved = vmcs_read(GUEST_BNDCFGS);
7871 	u64 bndcfgs;
7872 
7873 	if (!(ctrl_enter_rev.clr & ENT_LOAD_BNDCFGS)) {
7874 		printf("\"Load-IA32-BNDCFGS\" entry control not supported\n");
7875 		return;
7876 	}
7877 
7878 	vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7879 
7880 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7881 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7882 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7883 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7884 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7885 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7886 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7887 
7888 	vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7889 
7890 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7891 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7892 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7893 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7894 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7895 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7896 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7897 
7898 	vmcs_write(GUEST_BNDCFGS, bndcfgs_saved);
7899 }
7900 
7901 #define	GUEST_SEG_UNUSABLE_MASK	(1u << 16)
7902 #define	GUEST_SEG_SEL_TI_MASK	(1u << 2)
7903 
7904 
7905 #define	TEST_SEGMENT_SEL(test, xfail, sel, val)				\
7906 do {									\
7907 	vmcs_write(sel, val);						\
7908 	test_guest_state(test " segment", xfail, val, xstr(sel));	\
7909 } while (0)
7910 
7911 #define	TEST_INVALID_SEG_SEL(sel, val) \
7912 	TEST_SEGMENT_SEL("Invalid: " xstr(val), true, sel, val);
7913 
7914 #define	TEST_VALID_SEG_SEL(sel, val) \
7915 	TEST_SEGMENT_SEL("Valid: " xstr(val), false, sel, val);
7916 
7917 /*
7918  * The following checks are done on the Selector field of the Guest Segment
7919  * Registers:
7920  *    - TR. The TI flag (bit 2) must be 0.
7921  *    - LDTR. If LDTR is usable, the TI flag (bit 2) must be 0.
7922  *    - SS. If the guest will not be virtual-8086 and the "unrestricted
7923  *	guest" VM-execution control is 0, the RPL (bits 1:0) must equal
7924  *	the RPL of the selector field for CS.
7925  *
7926  *  [Intel SDM]
7927  */
7928 static void test_guest_segment_sel_fields(void)
7929 {
7930 	u16 sel_saved;
7931 	u32 ar_saved;
7932 	u32 cpu_ctrl0_saved;
7933 	u32 cpu_ctrl1_saved;
7934 	u16 cs_rpl_bits;
7935 
7936 	/*
7937 	 * Test for GUEST_SEL_TR
7938 	 */
7939 	sel_saved = vmcs_read(GUEST_SEL_TR);
7940 	TEST_INVALID_SEG_SEL(GUEST_SEL_TR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7941 	vmcs_write(GUEST_SEL_TR, sel_saved);
7942 
7943 	/*
7944 	 * Test for GUEST_SEL_LDTR
7945 	 */
7946 	sel_saved = vmcs_read(GUEST_SEL_LDTR);
7947 	ar_saved = vmcs_read(GUEST_AR_LDTR);
7948 	/* LDTR is set unusable */
7949 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7950 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7951 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7952 	/* LDTR is set usable */
7953 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7954 	TEST_INVALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7955 
7956 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7957 
7958 	vmcs_write(GUEST_AR_LDTR, ar_saved);
7959 	vmcs_write(GUEST_SEL_LDTR, sel_saved);
7960 
7961 	/*
7962 	 * Test for GUEST_SEL_SS
7963 	 */
7964 	cpu_ctrl0_saved = vmcs_read(CPU_EXEC_CTRL0);
7965 	cpu_ctrl1_saved = vmcs_read(CPU_EXEC_CTRL1);
7966 	ar_saved = vmcs_read(GUEST_AR_SS);
7967 	/* Turn off "unrestricted guest" vm-execution control */
7968 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved & ~CPU_URG);
7969 	cs_rpl_bits = vmcs_read(GUEST_SEL_CS) & 0x3;
7970 	sel_saved = vmcs_read(GUEST_SEL_SS);
7971 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7972 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7973 	/* Make SS usable if it's unusable or vice-versa */
7974 	if (ar_saved & GUEST_SEG_UNUSABLE_MASK)
7975 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7976 	else
7977 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7978 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7979 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7980 
7981 	/* Need a valid EPTP as the passing case fully enters the guest. */
7982 	if (enable_unrestricted_guest(true))
7983 		goto skip_ss_tests;
7984 
7985 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7986 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7987 
7988 	/* Make SS usable if it's unusable or vice-versa */
7989 	if (vmcs_read(GUEST_AR_SS) & GUEST_SEG_UNUSABLE_MASK)
7990 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7991 	else
7992 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7993 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7994 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7995 skip_ss_tests:
7996 
7997 	vmcs_write(GUEST_AR_SS, ar_saved);
7998 	vmcs_write(GUEST_SEL_SS, sel_saved);
7999 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrl0_saved);
8000 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved);
8001 }
8002 
8003 #define	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(xfail, seg_base)			\
8004 do {										\
8005 	addr_saved = vmcs_read(seg_base);					\
8006 	for (i = 32; i < 63; i = i + 4) {					\
8007 		addr = addr_saved | 1ull << i;					\
8008 		vmcs_write(seg_base, addr);					\
8009 		test_guest_state("seg.BASE[63:32] != 0, usable = " xstr(xfail),	\
8010 				 xfail, addr, xstr(seg_base));			\
8011 	}									\
8012 	vmcs_write(seg_base, addr_saved);					\
8013 } while (0)
8014 
8015 #define	TEST_SEGMENT_BASE_ADDR_CANONICAL(xfail, seg_base)		  \
8016 do {									  \
8017 	addr_saved = vmcs_read(seg_base);				  \
8018 	vmcs_write(seg_base, NONCANONICAL);				  \
8019 	test_guest_state("seg.BASE non-canonical, usable = " xstr(xfail), \
8020 			 xfail, NONCANONICAL, xstr(seg_base));		  \
8021 	vmcs_write(seg_base, addr_saved);				  \
8022 } while (0)
8023 
8024 /*
8025  * The following checks are done on the Base Address field of the Guest
8026  * Segment Registers on processors that support Intel 64 architecture:
8027  *    - TR, FS, GS : The address must be canonical.
8028  *    - LDTR : If LDTR is usable, the address must be canonical.
8029  *    - CS : Bits 63:32 of the address must be zero.
8030  *    - SS, DS, ES : If the register is usable, bits 63:32 of the address
8031  *	must be zero.
8032  *
8033  *  [Intel SDM]
8034  */
8035 static void test_guest_segment_base_addr_fields(void)
8036 {
8037 	u64 addr_saved;
8038 	u64 addr;
8039 	u32 ar_saved;
8040 	int i;
8041 
8042 	/*
8043 	 * The address of TR, FS, GS and LDTR must be canonical.
8044 	 */
8045 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_TR);
8046 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_FS);
8047 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_GS);
8048 	ar_saved = vmcs_read(GUEST_AR_LDTR);
8049 	/* Make LDTR unusable */
8050 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8051 	TEST_SEGMENT_BASE_ADDR_CANONICAL(false, GUEST_BASE_LDTR);
8052 	/* Make LDTR usable */
8053 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8054 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_LDTR);
8055 
8056 	vmcs_write(GUEST_AR_LDTR, ar_saved);
8057 
8058 	/*
8059 	 * Bits 63:32 in CS, SS, DS and ES base address must be zero
8060 	 */
8061 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_CS);
8062 	ar_saved = vmcs_read(GUEST_AR_SS);
8063 	/* Make SS unusable */
8064 	vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8065 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_SS);
8066 	/* Make SS usable */
8067 	vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8068 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_SS);
8069 	vmcs_write(GUEST_AR_SS, ar_saved);
8070 
8071 	ar_saved = vmcs_read(GUEST_AR_DS);
8072 	/* Make DS unusable */
8073 	vmcs_write(GUEST_AR_DS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8074 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_DS);
8075 	/* Make DS usable */
8076 	vmcs_write(GUEST_AR_DS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8077 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_DS);
8078 	vmcs_write(GUEST_AR_DS, ar_saved);
8079 
8080 	ar_saved = vmcs_read(GUEST_AR_ES);
8081 	/* Make ES unusable */
8082 	vmcs_write(GUEST_AR_ES, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8083 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_ES);
8084 	/* Make ES usable */
8085 	vmcs_write(GUEST_AR_ES, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8086 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_ES);
8087 	vmcs_write(GUEST_AR_ES, ar_saved);
8088 }
8089 
8090 /*
8091  * Check that the virtual CPU checks the VMX Guest State Area as
8092  * documented in the Intel SDM.
8093  */
8094 static void vmx_guest_state_area_test(void)
8095 {
8096 	vmx_set_test_stage(1);
8097 	test_set_guest(guest_state_test_main);
8098 
8099 	/*
8100 	 * The IA32_SYSENTER_ESP field and the IA32_SYSENTER_EIP field
8101 	 * must each contain a canonical address.
8102 	 */
8103 	test_canonical(GUEST_SYSENTER_ESP, "GUEST_SYSENTER_ESP", false);
8104 	test_canonical(GUEST_SYSENTER_EIP, "GUEST_SYSENTER_EIP", false);
8105 
8106 	test_guest_dr7();
8107 	test_load_guest_pat();
8108 	test_guest_efer();
8109 	test_load_guest_perf_global_ctrl();
8110 	test_load_guest_bndcfgs();
8111 
8112 	test_guest_segment_sel_fields();
8113 	test_guest_segment_base_addr_fields();
8114 
8115 	test_canonical(GUEST_BASE_GDTR, "GUEST_BASE_GDTR", false);
8116 	test_canonical(GUEST_BASE_IDTR, "GUEST_BASE_IDTR", false);
8117 
8118 	u32 guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_GDTR);
8119 	int i;
8120 	for (i = 16; i <= 31; i++) {
8121 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8122 		vmcs_write(GUEST_LIMIT_GDTR, tmp);
8123 		test_guest_state("GDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_GDTR");
8124 	}
8125 	vmcs_write(GUEST_LIMIT_GDTR, guest_desc_limit_saved);
8126 
8127 	guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_IDTR);
8128 	for (i = 16; i <= 31; i++) {
8129 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8130 		vmcs_write(GUEST_LIMIT_IDTR, tmp);
8131 		test_guest_state("IDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_IDTR");
8132 	}
8133 	vmcs_write(GUEST_LIMIT_IDTR, guest_desc_limit_saved);
8134 
8135 	/*
8136 	 * Let the guest finish execution
8137 	 */
8138 	vmx_set_test_stage(2);
8139 	enter_guest();
8140 }
8141 
8142 extern void unrestricted_guest_main(void);
8143 asm (".code32\n"
8144 	"unrestricted_guest_main:\n"
8145 	"vmcall\n"
8146 	"nop\n"
8147 	"mov $1, %edi\n"
8148 	"call hypercall\n"
8149 	".code64\n");
8150 
8151 static void setup_unrestricted_guest(void)
8152 {
8153 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) & ~(X86_CR0_PG));
8154 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) & ~ENT_GUEST_64);
8155 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) & ~EFER_LMA);
8156 	vmcs_write(GUEST_RIP, virt_to_phys(unrestricted_guest_main));
8157 }
8158 
8159 static void unsetup_unrestricted_guest(void)
8160 {
8161 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) | X86_CR0_PG);
8162 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_GUEST_64);
8163 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) | EFER_LMA);
8164 	vmcs_write(GUEST_RIP, (u64) phys_to_virt(vmcs_read(GUEST_RIP)));
8165 	vmcs_write(GUEST_RSP, (u64) phys_to_virt(vmcs_read(GUEST_RSP)));
8166 }
8167 
8168 /*
8169  * If "unrestricted guest" secondary VM-execution control is set, guests
8170  * can run in unpaged protected mode.
8171  */
8172 static void vmentry_unrestricted_guest_test(void)
8173 {
8174 	if (enable_unrestricted_guest(true)) {
8175 		report_skip("Unrestricted guest not supported");
8176 		return;
8177 	}
8178 
8179 	test_set_guest(unrestricted_guest_main);
8180 	setup_unrestricted_guest();
8181 	test_guest_state("Unrestricted guest test", false, CPU_URG, "CPU_URG");
8182 
8183 	/*
8184 	 * Let the guest finish execution as a regular guest
8185 	 */
8186 	unsetup_unrestricted_guest();
8187 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) & ~CPU_URG);
8188 	enter_guest();
8189 }
8190 
8191 static bool valid_vmcs_for_vmentry(void)
8192 {
8193 	struct vmcs *current_vmcs = NULL;
8194 
8195 	if (vmcs_save(&current_vmcs))
8196 		return false;
8197 
8198 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
8199 }
8200 
8201 static void try_vmentry_in_movss_shadow(void)
8202 {
8203 	u32 vm_inst_err;
8204 	u32 flags;
8205 	bool early_failure = false;
8206 	u32 expected_flags = X86_EFLAGS_FIXED;
8207 	bool valid_vmcs = valid_vmcs_for_vmentry();
8208 
8209 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
8210 
8211 	/*
8212 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
8213 	 * unsupported VMCS component").
8214 	 */
8215 	vmcs_write(~0u, 0);
8216 
8217 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
8218 			      "vmwrite %%rsp, %%rdx;"
8219 			      "mov 0f, %%rax;"
8220 			      "mov %[host_rip], %%edx;"
8221 			      "vmwrite %%rax, %%rdx;"
8222 			      "mov $-1, %%ah;"
8223 			      "sahf;"
8224 			      "mov %%ss, %%ax;"
8225 			      "mov %%ax, %%ss;"
8226 			      "vmlaunch;"
8227 			      "mov $1, %[early_failure];"
8228 			      "0: lahf;"
8229 			      "movzbl %%ah, %[flags]"
8230 			      : [early_failure] "+r" (early_failure),
8231 				[flags] "=&a" (flags)
8232 			      : [host_rsp] "i" (HOST_RSP),
8233 				[host_rip] "i" (HOST_RIP)
8234 			      : "rdx", "cc", "memory");
8235 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
8236 
8237 	report(early_failure, "Early VM-entry failure");
8238 	report(flags == expected_flags, "RFLAGS[8:0] is %x (actual %x)",
8239 	       expected_flags, flags);
8240 	if (valid_vmcs)
8241 		report(vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
8242 		       "VM-instruction error is %d (actual %d)",
8243 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
8244 }
8245 
8246 static void vmentry_movss_shadow_test(void)
8247 {
8248 	struct vmcs *orig_vmcs;
8249 
8250 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
8251 
8252 	/*
8253 	 * Set the launched flag on the current VMCS to verify the correct
8254 	 * error priority, below.
8255 	 */
8256 	test_set_guest(v2_null_test_guest);
8257 	enter_guest();
8258 
8259 	/*
8260 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
8261 	 * fail due to invalid guest state (if we make it that far).
8262 	 */
8263 	vmcs_write(GUEST_RFLAGS, 0);
8264 
8265 	/*
8266 	 * "VM entry with events blocked by MOV SS" takes precedence over
8267 	 * "VMLAUNCH with non-clear VMCS."
8268 	 */
8269 	report_prefix_push("valid current-VMCS");
8270 	try_vmentry_in_movss_shadow();
8271 	report_prefix_pop();
8272 
8273 	/*
8274 	 * VMfailInvalid takes precedence over "VM entry with events
8275 	 * blocked by MOV SS."
8276 	 */
8277 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
8278 	report_prefix_push("no current-VMCS");
8279 	try_vmentry_in_movss_shadow();
8280 	report_prefix_pop();
8281 
8282 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
8283 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8284 }
8285 
8286 static void vmx_ldtr_test_guest(void)
8287 {
8288 	u16 ldtr = sldt();
8289 
8290 	report(ldtr == NP_SEL, "Expected %x for L2 LDTR selector (got %x)",
8291 	       NP_SEL, ldtr);
8292 }
8293 
8294 /*
8295  * Ensure that the L1 LDTR is set to 0 on VM-exit.
8296  */
8297 static void vmx_ldtr_test(void)
8298 {
8299 	const u8 ldt_ar = 0x82; /* Present LDT */
8300 	u16 sel = FIRST_SPARE_SEL;
8301 
8302 	/* Set up a non-zero L1 LDTR prior to VM-entry. */
8303 	set_gdt_entry(sel, 0, 0, ldt_ar, 0);
8304 	lldt(sel);
8305 
8306 	test_set_guest(vmx_ldtr_test_guest);
8307 	/*
8308 	 * Set up a different LDTR for L2. The actual GDT contents are
8309 	 * irrelevant, since we stuff the hidden descriptor state
8310 	 * straight into the VMCS rather than reading it from the GDT.
8311 	 */
8312 	vmcs_write(GUEST_SEL_LDTR, NP_SEL);
8313 	vmcs_write(GUEST_AR_LDTR, ldt_ar);
8314 	enter_guest();
8315 
8316 	/*
8317 	 * VM-exit should clear LDTR (and make it unusable, but we
8318 	 * won't verify that here).
8319 	 */
8320 	sel = sldt();
8321 	report(!sel, "Expected 0 for L1 LDTR selector (got %x)", sel);
8322 }
8323 
8324 static void vmx_single_vmcall_guest(void)
8325 {
8326 	vmcall();
8327 }
8328 
8329 static void vmx_cr_load_test(void)
8330 {
8331 	unsigned long cr3, cr4, orig_cr3, orig_cr4;
8332 	u32 ctrls[2] = {0};
8333 	pgd_t *pml5;
8334 
8335 	orig_cr4 = read_cr4();
8336 	orig_cr3 = read_cr3();
8337 
8338 	if (!this_cpu_has(X86_FEATURE_PCID)) {
8339 		report_skip("PCID not detected");
8340 		return;
8341 	}
8342 	if (!this_cpu_has(X86_FEATURE_MCE)) {
8343 		report_skip("MCE not detected");
8344 		return;
8345 	}
8346 
8347 	TEST_ASSERT(!(orig_cr3 & X86_CR3_PCID_MASK));
8348 
8349 	/* Enable PCID for L1. */
8350 	cr4 = orig_cr4 | X86_CR4_PCIDE;
8351 	cr3 = orig_cr3 | 0x1;
8352 	TEST_ASSERT(!write_cr4_checking(cr4));
8353 	write_cr3(cr3);
8354 
8355 	test_set_guest(vmx_single_vmcall_guest);
8356 	vmcs_write(HOST_CR4, cr4);
8357 	vmcs_write(HOST_CR3, cr3);
8358 	enter_guest();
8359 
8360 	/*
8361 	 * No exception is expected.
8362 	 *
8363 	 * NB. KVM loads the last guest write to CR4 into CR4 read
8364 	 *     shadow. In order to trigger an exit to KVM, we can toggle a
8365 	 *     bit that is owned by KVM. We use CR4.MCE, which shall
8366 	 *     have no side effect because normally no guest MCE (e.g., as the
8367 	 *     result of bad memory) would happen during this test.
8368 	 */
8369 	TEST_ASSERT(!write_cr4_checking(cr4 ^ X86_CR4_MCE));
8370 
8371 	/* Cleanup L1 state. */
8372 	write_cr3(orig_cr3);
8373 	TEST_ASSERT(!write_cr4_checking(orig_cr4));
8374 
8375 	if (!this_cpu_has(X86_FEATURE_LA57))
8376 		goto done;
8377 
8378 	/*
8379 	 * Allocate a full page for PML5 to guarantee alignment, though only
8380 	 * the first entry needs to be filled (the test's virtual addresses
8381 	 * most definitely do not have any of bits 56:48 set).
8382 	 */
8383 	pml5 = alloc_page();
8384 	*pml5 = orig_cr3 | PT_PRESENT_MASK | PT_WRITABLE_MASK;
8385 
8386 	/*
8387 	 * Transition to/from 5-level paging in the host via VM-Exit.  CR4.LA57
8388 	 * can't be toggled while long is active via MOV CR4, but there are no
8389 	 * such restrictions on VM-Exit.
8390 	 */
8391 lol_5level:
8392 	vmcs_write(HOST_CR4, orig_cr4 | X86_CR4_LA57);
8393 	vmcs_write(HOST_CR3, virt_to_phys(pml5));
8394 	enter_guest();
8395 
8396 	/*
8397 	 * VMREAD with a memory operand to verify KVM detects the LA57 change,
8398 	 * e.g. uses the correct guest root level in gva_to_gpa().
8399 	 */
8400 	TEST_ASSERT(vmcs_readm(HOST_CR3) == virt_to_phys(pml5));
8401 	TEST_ASSERT(vmcs_readm(HOST_CR4) == (orig_cr4 | X86_CR4_LA57));
8402 
8403 	vmcs_write(HOST_CR4, orig_cr4);
8404 	vmcs_write(HOST_CR3, orig_cr3);
8405 	enter_guest();
8406 
8407 	TEST_ASSERT(vmcs_readm(HOST_CR3) == orig_cr3);
8408 	TEST_ASSERT(vmcs_readm(HOST_CR4) == orig_cr4);
8409 
8410 	/*
8411 	 * And now do the same LA57 shenanigans with EPT enabled.  KVM uses
8412 	 * two separate MMUs when L1 uses TDP, whereas the above shadow paging
8413 	 * version shares an MMU between L1 and L2.
8414 	 *
8415 	 * If the saved execution controls are non-zero then the EPT version
8416 	 * has already run.  In that case, restore the old controls.  If EPT
8417 	 * setup fails, e.g. EPT isn't supported, fall through and finish up.
8418 	 */
8419 	if (ctrls[0]) {
8420 		vmcs_write(CPU_EXEC_CTRL0, ctrls[0]);
8421 		vmcs_write(CPU_EXEC_CTRL1, ctrls[1]);
8422 	} else if (!setup_ept(false)) {
8423 		ctrls[0] = vmcs_read(CPU_EXEC_CTRL0);
8424 		ctrls[1]  = vmcs_read(CPU_EXEC_CTRL1);
8425 		goto lol_5level;
8426 	}
8427 
8428 	free_page(pml5);
8429 
8430 done:
8431 	skip_exit_vmcall();
8432 	enter_guest();
8433 }
8434 
8435 static void vmx_cr4_osxsave_test_guest(void)
8436 {
8437 	write_cr4(read_cr4() & ~X86_CR4_OSXSAVE);
8438 }
8439 
8440 /*
8441  * Ensure that kvm recalculates the L1 guest's CPUID.01H:ECX.OSXSAVE
8442  * after VM-exit from an L2 guest that sets CR4.OSXSAVE to a different
8443  * value than in L1.
8444  */
8445 static void vmx_cr4_osxsave_test(void)
8446 {
8447 	if (!this_cpu_has(X86_FEATURE_XSAVE)) {
8448 		report_skip("XSAVE not detected");
8449 		return;
8450 	}
8451 
8452 	if (!(read_cr4() & X86_CR4_OSXSAVE)) {
8453 		unsigned long cr4 = read_cr4() | X86_CR4_OSXSAVE;
8454 
8455 		write_cr4(cr4);
8456 		vmcs_write(GUEST_CR4, cr4);
8457 		vmcs_write(HOST_CR4, cr4);
8458 	}
8459 
8460 	TEST_ASSERT(cpuid_osxsave());
8461 
8462 	test_set_guest(vmx_cr4_osxsave_test_guest);
8463 	enter_guest();
8464 
8465 	TEST_ASSERT(cpuid_osxsave());
8466 }
8467 
8468 static void vmx_nm_test_guest(void)
8469 {
8470 	write_cr0(read_cr0() | X86_CR0_TS);
8471 	asm volatile("fnop");
8472 }
8473 
8474 static void check_nm_exit(const char *test)
8475 {
8476 	u32 reason = vmcs_read(EXI_REASON);
8477 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
8478 	const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8479 		NM_VECTOR;
8480 
8481 	report(reason == VMX_EXC_NMI && intr_info == expected, "%s", test);
8482 }
8483 
8484 /*
8485  * This test checks that:
8486  *
8487  * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then
8488  *     a subsequent #NM VM-exit is reflected to L1.
8489  *
8490  * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a
8491  *     subsequent #NM VM-exit is reflected to L1.
8492  */
8493 static void vmx_nm_test(void)
8494 {
8495 	unsigned long cr0 = read_cr0();
8496 
8497 	test_set_guest(vmx_nm_test_guest);
8498 
8499 	/*
8500 	 * L1 wants to intercept #NM exceptions encountered in L2.
8501 	 */
8502 	vmcs_write(EXC_BITMAP, 1 << NM_VECTOR);
8503 
8504 	/*
8505 	 * Launch L2 with CR0.TS clear, but don't claim host ownership of
8506 	 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop,
8507 	 * which will raise #NM. L0 should reflect the #NM VM-exit to L1.
8508 	 */
8509 	vmcs_write(CR0_MASK, 0);
8510 	vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS);
8511 	enter_guest();
8512 	check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1");
8513 
8514 	/*
8515 	 * Re-enter L2 at the fnop instruction, with CR0.TS clear but
8516 	 * CR0.EM set. The fnop will still raise #NM, and L0 should
8517 	 * reflect the #NM VM-exit to L1.
8518 	 */
8519 	vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM);
8520 	enter_guest();
8521 	check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1");
8522 
8523 	/*
8524 	 * Re-enter L2 at the fnop instruction, with both CR0.TS and
8525 	 * CR0.EM clear. There will be no #NM, and the L2 guest should
8526 	 * exit normally.
8527 	 */
8528 	vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM));
8529 	enter_guest();
8530 }
8531 
8532 bool vmx_pending_event_ipi_fired;
8533 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
8534 {
8535 	vmx_pending_event_ipi_fired = true;
8536 	eoi();
8537 }
8538 
8539 bool vmx_pending_event_guest_run;
8540 static void vmx_pending_event_guest(void)
8541 {
8542 	vmcall();
8543 	vmx_pending_event_guest_run = true;
8544 }
8545 
8546 static void vmx_pending_event_test_core(bool guest_hlt)
8547 {
8548 	int ipi_vector = 0xf1;
8549 
8550 	vmx_pending_event_ipi_fired = false;
8551 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
8552 
8553 	vmx_pending_event_guest_run = false;
8554 	test_set_guest(vmx_pending_event_guest);
8555 
8556 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
8557 
8558 	enter_guest();
8559 	skip_exit_vmcall();
8560 
8561 	if (guest_hlt)
8562 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8563 
8564 	irq_disable();
8565 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
8566 				   APIC_DM_FIXED | ipi_vector,
8567 				   0);
8568 
8569 	enter_guest();
8570 
8571 	assert_exit_reason(VMX_EXTINT);
8572 	report(!vmx_pending_event_guest_run,
8573 	       "Guest did not run before host received IPI");
8574 
8575 	irq_enable();
8576 	asm volatile ("nop");
8577 	irq_disable();
8578 	report(vmx_pending_event_ipi_fired,
8579 	       "Got pending interrupt after IRQ enabled");
8580 
8581 	if (guest_hlt)
8582 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8583 
8584 	enter_guest();
8585 	report(vmx_pending_event_guest_run,
8586 	       "Guest finished running when no interrupt");
8587 }
8588 
8589 static void vmx_pending_event_test(void)
8590 {
8591 	vmx_pending_event_test_core(false);
8592 }
8593 
8594 static void vmx_pending_event_hlt_test(void)
8595 {
8596 	vmx_pending_event_test_core(true);
8597 }
8598 
8599 static int vmx_window_test_db_count;
8600 
8601 static void vmx_window_test_db_handler(struct ex_regs *regs)
8602 {
8603 	vmx_window_test_db_count++;
8604 }
8605 
8606 static void vmx_nmi_window_test_guest(void)
8607 {
8608 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8609 
8610 	asm volatile("vmcall\n\t"
8611 		     "nop\n\t");
8612 
8613 	handle_exception(DB_VECTOR, NULL);
8614 }
8615 
8616 static void verify_nmi_window_exit(u64 rip)
8617 {
8618 	u32 exit_reason = vmcs_read(EXI_REASON);
8619 
8620 	report(exit_reason == VMX_NMI_WINDOW,
8621 	       "Exit reason (%d) is 'NMI window'", exit_reason);
8622 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8623 	       vmcs_read(GUEST_RIP), rip);
8624 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8625 }
8626 
8627 static void vmx_nmi_window_test(void)
8628 {
8629 	u64 nop_addr;
8630 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8631 
8632 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
8633 		report_skip("CPU does not support the \"Virtual NMIs\" VM-execution control.");
8634 		return;
8635 	}
8636 
8637 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
8638 		report_skip("CPU does not support the \"NMI-window exiting\" VM-execution control.");
8639 		return;
8640 	}
8641 
8642 	vmx_window_test_db_count = 0;
8643 
8644 	report_prefix_push("NMI-window");
8645 	test_set_guest(vmx_nmi_window_test_guest);
8646 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
8647 	enter_guest();
8648 	skip_exit_vmcall();
8649 	nop_addr = vmcs_read(GUEST_RIP);
8650 
8651 	/*
8652 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
8653 	 * RIP will not advance.
8654 	 */
8655 	report_prefix_push("active, no blocking");
8656 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8657 	enter_guest();
8658 	verify_nmi_window_exit(nop_addr);
8659 	report_prefix_pop();
8660 
8661 	/*
8662 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
8663 	 * a VM-exit on the next instruction after the nop. (The nop
8664 	 * is one byte.)
8665 	 */
8666 	report_prefix_push("active, blocking by MOV-SS");
8667 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8668 	enter_guest();
8669 	verify_nmi_window_exit(nop_addr + 1);
8670 	report_prefix_pop();
8671 
8672 	/*
8673 	 * Ask for "NMI-window exiting" (with event injection), and
8674 	 * expect a VM-exit after the event is injected. (RIP should
8675 	 * be at the address specified in the IDT entry for #DB.)
8676 	 */
8677 	report_prefix_push("active, no blocking, injecting #DB");
8678 	vmcs_write(ENT_INTR_INFO,
8679 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8680 	enter_guest();
8681 	verify_nmi_window_exit((u64)db_fault_addr);
8682 	report_prefix_pop();
8683 
8684 	/*
8685 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
8686 	 * a VM-exit after the next IRET (i.e. after the #DB handler
8687 	 * returns). So, RIP should be back at one byte past the nop.
8688 	 */
8689 	report_prefix_push("active, blocking by NMI");
8690 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
8691 	enter_guest();
8692 	verify_nmi_window_exit(nop_addr + 1);
8693 	report(vmx_window_test_db_count == 1,
8694 	       "#DB handler executed once (actual %d times)",
8695 	       vmx_window_test_db_count);
8696 	report_prefix_pop();
8697 
8698 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8699 		report_skip("CPU does not support activity state HLT.");
8700 	} else {
8701 		/*
8702 		 * Ask for "NMI-window exiting" when entering activity
8703 		 * state HLT, and expect an immediate VM-exit. RIP is
8704 		 * still one byte past the nop.
8705 		 */
8706 		report_prefix_push("halted, no blocking");
8707 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8708 		enter_guest();
8709 		verify_nmi_window_exit(nop_addr + 1);
8710 		report_prefix_pop();
8711 
8712 		/*
8713 		 * Ask for "NMI-window exiting" when entering activity
8714 		 * state HLT (with event injection), and expect a
8715 		 * VM-exit after the event is injected. (RIP should be
8716 		 * at the address specified in the IDT entry for #DB.)
8717 		 */
8718 		report_prefix_push("halted, no blocking, injecting #DB");
8719 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8720 		vmcs_write(ENT_INTR_INFO,
8721 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8722 			   DB_VECTOR);
8723 		enter_guest();
8724 		verify_nmi_window_exit((u64)db_fault_addr);
8725 		report_prefix_pop();
8726 	}
8727 
8728 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8729 	enter_guest();
8730 	report_prefix_pop();
8731 }
8732 
8733 static void vmx_intr_window_test_guest(void)
8734 {
8735 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8736 
8737 	/*
8738 	 * The two consecutive STIs are to ensure that only the first
8739 	 * one has a shadow. Note that NOP and STI are one byte
8740 	 * instructions.
8741 	 */
8742 	asm volatile("vmcall\n\t"
8743 		     "nop\n\t"
8744 		     "sti\n\t"
8745 		     "sti\n\t");
8746 
8747 	handle_exception(DB_VECTOR, NULL);
8748 }
8749 
8750 static void verify_intr_window_exit(u64 rip)
8751 {
8752 	u32 exit_reason = vmcs_read(EXI_REASON);
8753 
8754 	report(exit_reason == VMX_INTR_WINDOW,
8755 	       "Exit reason (%d) is 'interrupt window'", exit_reason);
8756 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8757 	       vmcs_read(GUEST_RIP), rip);
8758 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8759 }
8760 
8761 static void vmx_intr_window_test(void)
8762 {
8763 	u64 vmcall_addr;
8764 	u64 nop_addr;
8765 	unsigned int orig_db_gate_type;
8766 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8767 
8768 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
8769 		report_skip("CPU does not support the \"interrupt-window exiting\" VM-execution control.");
8770 		return;
8771 	}
8772 
8773 	/*
8774 	 * Change the IDT entry for #DB from interrupt gate to trap gate,
8775 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
8776 	 * be disabled after vectoring a #DB.
8777 	 */
8778 	orig_db_gate_type = boot_idt[DB_VECTOR].type;
8779 	boot_idt[DB_VECTOR].type = 15;
8780 
8781 	report_prefix_push("interrupt-window");
8782 	test_set_guest(vmx_intr_window_test_guest);
8783 	enter_guest();
8784 	assert_exit_reason(VMX_VMCALL);
8785 	vmcall_addr = vmcs_read(GUEST_RIP);
8786 
8787 	/*
8788 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
8789 	 * no blocking; expect an immediate VM-exit. Note that we have
8790 	 * not advanced past the vmcall instruction yet, so RIP should
8791 	 * point to the vmcall instruction.
8792 	 */
8793 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
8794 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8795 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
8796 	enter_guest();
8797 	verify_intr_window_exit(vmcall_addr);
8798 	report_prefix_pop();
8799 
8800 	/*
8801 	 * Ask for "interrupt-window exiting" (with event injection)
8802 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
8803 	 * the event is injected. That is, RIP should should be at the
8804 	 * address specified in the IDT entry for #DB.
8805 	 */
8806 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #DB");
8807 	vmcs_write(ENT_INTR_INFO,
8808 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8809 	vmcall_addr = vmcs_read(GUEST_RIP);
8810 	enter_guest();
8811 	verify_intr_window_exit((u64)db_fault_addr);
8812 	report_prefix_pop();
8813 
8814 	/*
8815 	 * Let the L2 guest run through the IRET, back to the VMCALL.
8816 	 * We have to clear the "interrupt-window exiting"
8817 	 * VM-execution control, or it would just keep causing
8818 	 * VM-exits. Then, advance past the VMCALL and set the
8819 	 * "interrupt-window exiting" VM-execution control again.
8820 	 */
8821 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8822 	enter_guest();
8823 	skip_exit_vmcall();
8824 	nop_addr = vmcs_read(GUEST_RIP);
8825 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8826 
8827 	/*
8828 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
8829 	 * RFLAGS.IF set, and expect a VM-exit on the next
8830 	 * instruction. (NOP is one byte.)
8831 	 */
8832 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
8833 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8834 	enter_guest();
8835 	verify_intr_window_exit(nop_addr + 1);
8836 	report_prefix_pop();
8837 
8838 	/*
8839 	 * Back up to the NOP and ask for "interrupt-window exiting"
8840 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
8841 	 * on the next instruction. (NOP is one byte.)
8842 	 */
8843 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
8844 	vmcs_write(GUEST_RIP, nop_addr);
8845 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
8846 	enter_guest();
8847 	verify_intr_window_exit(nop_addr + 1);
8848 	report_prefix_pop();
8849 
8850 	/*
8851 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
8852 	 * and expect a VM-exit on the instruction following the STI
8853 	 * shadow. Only the first STI (which is one byte past the NOP)
8854 	 * should have a shadow. The second STI (which is two bytes
8855 	 * past the NOP) has no shadow. Therefore, the interrupt
8856 	 * window opens at three bytes past the NOP.
8857 	 */
8858 	report_prefix_push("active, RFLAGS.IF = 0");
8859 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8860 	enter_guest();
8861 	verify_intr_window_exit(nop_addr + 3);
8862 	report_prefix_pop();
8863 
8864 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8865 		report_skip("CPU does not support activity state HLT.");
8866 	} else {
8867 		/*
8868 		 * Ask for "interrupt-window exiting" when entering
8869 		 * activity state HLT, and expect an immediate
8870 		 * VM-exit. RIP is still three bytes past the nop.
8871 		 */
8872 		report_prefix_push("halted, no blocking");
8873 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8874 		enter_guest();
8875 		verify_intr_window_exit(nop_addr + 3);
8876 		report_prefix_pop();
8877 
8878 		/*
8879 		 * Ask for "interrupt-window exiting" when entering
8880 		 * activity state HLT (with event injection), and
8881 		 * expect a VM-exit after the event is injected. That
8882 		 * is, RIP should should be at the address specified
8883 		 * in the IDT entry for #DB.
8884 		 */
8885 		report_prefix_push("halted, no blocking, injecting #DB");
8886 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8887 		vmcs_write(ENT_INTR_INFO,
8888 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8889 			   DB_VECTOR);
8890 		enter_guest();
8891 		verify_intr_window_exit((u64)db_fault_addr);
8892 		report_prefix_pop();
8893 	}
8894 
8895 	boot_idt[DB_VECTOR].type = orig_db_gate_type;
8896 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8897 	enter_guest();
8898 	report_prefix_pop();
8899 }
8900 
8901 #define GUEST_TSC_OFFSET (1u << 30)
8902 
8903 static u64 guest_tsc;
8904 
8905 static void vmx_store_tsc_test_guest(void)
8906 {
8907 	guest_tsc = rdtsc();
8908 }
8909 
8910 /*
8911  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
8912  * list, the value saved is not subject to the TSC offset that is
8913  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
8914  */
8915 static void vmx_store_tsc_test(void)
8916 {
8917 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
8918 	u64 low, high;
8919 
8920 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
8921 		report_skip("'Use TSC offsetting' not supported");
8922 		return;
8923 	}
8924 
8925 	test_set_guest(vmx_store_tsc_test_guest);
8926 
8927 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
8928 	vmcs_write(EXI_MSR_ST_CNT, 1);
8929 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
8930 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
8931 
8932 	low = rdtsc();
8933 	enter_guest();
8934 	high = rdtsc();
8935 
8936 	report(low + GUEST_TSC_OFFSET <= guest_tsc &&
8937 	       guest_tsc <= high + GUEST_TSC_OFFSET,
8938 	       "RDTSC value in the guest (%lu) is in range [%lu, %lu]",
8939 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
8940 	report(low <= msr_entry.value && msr_entry.value <= high,
8941 	       "IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
8942 	       msr_entry.value, low, high);
8943 }
8944 
8945 static void vmx_preemption_timer_zero_test_db_handler(struct ex_regs *regs)
8946 {
8947 }
8948 
8949 static void vmx_preemption_timer_zero_test_guest(void)
8950 {
8951 	while (vmx_get_test_stage() < 3)
8952 		vmcall();
8953 }
8954 
8955 static void vmx_preemption_timer_zero_activate_preemption_timer(void)
8956 {
8957 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
8958 	vmcs_write(PREEMPT_TIMER_VALUE, 0);
8959 }
8960 
8961 static void vmx_preemption_timer_zero_advance_past_vmcall(void)
8962 {
8963 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
8964 	enter_guest();
8965 	skip_exit_vmcall();
8966 }
8967 
8968 static void vmx_preemption_timer_zero_inject_db(bool intercept_db)
8969 {
8970 	vmx_preemption_timer_zero_activate_preemption_timer();
8971 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK |
8972 		   INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8973 	vmcs_write(EXC_BITMAP, intercept_db ? 1 << DB_VECTOR : 0);
8974 	enter_guest();
8975 }
8976 
8977 static void vmx_preemption_timer_zero_set_pending_dbg(u32 exception_bitmap)
8978 {
8979 	vmx_preemption_timer_zero_activate_preemption_timer();
8980 	vmcs_write(GUEST_PENDING_DEBUG, BIT(12) | DR_TRAP1);
8981 	vmcs_write(EXC_BITMAP, exception_bitmap);
8982 	enter_guest();
8983 }
8984 
8985 static void vmx_preemption_timer_zero_expect_preempt_at_rip(u64 expected_rip)
8986 {
8987 	u32 reason = (u32)vmcs_read(EXI_REASON);
8988 	u64 guest_rip = vmcs_read(GUEST_RIP);
8989 
8990 	report(reason == VMX_PREEMPT && guest_rip == expected_rip,
8991 	       "Exit reason is 0x%x (expected 0x%x) and guest RIP is %lx (0x%lx expected).",
8992 	       reason, VMX_PREEMPT, guest_rip, expected_rip);
8993 }
8994 
8995 /*
8996  * This test ensures that when the VMX preemption timer is zero at
8997  * VM-entry, a VM-exit occurs after any event injection and after any
8998  * pending debug exceptions are raised, but before execution of any
8999  * guest instructions.
9000  */
9001 static void vmx_preemption_timer_zero_test(void)
9002 {
9003 	u64 db_fault_address = (u64)get_idt_addr(&boot_idt[DB_VECTOR]);
9004 	handler old_db;
9005 	u32 reason;
9006 
9007 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9008 		report_skip("'Activate VMX-preemption timer' not supported");
9009 		return;
9010 	}
9011 
9012 	/*
9013 	 * Install a custom #DB handler that doesn't abort.
9014 	 */
9015 	old_db = handle_exception(DB_VECTOR,
9016 				  vmx_preemption_timer_zero_test_db_handler);
9017 
9018 	test_set_guest(vmx_preemption_timer_zero_test_guest);
9019 
9020 	/*
9021 	 * VMX-preemption timer should fire after event injection.
9022 	 */
9023 	vmx_set_test_stage(0);
9024 	vmx_preemption_timer_zero_inject_db(0);
9025 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
9026 	vmx_preemption_timer_zero_advance_past_vmcall();
9027 
9028 	/*
9029 	 * VMX-preemption timer should fire after event injection.
9030 	 * Exception bitmap is irrelevant, since you can't intercept
9031 	 * an event that you injected.
9032 	 */
9033 	vmx_set_test_stage(1);
9034 	vmx_preemption_timer_zero_inject_db(true);
9035 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
9036 	vmx_preemption_timer_zero_advance_past_vmcall();
9037 
9038 	/*
9039 	 * VMX-preemption timer should fire after pending debug exceptions
9040 	 * have delivered a #DB trap.
9041 	 */
9042 	vmx_set_test_stage(2);
9043 	vmx_preemption_timer_zero_set_pending_dbg(0);
9044 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
9045 	vmx_preemption_timer_zero_advance_past_vmcall();
9046 
9047 	/*
9048 	 * VMX-preemption timer would fire after pending debug exceptions
9049 	 * have delivered a #DB trap, but in this case, the #DB trap is
9050 	 * intercepted.
9051 	 */
9052 	vmx_set_test_stage(3);
9053 	vmx_preemption_timer_zero_set_pending_dbg(1 << DB_VECTOR);
9054 	reason = (u32)vmcs_read(EXI_REASON);
9055 	report(reason == VMX_EXC_NMI, "Exit reason is 0x%x (expected 0x%x)",
9056 	       reason, VMX_EXC_NMI);
9057 
9058 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9059 	enter_guest();
9060 
9061 	handle_exception(DB_VECTOR, old_db);
9062 }
9063 
9064 static u64 vmx_preemption_timer_tf_test_prev_rip;
9065 
9066 static void vmx_preemption_timer_tf_test_db_handler(struct ex_regs *regs)
9067 {
9068 	extern char vmx_preemption_timer_tf_test_endloop;
9069 
9070 	if (vmx_get_test_stage() == 2) {
9071 		/*
9072 		 * Stage 2 means that we're done, one way or another.
9073 		 * Arrange for the iret to drop us out of the wbinvd
9074 		 * loop and stop single-stepping.
9075 		 */
9076 		regs->rip = (u64)&vmx_preemption_timer_tf_test_endloop;
9077 		regs->rflags &= ~X86_EFLAGS_TF;
9078 	} else if (regs->rip == vmx_preemption_timer_tf_test_prev_rip) {
9079 		/*
9080 		 * The RIP should alternate between the wbinvd and the
9081 		 * jmp instruction in the code below. If we ever see
9082 		 * the same instruction twice in a row, that means a
9083 		 * single-step trap has been dropped. Let the
9084 		 * hypervisor know about the failure by executing a
9085 		 * VMCALL.
9086 		 */
9087 		vmcall();
9088 	}
9089 	vmx_preemption_timer_tf_test_prev_rip = regs->rip;
9090 }
9091 
9092 static void vmx_preemption_timer_tf_test_guest(void)
9093 {
9094 	/*
9095 	 * The hypervisor doesn't intercept WBINVD, so the loop below
9096 	 * shouldn't be a problem--it's just two instructions
9097 	 * executing in VMX non-root mode. However, when the
9098 	 * hypervisor is running in a virtual environment, the parent
9099 	 * hypervisor might intercept WBINVD and emulate it. If the
9100 	 * parent hypervisor is broken, the single-step trap after the
9101 	 * WBINVD might be lost.
9102 	 */
9103 	asm volatile("vmcall\n\t"
9104 		     "0: wbinvd\n\t"
9105 		     "1: jmp 0b\n\t"
9106 		     "vmx_preemption_timer_tf_test_endloop:");
9107 }
9108 
9109 /*
9110  * Ensure that the delivery of a "VMX-preemption timer expired"
9111  * VM-exit doesn't disrupt single-stepping in the guest. Note that
9112  * passing this test doesn't ensure correctness, because the test will
9113  * only fail if the VMX-preemtion timer fires at the right time (or
9114  * the wrong time, as it were).
9115  */
9116 static void vmx_preemption_timer_tf_test(void)
9117 {
9118 	handler old_db;
9119 	u32 reason;
9120 	int i;
9121 
9122 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9123 		report_skip("'Activate VMX-preemption timer' not supported");
9124 		return;
9125 	}
9126 
9127 	old_db = handle_exception(DB_VECTOR,
9128 				  vmx_preemption_timer_tf_test_db_handler);
9129 
9130 	test_set_guest(vmx_preemption_timer_tf_test_guest);
9131 
9132 	enter_guest();
9133 	skip_exit_vmcall();
9134 
9135 	vmx_set_test_stage(1);
9136 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9137 	vmcs_write(PREEMPT_TIMER_VALUE, 50000);
9138 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9139 
9140 	/*
9141 	 * The only exit we should see is "VMX-preemption timer
9142 	 * expired."  If we get a VMCALL exit, that means the #DB
9143 	 * handler has detected a missing single-step trap. It doesn't
9144 	 * matter where the guest RIP is when the VMX-preemption timer
9145 	 * expires (whether it's in the WBINVD loop or in the #DB
9146 	 * handler)--a single-step trap should never be discarded.
9147 	 */
9148 	for (i = 0; i < 10000; i++) {
9149 		enter_guest();
9150 		reason = (u32)vmcs_read(EXI_REASON);
9151 		if (reason == VMX_PREEMPT)
9152 			continue;
9153 		TEST_ASSERT(reason == VMX_VMCALL);
9154 		skip_exit_insn();
9155 		break;
9156 	}
9157 
9158 	report(reason == VMX_PREEMPT, "No single-step traps skipped");
9159 
9160 	vmx_set_test_stage(2);
9161 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9162 	enter_guest();
9163 
9164 	handle_exception(DB_VECTOR, old_db);
9165 }
9166 
9167 #define VMX_PREEMPTION_TIMER_EXPIRY_CYCLES 1000000
9168 
9169 static u64 vmx_preemption_timer_expiry_start;
9170 static u64 vmx_preemption_timer_expiry_finish;
9171 
9172 static void vmx_preemption_timer_expiry_test_guest(void)
9173 {
9174 	vmcall();
9175 	vmx_preemption_timer_expiry_start = fenced_rdtsc();
9176 
9177 	while (vmx_get_test_stage() == 0)
9178 		vmx_preemption_timer_expiry_finish = fenced_rdtsc();
9179 }
9180 
9181 /*
9182  * Test that the VMX-preemption timer is not excessively delayed.
9183  *
9184  * Per the SDM, volume 3, VM-entry starts the VMX-preemption timer
9185  * with the unsigned value in the VMX-preemption timer-value field,
9186  * and the VMX-preemption timer counts down by 1 every time bit X in
9187  * the TSC changes due to a TSC increment (where X is
9188  * IA32_VMX_MISC[4:0]). If the timer counts down to zero in any state
9189  * other than the wait-for-SIPI state, the logical processor
9190  * transitions to the C0 C-state and causes a VM-exit.
9191  *
9192  * The guest code above reads the starting TSC after VM-entry. At this
9193  * point, the VMX-preemption timer has already been activated. Next,
9194  * the guest code reads the current TSC in a loop, storing the value
9195  * read to memory.
9196  *
9197  * If the RDTSC in the loop reads a value past the VMX-preemption
9198  * timer deadline, then the VMX-preemption timer VM-exit must be
9199  * delivered before the next instruction retires. Even if a higher
9200  * priority SMI is delivered first, the VMX-preemption timer VM-exit
9201  * must be delivered before the next instruction retires. Hence, a TSC
9202  * value past the VMX-preemption timer deadline might be read, but it
9203  * cannot be stored. If a TSC value past the deadline *is* stored,
9204  * then the architectural specification has been violated.
9205  */
9206 static void vmx_preemption_timer_expiry_test(void)
9207 {
9208 	u32 preemption_timer_value;
9209 	union vmx_misc misc;
9210 	u64 tsc_deadline;
9211 	u32 reason;
9212 
9213 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9214 		report_skip("'Activate VMX-preemption timer' not supported");
9215 		return;
9216 	}
9217 
9218 	test_set_guest(vmx_preemption_timer_expiry_test_guest);
9219 
9220 	enter_guest();
9221 	skip_exit_vmcall();
9222 
9223 	misc.val = rdmsr(MSR_IA32_VMX_MISC);
9224 	preemption_timer_value =
9225 		VMX_PREEMPTION_TIMER_EXPIRY_CYCLES >> misc.pt_bit;
9226 
9227 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9228 	vmcs_write(PREEMPT_TIMER_VALUE, preemption_timer_value);
9229 	vmx_set_test_stage(0);
9230 
9231 	enter_guest();
9232 	reason = (u32)vmcs_read(EXI_REASON);
9233 	TEST_ASSERT(reason == VMX_PREEMPT);
9234 
9235 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9236 	vmx_set_test_stage(1);
9237 	enter_guest();
9238 
9239 	tsc_deadline = ((vmx_preemption_timer_expiry_start >> misc.pt_bit) <<
9240 			misc.pt_bit) + (preemption_timer_value << misc.pt_bit);
9241 
9242 	report(vmx_preemption_timer_expiry_finish < tsc_deadline,
9243 	       "Last stored guest TSC (%lu) < TSC deadline (%lu)",
9244 	       vmx_preemption_timer_expiry_finish, tsc_deadline);
9245 }
9246 
9247 static void vmx_db_test_guest(void)
9248 {
9249 	/*
9250 	 * For a hardware generated single-step #DB.
9251 	 */
9252 	asm volatile("vmcall;"
9253 		     "nop;"
9254 		     ".Lpost_nop:");
9255 	/*
9256 	 * ...in a MOVSS shadow, with pending debug exceptions.
9257 	 */
9258 	asm volatile("vmcall;"
9259 		     "nop;"
9260 		     ".Lpost_movss_nop:");
9261 	/*
9262 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
9263 	 * emulates it in software.)
9264 	 */
9265 	asm volatile("vmcall;"
9266 		     "wbinvd;"
9267 		     ".Lpost_wbinvd:");
9268 	/*
9269 	 * ...in a MOVSS shadow, with pending debug exceptions.
9270 	 */
9271 	asm volatile("vmcall;"
9272 		     "wbinvd;"
9273 		     ".Lpost_movss_wbinvd:");
9274 	/*
9275 	 * For a hardware generated single-step #DB in a transactional region.
9276 	 */
9277 	asm volatile("vmcall;"
9278 		     ".Lxbegin: xbegin .Lskip_rtm;"
9279 		     "xend;"
9280 		     ".Lskip_rtm:");
9281 }
9282 
9283 /*
9284  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
9285  * L2. No #DB is delivered and L2 continues to the next point of
9286  * interest.
9287  */
9288 static void dismiss_db(void)
9289 {
9290 	vmcs_write(GUEST_PENDING_DEBUG, 0);
9291 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
9292 	enter_guest();
9293 }
9294 
9295 /*
9296  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
9297  * Then throw away the #DB exception and resume L2.
9298  */
9299 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
9300 			  void *expected_rip, u64 expected_exit_qual,
9301 			  u64 expected_dr6)
9302 {
9303 	u32 reason = vmcs_read(EXI_REASON);
9304 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
9305 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
9306 	u64 guest_rip = vmcs_read(GUEST_RIP);
9307 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
9308 	u64 dr6 = read_dr6();
9309 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
9310 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
9311 
9312 	report(reason == VMX_EXC_NMI && intr_info == expected_intr_info,
9313 	       "Expected #DB VM-exit");
9314 	report((u64)expected_rip == guest_rip, "Expected RIP %p (actual %lx)",
9315 	       expected_rip, guest_rip);
9316 	report_xfail(xfail_pdbg, 0 == guest_pending_dbg,
9317 		     "Expected pending debug exceptions 0 (actual %lx)",
9318 		     guest_pending_dbg);
9319 	report_xfail(xfail_qual, expected_exit_qual == exit_qual,
9320 		     "Expected exit qualification %lx (actual %lx)",
9321 		     expected_exit_qual, exit_qual);
9322 	report_xfail(xfail_dr6, expected_dr6 == dr6,
9323 		     "Expected DR6 %lx (actual %lx)", expected_dr6, dr6);
9324 	dismiss_db();
9325 }
9326 
9327 /*
9328  * Assuming the guest has just exited on a VMCALL instruction, skip
9329  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
9330  * pending debug exceptions are non-zero, set the VMCS up as if the
9331  * previous instruction was a MOVSS that generated the indicated
9332  * pending debug exceptions. Then enter L2.
9333  */
9334 static void single_step_guest(const char *test_name, u64 starting_dr6,
9335 			      u64 pending_debug_exceptions)
9336 {
9337 	printf("\n%s\n", test_name);
9338 	skip_exit_vmcall();
9339 	write_dr6(starting_dr6);
9340 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9341 	if (pending_debug_exceptions) {
9342 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
9343 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
9344 	}
9345 	enter_guest();
9346 }
9347 
9348 /*
9349  * When L1 intercepts #DB, verify that a single-step trap clears
9350  * pending debug exceptions, populates the exit qualification field
9351  * properly, and that DR6 is not prematurely clobbered. In a
9352  * (simulated) MOVSS shadow, make sure that the pending debug
9353  * exception bits are properly accumulated into the exit qualification
9354  * field.
9355  */
9356 static void vmx_db_test(void)
9357 {
9358 	/*
9359 	 * We are going to set a few arbitrary bits in DR6 to verify that
9360 	 * (a) DR6 is not modified by an intercepted #DB, and
9361 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
9362          *     the exit qualification field for a subsequent #DB exception.
9363 	 */
9364 	const u64 starting_dr6 = DR6_RESERVED | BIT(13) | DR_TRAP3 | DR_TRAP1;
9365 	extern char post_nop asm(".Lpost_nop");
9366 	extern char post_movss_nop asm(".Lpost_movss_nop");
9367 	extern char post_wbinvd asm(".Lpost_wbinvd");
9368 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
9369 	extern char xbegin asm(".Lxbegin");
9370 	extern char skip_rtm asm(".Lskip_rtm");
9371 
9372 	/*
9373 	 * L1 wants to intercept #DB exceptions encountered in L2.
9374 	 */
9375 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
9376 
9377 	/*
9378 	 * Start L2 and run it up to the first point of interest.
9379 	 */
9380 	test_set_guest(vmx_db_test_guest);
9381 	enter_guest();
9382 
9383 	/*
9384 	 * Hardware-delivered #DB trap for single-step sets the
9385 	 * standard that L0 has to follow for emulated instructions.
9386 	 */
9387 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
9388 	check_db_exit(false, false, false, &post_nop, DR_STEP, starting_dr6);
9389 
9390 	/*
9391 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
9392 	 * also sets the standard that L0 has to follow for emulated
9393 	 * instructions. Here, we establish the VMCS pending debug
9394 	 * exceptions to indicate that the simulated MOVSS triggered a
9395 	 * data breakpoint as well as the single-step trap.
9396 	 */
9397 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
9398 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0 );
9399 	check_db_exit(false, false, false, &post_movss_nop, DR_STEP | DR_TRAP0,
9400 		      starting_dr6);
9401 
9402 	/*
9403 	 * L0 synthesized #DB trap for single-step is buggy, because
9404 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
9405 	 * reconstitute the exit qualification from the prematurely
9406 	 * modified DR6, but fails miserably.
9407 	 */
9408 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
9409 	check_db_exit(false, false, false, &post_wbinvd, DR_STEP, starting_dr6);
9410 
9411 	/*
9412 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
9413 	 * even worse, because L0 also leaves the pending debug
9414 	 * exceptions in the VMCS instead of accumulating them into
9415 	 * the exit qualification field for the #DB exception.
9416 	 */
9417 	single_step_guest("Software synthesized single-step in MOVSS shadow",
9418 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0);
9419 	check_db_exit(true, false, true, &post_movss_wbinvd, DR_STEP | DR_TRAP0,
9420 		      starting_dr6);
9421 
9422 	/*
9423 	 * Optional RTM test for hardware that supports RTM, to
9424 	 * demonstrate that the current volume 3 of the SDM
9425 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
9426 	 * qualification for debug exceptions is not reserved. It is
9427 	 * set to 1 if a debug exception (#DB) or a breakpoint
9428 	 * exception (#BP) occurs inside an RTM region while advanced
9429 	 * debugging of RTM transactional regions is enabled.
9430 	 */
9431 	if (this_cpu_has(X86_FEATURE_RTM)) {
9432 		vmcs_write(ENT_CONTROLS,
9433 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
9434 		/*
9435 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
9436 		 * in the guest to enable advanced debugging of RTM
9437 		 * transactional regions.
9438 		 */
9439 		vmcs_write(GUEST_DR7, BIT(11));
9440 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
9441 		single_step_guest("Hardware delivered single-step in "
9442 				  "transactional region", starting_dr6, 0);
9443 		check_db_exit(false, false, false, &xbegin, BIT(16),
9444 			      starting_dr6);
9445 	} else {
9446 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
9447 		enter_guest();
9448 	}
9449 }
9450 
9451 static void enable_vid(void)
9452 {
9453 	void *virtual_apic_page;
9454 
9455 	assert(cpu_has_apicv());
9456 
9457 	disable_intercept_for_x2apic_msrs();
9458 
9459 	virtual_apic_page = alloc_page();
9460 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
9461 
9462 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
9463 
9464 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
9465 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
9466 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
9467 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
9468 
9469 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
9470 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
9471 }
9472 
9473 static void trigger_ioapic_scan_thread(void *data)
9474 {
9475 	/* Wait until other CPU entered L2 */
9476 	while (vmx_get_test_stage() != 1)
9477 		;
9478 
9479 	/* Trigger ioapic scan */
9480 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
9481 	vmx_set_test_stage(2);
9482 }
9483 
9484 static void irq_79_handler_guest(isr_regs_t *regs)
9485 {
9486 	eoi();
9487 
9488 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
9489 	vmcall();
9490 }
9491 
9492 /*
9493  * Constant for num of busy-loop iterations after which
9494  * a timer interrupt should have happened in host
9495  */
9496 #define TIMER_INTERRUPT_DELAY 100000000
9497 
9498 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
9499 {
9500 	handle_irq(0x79, irq_79_handler_guest);
9501 	irq_enable();
9502 
9503 	/* Signal to L1 CPU to trigger ioapic scan */
9504 	vmx_set_test_stage(1);
9505 	/* Wait until L1 CPU to trigger ioapic scan */
9506 	while (vmx_get_test_stage() != 2)
9507 		;
9508 
9509 	/*
9510 	 * Wait for L0 timer interrupt to be raised while we run in L2
9511 	 * such that L0 will process the IOAPIC scan request before
9512 	 * resuming L2
9513 	 */
9514 	delay(TIMER_INTERRUPT_DELAY);
9515 
9516 	asm volatile ("int $0x79");
9517 }
9518 
9519 static void vmx_eoi_bitmap_ioapic_scan_test(void)
9520 {
9521 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
9522 		report_skip(__func__);
9523 		return;
9524 	}
9525 
9526 	enable_vid();
9527 
9528 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
9529 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
9530 
9531 	/*
9532 	 * Launch L2.
9533 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
9534 	 * In case the reason isn't VMX_VMCALL, the asserion inside
9535 	 * skip_exit_vmcall() will fail.
9536 	 */
9537 	enter_guest();
9538 	skip_exit_vmcall();
9539 
9540 	/* Let L2 finish */
9541 	enter_guest();
9542 	report_pass(__func__);
9543 }
9544 
9545 #define HLT_WITH_RVI_VECTOR		(0xf1)
9546 
9547 bool vmx_hlt_with_rvi_guest_isr_fired;
9548 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
9549 {
9550 	vmx_hlt_with_rvi_guest_isr_fired = true;
9551 	eoi();
9552 }
9553 
9554 static void vmx_hlt_with_rvi_guest(void)
9555 {
9556 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
9557 
9558 	irq_enable();
9559 	asm volatile ("nop");
9560 
9561 	vmcall();
9562 }
9563 
9564 static void vmx_hlt_with_rvi_test(void)
9565 {
9566 	if (!cpu_has_apicv()) {
9567 		report_skip(__func__);
9568 		return;
9569 	}
9570 
9571 	enable_vid();
9572 
9573 	vmx_hlt_with_rvi_guest_isr_fired = false;
9574 	test_set_guest(vmx_hlt_with_rvi_guest);
9575 
9576 	enter_guest();
9577 	skip_exit_vmcall();
9578 
9579 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
9580 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
9581 	enter_guest();
9582 
9583 	report(vmx_hlt_with_rvi_guest_isr_fired, "Interrupt raised in guest");
9584 }
9585 
9586 static void set_irq_line_thread(void *data)
9587 {
9588 	/* Wait until other CPU entered L2 */
9589 	while (vmx_get_test_stage() != 1)
9590 		;
9591 
9592 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
9593 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9594 	vmx_set_test_stage(2);
9595 }
9596 
9597 static bool irq_78_handler_vmcall_before_eoi;
9598 static void irq_78_handler_guest(isr_regs_t *regs)
9599 {
9600 	set_irq_line(0xf, 0);
9601 	if (irq_78_handler_vmcall_before_eoi)
9602 		vmcall();
9603 	eoi();
9604 	vmcall();
9605 }
9606 
9607 static void vmx_apic_passthrough_guest(void)
9608 {
9609 	handle_irq(0x78, irq_78_handler_guest);
9610 	irq_enable();
9611 
9612 	/* If requested, wait for other CPU to trigger ioapic scan */
9613 	if (vmx_get_test_stage() < 1) {
9614 		vmx_set_test_stage(1);
9615 		while (vmx_get_test_stage() != 2)
9616 			;
9617 	}
9618 
9619 	set_irq_line(0xf, 1);
9620 }
9621 
9622 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
9623 {
9624 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
9625 		report_skip(__func__);
9626 		return;
9627 	}
9628 
9629 	/* Test device is required for generating IRQs */
9630 	if (!test_device_enabled()) {
9631 		report_skip(__func__);
9632 		return;
9633 	}
9634 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9635 	u64 cpu_ctrl_1 = 0;
9636 
9637 	disable_intercept_for_x2apic_msrs();
9638 
9639 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9640 
9641 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9642 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9643 
9644 	if (set_irq_line_from_thread) {
9645 		irq_78_handler_vmcall_before_eoi = false;
9646 		on_cpu_async(1, set_irq_line_thread, NULL);
9647 	} else {
9648 		irq_78_handler_vmcall_before_eoi = true;
9649 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9650 		vmx_set_test_stage(2);
9651 	}
9652 	test_set_guest(vmx_apic_passthrough_guest);
9653 
9654 	if (irq_78_handler_vmcall_before_eoi) {
9655 		/* Before EOI remote_irr should still be set */
9656 		enter_guest();
9657 		skip_exit_vmcall();
9658 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
9659 			"IOAPIC pass-through: remote_irr=1 before EOI");
9660 	}
9661 
9662 	/* After EOI remote_irr should be cleared */
9663 	enter_guest();
9664 	skip_exit_vmcall();
9665 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
9666 		"IOAPIC pass-through: remote_irr=0 after EOI");
9667 
9668 	/* Let L2 finish */
9669 	enter_guest();
9670 	report_pass(__func__);
9671 }
9672 
9673 static void vmx_apic_passthrough_test(void)
9674 {
9675 	vmx_apic_passthrough(false);
9676 }
9677 
9678 static void vmx_apic_passthrough_thread_test(void)
9679 {
9680 	vmx_apic_passthrough(true);
9681 }
9682 
9683 static void vmx_apic_passthrough_tpr_threshold_guest(void)
9684 {
9685 	cli();
9686 	apic_set_tpr(0);
9687 }
9688 
9689 static bool vmx_apic_passthrough_tpr_threshold_ipi_isr_fired;
9690 static void vmx_apic_passthrough_tpr_threshold_ipi_isr(isr_regs_t *regs)
9691 {
9692 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = true;
9693 	eoi();
9694 }
9695 
9696 static void vmx_apic_passthrough_tpr_threshold_test(void)
9697 {
9698 	int ipi_vector = 0xe1;
9699 
9700 	disable_intercept_for_x2apic_msrs();
9701 	vmcs_clear_bits(PIN_CONTROLS, PIN_EXTINT);
9702 
9703 	/* Raise L0 TPR-threshold by queueing vector in LAPIC IRR */
9704 	cli();
9705 	apic_set_tpr((ipi_vector >> 4) + 1);
9706 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
9707 			APIC_DM_FIXED | ipi_vector,
9708 			0);
9709 
9710 	test_set_guest(vmx_apic_passthrough_tpr_threshold_guest);
9711 	enter_guest();
9712 
9713 	report(apic_get_tpr() == 0, "TPR was zero by guest");
9714 
9715 	/* Clean pending self-IPI */
9716 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = false;
9717 	handle_irq(ipi_vector, vmx_apic_passthrough_tpr_threshold_ipi_isr);
9718 	sti();
9719 	asm volatile ("nop");
9720 	report(vmx_apic_passthrough_tpr_threshold_ipi_isr_fired, "self-IPI fired");
9721 
9722 	report_pass(__func__);
9723 }
9724 
9725 static u64 init_signal_test_exit_reason;
9726 static bool init_signal_test_thread_continued;
9727 
9728 static void init_signal_test_thread(void *data)
9729 {
9730 	struct vmcs *test_vmcs = data;
9731 
9732 	/* Enter VMX operation (i.e. exec VMXON) */
9733 	u64 *ap_vmxon_region = alloc_page();
9734 	enable_vmx();
9735 	init_vmx(ap_vmxon_region);
9736 	_vmx_on(ap_vmxon_region);
9737 
9738 	/* Signal CPU have entered VMX operation */
9739 	vmx_set_test_stage(1);
9740 
9741 	/* Wait for BSP CPU to send INIT signal */
9742 	while (vmx_get_test_stage() != 2)
9743 		;
9744 
9745 	/*
9746 	 * Signal that we continue as usual as INIT signal
9747 	 * should be blocked while CPU is in VMX operation
9748 	 */
9749 	vmx_set_test_stage(3);
9750 
9751 	/* Wait for signal to enter VMX non-root mode */
9752 	while (vmx_get_test_stage() != 4)
9753 		;
9754 
9755 	/* Enter VMX non-root mode */
9756 	test_set_guest(v2_null_test_guest);
9757 	make_vmcs_current(test_vmcs);
9758 	enter_guest();
9759 	/* Save exit reason for BSP CPU to compare to expected result */
9760 	init_signal_test_exit_reason = vmcs_read(EXI_REASON);
9761 	/* VMCLEAR test-vmcs so it could be loaded by BSP CPU */
9762 	vmcs_clear(test_vmcs);
9763 	launched = false;
9764 	/* Signal that CPU exited to VMX root mode */
9765 	vmx_set_test_stage(5);
9766 
9767 	/* Wait for BSP CPU to signal to exit VMX operation */
9768 	while (vmx_get_test_stage() != 6)
9769 		;
9770 
9771 	/* Exit VMX operation (i.e. exec VMXOFF) */
9772 	vmx_off();
9773 
9774 	/*
9775 	 * Signal to BSP CPU that we continue as usual as INIT signal
9776 	 * should have been consumed by VMX_INIT exit from guest
9777 	 */
9778 	vmx_set_test_stage(7);
9779 
9780 	/* Wait for BSP CPU to signal to enter VMX operation */
9781 	while (vmx_get_test_stage() != 8)
9782 		;
9783 	/* Enter VMX operation (i.e. exec VMXON) */
9784 	_vmx_on(ap_vmxon_region);
9785 	/* Signal to BSP we are in VMX operation */
9786 	vmx_set_test_stage(9);
9787 
9788 	/* Wait for BSP CPU to send INIT signal */
9789 	while (vmx_get_test_stage() != 10)
9790 		;
9791 
9792 	/* Exit VMX operation (i.e. exec VMXOFF) */
9793 	vmx_off();
9794 
9795 	/*
9796 	 * Exiting VMX operation should result in latched
9797 	 * INIT signal being processed. Therefore, we should
9798 	 * never reach the below code. Thus, signal to BSP
9799 	 * CPU if we have reached here so it is able to
9800 	 * report an issue if it happens.
9801 	 */
9802 	init_signal_test_thread_continued = true;
9803 }
9804 
9805 #define INIT_SIGNAL_TEST_DELAY	100000000ULL
9806 
9807 static void vmx_init_signal_test(void)
9808 {
9809 	struct vmcs *test_vmcs;
9810 
9811 	if (cpu_count() < 2) {
9812 		report_skip(__func__);
9813 		return;
9814 	}
9815 
9816 	/* VMCLEAR test-vmcs so it could be loaded by other CPU */
9817 	vmcs_save(&test_vmcs);
9818 	vmcs_clear(test_vmcs);
9819 
9820 	vmx_set_test_stage(0);
9821 	on_cpu_async(1, init_signal_test_thread, test_vmcs);
9822 
9823 	/* Wait for other CPU to enter VMX operation */
9824 	while (vmx_get_test_stage() != 1)
9825 		;
9826 
9827 	/* Send INIT signal to other CPU */
9828 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9829 				   id_map[1]);
9830 	/* Signal other CPU we have sent INIT signal */
9831 	vmx_set_test_stage(2);
9832 
9833 	/*
9834 	 * Wait reasonable amount of time for INIT signal to
9835 	 * be received on other CPU and verify that other CPU
9836 	 * have proceed as usual to next test stage as INIT
9837 	 * signal should be blocked while other CPU in
9838 	 * VMX operation
9839 	 */
9840 	delay(INIT_SIGNAL_TEST_DELAY);
9841 	report(vmx_get_test_stage() == 3,
9842 	       "INIT signal blocked when CPU in VMX operation");
9843 	/* No point to continue if we failed at this point */
9844 	if (vmx_get_test_stage() != 3)
9845 		return;
9846 
9847 	/* Signal other CPU to enter VMX non-root mode */
9848 	init_signal_test_exit_reason = -1ull;
9849 	vmx_set_test_stage(4);
9850 	/*
9851 	 * Wait reasonable amont of time for other CPU
9852 	 * to exit to VMX root mode
9853 	 */
9854 	delay(INIT_SIGNAL_TEST_DELAY);
9855 	if (vmx_get_test_stage() != 5) {
9856 		report_fail("Pending INIT signal didn't result in VMX exit");
9857 		return;
9858 	}
9859 	report(init_signal_test_exit_reason == VMX_INIT,
9860 			"INIT signal during VMX non-root mode result in exit-reason %s (%lu)",
9861 			exit_reason_description(init_signal_test_exit_reason),
9862 			init_signal_test_exit_reason);
9863 
9864 	/* Run guest to completion */
9865 	make_vmcs_current(test_vmcs);
9866 	enter_guest();
9867 
9868 	/* Signal other CPU to exit VMX operation */
9869 	init_signal_test_thread_continued = false;
9870 	vmx_set_test_stage(6);
9871 
9872 	/* Wait reasonable amount of time for other CPU to exit VMX operation */
9873 	delay(INIT_SIGNAL_TEST_DELAY);
9874 	report(vmx_get_test_stage() == 7,
9875 	       "INIT signal consumed on VMX_INIT exit");
9876 	/* No point to continue if we failed at this point */
9877 	if (vmx_get_test_stage() != 7)
9878 		return;
9879 
9880 	/* Signal other CPU to enter VMX operation */
9881 	vmx_set_test_stage(8);
9882 	/* Wait for other CPU to enter VMX operation */
9883 	while (vmx_get_test_stage() != 9)
9884 		;
9885 
9886 	/* Send INIT signal to other CPU */
9887 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9888 				   id_map[1]);
9889 	/* Signal other CPU we have sent INIT signal */
9890 	vmx_set_test_stage(10);
9891 
9892 	/*
9893 	 * Wait reasonable amount of time for other CPU
9894 	 * to exit VMX operation and process INIT signal
9895 	 */
9896 	delay(INIT_SIGNAL_TEST_DELAY);
9897 	report(!init_signal_test_thread_continued,
9898 	       "INIT signal processed after exit VMX operation");
9899 
9900 	/*
9901 	 * TODO: Send SIPI to other CPU to sipi_entry (See x86/cstart64.S)
9902 	 * to re-init it to kvm-unit-tests standard environment.
9903 	 * Somehow (?) verify that SIPI was indeed received.
9904 	 */
9905 }
9906 
9907 #define SIPI_SIGNAL_TEST_DELAY	100000000ULL
9908 
9909 static void vmx_sipi_test_guest(void)
9910 {
9911 	if (apic_id() == 0) {
9912 		/* wait AP enter guest with activity=WAIT_SIPI */
9913 		while (vmx_get_test_stage() != 1)
9914 			;
9915 		delay(SIPI_SIGNAL_TEST_DELAY);
9916 
9917 		/* First SIPI signal */
9918 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9919 		report_pass("BSP(L2): Send first SIPI to cpu[%d]", id_map[1]);
9920 
9921 		/* wait AP enter guest */
9922 		while (vmx_get_test_stage() != 2)
9923 			;
9924 		delay(SIPI_SIGNAL_TEST_DELAY);
9925 
9926 		/* Second SIPI signal should be ignored since AP is not in WAIT_SIPI state */
9927 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9928 		report_pass("BSP(L2): Send second SIPI to cpu[%d]", id_map[1]);
9929 
9930 		/* Delay a while to check whether second SIPI would cause VMExit */
9931 		delay(SIPI_SIGNAL_TEST_DELAY);
9932 
9933 		/* Test is done, notify AP to exit test */
9934 		vmx_set_test_stage(3);
9935 
9936 		/* wait AP exit non-root mode */
9937 		while (vmx_get_test_stage() != 5)
9938 			;
9939 	} else {
9940 		/* wait BSP notify test is done */
9941 		while (vmx_get_test_stage() != 3)
9942 			;
9943 
9944 		/* AP exit guest */
9945 		vmx_set_test_stage(4);
9946 	}
9947 }
9948 
9949 static void sipi_test_ap_thread(void *data)
9950 {
9951 	struct vmcs *ap_vmcs;
9952 	u64 *ap_vmxon_region;
9953 	void *ap_stack, *ap_syscall_stack;
9954 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9955 	u64 cpu_ctrl_1 = 0;
9956 
9957 	/* Enter VMX operation (i.e. exec VMXON) */
9958 	ap_vmxon_region = alloc_page();
9959 	enable_vmx();
9960 	init_vmx(ap_vmxon_region);
9961 	_vmx_on(ap_vmxon_region);
9962 	init_vmcs(&ap_vmcs);
9963 	make_vmcs_current(ap_vmcs);
9964 
9965 	/* Set stack for AP */
9966 	ap_stack = alloc_page();
9967 	ap_syscall_stack = alloc_page();
9968 	vmcs_write(GUEST_RSP, (u64)(ap_stack + PAGE_SIZE - 1));
9969 	vmcs_write(GUEST_SYSENTER_ESP, (u64)(ap_syscall_stack + PAGE_SIZE - 1));
9970 
9971 	/* passthrough lapic to L2 */
9972 	disable_intercept_for_x2apic_msrs();
9973 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9974 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9975 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9976 
9977 	/* Set guest activity state to wait-for-SIPI state */
9978 	vmcs_write(GUEST_ACTV_STATE, ACTV_WAIT_SIPI);
9979 
9980 	vmx_set_test_stage(1);
9981 
9982 	/* AP enter guest */
9983 	enter_guest();
9984 
9985 	if (vmcs_read(EXI_REASON) == VMX_SIPI) {
9986 		report_pass("AP: Handle SIPI VMExit");
9987 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
9988 		vmx_set_test_stage(2);
9989 	} else {
9990 		report_fail("AP: Unexpected VMExit, reason=%ld", vmcs_read(EXI_REASON));
9991 		vmx_off();
9992 		return;
9993 	}
9994 
9995 	/* AP enter guest */
9996 	enter_guest();
9997 
9998 	report(vmcs_read(EXI_REASON) != VMX_SIPI,
9999 		"AP: should no SIPI VMExit since activity is not in WAIT_SIPI state");
10000 
10001 	/* notify BSP that AP is already exit from non-root mode */
10002 	vmx_set_test_stage(5);
10003 
10004 	/* Leave VMX operation */
10005 	vmx_off();
10006 }
10007 
10008 static void vmx_sipi_signal_test(void)
10009 {
10010 	if (!(rdmsr(MSR_IA32_VMX_MISC) & MSR_IA32_VMX_MISC_ACTIVITY_WAIT_SIPI)) {
10011 		printf("\tACTIVITY_WAIT_SIPI state is not supported.\n");
10012 		return;
10013 	}
10014 
10015 	if (cpu_count() < 2) {
10016 		report_skip(__func__);
10017 		return;
10018 	}
10019 
10020 	u64 cpu_ctrl_0 = CPU_SECONDARY;
10021 	u64 cpu_ctrl_1 = 0;
10022 
10023 	/* passthrough lapic to L2 */
10024 	disable_intercept_for_x2apic_msrs();
10025 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
10026 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
10027 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
10028 
10029 	test_set_guest(vmx_sipi_test_guest);
10030 
10031 	/* update CR3 on AP */
10032 	on_cpu(1, update_cr3, (void *)read_cr3());
10033 
10034 	/* start AP */
10035 	on_cpu_async(1, sipi_test_ap_thread, NULL);
10036 
10037 	vmx_set_test_stage(0);
10038 
10039 	/* BSP enter guest */
10040 	enter_guest();
10041 }
10042 
10043 
10044 enum vmcs_access {
10045 	ACCESS_VMREAD,
10046 	ACCESS_VMWRITE,
10047 	ACCESS_NONE,
10048 };
10049 
10050 struct vmcs_shadow_test_common {
10051 	enum vmcs_access op;
10052 	enum Reason reason;
10053 	u64 field;
10054 	u64 value;
10055 	u64 flags;
10056 	u64 time;
10057 } l1_l2_common;
10058 
10059 static inline u64 vmread_flags(u64 field, u64 *val)
10060 {
10061 	u64 flags;
10062 
10063 	asm volatile ("vmread %2, %1; pushf; pop %0"
10064 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
10065 	return flags & X86_EFLAGS_ALU;
10066 }
10067 
10068 static inline u64 vmwrite_flags(u64 field, u64 val)
10069 {
10070 	u64 flags;
10071 
10072 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
10073 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
10074 	return flags & X86_EFLAGS_ALU;
10075 }
10076 
10077 static void vmx_vmcs_shadow_test_guest(void)
10078 {
10079 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10080 	u64 start;
10081 
10082 	while (c->op != ACCESS_NONE) {
10083 		start = rdtsc();
10084 		switch (c->op) {
10085 		default:
10086 			c->flags = -1ull;
10087 			break;
10088 		case ACCESS_VMREAD:
10089 			c->flags = vmread_flags(c->field, &c->value);
10090 			break;
10091 		case ACCESS_VMWRITE:
10092 			c->flags = vmwrite_flags(c->field, 0);
10093 			break;
10094 		}
10095 		c->time = rdtsc() - start;
10096 		vmcall();
10097 	}
10098 }
10099 
10100 static u64 vmread_from_shadow(u64 field)
10101 {
10102 	struct vmcs *primary;
10103 	struct vmcs *shadow;
10104 	u64 value;
10105 
10106 	TEST_ASSERT(!vmcs_save(&primary));
10107 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10108 	TEST_ASSERT(!make_vmcs_current(shadow));
10109 	value = vmcs_read(field);
10110 	TEST_ASSERT(!make_vmcs_current(primary));
10111 	return value;
10112 }
10113 
10114 static u64 vmwrite_to_shadow(u64 field, u64 value)
10115 {
10116 	struct vmcs *primary;
10117 	struct vmcs *shadow;
10118 
10119 	TEST_ASSERT(!vmcs_save(&primary));
10120 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10121 	TEST_ASSERT(!make_vmcs_current(shadow));
10122 	vmcs_write(field, value);
10123 	value = vmcs_read(field);
10124 	TEST_ASSERT(!make_vmcs_current(primary));
10125 	return value;
10126 }
10127 
10128 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
10129 {
10130 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10131 
10132 	c->op = access;
10133 	vmcs_write(VMX_INST_ERROR, 0);
10134 	enter_guest();
10135 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
10136 	if (c->reason != VMX_VMCALL) {
10137 		skip_exit_insn();
10138 		enter_guest();
10139 	}
10140 	skip_exit_vmcall();
10141 }
10142 
10143 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
10144 {
10145 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10146 	struct vmcs *shadow;
10147 	u64 value;
10148 	uintptr_t flags[2];
10149 	bool good_shadow;
10150 	u32 vmx_inst_error;
10151 
10152 	report_prefix_pushf("field %lx", field);
10153 	c->field = field;
10154 
10155 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10156 	if (shadow != (struct vmcs *)-1ull) {
10157 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
10158 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
10159 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
10160 	} else {
10161 		/*
10162 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
10163 		 * shadowed-fields should fail with setting RFLAGS.CF.
10164 		 */
10165 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
10166 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
10167 		good_shadow = false;
10168 	}
10169 
10170 	/* Intercept both VMREAD and VMWRITE. */
10171 	report_prefix_push("no VMREAD/VMWRITE permission");
10172 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
10173 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
10174 		set_bit(field, bitmap[ACCESS_VMREAD]);
10175 		set_bit(field, bitmap[ACCESS_VMWRITE]);
10176 	}
10177 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10178 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10179 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10180 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10181 	report_prefix_pop();
10182 
10183 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
10184 		goto out;
10185 
10186 	/* Permit shadowed VMREAD. */
10187 	report_prefix_push("VMREAD permission only");
10188 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10189 	set_bit(field, bitmap[ACCESS_VMWRITE]);
10190 	if (good_shadow)
10191 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10192 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10193 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10194 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10195 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10196 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10197 	       c->time);
10198 	report(c->flags == flags[ACCESS_VMREAD],
10199 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10200 	       c->flags, flags[ACCESS_VMREAD]);
10201 	if (good_shadow)
10202 		report(c->value == value,
10203 		       "value read from shadow (%lx) is as expected (%lx)",
10204 		       c->value, value);
10205 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10206 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10207 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10208 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10209 	report_prefix_pop();
10210 
10211 	/* Permit shadowed VMWRITE. */
10212 	report_prefix_push("VMWRITE permission only");
10213 	set_bit(field, bitmap[ACCESS_VMREAD]);
10214 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10215 	if (good_shadow)
10216 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10217 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10218 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10219 	report(c->reason == VMX_VMCALL,
10220 		"shadowed for VMWRITE (in %ld cycles)",
10221 		c->time);
10222 	report(c->flags == flags[ACCESS_VMREAD],
10223 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10224 	       c->flags, flags[ACCESS_VMREAD]);
10225 	if (good_shadow) {
10226 		value = vmread_from_shadow(field);
10227 		report(value == 0,
10228 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10229 		       0ul);
10230 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10231 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10232 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10233 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10234 	}
10235 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10236 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10237 	report_prefix_pop();
10238 
10239 	/* Permit shadowed VMREAD and VMWRITE. */
10240 	report_prefix_push("VMREAD and VMWRITE permission");
10241 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10242 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10243 	if (good_shadow)
10244 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10245 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10246 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10247 	report(c->reason == VMX_VMCALL,
10248 		"shadowed for VMWRITE (in %ld cycles)",
10249 		c->time);
10250 	report(c->flags == flags[ACCESS_VMREAD],
10251 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10252 	       c->flags, flags[ACCESS_VMREAD]);
10253 	if (good_shadow) {
10254 		value = vmread_from_shadow(field);
10255 		report(value == 0,
10256 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10257 		       0ul);
10258 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10259 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10260 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10261 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10262 	}
10263 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10264 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10265 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10266 	       c->time);
10267 	report(c->flags == flags[ACCESS_VMREAD],
10268 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10269 	       c->flags, flags[ACCESS_VMREAD]);
10270 	if (good_shadow)
10271 		report(c->value == 0,
10272 		       "value read from shadow (%lx) is as expected (%lx)",
10273 		       c->value, 0ul);
10274 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10275 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10276 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10277 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10278 	report_prefix_pop();
10279 
10280 out:
10281 	report_prefix_pop();
10282 }
10283 
10284 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
10285 {
10286 	unsigned base;
10287 	unsigned index;
10288 	unsigned bit;
10289 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
10290 
10291 	/* Run test on all possible valid VMCS fields */
10292 	for (base = 0;
10293 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
10294 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
10295 		for (index = 0; index <= highest_index; index++)
10296 			vmcs_shadow_test_field(bitmap, base + index);
10297 
10298 	/*
10299 	 * Run tests on some invalid VMCS fields
10300 	 * (Have reserved bit set).
10301 	 */
10302 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
10303 		vmcs_shadow_test_field(bitmap, (1ull << bit));
10304 }
10305 
10306 static void vmx_vmcs_shadow_test(void)
10307 {
10308 	u8 *bitmap[2];
10309 	struct vmcs *shadow;
10310 
10311 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
10312 		printf("\t'Activate secondary controls' not supported.\n");
10313 		return;
10314 	}
10315 
10316 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
10317 		printf("\t'VMCS shadowing' not supported.\n");
10318 		return;
10319 	}
10320 
10321 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
10322 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
10323 		printf("\tVMWRITE can't modify VM-exit information fields.\n");
10324 		return;
10325 	}
10326 
10327 	test_set_guest(vmx_vmcs_shadow_test_guest);
10328 
10329 	bitmap[ACCESS_VMREAD] = alloc_page();
10330 	bitmap[ACCESS_VMWRITE] = alloc_page();
10331 
10332 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
10333 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
10334 
10335 	shadow = alloc_page();
10336 	shadow->hdr.revision_id = basic.revision;
10337 	shadow->hdr.shadow_vmcs = 1;
10338 	TEST_ASSERT(!vmcs_clear(shadow));
10339 
10340 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
10341 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
10342 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
10343 
10344 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
10345 	report_prefix_push("valid link pointer");
10346 	vmx_vmcs_shadow_test_body(bitmap);
10347 	report_prefix_pop();
10348 
10349 	vmcs_write(VMCS_LINK_PTR, -1ull);
10350 	report_prefix_push("invalid link pointer");
10351 	vmx_vmcs_shadow_test_body(bitmap);
10352 	report_prefix_pop();
10353 
10354 	l1_l2_common.op = ACCESS_NONE;
10355 	enter_guest();
10356 }
10357 
10358 /*
10359  * This test monitors the difference between a guest RDTSC instruction
10360  * and the IA32_TIME_STAMP_COUNTER MSR value stored in the VMCS12
10361  * VM-exit MSR-store list when taking a VM-exit on the instruction
10362  * following RDTSC.
10363  */
10364 #define RDTSC_DIFF_ITERS 100000
10365 #define RDTSC_DIFF_FAILS 100
10366 #define HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD 750
10367 
10368 /*
10369  * Set 'use TSC offsetting' and set the guest offset to the
10370  * inverse of the host's current TSC value, so that the guest starts running
10371  * with an effective TSC value of 0.
10372  */
10373 static void reset_guest_tsc_to_zero(void)
10374 {
10375 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
10376 	vmcs_write(TSC_OFFSET, -rdtsc());
10377 }
10378 
10379 static void rdtsc_vmexit_diff_test_guest(void)
10380 {
10381 	int i;
10382 
10383 	for (i = 0; i < RDTSC_DIFF_ITERS; i++)
10384 		/* Ensure rdtsc is the last instruction before the vmcall. */
10385 		asm volatile("rdtsc; vmcall" : : : "eax", "edx");
10386 }
10387 
10388 /*
10389  * This function only considers the "use TSC offsetting" VM-execution
10390  * control.  It does not handle "use TSC scaling" (because the latter
10391  * isn't available to the host today.)
10392  */
10393 static unsigned long long host_time_to_guest_time(unsigned long long t)
10394 {
10395 	TEST_ASSERT(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
10396 		    !(vmcs_read(CPU_EXEC_CTRL1) & CPU_USE_TSC_SCALING));
10397 
10398 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_USE_TSC_OFFSET)
10399 		t += vmcs_read(TSC_OFFSET);
10400 
10401 	return t;
10402 }
10403 
10404 static unsigned long long rdtsc_vmexit_diff_test_iteration(void)
10405 {
10406 	unsigned long long guest_tsc, host_to_guest_tsc;
10407 
10408 	enter_guest();
10409 	skip_exit_vmcall();
10410 	guest_tsc = (u32) regs.rax + (regs.rdx << 32);
10411 	host_to_guest_tsc = host_time_to_guest_time(exit_msr_store[0].value);
10412 
10413 	return host_to_guest_tsc - guest_tsc;
10414 }
10415 
10416 static void rdtsc_vmexit_diff_test(void)
10417 {
10418 	unsigned long long delta;
10419 	int fail = 0;
10420 	int i;
10421 
10422 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET))
10423 		test_skip("CPU doesn't support the 'use TSC offsetting' processor-based VM-execution control.\n");
10424 
10425 	test_set_guest(rdtsc_vmexit_diff_test_guest);
10426 
10427 	reset_guest_tsc_to_zero();
10428 
10429 	/*
10430 	 * Set up the VMCS12 VM-exit MSR-store list to store just one
10431 	 * MSR: IA32_TIME_STAMP_COUNTER. Note that the value stored is
10432 	 * in the host time domain (i.e., it is not adjusted according
10433 	 * to the TSC multiplier and TSC offset fields in the VMCS12,
10434 	 * as a guest RDTSC would be.)
10435 	 */
10436 	exit_msr_store = alloc_page();
10437 	exit_msr_store[0].index = MSR_IA32_TSC;
10438 	vmcs_write(EXI_MSR_ST_CNT, 1);
10439 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(exit_msr_store));
10440 
10441 	for (i = 0; i < RDTSC_DIFF_ITERS && fail < RDTSC_DIFF_FAILS; i++) {
10442 		delta = rdtsc_vmexit_diff_test_iteration();
10443 		if (delta >= HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD)
10444 			fail++;
10445 	}
10446 
10447 	enter_guest();
10448 
10449 	report(fail < RDTSC_DIFF_FAILS,
10450 	       "RDTSC to VM-exit delta too high in %d of %d iterations, last = %llu",
10451 	       fail, i, delta);
10452 }
10453 
10454 static int invalid_msr_init(struct vmcs *vmcs)
10455 {
10456 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
10457 		printf("\tPreemption timer is not supported\n");
10458 		return VMX_TEST_EXIT;
10459 	}
10460 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
10461 	preempt_val = 10000000;
10462 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
10463 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
10464 
10465 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
10466 		printf("\tSave preemption value is not supported\n");
10467 
10468 	vmcs_write(ENT_MSR_LD_CNT, 1);
10469 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
10470 
10471 	return VMX_TEST_START;
10472 }
10473 
10474 
10475 static void invalid_msr_main(void)
10476 {
10477 	report_fail("Invalid MSR load");
10478 }
10479 
10480 static int invalid_msr_exit_handler(union exit_reason exit_reason)
10481 {
10482 	report_fail("Invalid MSR load");
10483 	print_vmexit_info(exit_reason);
10484 	return VMX_TEST_EXIT;
10485 }
10486 
10487 static int invalid_msr_entry_failure(struct vmentry_result *result)
10488 {
10489 	report(result->exit_reason.failed_vmentry &&
10490 	       result->exit_reason.basic == VMX_FAIL_MSR, "Invalid MSR load");
10491 	return VMX_TEST_VMEXIT;
10492 }
10493 
10494 /*
10495  * The max number of MSRs in an atomic switch MSR list is:
10496  * (111B + 1) * 512 = 4096
10497  *
10498  * Each list entry consumes:
10499  * 4-byte MSR index + 4 bytes reserved + 8-byte data = 16 bytes
10500  *
10501  * Allocate 128 kB to cover max_msr_list_size (i.e., 64 kB) and then some.
10502  */
10503 static const u32 msr_list_page_order = 5;
10504 
10505 static void atomic_switch_msr_limit_test_guest(void)
10506 {
10507 	vmcall();
10508 }
10509 
10510 static void populate_msr_list(struct vmx_msr_entry *msr_list,
10511 			      size_t byte_capacity, int count)
10512 {
10513 	int i;
10514 
10515 	for (i = 0; i < count; i++) {
10516 		msr_list[i].index = MSR_IA32_TSC;
10517 		msr_list[i].reserved = 0;
10518 		msr_list[i].value = 0x1234567890abcdef;
10519 	}
10520 
10521 	memset(msr_list + count, 0xff,
10522 	       byte_capacity - count * sizeof(*msr_list));
10523 }
10524 
10525 static int max_msr_list_size(void)
10526 {
10527 	u32 vmx_misc = rdmsr(MSR_IA32_VMX_MISC);
10528 	u32 factor = ((vmx_misc & GENMASK(27, 25)) >> 25) + 1;
10529 
10530 	return factor * 512;
10531 }
10532 
10533 static void atomic_switch_msrs_test(int count)
10534 {
10535 	struct vmx_msr_entry *vm_enter_load;
10536         struct vmx_msr_entry *vm_exit_load;
10537         struct vmx_msr_entry *vm_exit_store;
10538 	int max_allowed = max_msr_list_size();
10539 	int byte_capacity = 1ul << (msr_list_page_order + PAGE_SHIFT);
10540 	/* Exceeding the max MSR list size at exit trigers KVM to abort. */
10541 	int exit_count = count > max_allowed ? max_allowed : count;
10542 	int cleanup_count = count > max_allowed ? 2 : 1;
10543 	int i;
10544 
10545 	/*
10546 	 * Check for the IA32_TSC MSR,
10547 	 * available with the "TSC flag" and used to populate the MSR lists.
10548 	 */
10549 	if (!(cpuid(1).d & (1 << 4))) {
10550 		report_skip(__func__);
10551 		return;
10552 	}
10553 
10554 	/* Set L2 guest. */
10555 	test_set_guest(atomic_switch_msr_limit_test_guest);
10556 
10557 	/* Setup atomic MSR switch lists. */
10558 	vm_enter_load = alloc_pages(msr_list_page_order);
10559 	vm_exit_load = alloc_pages(msr_list_page_order);
10560 	vm_exit_store = alloc_pages(msr_list_page_order);
10561 
10562 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)vm_enter_load);
10563 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)vm_exit_load);
10564 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)vm_exit_store);
10565 
10566 	/*
10567 	 * VM-Enter should succeed up to the max number of MSRs per list, and
10568 	 * should not consume junk beyond the last entry.
10569 	 */
10570 	populate_msr_list(vm_enter_load, byte_capacity, count);
10571 	populate_msr_list(vm_exit_load, byte_capacity, exit_count);
10572 	populate_msr_list(vm_exit_store, byte_capacity, exit_count);
10573 
10574 	vmcs_write(ENT_MSR_LD_CNT, count);
10575 	vmcs_write(EXI_MSR_LD_CNT, exit_count);
10576 	vmcs_write(EXI_MSR_ST_CNT, exit_count);
10577 
10578 	if (count <= max_allowed) {
10579 		enter_guest();
10580 		assert_exit_reason(VMX_VMCALL);
10581 		skip_exit_vmcall();
10582 	} else {
10583 		u32 exit_qual;
10584 
10585 		test_guest_state("Invalid MSR Load Count", true, count,
10586 				 "ENT_MSR_LD_CNT");
10587 
10588 		exit_qual = vmcs_read(EXI_QUALIFICATION);
10589 		report(exit_qual == max_allowed + 1, "exit_qual, %u, is %u.",
10590 		       exit_qual, max_allowed + 1);
10591 	}
10592 
10593 	/* Cleanup. */
10594 	vmcs_write(ENT_MSR_LD_CNT, 0);
10595 	vmcs_write(EXI_MSR_LD_CNT, 0);
10596 	vmcs_write(EXI_MSR_ST_CNT, 0);
10597 	for (i = 0; i < cleanup_count; i++) {
10598 		enter_guest();
10599 		skip_exit_vmcall();
10600 	}
10601 	free_pages_by_order(vm_enter_load, msr_list_page_order);
10602 	free_pages_by_order(vm_exit_load, msr_list_page_order);
10603 	free_pages_by_order(vm_exit_store, msr_list_page_order);
10604 }
10605 
10606 static void atomic_switch_max_msrs_test(void)
10607 {
10608 	atomic_switch_msrs_test(max_msr_list_size());
10609 }
10610 
10611 static void atomic_switch_overflow_msrs_test(void)
10612 {
10613 	if (test_device_enabled())
10614 		atomic_switch_msrs_test(max_msr_list_size() + 1);
10615 	else
10616 		test_skip("Test is only supported on KVM");
10617 }
10618 
10619 static void vmx_pf_exception_test_guest(void)
10620 {
10621 	ac_test_run(PT_LEVEL_PML4);
10622 }
10623 
10624 static void vmx_pf_exception_test(void)
10625 {
10626 	u64 efer;
10627 	struct cpuid cpuid;
10628 
10629 	test_set_guest(vmx_pf_exception_test_guest);
10630 
10631 	enter_guest();
10632 
10633 	while (vmcs_read(EXI_REASON) != VMX_VMCALL) {
10634 		switch (vmcs_read(EXI_REASON)) {
10635 		case VMX_RDMSR:
10636 			assert(regs.rcx == MSR_EFER);
10637 			efer = vmcs_read(GUEST_EFER);
10638 			regs.rdx = efer >> 32;
10639 			regs.rax = efer & 0xffffffff;
10640 			break;
10641 		case VMX_WRMSR:
10642 			assert(regs.rcx == MSR_EFER);
10643 			efer = regs.rdx << 32 | (regs.rax & 0xffffffff);
10644 			vmcs_write(GUEST_EFER, efer);
10645 			break;
10646 		case VMX_CPUID:
10647 			cpuid = (struct cpuid) {0, 0, 0, 0};
10648 			cpuid = raw_cpuid(regs.rax, regs.rcx);
10649 			regs.rax = cpuid.a;
10650 			regs.rbx = cpuid.b;
10651 			regs.rcx = cpuid.c;
10652 			regs.rdx = cpuid.d;
10653 			break;
10654 		default:
10655 			assert_msg(false,
10656 				"Unexpected exit to L1, exit_reason: %s (0x%lx)",
10657 				exit_reason_description(vmcs_read(EXI_REASON)),
10658 				vmcs_read(EXI_REASON));
10659 		}
10660 		skip_exit_insn();
10661 		enter_guest();
10662 	}
10663 
10664 	assert_exit_reason(VMX_VMCALL);
10665 }
10666 #define TEST(name) { #name, .v2 = name }
10667 
10668 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
10669 struct vmx_test vmx_tests[] = {
10670 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
10671 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
10672 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
10673 		preemption_timer_exit_handler, NULL, {0} },
10674 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
10675 		test_ctrl_pat_exit_handler, NULL, {0} },
10676 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
10677 		test_ctrl_efer_exit_handler, NULL, {0} },
10678 	{ "CR shadowing", NULL, cr_shadowing_main,
10679 		cr_shadowing_exit_handler, NULL, {0} },
10680 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
10681 		NULL, {0} },
10682 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
10683 		insn_intercept_exit_handler, NULL, {0} },
10684 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
10685 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
10686 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
10687 	{ "interrupt", interrupt_init, interrupt_main,
10688 		interrupt_exit_handler, NULL, {0} },
10689 	{ "nmi_hlt", nmi_hlt_init, nmi_hlt_main,
10690 		nmi_hlt_exit_handler, NULL, {0} },
10691 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
10692 		NULL, {0} },
10693 	{ "MSR switch", msr_switch_init, msr_switch_main,
10694 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
10695 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
10696 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
10697 		disable_rdtscp_exit_handler, NULL, {0} },
10698 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
10699 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
10700 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
10701 		exit_monitor_from_l2_handler, NULL, {0} },
10702 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
10703 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
10704 	/* Basic V2 tests. */
10705 	TEST(v2_null_test),
10706 	TEST(v2_multiple_entries_test),
10707 	TEST(fixture_test_case1),
10708 	TEST(fixture_test_case2),
10709 	/* Opcode tests. */
10710 	TEST(invvpid_test),
10711 	/* VM-entry tests */
10712 	TEST(vmx_controls_test),
10713 	TEST(vmx_host_state_area_test),
10714 	TEST(vmx_guest_state_area_test),
10715 	TEST(vmentry_movss_shadow_test),
10716 	TEST(vmentry_unrestricted_guest_test),
10717 	/* APICv tests */
10718 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
10719 	TEST(vmx_hlt_with_rvi_test),
10720 	TEST(apic_reg_virt_test),
10721 	TEST(virt_x2apic_mode_test),
10722 	/* APIC pass-through tests */
10723 	TEST(vmx_apic_passthrough_test),
10724 	TEST(vmx_apic_passthrough_thread_test),
10725 	TEST(vmx_apic_passthrough_tpr_threshold_test),
10726 	TEST(vmx_init_signal_test),
10727 	TEST(vmx_sipi_signal_test),
10728 	/* VMCS Shadowing tests */
10729 	TEST(vmx_vmcs_shadow_test),
10730 	/* Regression tests */
10731 	TEST(vmx_ldtr_test),
10732 	TEST(vmx_cr_load_test),
10733 	TEST(vmx_cr4_osxsave_test),
10734 	TEST(vmx_nm_test),
10735 	TEST(vmx_db_test),
10736 	TEST(vmx_nmi_window_test),
10737 	TEST(vmx_intr_window_test),
10738 	TEST(vmx_pending_event_test),
10739 	TEST(vmx_pending_event_hlt_test),
10740 	TEST(vmx_store_tsc_test),
10741 	TEST(vmx_preemption_timer_zero_test),
10742 	TEST(vmx_preemption_timer_tf_test),
10743 	TEST(vmx_preemption_timer_expiry_test),
10744 	/* EPT access tests. */
10745 	TEST(ept_access_test_not_present),
10746 	TEST(ept_access_test_read_only),
10747 	TEST(ept_access_test_write_only),
10748 	TEST(ept_access_test_read_write),
10749 	TEST(ept_access_test_execute_only),
10750 	TEST(ept_access_test_read_execute),
10751 	TEST(ept_access_test_write_execute),
10752 	TEST(ept_access_test_read_write_execute),
10753 	TEST(ept_access_test_reserved_bits),
10754 	TEST(ept_access_test_ignored_bits),
10755 	TEST(ept_access_test_paddr_not_present_ad_disabled),
10756 	TEST(ept_access_test_paddr_not_present_ad_enabled),
10757 	TEST(ept_access_test_paddr_read_only_ad_disabled),
10758 	TEST(ept_access_test_paddr_read_only_ad_enabled),
10759 	TEST(ept_access_test_paddr_read_write),
10760 	TEST(ept_access_test_paddr_read_write_execute),
10761 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
10762 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
10763 	TEST(ept_access_test_paddr_not_present_page_fault),
10764 	TEST(ept_access_test_force_2m_page),
10765 	/* Atomic MSR switch tests. */
10766 	TEST(atomic_switch_max_msrs_test),
10767 	TEST(atomic_switch_overflow_msrs_test),
10768 	TEST(rdtsc_vmexit_diff_test),
10769 	TEST(vmx_mtf_test),
10770 	TEST(vmx_mtf_pdpte_test),
10771 	TEST(vmx_pf_exception_test),
10772 	{ NULL, NULL, NULL, NULL, NULL, {0} },
10773 };
10774