xref: /kvm-unit-tests/x86/vmx_tests.c (revision 4143fbfd524b140b81122286c3984bd1aaac82b7)
1 /*
2  * All test cases of nested virtualization should be in this file
3  *
4  * Author : Arthur Chunqi Li <yzt356@gmail.com>
5  */
6 
7 #include <asm/debugreg.h>
8 
9 #include "vmx.h"
10 #include "msr.h"
11 #include "processor.h"
12 #include "vm.h"
13 #include "pci.h"
14 #include "fwcfg.h"
15 #include "isr.h"
16 #include "desc.h"
17 #include "apic.h"
18 #include "types.h"
19 #include "vmalloc.h"
20 #include "alloc_page.h"
21 #include "smp.h"
22 #include "delay.h"
23 #include "access.h"
24 #include "x86/usermode.h"
25 
26 /*
27  * vmcs.GUEST_PENDING_DEBUG has the same format as DR6, although some bits that
28  * are legal in DR6 are reserved in vmcs.GUEST_PENDING_DEBUG.  And if any data
29  * or I/O breakpoint matches *and* was enabled, bit 12 is also set.
30  */
31 #define PENDING_DBG_TRAP	BIT(12)
32 
33 #define VPID_CAP_INVVPID_TYPES_SHIFT 40
34 
35 u64 ia32_pat;
36 u64 ia32_efer;
37 void *io_bitmap_a, *io_bitmap_b;
38 u16 ioport;
39 
40 unsigned long *pml4;
41 u64 eptp;
42 void *data_page1, *data_page2;
43 
44 phys_addr_t pci_physaddr;
45 
46 void *pml_log;
47 #define PML_INDEX 512
48 
49 static inline unsigned ffs(unsigned x)
50 {
51 	int pos = -1;
52 
53 	__asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0"
54 			     : "+r"(pos) : "rm"(x) : "eax");
55 	return pos + 1;
56 }
57 
58 static inline void vmcall(void)
59 {
60 	asm volatile("vmcall");
61 }
62 
63 static void basic_guest_main(void)
64 {
65 	report_pass("Basic VMX test");
66 }
67 
68 static int basic_exit_handler(union exit_reason exit_reason)
69 {
70 	report_fail("Basic VMX test");
71 	print_vmexit_info(exit_reason);
72 	return VMX_TEST_EXIT;
73 }
74 
75 static void vmenter_main(void)
76 {
77 	u64 rax;
78 	u64 rsp, resume_rsp;
79 
80 	report_pass("test vmlaunch");
81 
82 	asm volatile(
83 		"mov %%rsp, %0\n\t"
84 		"mov %3, %%rax\n\t"
85 		"vmcall\n\t"
86 		"mov %%rax, %1\n\t"
87 		"mov %%rsp, %2\n\t"
88 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
89 		: "g"(0xABCD));
90 	report((rax == 0xFFFF) && (rsp == resume_rsp), "test vmresume");
91 }
92 
93 static int vmenter_exit_handler(union exit_reason exit_reason)
94 {
95 	u64 guest_rip = vmcs_read(GUEST_RIP);
96 
97 	switch (exit_reason.basic) {
98 	case VMX_VMCALL:
99 		if (regs.rax != 0xABCD) {
100 			report_fail("test vmresume");
101 			return VMX_TEST_VMEXIT;
102 		}
103 		regs.rax = 0xFFFF;
104 		vmcs_write(GUEST_RIP, guest_rip + 3);
105 		return VMX_TEST_RESUME;
106 	default:
107 		report_fail("test vmresume");
108 		print_vmexit_info(exit_reason);
109 	}
110 	return VMX_TEST_VMEXIT;
111 }
112 
113 u32 preempt_scale;
114 volatile unsigned long long tsc_val;
115 volatile u32 preempt_val;
116 u64 saved_rip;
117 
118 static int preemption_timer_init(struct vmcs *vmcs)
119 {
120 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
121 		printf("\tPreemption timer is not supported\n");
122 		return VMX_TEST_EXIT;
123 	}
124 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
125 	preempt_val = 10000000;
126 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
127 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
128 
129 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
130 		printf("\tSave preemption value is not supported\n");
131 
132 	return VMX_TEST_START;
133 }
134 
135 static void preemption_timer_main(void)
136 {
137 	tsc_val = rdtsc();
138 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
139 		vmx_set_test_stage(0);
140 		vmcall();
141 		if (vmx_get_test_stage() == 1)
142 			vmcall();
143 	}
144 	vmx_set_test_stage(1);
145 	while (vmx_get_test_stage() == 1) {
146 		if (((rdtsc() - tsc_val) >> preempt_scale)
147 				> 10 * preempt_val) {
148 			vmx_set_test_stage(2);
149 			vmcall();
150 		}
151 	}
152 	tsc_val = rdtsc();
153 	asm volatile ("hlt");
154 	vmcall();
155 	vmx_set_test_stage(5);
156 	vmcall();
157 }
158 
159 static int preemption_timer_exit_handler(union exit_reason exit_reason)
160 {
161 	bool guest_halted;
162 	u64 guest_rip;
163 	u32 insn_len;
164 	u32 ctrl_exit;
165 
166 	guest_rip = vmcs_read(GUEST_RIP);
167 	insn_len = vmcs_read(EXI_INST_LEN);
168 	switch (exit_reason.basic) {
169 	case VMX_PREEMPT:
170 		switch (vmx_get_test_stage()) {
171 		case 1:
172 		case 2:
173 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val,
174 			       "busy-wait for preemption timer");
175 			vmx_set_test_stage(3);
176 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
177 			return VMX_TEST_RESUME;
178 		case 3:
179 			guest_halted =
180 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
181 			report(((rdtsc() - tsc_val) >> preempt_scale) >= preempt_val
182 			        && guest_halted,
183 			       "preemption timer during hlt");
184 			vmx_set_test_stage(4);
185 			vmcs_write(PIN_CONTROLS,
186 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
187 			vmcs_write(EXI_CONTROLS,
188 				   vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT);
189 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
190 			return VMX_TEST_RESUME;
191 		case 4:
192 			report(saved_rip == guest_rip,
193 			       "preemption timer with 0 value");
194 			break;
195 		default:
196 			report_fail("Invalid stage.");
197 			print_vmexit_info(exit_reason);
198 			break;
199 		}
200 		break;
201 	case VMX_VMCALL:
202 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
203 		switch (vmx_get_test_stage()) {
204 		case 0:
205 			report(vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val,
206 			       "Keep preemption value");
207 			vmx_set_test_stage(1);
208 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
209 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
210 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
211 			vmcs_write(EXI_CONTROLS, ctrl_exit);
212 			return VMX_TEST_RESUME;
213 		case 1:
214 			report(vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val,
215 			       "Save preemption value");
216 			return VMX_TEST_RESUME;
217 		case 2:
218 			report_fail("busy-wait for preemption timer");
219 			vmx_set_test_stage(3);
220 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
221 			return VMX_TEST_RESUME;
222 		case 3:
223 			report_fail("preemption timer during hlt");
224 			vmx_set_test_stage(4);
225 			/* fall through */
226 		case 4:
227 			vmcs_write(PIN_CONTROLS,
228 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
229 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
230 			saved_rip = guest_rip + insn_len;
231 			return VMX_TEST_RESUME;
232 		case 5:
233 			report_fail("preemption timer with 0 value (vmcall stage 5)");
234 			break;
235 		default:
236 			// Should not reach here
237 			report_fail("unexpected stage, %d",
238 				    vmx_get_test_stage());
239 			print_vmexit_info(exit_reason);
240 			return VMX_TEST_VMEXIT;
241 		}
242 		break;
243 	default:
244 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
245 		print_vmexit_info(exit_reason);
246 	}
247 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
248 	return VMX_TEST_VMEXIT;
249 }
250 
251 static void msr_bmp_init(void)
252 {
253 	void *msr_bitmap;
254 	u32 ctrl_cpu0;
255 
256 	msr_bitmap = alloc_page();
257 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
258 	ctrl_cpu0 |= CPU_MSR_BITMAP;
259 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
260 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
261 }
262 
263 static void *get_msr_bitmap(void)
264 {
265 	void *msr_bitmap;
266 
267 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
268 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
269 	} else {
270 		msr_bitmap = alloc_page();
271 		memset(msr_bitmap, 0xff, PAGE_SIZE);
272 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
273 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
274 	}
275 
276 	return msr_bitmap;
277 }
278 
279 static void disable_intercept_for_x2apic_msrs(void)
280 {
281 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
282 	u32 msr;
283 
284 	for (msr = APIC_BASE_MSR;
285 		 msr < (APIC_BASE_MSR+0xff);
286 		 msr += BITS_PER_LONG) {
287 		unsigned int word = msr / BITS_PER_LONG;
288 
289 		msr_bitmap[word] = 0;
290 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
291 	}
292 }
293 
294 static int test_ctrl_pat_init(struct vmcs *vmcs)
295 {
296 	u64 ctrl_ent;
297 	u64 ctrl_exi;
298 
299 	msr_bmp_init();
300 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
301 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
302 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
303 		printf("\tSave/load PAT is not supported\n");
304 		return 1;
305 	}
306 
307 	ctrl_ent = vmcs_read(ENT_CONTROLS);
308 	ctrl_exi = vmcs_read(EXI_CONTROLS);
309 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
310 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
311 	vmcs_write(ENT_CONTROLS, ctrl_ent);
312 	vmcs_write(EXI_CONTROLS, ctrl_exi);
313 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
314 	vmcs_write(GUEST_PAT, 0x0);
315 	vmcs_write(HOST_PAT, ia32_pat);
316 	return VMX_TEST_START;
317 }
318 
319 static void test_ctrl_pat_main(void)
320 {
321 	u64 guest_ia32_pat;
322 
323 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
324 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
325 		printf("\tENT_LOAD_PAT is not supported.\n");
326 	else {
327 		if (guest_ia32_pat != 0) {
328 			report_fail("Entry load PAT");
329 			return;
330 		}
331 	}
332 	wrmsr(MSR_IA32_CR_PAT, 0x6);
333 	vmcall();
334 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
335 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
336 		report(guest_ia32_pat == ia32_pat, "Entry load PAT");
337 }
338 
339 static int test_ctrl_pat_exit_handler(union exit_reason exit_reason)
340 {
341 	u64 guest_rip;
342 	u64 guest_pat;
343 
344 	guest_rip = vmcs_read(GUEST_RIP);
345 	switch (exit_reason.basic) {
346 	case VMX_VMCALL:
347 		guest_pat = vmcs_read(GUEST_PAT);
348 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
349 			printf("\tEXI_SAVE_PAT is not supported\n");
350 			vmcs_write(GUEST_PAT, 0x6);
351 		} else {
352 			report(guest_pat == 0x6, "Exit save PAT");
353 		}
354 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
355 			printf("\tEXI_LOAD_PAT is not supported\n");
356 		else
357 			report(rdmsr(MSR_IA32_CR_PAT) == ia32_pat,
358 			       "Exit load PAT");
359 		vmcs_write(GUEST_PAT, ia32_pat);
360 		vmcs_write(GUEST_RIP, guest_rip + 3);
361 		return VMX_TEST_RESUME;
362 	default:
363 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
364 		break;
365 	}
366 	return VMX_TEST_VMEXIT;
367 }
368 
369 static int test_ctrl_efer_init(struct vmcs *vmcs)
370 {
371 	u64 ctrl_ent;
372 	u64 ctrl_exi;
373 
374 	msr_bmp_init();
375 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
376 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
377 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
378 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
379 	ia32_efer = rdmsr(MSR_EFER);
380 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
381 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
382 	return VMX_TEST_START;
383 }
384 
385 static void test_ctrl_efer_main(void)
386 {
387 	u64 guest_ia32_efer;
388 
389 	guest_ia32_efer = rdmsr(MSR_EFER);
390 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
391 		printf("\tENT_LOAD_EFER is not supported.\n");
392 	else {
393 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
394 			report_fail("Entry load EFER");
395 			return;
396 		}
397 	}
398 	wrmsr(MSR_EFER, ia32_efer);
399 	vmcall();
400 	guest_ia32_efer = rdmsr(MSR_EFER);
401 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
402 		report(guest_ia32_efer == ia32_efer, "Entry load EFER");
403 }
404 
405 static int test_ctrl_efer_exit_handler(union exit_reason exit_reason)
406 {
407 	u64 guest_rip;
408 	u64 guest_efer;
409 
410 	guest_rip = vmcs_read(GUEST_RIP);
411 	switch (exit_reason.basic) {
412 	case VMX_VMCALL:
413 		guest_efer = vmcs_read(GUEST_EFER);
414 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
415 			printf("\tEXI_SAVE_EFER is not supported\n");
416 			vmcs_write(GUEST_EFER, ia32_efer);
417 		} else {
418 			report(guest_efer == ia32_efer, "Exit save EFER");
419 		}
420 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
421 			printf("\tEXI_LOAD_EFER is not supported\n");
422 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
423 		} else {
424 			report(rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX),
425 			       "Exit load EFER");
426 		}
427 		vmcs_write(GUEST_PAT, ia32_efer);
428 		vmcs_write(GUEST_RIP, guest_rip + 3);
429 		return VMX_TEST_RESUME;
430 	default:
431 		printf("ERROR : Unknown exit reason, 0x%x.\n", exit_reason.full);
432 		break;
433 	}
434 	return VMX_TEST_VMEXIT;
435 }
436 
437 u32 guest_cr0, guest_cr4;
438 
439 static void cr_shadowing_main(void)
440 {
441 	u32 cr0, cr4, tmp;
442 
443 	// Test read through
444 	vmx_set_test_stage(0);
445 	guest_cr0 = read_cr0();
446 	if (vmx_get_test_stage() == 1)
447 		report_fail("Read through CR0");
448 	else
449 		vmcall();
450 	vmx_set_test_stage(1);
451 	guest_cr4 = read_cr4();
452 	if (vmx_get_test_stage() == 2)
453 		report_fail("Read through CR4");
454 	else
455 		vmcall();
456 	// Test write through
457 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
458 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
459 	vmx_set_test_stage(2);
460 	write_cr0(guest_cr0);
461 	if (vmx_get_test_stage() == 3)
462 		report_fail("Write through CR0");
463 	else
464 		vmcall();
465 	vmx_set_test_stage(3);
466 	write_cr4(guest_cr4);
467 	if (vmx_get_test_stage() == 4)
468 		report_fail("Write through CR4");
469 	else
470 		vmcall();
471 	// Test read shadow
472 	vmx_set_test_stage(4);
473 	vmcall();
474 	cr0 = read_cr0();
475 	if (vmx_get_test_stage() != 5)
476 		report(cr0 == guest_cr0, "Read shadowing CR0");
477 	vmx_set_test_stage(5);
478 	cr4 = read_cr4();
479 	if (vmx_get_test_stage() != 6)
480 		report(cr4 == guest_cr4, "Read shadowing CR4");
481 	// Test write shadow (same value with shadow)
482 	vmx_set_test_stage(6);
483 	write_cr0(guest_cr0);
484 	if (vmx_get_test_stage() == 7)
485 		report_fail("Write shadowing CR0 (same value with shadow)");
486 	else
487 		vmcall();
488 	vmx_set_test_stage(7);
489 	write_cr4(guest_cr4);
490 	if (vmx_get_test_stage() == 8)
491 		report_fail("Write shadowing CR4 (same value with shadow)");
492 	else
493 		vmcall();
494 	// Test write shadow (different value)
495 	vmx_set_test_stage(8);
496 	tmp = guest_cr0 ^ X86_CR0_TS;
497 	asm volatile("mov %0, %%rsi\n\t"
498 		"mov %%rsi, %%cr0\n\t"
499 		::"m"(tmp)
500 		:"rsi", "memory", "cc");
501 	report(vmx_get_test_stage() == 9,
502 	       "Write shadowing different X86_CR0_TS");
503 	vmx_set_test_stage(9);
504 	tmp = guest_cr0 ^ X86_CR0_MP;
505 	asm volatile("mov %0, %%rsi\n\t"
506 		"mov %%rsi, %%cr0\n\t"
507 		::"m"(tmp)
508 		:"rsi", "memory", "cc");
509 	report(vmx_get_test_stage() == 10,
510 	       "Write shadowing different X86_CR0_MP");
511 	vmx_set_test_stage(10);
512 	tmp = guest_cr4 ^ X86_CR4_TSD;
513 	asm volatile("mov %0, %%rsi\n\t"
514 		"mov %%rsi, %%cr4\n\t"
515 		::"m"(tmp)
516 		:"rsi", "memory", "cc");
517 	report(vmx_get_test_stage() == 11,
518 	       "Write shadowing different X86_CR4_TSD");
519 	vmx_set_test_stage(11);
520 	tmp = guest_cr4 ^ X86_CR4_DE;
521 	asm volatile("mov %0, %%rsi\n\t"
522 		"mov %%rsi, %%cr4\n\t"
523 		::"m"(tmp)
524 		:"rsi", "memory", "cc");
525 	report(vmx_get_test_stage() == 12,
526 	       "Write shadowing different X86_CR4_DE");
527 }
528 
529 static int cr_shadowing_exit_handler(union exit_reason exit_reason)
530 {
531 	u64 guest_rip;
532 	u32 insn_len;
533 	u32 exit_qual;
534 
535 	guest_rip = vmcs_read(GUEST_RIP);
536 	insn_len = vmcs_read(EXI_INST_LEN);
537 	exit_qual = vmcs_read(EXI_QUALIFICATION);
538 	switch (exit_reason.basic) {
539 	case VMX_VMCALL:
540 		switch (vmx_get_test_stage()) {
541 		case 0:
542 			report(guest_cr0 == vmcs_read(GUEST_CR0),
543 			       "Read through CR0");
544 			break;
545 		case 1:
546 			report(guest_cr4 == vmcs_read(GUEST_CR4),
547 			       "Read through CR4");
548 			break;
549 		case 2:
550 			report(guest_cr0 == vmcs_read(GUEST_CR0),
551 			       "Write through CR0");
552 			break;
553 		case 3:
554 			report(guest_cr4 == vmcs_read(GUEST_CR4),
555 			       "Write through CR4");
556 			break;
557 		case 4:
558 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
559 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
560 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
561 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
562 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
563 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
564 			break;
565 		case 6:
566 			report(guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)),
567 			       "Write shadowing CR0 (same value)");
568 			break;
569 		case 7:
570 			report(guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)),
571 			       "Write shadowing CR4 (same value)");
572 			break;
573 		default:
574 			// Should not reach here
575 			report_fail("unexpected stage, %d",
576 				    vmx_get_test_stage());
577 			print_vmexit_info(exit_reason);
578 			return VMX_TEST_VMEXIT;
579 		}
580 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
581 		return VMX_TEST_RESUME;
582 	case VMX_CR:
583 		switch (vmx_get_test_stage()) {
584 		case 4:
585 			report_fail("Read shadowing CR0");
586 			vmx_inc_test_stage();
587 			break;
588 		case 5:
589 			report_fail("Read shadowing CR4");
590 			vmx_inc_test_stage();
591 			break;
592 		case 6:
593 			report_fail("Write shadowing CR0 (same value)");
594 			vmx_inc_test_stage();
595 			break;
596 		case 7:
597 			report_fail("Write shadowing CR4 (same value)");
598 			vmx_inc_test_stage();
599 			break;
600 		case 8:
601 		case 9:
602 			// 0x600 encodes "mov %esi, %cr0"
603 			if (exit_qual == 0x600)
604 				vmx_inc_test_stage();
605 			break;
606 		case 10:
607 		case 11:
608 			// 0x604 encodes "mov %esi, %cr4"
609 			if (exit_qual == 0x604)
610 				vmx_inc_test_stage();
611 			break;
612 		default:
613 			// Should not reach here
614 			report_fail("unexpected stage, %d",
615 				    vmx_get_test_stage());
616 			print_vmexit_info(exit_reason);
617 			return VMX_TEST_VMEXIT;
618 		}
619 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
620 		return VMX_TEST_RESUME;
621 	default:
622 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
623 		print_vmexit_info(exit_reason);
624 	}
625 	return VMX_TEST_VMEXIT;
626 }
627 
628 static int iobmp_init(struct vmcs *vmcs)
629 {
630 	u32 ctrl_cpu0;
631 
632 	io_bitmap_a = alloc_page();
633 	io_bitmap_b = alloc_page();
634 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
635 	ctrl_cpu0 |= CPU_IO_BITMAP;
636 	ctrl_cpu0 &= (~CPU_IO);
637 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
638 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
639 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
640 	return VMX_TEST_START;
641 }
642 
643 static void iobmp_main(void)
644 {
645 	// stage 0, test IO pass
646 	vmx_set_test_stage(0);
647 	inb(0x5000);
648 	outb(0x0, 0x5000);
649 	report(vmx_get_test_stage() == 0, "I/O bitmap - I/O pass");
650 	// test IO width, in/out
651 	((u8 *)io_bitmap_a)[0] = 0xFF;
652 	vmx_set_test_stage(2);
653 	inb(0x0);
654 	report(vmx_get_test_stage() == 3, "I/O bitmap - trap in");
655 	vmx_set_test_stage(3);
656 	outw(0x0, 0x0);
657 	report(vmx_get_test_stage() == 4, "I/O bitmap - trap out");
658 	vmx_set_test_stage(4);
659 	inl(0x0);
660 	report(vmx_get_test_stage() == 5, "I/O bitmap - I/O width, long");
661 	// test low/high IO port
662 	vmx_set_test_stage(5);
663 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
664 	inb(0x5000);
665 	report(vmx_get_test_stage() == 6, "I/O bitmap - I/O port, low part");
666 	vmx_set_test_stage(6);
667 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
668 	inb(0x9000);
669 	report(vmx_get_test_stage() == 7, "I/O bitmap - I/O port, high part");
670 	// test partial pass
671 	vmx_set_test_stage(7);
672 	inl(0x4FFF);
673 	report(vmx_get_test_stage() == 8, "I/O bitmap - partial pass");
674 	// test overrun
675 	vmx_set_test_stage(8);
676 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
677 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
678 	inl(0xFFFF);
679 	report(vmx_get_test_stage() == 9, "I/O bitmap - overrun");
680 	vmx_set_test_stage(9);
681 	vmcall();
682 	outb(0x0, 0x0);
683 	report(vmx_get_test_stage() == 9,
684 	       "I/O bitmap - ignore unconditional exiting");
685 	vmx_set_test_stage(10);
686 	vmcall();
687 	outb(0x0, 0x0);
688 	report(vmx_get_test_stage() == 11,
689 	       "I/O bitmap - unconditional exiting");
690 }
691 
692 static int iobmp_exit_handler(union exit_reason exit_reason)
693 {
694 	u64 guest_rip;
695 	ulong exit_qual;
696 	u32 insn_len, ctrl_cpu0;
697 
698 	guest_rip = vmcs_read(GUEST_RIP);
699 	exit_qual = vmcs_read(EXI_QUALIFICATION);
700 	insn_len = vmcs_read(EXI_INST_LEN);
701 	switch (exit_reason.basic) {
702 	case VMX_IO:
703 		switch (vmx_get_test_stage()) {
704 		case 0:
705 		case 1:
706 			vmx_inc_test_stage();
707 			break;
708 		case 2:
709 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE,
710 			       "I/O bitmap - I/O width, byte");
711 			report(exit_qual & VMX_IO_IN,
712 			       "I/O bitmap - I/O direction, in");
713 			vmx_inc_test_stage();
714 			break;
715 		case 3:
716 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD,
717 			       "I/O bitmap - I/O width, word");
718 			report(!(exit_qual & VMX_IO_IN),
719 			       "I/O bitmap - I/O direction, out");
720 			vmx_inc_test_stage();
721 			break;
722 		case 4:
723 			report((exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG,
724 			       "I/O bitmap - I/O width, long");
725 			vmx_inc_test_stage();
726 			break;
727 		case 5:
728 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
729 				vmx_inc_test_stage();
730 			break;
731 		case 6:
732 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
733 				vmx_inc_test_stage();
734 			break;
735 		case 7:
736 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
737 				vmx_inc_test_stage();
738 			break;
739 		case 8:
740 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
741 				vmx_inc_test_stage();
742 			break;
743 		case 9:
744 		case 10:
745 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
746 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
747 			vmx_inc_test_stage();
748 			break;
749 		default:
750 			// Should not reach here
751 			report_fail("unexpected stage, %d",
752 				    vmx_get_test_stage());
753 			print_vmexit_info(exit_reason);
754 			return VMX_TEST_VMEXIT;
755 		}
756 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
757 		return VMX_TEST_RESUME;
758 	case VMX_VMCALL:
759 		switch (vmx_get_test_stage()) {
760 		case 9:
761 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
762 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
763 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
764 			break;
765 		case 10:
766 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
767 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
768 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
769 			break;
770 		default:
771 			// Should not reach here
772 			report_fail("unexpected stage, %d",
773 				    vmx_get_test_stage());
774 			print_vmexit_info(exit_reason);
775 			return VMX_TEST_VMEXIT;
776 		}
777 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
778 		return VMX_TEST_RESUME;
779 	default:
780 		printf("guest_rip = %#lx\n", guest_rip);
781 		printf("\tERROR : Unknown exit reason, 0x%x\n", exit_reason.full);
782 		break;
783 	}
784 	return VMX_TEST_VMEXIT;
785 }
786 
787 #define INSN_CPU0		0
788 #define INSN_CPU1		1
789 #define INSN_ALWAYS_TRAP	2
790 
791 #define FIELD_EXIT_QUAL		(1 << 0)
792 #define FIELD_INSN_INFO		(1 << 1)
793 
794 asm(
795 	"insn_hlt: hlt;ret\n\t"
796 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
797 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
798 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
799 	"insn_rdtsc: rdtsc;ret\n\t"
800 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
801 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
802 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
803 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
804 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
805 	"insn_pause: pause;ret\n\t"
806 	"insn_wbinvd: wbinvd;ret\n\t"
807 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
808 	"insn_invd: invd;ret\n\t"
809 	"insn_sgdt: sgdt gdt_descr;ret\n\t"
810 	"insn_lgdt: lgdt gdt_descr;ret\n\t"
811 	"insn_sidt: sidt idt_descr;ret\n\t"
812 	"insn_lidt: lidt idt_descr;ret\n\t"
813 	"insn_sldt: sldt %ax;ret\n\t"
814 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
815 	"insn_str: str %ax;ret\n\t"
816 	"insn_rdrand: rdrand %rax;ret\n\t"
817 	"insn_rdseed: rdseed %rax;ret\n\t"
818 );
819 extern void insn_hlt(void);
820 extern void insn_invlpg(void);
821 extern void insn_mwait(void);
822 extern void insn_rdpmc(void);
823 extern void insn_rdtsc(void);
824 extern void insn_cr3_load(void);
825 extern void insn_cr3_store(void);
826 extern void insn_cr8_load(void);
827 extern void insn_cr8_store(void);
828 extern void insn_monitor(void);
829 extern void insn_pause(void);
830 extern void insn_wbinvd(void);
831 extern void insn_sgdt(void);
832 extern void insn_lgdt(void);
833 extern void insn_sidt(void);
834 extern void insn_lidt(void);
835 extern void insn_sldt(void);
836 extern void insn_lldt(void);
837 extern void insn_str(void);
838 extern void insn_cpuid(void);
839 extern void insn_invd(void);
840 extern void insn_rdrand(void);
841 extern void insn_rdseed(void);
842 
843 u32 cur_insn;
844 u64 cr3;
845 
846 #define X86_FEATURE_MONITOR	(1 << 3)
847 
848 typedef bool (*supported_fn)(void);
849 
850 static bool monitor_supported(void)
851 {
852 	return this_cpu_has(X86_FEATURE_MWAIT);
853 }
854 
855 struct insn_table {
856 	const char *name;
857 	u32 flag;
858 	void (*insn_func)(void);
859 	u32 type;
860 	u32 reason;
861 	ulong exit_qual;
862 	u32 insn_info;
863 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
864 	// which field need to be tested, reason is always tested
865 	u32 test_field;
866 	const supported_fn supported_fn;
867 	u8 disabled;
868 };
869 
870 /*
871  * Add more test cases of instruction intercept here. Elements in this
872  * table is:
873  *	name/control flag/insn function/type/exit reason/exit qulification/
874  *	instruction info/field to test
875  * The last field defines which fields (exit_qual and insn_info) need to be
876  * tested in exit handler. If set to 0, only "reason" is checked.
877  */
878 static struct insn_table insn_table[] = {
879 	// Flags for Primary Processor-Based VM-Execution Controls
880 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
881 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
882 		0x12345678, 0, FIELD_EXIT_QUAL},
883 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0, &monitor_supported},
884 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0},
885 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
886 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
887 		FIELD_EXIT_QUAL},
888 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
889 		FIELD_EXIT_QUAL},
890 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
891 		FIELD_EXIT_QUAL},
892 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
893 		FIELD_EXIT_QUAL},
894 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0, &monitor_supported},
895 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
896 	// Flags for Secondary Processor-Based VM-Execution Controls
897 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
898 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
899 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
900 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
901 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
902 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
903 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
904 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
905 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
906 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
907 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
908 	// Instructions always trap
909 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
910 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
911 	// Instructions never trap
912 	{NULL},
913 };
914 
915 static int insn_intercept_init(struct vmcs *vmcs)
916 {
917 	u32 ctrl_cpu, cur_insn;
918 
919 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
920 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
921 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
922 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
923 	cr3 = read_cr3();
924 
925 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
926 		if (insn_table[cur_insn].supported_fn == NULL)
927 			continue;
928 		insn_table[cur_insn].disabled = !insn_table[cur_insn].supported_fn();
929 	}
930 	return VMX_TEST_START;
931 }
932 
933 static void insn_intercept_main(void)
934 {
935 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
936 		vmx_set_test_stage(cur_insn * 2);
937 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
938 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
939 		    (insn_table[cur_insn].type == INSN_CPU1 &&
940 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
941 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
942 			       insn_table[cur_insn].type - INSN_CPU0,
943 			       insn_table[cur_insn].name);
944 			continue;
945 		}
946 
947 		if (insn_table[cur_insn].disabled) {
948 			printf("\tFeature required for %s is not supported.\n",
949 			       insn_table[cur_insn].name);
950 			continue;
951 		}
952 
953 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
954 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
955 		    (insn_table[cur_insn].type == INSN_CPU1 &&
956 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
957 			/* skip hlt, it stalls the guest and is tested below */
958 			if (insn_table[cur_insn].insn_func != insn_hlt)
959 				insn_table[cur_insn].insn_func();
960 			report(vmx_get_test_stage() == cur_insn * 2,
961 					"execute %s",
962 					insn_table[cur_insn].name);
963 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
964 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
965 			       insn_table[cur_insn].type - INSN_CPU0,
966 			       insn_table[cur_insn].name);
967 
968 		vmcall();
969 
970 		insn_table[cur_insn].insn_func();
971 		report(vmx_get_test_stage() == cur_insn * 2 + 1,
972 				"intercept %s",
973 				insn_table[cur_insn].name);
974 
975 		vmx_set_test_stage(cur_insn * 2 + 1);
976 		vmcall();
977 	}
978 }
979 
980 static int insn_intercept_exit_handler(union exit_reason exit_reason)
981 {
982 	u64 guest_rip;
983 	ulong exit_qual;
984 	u32 insn_len;
985 	u32 insn_info;
986 	bool pass;
987 
988 	guest_rip = vmcs_read(GUEST_RIP);
989 	exit_qual = vmcs_read(EXI_QUALIFICATION);
990 	insn_len = vmcs_read(EXI_INST_LEN);
991 	insn_info = vmcs_read(EXI_INST_INFO);
992 
993 	if (exit_reason.basic == VMX_VMCALL) {
994 		u32 val = 0;
995 
996 		if (insn_table[cur_insn].type == INSN_CPU0)
997 			val = vmcs_read(CPU_EXEC_CTRL0);
998 		else if (insn_table[cur_insn].type == INSN_CPU1)
999 			val = vmcs_read(CPU_EXEC_CTRL1);
1000 
1001 		if (vmx_get_test_stage() & 1)
1002 			val &= ~insn_table[cur_insn].flag;
1003 		else
1004 			val |= insn_table[cur_insn].flag;
1005 
1006 		if (insn_table[cur_insn].type == INSN_CPU0)
1007 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
1008 		else if (insn_table[cur_insn].type == INSN_CPU1)
1009 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
1010 	} else {
1011 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
1012 			insn_table[cur_insn].reason == exit_reason.full;
1013 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
1014 		    insn_table[cur_insn].exit_qual != exit_qual)
1015 			pass = false;
1016 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
1017 		    insn_table[cur_insn].insn_info != insn_info)
1018 			pass = false;
1019 		if (pass)
1020 			vmx_inc_test_stage();
1021 	}
1022 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1023 	return VMX_TEST_RESUME;
1024 }
1025 
1026 /**
1027  * __setup_ept - Setup the VMCS fields to enable Extended Page Tables (EPT)
1028  * @hpa:	Host physical address of the top-level, a.k.a. root, EPT table
1029  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1030  *
1031  * Returns 0 on success, 1 on failure.
1032  *
1033  * Note that @hpa doesn't need to point at actual memory if VM-Launch is
1034  * expected to fail, e.g. setup_dummy_ept() arbitrarily passes '0' to satisfy
1035  * the various EPTP consistency checks, but doesn't ensure backing for HPA '0'.
1036  */
1037 static int __setup_ept(u64 hpa, bool enable_ad)
1038 {
1039 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1040 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1041 		printf("\tEPT is not supported\n");
1042 		return 1;
1043 	}
1044 	if (!(ept_vpid.val & EPT_CAP_WB)) {
1045 		printf("\tWB memtype for EPT walks not supported\n");
1046 		return 1;
1047 	}
1048 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
1049 		printf("\tPWL4 is not supported\n");
1050 		return 1;
1051 	}
1052 
1053 	eptp = EPT_MEM_TYPE_WB;
1054 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1055 	eptp |= hpa;
1056 	if (enable_ad)
1057 		eptp |= EPTP_AD_FLAG;
1058 
1059 	vmcs_write(EPTP, eptp);
1060 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY);
1061 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT);
1062 
1063 	return 0;
1064 }
1065 
1066 /**
1067  * setup_ept - Enable Extended Page Tables (EPT) and setup an identity map
1068  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1069  *
1070  * Returns 0 on success, 1 on failure.
1071  *
1072  * This is the "real" function for setting up EPT tables, i.e. use this for
1073  * tests that need to run code in the guest with EPT enabled.
1074  */
1075 static int setup_ept(bool enable_ad)
1076 {
1077 	unsigned long end_of_memory;
1078 
1079 	pml4 = alloc_page();
1080 
1081 	if (__setup_ept(virt_to_phys(pml4), enable_ad))
1082 		return 1;
1083 
1084 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1085 	if (end_of_memory < (1ul << 32))
1086 		end_of_memory = (1ul << 32);
1087 	/* Cannot use large EPT pages if we need to track EPT
1088 	 * accessed/dirty bits at 4K granularity.
1089 	 */
1090 	setup_ept_range(pml4, 0, end_of_memory, 0,
1091 			!enable_ad && ept_2m_supported(),
1092 			EPT_WA | EPT_RA | EPT_EA);
1093 	return 0;
1094 }
1095 
1096 /**
1097  * setup_dummy_ept - Enable Extended Page Tables (EPT) with a dummy root HPA
1098  *
1099  * Setup EPT using a semi-arbitrary dummy root HPA.  This function is intended
1100  * for use by tests that need EPT enabled to verify dependent VMCS controls
1101  * but never expect to fully enter the guest, i.e. don't need setup the actual
1102  * EPT tables.
1103  */
1104 static void setup_dummy_ept(void)
1105 {
1106 	if (__setup_ept(0, false))
1107 		report_abort("EPT setup unexpectedly failed");
1108 }
1109 
1110 static int enable_unrestricted_guest(bool need_valid_ept)
1111 {
1112 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1113 	    !(ctrl_cpu_rev[1].clr & CPU_URG) ||
1114 	    !(ctrl_cpu_rev[1].clr & CPU_EPT))
1115 		return 1;
1116 
1117 	if (need_valid_ept)
1118 		setup_ept(false);
1119 	else
1120 		setup_dummy_ept();
1121 
1122 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
1123 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
1124 
1125 	return 0;
1126 }
1127 
1128 static void ept_enable_ad_bits(void)
1129 {
1130 	eptp |= EPTP_AD_FLAG;
1131 	vmcs_write(EPTP, eptp);
1132 }
1133 
1134 static void ept_disable_ad_bits(void)
1135 {
1136 	eptp &= ~EPTP_AD_FLAG;
1137 	vmcs_write(EPTP, eptp);
1138 }
1139 
1140 static int ept_ad_enabled(void)
1141 {
1142 	return eptp & EPTP_AD_FLAG;
1143 }
1144 
1145 static void ept_enable_ad_bits_or_skip_test(void)
1146 {
1147 	if (!ept_ad_bits_supported())
1148 		test_skip("EPT AD bits not supported.");
1149 	ept_enable_ad_bits();
1150 }
1151 
1152 static int apic_version;
1153 
1154 static int ept_init_common(bool have_ad)
1155 {
1156 	int ret;
1157 	struct pci_dev pcidev;
1158 
1159 	/* INVEPT is required by the EPT violation handler. */
1160 	if (!is_invept_type_supported(INVEPT_SINGLE))
1161 		return VMX_TEST_EXIT;
1162 
1163 	if (setup_ept(have_ad))
1164 		return VMX_TEST_EXIT;
1165 
1166 	data_page1 = alloc_page();
1167 	data_page2 = alloc_page();
1168 	*((u32 *)data_page1) = MAGIC_VAL_1;
1169 	*((u32 *)data_page2) = MAGIC_VAL_2;
1170 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1171 			EPT_RA | EPT_WA | EPT_EA);
1172 
1173 	apic_version = apic_read(APIC_LVR);
1174 
1175 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1176 	if (ret != PCIDEVADDR_INVALID) {
1177 		pci_dev_init(&pcidev, ret);
1178 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1179 	}
1180 
1181 	return VMX_TEST_START;
1182 }
1183 
1184 static int ept_init(struct vmcs *vmcs)
1185 {
1186 	return ept_init_common(false);
1187 }
1188 
1189 static void ept_common(void)
1190 {
1191 	vmx_set_test_stage(0);
1192 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1193 			*((u32 *)data_page1) != MAGIC_VAL_1)
1194 		report_fail("EPT basic framework - read");
1195 	else {
1196 		*((u32 *)data_page2) = MAGIC_VAL_3;
1197 		vmcall();
1198 		if (vmx_get_test_stage() == 1) {
1199 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1200 					*((u32 *)data_page2) == MAGIC_VAL_2)
1201 				report_pass("EPT basic framework");
1202 			else
1203 				report_pass("EPT basic framework - remap");
1204 		}
1205 	}
1206 	// Test EPT Misconfigurations
1207 	vmx_set_test_stage(1);
1208 	vmcall();
1209 	*((u32 *)data_page1) = MAGIC_VAL_1;
1210 	if (vmx_get_test_stage() != 2) {
1211 		report_fail("EPT misconfigurations");
1212 		goto t1;
1213 	}
1214 	vmx_set_test_stage(2);
1215 	vmcall();
1216 	*((u32 *)data_page1) = MAGIC_VAL_1;
1217 	report(vmx_get_test_stage() == 3, "EPT misconfigurations");
1218 t1:
1219 	// Test EPT violation
1220 	vmx_set_test_stage(3);
1221 	vmcall();
1222 	*((u32 *)data_page1) = MAGIC_VAL_1;
1223 	report(vmx_get_test_stage() == 4, "EPT violation - page permission");
1224 	// Violation caused by EPT paging structure
1225 	vmx_set_test_stage(4);
1226 	vmcall();
1227 	*((u32 *)data_page1) = MAGIC_VAL_2;
1228 	report(vmx_get_test_stage() == 5, "EPT violation - paging structure");
1229 
1230 	// MMIO Read/Write
1231 	vmx_set_test_stage(5);
1232 	vmcall();
1233 
1234 	*(u32 volatile *)pci_physaddr;
1235 	report(vmx_get_test_stage() == 6, "MMIO EPT violation - read");
1236 
1237 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1238 	report(vmx_get_test_stage() == 7, "MMIO EPT violation - write");
1239 }
1240 
1241 static void ept_main(void)
1242 {
1243 	ept_common();
1244 
1245 	// Test EPT access to L1 MMIO
1246 	vmx_set_test_stage(7);
1247 	report(*((u32 *)0xfee00030UL) == apic_version, "EPT - MMIO access");
1248 
1249 	// Test invalid operand for INVEPT
1250 	vmcall();
1251 	report(vmx_get_test_stage() == 8, "EPT - unsupported INVEPT");
1252 }
1253 
1254 static bool invept_test(int type, u64 eptp)
1255 {
1256 	bool ret, supported;
1257 
1258 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1259 	ret = __invept(type, eptp);
1260 
1261 	if (ret == !supported)
1262 		return false;
1263 
1264 	if (!supported)
1265 		printf("WARNING: unsupported invept passed!\n");
1266 	else
1267 		printf("WARNING: invept failed!\n");
1268 
1269 	return true;
1270 }
1271 
1272 static int pml_exit_handler(union exit_reason exit_reason)
1273 {
1274 	u16 index, count;
1275 	u64 *pmlbuf = pml_log;
1276 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1277 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1278 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1279 
1280 	switch (exit_reason.basic) {
1281 	case VMX_VMCALL:
1282 		switch (vmx_get_test_stage()) {
1283 		case 0:
1284 			index = vmcs_read(GUEST_PML_INDEX);
1285 			for (count = index + 1; count < PML_INDEX; count++) {
1286 				if (pmlbuf[count] == (u64)data_page2) {
1287 					vmx_inc_test_stage();
1288 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1289 					break;
1290 				}
1291 			}
1292 			break;
1293 		case 1:
1294 			index = vmcs_read(GUEST_PML_INDEX);
1295 			/* Keep clearing the dirty bit till a overflow */
1296 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1297 			break;
1298 		default:
1299 			report_fail("unexpected stage, %d.",
1300 			       vmx_get_test_stage());
1301 			print_vmexit_info(exit_reason);
1302 			return VMX_TEST_VMEXIT;
1303 		}
1304 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1305 		return VMX_TEST_RESUME;
1306 	case VMX_PML_FULL:
1307 		vmx_inc_test_stage();
1308 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1309 		return VMX_TEST_RESUME;
1310 	default:
1311 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1312 		print_vmexit_info(exit_reason);
1313 	}
1314 	return VMX_TEST_VMEXIT;
1315 }
1316 
1317 static int ept_exit_handler_common(union exit_reason exit_reason, bool have_ad)
1318 {
1319 	u64 guest_rip;
1320 	u64 guest_cr3;
1321 	u32 insn_len;
1322 	u32 exit_qual;
1323 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte,
1324 			     guest_pte_addr;
1325 
1326 	guest_rip = vmcs_read(GUEST_RIP);
1327 	guest_cr3 = vmcs_read(GUEST_CR3);
1328 	insn_len = vmcs_read(EXI_INST_LEN);
1329 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1330 	pteval_t *ptep;
1331 	switch (exit_reason.basic) {
1332 	case VMX_VMCALL:
1333 		switch (vmx_get_test_stage()) {
1334 		case 0:
1335 			check_ept_ad(pml4, guest_cr3,
1336 				     (unsigned long)data_page1,
1337 				     have_ad ? EPT_ACCESS_FLAG : 0,
1338 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1339 			check_ept_ad(pml4, guest_cr3,
1340 				     (unsigned long)data_page2,
1341 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1342 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1343 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1344 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1345 			if (have_ad)
1346 				invept(INVEPT_SINGLE, eptp);
1347 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1348 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1349 				vmx_inc_test_stage();
1350 				install_ept(pml4, (unsigned long)data_page2,
1351 						(unsigned long)data_page2,
1352 						EPT_RA | EPT_WA | EPT_EA);
1353 			} else
1354 				report_fail("EPT basic framework - write");
1355 			break;
1356 		case 1:
1357 			install_ept(pml4, (unsigned long)data_page1,
1358  				(unsigned long)data_page1, EPT_WA);
1359 			invept(INVEPT_SINGLE, eptp);
1360 			break;
1361 		case 2:
1362 			install_ept(pml4, (unsigned long)data_page1,
1363  				(unsigned long)data_page1,
1364  				EPT_RA | EPT_WA | EPT_EA |
1365  				(2 << EPT_MEM_TYPE_SHIFT));
1366 			invept(INVEPT_SINGLE, eptp);
1367 			break;
1368 		case 3:
1369 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1370 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1371 						1, &data_page1_pte));
1372 			set_ept_pte(pml4, (unsigned long)data_page1,
1373 				1, data_page1_pte & ~EPT_PRESENT);
1374 			invept(INVEPT_SINGLE, eptp);
1375 			break;
1376 		case 4:
1377 			ptep = get_pte_level((pgd_t *)guest_cr3, data_page1, /*level=*/2);
1378 			guest_pte_addr = virt_to_phys(ptep) & PAGE_MASK;
1379 
1380 			TEST_ASSERT(get_ept_pte(pml4, guest_pte_addr, 2, &data_page1_pte_pte));
1381 			set_ept_pte(pml4, guest_pte_addr, 2,
1382 				data_page1_pte_pte & ~EPT_PRESENT);
1383 			invept(INVEPT_SINGLE, eptp);
1384 			break;
1385 		case 5:
1386 			install_ept(pml4, (unsigned long)pci_physaddr,
1387 				(unsigned long)pci_physaddr, 0);
1388 			invept(INVEPT_SINGLE, eptp);
1389 			break;
1390 		case 7:
1391 			if (!invept_test(0, eptp))
1392 				vmx_inc_test_stage();
1393 			break;
1394 		// Should not reach here
1395 		default:
1396 			report_fail("ERROR - unexpected stage, %d.",
1397 			       vmx_get_test_stage());
1398 			print_vmexit_info(exit_reason);
1399 			return VMX_TEST_VMEXIT;
1400 		}
1401 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1402 		return VMX_TEST_RESUME;
1403 	case VMX_EPT_MISCONFIG:
1404 		switch (vmx_get_test_stage()) {
1405 		case 1:
1406 		case 2:
1407 			vmx_inc_test_stage();
1408 			install_ept(pml4, (unsigned long)data_page1,
1409  				(unsigned long)data_page1,
1410  				EPT_RA | EPT_WA | EPT_EA);
1411 			invept(INVEPT_SINGLE, eptp);
1412 			break;
1413 		// Should not reach here
1414 		default:
1415 			report_fail("ERROR - unexpected stage, %d.",
1416 			       vmx_get_test_stage());
1417 			print_vmexit_info(exit_reason);
1418 			return VMX_TEST_VMEXIT;
1419 		}
1420 		return VMX_TEST_RESUME;
1421 	case VMX_EPT_VIOLATION:
1422 		/*
1423 		 * Exit-qualifications are masked not to account for advanced
1424 		 * VM-exit information. Once KVM supports this feature, this
1425 		 * masking should be removed.
1426 		 */
1427 		exit_qual &= ~EPT_VLT_GUEST_MASK;
1428 
1429 		switch(vmx_get_test_stage()) {
1430 		case 3:
1431 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1432 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1433 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1434 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1435 					EPT_VLT_PADDR))
1436 				vmx_inc_test_stage();
1437 			set_ept_pte(pml4, (unsigned long)data_page1,
1438 				1, data_page1_pte | (EPT_PRESENT));
1439 			invept(INVEPT_SINGLE, eptp);
1440 			break;
1441 		case 4:
1442 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1443 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1444 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1445 			if (exit_qual == (EPT_VLT_RD |
1446 					  (have_ad ? EPT_VLT_WR : 0) |
1447 					  EPT_VLT_LADDR_VLD))
1448 				vmx_inc_test_stage();
1449 			set_ept_pte(pml4, guest_pte_addr, 2,
1450 				data_page1_pte_pte | (EPT_PRESENT));
1451 			invept(INVEPT_SINGLE, eptp);
1452 			break;
1453 		case 5:
1454 			if (exit_qual & EPT_VLT_RD)
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);
1459 			invept(INVEPT_SINGLE, eptp);
1460 			break;
1461 		case 6:
1462 			if (exit_qual & EPT_VLT_WR)
1463 				vmx_inc_test_stage();
1464 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1465 						1, &memaddr_pte));
1466 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1467 			invept(INVEPT_SINGLE, eptp);
1468 			break;
1469 		default:
1470 			// Should not reach here
1471 			report_fail("ERROR : unexpected stage, %d",
1472 			       vmx_get_test_stage());
1473 			print_vmexit_info(exit_reason);
1474 			return VMX_TEST_VMEXIT;
1475 		}
1476 		return VMX_TEST_RESUME;
1477 	default:
1478 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1479 		print_vmexit_info(exit_reason);
1480 	}
1481 	return VMX_TEST_VMEXIT;
1482 }
1483 
1484 static int ept_exit_handler(union exit_reason exit_reason)
1485 {
1486 	return ept_exit_handler_common(exit_reason, false);
1487 }
1488 
1489 static int eptad_init(struct vmcs *vmcs)
1490 {
1491 	int r = ept_init_common(true);
1492 
1493 	if (r == VMX_TEST_EXIT)
1494 		return r;
1495 
1496 	if (!ept_ad_bits_supported()) {
1497 		printf("\tEPT A/D bits are not supported");
1498 		return VMX_TEST_EXIT;
1499 	}
1500 
1501 	return r;
1502 }
1503 
1504 static int pml_init(struct vmcs *vmcs)
1505 {
1506 	u32 ctrl_cpu;
1507 	int r = eptad_init(vmcs);
1508 
1509 	if (r == VMX_TEST_EXIT)
1510 		return r;
1511 
1512 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1513 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1514 		printf("\tPML is not supported");
1515 		return VMX_TEST_EXIT;
1516 	}
1517 
1518 	pml_log = alloc_page();
1519 	vmcs_write(PMLADDR, (u64)pml_log);
1520 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1521 
1522 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1523 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1524 
1525 	return VMX_TEST_START;
1526 }
1527 
1528 static void pml_main(void)
1529 {
1530 	int count = 0;
1531 
1532 	vmx_set_test_stage(0);
1533 	*((u32 *)data_page2) = 0x1;
1534 	vmcall();
1535 	report(vmx_get_test_stage() == 1, "PML - Dirty GPA Logging");
1536 
1537 	while (vmx_get_test_stage() == 1) {
1538 		vmcall();
1539 		*((u32 *)data_page2) = 0x1;
1540 		if (count++ > PML_INDEX)
1541 			break;
1542 	}
1543 	report(vmx_get_test_stage() == 2, "PML Full Event");
1544 }
1545 
1546 static void eptad_main(void)
1547 {
1548 	ept_common();
1549 }
1550 
1551 static int eptad_exit_handler(union exit_reason exit_reason)
1552 {
1553 	return ept_exit_handler_common(exit_reason, true);
1554 }
1555 
1556 #define TIMER_VECTOR	222
1557 
1558 static volatile bool timer_fired;
1559 
1560 static void timer_isr(isr_regs_t *regs)
1561 {
1562 	timer_fired = true;
1563 	apic_write(APIC_EOI, 0);
1564 }
1565 
1566 static int interrupt_init(struct vmcs *vmcs)
1567 {
1568 	msr_bmp_init();
1569 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1570 	handle_irq(TIMER_VECTOR, timer_isr);
1571 	return VMX_TEST_START;
1572 }
1573 
1574 static void interrupt_main(void)
1575 {
1576 	long long start, loops;
1577 
1578 	vmx_set_test_stage(0);
1579 
1580 	apic_write(APIC_LVTT, TIMER_VECTOR);
1581 	irq_enable();
1582 
1583 	apic_write(APIC_TMICT, 1);
1584 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1585 		asm volatile ("nop");
1586 	report(timer_fired, "direct interrupt while running guest");
1587 
1588 	apic_write(APIC_TMICT, 0);
1589 	irq_disable();
1590 	vmcall();
1591 	timer_fired = false;
1592 	apic_write(APIC_TMICT, 1);
1593 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1594 		asm volatile ("nop");
1595 	report(timer_fired, "intercepted interrupt while running guest");
1596 
1597 	irq_enable();
1598 	apic_write(APIC_TMICT, 0);
1599 	irq_disable();
1600 	vmcall();
1601 	timer_fired = false;
1602 	start = rdtsc();
1603 	apic_write(APIC_TMICT, 1000000);
1604 
1605 	safe_halt();
1606 
1607 	report(rdtsc() - start > 1000000 && timer_fired,
1608 	       "direct interrupt + hlt");
1609 
1610 	apic_write(APIC_TMICT, 0);
1611 	irq_disable();
1612 	vmcall();
1613 	timer_fired = false;
1614 	start = rdtsc();
1615 	apic_write(APIC_TMICT, 1000000);
1616 
1617 	safe_halt();
1618 
1619 	report(rdtsc() - start > 10000 && timer_fired,
1620 	       "intercepted interrupt + hlt");
1621 
1622 	apic_write(APIC_TMICT, 0);
1623 	irq_disable();
1624 	vmcall();
1625 	timer_fired = false;
1626 	start = rdtsc();
1627 	apic_write(APIC_TMICT, 1000000);
1628 
1629 	irq_enable();
1630 	asm volatile ("nop");
1631 	vmcall();
1632 
1633 	report(rdtsc() - start > 10000 && timer_fired,
1634 	       "direct interrupt + activity state hlt");
1635 
1636 	apic_write(APIC_TMICT, 0);
1637 	irq_disable();
1638 	vmcall();
1639 	timer_fired = false;
1640 	start = rdtsc();
1641 	apic_write(APIC_TMICT, 1000000);
1642 
1643 	irq_enable();
1644 	asm volatile ("nop");
1645 	vmcall();
1646 
1647 	report(rdtsc() - start > 10000 && timer_fired,
1648 	       "intercepted interrupt + activity state hlt");
1649 
1650 	apic_write(APIC_TMICT, 0);
1651 	irq_disable();
1652 	vmx_set_test_stage(7);
1653 	vmcall();
1654 	timer_fired = false;
1655 	apic_write(APIC_TMICT, 1);
1656 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1657 		asm volatile ("nop");
1658 	report(timer_fired,
1659 	       "running a guest with interrupt acknowledgement set");
1660 
1661 	apic_write(APIC_TMICT, 0);
1662 	irq_enable();
1663 	timer_fired = false;
1664 	vmcall();
1665 	report(timer_fired, "Inject an event to a halted guest");
1666 }
1667 
1668 static int interrupt_exit_handler(union exit_reason exit_reason)
1669 {
1670 	u64 guest_rip = vmcs_read(GUEST_RIP);
1671 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1672 
1673 	switch (exit_reason.basic) {
1674 	case VMX_VMCALL:
1675 		switch (vmx_get_test_stage()) {
1676 		case 0:
1677 		case 2:
1678 		case 5:
1679 			vmcs_write(PIN_CONTROLS,
1680 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1681 			break;
1682 		case 7:
1683 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1684 			vmcs_write(PIN_CONTROLS,
1685 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1686 			break;
1687 		case 1:
1688 		case 3:
1689 			vmcs_write(PIN_CONTROLS,
1690 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1691 			break;
1692 		case 4:
1693 		case 6:
1694 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1695 			break;
1696 
1697 		case 8:
1698 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1699 			vmcs_write(ENT_INTR_INFO,
1700 				   TIMER_VECTOR |
1701 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1702 				   INTR_INFO_VALID_MASK);
1703 			break;
1704 		}
1705 		vmx_inc_test_stage();
1706 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1707 		return VMX_TEST_RESUME;
1708 	case VMX_EXTINT:
1709 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1710 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1711 			handle_external_interrupt(vector);
1712 		} else {
1713 			irq_enable();
1714 			asm volatile ("nop");
1715 			irq_disable();
1716 		}
1717 		if (vmx_get_test_stage() >= 2)
1718 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1719 		return VMX_TEST_RESUME;
1720 	default:
1721 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
1722 		print_vmexit_info(exit_reason);
1723 	}
1724 
1725 	return VMX_TEST_VMEXIT;
1726 }
1727 
1728 
1729 static volatile int nmi_fired;
1730 
1731 #define NMI_DELAY 100000000ULL
1732 
1733 static void nmi_isr(isr_regs_t *regs)
1734 {
1735 	nmi_fired = true;
1736 }
1737 
1738 static int nmi_hlt_init(struct vmcs *vmcs)
1739 {
1740 	msr_bmp_init();
1741 	handle_irq(NMI_VECTOR, nmi_isr);
1742 	vmcs_write(PIN_CONTROLS,
1743 		   vmcs_read(PIN_CONTROLS) & ~PIN_NMI);
1744 	vmcs_write(PIN_CONTROLS,
1745 		   vmcs_read(PIN_CONTROLS) & ~PIN_VIRT_NMI);
1746 	return VMX_TEST_START;
1747 }
1748 
1749 static void nmi_message_thread(void *data)
1750 {
1751     while (vmx_get_test_stage() != 1)
1752         pause();
1753 
1754     delay(NMI_DELAY);
1755     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1756 
1757     while (vmx_get_test_stage() != 2)
1758         pause();
1759 
1760     delay(NMI_DELAY);
1761     apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1762 }
1763 
1764 static void nmi_hlt_main(void)
1765 {
1766     long long start;
1767 
1768     if (cpu_count() < 2) {
1769         report_skip(__func__);
1770         vmx_set_test_stage(-1);
1771         return;
1772     }
1773 
1774     vmx_set_test_stage(0);
1775     on_cpu_async(1, nmi_message_thread, NULL);
1776     start = rdtsc();
1777     vmx_set_test_stage(1);
1778     asm volatile ("hlt");
1779     report((rdtsc() - start > NMI_DELAY) && nmi_fired,
1780             "direct NMI + hlt");
1781     if (!nmi_fired)
1782         vmx_set_test_stage(-1);
1783     nmi_fired = false;
1784 
1785     vmcall();
1786 
1787     start = rdtsc();
1788     vmx_set_test_stage(2);
1789     asm volatile ("hlt");
1790     report((rdtsc() - start > NMI_DELAY) && !nmi_fired,
1791             "intercepted NMI + hlt");
1792     if (nmi_fired) {
1793         report(!nmi_fired, "intercepted NMI was dispatched");
1794         vmx_set_test_stage(-1);
1795         return;
1796     }
1797     vmx_set_test_stage(3);
1798 }
1799 
1800 static int nmi_hlt_exit_handler(union exit_reason exit_reason)
1801 {
1802     u64 guest_rip = vmcs_read(GUEST_RIP);
1803     u32 insn_len = vmcs_read(EXI_INST_LEN);
1804 
1805     switch (vmx_get_test_stage()) {
1806     case 1:
1807         if (exit_reason.basic != VMX_VMCALL) {
1808             report_fail("VMEXIT not due to vmcall. Exit reason 0x%x",
1809                         exit_reason.full);
1810             print_vmexit_info(exit_reason);
1811             return VMX_TEST_VMEXIT;
1812         }
1813 
1814         vmcs_write(PIN_CONTROLS,
1815                vmcs_read(PIN_CONTROLS) | PIN_NMI);
1816         vmcs_write(PIN_CONTROLS,
1817                vmcs_read(PIN_CONTROLS) | PIN_VIRT_NMI);
1818         vmcs_write(GUEST_RIP, guest_rip + insn_len);
1819         break;
1820 
1821     case 2:
1822         if (exit_reason.basic != VMX_EXC_NMI) {
1823             report_fail("VMEXIT not due to NMI intercept. Exit reason 0x%x",
1824                         exit_reason.full);
1825             print_vmexit_info(exit_reason);
1826             return VMX_TEST_VMEXIT;
1827         }
1828         report_pass("NMI intercept while running guest");
1829         vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1830         break;
1831 
1832     case 3:
1833         break;
1834 
1835     default:
1836         return VMX_TEST_VMEXIT;
1837     }
1838 
1839     if (vmx_get_test_stage() == 3)
1840         return VMX_TEST_VMEXIT;
1841 
1842     return VMX_TEST_RESUME;
1843 }
1844 
1845 
1846 static int dbgctls_init(struct vmcs *vmcs)
1847 {
1848 	u64 dr7 = 0x402;
1849 	u64 zero = 0;
1850 
1851 	msr_bmp_init();
1852 	asm volatile(
1853 		"mov %0,%%dr0\n\t"
1854 		"mov %0,%%dr1\n\t"
1855 		"mov %0,%%dr2\n\t"
1856 		"mov %1,%%dr7\n\t"
1857 		: : "r" (zero), "r" (dr7));
1858 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1859 	vmcs_write(GUEST_DR7, 0x404);
1860 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1861 
1862 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1863 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1864 
1865 	return VMX_TEST_START;
1866 }
1867 
1868 static void dbgctls_main(void)
1869 {
1870 	u64 dr7, debugctl;
1871 
1872 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1873 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1874 	/* Commented out: KVM does not support DEBUGCTL so far */
1875 	(void)debugctl;
1876 	report(dr7 == 0x404, "Load debug controls" /* && debugctl == 0x2 */);
1877 
1878 	dr7 = 0x408;
1879 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1880 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1881 
1882 	vmx_set_test_stage(0);
1883 	vmcall();
1884 	report(vmx_get_test_stage() == 1, "Save debug controls");
1885 
1886 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1887 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1888 		printf("\tDebug controls are always loaded/saved\n");
1889 		return;
1890 	}
1891 	vmx_set_test_stage(2);
1892 	vmcall();
1893 
1894 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1895 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1896 	/* Commented out: KVM does not support DEBUGCTL so far */
1897 	(void)debugctl;
1898 	report(dr7 == 0x402,
1899 	       "Guest=host debug controls" /* && debugctl == 0x1 */);
1900 
1901 	dr7 = 0x408;
1902 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1903 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1904 
1905 	vmx_set_test_stage(3);
1906 	vmcall();
1907 	report(vmx_get_test_stage() == 4, "Don't save debug controls");
1908 }
1909 
1910 static int dbgctls_exit_handler(union exit_reason exit_reason)
1911 {
1912 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1913 	u64 guest_rip = vmcs_read(GUEST_RIP);
1914 	u64 dr7, debugctl;
1915 
1916 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1917 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1918 
1919 	switch (exit_reason.basic) {
1920 	case VMX_VMCALL:
1921 		switch (vmx_get_test_stage()) {
1922 		case 0:
1923 			if (dr7 == 0x400 && debugctl == 0 &&
1924 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1925 			    Commented out: KVM does not support DEBUGCTL so far
1926 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1927 				vmx_inc_test_stage();
1928 			break;
1929 		case 2:
1930 			dr7 = 0x402;
1931 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1932 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1933 			vmcs_write(GUEST_DR7, 0x404);
1934 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1935 
1936 			vmcs_write(ENT_CONTROLS,
1937 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1938 			vmcs_write(EXI_CONTROLS,
1939 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1940 			break;
1941 		case 3:
1942 			if (dr7 == 0x400 && debugctl == 0 &&
1943 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1944 			    Commented out: KVM does not support DEBUGCTL so far
1945 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1946 				vmx_inc_test_stage();
1947 			break;
1948 		}
1949 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1950 		return VMX_TEST_RESUME;
1951 	default:
1952 		report_fail("Unknown exit reason, %d", exit_reason.full);
1953 		print_vmexit_info(exit_reason);
1954 	}
1955 	return VMX_TEST_VMEXIT;
1956 }
1957 
1958 struct vmx_msr_entry {
1959 	u32 index;
1960 	u32 reserved;
1961 	u64 value;
1962 } __attribute__((packed));
1963 
1964 #define MSR_MAGIC 0x31415926
1965 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1966 
1967 static int msr_switch_init(struct vmcs *vmcs)
1968 {
1969 	msr_bmp_init();
1970 	exit_msr_store = alloc_page();
1971 	exit_msr_load = alloc_page();
1972 	entry_msr_load = alloc_page();
1973 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1974 	entry_msr_load[0].value = MSR_MAGIC;
1975 
1976 	vmx_set_test_stage(1);
1977 	vmcs_write(ENT_MSR_LD_CNT, 1);
1978 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1979 	vmcs_write(EXI_MSR_ST_CNT, 1);
1980 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1981 	vmcs_write(EXI_MSR_LD_CNT, 1);
1982 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1983 	return VMX_TEST_START;
1984 }
1985 
1986 static void msr_switch_main(void)
1987 {
1988 	if (vmx_get_test_stage() == 1) {
1989 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC,
1990 		       "VM entry MSR load");
1991 		vmx_set_test_stage(2);
1992 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
1993 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
1994 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
1995 		exit_msr_load[0].value = MSR_MAGIC + 2;
1996 	}
1997 	vmcall();
1998 }
1999 
2000 static int msr_switch_exit_handler(union exit_reason exit_reason)
2001 {
2002 	if (exit_reason.basic == VMX_VMCALL && vmx_get_test_stage() == 2) {
2003 		report(exit_msr_store[0].value == MSR_MAGIC + 1,
2004 		       "VM exit MSR store");
2005 		report(rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2,
2006 		       "VM exit MSR load");
2007 		vmx_set_test_stage(3);
2008 		entry_msr_load[0].index = MSR_FS_BASE;
2009 		return VMX_TEST_RESUME;
2010 	}
2011 	printf("ERROR %s: unexpected stage=%u or reason=0x%x\n",
2012 		__func__, vmx_get_test_stage(), exit_reason.full);
2013 	return VMX_TEST_EXIT;
2014 }
2015 
2016 static int msr_switch_entry_failure(struct vmentry_result *result)
2017 {
2018 	if (result->vm_fail) {
2019 		printf("ERROR %s: VM-Fail on %s\n", __func__, result->instr);
2020 		return VMX_TEST_EXIT;
2021 	}
2022 
2023 	if (result->exit_reason.failed_vmentry &&
2024 	    result->exit_reason.basic == VMX_FAIL_MSR &&
2025 	    vmx_get_test_stage() == 3) {
2026 		report(vmcs_read(EXI_QUALIFICATION) == 1,
2027 		       "VM entry MSR load: try to load FS_BASE");
2028 		return VMX_TEST_VMEXIT;
2029 	}
2030 	printf("ERROR %s: unexpected stage=%u or reason=%x\n",
2031 		__func__, vmx_get_test_stage(), result->exit_reason.full);
2032 	return VMX_TEST_EXIT;
2033 }
2034 
2035 static int vmmcall_init(struct vmcs *vmcs)
2036 {
2037 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
2038 	return VMX_TEST_START;
2039 }
2040 
2041 static void vmmcall_main(void)
2042 {
2043 	asm volatile(
2044 		"mov $0xABCD, %%rax\n\t"
2045 		"vmmcall\n\t"
2046 		::: "rax");
2047 
2048 	report_fail("VMMCALL");
2049 }
2050 
2051 static int vmmcall_exit_handler(union exit_reason exit_reason)
2052 {
2053 	switch (exit_reason.basic) {
2054 	case VMX_VMCALL:
2055 		printf("here\n");
2056 		report_fail("VMMCALL triggers #UD");
2057 		break;
2058 	case VMX_EXC_NMI:
2059 		report((vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR,
2060 		       "VMMCALL triggers #UD");
2061 		break;
2062 	default:
2063 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2064 		print_vmexit_info(exit_reason);
2065 	}
2066 
2067 	return VMX_TEST_VMEXIT;
2068 }
2069 
2070 static int disable_rdtscp_init(struct vmcs *vmcs)
2071 {
2072 	u32 ctrl_cpu1;
2073 
2074 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
2075 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
2076 		ctrl_cpu1 &= ~CPU_RDTSCP;
2077 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
2078 	}
2079 
2080 	return VMX_TEST_START;
2081 }
2082 
2083 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
2084 {
2085 	switch (vmx_get_test_stage()) {
2086 	case 0:
2087 		report_pass("RDTSCP triggers #UD");
2088 		vmx_inc_test_stage();
2089 		regs->rip += 3;
2090 		break;
2091 	case 2:
2092 		report_pass("RDPID triggers #UD");
2093 		vmx_inc_test_stage();
2094 		regs->rip += 4;
2095 		break;
2096 	}
2097 	return;
2098 
2099 }
2100 
2101 static void disable_rdtscp_main(void)
2102 {
2103 	/* Test that #UD is properly injected in L2.  */
2104 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2105 
2106 	vmx_set_test_stage(0);
2107 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2108 	vmcall();
2109 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2110 
2111 	handle_exception(UD_VECTOR, 0);
2112 	vmcall();
2113 }
2114 
2115 static int disable_rdtscp_exit_handler(union exit_reason exit_reason)
2116 {
2117 	switch (exit_reason.basic) {
2118 	case VMX_VMCALL:
2119 		switch (vmx_get_test_stage()) {
2120 		case 0:
2121 			report_fail("RDTSCP triggers #UD");
2122 			vmx_inc_test_stage();
2123 			/* fallthrough */
2124 		case 1:
2125 			vmx_inc_test_stage();
2126 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2127 			return VMX_TEST_RESUME;
2128 		case 2:
2129 			report_fail("RDPID triggers #UD");
2130 			break;
2131 		}
2132 		break;
2133 
2134 	default:
2135 		report_fail("Unknown exit reason, 0x%x", exit_reason.full);
2136 		print_vmexit_info(exit_reason);
2137 	}
2138 	return VMX_TEST_VMEXIT;
2139 }
2140 
2141 static int int3_init(struct vmcs *vmcs)
2142 {
2143 	vmcs_write(EXC_BITMAP, ~0u);
2144 	return VMX_TEST_START;
2145 }
2146 
2147 static void int3_guest_main(void)
2148 {
2149 	asm volatile ("int3");
2150 }
2151 
2152 static int int3_exit_handler(union exit_reason exit_reason)
2153 {
2154 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2155 
2156 	report(exit_reason.basic == VMX_EXC_NMI &&
2157 	       (intr_info & INTR_INFO_VALID_MASK) &&
2158 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2159 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2160 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2161 	       "L1 intercepts #BP");
2162 
2163 	return VMX_TEST_VMEXIT;
2164 }
2165 
2166 static int into_init(struct vmcs *vmcs)
2167 {
2168 	vmcs_write(EXC_BITMAP, ~0u);
2169 	return VMX_TEST_START;
2170 }
2171 
2172 static void into_guest_main(void)
2173 {
2174 	struct far_pointer32 fp = {
2175 		.offset = (uintptr_t)&&into,
2176 		.selector = KERNEL_CS32,
2177 	};
2178 	uintptr_t rsp;
2179 
2180 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
2181 
2182 	if (fp.offset != (uintptr_t)&&into) {
2183 		printf("Code address too high.\n");
2184 		return;
2185 	}
2186 	if ((u32)rsp != rsp) {
2187 		printf("Stack address too high.\n");
2188 		return;
2189 	}
2190 
2191 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2192 	return;
2193 into:
2194 	asm volatile (".code32;"
2195 		      "movl $0x7fffffff, %eax;"
2196 		      "addl %eax, %eax;"
2197 		      "into;"
2198 		      "lret;"
2199 		      ".code64");
2200 	__builtin_unreachable();
2201 }
2202 
2203 static int into_exit_handler(union exit_reason exit_reason)
2204 {
2205 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2206 
2207 	report(exit_reason.basic == VMX_EXC_NMI &&
2208 	       (intr_info & INTR_INFO_VALID_MASK) &&
2209 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2210 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2211 	        INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION,
2212 	       "L1 intercepts #OF");
2213 
2214 	return VMX_TEST_VMEXIT;
2215 }
2216 
2217 static void exit_monitor_from_l2_main(void)
2218 {
2219 	printf("Calling exit(0) from l2...\n");
2220 	exit(0);
2221 }
2222 
2223 static int exit_monitor_from_l2_handler(union exit_reason exit_reason)
2224 {
2225 	report_fail("The guest should have killed the VMM");
2226 	return VMX_TEST_EXIT;
2227 }
2228 
2229 static void assert_exit_reason(u64 expected)
2230 {
2231 	u64 actual = vmcs_read(EXI_REASON);
2232 
2233 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2234 			   exit_reason_description(expected),
2235 			   exit_reason_description(actual));
2236 }
2237 
2238 static void skip_exit_insn(void)
2239 {
2240 	u64 guest_rip = vmcs_read(GUEST_RIP);
2241 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2242 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2243 }
2244 
2245 static void skip_exit_vmcall(void)
2246 {
2247 	assert_exit_reason(VMX_VMCALL);
2248 	skip_exit_insn();
2249 }
2250 
2251 static void v2_null_test_guest(void)
2252 {
2253 }
2254 
2255 static void v2_null_test(void)
2256 {
2257 	test_set_guest(v2_null_test_guest);
2258 	enter_guest();
2259 	report_pass(__func__);
2260 }
2261 
2262 static void v2_multiple_entries_test_guest(void)
2263 {
2264 	vmx_set_test_stage(1);
2265 	vmcall();
2266 	vmx_set_test_stage(2);
2267 }
2268 
2269 static void v2_multiple_entries_test(void)
2270 {
2271 	test_set_guest(v2_multiple_entries_test_guest);
2272 	enter_guest();
2273 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2274 	skip_exit_vmcall();
2275 	enter_guest();
2276 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2277 	report_pass(__func__);
2278 }
2279 
2280 static int fixture_test_data = 1;
2281 
2282 static void fixture_test_teardown(void *data)
2283 {
2284 	*((int *) data) = 1;
2285 }
2286 
2287 static void fixture_test_guest(void)
2288 {
2289 	fixture_test_data++;
2290 }
2291 
2292 
2293 static void fixture_test_setup(void)
2294 {
2295 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2296 			   "fixture_test_teardown didn't run?!");
2297 	fixture_test_data = 2;
2298 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2299 	test_set_guest(fixture_test_guest);
2300 }
2301 
2302 static void fixture_test_case1(void)
2303 {
2304 	fixture_test_setup();
2305 	TEST_ASSERT_EQ(2, fixture_test_data);
2306 	enter_guest();
2307 	TEST_ASSERT_EQ(3, fixture_test_data);
2308 	report_pass(__func__);
2309 }
2310 
2311 static void fixture_test_case2(void)
2312 {
2313 	fixture_test_setup();
2314 	TEST_ASSERT_EQ(2, fixture_test_data);
2315 	enter_guest();
2316 	TEST_ASSERT_EQ(3, fixture_test_data);
2317 	report_pass(__func__);
2318 }
2319 
2320 enum ept_access_op {
2321 	OP_READ,
2322 	OP_WRITE,
2323 	OP_EXEC,
2324 	OP_FLUSH_TLB,
2325 	OP_EXIT,
2326 };
2327 
2328 static struct ept_access_test_data {
2329 	unsigned long gpa;
2330 	unsigned long *gva;
2331 	unsigned long hpa;
2332 	unsigned long *hva;
2333 	enum ept_access_op op;
2334 } ept_access_test_data;
2335 
2336 extern unsigned char ret42_start;
2337 extern unsigned char ret42_end;
2338 
2339 /* Returns 42. */
2340 asm(
2341 	".align 64\n"
2342 	"ret42_start:\n"
2343 	"mov $42, %eax\n"
2344 	"ret\n"
2345 	"ret42_end:\n"
2346 );
2347 
2348 static void
2349 diagnose_ept_violation_qual(u64 expected, u64 actual)
2350 {
2351 
2352 #define DIAGNOSE(flag)							\
2353 do {									\
2354 	if ((expected & flag) != (actual & flag))			\
2355 		printf(#flag " %sexpected\n",				\
2356 		       (expected & flag) ? "" : "un");			\
2357 } while (0)
2358 
2359 	DIAGNOSE(EPT_VLT_RD);
2360 	DIAGNOSE(EPT_VLT_WR);
2361 	DIAGNOSE(EPT_VLT_FETCH);
2362 	DIAGNOSE(EPT_VLT_PERM_RD);
2363 	DIAGNOSE(EPT_VLT_PERM_WR);
2364 	DIAGNOSE(EPT_VLT_PERM_EX);
2365 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2366 	DIAGNOSE(EPT_VLT_PADDR);
2367 
2368 #undef DIAGNOSE
2369 }
2370 
2371 static void do_ept_access_op(enum ept_access_op op)
2372 {
2373 	ept_access_test_data.op = op;
2374 	enter_guest();
2375 }
2376 
2377 /*
2378  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2379  * needed by tests that modify guest PTEs.
2380  */
2381 static void ept_access_test_guest_flush_tlb(void)
2382 {
2383 	do_ept_access_op(OP_FLUSH_TLB);
2384 	skip_exit_vmcall();
2385 }
2386 
2387 /*
2388  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2389  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2390  * a huge page.
2391  */
2392 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2393 				 unsigned long clear, unsigned long set)
2394 {
2395 	struct ept_access_test_data *data = &ept_access_test_data;
2396 	unsigned long orig_pte;
2397 	unsigned long pte;
2398 
2399 	/* Screw with the mapping at the requested level. */
2400 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2401 	pte = orig_pte;
2402 	if (mkhuge)
2403 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2404 	else
2405 		pte = orig_pte;
2406 	pte = (pte & ~clear) | set;
2407 	set_ept_pte(pml4, gpa, level, pte);
2408 	invept(INVEPT_SINGLE, eptp);
2409 
2410 	return orig_pte;
2411 }
2412 
2413 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2414 {
2415 	set_ept_pte(pml4, gpa, level, orig_pte);
2416 	invept(INVEPT_SINGLE, eptp);
2417 }
2418 
2419 static void do_ept_violation(bool leaf, enum ept_access_op op,
2420 			     u64 expected_qual, u64 expected_paddr)
2421 {
2422 	u64 qual;
2423 
2424 	/* Try the access and observe the violation. */
2425 	do_ept_access_op(op);
2426 
2427 	assert_exit_reason(VMX_EPT_VIOLATION);
2428 
2429 	qual = vmcs_read(EXI_QUALIFICATION);
2430 
2431 	/* Mask undefined bits (which may later be defined in certain cases). */
2432 	qual &= ~(EPT_VLT_GUEST_USER | EPT_VLT_GUEST_RW | EPT_VLT_GUEST_EX |
2433 		 EPT_VLT_PERM_USER_EX);
2434 
2435 	diagnose_ept_violation_qual(expected_qual, qual);
2436 	TEST_EXPECT_EQ(expected_qual, qual);
2437 
2438 	#if 0
2439 	/* Disable for now otherwise every test will fail */
2440 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2441 		       (unsigned long) (
2442 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2443 	#endif
2444 	/*
2445 	 * TODO: tests that probe expected_paddr in pages other than the one at
2446 	 * the beginning of the 1g region.
2447 	 */
2448 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2449 }
2450 
2451 static void
2452 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2453 			      unsigned long set, enum ept_access_op op,
2454 			      u64 expected_qual)
2455 {
2456 	struct ept_access_test_data *data = &ept_access_test_data;
2457 	unsigned long orig_pte;
2458 
2459 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2460 
2461 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2462 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2463 					 data->gpa);
2464 
2465 	/* Fix the violation and resume the op loop. */
2466 	ept_untwiddle(data->gpa, level, orig_pte);
2467 	enter_guest();
2468 	skip_exit_vmcall();
2469 }
2470 
2471 static void
2472 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2473 		       enum ept_access_op op, u64 expected_qual)
2474 {
2475 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2476 				      expected_qual);
2477 	if (ept_huge_pages_supported(level))
2478 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2479 					      expected_qual);
2480 }
2481 
2482 static void ept_violation(unsigned long clear, unsigned long set,
2483 			  enum ept_access_op op, u64 expected_qual)
2484 {
2485 	ept_violation_at_level(1, clear, set, op, expected_qual);
2486 	ept_violation_at_level(2, clear, set, op, expected_qual);
2487 	ept_violation_at_level(3, clear, set, op, expected_qual);
2488 	ept_violation_at_level(4, clear, set, op, expected_qual);
2489 }
2490 
2491 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2492 				       u64 expected_qual)
2493 {
2494 	ept_violation(EPT_PRESENT, access, op,
2495 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2496 }
2497 
2498 /*
2499  * For translations that don't involve a GVA, that is physical address (paddr)
2500  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2501  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2502  * translations don't involve GVAs, such as when the hardware does the guest
2503  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2504  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2505  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2506  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2507  *
2508  * Note that paddr violations can also be triggered by loading PAE page tables
2509  * with wonky addresses. We don't test that yet.
2510  *
2511  * This function modifies the EPT entry that maps the GPA that the guest page
2512  * table entry mapping ept_access_test_data.gva resides on.
2513  *
2514  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2515  *
2516  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2517  *
2518  *	@op		Guest operation to perform with
2519  *			ept_access_test_data.gva.
2520  *
2521  *	@expect_violation
2522  *			Is a violation expected during the paddr access?
2523  *
2524  *	@expected_qual	Expected qualification for the EPT violation.
2525  *			EPT_VLT_PADDR should be clear.
2526  */
2527 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2528 			     enum ept_access_op op, bool expect_violation,
2529 			     u64 expected_qual)
2530 {
2531 	struct ept_access_test_data *data = &ept_access_test_data;
2532 	unsigned long *ptep;
2533 	unsigned long gpa;
2534 	unsigned long orig_epte;
2535 	unsigned long epte;
2536 	int i;
2537 
2538 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2539 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2540 	TEST_ASSERT(ptep);
2541 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2542 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2543 	ept_access_test_guest_flush_tlb();
2544 
2545 	/*
2546 	 * Now modify the access bits on the EPT entry for the GPA that the
2547 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2548 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2549 	 * constructed our test such that those other 511 PTEs aren't used by
2550 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2551 	 * PTE we're modifying is at the beginning of a 4K page and the
2552 	 * following 511 entries are also under our control (and not touched by
2553 	 * the guest).
2554 	 */
2555 	gpa = virt_to_phys(ptep);
2556 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2557 	/*
2558 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2559 	 * otherwise our level=1 twiddling below will fail. We use the
2560 	 * identity map (gpa = gpa) since page tables are shared with the host.
2561 	 */
2562 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2563 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2564 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2565 
2566 	if (expect_violation) {
2567 		do_ept_violation(/*leaf=*/true, op,
2568 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2569 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2570 		do_ept_access_op(op);
2571 	} else {
2572 		do_ept_access_op(op);
2573 		if (ept_ad_enabled()) {
2574 			for (i = EPT_PAGE_LEVEL; i > 0; i--) {
2575 				TEST_ASSERT(get_ept_pte(pml4, gpa, i, &epte));
2576 				TEST_ASSERT(epte & EPT_ACCESS_FLAG);
2577 				if (i == 1)
2578 					TEST_ASSERT(epte & EPT_DIRTY_FLAG);
2579 				else
2580 					TEST_ASSERT_EQ(epte & EPT_DIRTY_FLAG, 0);
2581 			}
2582 		}
2583 
2584 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2585 	}
2586 
2587 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2588 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2589 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2590 
2591 	skip_exit_vmcall();
2592 }
2593 
2594 static void ept_access_allowed_paddr(unsigned long ept_access,
2595 				     unsigned long pte_ad,
2596 				     enum ept_access_op op)
2597 {
2598 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2599 			 /*expected_qual=*/-1);
2600 }
2601 
2602 static void ept_access_violation_paddr(unsigned long ept_access,
2603 				       unsigned long pte_ad,
2604 				       enum ept_access_op op,
2605 				       u64 expected_qual)
2606 {
2607 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2608 			 expected_qual);
2609 }
2610 
2611 
2612 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2613 					unsigned long clear,
2614 					unsigned long set,
2615 					enum ept_access_op op)
2616 {
2617 	struct ept_access_test_data *data = &ept_access_test_data;
2618 	unsigned long orig_pte;
2619 
2620 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2621 
2622 	/* No violation. Should proceed to vmcall. */
2623 	do_ept_access_op(op);
2624 	skip_exit_vmcall();
2625 
2626 	ept_untwiddle(data->gpa, level, orig_pte);
2627 }
2628 
2629 static void ept_allowed_at_level(int level, unsigned long clear,
2630 				 unsigned long set, enum ept_access_op op)
2631 {
2632 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2633 	if (ept_huge_pages_supported(level))
2634 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2635 }
2636 
2637 static void ept_allowed(unsigned long clear, unsigned long set,
2638 			enum ept_access_op op)
2639 {
2640 	ept_allowed_at_level(1, clear, set, op);
2641 	ept_allowed_at_level(2, clear, set, op);
2642 	ept_allowed_at_level(3, clear, set, op);
2643 	ept_allowed_at_level(4, clear, set, op);
2644 }
2645 
2646 static void ept_ignored_bit(int bit)
2647 {
2648 	/* Set the bit. */
2649 	ept_allowed(0, 1ul << bit, OP_READ);
2650 	ept_allowed(0, 1ul << bit, OP_WRITE);
2651 	ept_allowed(0, 1ul << bit, OP_EXEC);
2652 
2653 	/* Clear the bit. */
2654 	ept_allowed(1ul << bit, 0, OP_READ);
2655 	ept_allowed(1ul << bit, 0, OP_WRITE);
2656 	ept_allowed(1ul << bit, 0, OP_EXEC);
2657 }
2658 
2659 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2660 {
2661 	ept_allowed(EPT_PRESENT, access, op);
2662 }
2663 
2664 
2665 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2666 					     unsigned long clear,
2667 					     unsigned long set,
2668 					     enum ept_access_op op)
2669 {
2670 	struct ept_access_test_data *data = &ept_access_test_data;
2671 	unsigned long orig_pte;
2672 
2673 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2674 
2675 	do_ept_access_op(op);
2676 	assert_exit_reason(VMX_EPT_MISCONFIG);
2677 
2678 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2679 	#if 0
2680 	/* broken: */
2681 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2682 	#endif
2683 	#if 0
2684 	/*
2685 	 * broken:
2686 	 * According to description of exit qual for EPT violation,
2687 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2688 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2689 	 * to be set for msiconfig.
2690 	 */
2691 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2692 		       (unsigned long) (
2693 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2694 	#endif
2695 
2696 	/* Fix the violation and resume the op loop. */
2697 	ept_untwiddle(data->gpa, level, orig_pte);
2698 	enter_guest();
2699 	skip_exit_vmcall();
2700 }
2701 
2702 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2703 					  unsigned long clear,
2704 					  unsigned long set)
2705 {
2706 	/* The op shouldn't matter (read, write, exec), so try them all! */
2707 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2708 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2709 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2710 }
2711 
2712 static void ept_misconfig_at_level(int level, unsigned long clear,
2713 				   unsigned long set)
2714 {
2715 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2716 	if (ept_huge_pages_supported(level))
2717 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2718 }
2719 
2720 static void ept_misconfig(unsigned long clear, unsigned long set)
2721 {
2722 	ept_misconfig_at_level(1, clear, set);
2723 	ept_misconfig_at_level(2, clear, set);
2724 	ept_misconfig_at_level(3, clear, set);
2725 	ept_misconfig_at_level(4, clear, set);
2726 }
2727 
2728 static void ept_access_misconfig(unsigned long access)
2729 {
2730 	ept_misconfig(EPT_PRESENT, access);
2731 }
2732 
2733 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2734 {
2735 	/* Setting the bit causes a misconfig. */
2736 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2737 
2738 	/* Making the entry non-present turns reserved bits into ignored. */
2739 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2740 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2741 }
2742 
2743 static void ept_reserved_bit_at_level_huge(int level, int bit)
2744 {
2745 	/* Setting the bit causes a misconfig. */
2746 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2747 
2748 	/* Making the entry non-present turns reserved bits into ignored. */
2749 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2750 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2751 }
2752 
2753 static void ept_reserved_bit_at_level(int level, int bit)
2754 {
2755 	/* Setting the bit causes a misconfig. */
2756 	ept_misconfig_at_level(level, 0, 1ul << bit);
2757 
2758 	/* Making the entry non-present turns reserved bits into ignored. */
2759 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2760 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2761 }
2762 
2763 static void ept_reserved_bit(int bit)
2764 {
2765 	ept_reserved_bit_at_level(1, bit);
2766 	ept_reserved_bit_at_level(2, bit);
2767 	ept_reserved_bit_at_level(3, bit);
2768 	ept_reserved_bit_at_level(4, bit);
2769 }
2770 
2771 #define PAGE_2M_ORDER 9
2772 #define PAGE_1G_ORDER 18
2773 
2774 static void *get_1g_page(void)
2775 {
2776 	static void *alloc;
2777 
2778 	if (!alloc)
2779 		alloc = alloc_pages(PAGE_1G_ORDER);
2780 	return alloc;
2781 }
2782 
2783 static void ept_access_test_teardown(void *unused)
2784 {
2785 	/* Exit the guest cleanly. */
2786 	do_ept_access_op(OP_EXIT);
2787 }
2788 
2789 static void ept_access_test_guest(void)
2790 {
2791 	struct ept_access_test_data *data = &ept_access_test_data;
2792 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2793 
2794 	while (true) {
2795 		switch (data->op) {
2796 		case OP_READ:
2797 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2798 			break;
2799 		case OP_WRITE:
2800 			*data->gva = MAGIC_VAL_2;
2801 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2802 			*data->gva = MAGIC_VAL_1;
2803 			break;
2804 		case OP_EXEC:
2805 			TEST_ASSERT_EQ(42, code());
2806 			break;
2807 		case OP_FLUSH_TLB:
2808 			write_cr3(read_cr3());
2809 			break;
2810 		case OP_EXIT:
2811 			return;
2812 		default:
2813 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2814 		}
2815 		vmcall();
2816 	}
2817 }
2818 
2819 static void ept_access_test_setup(void)
2820 {
2821 	struct ept_access_test_data *data = &ept_access_test_data;
2822 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2823 	unsigned long size = npages * PAGE_SIZE;
2824 	unsigned long *page_table = current_page_table();
2825 	unsigned long pte;
2826 
2827 	if (setup_ept(false))
2828 		test_skip("EPT not supported");
2829 
2830 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2831 	if (cpuid_maxphyaddr() < 40)
2832 		test_skip("Test needs MAXPHYADDR >= 40");
2833 
2834 	test_set_guest(ept_access_test_guest);
2835 	test_add_teardown(ept_access_test_teardown, NULL);
2836 
2837 	data->hva = get_1g_page();
2838 	TEST_ASSERT(data->hva);
2839 	data->hpa = virt_to_phys(data->hva);
2840 
2841 	data->gpa = 1ul << 39;
2842 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2843 				   size);
2844 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2845 	install_pages(page_table, data->gpa, size, data->gva);
2846 
2847 	/*
2848 	 * Make sure nothing's mapped here so the tests that screw with the
2849 	 * pml4 entry don't inadvertently break something.
2850 	 */
2851 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2852 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2853 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2854 
2855 	data->hva[0] = MAGIC_VAL_1;
2856 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2857 }
2858 
2859 static void ept_access_test_not_present(void)
2860 {
2861 	ept_access_test_setup();
2862 	/* --- */
2863 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2864 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2865 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2866 }
2867 
2868 static void ept_access_test_read_only(void)
2869 {
2870 	ept_access_test_setup();
2871 
2872 	/* r-- */
2873 	ept_access_allowed(EPT_RA, OP_READ);
2874 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2875 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2876 }
2877 
2878 static void ept_access_test_write_only(void)
2879 {
2880 	ept_access_test_setup();
2881 	/* -w- */
2882 	ept_access_misconfig(EPT_WA);
2883 }
2884 
2885 static void ept_access_test_read_write(void)
2886 {
2887 	ept_access_test_setup();
2888 	/* rw- */
2889 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2890 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2891 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2892 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2893 }
2894 
2895 
2896 static void ept_access_test_execute_only(void)
2897 {
2898 	ept_access_test_setup();
2899 	/* --x */
2900 	if (ept_execute_only_supported()) {
2901 		ept_access_violation(EPT_EA, OP_READ,
2902 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2903 		ept_access_violation(EPT_EA, OP_WRITE,
2904 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2905 		ept_access_allowed(EPT_EA, OP_EXEC);
2906 	} else {
2907 		ept_access_misconfig(EPT_EA);
2908 	}
2909 }
2910 
2911 static void ept_access_test_read_execute(void)
2912 {
2913 	ept_access_test_setup();
2914 	/* r-x */
2915 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2916 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2917 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2918 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2919 }
2920 
2921 static void ept_access_test_write_execute(void)
2922 {
2923 	ept_access_test_setup();
2924 	/* -wx */
2925 	ept_access_misconfig(EPT_WA | EPT_EA);
2926 }
2927 
2928 static void ept_access_test_read_write_execute(void)
2929 {
2930 	ept_access_test_setup();
2931 	/* rwx */
2932 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2933 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2934 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2935 }
2936 
2937 static void ept_access_test_reserved_bits(void)
2938 {
2939 	int i;
2940 	int maxphyaddr;
2941 
2942 	ept_access_test_setup();
2943 
2944 	/* Reserved bits above maxphyaddr. */
2945 	maxphyaddr = cpuid_maxphyaddr();
2946 	for (i = maxphyaddr; i <= 51; i++) {
2947 		report_prefix_pushf("reserved_bit=%d", i);
2948 		ept_reserved_bit(i);
2949 		report_prefix_pop();
2950 	}
2951 
2952 	/* Level-specific reserved bits. */
2953 	ept_reserved_bit_at_level_nohuge(2, 3);
2954 	ept_reserved_bit_at_level_nohuge(2, 4);
2955 	ept_reserved_bit_at_level_nohuge(2, 5);
2956 	ept_reserved_bit_at_level_nohuge(2, 6);
2957 	/* 2M alignment. */
2958 	for (i = 12; i < 20; i++) {
2959 		report_prefix_pushf("reserved_bit=%d", i);
2960 		ept_reserved_bit_at_level_huge(2, i);
2961 		report_prefix_pop();
2962 	}
2963 	ept_reserved_bit_at_level_nohuge(3, 3);
2964 	ept_reserved_bit_at_level_nohuge(3, 4);
2965 	ept_reserved_bit_at_level_nohuge(3, 5);
2966 	ept_reserved_bit_at_level_nohuge(3, 6);
2967 	/* 1G alignment. */
2968 	for (i = 12; i < 29; i++) {
2969 		report_prefix_pushf("reserved_bit=%d", i);
2970 		ept_reserved_bit_at_level_huge(3, i);
2971 		report_prefix_pop();
2972 	}
2973 	ept_reserved_bit_at_level(4, 3);
2974 	ept_reserved_bit_at_level(4, 4);
2975 	ept_reserved_bit_at_level(4, 5);
2976 	ept_reserved_bit_at_level(4, 6);
2977 	ept_reserved_bit_at_level(4, 7);
2978 }
2979 
2980 static void ept_access_test_ignored_bits(void)
2981 {
2982 	ept_access_test_setup();
2983 	/*
2984 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2985 	 * far as translation is concerned even if AD bits are enabled in the
2986 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2987 	 * control is 0.
2988 	 */
2989 	ept_ignored_bit(8);
2990 	ept_ignored_bit(9);
2991 	ept_ignored_bit(10);
2992 	ept_ignored_bit(11);
2993 	ept_ignored_bit(52);
2994 	ept_ignored_bit(53);
2995 	ept_ignored_bit(54);
2996 	ept_ignored_bit(55);
2997 	ept_ignored_bit(56);
2998 	ept_ignored_bit(57);
2999 	ept_ignored_bit(58);
3000 	ept_ignored_bit(59);
3001 	ept_ignored_bit(60);
3002 	ept_ignored_bit(61);
3003 	ept_ignored_bit(62);
3004 	ept_ignored_bit(63);
3005 }
3006 
3007 static void ept_access_test_paddr_not_present_ad_disabled(void)
3008 {
3009 	ept_access_test_setup();
3010 	ept_disable_ad_bits();
3011 
3012 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
3013 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
3014 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
3015 }
3016 
3017 static void ept_access_test_paddr_not_present_ad_enabled(void)
3018 {
3019 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
3020 
3021 	ept_access_test_setup();
3022 	ept_enable_ad_bits_or_skip_test();
3023 
3024 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
3025 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
3026 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
3027 }
3028 
3029 static void ept_access_test_paddr_read_only_ad_disabled(void)
3030 {
3031 	/*
3032 	 * When EPT AD bits are disabled, all accesses to guest paging
3033 	 * structures are reported separately as a read and (after
3034 	 * translation of the GPA to host physical address) a read+write
3035 	 * if the A/D bits have to be set.
3036 	 */
3037 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3038 
3039 	ept_access_test_setup();
3040 	ept_disable_ad_bits();
3041 
3042 	/* Can't update A bit, so all accesses fail. */
3043 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3044 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3045 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3046 	/* AD bits disabled, so only writes try to update the D bit. */
3047 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
3048 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3049 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
3050 	/* Both A and D already set, so read-only is OK. */
3051 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
3052 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
3053 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
3054 }
3055 
3056 static void ept_access_test_paddr_read_only_ad_enabled(void)
3057 {
3058 	/*
3059 	 * When EPT AD bits are enabled, all accesses to guest paging
3060 	 * structures are considered writes as far as EPT translation
3061 	 * is concerned.
3062 	 */
3063 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3064 
3065 	ept_access_test_setup();
3066 	ept_enable_ad_bits_or_skip_test();
3067 
3068 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3069 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3070 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3071 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
3072 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3073 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
3074 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
3075 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
3076 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
3077 }
3078 
3079 static void ept_access_test_paddr_read_write(void)
3080 {
3081 	ept_access_test_setup();
3082 	/* Read-write access to paging structure. */
3083 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
3084 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
3085 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
3086 }
3087 
3088 static void ept_access_test_paddr_read_write_execute(void)
3089 {
3090 	ept_access_test_setup();
3091 	/* RWX access to paging structure. */
3092 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
3093 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
3094 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
3095 }
3096 
3097 static void ept_access_test_paddr_read_execute_ad_disabled(void)
3098 {
3099   	/*
3100 	 * When EPT AD bits are disabled, all accesses to guest paging
3101 	 * structures are reported separately as a read and (after
3102 	 * translation of the GPA to host physical address) a read+write
3103 	 * if the A/D bits have to be set.
3104 	 */
3105 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3106 
3107 	ept_access_test_setup();
3108 	ept_disable_ad_bits();
3109 
3110 	/* Can't update A bit, so all accesses fail. */
3111 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3112 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3113 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3114 	/* AD bits disabled, so only writes try to update the D bit. */
3115 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
3116 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3117 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3118 	/* Both A and D already set, so read-only is OK. */
3119 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3120 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3121 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3122 }
3123 
3124 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3125 {
3126 	/*
3127 	 * When EPT AD bits are enabled, all accesses to guest paging
3128 	 * structures are considered writes as far as EPT translation
3129 	 * is concerned.
3130 	 */
3131 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3132 
3133 	ept_access_test_setup();
3134 	ept_enable_ad_bits_or_skip_test();
3135 
3136 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3137 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3138 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3139 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3140 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3141 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3142 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3143 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3144 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3145 }
3146 
3147 static void ept_access_test_paddr_not_present_page_fault(void)
3148 {
3149 	ept_access_test_setup();
3150 	/*
3151 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3152 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3153 	 * Thus guest page fault before host takes EPT violation for trying to
3154 	 * update A bit.
3155 	 */
3156 }
3157 
3158 static void ept_access_test_force_2m_page(void)
3159 {
3160 	ept_access_test_setup();
3161 
3162 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3163 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3164 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3165 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3166 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3167 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3168 }
3169 
3170 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3171 {
3172 	if (!is_invvpid_type_supported(type))
3173 		return false;
3174 
3175 	if (vpid >> 16)
3176 		return false;
3177 
3178 	if (type != INVVPID_ALL && !vpid)
3179 		return false;
3180 
3181 	if (type == INVVPID_ADDR && !is_canonical(gla))
3182 		return false;
3183 
3184 	return true;
3185 }
3186 
3187 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3188 {
3189 	int rc;
3190 	bool valid = invvpid_valid(type, vpid, gla);
3191 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3192 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3193 	/*
3194 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3195 	 * that we can tell if it is updated by INVVPID.
3196 	 */
3197 	vmcs_read(~0);
3198 	rc = __invvpid(type, vpid, gla);
3199 	report(!rc == valid, "INVVPID type %ld VPID %lx GLA %lx %s", type,
3200 	       vpid, gla,
3201 	       valid ? "passes" : "fails");
3202 	report(vmcs_read(VMX_INST_ERROR) == expected,
3203 	       "After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3204 	       rc ? "failed" : "successful",
3205 	       expected, vmcs_read(VMX_INST_ERROR));
3206 }
3207 
3208 static inline unsigned long get_first_supported_invvpid_type(void)
3209 {
3210 	u64 type = ffs(ept_vpid.val >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3211 
3212 	__TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3213 	return type;
3214 }
3215 
3216 static void ds_invvpid(void *data)
3217 {
3218 	asm volatile("invvpid %0, %1"
3219 		     :
3220 		     : "m"(*(struct invvpid_operand *)data),
3221 		       "r"(get_first_supported_invvpid_type()));
3222 }
3223 
3224 /*
3225  * The SS override is ignored in 64-bit mode, so we use an addressing
3226  * mode with %rsp as the base register to generate an implicit SS
3227  * reference.
3228  */
3229 static void ss_invvpid(void *data)
3230 {
3231 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3232 		     : "+r"(data)
3233 		     : "r"(get_first_supported_invvpid_type()));
3234 }
3235 
3236 static void invvpid_test_gp(void)
3237 {
3238 	bool fault;
3239 
3240 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3241 				   (void *)NONCANONICAL);
3242 	report(fault, "INVVPID with non-canonical DS operand raises #GP");
3243 }
3244 
3245 static void invvpid_test_ss(void)
3246 {
3247 	bool fault;
3248 
3249 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3250 				   (void *)NONCANONICAL);
3251 	report(fault, "INVVPID with non-canonical SS operand raises #SS");
3252 }
3253 
3254 static void invvpid_test_pf(void)
3255 {
3256 	void *vpage = alloc_vpage();
3257 	bool fault;
3258 
3259 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3260 	report(fault, "INVVPID with unmapped operand raises #PF");
3261 }
3262 
3263 static void try_compat_invvpid(void *unused)
3264 {
3265 	struct far_pointer32 fp = {
3266 		.offset = (uintptr_t)&&invvpid,
3267 		.selector = KERNEL_CS32,
3268 	};
3269 	uintptr_t rsp;
3270 
3271 	asm volatile ("mov %%rsp, %0" : "=r"(rsp));
3272 
3273 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3274 			"Code address too high.");
3275 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3276 
3277 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3278 	return;
3279 invvpid:
3280 	asm volatile (".code32;"
3281 		      "invvpid (%eax), %eax;"
3282 		      "lret;"
3283 		      ".code64");
3284 	__builtin_unreachable();
3285 }
3286 
3287 static void invvpid_test_compatibility_mode(void)
3288 {
3289 	bool fault;
3290 
3291 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3292 	report(fault, "Compatibility mode INVVPID raises #UD");
3293 }
3294 
3295 static void invvpid_test_not_in_vmx_operation(void)
3296 {
3297 	bool fault;
3298 
3299 	TEST_ASSERT(!vmx_off());
3300 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3301 	report(fault, "INVVPID outside of VMX operation raises #UD");
3302 	TEST_ASSERT(!vmx_on());
3303 }
3304 
3305 /*
3306  * This does not test real-address mode, virtual-8086 mode, protected mode,
3307  * or CPL > 0.
3308  */
3309 static void invvpid_test(void)
3310 {
3311 	int i;
3312 	unsigned types = 0;
3313 	unsigned type;
3314 
3315 	if (!is_vpid_supported())
3316 		test_skip("VPID not supported");
3317 
3318 	if (!is_invvpid_supported())
3319 		test_skip("INVVPID not supported.\n");
3320 
3321 	if (is_invvpid_type_supported(INVVPID_ADDR))
3322 		types |= 1u << INVVPID_ADDR;
3323 	if (is_invvpid_type_supported(INVVPID_CONTEXT_GLOBAL))
3324 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3325 	if (is_invvpid_type_supported(INVVPID_ALL))
3326 		types |= 1u << INVVPID_ALL;
3327 	if (is_invvpid_type_supported(INVVPID_CONTEXT_LOCAL))
3328 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3329 
3330 	if (!types)
3331 		test_skip("No INVVPID types supported.\n");
3332 
3333 	for (i = -127; i < 128; i++)
3334 		try_invvpid(i, 0xffff, 0);
3335 
3336 	/*
3337 	 * VPID must not be more than 16 bits.
3338 	 */
3339 	for (i = 0; i < 64; i++)
3340 		for (type = 0; type < 4; type++)
3341 			if (types & (1u << type))
3342 				try_invvpid(type, 1ul << i, 0);
3343 
3344 	/*
3345 	 * VPID must not be zero, except for "all contexts."
3346 	 */
3347 	for (type = 0; type < 4; type++)
3348 		if (types & (1u << type))
3349 			try_invvpid(type, 0, 0);
3350 
3351 	/*
3352 	 * The gla operand is only validated for single-address INVVPID.
3353 	 */
3354 	if (types & (1u << INVVPID_ADDR))
3355 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3356 
3357 	invvpid_test_gp();
3358 	invvpid_test_ss();
3359 	invvpid_test_pf();
3360 	invvpid_test_compatibility_mode();
3361 	invvpid_test_not_in_vmx_operation();
3362 }
3363 
3364 /*
3365  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3366  * at least as far as the guest-state checks. Returns false if the
3367  * VMLAUNCH fails early and execution falls through to the next
3368  * instruction.
3369  */
3370 static bool vmlaunch_succeeds(void)
3371 {
3372 	u32 exit_reason;
3373 
3374 	/*
3375 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3376 	 * unsupported VMCS component"). The caller can then check
3377 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3378 	 */
3379 	vmcs_write(~0u, 0);
3380 
3381 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3382 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3383 				   :
3384 				   : "r" ((u64)HOST_RSP)
3385 				   : "cc", "memory"
3386 				   : success);
3387 	return false;
3388 success:
3389 	exit_reason = vmcs_read(EXI_REASON);
3390 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3391 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3392 	return true;
3393 }
3394 
3395 /*
3396  * Try to launch the current VMCS.
3397  */
3398 static void test_vmx_vmlaunch(u32 xerror)
3399 {
3400 	bool success = vmlaunch_succeeds();
3401 	u32 vmx_inst_err;
3402 
3403 	report(success == !xerror, "vmlaunch %s",
3404 	       !xerror ? "succeeds" : "fails");
3405 	if (!success && xerror) {
3406 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3407 		report(vmx_inst_err == xerror,
3408 		       "VMX inst error is %d (actual %d)", xerror,
3409 		       vmx_inst_err);
3410 	}
3411 }
3412 
3413 /*
3414  * Try to launch the current VMCS, and expect one of two possible
3415  * errors (or success) codes.
3416  */
3417 static void test_vmx_vmlaunch2(u32 xerror1, u32 xerror2)
3418 {
3419 	bool success = vmlaunch_succeeds();
3420 	u32 vmx_inst_err;
3421 
3422 	if (!xerror1 == !xerror2)
3423 		report(success == !xerror1, "vmlaunch %s",
3424 		       !xerror1 ? "succeeds" : "fails");
3425 
3426 	if (!success && (xerror1 || xerror2)) {
3427 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3428 		report(vmx_inst_err == xerror1 || vmx_inst_err == xerror2,
3429 		       "VMX inst error is %d or %d (actual %d)", xerror1,
3430 		       xerror2, vmx_inst_err);
3431 	}
3432 }
3433 
3434 static void test_vmx_invalid_controls(void)
3435 {
3436 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_CONTROL_FIELD);
3437 }
3438 
3439 static void test_vmx_valid_controls(void)
3440 {
3441 	test_vmx_vmlaunch(0);
3442 }
3443 
3444 /*
3445  * Test a particular value of a VM-execution control bit, if the value
3446  * is required or if the value is zero.
3447  */
3448 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3449 				    enum Encoding encoding, unsigned bit,
3450 				    unsigned val)
3451 {
3452 	u32 mask = 1u << bit;
3453 	bool expected;
3454 	u32 controls;
3455 
3456 	if (msr.set & mask)
3457 		TEST_ASSERT(msr.clr & mask);
3458 
3459 	/*
3460 	 * We can't arbitrarily turn on a control bit, because it may
3461 	 * introduce dependencies on other VMCS fields. So, we only
3462 	 * test turning on bits that have a required setting.
3463 	 */
3464 	if (val && (msr.clr & mask) && !(msr.set & mask))
3465 		return;
3466 
3467 	report_prefix_pushf("%s %s bit %d",
3468 			    val ? "Set" : "Clear", name, bit);
3469 
3470 	controls = vmcs_read(encoding);
3471 	if (val) {
3472 		vmcs_write(encoding, msr.set | mask);
3473 		expected = (msr.clr & mask);
3474 	} else {
3475 		vmcs_write(encoding, msr.set & ~mask);
3476 		expected = !(msr.set & mask);
3477 	}
3478 	if (expected)
3479 		test_vmx_valid_controls();
3480 	else
3481 		test_vmx_invalid_controls();
3482 	vmcs_write(encoding, controls);
3483 	report_prefix_pop();
3484 }
3485 
3486 /*
3487  * Test reserved values of a VM-execution control bit, based on the
3488  * allowed bit settings from the corresponding VMX capability MSR.
3489  */
3490 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3491 			      enum Encoding encoding, unsigned bit)
3492 {
3493 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3494 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3495 }
3496 
3497 /*
3498  * Reserved bits in the pin-based VM-execution controls must be set
3499  * properly. Software may consult the VMX capability MSRs to determine
3500  * the proper settings.
3501  * [Intel SDM]
3502  */
3503 static void test_pin_based_ctls(void)
3504 {
3505 	unsigned bit;
3506 
3507 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3508 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3509 	for (bit = 0; bit < 32; bit++)
3510 		test_rsvd_ctl_bit("pin-based controls",
3511 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3512 }
3513 
3514 /*
3515  * Reserved bits in the primary processor-based VM-execution controls
3516  * must be set properly. Software may consult the VMX capability MSRs
3517  * to determine the proper settings.
3518  * [Intel SDM]
3519  */
3520 static void test_primary_processor_based_ctls(void)
3521 {
3522 	unsigned bit;
3523 
3524 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3525 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3526 	for (bit = 0; bit < 32; bit++)
3527 		test_rsvd_ctl_bit("primary processor-based controls",
3528 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3529 }
3530 
3531 /*
3532  * If the "activate secondary controls" primary processor-based
3533  * VM-execution control is 1, reserved bits in the secondary
3534  * processor-based VM-execution controls must be cleared. Software may
3535  * consult the VMX capability MSRs to determine which bits are
3536  * reserved.
3537  * If the "activate secondary controls" primary processor-based
3538  * VM-execution control is 0 (or if the processor does not support the
3539  * 1-setting of that control), no checks are performed on the
3540  * secondary processor-based VM-execution controls.
3541  * [Intel SDM]
3542  */
3543 static void test_secondary_processor_based_ctls(void)
3544 {
3545 	u32 primary;
3546 	u32 secondary;
3547 	unsigned bit;
3548 
3549 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3550 		return;
3551 
3552 	primary = vmcs_read(CPU_EXEC_CTRL0);
3553 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3554 
3555 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3556 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3557 	for (bit = 0; bit < 32; bit++)
3558 		test_rsvd_ctl_bit("secondary processor-based controls",
3559 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3560 
3561 	/*
3562 	 * When the "activate secondary controls" VM-execution control
3563 	 * is clear, there are no checks on the secondary controls.
3564 	 */
3565 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3566 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3567 	report(vmlaunch_succeeds(),
3568 	       "Secondary processor-based controls ignored");
3569 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3570 	vmcs_write(CPU_EXEC_CTRL0, primary);
3571 }
3572 
3573 static void try_cr3_target_count(unsigned i, unsigned max)
3574 {
3575 	report_prefix_pushf("CR3 target count 0x%x", i);
3576 	vmcs_write(CR3_TARGET_COUNT, i);
3577 	if (i <= max)
3578 		test_vmx_valid_controls();
3579 	else
3580 		test_vmx_invalid_controls();
3581 	report_prefix_pop();
3582 }
3583 
3584 /*
3585  * The CR3-target count must not be greater than 4. Future processors
3586  * may support a different number of CR3-target values. Software
3587  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3588  * number of values supported.
3589  * [Intel SDM]
3590  */
3591 static void test_cr3_targets(void)
3592 {
3593 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3594 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3595 	unsigned i;
3596 
3597 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3598 	TEST_ASSERT(supported_targets <= 256);
3599 
3600 	try_cr3_target_count(-1u, supported_targets);
3601 	try_cr3_target_count(0x80000000, supported_targets);
3602 	try_cr3_target_count(0x7fffffff, supported_targets);
3603 	for (i = 0; i <= supported_targets + 1; i++)
3604 		try_cr3_target_count(i, supported_targets);
3605 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3606 
3607 	/* VMWRITE to nonexistent target fields should fail. */
3608 	for (i = supported_targets; i < 256; i++)
3609 		TEST_ASSERT(vmcs_write(CR3_TARGET_0 + i*2, 0));
3610 }
3611 
3612 /*
3613  * Test a particular address setting in the VMCS
3614  */
3615 static void test_vmcs_addr(const char *name,
3616 			   enum Encoding encoding,
3617 			   u64 align,
3618 			   bool ignored,
3619 			   bool skip_beyond_mapped_ram,
3620 			   u64 addr)
3621 {
3622 	report_prefix_pushf("%s = %lx", name, addr);
3623 	vmcs_write(encoding, addr);
3624 	if (skip_beyond_mapped_ram &&
3625 	    addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3626 	    addr < (1ul << cpuid_maxphyaddr()))
3627 		printf("Skipping physical address beyond mapped RAM\n");
3628 	else if (ignored || (IS_ALIGNED(addr, align) &&
3629 	    addr < (1ul << cpuid_maxphyaddr())))
3630 		test_vmx_valid_controls();
3631 	else
3632 		test_vmx_invalid_controls();
3633 	report_prefix_pop();
3634 }
3635 
3636 /*
3637  * Test interesting values for a VMCS address
3638  */
3639 static void test_vmcs_addr_values(const char *name,
3640 				  enum Encoding encoding,
3641 				  u64 align,
3642 				  bool ignored,
3643 				  bool skip_beyond_mapped_ram,
3644 				  u32 bit_start, u32 bit_end)
3645 {
3646 	unsigned i;
3647 	u64 orig_val = vmcs_read(encoding);
3648 
3649 	for (i = bit_start; i <= bit_end; i++)
3650 		test_vmcs_addr(name, encoding, align, ignored,
3651 			       skip_beyond_mapped_ram, 1ul << i);
3652 
3653 	test_vmcs_addr(name, encoding, align, ignored,
3654 		       skip_beyond_mapped_ram, PAGE_SIZE - 1);
3655 	test_vmcs_addr(name, encoding, align, ignored,
3656 		       skip_beyond_mapped_ram, PAGE_SIZE);
3657 	test_vmcs_addr(name, encoding, align, ignored,
3658 		       skip_beyond_mapped_ram,
3659 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3660 	test_vmcs_addr(name, encoding, align, ignored,
3661 		       skip_beyond_mapped_ram, -1ul);
3662 
3663 	vmcs_write(encoding, orig_val);
3664 }
3665 
3666 /*
3667  * Test a physical address reference in the VMCS, when the corresponding
3668  * feature is enabled and when the corresponding feature is disabled.
3669  */
3670 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3671 				     const char *field_name,
3672 				     const char *control_name, u64 align,
3673 				     bool skip_beyond_mapped_ram,
3674 				     bool control_primary)
3675 {
3676 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3677 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3678 	u64 page_addr;
3679 
3680 	if (control_primary) {
3681 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3682 			return;
3683 	} else {
3684 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3685 			return;
3686 	}
3687 
3688 	page_addr = vmcs_read(field);
3689 
3690 	report_prefix_pushf("%s enabled", control_name);
3691 	if (control_primary) {
3692 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3693 	} else {
3694 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3695 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3696 	}
3697 
3698 	test_vmcs_addr_values(field_name, field, align, false,
3699 			      skip_beyond_mapped_ram, 0, 63);
3700 	report_prefix_pop();
3701 
3702 	report_prefix_pushf("%s disabled", control_name);
3703 	if (control_primary) {
3704 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3705 	} else {
3706 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3707 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3708 	}
3709 
3710 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3711 	report_prefix_pop();
3712 
3713 	vmcs_write(field, page_addr);
3714 	vmcs_write(CPU_EXEC_CTRL0, primary);
3715 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3716 }
3717 
3718 /*
3719  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3720  * each I/O-bitmap address must be 0. Neither address should set any
3721  * bits beyond the processor's physical-address width.
3722  * [Intel SDM]
3723  */
3724 static void test_io_bitmaps(void)
3725 {
3726 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3727 				 "I/O bitmap A", "Use I/O bitmaps",
3728 				 PAGE_SIZE, false, true);
3729 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3730 				 "I/O bitmap B", "Use I/O bitmaps",
3731 				 PAGE_SIZE, false, true);
3732 }
3733 
3734 /*
3735  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3736  * the MSR-bitmap address must be 0. The address should not set any
3737  * bits beyond the processor's physical-address width.
3738  * [Intel SDM]
3739  */
3740 static void test_msr_bitmap(void)
3741 {
3742 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3743 				 "MSR bitmap", "Use MSR bitmaps",
3744 				 PAGE_SIZE, false, true);
3745 }
3746 
3747 /*
3748  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3749  * address must satisfy the following checks:
3750  * - Bits 11:0 of the address must be 0.
3751  * - The address should not set any bits beyond the processor's
3752  *   physical-address width.
3753  * [Intel SDM]
3754  */
3755 static void test_apic_virt_addr(void)
3756 {
3757 	/*
3758 	 * Ensure the processor will never use the virtual-APIC page, since
3759 	 * we will point it to invalid RAM.  Otherwise KVM is puzzled about
3760 	 * what we're trying to achieve and fails vmentry.
3761 	 */
3762 	u32 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
3763 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0 | CPU_CR8_LOAD | CPU_CR8_STORE);
3764 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3765 				 "virtual-APIC address", "Use TPR shadow",
3766 				 PAGE_SIZE, false, true);
3767 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
3768 }
3769 
3770 /*
3771  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3772  * APIC-access address must satisfy the following checks:
3773  *  - Bits 11:0 of the address must be 0.
3774  *  - The address should not set any bits beyond the processor's
3775  *    physical-address width.
3776  * [Intel SDM]
3777  */
3778 static void test_apic_access_addr(void)
3779 {
3780 	void *apic_access_page = alloc_page();
3781 
3782 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3783 
3784 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3785 				 "APIC-access address",
3786 				 "virtualize APIC-accesses", PAGE_SIZE,
3787 				 true, false);
3788 }
3789 
3790 static bool set_bit_pattern(u8 mask, u32 *secondary)
3791 {
3792 	u8 i;
3793 	bool flag = false;
3794 	u32 test_bits[3] = {
3795 		CPU_VIRT_X2APIC,
3796 		CPU_APIC_REG_VIRT,
3797 		CPU_VINTD
3798 	};
3799 
3800         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3801 		if ((mask & (1u << i)) &&
3802 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3803 			*secondary |= test_bits[i];
3804 			flag = true;
3805 		}
3806 	}
3807 
3808 	return (flag);
3809 }
3810 
3811 /*
3812  * If the "use TPR shadow" VM-execution control is 0, the following
3813  * VM-execution controls must also be 0:
3814  * 	- virtualize x2APIC mode
3815  *	- APIC-register virtualization
3816  *	- virtual-interrupt delivery
3817  *    [Intel SDM]
3818  *
3819  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3820  *    "virtualize APIC accesses" VM-execution control must be 0.
3821  *    [Intel SDM]
3822  */
3823 static void test_apic_virtual_ctls(void)
3824 {
3825 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3826 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3827 	u32 primary = saved_primary;
3828 	u32 secondary = saved_secondary;
3829 	bool is_ctrl_valid = false;
3830 	char str[10] = "disabled";
3831 	u8 i = 0, j;
3832 
3833 	/*
3834 	 * First test
3835 	 */
3836 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3837 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3838 		return;
3839 
3840 	primary |= CPU_SECONDARY;
3841 	primary &= ~CPU_TPR_SHADOW;
3842 	vmcs_write(CPU_EXEC_CTRL0, primary);
3843 
3844 	while (1) {
3845 		for (j = 1; j < 8; j++) {
3846 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3847 			if (primary & CPU_TPR_SHADOW) {
3848 				is_ctrl_valid = true;
3849 			} else {
3850 				if (! set_bit_pattern(j, &secondary))
3851 					is_ctrl_valid = true;
3852 				else
3853 					is_ctrl_valid = false;
3854 			}
3855 
3856 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3857 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3858 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3859 			if (is_ctrl_valid)
3860 				test_vmx_valid_controls();
3861 			else
3862 				test_vmx_invalid_controls();
3863 			report_prefix_pop();
3864 		}
3865 
3866 		if (i == 1)
3867 			break;
3868 		i++;
3869 
3870 		primary |= CPU_TPR_SHADOW;
3871 		vmcs_write(CPU_EXEC_CTRL0, primary);
3872 		strcpy(str, "enabled");
3873 	}
3874 
3875 	/*
3876 	 * Second test
3877 	 */
3878 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3879 
3880 	primary = saved_primary;
3881 	secondary = saved_secondary;
3882 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3883 		return;
3884 
3885 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3886 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3887 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3888 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3889 	test_vmx_valid_controls();
3890 	report_prefix_pop();
3891 
3892 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3893 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3894 	test_vmx_valid_controls();
3895 	report_prefix_pop();
3896 
3897 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3898 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3899 	test_vmx_invalid_controls();
3900 	report_prefix_pop();
3901 
3902 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3903 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3904 	test_vmx_valid_controls();
3905 	report_prefix_pop();
3906 
3907 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3908 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3909 }
3910 
3911 /*
3912  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3913  * "external-interrupt exiting" VM-execution control must be 1.
3914  * [Intel SDM]
3915  */
3916 static void test_virtual_intr_ctls(void)
3917 {
3918 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3919 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3920 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3921 	u32 primary = saved_primary;
3922 	u32 secondary = saved_secondary;
3923 	u32 pin = saved_pin;
3924 
3925 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3926 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
3927 		return;
3928 
3929 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3930 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
3931 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3932 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
3933 	test_vmx_valid_controls();
3934 	report_prefix_pop();
3935 
3936 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
3937 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3938 	test_vmx_invalid_controls();
3939 	report_prefix_pop();
3940 
3941 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
3942 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
3943 	test_vmx_valid_controls();
3944 	report_prefix_pop();
3945 
3946 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3947 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3948 	test_vmx_invalid_controls();
3949 	report_prefix_pop();
3950 
3951 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3952 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3953 	vmcs_write(PIN_CONTROLS, saved_pin);
3954 }
3955 
3956 static void test_pi_desc_addr(u64 addr, bool is_ctrl_valid)
3957 {
3958 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
3959 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
3960 	if (is_ctrl_valid)
3961 		test_vmx_valid_controls();
3962 	else
3963 		test_vmx_invalid_controls();
3964 	report_prefix_pop();
3965 }
3966 
3967 /*
3968  * If the "process posted interrupts" VM-execution control is 1, the
3969  * following must be true:
3970  *
3971  *	- The "virtual-interrupt delivery" VM-execution control is 1.
3972  *	- The "acknowledge interrupt on exit" VM-exit control is 1.
3973  *	- The posted-interrupt notification vector has a value in the
3974  *	- range 0 - 255 (bits 15:8 are all 0).
3975  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
3976  *	- The posted-interrupt descriptor address does not set any bits
3977  *	  beyond the processor's physical-address width.
3978  * [Intel SDM]
3979  */
3980 static void test_posted_intr(void)
3981 {
3982 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3983 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3984 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3985 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
3986 	u32 primary = saved_primary;
3987 	u32 secondary = saved_secondary;
3988 	u32 pin = saved_pin;
3989 	u32 exit_ctl = exit_ctl_saved;
3990 	u16 vec;
3991 	int i;
3992 
3993 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
3994 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3995 	    (ctrl_exit_rev.clr & EXI_INTA)))
3996 		return;
3997 
3998 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3999 
4000 	/*
4001 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
4002 	 */
4003 	pin |= PIN_POST_INTR;
4004 	vmcs_write(PIN_CONTROLS, pin);
4005 	secondary &= ~CPU_VINTD;
4006 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4007 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
4008 	test_vmx_invalid_controls();
4009 	report_prefix_pop();
4010 
4011 	secondary |= CPU_VINTD;
4012 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4013 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
4014 	test_vmx_invalid_controls();
4015 	report_prefix_pop();
4016 
4017 	exit_ctl &= ~EXI_INTA;
4018 	vmcs_write(EXI_CONTROLS, exit_ctl);
4019 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
4020 	test_vmx_invalid_controls();
4021 	report_prefix_pop();
4022 
4023 	exit_ctl |= EXI_INTA;
4024 	vmcs_write(EXI_CONTROLS, exit_ctl);
4025 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4026 	test_vmx_valid_controls();
4027 	report_prefix_pop();
4028 
4029 	secondary &= ~CPU_VINTD;
4030 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4031 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
4032 	test_vmx_invalid_controls();
4033 	report_prefix_pop();
4034 
4035 	secondary |= CPU_VINTD;
4036 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4037 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
4038 	test_vmx_valid_controls();
4039 	report_prefix_pop();
4040 
4041 	/*
4042 	 * Test posted-interrupt notification vector
4043 	 */
4044 	for (i = 0; i < 8; i++) {
4045 		vec = (1ul << i);
4046 		vmcs_write(PINV, vec);
4047 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4048 		test_vmx_valid_controls();
4049 		report_prefix_pop();
4050 	}
4051 	for (i = 8; i < 16; i++) {
4052 		vec = (1ul << i);
4053 		vmcs_write(PINV, vec);
4054 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4055 		test_vmx_invalid_controls();
4056 		report_prefix_pop();
4057 	}
4058 
4059 	vec &= ~(0xff << 8);
4060 	vmcs_write(PINV, vec);
4061 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
4062 	test_vmx_valid_controls();
4063 	report_prefix_pop();
4064 
4065 	/*
4066 	 * Test posted-interrupt descriptor address
4067 	 */
4068 	for (i = 0; i < 6; i++) {
4069 		test_pi_desc_addr(1ul << i, false);
4070 	}
4071 
4072 	test_pi_desc_addr(0xf0, false);
4073 	test_pi_desc_addr(0xff, false);
4074 	test_pi_desc_addr(0x0f, false);
4075 	test_pi_desc_addr(0x8000, true);
4076 	test_pi_desc_addr(0x00, true);
4077 	test_pi_desc_addr(0xc000, true);
4078 
4079 	test_vmcs_addr_values("process-posted interrupts",
4080 			       POSTED_INTR_DESC_ADDR, 64,
4081 			       false, false, 0, 63);
4082 
4083 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4084 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4085 	vmcs_write(PIN_CONTROLS, saved_pin);
4086 }
4087 
4088 static void test_apic_ctls(void)
4089 {
4090 	test_apic_virt_addr();
4091 	test_apic_access_addr();
4092 	test_apic_virtual_ctls();
4093 	test_virtual_intr_ctls();
4094 	test_posted_intr();
4095 }
4096 
4097 /*
4098  * If the "enable VPID" VM-execution control is 1, the value of the
4099  * of the VPID VM-execution control field must not be 0000H.
4100  * [Intel SDM]
4101  */
4102 static void test_vpid(void)
4103 {
4104 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4105 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4106 	u16 vpid = 0x0000;
4107 	int i;
4108 
4109 	if (!is_vpid_supported()) {
4110 		printf("Secondary controls and/or VPID not supported\n");
4111 		return;
4112 	}
4113 
4114 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
4115 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
4116 	vmcs_write(VPID, vpid);
4117 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
4118 	test_vmx_valid_controls();
4119 	report_prefix_pop();
4120 
4121 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
4122 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4123 	test_vmx_invalid_controls();
4124 	report_prefix_pop();
4125 
4126 	for (i = 0; i < 16; i++) {
4127 		vpid = (short)1 << i;;
4128 		vmcs_write(VPID, vpid);
4129 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4130 		test_vmx_valid_controls();
4131 		report_prefix_pop();
4132 	}
4133 
4134 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4135 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4136 }
4137 
4138 static void set_vtpr(unsigned vtpr)
4139 {
4140 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
4141 }
4142 
4143 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
4144 {
4145 	bool valid = true;
4146 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4147 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4148 
4149 	if ((primary & CPU_TPR_SHADOW) &&
4150 	    (!(primary & CPU_SECONDARY) ||
4151 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
4152 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
4153 
4154 	set_vtpr(vtpr);
4155 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
4156 	    threshold, (vtpr >> 4) & 0xf);
4157 	if (valid)
4158 		test_vmx_valid_controls();
4159 	else
4160 		test_vmx_invalid_controls();
4161 	report_prefix_pop();
4162 }
4163 
4164 static void test_invalid_event_injection(void)
4165 {
4166 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
4167 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
4168 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
4169 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
4170 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
4171 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4172 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4173 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4174 	u32 cnt;
4175 
4176 	/* Setup */
4177 	report_prefix_push("invalid event injection");
4178 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4179 	vmcs_write(ENT_INST_LEN, 0x00000001);
4180 
4181 	/* The field's interruption type is not set to a reserved value. */
4182 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4183 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4184 			    "RESERVED interruption type invalid [-]",
4185 			    ent_intr_info);
4186 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4187 	test_vmx_invalid_controls();
4188 	report_prefix_pop();
4189 
4190 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4191 			DE_VECTOR;
4192 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4193 			    "RESERVED interruption type invalid [+]",
4194 			    ent_intr_info);
4195 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4196 	test_vmx_valid_controls();
4197 	report_prefix_pop();
4198 
4199 	/* If the interruption type is other event, the vector is 0. */
4200 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4201 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4202 			    "(OTHER EVENT && vector != 0) invalid [-]",
4203 			    ent_intr_info);
4204 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4205 	test_vmx_invalid_controls();
4206 	report_prefix_pop();
4207 
4208 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4209 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4210 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4211 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4212 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4213 	test_vmx_invalid_controls();
4214 	report_prefix_pop();
4215 
4216 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4217 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4218 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4219 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4220 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4221 	test_vmx_valid_controls();
4222 	report_prefix_pop();
4223 
4224 	/*
4225 	 * If the interruption type
4226 	 * is HW exception, the vector is at most 31.
4227 	 */
4228 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4229 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4230 			    "(HW exception && vector > 31) invalid [-]",
4231 			    ent_intr_info);
4232 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4233 	test_vmx_invalid_controls();
4234 	report_prefix_pop();
4235 
4236 	/*
4237 	 * deliver-error-code is 1 iff either
4238 	 * (a) the "unrestricted guest" VM-execution control is 0
4239 	 * (b) CR0.PE is set.
4240 	 */
4241 
4242 	/* Assert that unrestricted guest is disabled or unsupported */
4243 	assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
4244 	       !(secondary_save & CPU_URG));
4245 
4246 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4247 			GP_VECTOR;
4248 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4249 			    "error code <-> (!URG || prot_mode) [-]",
4250 			    ent_intr_info);
4251 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4252 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4253 	test_vmx_invalid_controls();
4254 	report_prefix_pop();
4255 
4256 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4257 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4258 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4259 			    "error code <-> (!URG || prot_mode) [+]",
4260 			    ent_intr_info);
4261 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4262 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4263 	test_vmx_valid_controls();
4264 	report_prefix_pop();
4265 
4266 	if (enable_unrestricted_guest(false))
4267 		goto skip_unrestricted_guest;
4268 
4269 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4270 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4271 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4272 			    "error code <-> (!URG || prot_mode) [-]",
4273 			    ent_intr_info);
4274 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4275 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4276 	test_vmx_invalid_controls();
4277 	report_prefix_pop();
4278 
4279 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4280 			GP_VECTOR;
4281 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4282 			    "error code <-> (!URG || prot_mode) [-]",
4283 			    ent_intr_info);
4284 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4285 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4286 	test_vmx_invalid_controls();
4287 	report_prefix_pop();
4288 
4289 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4290 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4291 
4292 skip_unrestricted_guest:
4293 	vmcs_write(GUEST_CR0, guest_cr0_save);
4294 
4295 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4296 	report_prefix_push("error code <-> HW exception");
4297 	for (cnt = 0; cnt < 8; cnt++) {
4298 		u32 exception_type_mask = cnt << 8;
4299 		u32 deliver_error_code_mask =
4300 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4301 			INTR_INFO_DELIVER_CODE_MASK : 0;
4302 
4303 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4304 				exception_type_mask | GP_VECTOR;
4305 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4306 				    ent_intr_info);
4307 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4308 		test_vmx_invalid_controls();
4309 		report_prefix_pop();
4310 	}
4311 	report_prefix_pop();
4312 
4313 	/*
4314 	 * deliver-error-code is 1 iff the the vector
4315 	 * indicates an exception that would normally deliver an error code
4316 	 */
4317 	report_prefix_push("error code <-> vector delivers error code");
4318 	for (cnt = 0; cnt < 32; cnt++) {
4319 		bool has_error_code = false;
4320 		u32 deliver_error_code_mask;
4321 
4322 		switch (cnt) {
4323 		case DF_VECTOR:
4324 		case TS_VECTOR:
4325 		case NP_VECTOR:
4326 		case SS_VECTOR:
4327 		case GP_VECTOR:
4328 		case PF_VECTOR:
4329 		case AC_VECTOR:
4330 			has_error_code = true;
4331 		case CP_VECTOR:
4332 			/* Some CPUs have error code and some do not, skip */
4333 			continue;
4334 		}
4335 
4336 		/* Negative case */
4337 		deliver_error_code_mask = has_error_code ?
4338 						0 :
4339 						INTR_INFO_DELIVER_CODE_MASK;
4340 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4341 				INTR_TYPE_HARD_EXCEPTION | cnt;
4342 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4343 				    ent_intr_info);
4344 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4345 		test_vmx_invalid_controls();
4346 		report_prefix_pop();
4347 
4348 		/* Positive case */
4349 		deliver_error_code_mask = has_error_code ?
4350 						INTR_INFO_DELIVER_CODE_MASK :
4351 						0;
4352 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4353 				INTR_TYPE_HARD_EXCEPTION | cnt;
4354 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4355 				    ent_intr_info);
4356 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4357 		test_vmx_valid_controls();
4358 		report_prefix_pop();
4359 	}
4360 	report_prefix_pop();
4361 
4362 	/* Reserved bits in the field (30:12) are 0. */
4363 	report_prefix_push("reserved bits clear");
4364 	for (cnt = 12; cnt <= 30; cnt++) {
4365 		ent_intr_info = ent_intr_info_base |
4366 				INTR_INFO_DELIVER_CODE_MASK |
4367 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4368 				(1U << cnt);
4369 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4370 				    ent_intr_info);
4371 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4372 		test_vmx_invalid_controls();
4373 		report_prefix_pop();
4374 	}
4375 	report_prefix_pop();
4376 
4377 	/*
4378 	 * If deliver-error-code is 1
4379 	 * bits 31:16 of the VM-entry exception error-code field are 0.
4380 	 */
4381 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4382 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4383 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4384 			    "VM-entry exception error code[31:16] clear",
4385 			    ent_intr_info);
4386 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4387 	for (cnt = 16; cnt <= 31; cnt++) {
4388 		ent_intr_err = 1U << cnt;
4389 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4390 				    ent_intr_err);
4391 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4392 		test_vmx_invalid_controls();
4393 		report_prefix_pop();
4394 	}
4395 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4396 	report_prefix_pop();
4397 
4398 	/*
4399 	 * If the interruption type is software interrupt, software exception,
4400 	 * or privileged software exception, the VM-entry instruction-length
4401 	 * field is in the range 0 - 15.
4402 	 */
4403 
4404 	for (cnt = 0; cnt < 3; cnt++) {
4405 		switch (cnt) {
4406 		case 0:
4407 			ent_intr_info = ent_intr_info_base |
4408 					INTR_TYPE_SOFT_INTR;
4409 			break;
4410 		case 1:
4411 			ent_intr_info = ent_intr_info_base |
4412 					INTR_TYPE_SOFT_EXCEPTION;
4413 			break;
4414 		case 2:
4415 			ent_intr_info = ent_intr_info_base |
4416 					INTR_TYPE_PRIV_SW_EXCEPTION;
4417 			break;
4418 		}
4419 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4420 				    "VM-entry instruction-length check",
4421 				    ent_intr_info);
4422 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4423 
4424 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4425 		ent_intr_len = -1;
4426 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4427 				    ent_intr_len);
4428 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4429 		test_vmx_invalid_controls();
4430 		report_prefix_pop();
4431 
4432 		/* Instruction length set to 16 should fail */
4433 		ent_intr_len = 0x00000010;
4434 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4435 				    ent_intr_len);
4436 		vmcs_write(ENT_INST_LEN, 0x00000010);
4437 		test_vmx_invalid_controls();
4438 		report_prefix_pop();
4439 
4440 		report_prefix_pop();
4441 	}
4442 
4443 	/* Cleanup */
4444 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4445 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4446 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4447 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4448 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4449 	vmcs_write(GUEST_CR0, guest_cr0_save);
4450 	report_prefix_pop();
4451 }
4452 
4453 /*
4454  * Test interesting vTPR values for a given TPR threshold.
4455  */
4456 static void test_vtpr_values(unsigned threshold)
4457 {
4458 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4459 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4460 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4461 }
4462 
4463 static void try_tpr_threshold(unsigned threshold)
4464 {
4465 	bool valid = true;
4466 
4467 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4468 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4469 
4470 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4471 	    (secondary & CPU_VINTD)))
4472 		valid = !(threshold >> 4);
4473 
4474 	set_vtpr(-1);
4475 	vmcs_write(TPR_THRESHOLD, threshold);
4476 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4477 	if (valid)
4478 		test_vmx_valid_controls();
4479 	else
4480 		test_vmx_invalid_controls();
4481 	report_prefix_pop();
4482 
4483 	if (valid)
4484 		test_vtpr_values(threshold);
4485 }
4486 
4487 /*
4488  * Test interesting TPR threshold values.
4489  */
4490 static void test_tpr_threshold_values(void)
4491 {
4492 	unsigned i;
4493 
4494 	for (i = 0; i < 0x10; i++)
4495 		try_tpr_threshold(i);
4496 	for (i = 4; i < 32; i++)
4497 		try_tpr_threshold(1u << i);
4498 	try_tpr_threshold(-1u);
4499 	try_tpr_threshold(0x7fffffff);
4500 }
4501 
4502 /*
4503  * This test covers the following two VM entry checks:
4504  *
4505  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4506  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4507  *         31:4 of the TPR threshold VM-execution control field must
4508 	   be 0.
4509  *         [Intel SDM]
4510  *
4511  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4512  *          "virtual-interrupt delivery" VM-execution control is 0
4513  *          and the "virtualize APIC accesses" VM-execution control
4514  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4515  *          control field must not be greater than the value of bits
4516  *          7:4 of VTPR.
4517  *          [Intel SDM]
4518  */
4519 static void test_tpr_threshold(void)
4520 {
4521 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4522 	u64 apic_virt_addr = vmcs_read(APIC_VIRT_ADDR);
4523 	u64 threshold = vmcs_read(TPR_THRESHOLD);
4524 	void *virtual_apic_page;
4525 
4526 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4527 		return;
4528 
4529 	virtual_apic_page = alloc_page();
4530 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4531 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4532 
4533 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4534 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4535 	test_tpr_threshold_values();
4536 	report_prefix_pop();
4537 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4538 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4539 	test_tpr_threshold_values();
4540 	report_prefix_pop();
4541 
4542 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4543 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES))))
4544 		goto out;
4545 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4546 
4547 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4548 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4549 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4550 		test_tpr_threshold_values();
4551 		report_prefix_pop();
4552 
4553 		vmcs_write(CPU_EXEC_CTRL0,
4554 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4555 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4556 		test_tpr_threshold_values();
4557 		report_prefix_pop();
4558 	}
4559 
4560 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4561 		vmcs_write(CPU_EXEC_CTRL0,
4562 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4563 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4564 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4565 		test_tpr_threshold_values();
4566 		report_prefix_pop();
4567 
4568 		vmcs_write(CPU_EXEC_CTRL0,
4569 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4570 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4571 		test_tpr_threshold_values();
4572 		report_prefix_pop();
4573 	}
4574 
4575 	if ((ctrl_cpu_rev[1].clr &
4576 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4577 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4578 		vmcs_write(CPU_EXEC_CTRL0,
4579 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4580 		vmcs_write(CPU_EXEC_CTRL1,
4581 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4582 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4583 		test_tpr_threshold_values();
4584 		report_prefix_pop();
4585 
4586 		vmcs_write(CPU_EXEC_CTRL0,
4587 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4588 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4589 		test_tpr_threshold_values();
4590 		report_prefix_pop();
4591 	}
4592 
4593 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4594 out:
4595 	vmcs_write(TPR_THRESHOLD, threshold);
4596 	vmcs_write(APIC_VIRT_ADDR, apic_virt_addr);
4597 	vmcs_write(CPU_EXEC_CTRL0, primary);
4598 }
4599 
4600 /*
4601  * This test verifies the following two vmentry checks:
4602  *
4603  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4604  *  VM-execution control must be 0.
4605  *  [Intel SDM]
4606  *
4607  *  If the "virtual NMIs" VM-execution control is 0, the "NMI-window
4608  *  exiting" VM-execution control must be 0.
4609  *  [Intel SDM]
4610  */
4611 static void test_nmi_ctrls(void)
4612 {
4613 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4614 
4615 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4616 	    (PIN_NMI | PIN_VIRT_NMI)) {
4617 		printf("NMI exiting and Virtual NMIs are not supported !\n");
4618 		return;
4619 	}
4620 
4621 	/* Save the controls so that we can restore them after our tests */
4622 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4623 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4624 
4625 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4626 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4627 
4628 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4629 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4630 	test_vmx_valid_controls();
4631 	report_prefix_pop();
4632 
4633 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4634 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4635 	test_vmx_invalid_controls();
4636 	report_prefix_pop();
4637 
4638 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4639 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4640 	test_vmx_valid_controls();
4641 	report_prefix_pop();
4642 
4643 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4644 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4645 	test_vmx_valid_controls();
4646 	report_prefix_pop();
4647 
4648 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4649 		report_info("NMI-window exiting is not supported, skipping...");
4650 		goto done;
4651 	}
4652 
4653 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4654 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4655 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4656 	test_vmx_invalid_controls();
4657 	report_prefix_pop();
4658 
4659 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4660 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4661 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4662 	test_vmx_valid_controls();
4663 	report_prefix_pop();
4664 
4665 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4666 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4667 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4668 	test_vmx_valid_controls();
4669 	report_prefix_pop();
4670 
4671 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4672 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4673 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4674 	test_vmx_valid_controls();
4675 	report_prefix_pop();
4676 
4677 	/* Restore the controls to their original values */
4678 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4679 done:
4680 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4681 }
4682 
4683 static void test_eptp_ad_bit(u64 eptp, bool is_ctrl_valid)
4684 {
4685 	vmcs_write(EPTP, eptp);
4686 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4687 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4688 	if (is_ctrl_valid)
4689 		test_vmx_valid_controls();
4690 	else
4691 		test_vmx_invalid_controls();
4692 	report_prefix_pop();
4693 
4694 }
4695 
4696 /*
4697  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4698  *    control field must satisfy the following checks:
4699  *
4700  *     - The EPT memory type (bits 2:0) must be a value supported by the
4701  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4702  *     - Bits 5:3 (1 less than the EPT page-walk length) must indicate a
4703  *	 supported EPT page-walk length.
4704  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4705  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4706  *	 indicating that the processor does not support accessed and dirty
4707  *	 dirty flags for EPT.
4708  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4709  *	 physical-address width) must all be 0.
4710  *
4711  * 2. If the "unrestricted guest" VM-execution control is 1, the
4712  *    "enable EPT" VM-execution control must also be 1.
4713  */
4714 static void test_ept_eptp(void)
4715 {
4716 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4717 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4718 	u64 eptp_saved = vmcs_read(EPTP);
4719 	u32 primary = primary_saved;
4720 	u32 secondary = secondary_saved;
4721 	u64 eptp = eptp_saved;
4722 	u32 i, maxphysaddr;
4723 	u64 j, resv_bits_mask = 0;
4724 
4725 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4726 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4727 		printf("\"CPU secondary\" and/or \"enable EPT\" execution controls are not supported !\n");
4728 		return;
4729 	}
4730 
4731 	/* Support for 4-level EPT is mandatory. */
4732 	report(is_4_level_ept_supported(), "4-level EPT support check");
4733 
4734 	primary |= CPU_SECONDARY;
4735 	vmcs_write(CPU_EXEC_CTRL0, primary);
4736 	secondary |= CPU_EPT;
4737 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4738 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4739 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4740 	vmcs_write(EPTP, eptp);
4741 
4742 	for (i = 0; i < 8; i++) {
4743 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4744 		vmcs_write(EPTP, eptp);
4745 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4746 		    eptp & EPT_MEM_TYPE_MASK);
4747 		if (is_ept_memtype_supported(i))
4748 			test_vmx_valid_controls();
4749 		else
4750 			test_vmx_invalid_controls();
4751 		report_prefix_pop();
4752 	}
4753 
4754 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4755 
4756 	/*
4757 	 * Page walk length (bits 5:3).  Note, the value in VMCS.EPTP "is 1
4758 	 * less than the EPT page-walk length".
4759 	 */
4760 	for (i = 0; i < 8; i++) {
4761 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4762 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4763 
4764 		vmcs_write(EPTP, eptp);
4765 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4766 		    eptp & EPTP_PG_WALK_LEN_MASK);
4767 		if (i == 3 || (i == 4 && is_5_level_ept_supported()))
4768 			test_vmx_valid_controls();
4769 		else
4770 			test_vmx_invalid_controls();
4771 		report_prefix_pop();
4772 	}
4773 
4774 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4775 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4776 
4777 	/*
4778 	 * Accessed and dirty flag (bit 6)
4779 	 */
4780 	if (ept_ad_bits_supported()) {
4781 		report_info("Processor supports accessed and dirty flag");
4782 		eptp &= ~EPTP_AD_FLAG;
4783 		test_eptp_ad_bit(eptp, true);
4784 
4785 		eptp |= EPTP_AD_FLAG;
4786 		test_eptp_ad_bit(eptp, true);
4787 	} else {
4788 		report_info("Processor does not supports accessed and dirty flag");
4789 		eptp &= ~EPTP_AD_FLAG;
4790 		test_eptp_ad_bit(eptp, true);
4791 
4792 		eptp |= EPTP_AD_FLAG;
4793 		test_eptp_ad_bit(eptp, false);
4794 	}
4795 
4796 	/*
4797 	 * Reserved bits [11:7] and [63:N]
4798 	 */
4799 	for (i = 0; i < 32; i++) {
4800 		eptp = (eptp &
4801 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4802 		    (i << EPTP_RESERV_BITS_SHIFT);
4803 		vmcs_write(EPTP, eptp);
4804 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4805 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4806 		    EPTP_RESERV_BITS_MASK);
4807 		if (i == 0)
4808 			test_vmx_valid_controls();
4809 		else
4810 			test_vmx_invalid_controls();
4811 		report_prefix_pop();
4812 	}
4813 
4814 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4815 
4816 	maxphysaddr = cpuid_maxphyaddr();
4817 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4818 		resv_bits_mask |= 1ul << i;
4819 	}
4820 
4821 	for (j = maxphysaddr - 1; j <= 63; j++) {
4822 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4823 		    (j < maxphysaddr ? 0 : 1ul << j);
4824 		vmcs_write(EPTP, eptp);
4825 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4826 		    (eptp >> maxphysaddr) & resv_bits_mask);
4827 		if (j < maxphysaddr)
4828 			test_vmx_valid_controls();
4829 		else
4830 			test_vmx_invalid_controls();
4831 		report_prefix_pop();
4832 	}
4833 
4834 	secondary &= ~(CPU_EPT | CPU_URG);
4835 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4836 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4837 	test_vmx_valid_controls();
4838 	report_prefix_pop();
4839 
4840 	if (!(ctrl_cpu_rev[1].clr & CPU_URG))
4841 		goto skip_unrestricted_guest;
4842 
4843 	secondary |= CPU_URG;
4844 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4845 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4846 	test_vmx_invalid_controls();
4847 	report_prefix_pop();
4848 
4849 	secondary |= CPU_EPT;
4850 	setup_dummy_ept();
4851 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4852 	test_vmx_valid_controls();
4853 	report_prefix_pop();
4854 
4855 skip_unrestricted_guest:
4856 	secondary &= ~CPU_URG;
4857 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4858 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4859 	test_vmx_valid_controls();
4860 	report_prefix_pop();
4861 
4862 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4863 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4864 	vmcs_write(EPTP, eptp_saved);
4865 }
4866 
4867 /*
4868  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4869  * VM-execution control must also be 1. In addition, the PML address
4870  * must satisfy the following checks:
4871  *
4872  *    * Bits 11:0 of the address must be 0.
4873  *    * The address should not set any bits beyond the processor's
4874  *	physical-address width.
4875  *
4876  *  [Intel SDM]
4877  */
4878 static void test_pml(void)
4879 {
4880 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4881 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4882 	u32 primary = primary_saved;
4883 	u32 secondary = secondary_saved;
4884 
4885 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4886 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
4887 		printf("\"Secondary execution\" control or \"enable EPT\" control or \"enable PML\" control is not supported !\n");
4888 		return;
4889 	}
4890 
4891 	primary |= CPU_SECONDARY;
4892 	vmcs_write(CPU_EXEC_CTRL0, primary);
4893 	secondary &= ~(CPU_PML | CPU_EPT);
4894 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4895 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
4896 	test_vmx_valid_controls();
4897 	report_prefix_pop();
4898 
4899 	secondary |= CPU_PML;
4900 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4901 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
4902 	test_vmx_invalid_controls();
4903 	report_prefix_pop();
4904 
4905 	secondary |= CPU_EPT;
4906 	setup_dummy_ept();
4907 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
4908 	test_vmx_valid_controls();
4909 	report_prefix_pop();
4910 
4911 	secondary &= ~CPU_PML;
4912 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4913 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
4914 	test_vmx_valid_controls();
4915 	report_prefix_pop();
4916 
4917 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
4918 				 PAGE_SIZE, false, false);
4919 
4920 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4921 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4922 }
4923 
4924  /*
4925  * If the "activate VMX-preemption timer" VM-execution control is 0, the
4926  * the "save VMX-preemption timer value" VM-exit control must also be 0.
4927  *
4928  *  [Intel SDM]
4929  */
4930 static void test_vmx_preemption_timer(void)
4931 {
4932 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4933 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
4934 	u32 pin = saved_pin;
4935 	u32 exit = saved_exit;
4936 
4937 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
4938 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
4939 		printf("\"Save-VMX-preemption-timer\" control and/or \"Enable-VMX-preemption-timer\" control is not supported\n");
4940 		return;
4941 	}
4942 
4943 	pin |= PIN_PREEMPT;
4944 	vmcs_write(PIN_CONTROLS, pin);
4945 	exit &= ~EXI_SAVE_PREEMPT;
4946 	vmcs_write(EXI_CONTROLS, exit);
4947 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
4948 	test_vmx_valid_controls();
4949 	report_prefix_pop();
4950 
4951 	exit |= EXI_SAVE_PREEMPT;
4952 	vmcs_write(EXI_CONTROLS, exit);
4953 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
4954 	test_vmx_valid_controls();
4955 	report_prefix_pop();
4956 
4957 	pin &= ~PIN_PREEMPT;
4958 	vmcs_write(PIN_CONTROLS, pin);
4959 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
4960 	test_vmx_invalid_controls();
4961 	report_prefix_pop();
4962 
4963 	exit &= ~EXI_SAVE_PREEMPT;
4964 	vmcs_write(EXI_CONTROLS, exit);
4965 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
4966 	test_vmx_valid_controls();
4967 	report_prefix_pop();
4968 
4969 	vmcs_write(PIN_CONTROLS, saved_pin);
4970 	vmcs_write(EXI_CONTROLS, saved_exit);
4971 }
4972 
4973 extern unsigned char test_mtf1;
4974 extern unsigned char test_mtf2;
4975 extern unsigned char test_mtf3;
4976 extern unsigned char test_mtf4;
4977 
4978 static void test_mtf_guest(void)
4979 {
4980 	asm ("vmcall;\n\t"
4981 	     "out %al, $0x80;\n\t"
4982 	     "test_mtf1:\n\t"
4983 	     "vmcall;\n\t"
4984 	     "out %al, $0x80;\n\t"
4985 	     "test_mtf2:\n\t"
4986 	     /*
4987 	      * Prepare for the 'MOV CR3' test. Attempt to induce a
4988 	      * general-protection fault by moving a non-canonical address into
4989 	      * CR3. The 'MOV CR3' instruction does not take an imm64 operand,
4990 	      * so we must MOV the desired value into a register first.
4991 	      *
4992 	      * MOV RAX is done before the VMCALL such that MTF is only enabled
4993 	      * for the instruction under test.
4994 	      */
4995 	     "mov $0xaaaaaaaaaaaaaaaa, %rax;\n\t"
4996 	     "vmcall;\n\t"
4997 	     "mov %rax, %cr3;\n\t"
4998 	     "test_mtf3:\n\t"
4999 	     "vmcall;\n\t"
5000 	     /*
5001 	      * ICEBP/INT1 instruction. Though the instruction is now
5002 	      * documented, don't rely on assemblers enumerating the
5003 	      * instruction. Resort to hand assembly.
5004 	      */
5005 	     ".byte 0xf1;\n\t"
5006 	     "vmcall;\n\t"
5007 	     "test_mtf4:\n\t"
5008 	     "mov $0, %eax;\n\t");
5009 }
5010 
5011 static void test_mtf_gp_handler(struct ex_regs *regs)
5012 {
5013 	regs->rip = (unsigned long) &test_mtf3;
5014 }
5015 
5016 static void test_mtf_db_handler(struct ex_regs *regs)
5017 {
5018 }
5019 
5020 static void enable_mtf(void)
5021 {
5022 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5023 
5024 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 | CPU_MTF);
5025 }
5026 
5027 static void disable_mtf(void)
5028 {
5029 	u32 ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5030 
5031 	vmcs_write(CPU_EXEC_CTRL0, ctrl0 & ~CPU_MTF);
5032 }
5033 
5034 static void enable_tf(void)
5035 {
5036 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5037 
5038 	vmcs_write(GUEST_RFLAGS, rflags | X86_EFLAGS_TF);
5039 }
5040 
5041 static void disable_tf(void)
5042 {
5043 	unsigned long rflags = vmcs_read(GUEST_RFLAGS);
5044 
5045 	vmcs_write(GUEST_RFLAGS, rflags & ~X86_EFLAGS_TF);
5046 }
5047 
5048 static void report_mtf(const char *insn_name, unsigned long exp_rip)
5049 {
5050 	unsigned long rip = vmcs_read(GUEST_RIP);
5051 
5052 	assert_exit_reason(VMX_MTF);
5053 	report(rip == exp_rip, "MTF VM-exit after %s. RIP: 0x%lx (expected 0x%lx)",
5054 	       insn_name, rip, exp_rip);
5055 }
5056 
5057 static void vmx_mtf_test(void)
5058 {
5059 	unsigned long pending_dbg;
5060 	handler old_gp, old_db;
5061 
5062 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5063 		printf("CPU does not support the 'monitor trap flag' processor-based VM-execution control.\n");
5064 		return;
5065 	}
5066 
5067 	test_set_guest(test_mtf_guest);
5068 
5069 	/* Expect an MTF VM-exit after OUT instruction */
5070 	enter_guest();
5071 	skip_exit_vmcall();
5072 
5073 	enable_mtf();
5074 	enter_guest();
5075 	report_mtf("OUT", (unsigned long) &test_mtf1);
5076 	disable_mtf();
5077 
5078 	/*
5079 	 * Concurrent #DB trap and MTF on instruction boundary. Expect MTF
5080 	 * VM-exit with populated 'pending debug exceptions' VMCS field.
5081 	 */
5082 	enter_guest();
5083 	skip_exit_vmcall();
5084 
5085 	enable_mtf();
5086 	enable_tf();
5087 
5088 	enter_guest();
5089 	report_mtf("OUT", (unsigned long) &test_mtf2);
5090 	pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
5091 	report(pending_dbg & DR6_BS,
5092 	       "'pending debug exceptions' field after MTF VM-exit: 0x%lx (expected 0x%lx)",
5093 	       pending_dbg, (unsigned long) DR6_BS);
5094 
5095 	disable_mtf();
5096 	disable_tf();
5097 	vmcs_write(GUEST_PENDING_DEBUG, 0);
5098 
5099 	/*
5100 	 * #GP exception takes priority over MTF. Expect MTF VM-exit with RIP
5101 	 * advanced to first instruction of #GP handler.
5102 	 */
5103 	enter_guest();
5104 	skip_exit_vmcall();
5105 
5106 	old_gp = handle_exception(GP_VECTOR, test_mtf_gp_handler);
5107 
5108 	enable_mtf();
5109 	enter_guest();
5110 	report_mtf("MOV CR3", (unsigned long) get_idt_addr(&boot_idt[GP_VECTOR]));
5111 	disable_mtf();
5112 
5113 	/*
5114 	 * Concurrent MTF and privileged software exception (i.e. ICEBP/INT1).
5115 	 * MTF should follow the delivery of #DB trap, though the SDM doesn't
5116 	 * provide clear indication of the relative priority.
5117 	 */
5118 	enter_guest();
5119 	skip_exit_vmcall();
5120 
5121 	handle_exception(GP_VECTOR, old_gp);
5122 	old_db = handle_exception(DB_VECTOR, test_mtf_db_handler);
5123 
5124 	enable_mtf();
5125 	enter_guest();
5126 	report_mtf("INT1", (unsigned long) get_idt_addr(&boot_idt[DB_VECTOR]));
5127 	disable_mtf();
5128 
5129 	enter_guest();
5130 	skip_exit_vmcall();
5131 	handle_exception(DB_VECTOR, old_db);
5132 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK | INTR_TYPE_OTHER_EVENT);
5133 	enter_guest();
5134 	report_mtf("injected MTF", (unsigned long) &test_mtf4);
5135 	enter_guest();
5136 }
5137 
5138 extern char vmx_mtf_pdpte_guest_begin;
5139 extern char vmx_mtf_pdpte_guest_end;
5140 
5141 asm("vmx_mtf_pdpte_guest_begin:\n\t"
5142     "mov %cr0, %rax\n\t"    /* save CR0 with PG=1                 */
5143     "vmcall\n\t"            /* on return from this CR0.PG=0       */
5144     "mov %rax, %cr0\n\t"    /* restore CR0.PG=1 to enter PAE mode */
5145     "vmcall\n\t"
5146     "retq\n\t"
5147     "vmx_mtf_pdpte_guest_end:");
5148 
5149 static void vmx_mtf_pdpte_test(void)
5150 {
5151 	void *test_mtf_pdpte_guest;
5152 	pteval_t *pdpt;
5153 	u32 guest_ar_cs;
5154 	u64 guest_efer;
5155 	pteval_t *pte;
5156 	u64 guest_cr0;
5157 	u64 guest_cr3;
5158 	u64 guest_cr4;
5159 	u64 ent_ctls;
5160 	int i;
5161 
5162 	if (setup_ept(false))
5163 		return;
5164 
5165 	if (!(ctrl_cpu_rev[0].clr & CPU_MTF)) {
5166 		printf("CPU does not support 'monitor trap flag.'\n");
5167 		return;
5168 	}
5169 
5170 	if (!(ctrl_cpu_rev[1].clr & CPU_URG)) {
5171 		printf("CPU does not support 'unrestricted guest.'\n");
5172 		return;
5173 	}
5174 
5175 	vmcs_write(EXC_BITMAP, ~0);
5176 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
5177 
5178 	/*
5179 	 * Copy the guest code to an identity-mapped page.
5180 	 */
5181 	test_mtf_pdpte_guest = alloc_page();
5182 	memcpy(test_mtf_pdpte_guest, &vmx_mtf_pdpte_guest_begin,
5183 	       &vmx_mtf_pdpte_guest_end - &vmx_mtf_pdpte_guest_begin);
5184 
5185 	test_set_guest(test_mtf_pdpte_guest);
5186 
5187 	enter_guest();
5188 	skip_exit_vmcall();
5189 
5190 	/*
5191 	 * Put the guest in non-paged 32-bit protected mode, ready to enter
5192 	 * PAE mode when CR0.PG is set. CR4.PAE will already have been set
5193 	 * when the guest started out in long mode.
5194 	 */
5195 	ent_ctls = vmcs_read(ENT_CONTROLS);
5196 	vmcs_write(ENT_CONTROLS, ent_ctls & ~ENT_GUEST_64);
5197 
5198 	guest_efer = vmcs_read(GUEST_EFER);
5199 	vmcs_write(GUEST_EFER, guest_efer & ~(EFER_LMA | EFER_LME));
5200 
5201 	/*
5202 	 * Set CS access rights bits for 32-bit protected mode:
5203 	 * 3:0    B execute/read/accessed
5204 	 * 4      1 code or data
5205 	 * 6:5    0 descriptor privilege level
5206 	 * 7      1 present
5207 	 * 11:8   0 reserved
5208 	 * 12     0 available for use by system software
5209 	 * 13     0 64 bit mode not active
5210 	 * 14     1 default operation size 32-bit segment
5211 	 * 15     1 page granularity: segment limit in 4K units
5212 	 * 16     0 segment usable
5213 	 * 31:17  0 reserved
5214 	 */
5215 	guest_ar_cs = vmcs_read(GUEST_AR_CS);
5216 	vmcs_write(GUEST_AR_CS, 0xc09b);
5217 
5218 	guest_cr0 = vmcs_read(GUEST_CR0);
5219 	vmcs_write(GUEST_CR0, guest_cr0 & ~X86_CR0_PG);
5220 
5221 	guest_cr4 = vmcs_read(GUEST_CR4);
5222 	vmcs_write(GUEST_CR4, guest_cr4 & ~X86_CR4_PCIDE);
5223 
5224 	guest_cr3 = vmcs_read(GUEST_CR3);
5225 
5226 	/*
5227 	 * Turn the 4-level page table into a PAE page table by following the 0th
5228 	 * PML4 entry to a PDPT page, and grab the first four PDPTEs from that
5229 	 * page.
5230 	 *
5231 	 * Why does this work?
5232 	 *
5233 	 * PAE uses 32-bit addressing which implies:
5234 	 * Bits 11:0   page offset
5235 	 * Bits 20:12  entry into 512-entry page table
5236 	 * Bits 29:21  entry into a 512-entry directory table
5237 	 * Bits 31:30  entry into the page directory pointer table.
5238 	 * Bits 63:32  zero
5239 	 *
5240 	 * As only 2 bits are needed to select the PDPTEs for the entire
5241 	 * 32-bit address space, take the first 4 PDPTEs in the level 3 page
5242 	 * directory pointer table. It doesn't matter which of these PDPTEs
5243 	 * are present because they must cover the guest code given that it
5244 	 * has already run successfully.
5245 	 *
5246 	 * Get a pointer to PTE for GVA=0 in the page directory pointer table
5247 	 */
5248 	pte = get_pte_level(
5249             (pgd_t *)phys_to_virt(guest_cr3 & ~X86_CR3_PCID_MASK), 0,
5250             PDPT_LEVEL);
5251 
5252 	/*
5253 	 * Need some memory for the 4-entry PAE page directory pointer
5254 	 * table. Use the end of the identity-mapped page where the guest code
5255 	 * is stored. There is definitely space as the guest code is only a
5256 	 * few bytes.
5257 	 */
5258 	pdpt = test_mtf_pdpte_guest + PAGE_SIZE - 4 * sizeof(pteval_t);
5259 
5260 	/*
5261 	 * Copy the first four PDPTEs into the PAE page table with reserved
5262 	 * bits cleared. Note that permission bits from the PML4E and PDPTE
5263 	 * are not propagated.
5264 	 */
5265 	for (i = 0; i < 4; i++) {
5266 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_RSVD_MASK),
5267 				   "PDPTE has invalid reserved bits");
5268 		TEST_ASSERT_EQ_MSG(0, (pte[i] & PDPTE64_PAGE_SIZE_MASK),
5269 				   "Cannot use 1GB super pages for PAE");
5270 		pdpt[i] = pte[i] & ~(PAE_PDPTE_RSVD_MASK);
5271 	}
5272 	vmcs_write(GUEST_CR3, virt_to_phys(pdpt));
5273 
5274 	enable_mtf();
5275 	enter_guest();
5276 	assert_exit_reason(VMX_MTF);
5277 	disable_mtf();
5278 
5279 	/*
5280 	 * The four PDPTEs should have been loaded into the VMCS when
5281 	 * the guest set CR0.PG to enter PAE mode.
5282 	 */
5283 	for (i = 0; i < 4; i++) {
5284 		u64 pdpte = vmcs_read(GUEST_PDPTE + 2 * i);
5285 
5286 		report(pdpte == pdpt[i], "PDPTE%d is 0x%lx (expected 0x%lx)",
5287 		       i, pdpte, pdpt[i]);
5288 	}
5289 
5290 	/*
5291 	 * Now, try to enter the guest in PAE mode. If the PDPTEs in the
5292 	 * vmcs are wrong, this will fail.
5293 	 */
5294 	enter_guest();
5295 	skip_exit_vmcall();
5296 
5297 	/*
5298 	 * Return guest to 64-bit mode and wrap up.
5299 	 */
5300 	vmcs_write(ENT_CONTROLS, ent_ctls);
5301 	vmcs_write(GUEST_EFER, guest_efer);
5302 	vmcs_write(GUEST_AR_CS, guest_ar_cs);
5303 	vmcs_write(GUEST_CR0, guest_cr0);
5304 	vmcs_write(GUEST_CR4, guest_cr4);
5305 	vmcs_write(GUEST_CR3, guest_cr3);
5306 
5307 	enter_guest();
5308 }
5309 
5310 /*
5311  * Tests for VM-execution control fields
5312  */
5313 static void test_vm_execution_ctls(void)
5314 {
5315 	test_pin_based_ctls();
5316 	test_primary_processor_based_ctls();
5317 	test_secondary_processor_based_ctls();
5318 	test_cr3_targets();
5319 	test_io_bitmaps();
5320 	test_msr_bitmap();
5321 	test_apic_ctls();
5322 	test_tpr_threshold();
5323 	test_nmi_ctrls();
5324 	test_pml();
5325 	test_vpid();
5326 	test_ept_eptp();
5327 	test_vmx_preemption_timer();
5328 }
5329 
5330  /*
5331   * The following checks are performed for the VM-entry MSR-load address if
5332   * the VM-entry MSR-load count field is non-zero:
5333   *
5334   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
5335   *      The address should not set any bits beyond the processor's
5336   *      physical-address width.
5337   *
5338   *    - The address of the last byte in the VM-entry MSR-load area
5339   *      should not set any bits beyond the processor's physical-address
5340   *      width. The address of this last byte is VM-entry MSR-load address
5341   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5342   *      uses more bits than the processor's physical-address width.)
5343   *
5344   *
5345   *  [Intel SDM]
5346   */
5347 static void test_entry_msr_load(void)
5348 {
5349 	entry_msr_load = alloc_page();
5350 	u64 tmp;
5351 	u32 entry_msr_ld_cnt = 1;
5352 	int i;
5353 	u32 addr_len = 64;
5354 
5355 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
5356 
5357 	/* Check first 4 bits of VM-entry MSR-load address */
5358 	for (i = 0; i < 4; i++) {
5359 		tmp = (u64)entry_msr_load | 1ull << i;
5360 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
5361 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
5362 				    tmp & 0xf);
5363 		test_vmx_invalid_controls();
5364 		report_prefix_pop();
5365 	}
5366 
5367 	if (basic.val & (1ul << 48))
5368 		addr_len = 32;
5369 
5370 	test_vmcs_addr_values("VM-entry-MSR-load address",
5371 				ENTER_MSR_LD_ADDR, 16, false, false,
5372 				4, addr_len - 1);
5373 
5374 	/*
5375 	 * Check last byte of VM-entry MSR-load address
5376 	 */
5377 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
5378 
5379 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5380 							i < 64; i++) {
5381 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
5382 			1ul << i;
5383 		vmcs_write(ENTER_MSR_LD_ADDR,
5384 			   tmp - (entry_msr_ld_cnt * 16 - 1));
5385 		test_vmx_invalid_controls();
5386 	}
5387 
5388 	vmcs_write(ENT_MSR_LD_CNT, 2);
5389 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5390 	test_vmx_invalid_controls();
5391 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5392 	test_vmx_valid_controls();
5393 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5394 	test_vmx_valid_controls();
5395 }
5396 
5397 static struct vmx_state_area_test_data {
5398 	u32 msr;
5399 	u64 exp;
5400 	bool enabled;
5401 } vmx_state_area_test_data;
5402 
5403 static void guest_state_test_main(void)
5404 {
5405 	u64 obs;
5406 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
5407 
5408 	while (1) {
5409 		if (vmx_get_test_stage() == 2)
5410 			break;
5411 
5412 		if (data->enabled) {
5413 			obs = rdmsr(data->msr);
5414 			report(data->exp == obs,
5415 			       "Guest state is 0x%lx (expected 0x%lx)",
5416 			       obs, data->exp);
5417 		}
5418 
5419 		vmcall();
5420 	}
5421 
5422 	asm volatile("fnop");
5423 }
5424 
5425 static void test_guest_state(const char *test, bool xfail, u64 field,
5426 			     const char * field_name)
5427 {
5428 	struct vmentry_result result;
5429 	u8 abort_flags;
5430 
5431 	abort_flags = ABORT_ON_EARLY_VMENTRY_FAIL;
5432 	if (!xfail)
5433 		abort_flags = ABORT_ON_INVALID_GUEST_STATE;
5434 
5435 	__enter_guest(abort_flags, &result);
5436 
5437 	report(result.exit_reason.failed_vmentry == xfail &&
5438 	       ((xfail && result.exit_reason.basic == VMX_FAIL_STATE) ||
5439 	        (!xfail && result.exit_reason.basic == VMX_VMCALL)) &&
5440 		(!xfail || vmcs_read(EXI_QUALIFICATION) == ENTRY_FAIL_DEFAULT),
5441 	        "%s, %s = %lx", test, field_name, field);
5442 
5443 	if (!result.exit_reason.failed_vmentry)
5444 		skip_exit_insn();
5445 }
5446 
5447 /*
5448  * Tests for VM-entry control fields
5449  */
5450 static void test_vm_entry_ctls(void)
5451 {
5452 	test_invalid_event_injection();
5453 	test_entry_msr_load();
5454 }
5455 
5456 /*
5457  * The following checks are performed for the VM-exit MSR-store address if
5458  * the VM-exit MSR-store count field is non-zero:
5459  *
5460  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
5461  *      The address should not set any bits beyond the processor's
5462  *      physical-address width.
5463  *
5464  *    - The address of the last byte in the VM-exit MSR-store area
5465  *      should not set any bits beyond the processor's physical-address
5466  *      width. The address of this last byte is VM-exit MSR-store address
5467  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5468  *      uses more bits than the processor's physical-address width.)
5469  *
5470  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
5471  * in the range 63:32.
5472  *
5473  *  [Intel SDM]
5474  */
5475 static void test_exit_msr_store(void)
5476 {
5477 	exit_msr_store = alloc_page();
5478 	u64 tmp;
5479 	u32 exit_msr_st_cnt = 1;
5480 	int i;
5481 	u32 addr_len = 64;
5482 
5483 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
5484 
5485 	/* Check first 4 bits of VM-exit MSR-store address */
5486 	for (i = 0; i < 4; i++) {
5487 		tmp = (u64)exit_msr_store | 1ull << i;
5488 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
5489 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
5490 				    tmp & 0xf);
5491 		test_vmx_invalid_controls();
5492 		report_prefix_pop();
5493 	}
5494 
5495 	if (basic.val & (1ul << 48))
5496 		addr_len = 32;
5497 
5498 	test_vmcs_addr_values("VM-exit-MSR-store address",
5499 				EXIT_MSR_ST_ADDR, 16, false, false,
5500 				4, addr_len - 1);
5501 
5502 	/*
5503 	 * Check last byte of VM-exit MSR-store address
5504 	 */
5505 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
5506 
5507 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5508 							i < 64; i++) {
5509 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
5510 			1ul << i;
5511 		vmcs_write(EXIT_MSR_ST_ADDR,
5512 			   tmp - (exit_msr_st_cnt * 16 - 1));
5513 		test_vmx_invalid_controls();
5514 	}
5515 
5516 	vmcs_write(EXI_MSR_ST_CNT, 2);
5517 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5518 	test_vmx_invalid_controls();
5519 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5520 	test_vmx_valid_controls();
5521 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5522 	test_vmx_valid_controls();
5523 }
5524 
5525 /*
5526  * Tests for VM-exit controls
5527  */
5528 static void test_vm_exit_ctls(void)
5529 {
5530 	test_exit_msr_store();
5531 }
5532 
5533 /*
5534  * Check that the virtual CPU checks all of the VMX controls as
5535  * documented in the Intel SDM.
5536  */
5537 static void vmx_controls_test(void)
5538 {
5539 	/*
5540 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
5541 	 * fail due to invalid guest state, should we make it that
5542 	 * far.
5543 	 */
5544 	vmcs_write(GUEST_RFLAGS, 0);
5545 
5546 	test_vm_execution_ctls();
5547 	test_vm_exit_ctls();
5548 	test_vm_entry_ctls();
5549 }
5550 
5551 struct apic_reg_virt_config {
5552 	bool apic_register_virtualization;
5553 	bool use_tpr_shadow;
5554 	bool virtualize_apic_accesses;
5555 	bool virtualize_x2apic_mode;
5556 	bool activate_secondary_controls;
5557 };
5558 
5559 struct apic_reg_test {
5560 	const char *name;
5561 	struct apic_reg_virt_config apic_reg_virt_config;
5562 };
5563 
5564 struct apic_reg_virt_expectation {
5565 	enum Reason rd_exit_reason;
5566 	enum Reason wr_exit_reason;
5567 	u32 val;
5568 	u32 (*virt_fn)(u32);
5569 
5570 	/*
5571 	 * If false, accessing the APIC access address from L2 is treated as a
5572 	 * normal memory operation, rather than triggering virtualization.
5573 	 */
5574 	bool virtualize_apic_accesses;
5575 };
5576 
5577 static u32 apic_virt_identity(u32 val)
5578 {
5579 	return val;
5580 }
5581 
5582 static u32 apic_virt_nibble1(u32 val)
5583 {
5584 	return val & 0xf0;
5585 }
5586 
5587 static u32 apic_virt_byte3(u32 val)
5588 {
5589 	return val & (0xff << 24);
5590 }
5591 
5592 static bool apic_reg_virt_exit_expectation(
5593 	u32 reg, struct apic_reg_virt_config *config,
5594 	struct apic_reg_virt_expectation *expectation)
5595 {
5596 	/* Good configs, where some L2 APIC accesses are virtualized. */
5597 	bool virtualize_apic_accesses_only =
5598 		config->virtualize_apic_accesses &&
5599 		!config->use_tpr_shadow &&
5600 		!config->apic_register_virtualization &&
5601 		!config->virtualize_x2apic_mode &&
5602 		config->activate_secondary_controls;
5603 	bool virtualize_apic_accesses_and_use_tpr_shadow =
5604 		config->virtualize_apic_accesses &&
5605 		config->use_tpr_shadow &&
5606 		!config->apic_register_virtualization &&
5607 		!config->virtualize_x2apic_mode &&
5608 		config->activate_secondary_controls;
5609 	bool apic_register_virtualization =
5610 		config->virtualize_apic_accesses &&
5611 		config->use_tpr_shadow &&
5612 		config->apic_register_virtualization &&
5613 		!config->virtualize_x2apic_mode &&
5614 		config->activate_secondary_controls;
5615 
5616 	expectation->val = MAGIC_VAL_1;
5617 	expectation->virt_fn = apic_virt_identity;
5618 	expectation->virtualize_apic_accesses =
5619 		config->virtualize_apic_accesses &&
5620 		config->activate_secondary_controls;
5621 	if (virtualize_apic_accesses_only) {
5622 		expectation->rd_exit_reason = VMX_APIC_ACCESS;
5623 		expectation->wr_exit_reason = VMX_APIC_ACCESS;
5624 	} else if (virtualize_apic_accesses_and_use_tpr_shadow) {
5625 		switch (reg) {
5626 		case APIC_TASKPRI:
5627 			expectation->rd_exit_reason = VMX_VMCALL;
5628 			expectation->wr_exit_reason = VMX_VMCALL;
5629 			expectation->virt_fn = apic_virt_nibble1;
5630 			break;
5631 		default:
5632 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5633 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5634 		}
5635 	} else if (apic_register_virtualization) {
5636 		expectation->rd_exit_reason = VMX_VMCALL;
5637 
5638 		switch (reg) {
5639 		case APIC_ID:
5640 		case APIC_EOI:
5641 		case APIC_LDR:
5642 		case APIC_DFR:
5643 		case APIC_SPIV:
5644 		case APIC_ESR:
5645 		case APIC_ICR:
5646 		case APIC_LVTT:
5647 		case APIC_LVTTHMR:
5648 		case APIC_LVTPC:
5649 		case APIC_LVT0:
5650 		case APIC_LVT1:
5651 		case APIC_LVTERR:
5652 		case APIC_TMICT:
5653 		case APIC_TDCR:
5654 			expectation->wr_exit_reason = VMX_APIC_WRITE;
5655 			break;
5656 		case APIC_LVR:
5657 		case APIC_ISR ... APIC_ISR + 0x70:
5658 		case APIC_TMR ... APIC_TMR + 0x70:
5659 		case APIC_IRR ... APIC_IRR + 0x70:
5660 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5661 			break;
5662 		case APIC_TASKPRI:
5663 			expectation->wr_exit_reason = VMX_VMCALL;
5664 			expectation->virt_fn = apic_virt_nibble1;
5665 			break;
5666 		case APIC_ICR2:
5667 			expectation->wr_exit_reason = VMX_VMCALL;
5668 			expectation->virt_fn = apic_virt_byte3;
5669 			break;
5670 		default:
5671 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5672 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5673 		}
5674 	} else if (!expectation->virtualize_apic_accesses) {
5675 		/*
5676 		 * No APIC registers are directly virtualized. This includes
5677 		 * VTPR, which can be virtualized through MOV to/from CR8 via
5678 		 * the use TPR shadow control, but not through directly
5679 		 * accessing VTPR.
5680 		 */
5681 		expectation->rd_exit_reason = VMX_VMCALL;
5682 		expectation->wr_exit_reason = VMX_VMCALL;
5683 	} else {
5684 		printf("Cannot parse APIC register virtualization config:\n"
5685 		       "\tvirtualize_apic_accesses: %d\n"
5686 		       "\tuse_tpr_shadow: %d\n"
5687 		       "\tapic_register_virtualization: %d\n"
5688 		       "\tvirtualize_x2apic_mode: %d\n"
5689 		       "\tactivate_secondary_controls: %d\n",
5690 		       config->virtualize_apic_accesses,
5691 		       config->use_tpr_shadow,
5692 		       config->apic_register_virtualization,
5693 		       config->virtualize_x2apic_mode,
5694 		       config->activate_secondary_controls);
5695 
5696 		return false;
5697 	}
5698 
5699 	return true;
5700 }
5701 
5702 struct apic_reg_test apic_reg_tests[] = {
5703 	/* Good configs, where some L2 APIC accesses are virtualized. */
5704 	{
5705 		.name = "Virtualize APIC accesses",
5706 		.apic_reg_virt_config = {
5707 			.virtualize_apic_accesses = true,
5708 			.use_tpr_shadow = false,
5709 			.apic_register_virtualization = false,
5710 			.virtualize_x2apic_mode = false,
5711 			.activate_secondary_controls = true,
5712 		},
5713 	},
5714 	{
5715 		.name = "Virtualize APIC accesses + Use TPR shadow",
5716 		.apic_reg_virt_config = {
5717 			.virtualize_apic_accesses = true,
5718 			.use_tpr_shadow = true,
5719 			.apic_register_virtualization = false,
5720 			.virtualize_x2apic_mode = false,
5721 			.activate_secondary_controls = true,
5722 		},
5723 	},
5724 	{
5725 		.name = "APIC-register virtualization",
5726 		.apic_reg_virt_config = {
5727 			.virtualize_apic_accesses = true,
5728 			.use_tpr_shadow = true,
5729 			.apic_register_virtualization = true,
5730 			.virtualize_x2apic_mode = false,
5731 			.activate_secondary_controls = true,
5732 		},
5733 	},
5734 
5735 	/*
5736 	 * Test that the secondary processor-based VM-execution controls are
5737 	 * correctly ignored when "activate secondary controls" is disabled.
5738 	 */
5739 	{
5740 		.name = "Activate secondary controls off",
5741 		.apic_reg_virt_config = {
5742 			.virtualize_apic_accesses = true,
5743 			.use_tpr_shadow = false,
5744 			.apic_register_virtualization = true,
5745 			.virtualize_x2apic_mode = true,
5746 			.activate_secondary_controls = false,
5747 		},
5748 	},
5749 	{
5750 		.name = "Activate secondary controls off + Use TPR shadow",
5751 		.apic_reg_virt_config = {
5752 			.virtualize_apic_accesses = true,
5753 			.use_tpr_shadow = true,
5754 			.apic_register_virtualization = true,
5755 			.virtualize_x2apic_mode = true,
5756 			.activate_secondary_controls = false,
5757 		},
5758 	},
5759 
5760 	/*
5761 	 * Test that the APIC access address is treated like an arbitrary memory
5762 	 * address when "virtualize APIC accesses" is disabled.
5763 	 */
5764 	{
5765 		.name = "Virtualize APIC accesses off + Use TPR shadow",
5766 		.apic_reg_virt_config = {
5767 			.virtualize_apic_accesses = false,
5768 			.use_tpr_shadow = true,
5769 			.apic_register_virtualization = true,
5770 			.virtualize_x2apic_mode = true,
5771 			.activate_secondary_controls = true,
5772 		},
5773 	},
5774 
5775 	/*
5776 	 * Test that VM entry fails due to invalid controls when
5777 	 * "APIC-register virtualization" is enabled while "use TPR shadow" is
5778 	 * disabled.
5779 	 */
5780 	{
5781 		.name = "APIC-register virtualization + Use TPR shadow off",
5782 		.apic_reg_virt_config = {
5783 			.virtualize_apic_accesses = true,
5784 			.use_tpr_shadow = false,
5785 			.apic_register_virtualization = true,
5786 			.virtualize_x2apic_mode = false,
5787 			.activate_secondary_controls = true,
5788 		},
5789 	},
5790 
5791 	/*
5792 	 * Test that VM entry fails due to invalid controls when
5793 	 * "Virtualize x2APIC mode" is enabled while "use TPR shadow" is
5794 	 * disabled.
5795 	 */
5796 	{
5797 		.name = "Virtualize x2APIC mode + Use TPR shadow off",
5798 		.apic_reg_virt_config = {
5799 			.virtualize_apic_accesses = false,
5800 			.use_tpr_shadow = false,
5801 			.apic_register_virtualization = false,
5802 			.virtualize_x2apic_mode = true,
5803 			.activate_secondary_controls = true,
5804 		},
5805 	},
5806 	{
5807 		.name = "Virtualize x2APIC mode + Use TPR shadow off v2",
5808 		.apic_reg_virt_config = {
5809 			.virtualize_apic_accesses = false,
5810 			.use_tpr_shadow = false,
5811 			.apic_register_virtualization = true,
5812 			.virtualize_x2apic_mode = true,
5813 			.activate_secondary_controls = true,
5814 		},
5815 	},
5816 
5817 	/*
5818 	 * Test that VM entry fails due to invalid controls when
5819 	 * "virtualize x2APIC mode" is enabled while "virtualize APIC accesses"
5820 	 * is enabled.
5821 	 */
5822 	{
5823 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses",
5824 		.apic_reg_virt_config = {
5825 			.virtualize_apic_accesses = true,
5826 			.use_tpr_shadow = true,
5827 			.apic_register_virtualization = false,
5828 			.virtualize_x2apic_mode = true,
5829 			.activate_secondary_controls = true,
5830 		},
5831 	},
5832 	{
5833 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses v2",
5834 		.apic_reg_virt_config = {
5835 			.virtualize_apic_accesses = true,
5836 			.use_tpr_shadow = true,
5837 			.apic_register_virtualization = true,
5838 			.virtualize_x2apic_mode = true,
5839 			.activate_secondary_controls = true,
5840 		},
5841 	},
5842 };
5843 
5844 enum Apic_op {
5845 	APIC_OP_XAPIC_RD,
5846 	APIC_OP_XAPIC_WR,
5847 	TERMINATE,
5848 };
5849 
5850 static u32 vmx_xapic_read(u32 *apic_access_address, u32 reg)
5851 {
5852 	return *(volatile u32 *)((uintptr_t)apic_access_address + reg);
5853 }
5854 
5855 static void vmx_xapic_write(u32 *apic_access_address, u32 reg, u32 val)
5856 {
5857 	*(volatile u32 *)((uintptr_t)apic_access_address + reg) = val;
5858 }
5859 
5860 struct apic_reg_virt_guest_args {
5861 	enum Apic_op op;
5862 	u32 *apic_access_address;
5863 	u32 reg;
5864 	u32 val;
5865 	bool check_rd;
5866 	u32 (*virt_fn)(u32);
5867 } apic_reg_virt_guest_args;
5868 
5869 static void apic_reg_virt_guest(void)
5870 {
5871 	volatile struct apic_reg_virt_guest_args *args =
5872 		&apic_reg_virt_guest_args;
5873 
5874 	for (;;) {
5875 		enum Apic_op op = args->op;
5876 		u32 *apic_access_address = args->apic_access_address;
5877 		u32 reg = args->reg;
5878 		u32 val = args->val;
5879 		bool check_rd = args->check_rd;
5880 		u32 (*virt_fn)(u32) = args->virt_fn;
5881 
5882 		if (op == TERMINATE)
5883 			break;
5884 
5885 		if (op == APIC_OP_XAPIC_RD) {
5886 			u32 ret = vmx_xapic_read(apic_access_address, reg);
5887 
5888 			if (check_rd) {
5889 				u32 want = virt_fn(val);
5890 				u32 got = virt_fn(ret);
5891 
5892 				report(got == want,
5893 				       "read 0x%x, expected 0x%x.", got, want);
5894 			}
5895 		} else if (op == APIC_OP_XAPIC_WR) {
5896 			vmx_xapic_write(apic_access_address, reg, val);
5897 		}
5898 
5899 		/*
5900 		 * The L1 should always execute a vmcall after it's done testing
5901 		 * an individual APIC operation. This helps to validate that the
5902 		 * L1 and L2 are in sync with each other, as expected.
5903 		 */
5904 		vmcall();
5905 	}
5906 }
5907 
5908 static void test_xapic_rd(
5909 	u32 reg, struct apic_reg_virt_expectation *expectation,
5910 	u32 *apic_access_address, u32 *virtual_apic_page)
5911 {
5912 	u32 val = expectation->val;
5913 	u32 exit_reason_want = expectation->rd_exit_reason;
5914 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5915 
5916 	report_prefix_pushf("xapic - reading 0x%03x", reg);
5917 
5918 	/* Configure guest to do an xapic read */
5919 	args->op = APIC_OP_XAPIC_RD;
5920 	args->apic_access_address = apic_access_address;
5921 	args->reg = reg;
5922 	args->val = val;
5923 	args->check_rd = exit_reason_want == VMX_VMCALL;
5924 	args->virt_fn = expectation->virt_fn;
5925 
5926 	/* Setup virtual APIC page */
5927 	if (!expectation->virtualize_apic_accesses) {
5928 		apic_access_address[apic_reg_index(reg)] = val;
5929 		virtual_apic_page[apic_reg_index(reg)] = 0;
5930 	} else if (exit_reason_want == VMX_VMCALL) {
5931 		apic_access_address[apic_reg_index(reg)] = 0;
5932 		virtual_apic_page[apic_reg_index(reg)] = val;
5933 	}
5934 
5935 	/* Enter guest */
5936 	enter_guest();
5937 
5938 	/*
5939 	 * Validate the behavior and
5940 	 * pass a magic value back to the guest.
5941 	 */
5942 	if (exit_reason_want == VMX_APIC_ACCESS) {
5943 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5944 
5945 		assert_exit_reason(exit_reason_want);
5946 		report(apic_page_offset == reg,
5947 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
5948 		       apic_page_offset, reg);
5949 		skip_exit_insn();
5950 
5951 		/* Reenter guest so it can consume/check rcx and exit again. */
5952 		enter_guest();
5953 	} else if (exit_reason_want != VMX_VMCALL) {
5954 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
5955 	}
5956 
5957 	skip_exit_vmcall();
5958 	report_prefix_pop();
5959 }
5960 
5961 static void test_xapic_wr(
5962 	u32 reg, struct apic_reg_virt_expectation *expectation,
5963 	u32 *apic_access_address, u32 *virtual_apic_page)
5964 {
5965 	u32 val = expectation->val;
5966 	u32 exit_reason_want = expectation->wr_exit_reason;
5967 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5968 	bool virtualized =
5969 		expectation->virtualize_apic_accesses &&
5970 		(exit_reason_want == VMX_APIC_WRITE ||
5971 		 exit_reason_want == VMX_VMCALL);
5972 	bool checked = false;
5973 
5974 	report_prefix_pushf("xapic - writing 0x%x to 0x%03x", val, reg);
5975 
5976 	/* Configure guest to do an xapic read */
5977 	args->op = APIC_OP_XAPIC_WR;
5978 	args->apic_access_address = apic_access_address;
5979 	args->reg = reg;
5980 	args->val = val;
5981 
5982 	/* Setup virtual APIC page */
5983 	if (virtualized || !expectation->virtualize_apic_accesses) {
5984 		apic_access_address[apic_reg_index(reg)] = 0;
5985 		virtual_apic_page[apic_reg_index(reg)] = 0;
5986 	}
5987 
5988 	/* Enter guest */
5989 	enter_guest();
5990 
5991 	/*
5992 	 * Validate the behavior and
5993 	 * pass a magic value back to the guest.
5994 	 */
5995 	if (exit_reason_want == VMX_APIC_ACCESS) {
5996 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5997 
5998 		assert_exit_reason(exit_reason_want);
5999 		report(apic_page_offset == reg,
6000 		       "got APIC access exit @ page offset 0x%03x, want 0x%03x",
6001 		       apic_page_offset, reg);
6002 		skip_exit_insn();
6003 
6004 		/* Reenter guest so it can consume/check rcx and exit again. */
6005 		enter_guest();
6006 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6007 		assert_exit_reason(exit_reason_want);
6008 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6009 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%x",
6010 		       apic_reg_index(reg),
6011 		       virtual_apic_page[apic_reg_index(reg)], val);
6012 		checked = true;
6013 
6014 		/* Reenter guest so it can consume/check rcx and exit again. */
6015 		enter_guest();
6016 	} else if (exit_reason_want != VMX_VMCALL) {
6017 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6018 	}
6019 
6020 	assert_exit_reason(VMX_VMCALL);
6021 	if (virtualized && !checked) {
6022 		u32 want = expectation->virt_fn(val);
6023 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6024 		got = expectation->virt_fn(got);
6025 
6026 		report(got == want, "exitless write; val is 0x%x, want 0x%x",
6027 		       got, want);
6028 	} else if (!expectation->virtualize_apic_accesses && !checked) {
6029 		u32 got = apic_access_address[apic_reg_index(reg)];
6030 
6031 		report(got == val,
6032 		       "non-virtualized write; val is 0x%x, want 0x%x", got,
6033 		       val);
6034 	} else if (!expectation->virtualize_apic_accesses && checked) {
6035 		report_fail("Non-virtualized write was prematurely checked!");
6036 	}
6037 
6038 	skip_exit_vmcall();
6039 	report_prefix_pop();
6040 }
6041 
6042 enum Config_type {
6043 	CONFIG_TYPE_GOOD,
6044 	CONFIG_TYPE_UNSUPPORTED,
6045 	CONFIG_TYPE_VMENTRY_FAILS_EARLY,
6046 };
6047 
6048 static enum Config_type configure_apic_reg_virt_test(
6049 	struct apic_reg_virt_config *apic_reg_virt_config)
6050 {
6051 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6052 	u32 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6053 	/* Configs where L2 entry fails early, due to invalid controls. */
6054 	bool use_tpr_shadow_incorrectly_off =
6055 		!apic_reg_virt_config->use_tpr_shadow &&
6056 		(apic_reg_virt_config->apic_register_virtualization ||
6057 		 apic_reg_virt_config->virtualize_x2apic_mode) &&
6058 		apic_reg_virt_config->activate_secondary_controls;
6059 	bool virtualize_apic_accesses_incorrectly_on =
6060 		apic_reg_virt_config->virtualize_apic_accesses &&
6061 		apic_reg_virt_config->virtualize_x2apic_mode &&
6062 		apic_reg_virt_config->activate_secondary_controls;
6063 	bool vmentry_fails_early =
6064 		use_tpr_shadow_incorrectly_off ||
6065 		virtualize_apic_accesses_incorrectly_on;
6066 
6067 	if (apic_reg_virt_config->activate_secondary_controls) {
6068 		if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
6069 			printf("VM-execution control \"activate secondary controls\" NOT supported.\n");
6070 			return CONFIG_TYPE_UNSUPPORTED;
6071 		}
6072 		cpu_exec_ctrl0 |= CPU_SECONDARY;
6073 	} else {
6074 		cpu_exec_ctrl0 &= ~CPU_SECONDARY;
6075 	}
6076 
6077 	if (apic_reg_virt_config->virtualize_apic_accesses) {
6078 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES)) {
6079 			printf("VM-execution control \"virtualize APIC accesses\" NOT supported.\n");
6080 			return CONFIG_TYPE_UNSUPPORTED;
6081 		}
6082 		cpu_exec_ctrl1 |= CPU_VIRT_APIC_ACCESSES;
6083 	} else {
6084 		cpu_exec_ctrl1 &= ~CPU_VIRT_APIC_ACCESSES;
6085 	}
6086 
6087 	if (apic_reg_virt_config->use_tpr_shadow) {
6088 		if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6089 			printf("VM-execution control \"use TPR shadow\" NOT supported.\n");
6090 			return CONFIG_TYPE_UNSUPPORTED;
6091 		}
6092 		cpu_exec_ctrl0 |= CPU_TPR_SHADOW;
6093 	} else {
6094 		cpu_exec_ctrl0 &= ~CPU_TPR_SHADOW;
6095 	}
6096 
6097 	if (apic_reg_virt_config->apic_register_virtualization) {
6098 		if (!(ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT)) {
6099 			printf("VM-execution control \"APIC-register virtualization\" NOT supported.\n");
6100 			return CONFIG_TYPE_UNSUPPORTED;
6101 		}
6102 		cpu_exec_ctrl1 |= CPU_APIC_REG_VIRT;
6103 	} else {
6104 		cpu_exec_ctrl1 &= ~CPU_APIC_REG_VIRT;
6105 	}
6106 
6107 	if (apic_reg_virt_config->virtualize_x2apic_mode) {
6108 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_X2APIC)) {
6109 			printf("VM-execution control \"virtualize x2APIC mode\" NOT supported.\n");
6110 			return CONFIG_TYPE_UNSUPPORTED;
6111 		}
6112 		cpu_exec_ctrl1 |= CPU_VIRT_X2APIC;
6113 	} else {
6114 		cpu_exec_ctrl1 &= ~CPU_VIRT_X2APIC;
6115 	}
6116 
6117 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6118 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6119 
6120 	if (vmentry_fails_early)
6121 		return CONFIG_TYPE_VMENTRY_FAILS_EARLY;
6122 
6123 	return CONFIG_TYPE_GOOD;
6124 }
6125 
6126 static bool cpu_has_apicv(void)
6127 {
6128 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
6129 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
6130 		(ctrl_pin_rev.clr & PIN_POST_INTR));
6131 }
6132 
6133 /* Validates APIC register access across valid virtualization configurations. */
6134 static void apic_reg_virt_test(void)
6135 {
6136 	u32 *apic_access_address;
6137 	u32 *virtual_apic_page;
6138 	u64 control;
6139 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6140 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6141 	int i;
6142 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
6143 
6144 	if (!cpu_has_apicv()) {
6145 		report_skip(__func__);
6146 		return;
6147 	}
6148 
6149 	control = cpu_exec_ctrl1;
6150 	control &= ~CPU_VINTD;
6151 	vmcs_write(CPU_EXEC_CTRL1, control);
6152 
6153 	test_set_guest(apic_reg_virt_guest);
6154 
6155 	/*
6156 	 * From the SDM: The 1-setting of the "virtualize APIC accesses"
6157 	 * VM-execution is guaranteed to apply only if translations to the
6158 	 * APIC-access address use a 4-KByte page.
6159 	 */
6160 	apic_access_address = alloc_page();
6161 	force_4k_page(apic_access_address);
6162 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_address));
6163 
6164 	virtual_apic_page = alloc_page();
6165 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6166 
6167 	for (i = 0; i < ARRAY_SIZE(apic_reg_tests); i++) {
6168 		struct apic_reg_test *apic_reg_test = &apic_reg_tests[i];
6169 		struct apic_reg_virt_config *apic_reg_virt_config =
6170 				&apic_reg_test->apic_reg_virt_config;
6171 		enum Config_type config_type;
6172 		u32 reg;
6173 
6174 		printf("--- %s test ---\n", apic_reg_test->name);
6175 		config_type =
6176 			configure_apic_reg_virt_test(apic_reg_virt_config);
6177 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6178 			printf("Skip because of missing features.\n");
6179 			continue;
6180 		}
6181 
6182 		if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6183 			enter_guest_with_bad_controls();
6184 			continue;
6185 		}
6186 
6187 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6188 			struct apic_reg_virt_expectation expectation = {};
6189 			bool ok;
6190 
6191 			ok = apic_reg_virt_exit_expectation(
6192 				reg, apic_reg_virt_config, &expectation);
6193 			if (!ok) {
6194 				report_fail("Malformed test.");
6195 				break;
6196 			}
6197 
6198 			test_xapic_rd(reg, &expectation, apic_access_address,
6199 				      virtual_apic_page);
6200 			test_xapic_wr(reg, &expectation, apic_access_address,
6201 				      virtual_apic_page);
6202 		}
6203 	}
6204 
6205 	/* Terminate the guest */
6206 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6207 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6208 	args->op = TERMINATE;
6209 	enter_guest();
6210 	assert_exit_reason(VMX_VMCALL);
6211 }
6212 
6213 struct virt_x2apic_mode_config {
6214 	struct apic_reg_virt_config apic_reg_virt_config;
6215 	bool virtual_interrupt_delivery;
6216 	bool use_msr_bitmaps;
6217 	bool disable_x2apic_msr_intercepts;
6218 	bool disable_x2apic;
6219 };
6220 
6221 struct virt_x2apic_mode_test_case {
6222 	const char *name;
6223 	struct virt_x2apic_mode_config virt_x2apic_mode_config;
6224 };
6225 
6226 enum Virt_x2apic_mode_behavior_type {
6227 	X2APIC_ACCESS_VIRTUALIZED,
6228 	X2APIC_ACCESS_PASSED_THROUGH,
6229 	X2APIC_ACCESS_TRIGGERS_GP,
6230 };
6231 
6232 struct virt_x2apic_mode_expectation {
6233 	enum Reason rd_exit_reason;
6234 	enum Reason wr_exit_reason;
6235 
6236 	/*
6237 	 * RDMSR and WRMSR handle 64-bit values. However, except for ICR, all of
6238 	 * the x2APIC registers are 32 bits. Notice:
6239 	 *   1. vmx_x2apic_read() clears the upper 32 bits for 32-bit registers.
6240 	 *   2. vmx_x2apic_write() expects the val arg to be well-formed.
6241 	 */
6242 	u64 rd_val;
6243 	u64 wr_val;
6244 
6245 	/*
6246 	 * Compares input to virtualized output;
6247 	 * 1st arg is pointer to return expected virtualization output.
6248 	 */
6249 	u64 (*virt_fn)(u64);
6250 
6251 	enum Virt_x2apic_mode_behavior_type rd_behavior;
6252 	enum Virt_x2apic_mode_behavior_type wr_behavior;
6253 	bool wr_only;
6254 };
6255 
6256 static u64 virt_x2apic_mode_identity(u64 val)
6257 {
6258 	return val;
6259 }
6260 
6261 static u64 virt_x2apic_mode_nibble1(u64 val)
6262 {
6263 	return val & 0xf0;
6264 }
6265 
6266 static void virt_x2apic_mode_rd_expectation(
6267 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6268 	bool apic_register_virtualization, bool virtual_interrupt_delivery,
6269 	struct virt_x2apic_mode_expectation *expectation)
6270 {
6271 	bool readable =
6272 		!x2apic_reg_reserved(reg) &&
6273 		reg != APIC_EOI;
6274 
6275 	expectation->rd_exit_reason = VMX_VMCALL;
6276 	expectation->virt_fn = virt_x2apic_mode_identity;
6277 	if (virt_x2apic_mode_on && apic_register_virtualization) {
6278 		expectation->rd_val = MAGIC_VAL_1;
6279 		if (reg == APIC_PROCPRI && virtual_interrupt_delivery)
6280 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6281 		else if (reg == APIC_TASKPRI)
6282 			expectation->virt_fn = virt_x2apic_mode_nibble1;
6283 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6284 	} else if (virt_x2apic_mode_on && !apic_register_virtualization &&
6285 		   reg == APIC_TASKPRI) {
6286 		expectation->rd_val = MAGIC_VAL_1;
6287 		expectation->virt_fn = virt_x2apic_mode_nibble1;
6288 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
6289 	} else if (!disable_x2apic && readable) {
6290 		expectation->rd_val = apic_read(reg);
6291 		expectation->rd_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6292 	} else {
6293 		expectation->rd_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6294 	}
6295 }
6296 
6297 /*
6298  * get_x2apic_wr_val() creates an innocuous write value for an x2APIC register.
6299  *
6300  * For writable registers, get_x2apic_wr_val() deposits the write value into the
6301  * val pointer arg and returns true. For non-writable registers, val is not
6302  * modified and get_x2apic_wr_val() returns false.
6303  */
6304 static bool get_x2apic_wr_val(u32 reg, u64 *val)
6305 {
6306 	switch (reg) {
6307 	case APIC_TASKPRI:
6308 		/* Bits 31:8 are reserved. */
6309 		*val &= 0xff;
6310 		break;
6311 	case APIC_EOI:
6312 	case APIC_ESR:
6313 	case APIC_TMICT:
6314 		/*
6315 		 * EOI, ESR: WRMSR of a non-zero value causes #GP(0).
6316 		 * TMICT: A write of 0 to the initial-count register effectively
6317 		 *        stops the local APIC timer, in both one-shot and
6318 		 *        periodic mode.
6319 		 */
6320 		*val = 0;
6321 		break;
6322 	case APIC_SPIV:
6323 	case APIC_LVTT:
6324 	case APIC_LVTTHMR:
6325 	case APIC_LVTPC:
6326 	case APIC_LVT0:
6327 	case APIC_LVT1:
6328 	case APIC_LVTERR:
6329 	case APIC_TDCR:
6330 		/*
6331 		 * To avoid writing a 1 to a reserved bit or causing some other
6332 		 * unintended side effect, read the current value and use it as
6333 		 * the write value.
6334 		 */
6335 		*val = apic_read(reg);
6336 		break;
6337 	case APIC_CMCI:
6338 		if (!apic_lvt_entry_supported(6))
6339 			return false;
6340 		*val = apic_read(reg);
6341 		break;
6342 	case APIC_ICR:
6343 		*val = 0x40000 | 0xf1;
6344 		break;
6345 	case APIC_SELF_IPI:
6346 		/*
6347 		 * With special processing (i.e., virtualize x2APIC mode +
6348 		 * virtual interrupt delivery), writing zero causes an
6349 		 * APIC-write VM exit. We plan to add a test for enabling
6350 		 * "virtual-interrupt delivery" in VMCS12, and that's where we
6351 		 * will test a self IPI with special processing.
6352 		 */
6353 		*val = 0x0;
6354 		break;
6355 	default:
6356 		return false;
6357 	}
6358 
6359 	return true;
6360 }
6361 
6362 static bool special_processing_applies(u32 reg, u64 *val,
6363 				       bool virt_int_delivery)
6364 {
6365 	bool special_processing =
6366 		(reg == APIC_TASKPRI) ||
6367 		(virt_int_delivery &&
6368 		 (reg == APIC_EOI || reg == APIC_SELF_IPI));
6369 
6370 	if (special_processing) {
6371 		TEST_ASSERT(get_x2apic_wr_val(reg, val));
6372 		return true;
6373 	}
6374 
6375 	return false;
6376 }
6377 
6378 static void virt_x2apic_mode_wr_expectation(
6379 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
6380 	bool virt_int_delivery,
6381 	struct virt_x2apic_mode_expectation *expectation)
6382 {
6383 	expectation->wr_exit_reason = VMX_VMCALL;
6384 	expectation->wr_val = MAGIC_VAL_1;
6385 	expectation->wr_only = false;
6386 
6387 	if (virt_x2apic_mode_on &&
6388 	    special_processing_applies(reg, &expectation->wr_val,
6389 				       virt_int_delivery)) {
6390 		expectation->wr_behavior = X2APIC_ACCESS_VIRTUALIZED;
6391 		if (reg == APIC_SELF_IPI)
6392 			expectation->wr_exit_reason = VMX_APIC_WRITE;
6393 	} else if (!disable_x2apic &&
6394 		   get_x2apic_wr_val(reg, &expectation->wr_val)) {
6395 		expectation->wr_behavior = X2APIC_ACCESS_PASSED_THROUGH;
6396 		if (reg == APIC_EOI || reg == APIC_SELF_IPI)
6397 			expectation->wr_only = true;
6398 		if (reg == APIC_ICR)
6399 			expectation->wr_exit_reason = VMX_EXTINT;
6400 	} else {
6401 		expectation->wr_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6402 		/*
6403 		 * Writing 1 to a reserved bit triggers a #GP.
6404 		 * Thus, set the write value to 0, which seems
6405 		 * the most likely to detect a missed #GP.
6406 		 */
6407 		expectation->wr_val = 0;
6408 	}
6409 }
6410 
6411 static void virt_x2apic_mode_exit_expectation(
6412 	u32 reg, struct virt_x2apic_mode_config *config,
6413 	struct virt_x2apic_mode_expectation *expectation)
6414 {
6415 	struct apic_reg_virt_config *base_config =
6416 		&config->apic_reg_virt_config;
6417 	bool virt_x2apic_mode_on =
6418 		base_config->virtualize_x2apic_mode &&
6419 		config->use_msr_bitmaps &&
6420 		config->disable_x2apic_msr_intercepts &&
6421 		base_config->activate_secondary_controls;
6422 
6423 	virt_x2apic_mode_wr_expectation(
6424 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6425 		config->virtual_interrupt_delivery, expectation);
6426 	virt_x2apic_mode_rd_expectation(
6427 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6428 		base_config->apic_register_virtualization,
6429 		config->virtual_interrupt_delivery, expectation);
6430 }
6431 
6432 struct virt_x2apic_mode_test_case virt_x2apic_mode_tests[] = {
6433 	/*
6434 	 * Baseline "virtualize x2APIC mode" configuration:
6435 	 *   - virtualize x2APIC mode
6436 	 *   - virtual-interrupt delivery
6437 	 *   - APIC-register virtualization
6438 	 *   - x2APIC MSR intercepts disabled
6439 	 *
6440 	 * Reads come from virtual APIC page, special processing applies to
6441 	 * VTPR, EOI, and SELF IPI, and all other writes pass through to L1
6442 	 * APIC.
6443 	 */
6444 	{
6445 		.name = "Baseline",
6446 		.virt_x2apic_mode_config = {
6447 			.virtual_interrupt_delivery = true,
6448 			.use_msr_bitmaps = true,
6449 			.disable_x2apic_msr_intercepts = true,
6450 			.disable_x2apic = false,
6451 			.apic_reg_virt_config = {
6452 				.apic_register_virtualization = true,
6453 				.use_tpr_shadow = true,
6454 				.virtualize_apic_accesses = false,
6455 				.virtualize_x2apic_mode = true,
6456 				.activate_secondary_controls = true,
6457 			},
6458 		},
6459 	},
6460 	{
6461 		.name = "Baseline w/ x2apic disabled",
6462 		.virt_x2apic_mode_config = {
6463 			.virtual_interrupt_delivery = true,
6464 			.use_msr_bitmaps = true,
6465 			.disable_x2apic_msr_intercepts = true,
6466 			.disable_x2apic = true,
6467 			.apic_reg_virt_config = {
6468 				.apic_register_virtualization = true,
6469 				.use_tpr_shadow = true,
6470 				.virtualize_apic_accesses = false,
6471 				.virtualize_x2apic_mode = true,
6472 				.activate_secondary_controls = true,
6473 			},
6474 		},
6475 	},
6476 
6477 	/*
6478 	 * Baseline, minus virtual-interrupt delivery. Reads come from virtual
6479 	 * APIC page, special processing applies to VTPR, and all other writes
6480 	 * pass through to L1 APIC.
6481 	 */
6482 	{
6483 		.name = "Baseline - virtual interrupt delivery",
6484 		.virt_x2apic_mode_config = {
6485 			.virtual_interrupt_delivery = false,
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 	/*
6500 	 * Baseline, minus APIC-register virtualization. x2APIC reads pass
6501 	 * through to L1's APIC, unless reading VTPR
6502 	 */
6503 	{
6504 		.name = "Virtualize x2APIC mode, no APIC reg virt",
6505 		.virt_x2apic_mode_config = {
6506 			.virtual_interrupt_delivery = true,
6507 			.use_msr_bitmaps = true,
6508 			.disable_x2apic_msr_intercepts = true,
6509 			.disable_x2apic = false,
6510 			.apic_reg_virt_config = {
6511 				.apic_register_virtualization = false,
6512 				.use_tpr_shadow = true,
6513 				.virtualize_apic_accesses = false,
6514 				.virtualize_x2apic_mode = true,
6515 				.activate_secondary_controls = true,
6516 			},
6517 		},
6518 	},
6519 	{
6520 		.name = "Virtualize x2APIC mode, no APIC reg virt, x2APIC off",
6521 		.virt_x2apic_mode_config = {
6522 			.virtual_interrupt_delivery = true,
6523 			.use_msr_bitmaps = true,
6524 			.disable_x2apic_msr_intercepts = true,
6525 			.disable_x2apic = true,
6526 			.apic_reg_virt_config = {
6527 				.apic_register_virtualization = false,
6528 				.use_tpr_shadow = true,
6529 				.virtualize_apic_accesses = false,
6530 				.virtualize_x2apic_mode = true,
6531 				.activate_secondary_controls = true,
6532 			},
6533 		},
6534 	},
6535 
6536 	/*
6537 	 * Enable "virtualize x2APIC mode" and "APIC-register virtualization",
6538 	 * and disable intercepts for the x2APIC MSRs, but fail to enable
6539 	 * "activate secondary controls" (i.e. L2 gets access to L1's x2APIC
6540 	 * MSRs).
6541 	 */
6542 	{
6543 		.name = "Fail to enable activate secondary controls",
6544 		.virt_x2apic_mode_config = {
6545 			.virtual_interrupt_delivery = true,
6546 			.use_msr_bitmaps = true,
6547 			.disable_x2apic_msr_intercepts = true,
6548 			.disable_x2apic = false,
6549 			.apic_reg_virt_config = {
6550 				.apic_register_virtualization = true,
6551 				.use_tpr_shadow = true,
6552 				.virtualize_apic_accesses = false,
6553 				.virtualize_x2apic_mode = true,
6554 				.activate_secondary_controls = false,
6555 			},
6556 		},
6557 	},
6558 
6559 	/*
6560 	 * Enable "APIC-register virtualization" and enable "activate secondary
6561 	 * controls" and disable intercepts for the x2APIC MSRs, but do not
6562 	 * enable the "virtualize x2APIC mode" VM-execution control (i.e. L2
6563 	 * gets access to L1's x2APIC MSRs).
6564 	 */
6565 	{
6566 		.name = "Fail to enable virtualize x2APIC mode",
6567 		.virt_x2apic_mode_config = {
6568 			.virtual_interrupt_delivery = true,
6569 			.use_msr_bitmaps = true,
6570 			.disable_x2apic_msr_intercepts = true,
6571 			.disable_x2apic = false,
6572 			.apic_reg_virt_config = {
6573 				.apic_register_virtualization = true,
6574 				.use_tpr_shadow = true,
6575 				.virtualize_apic_accesses = false,
6576 				.virtualize_x2apic_mode = false,
6577 				.activate_secondary_controls = true,
6578 			},
6579 		},
6580 	},
6581 
6582 	/*
6583 	 * Disable "Virtualize x2APIC mode", disable x2APIC MSR intercepts, and
6584 	 * enable "APIC-register virtualization" --> L2 gets L1's x2APIC MSRs.
6585 	 */
6586 	{
6587 		.name = "Baseline",
6588 		.virt_x2apic_mode_config = {
6589 			.virtual_interrupt_delivery = true,
6590 			.use_msr_bitmaps = true,
6591 			.disable_x2apic_msr_intercepts = true,
6592 			.disable_x2apic = false,
6593 			.apic_reg_virt_config = {
6594 				.apic_register_virtualization = true,
6595 				.use_tpr_shadow = true,
6596 				.virtualize_apic_accesses = false,
6597 				.virtualize_x2apic_mode = false,
6598 				.activate_secondary_controls = true,
6599 			},
6600 		},
6601 	},
6602 };
6603 
6604 enum X2apic_op {
6605 	X2APIC_OP_RD,
6606 	X2APIC_OP_WR,
6607 	X2APIC_TERMINATE,
6608 };
6609 
6610 static u64 vmx_x2apic_read(u32 reg)
6611 {
6612 	u32 msr_addr = x2apic_msr(reg);
6613 	u64 val;
6614 
6615 	val = rdmsr(msr_addr);
6616 
6617 	return val;
6618 }
6619 
6620 static void vmx_x2apic_write(u32 reg, u64 val)
6621 {
6622 	u32 msr_addr = x2apic_msr(reg);
6623 
6624 	wrmsr(msr_addr, val);
6625 }
6626 
6627 struct virt_x2apic_mode_guest_args {
6628 	enum X2apic_op op;
6629 	u32 reg;
6630 	u64 val;
6631 	bool should_gp;
6632 	u64 (*virt_fn)(u64);
6633 } virt_x2apic_mode_guest_args;
6634 
6635 static volatile bool handle_x2apic_gp_ran;
6636 static volatile u32 handle_x2apic_gp_insn_len;
6637 static void handle_x2apic_gp(struct ex_regs *regs)
6638 {
6639 	handle_x2apic_gp_ran = true;
6640 	regs->rip += handle_x2apic_gp_insn_len;
6641 }
6642 
6643 static handler setup_x2apic_gp_handler(void)
6644 {
6645 	handler old_handler;
6646 
6647 	old_handler = handle_exception(GP_VECTOR, handle_x2apic_gp);
6648 	/* RDMSR and WRMSR are both 2 bytes, assuming no prefixes. */
6649 	handle_x2apic_gp_insn_len = 2;
6650 
6651 	return old_handler;
6652 }
6653 
6654 static void teardown_x2apic_gp_handler(handler old_handler)
6655 {
6656 	handle_exception(GP_VECTOR, old_handler);
6657 
6658 	/*
6659 	 * Defensively reset instruction length, so that if the handler is
6660 	 * incorrectly used, it will loop infinitely, rather than run off into
6661 	 * la la land.
6662 	 */
6663 	handle_x2apic_gp_insn_len = 0;
6664 	handle_x2apic_gp_ran = false;
6665 }
6666 
6667 static void virt_x2apic_mode_guest(void)
6668 {
6669 	volatile struct virt_x2apic_mode_guest_args *args =
6670 		&virt_x2apic_mode_guest_args;
6671 
6672 	for (;;) {
6673 		enum X2apic_op op = args->op;
6674 		u32 reg = args->reg;
6675 		u64 val = args->val;
6676 		bool should_gp = args->should_gp;
6677 		u64 (*virt_fn)(u64) = args->virt_fn;
6678 		handler old_handler;
6679 
6680 		if (op == X2APIC_TERMINATE)
6681 			break;
6682 
6683 		if (should_gp) {
6684 			TEST_ASSERT(!handle_x2apic_gp_ran);
6685 			old_handler = setup_x2apic_gp_handler();
6686 		}
6687 
6688 		if (op == X2APIC_OP_RD) {
6689 			u64 ret = vmx_x2apic_read(reg);
6690 
6691 			if (!should_gp) {
6692 				u64 want = virt_fn(val);
6693 				u64 got = virt_fn(ret);
6694 
6695 				report(got == want,
6696 				       "APIC read; got 0x%lx, want 0x%lx.",
6697 				       got, want);
6698 			}
6699 		} else if (op == X2APIC_OP_WR) {
6700 			vmx_x2apic_write(reg, val);
6701 		}
6702 
6703 		if (should_gp) {
6704 			report(handle_x2apic_gp_ran,
6705 			       "x2APIC op triggered GP.");
6706 			teardown_x2apic_gp_handler(old_handler);
6707 		}
6708 
6709 		/*
6710 		 * The L1 should always execute a vmcall after it's done testing
6711 		 * an individual APIC operation. This helps to validate that the
6712 		 * L1 and L2 are in sync with each other, as expected.
6713 		 */
6714 		vmcall();
6715 	}
6716 }
6717 
6718 static void test_x2apic_rd(
6719 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6720 	u32 *virtual_apic_page)
6721 {
6722 	u64 val = expectation->rd_val;
6723 	u32 exit_reason_want = expectation->rd_exit_reason;
6724 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6725 
6726 	report_prefix_pushf("x2apic - reading 0x%03x", reg);
6727 
6728 	/* Configure guest to do an x2apic read */
6729 	args->op = X2APIC_OP_RD;
6730 	args->reg = reg;
6731 	args->val = val;
6732 	args->should_gp = expectation->rd_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6733 	args->virt_fn = expectation->virt_fn;
6734 
6735 	/* Setup virtual APIC page */
6736 	if (expectation->rd_behavior == X2APIC_ACCESS_VIRTUALIZED)
6737 		virtual_apic_page[apic_reg_index(reg)] = (u32)val;
6738 
6739 	/* Enter guest */
6740 	enter_guest();
6741 
6742 	if (exit_reason_want != VMX_VMCALL) {
6743 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6744 	}
6745 
6746 	skip_exit_vmcall();
6747 	report_prefix_pop();
6748 }
6749 
6750 static volatile bool handle_x2apic_ipi_ran;
6751 static void handle_x2apic_ipi(isr_regs_t *regs)
6752 {
6753 	handle_x2apic_ipi_ran = true;
6754 	eoi();
6755 }
6756 
6757 static void test_x2apic_wr(
6758 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6759 	u32 *virtual_apic_page)
6760 {
6761 	u64 val = expectation->wr_val;
6762 	u32 exit_reason_want = expectation->wr_exit_reason;
6763 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6764 	int ipi_vector = 0xf1;
6765 	u32 restore_val = 0;
6766 
6767 	report_prefix_pushf("x2apic - writing 0x%lx to 0x%03x", val, reg);
6768 
6769 	/* Configure guest to do an x2apic read */
6770 	args->op = X2APIC_OP_WR;
6771 	args->reg = reg;
6772 	args->val = val;
6773 	args->should_gp = expectation->wr_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6774 
6775 	/* Setup virtual APIC page */
6776 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED)
6777 		virtual_apic_page[apic_reg_index(reg)] = 0;
6778 	if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH && !expectation->wr_only)
6779 		restore_val = apic_read(reg);
6780 
6781 	/* Setup IPI handler */
6782 	handle_x2apic_ipi_ran = false;
6783 	handle_irq(ipi_vector, handle_x2apic_ipi);
6784 
6785 	/* Enter guest */
6786 	enter_guest();
6787 
6788 	/*
6789 	 * Validate the behavior and
6790 	 * pass a magic value back to the guest.
6791 	 */
6792 	if (exit_reason_want == VMX_EXTINT) {
6793 		assert_exit_reason(exit_reason_want);
6794 
6795 		/* Clear the external interrupt. */
6796 		irq_enable();
6797 		asm volatile ("nop");
6798 		irq_disable();
6799 		report(handle_x2apic_ipi_ran,
6800 		       "Got pending interrupt after IRQ enabled.");
6801 
6802 		enter_guest();
6803 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6804 		assert_exit_reason(exit_reason_want);
6805 		report(virtual_apic_page[apic_reg_index(reg)] == val,
6806 		       "got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%lx",
6807 		       apic_reg_index(reg),
6808 		       virtual_apic_page[apic_reg_index(reg)], val);
6809 
6810 		/* Reenter guest so it can consume/check rcx and exit again. */
6811 		enter_guest();
6812 	} else if (exit_reason_want != VMX_VMCALL) {
6813 		report_fail("Oops, bad exit expectation: %u.", exit_reason_want);
6814 	}
6815 
6816 	assert_exit_reason(VMX_VMCALL);
6817 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) {
6818 		u64 want = val;
6819 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6820 
6821 		report(got == want, "x2APIC write; got 0x%x, want 0x%lx", got,
6822 		       want);
6823 	} else if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH) {
6824 		if (!expectation->wr_only) {
6825 			u32 got = apic_read(reg);
6826 			bool ok;
6827 
6828 			/*
6829 			 * When L1's TPR is passed through to L2, the lower
6830 			 * nibble can be lost. For example, if L2 executes
6831 			 * WRMSR(0x808, 0x78), then, L1 might read 0x70.
6832 			 *
6833 			 * Here's how the lower nibble can get lost:
6834 			 *   1. L2 executes WRMSR(0x808, 0x78).
6835 			 *   2. L2 exits to L0 with a WRMSR exit.
6836 			 *   3. L0 emulates WRMSR, by writing L1's TPR.
6837 			 *   4. L0 re-enters L2.
6838 			 *   5. L2 exits to L0 (reason doesn't matter).
6839 			 *   6. L0 reflects L2's exit to L1.
6840 			 *   7. Before entering L1, L0 exits to user-space
6841 			 *      (e.g., to satisfy TPR access reporting).
6842 			 *   8. User-space executes KVM_SET_REGS ioctl, which
6843 			 *      clears the lower nibble of L1's TPR.
6844 			 */
6845 			if (reg == APIC_TASKPRI) {
6846 				got = apic_virt_nibble1(got);
6847 				val = apic_virt_nibble1(val);
6848 			}
6849 
6850 			ok = got == val;
6851 			report(ok,
6852 			       "non-virtualized write; val is 0x%x, want 0x%lx",
6853 			       got, val);
6854 			apic_write(reg, restore_val);
6855 		} else {
6856 			report_pass("non-virtualized and write-only OK");
6857 		}
6858 	}
6859 	skip_exit_insn();
6860 
6861 	report_prefix_pop();
6862 }
6863 
6864 static enum Config_type configure_virt_x2apic_mode_test(
6865 	struct virt_x2apic_mode_config *virt_x2apic_mode_config,
6866 	u8 *msr_bitmap_page)
6867 {
6868 	int msr;
6869 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6870 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6871 
6872 	/* x2apic-specific VMCS config */
6873 	if (virt_x2apic_mode_config->use_msr_bitmaps) {
6874 		/* virt_x2apic_mode_test() checks for MSR bitmaps support */
6875 		cpu_exec_ctrl0 |= CPU_MSR_BITMAP;
6876 	} else {
6877 		cpu_exec_ctrl0 &= ~CPU_MSR_BITMAP;
6878 	}
6879 
6880 	if (virt_x2apic_mode_config->virtual_interrupt_delivery) {
6881 		if (!(ctrl_cpu_rev[1].clr & CPU_VINTD)) {
6882 			report_skip("VM-execution control \"virtual-interrupt delivery\" NOT supported.\n");
6883 			return CONFIG_TYPE_UNSUPPORTED;
6884 		}
6885 		cpu_exec_ctrl1 |= CPU_VINTD;
6886 	} else {
6887 		cpu_exec_ctrl1 &= ~CPU_VINTD;
6888 	}
6889 
6890 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6891 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6892 
6893 	/* x2APIC MSR intercepts are usually off for "Virtualize x2APIC mode" */
6894 	for (msr = 0x800; msr <= 0x8ff; msr++) {
6895 		if (virt_x2apic_mode_config->disable_x2apic_msr_intercepts) {
6896 			clear_bit(msr, msr_bitmap_page + 0x000);
6897 			clear_bit(msr, msr_bitmap_page + 0x800);
6898 		} else {
6899 			set_bit(msr, msr_bitmap_page + 0x000);
6900 			set_bit(msr, msr_bitmap_page + 0x800);
6901 		}
6902 	}
6903 
6904 	/* x2APIC mode can impact virtualization */
6905 	reset_apic();
6906 	if (!virt_x2apic_mode_config->disable_x2apic)
6907 		enable_x2apic();
6908 
6909 	return configure_apic_reg_virt_test(
6910 		&virt_x2apic_mode_config->apic_reg_virt_config);
6911 }
6912 
6913 static void virt_x2apic_mode_test(void)
6914 {
6915 	u32 *virtual_apic_page;
6916 	u8 *msr_bitmap_page;
6917 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6918 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6919 	int i;
6920 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6921 
6922 	if (!cpu_has_apicv()) {
6923 		report_skip(__func__);
6924 		return;
6925 	}
6926 
6927 	/*
6928 	 * This is to exercise an issue in KVM's logic to merge L0's and L1's
6929 	 * MSR bitmaps. Previously, an L1 could get at L0's x2APIC MSRs by
6930 	 * writing the IA32_SPEC_CTRL MSR or the IA32_PRED_CMD MSRs. KVM would
6931 	 * then proceed to manipulate the MSR bitmaps, as if VMCS12 had the
6932 	 * "Virtualize x2APIC mod" control set, even when it didn't.
6933 	 */
6934 	if (has_spec_ctrl())
6935 		wrmsr(MSR_IA32_SPEC_CTRL, 1);
6936 
6937 	/*
6938 	 * Check that VMCS12 supports:
6939 	 *   - "Virtual-APIC address", indicated by "use TPR shadow"
6940 	 *   - "MSR-bitmap address", indicated by "use MSR bitmaps"
6941 	 */
6942 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6943 		report_skip("VM-execution control \"use TPR shadow\" NOT supported.\n");
6944 		return;
6945 	} else if (!(ctrl_cpu_rev[0].clr & CPU_MSR_BITMAP)) {
6946 		report_skip("VM-execution control \"use MSR bitmaps\" NOT supported.\n");
6947 		return;
6948 	}
6949 
6950 	test_set_guest(virt_x2apic_mode_guest);
6951 
6952 	virtual_apic_page = alloc_page();
6953 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6954 
6955 	msr_bitmap_page = alloc_page();
6956 	memset(msr_bitmap_page, 0xff, PAGE_SIZE);
6957 	vmcs_write(MSR_BITMAP, virt_to_phys(msr_bitmap_page));
6958 
6959 	for (i = 0; i < ARRAY_SIZE(virt_x2apic_mode_tests); i++) {
6960 		struct virt_x2apic_mode_test_case *virt_x2apic_mode_test_case =
6961 			&virt_x2apic_mode_tests[i];
6962 		struct virt_x2apic_mode_config *virt_x2apic_mode_config =
6963 			&virt_x2apic_mode_test_case->virt_x2apic_mode_config;
6964 		enum Config_type config_type;
6965 		u32 reg;
6966 
6967 		printf("--- %s test ---\n", virt_x2apic_mode_test_case->name);
6968 		config_type =
6969 			configure_virt_x2apic_mode_test(virt_x2apic_mode_config,
6970 							msr_bitmap_page);
6971 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6972 			report_skip("Skip because of missing features.\n");
6973 			continue;
6974 		} else if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6975 			enter_guest_with_bad_controls();
6976 			continue;
6977 		}
6978 
6979 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6980 			struct virt_x2apic_mode_expectation expectation;
6981 
6982 			virt_x2apic_mode_exit_expectation(
6983 				reg, virt_x2apic_mode_config, &expectation);
6984 
6985 			test_x2apic_rd(reg, &expectation, virtual_apic_page);
6986 			test_x2apic_wr(reg, &expectation, virtual_apic_page);
6987 		}
6988 	}
6989 
6990 
6991 	/* Terminate the guest */
6992 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6993 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6994 	args->op = X2APIC_TERMINATE;
6995 	enter_guest();
6996 	assert_exit_reason(VMX_VMCALL);
6997 }
6998 
6999 static void test_ctl_reg(const char *cr_name, u64 cr, u64 fixed0, u64 fixed1)
7000 {
7001 	u64 val;
7002 	u64 cr_saved = vmcs_read(cr);
7003 	int i;
7004 
7005 	val = fixed0 & fixed1;
7006 	if (cr == HOST_CR4)
7007 		vmcs_write(cr, val | X86_CR4_PAE);
7008 	else
7009 		vmcs_write(cr, val);
7010 	report_prefix_pushf("%s %lx", cr_name, val);
7011 	if (val == fixed0)
7012 		test_vmx_vmlaunch(0);
7013 	else
7014 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7015 	report_prefix_pop();
7016 
7017 	for (i = 0; i < 64; i++) {
7018 
7019 		/* Set a bit when the corresponding bit in fixed1 is 0 */
7020 		if ((fixed1 & (1ull << i)) == 0) {
7021 			if (cr == HOST_CR4 && ((1ull << i) & X86_CR4_SMEP ||
7022 					       (1ull << i) & X86_CR4_SMAP))
7023 				continue;
7024 
7025 			vmcs_write(cr, cr_saved | (1ull << i));
7026 			report_prefix_pushf("%s %llx", cr_name,
7027 						cr_saved | (1ull << i));
7028 			test_vmx_vmlaunch(
7029 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7030 			report_prefix_pop();
7031 		}
7032 
7033 		/* Unset a bit when the corresponding bit in fixed0 is 1 */
7034 		if (fixed0 & (1ull << i)) {
7035 			vmcs_write(cr, cr_saved & ~(1ull << i));
7036 			report_prefix_pushf("%s %llx", cr_name,
7037 						cr_saved & ~(1ull << i));
7038 			test_vmx_vmlaunch(
7039 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7040 			report_prefix_pop();
7041 		}
7042 	}
7043 
7044 	vmcs_write(cr, cr_saved);
7045 }
7046 
7047 /*
7048  * 1. The CR0 field must not set any bit to a value not supported in VMX
7049  *    operation.
7050  * 2. The CR4 field must not set any bit to a value not supported in VMX
7051  *    operation.
7052  * 3. On processors that support Intel 64 architecture, the CR3 field must
7053  *    be such that bits 63:52 and bits in the range 51:32 beyond the
7054  *    processor's physical-address width must be 0.
7055  *
7056  *  [Intel SDM]
7057  */
7058 static void test_host_ctl_regs(void)
7059 {
7060 	u64 fixed0, fixed1, cr3, cr3_saved;
7061 	int i;
7062 
7063 	/* Test CR0 */
7064 	fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0);
7065 	fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1);
7066 	test_ctl_reg("HOST_CR0", HOST_CR0, fixed0, fixed1);
7067 
7068 	/* Test CR4 */
7069 	fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0);
7070 	fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1) &
7071 		 ~(X86_CR4_SMEP | X86_CR4_SMAP);
7072 	test_ctl_reg("HOST_CR4", HOST_CR4, fixed0, fixed1);
7073 
7074 	/* Test CR3 */
7075 	cr3_saved = vmcs_read(HOST_CR3);
7076 	for (i = cpuid_maxphyaddr(); i < 64; i++) {
7077 		cr3 = cr3_saved | (1ul << i);
7078 		vmcs_write(HOST_CR3, cr3);
7079 		report_prefix_pushf("HOST_CR3 %lx", cr3);
7080 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7081 		report_prefix_pop();
7082 	}
7083 
7084 	vmcs_write(HOST_CR3, cr3_saved);
7085 }
7086 
7087 static void test_efer_vmlaunch(u32 fld, bool ok)
7088 {
7089 	if (fld == HOST_EFER) {
7090 		if (ok)
7091 			test_vmx_vmlaunch(0);
7092 		else
7093 			test_vmx_vmlaunch2(VMXERR_ENTRY_INVALID_CONTROL_FIELD,
7094 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7095 	} else {
7096 		test_guest_state("EFER test", !ok, GUEST_EFER, "GUEST_EFER");
7097 	}
7098 }
7099 
7100 static void test_efer_one(u32 fld, const char * fld_name, u64 efer,
7101 			  u32 ctrl_fld, u64 ctrl,
7102 			  int i, const char *efer_bit_name)
7103 {
7104 	bool ok;
7105 
7106 	ok = true;
7107 	if (ctrl_fld == EXI_CONTROLS && (ctrl & EXI_LOAD_EFER)) {
7108 		if (!!(efer & EFER_LMA) != !!(ctrl & EXI_HOST_64))
7109 			ok = false;
7110 		if (!!(efer & EFER_LME) != !!(ctrl & EXI_HOST_64))
7111 			ok = false;
7112 	}
7113 	if (ctrl_fld == ENT_CONTROLS && (ctrl & ENT_LOAD_EFER)) {
7114 		/* Check LMA too since CR0.PG is set.  */
7115 		if (!!(efer & EFER_LMA) != !!(ctrl & ENT_GUEST_64))
7116 			ok = false;
7117 		if (!!(efer & EFER_LME) != !!(ctrl & ENT_GUEST_64))
7118 			ok = false;
7119 	}
7120 
7121 	/*
7122 	 * Skip the test if it would enter the guest in 32-bit mode.
7123 	 * Perhaps write the test in assembly and make sure it
7124 	 * can be run in either mode?
7125 	 */
7126 	if (fld == GUEST_EFER && ok && !(ctrl & ENT_GUEST_64))
7127 		return;
7128 
7129 	vmcs_write(ctrl_fld, ctrl);
7130 	vmcs_write(fld, efer);
7131 	report_prefix_pushf("%s %s bit turned %s, controls %s",
7132 			    fld_name, efer_bit_name,
7133 			    (i & 1) ? "on" : "off",
7134 			    (i & 2) ? "on" : "off");
7135 
7136 	test_efer_vmlaunch(fld, ok);
7137 	report_prefix_pop();
7138 }
7139 
7140 static void test_efer_bit(u32 fld, const char * fld_name,
7141 			  u32 ctrl_fld, u64 ctrl_bit, u64 efer_bit,
7142 			  const char *efer_bit_name)
7143 {
7144 	u64 efer_saved = vmcs_read(fld);
7145 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7146 	int i;
7147 
7148 	for (i = 0; i < 4; i++) {
7149 		u64 efer = efer_saved & ~efer_bit;
7150 		u64 ctrl = ctrl_saved & ~ctrl_bit;
7151 
7152 		if (i & 1)
7153 			efer |= efer_bit;
7154 		if (i & 2)
7155 			ctrl |= ctrl_bit;
7156 
7157 		test_efer_one(fld, fld_name, efer, ctrl_fld, ctrl,
7158 			      i, efer_bit_name);
7159 	}
7160 
7161 	vmcs_write(ctrl_fld, ctrl_saved);
7162 	vmcs_write(fld, efer_saved);
7163 }
7164 
7165 static void test_efer(u32 fld, const char * fld_name, u32 ctrl_fld,
7166 		      u64 ctrl_bit1, u64 ctrl_bit2)
7167 {
7168 	u64 efer_saved = vmcs_read(fld);
7169 	u32 ctrl_saved = vmcs_read(ctrl_fld);
7170 	u64 efer_reserved_bits =  ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
7171 	u64 i;
7172 	u64 efer;
7173 
7174 	if (cpu_has_efer_nx())
7175 		efer_reserved_bits &= ~EFER_NX;
7176 
7177 	if (!ctrl_bit1) {
7178 		printf("\"Load-IA32-EFER\" exit control not supported\n");
7179 		goto test_entry_exit_mode;
7180 	}
7181 
7182 	report_prefix_pushf("%s %lx", fld_name, efer_saved);
7183 	test_efer_vmlaunch(fld, true);
7184 	report_prefix_pop();
7185 
7186 	/*
7187 	 * Check reserved bits
7188 	 */
7189 	vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit1);
7190 	for (i = 0; i < 64; i++) {
7191 		if ((1ull << i) & efer_reserved_bits) {
7192 			efer = efer_saved | (1ull << i);
7193 			vmcs_write(fld, efer);
7194 			report_prefix_pushf("%s %lx", fld_name, efer);
7195 			test_efer_vmlaunch(fld, true);
7196 			report_prefix_pop();
7197 		}
7198 	}
7199 
7200 	vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit1);
7201 	for (i = 0; i < 64; i++) {
7202 		if ((1ull << i) & efer_reserved_bits) {
7203 			efer = efer_saved | (1ull << i);
7204 			vmcs_write(fld, efer);
7205 			report_prefix_pushf("%s %lx", fld_name, efer);
7206 			test_efer_vmlaunch(fld, false);
7207 			report_prefix_pop();
7208 		}
7209 	}
7210 
7211 	vmcs_write(ctrl_fld, ctrl_saved);
7212 	vmcs_write(fld, efer_saved);
7213 
7214 	/*
7215 	 * Check LMA and LME bits
7216 	 */
7217 	test_efer_bit(fld, fld_name,
7218 		      ctrl_fld, ctrl_bit1,
7219 		      EFER_LMA,
7220 		      "EFER_LMA");
7221 	test_efer_bit(fld, fld_name,
7222 		      ctrl_fld, ctrl_bit1,
7223 		      EFER_LME,
7224 		      "EFER_LME");
7225 
7226 test_entry_exit_mode:
7227 	test_efer_bit(fld, fld_name,
7228 		      ctrl_fld, ctrl_bit2,
7229 		      EFER_LMA,
7230 		      "EFER_LMA");
7231 	test_efer_bit(fld, fld_name,
7232 		      ctrl_fld, ctrl_bit2,
7233 		      EFER_LME,
7234 		      "EFER_LME");
7235 }
7236 
7237 /*
7238  * If the 'load IA32_EFER' VM-exit control is 1, bits reserved in the
7239  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7240  * the values of the LMA and LME bits in the field must each be that of
7241  * the 'host address-space size' VM-exit control.
7242  *
7243  *  [Intel SDM]
7244  */
7245 static void test_host_efer(void)
7246 {
7247 	test_efer(HOST_EFER, "HOST_EFER", EXI_CONTROLS,
7248 		  ctrl_exit_rev.clr & EXI_LOAD_EFER,
7249 		  EXI_HOST_64);
7250 }
7251 
7252 /*
7253  * If the 'load IA32_EFER' VM-enter control is 1, bits reserved in the
7254  * IA32_EFER MSR must be 0 in the field for that register. In addition,
7255  * the values of the LMA and LME bits in the field must each be that of
7256  * the 'IA32e-mode guest' VM-exit control.
7257  */
7258 static void test_guest_efer(void)
7259 {
7260 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER)) {
7261 		printf("\"Load-IA32-EFER\" entry control not supported\n");
7262 		return;
7263 	}
7264 
7265 	vmcs_write(GUEST_EFER, rdmsr(MSR_EFER));
7266 	test_efer(GUEST_EFER, "GUEST_EFER", ENT_CONTROLS,
7267 		  ctrl_enter_rev.clr & ENT_LOAD_EFER,
7268 		  ENT_GUEST_64);
7269 }
7270 
7271 /*
7272  * PAT values higher than 8 are uninteresting since they're likely lumped
7273  * in with "8". We only test values above 8 one bit at a time,
7274  * in order to reduce the number of VM-Entries and keep the runtime reasonable.
7275  */
7276 #define	PAT_VAL_LIMIT	8
7277 
7278 static void test_pat(u32 field, const char * field_name, u32 ctrl_field,
7279 		     u64 ctrl_bit)
7280 {
7281 	u32 ctrl_saved = vmcs_read(ctrl_field);
7282 	u64 pat_saved = vmcs_read(field);
7283 	u64 i, val;
7284 	u32 j;
7285 	int error;
7286 
7287 	vmcs_clear_bits(ctrl_field, ctrl_bit);
7288 
7289 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7290 		/* Test PAT0..PAT7 fields */
7291 		for (j = 0; j < (i ? 8 : 1); j++) {
7292 			val = i << j * 8;
7293 			vmcs_write(field, val);
7294 			if (field == HOST_PAT) {
7295 				report_prefix_pushf("%s %lx", field_name, val);
7296 				test_vmx_vmlaunch(0);
7297 				report_prefix_pop();
7298 
7299 			} else {	// GUEST_PAT
7300 				test_guest_state("ENT_LOAD_PAT enabled", false,
7301 						 val, "GUEST_PAT");
7302 			}
7303 		}
7304 	}
7305 
7306 	vmcs_set_bits(ctrl_field, ctrl_bit);
7307 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
7308 		/* Test PAT0..PAT7 fields */
7309 		for (j = 0; j < (i ? 8 : 1); j++) {
7310 			val = i << j * 8;
7311 			vmcs_write(field, val);
7312 
7313 			if (field == HOST_PAT) {
7314 				report_prefix_pushf("%s %lx", field_name, val);
7315 				if (i == 0x2 || i == 0x3 || i >= 0x8)
7316 					error =
7317 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
7318 				else
7319 					error = 0;
7320 
7321 				test_vmx_vmlaunch(error);
7322 				report_prefix_pop();
7323 
7324 			} else {	// GUEST_PAT
7325 				error = (i == 0x2 || i == 0x3 || i >= 0x8);
7326 				test_guest_state("ENT_LOAD_PAT enabled", !!error,
7327 						 val, "GUEST_PAT");
7328 			}
7329 
7330 		}
7331 	}
7332 
7333 	vmcs_write(ctrl_field, ctrl_saved);
7334 	vmcs_write(field, pat_saved);
7335 }
7336 
7337 /*
7338  *  If the "load IA32_PAT" VM-exit control is 1, the value of the field
7339  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7340  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7341  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7342  *  6 (WB), or 7 (UC-).
7343  *
7344  *  [Intel SDM]
7345  */
7346 static void test_load_host_pat(void)
7347 {
7348 	/*
7349 	 * "load IA32_PAT" VM-exit control
7350 	 */
7351 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) {
7352 		printf("\"Load-IA32-PAT\" exit control not supported\n");
7353 		return;
7354 	}
7355 
7356 	test_pat(HOST_PAT, "HOST_PAT", EXI_CONTROLS, EXI_LOAD_PAT);
7357 }
7358 
7359 union cpuidA_eax {
7360 	struct {
7361 		unsigned int version_id:8;
7362 		unsigned int num_counters_gp:8;
7363 		unsigned int bit_width:8;
7364 		unsigned int mask_length:8;
7365 	} split;
7366 	unsigned int full;
7367 };
7368 
7369 union cpuidA_edx {
7370 	struct {
7371 		unsigned int num_counters_fixed:5;
7372 		unsigned int bit_width_fixed:8;
7373 		unsigned int reserved:9;
7374 	} split;
7375 	unsigned int full;
7376 };
7377 
7378 static bool valid_pgc(u64 val)
7379 {
7380 	struct cpuid id;
7381 	union cpuidA_eax eax;
7382 	union cpuidA_edx edx;
7383 	u64 mask;
7384 
7385 	id = cpuid(0xA);
7386 	eax.full = id.a;
7387 	edx.full = id.d;
7388 	mask = ~(((1ull << eax.split.num_counters_gp) - 1) |
7389 		 (((1ull << edx.split.num_counters_fixed) - 1) << 32));
7390 
7391 	return !(val & mask);
7392 }
7393 
7394 static void test_pgc_vmlaunch(u32 xerror, u32 xreason, bool xfail, bool host)
7395 {
7396 	u32 inst_err;
7397 	u64 obs;
7398 	bool success;
7399 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7400 
7401 	if (host) {
7402 		success = vmlaunch_succeeds();
7403 		obs = rdmsr(data->msr);
7404 		if (!success) {
7405 			inst_err = vmcs_read(VMX_INST_ERROR);
7406 			report(xerror == inst_err, "vmlaunch failed, "
7407 			       "VMX Inst Error is %d (expected %d)",
7408 			       inst_err, xerror);
7409 		} else {
7410 			report(!data->enabled || data->exp == obs,
7411 			       "Host state is 0x%lx (expected 0x%lx)",
7412 			       obs, data->exp);
7413 			report(success != xfail, "vmlaunch succeeded");
7414 		}
7415 	} else {
7416 		test_guest_state("load GUEST_PERF_GLOBAL_CTRL", xfail,
7417 				 GUEST_PERF_GLOBAL_CTRL,
7418 				 "GUEST_PERF_GLOBAL_CTRL");
7419 	}
7420 }
7421 
7422 /*
7423  * test_load_perf_global_ctrl is a generic function for testing the
7424  * "load IA32_PERF_GLOBAL_CTRL" VM-{Entry,Exit} controls. This test function
7425  * tests the provided ctrl_val when disabled and enabled.
7426  *
7427  * @nr: VMCS field number corresponding to the host/guest state field
7428  * @name: Name of the above VMCS field for printing in test report
7429  * @ctrl_nr: VMCS field number corresponding to the VM-{Entry,Exit} control
7430  * @ctrl_val: Bit to set on the ctrl_field
7431  */
7432 static void test_perf_global_ctrl(u32 nr, const char *name, u32 ctrl_nr,
7433 				  const char *ctrl_name, u64 ctrl_val)
7434 {
7435 	u64 ctrl_saved = vmcs_read(ctrl_nr);
7436 	u64 pgc_saved = vmcs_read(nr);
7437 	u64 i, val;
7438 	bool host = nr == HOST_PERF_GLOBAL_CTRL;
7439 	struct vmx_state_area_test_data *data = &vmx_state_area_test_data;
7440 
7441 	data->msr = MSR_CORE_PERF_GLOBAL_CTRL;
7442 	msr_bmp_init();
7443 	vmcs_write(ctrl_nr, ctrl_saved & ~ctrl_val);
7444 	data->enabled = false;
7445 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=0 on %s",
7446 			    ctrl_name);
7447 
7448 	for (i = 0; i < 64; i++) {
7449 		val = 1ull << i;
7450 		vmcs_write(nr, val);
7451 		report_prefix_pushf("%s = 0x%lx", name, val);
7452 		test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7453 		report_prefix_pop();
7454 	}
7455 	report_prefix_pop();
7456 
7457 	vmcs_write(ctrl_nr, ctrl_saved | ctrl_val);
7458 	data->enabled = true;
7459 	report_prefix_pushf("\"load IA32_PERF_GLOBAL_CTRL\"=1 on %s",
7460 			    ctrl_name);
7461 	for (i = 0; i < 64; i++) {
7462 		val = 1ull << i;
7463 		data->exp = val;
7464 		vmcs_write(nr, val);
7465 		report_prefix_pushf("%s = 0x%lx", name, val);
7466 		if (valid_pgc(val)) {
7467 			test_pgc_vmlaunch(0, VMX_VMCALL, false, host);
7468 		} else {
7469 			if (host)
7470 				test_pgc_vmlaunch(
7471 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
7472 					0,
7473 					true,
7474 					host);
7475 			else
7476 				test_pgc_vmlaunch(
7477 					0,
7478 					VMX_ENTRY_FAILURE | VMX_FAIL_STATE,
7479 					true,
7480 					host);
7481 		}
7482 		report_prefix_pop();
7483 	}
7484 
7485 	data->enabled = false;
7486 	report_prefix_pop();
7487 	vmcs_write(ctrl_nr, ctrl_saved);
7488 	vmcs_write(nr, pgc_saved);
7489 }
7490 
7491 static void test_load_host_perf_global_ctrl(void)
7492 {
7493 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PERF)) {
7494 		printf("\"load IA32_PERF_GLOBAL_CTRL\" exit control not supported\n");
7495 		return;
7496 	}
7497 
7498 	test_perf_global_ctrl(HOST_PERF_GLOBAL_CTRL, "HOST_PERF_GLOBAL_CTRL",
7499 				   EXI_CONTROLS, "EXI_CONTROLS", EXI_LOAD_PERF);
7500 }
7501 
7502 
7503 static void test_load_guest_perf_global_ctrl(void)
7504 {
7505 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PERF)) {
7506 		printf("\"load IA32_PERF_GLOBAL_CTRL\" entry control not supported\n");
7507 		return;
7508 	}
7509 
7510 	test_perf_global_ctrl(GUEST_PERF_GLOBAL_CTRL, "GUEST_PERF_GLOBAL_CTRL",
7511 				   ENT_CONTROLS, "ENT_CONTROLS", ENT_LOAD_PERF);
7512 }
7513 
7514 
7515 /*
7516  * test_vmcs_field - test a value for the given VMCS field
7517  * @field: VMCS field
7518  * @field_name: string name of VMCS field
7519  * @bit_start: starting bit
7520  * @bit_end: ending bit
7521  * @val: value that the bit range must or must not contain
7522  * @valid_val: whether value given in 'val' must be valid or not
7523  * @error: expected VMCS error when vmentry fails for an invalid value
7524  */
7525 static void test_vmcs_field(u64 field, const char *field_name, u32 bit_start,
7526 			    u32 bit_end, u64 val, bool valid_val, u32 error)
7527 {
7528 	u64 field_saved = vmcs_read(field);
7529 	u32 i;
7530 	u64 tmp;
7531 	u32 bit_on;
7532 	u64 mask = ~0ull;
7533 
7534 	mask = (mask >> bit_end) << bit_end;
7535 	mask = mask | ((1 << bit_start) - 1);
7536 	tmp = (field_saved & mask) | (val << bit_start);
7537 
7538 	vmcs_write(field, tmp);
7539 	report_prefix_pushf("%s %lx", field_name, tmp);
7540 	if (valid_val)
7541 		test_vmx_vmlaunch(0);
7542 	else
7543 		test_vmx_vmlaunch(error);
7544 	report_prefix_pop();
7545 
7546 	for (i = bit_start; i <= bit_end; i = i + 2) {
7547 		bit_on = ((1ull < i) & (val << bit_start)) ? 0 : 1;
7548 		if (bit_on)
7549 			tmp = field_saved | (1ull << i);
7550 		else
7551 			tmp = field_saved & ~(1ull << i);
7552 		vmcs_write(field, tmp);
7553 		report_prefix_pushf("%s %lx", field_name, tmp);
7554 		if (valid_val)
7555 			test_vmx_vmlaunch(error);
7556 		else
7557 			test_vmx_vmlaunch(0);
7558 		report_prefix_pop();
7559 	}
7560 
7561 	vmcs_write(field, field_saved);
7562 }
7563 
7564 static void test_canonical(u64 field, const char * field_name, bool host)
7565 {
7566 	u64 addr_saved = vmcs_read(field);
7567 
7568 	/*
7569 	 * Use the existing value if possible.  Writing a random canonical
7570 	 * value is not an option as doing so would corrupt the field being
7571 	 * tested and likely hose the test.
7572 	 */
7573 	if (is_canonical(addr_saved)) {
7574 		if (host) {
7575 			report_prefix_pushf("%s %lx", field_name, addr_saved);
7576 			test_vmx_vmlaunch(0);
7577 			report_prefix_pop();
7578 		} else {
7579 			test_guest_state("Test canonical address", false,
7580 					 addr_saved, field_name);
7581 		}
7582 	}
7583 
7584 	vmcs_write(field, NONCANONICAL);
7585 
7586 	if (host) {
7587 		report_prefix_pushf("%s %llx", field_name, NONCANONICAL);
7588 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7589 		report_prefix_pop();
7590 	} else {
7591 		test_guest_state("Test non-canonical address", true,
7592 				 NONCANONICAL, field_name);
7593 	}
7594 
7595 	vmcs_write(field, addr_saved);
7596 }
7597 
7598 #define TEST_RPL_TI_FLAGS(reg, name)				\
7599 	test_vmcs_field(reg, name, 0, 2, 0x0, true,		\
7600 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7601 
7602 #define TEST_CS_TR_FLAGS(reg, name)				\
7603 	test_vmcs_field(reg, name, 3, 15, 0x0000, false,	\
7604 			VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7605 
7606 /*
7607  * 1. In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the
7608  *    RPL (bits 1:0) and the TI flag (bit 2) must be 0.
7609  * 2. The selector fields for CS and TR cannot be 0000H.
7610  * 3. The selector field for SS cannot be 0000H if the "host address-space
7611  *    size" VM-exit control is 0.
7612  * 4. On processors that support Intel 64 architecture, the base-address
7613  *    fields for FS, GS and TR must contain canonical addresses.
7614  */
7615 static void test_host_segment_regs(void)
7616 {
7617 	u16 selector_saved;
7618 
7619 	/*
7620 	 * Test RPL and TI flags
7621 	 */
7622 	TEST_RPL_TI_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7623 	TEST_RPL_TI_FLAGS(HOST_SEL_SS, "HOST_SEL_SS");
7624 	TEST_RPL_TI_FLAGS(HOST_SEL_DS, "HOST_SEL_DS");
7625 	TEST_RPL_TI_FLAGS(HOST_SEL_ES, "HOST_SEL_ES");
7626 	TEST_RPL_TI_FLAGS(HOST_SEL_FS, "HOST_SEL_FS");
7627 	TEST_RPL_TI_FLAGS(HOST_SEL_GS, "HOST_SEL_GS");
7628 	TEST_RPL_TI_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7629 
7630 	/*
7631 	 * Test that CS and TR fields can not be 0x0000
7632 	 */
7633 	TEST_CS_TR_FLAGS(HOST_SEL_CS, "HOST_SEL_CS");
7634 	TEST_CS_TR_FLAGS(HOST_SEL_TR, "HOST_SEL_TR");
7635 
7636 	/*
7637 	 * SS field can not be 0x0000 if "host address-space size" VM-exit
7638 	 * control is 0
7639 	 */
7640 	selector_saved = vmcs_read(HOST_SEL_SS);
7641 	vmcs_write(HOST_SEL_SS, 0);
7642 	report_prefix_pushf("HOST_SEL_SS 0");
7643 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7644 		test_vmx_vmlaunch(0);
7645 	} else {
7646 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7647 	}
7648 	report_prefix_pop();
7649 
7650 	vmcs_write(HOST_SEL_SS, selector_saved);
7651 
7652 	/*
7653 	 * Base address for FS, GS and TR must be canonical
7654 	 */
7655 	test_canonical(HOST_BASE_FS, "HOST_BASE_FS", true);
7656 	test_canonical(HOST_BASE_GS, "HOST_BASE_GS", true);
7657 	test_canonical(HOST_BASE_TR, "HOST_BASE_TR", true);
7658 }
7659 
7660 /*
7661  *  On processors that support Intel 64 architecture, the base-address
7662  *  fields for GDTR and IDTR must contain canonical addresses.
7663  */
7664 static void test_host_desc_tables(void)
7665 {
7666 	test_canonical(HOST_BASE_GDTR, "HOST_BASE_GDTR", true);
7667 	test_canonical(HOST_BASE_IDTR, "HOST_BASE_IDTR", true);
7668 }
7669 
7670 /*
7671  * If the "host address-space size" VM-exit control is 0, the following must
7672  * hold:
7673  *    - The "IA-32e mode guest" VM-entry control is 0.
7674  *    - Bit 17 of the CR4 field (corresponding to CR4.PCIDE) is 0.
7675  *    - Bits 63:32 in the RIP field are 0.
7676  *
7677  * If the "host address-space size" VM-exit control is 1, the following must
7678  * hold:
7679  *    - Bit 5 of the CR4 field (corresponding to CR4.PAE) is 1.
7680  *    - The RIP field contains a canonical address.
7681  *
7682  */
7683 static void test_host_addr_size(void)
7684 {
7685 	u64 cr4_saved = vmcs_read(HOST_CR4);
7686 	u64 rip_saved = vmcs_read(HOST_RIP);
7687 	u64 entry_ctrl_saved = vmcs_read(ENT_CONTROLS);
7688 	int i;
7689 	u64 tmp;
7690 
7691 	if (vmcs_read(EXI_CONTROLS) & EXI_HOST_64) {
7692 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7693 		report_prefix_pushf("\"IA-32e mode guest\" enabled");
7694 		test_vmx_vmlaunch(0);
7695 		report_prefix_pop();
7696 
7697 		vmcs_write(HOST_CR4, cr4_saved | X86_CR4_PCIDE);
7698 		report_prefix_pushf("\"CR4.PCIDE\" set");
7699 		test_vmx_vmlaunch(0);
7700 		report_prefix_pop();
7701 
7702 		for (i = 32; i <= 63; i = i + 4) {
7703 			tmp = rip_saved | 1ull << i;
7704 			vmcs_write(HOST_RIP, tmp);
7705 			report_prefix_pushf("HOST_RIP %lx", tmp);
7706 			test_vmx_vmlaunch(0);
7707 			report_prefix_pop();
7708 		}
7709 
7710 		if (cr4_saved & X86_CR4_PAE) {
7711 			vmcs_write(HOST_CR4, cr4_saved  & ~X86_CR4_PAE);
7712 			report_prefix_pushf("\"CR4.PAE\" unset");
7713 			test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7714 		} else {
7715 			report_prefix_pushf("\"CR4.PAE\" set");
7716 			test_vmx_vmlaunch(0);
7717 		}
7718 		report_prefix_pop();
7719 
7720 		vmcs_write(HOST_RIP, NONCANONICAL);
7721 		report_prefix_pushf("HOST_RIP %llx", NONCANONICAL);
7722 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7723 		report_prefix_pop();
7724 
7725 		vmcs_write(ENT_CONTROLS, entry_ctrl_saved | ENT_GUEST_64);
7726 		vmcs_write(HOST_RIP, rip_saved);
7727 		vmcs_write(HOST_CR4, cr4_saved);
7728 
7729 		/* Restore host's active RIP and CR4 values. */
7730 		report_prefix_pushf("restore host state");
7731 		test_vmx_vmlaunch(0);
7732 		report_prefix_pop();
7733 	}
7734 }
7735 
7736 /*
7737  * Check that the virtual CPU checks the VMX Host State Area as
7738  * documented in the Intel SDM.
7739  */
7740 static void vmx_host_state_area_test(void)
7741 {
7742 	/*
7743 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
7744 	 * fail due to invalid guest state, should we make it that
7745 	 * far.
7746 	 */
7747 	vmcs_write(GUEST_RFLAGS, 0);
7748 
7749 	test_host_ctl_regs();
7750 
7751 	test_canonical(HOST_SYSENTER_ESP, "HOST_SYSENTER_ESP", true);
7752 	test_canonical(HOST_SYSENTER_EIP, "HOST_SYSENTER_EIP", true);
7753 
7754 	test_host_efer();
7755 	test_load_host_pat();
7756 	test_host_segment_regs();
7757 	test_host_desc_tables();
7758 	test_host_addr_size();
7759 	test_load_host_perf_global_ctrl();
7760 }
7761 
7762 /*
7763  * If the "load debug controls" VM-entry control is 1, bits 63:32 in
7764  * the DR7 field must be 0.
7765  *
7766  * [Intel SDM]
7767  */
7768 static void test_guest_dr7(void)
7769 {
7770 	u32 ent_saved = vmcs_read(ENT_CONTROLS);
7771 	u64 dr7_saved = vmcs_read(GUEST_DR7);
7772 	u64 val;
7773 	int i;
7774 
7775 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS) {
7776 		vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7777 		for (i = 0; i < 64; i++) {
7778 			val = 1ull << i;
7779 			vmcs_write(GUEST_DR7, val);
7780 			test_guest_state("ENT_LOAD_DBGCTLS disabled", false,
7781 					 val, "GUEST_DR7");
7782 		}
7783 	}
7784 	if (ctrl_enter_rev.clr & ENT_LOAD_DBGCTLS) {
7785 		vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_DBGCTLS);
7786 		for (i = 0; i < 64; i++) {
7787 			val = 1ull << i;
7788 			vmcs_write(GUEST_DR7, val);
7789 			test_guest_state("ENT_LOAD_DBGCTLS enabled", i >= 32,
7790 					 val, "GUEST_DR7");
7791 		}
7792 	}
7793 	vmcs_write(GUEST_DR7, dr7_saved);
7794 	vmcs_write(ENT_CONTROLS, ent_saved);
7795 }
7796 
7797 /*
7798  *  If the "load IA32_PAT" VM-entry control is 1, the value of the field
7799  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7800  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7801  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7802  *  6 (WB), or 7 (UC-).
7803  *
7804  *  [Intel SDM]
7805  */
7806 static void test_load_guest_pat(void)
7807 {
7808 	/*
7809 	 * "load IA32_PAT" VM-entry control
7810 	 */
7811 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
7812 		printf("\"Load-IA32-PAT\" entry control not supported\n");
7813 		return;
7814 	}
7815 
7816 	test_pat(GUEST_PAT, "GUEST_PAT", ENT_CONTROLS, ENT_LOAD_PAT);
7817 }
7818 
7819 #define MSR_IA32_BNDCFGS_RSVD_MASK	0x00000ffc
7820 
7821 /*
7822  * If the "load IA32_BNDCFGS" VM-entry control is 1, the following
7823  * checks are performed on the field for the IA32_BNDCFGS MSR:
7824  *
7825  *   - Bits reserved in the IA32_BNDCFGS MSR must be 0.
7826  *   - The linear address in bits 63:12 must be canonical.
7827  *
7828  *  [Intel SDM]
7829  */
7830 static void test_load_guest_bndcfgs(void)
7831 {
7832 	u64 bndcfgs_saved = vmcs_read(GUEST_BNDCFGS);
7833 	u64 bndcfgs;
7834 
7835 	if (!(ctrl_enter_rev.clr & ENT_LOAD_BNDCFGS)) {
7836 		printf("\"Load-IA32-BNDCFGS\" entry control not supported\n");
7837 		return;
7838 	}
7839 
7840 	vmcs_clear_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7841 
7842 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7843 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7844 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7845 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7846 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7847 	test_guest_state("ENT_LOAD_BNDCFGS disabled", false,
7848 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7849 
7850 	vmcs_set_bits(ENT_CONTROLS, ENT_LOAD_BNDCFGS);
7851 
7852 	vmcs_write(GUEST_BNDCFGS, NONCANONICAL);
7853 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7854 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7855 	bndcfgs = bndcfgs_saved | MSR_IA32_BNDCFGS_RSVD_MASK;
7856 	vmcs_write(GUEST_BNDCFGS, bndcfgs);
7857 	test_guest_state("ENT_LOAD_BNDCFGS enabled", true,
7858 			 GUEST_BNDCFGS, "GUEST_BNDCFGS");
7859 
7860 	vmcs_write(GUEST_BNDCFGS, bndcfgs_saved);
7861 }
7862 
7863 #define	GUEST_SEG_UNUSABLE_MASK	(1u << 16)
7864 #define	GUEST_SEG_SEL_TI_MASK	(1u << 2)
7865 
7866 
7867 #define	TEST_SEGMENT_SEL(test, xfail, sel, val)				\
7868 do {									\
7869 	vmcs_write(sel, val);						\
7870 	test_guest_state(test " segment", xfail, val, xstr(sel));	\
7871 } while (0)
7872 
7873 #define	TEST_INVALID_SEG_SEL(sel, val) \
7874 	TEST_SEGMENT_SEL("Invalid: " xstr(val), true, sel, val);
7875 
7876 #define	TEST_VALID_SEG_SEL(sel, val) \
7877 	TEST_SEGMENT_SEL("Valid: " xstr(val), false, sel, val);
7878 
7879 /*
7880  * The following checks are done on the Selector field of the Guest Segment
7881  * Registers:
7882  *    - TR. The TI flag (bit 2) must be 0.
7883  *    - LDTR. If LDTR is usable, the TI flag (bit 2) must be 0.
7884  *    - SS. If the guest will not be virtual-8086 and the "unrestricted
7885  *	guest" VM-execution control is 0, the RPL (bits 1:0) must equal
7886  *	the RPL of the selector field for CS.
7887  *
7888  *  [Intel SDM]
7889  */
7890 static void test_guest_segment_sel_fields(void)
7891 {
7892 	u16 sel_saved;
7893 	u32 ar_saved;
7894 	u32 cpu_ctrl0_saved;
7895 	u32 cpu_ctrl1_saved;
7896 	u16 cs_rpl_bits;
7897 
7898 	/*
7899 	 * Test for GUEST_SEL_TR
7900 	 */
7901 	sel_saved = vmcs_read(GUEST_SEL_TR);
7902 	TEST_INVALID_SEG_SEL(GUEST_SEL_TR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7903 	vmcs_write(GUEST_SEL_TR, sel_saved);
7904 
7905 	/*
7906 	 * Test for GUEST_SEL_LDTR
7907 	 */
7908 	sel_saved = vmcs_read(GUEST_SEL_LDTR);
7909 	ar_saved = vmcs_read(GUEST_AR_LDTR);
7910 	/* LDTR is set unusable */
7911 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7912 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7913 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7914 	/* LDTR is set usable */
7915 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7916 	TEST_INVALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved | GUEST_SEG_SEL_TI_MASK);
7917 
7918 	TEST_VALID_SEG_SEL(GUEST_SEL_LDTR, sel_saved & ~GUEST_SEG_SEL_TI_MASK);
7919 
7920 	vmcs_write(GUEST_AR_LDTR, ar_saved);
7921 	vmcs_write(GUEST_SEL_LDTR, sel_saved);
7922 
7923 	/*
7924 	 * Test for GUEST_SEL_SS
7925 	 */
7926 	cpu_ctrl0_saved = vmcs_read(CPU_EXEC_CTRL0);
7927 	cpu_ctrl1_saved = vmcs_read(CPU_EXEC_CTRL1);
7928 	ar_saved = vmcs_read(GUEST_AR_SS);
7929 	/* Turn off "unrestricted guest" vm-execution control */
7930 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved & ~CPU_URG);
7931 	cs_rpl_bits = vmcs_read(GUEST_SEL_CS) & 0x3;
7932 	sel_saved = vmcs_read(GUEST_SEL_SS);
7933 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7934 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7935 	/* Make SS usable if it's unusable or vice-versa */
7936 	if (ar_saved & GUEST_SEG_UNUSABLE_MASK)
7937 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7938 	else
7939 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7940 	TEST_INVALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7941 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7942 
7943 	/* Need a valid EPTP as the passing case fully enters the guest. */
7944 	if (enable_unrestricted_guest(true))
7945 		goto skip_ss_tests;
7946 
7947 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7948 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7949 
7950 	/* Make SS usable if it's unusable or vice-versa */
7951 	if (vmcs_read(GUEST_AR_SS) & GUEST_SEG_UNUSABLE_MASK)
7952 		vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
7953 	else
7954 		vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
7955 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (~cs_rpl_bits & 0x3)));
7956 	TEST_VALID_SEG_SEL(GUEST_SEL_SS, ((sel_saved & ~0x3) | (cs_rpl_bits & 0x3)));
7957 skip_ss_tests:
7958 
7959 	vmcs_write(GUEST_AR_SS, ar_saved);
7960 	vmcs_write(GUEST_SEL_SS, sel_saved);
7961 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrl0_saved);
7962 	vmcs_write(CPU_EXEC_CTRL1, cpu_ctrl1_saved);
7963 }
7964 
7965 #define	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(xfail, seg_base)			\
7966 do {										\
7967 	addr_saved = vmcs_read(seg_base);					\
7968 	for (i = 32; i < 63; i = i + 4) {					\
7969 		addr = addr_saved | 1ull << i;					\
7970 		vmcs_write(seg_base, addr);					\
7971 		test_guest_state("seg.BASE[63:32] != 0, usable = " xstr(xfail),	\
7972 				 xfail, addr, xstr(seg_base));			\
7973 	}									\
7974 	vmcs_write(seg_base, addr_saved);					\
7975 } while (0)
7976 
7977 #define	TEST_SEGMENT_BASE_ADDR_CANONICAL(xfail, seg_base)		  \
7978 do {									  \
7979 	addr_saved = vmcs_read(seg_base);				  \
7980 	vmcs_write(seg_base, NONCANONICAL);				  \
7981 	test_guest_state("seg.BASE non-canonical, usable = " xstr(xfail), \
7982 			 xfail, NONCANONICAL, xstr(seg_base));		  \
7983 	vmcs_write(seg_base, addr_saved);				  \
7984 } while (0)
7985 
7986 /*
7987  * The following checks are done on the Base Address field of the Guest
7988  * Segment Registers on processors that support Intel 64 architecture:
7989  *    - TR, FS, GS : The address must be canonical.
7990  *    - LDTR : If LDTR is usable, the address must be canonical.
7991  *    - CS : Bits 63:32 of the address must be zero.
7992  *    - SS, DS, ES : If the register is usable, bits 63:32 of the address
7993  *	must be zero.
7994  *
7995  *  [Intel SDM]
7996  */
7997 static void test_guest_segment_base_addr_fields(void)
7998 {
7999 	u64 addr_saved;
8000 	u64 addr;
8001 	u32 ar_saved;
8002 	int i;
8003 
8004 	/*
8005 	 * The address of TR, FS, GS and LDTR must be canonical.
8006 	 */
8007 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_TR);
8008 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_FS);
8009 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_GS);
8010 	ar_saved = vmcs_read(GUEST_AR_LDTR);
8011 	/* Make LDTR unusable */
8012 	vmcs_write(GUEST_AR_LDTR, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8013 	TEST_SEGMENT_BASE_ADDR_CANONICAL(false, GUEST_BASE_LDTR);
8014 	/* Make LDTR usable */
8015 	vmcs_write(GUEST_AR_LDTR, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8016 	TEST_SEGMENT_BASE_ADDR_CANONICAL(true, GUEST_BASE_LDTR);
8017 
8018 	vmcs_write(GUEST_AR_LDTR, ar_saved);
8019 
8020 	/*
8021 	 * Bits 63:32 in CS, SS, DS and ES base address must be zero
8022 	 */
8023 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_CS);
8024 	ar_saved = vmcs_read(GUEST_AR_SS);
8025 	/* Make SS unusable */
8026 	vmcs_write(GUEST_AR_SS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8027 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_SS);
8028 	/* Make SS usable */
8029 	vmcs_write(GUEST_AR_SS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8030 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_SS);
8031 	vmcs_write(GUEST_AR_SS, ar_saved);
8032 
8033 	ar_saved = vmcs_read(GUEST_AR_DS);
8034 	/* Make DS unusable */
8035 	vmcs_write(GUEST_AR_DS, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8036 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_DS);
8037 	/* Make DS usable */
8038 	vmcs_write(GUEST_AR_DS, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8039 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_DS);
8040 	vmcs_write(GUEST_AR_DS, ar_saved);
8041 
8042 	ar_saved = vmcs_read(GUEST_AR_ES);
8043 	/* Make ES unusable */
8044 	vmcs_write(GUEST_AR_ES, ar_saved | GUEST_SEG_UNUSABLE_MASK);
8045 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(false, GUEST_BASE_ES);
8046 	/* Make ES usable */
8047 	vmcs_write(GUEST_AR_ES, ar_saved & ~GUEST_SEG_UNUSABLE_MASK);
8048 	TEST_SEGMENT_BASE_ADDR_UPPER_BITS(true, GUEST_BASE_ES);
8049 	vmcs_write(GUEST_AR_ES, ar_saved);
8050 }
8051 
8052 /*
8053  * Check that the virtual CPU checks the VMX Guest State Area as
8054  * documented in the Intel SDM.
8055  */
8056 static void vmx_guest_state_area_test(void)
8057 {
8058 	vmx_set_test_stage(1);
8059 	test_set_guest(guest_state_test_main);
8060 
8061 	/*
8062 	 * The IA32_SYSENTER_ESP field and the IA32_SYSENTER_EIP field
8063 	 * must each contain a canonical address.
8064 	 */
8065 	test_canonical(GUEST_SYSENTER_ESP, "GUEST_SYSENTER_ESP", false);
8066 	test_canonical(GUEST_SYSENTER_EIP, "GUEST_SYSENTER_EIP", false);
8067 
8068 	test_guest_dr7();
8069 	test_load_guest_pat();
8070 	test_guest_efer();
8071 	test_load_guest_perf_global_ctrl();
8072 	test_load_guest_bndcfgs();
8073 
8074 	test_guest_segment_sel_fields();
8075 	test_guest_segment_base_addr_fields();
8076 
8077 	test_canonical(GUEST_BASE_GDTR, "GUEST_BASE_GDTR", false);
8078 	test_canonical(GUEST_BASE_IDTR, "GUEST_BASE_IDTR", false);
8079 
8080 	u32 guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_GDTR);
8081 	int i;
8082 	for (i = 16; i <= 31; i++) {
8083 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8084 		vmcs_write(GUEST_LIMIT_GDTR, tmp);
8085 		test_guest_state("GDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_GDTR");
8086 	}
8087 	vmcs_write(GUEST_LIMIT_GDTR, guest_desc_limit_saved);
8088 
8089 	guest_desc_limit_saved = vmcs_read(GUEST_LIMIT_IDTR);
8090 	for (i = 16; i <= 31; i++) {
8091 		u32 tmp = guest_desc_limit_saved | (1ull << i);
8092 		vmcs_write(GUEST_LIMIT_IDTR, tmp);
8093 		test_guest_state("IDT.limit > 0xffff", true, tmp, "GUEST_LIMIT_IDTR");
8094 	}
8095 	vmcs_write(GUEST_LIMIT_IDTR, guest_desc_limit_saved);
8096 
8097 	/*
8098 	 * Let the guest finish execution
8099 	 */
8100 	vmx_set_test_stage(2);
8101 	enter_guest();
8102 }
8103 
8104 extern void unrestricted_guest_main(void);
8105 asm (".code32\n"
8106 	"unrestricted_guest_main:\n"
8107 	"vmcall\n"
8108 	"nop\n"
8109 	"mov $1, %edi\n"
8110 	"call hypercall\n"
8111 	".code64\n");
8112 
8113 static void setup_unrestricted_guest(void)
8114 {
8115 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) & ~(X86_CR0_PG));
8116 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) & ~ENT_GUEST_64);
8117 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) & ~EFER_LMA);
8118 	vmcs_write(GUEST_RIP, virt_to_phys(unrestricted_guest_main));
8119 }
8120 
8121 static void unsetup_unrestricted_guest(void)
8122 {
8123 	vmcs_write(GUEST_CR0, vmcs_read(GUEST_CR0) | X86_CR0_PG);
8124 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_GUEST_64);
8125 	vmcs_write(GUEST_EFER, vmcs_read(GUEST_EFER) | EFER_LMA);
8126 	vmcs_write(GUEST_RIP, (u64) phys_to_virt(vmcs_read(GUEST_RIP)));
8127 	vmcs_write(GUEST_RSP, (u64) phys_to_virt(vmcs_read(GUEST_RSP)));
8128 }
8129 
8130 /*
8131  * If "unrestricted guest" secondary VM-execution control is set, guests
8132  * can run in unpaged protected mode.
8133  */
8134 static void vmentry_unrestricted_guest_test(void)
8135 {
8136 	if (enable_unrestricted_guest(true)) {
8137 		report_skip("Unrestricted guest not supported");
8138 		return;
8139 	}
8140 
8141 	test_set_guest(unrestricted_guest_main);
8142 	setup_unrestricted_guest();
8143 	test_guest_state("Unrestricted guest test", false, CPU_URG, "CPU_URG");
8144 
8145 	/*
8146 	 * Let the guest finish execution as a regular guest
8147 	 */
8148 	unsetup_unrestricted_guest();
8149 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) & ~CPU_URG);
8150 	enter_guest();
8151 }
8152 
8153 static bool valid_vmcs_for_vmentry(void)
8154 {
8155 	struct vmcs *current_vmcs = NULL;
8156 
8157 	if (vmcs_save(&current_vmcs))
8158 		return false;
8159 
8160 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
8161 }
8162 
8163 static void try_vmentry_in_movss_shadow(void)
8164 {
8165 	u32 vm_inst_err;
8166 	u32 flags;
8167 	bool early_failure = false;
8168 	u32 expected_flags = X86_EFLAGS_FIXED;
8169 	bool valid_vmcs = valid_vmcs_for_vmentry();
8170 
8171 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
8172 
8173 	/*
8174 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
8175 	 * unsupported VMCS component").
8176 	 */
8177 	vmcs_write(~0u, 0);
8178 
8179 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
8180 			      "vmwrite %%rsp, %%rdx;"
8181 			      "mov 0f, %%rax;"
8182 			      "mov %[host_rip], %%edx;"
8183 			      "vmwrite %%rax, %%rdx;"
8184 			      "mov $-1, %%ah;"
8185 			      "sahf;"
8186 			      "mov %%ss, %%ax;"
8187 			      "mov %%ax, %%ss;"
8188 			      "vmlaunch;"
8189 			      "mov $1, %[early_failure];"
8190 			      "0: lahf;"
8191 			      "movzbl %%ah, %[flags]"
8192 			      : [early_failure] "+r" (early_failure),
8193 				[flags] "=&a" (flags)
8194 			      : [host_rsp] "i" (HOST_RSP),
8195 				[host_rip] "i" (HOST_RIP)
8196 			      : "rdx", "cc", "memory");
8197 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
8198 
8199 	report(early_failure, "Early VM-entry failure");
8200 	report(flags == expected_flags, "RFLAGS[8:0] is %x (actual %x)",
8201 	       expected_flags, flags);
8202 	if (valid_vmcs)
8203 		report(vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
8204 		       "VM-instruction error is %d (actual %d)",
8205 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
8206 }
8207 
8208 static void vmentry_movss_shadow_test(void)
8209 {
8210 	struct vmcs *orig_vmcs;
8211 
8212 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
8213 
8214 	/*
8215 	 * Set the launched flag on the current VMCS to verify the correct
8216 	 * error priority, below.
8217 	 */
8218 	test_set_guest(v2_null_test_guest);
8219 	enter_guest();
8220 
8221 	/*
8222 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
8223 	 * fail due to invalid guest state (if we make it that far).
8224 	 */
8225 	vmcs_write(GUEST_RFLAGS, 0);
8226 
8227 	/*
8228 	 * "VM entry with events blocked by MOV SS" takes precedence over
8229 	 * "VMLAUNCH with non-clear VMCS."
8230 	 */
8231 	report_prefix_push("valid current-VMCS");
8232 	try_vmentry_in_movss_shadow();
8233 	report_prefix_pop();
8234 
8235 	/*
8236 	 * VMfailInvalid takes precedence over "VM entry with events
8237 	 * blocked by MOV SS."
8238 	 */
8239 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
8240 	report_prefix_push("no current-VMCS");
8241 	try_vmentry_in_movss_shadow();
8242 	report_prefix_pop();
8243 
8244 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
8245 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8246 }
8247 
8248 static void vmx_ldtr_test_guest(void)
8249 {
8250 	u16 ldtr = sldt();
8251 
8252 	report(ldtr == NP_SEL, "Expected %x for L2 LDTR selector (got %x)",
8253 	       NP_SEL, ldtr);
8254 }
8255 
8256 /*
8257  * Ensure that the L1 LDTR is set to 0 on VM-exit.
8258  */
8259 static void vmx_ldtr_test(void)
8260 {
8261 	const u8 ldt_ar = 0x82; /* Present LDT */
8262 	u16 sel = FIRST_SPARE_SEL;
8263 
8264 	/* Set up a non-zero L1 LDTR prior to VM-entry. */
8265 	set_gdt_entry(sel, 0, 0, ldt_ar, 0);
8266 	lldt(sel);
8267 
8268 	test_set_guest(vmx_ldtr_test_guest);
8269 	/*
8270 	 * Set up a different LDTR for L2. The actual GDT contents are
8271 	 * irrelevant, since we stuff the hidden descriptor state
8272 	 * straight into the VMCS rather than reading it from the GDT.
8273 	 */
8274 	vmcs_write(GUEST_SEL_LDTR, NP_SEL);
8275 	vmcs_write(GUEST_AR_LDTR, ldt_ar);
8276 	enter_guest();
8277 
8278 	/*
8279 	 * VM-exit should clear LDTR (and make it unusable, but we
8280 	 * won't verify that here).
8281 	 */
8282 	sel = sldt();
8283 	report(!sel, "Expected 0 for L1 LDTR selector (got %x)", sel);
8284 }
8285 
8286 static void vmx_single_vmcall_guest(void)
8287 {
8288 	vmcall();
8289 }
8290 
8291 static void vmx_cr_load_test(void)
8292 {
8293 	unsigned long cr3, cr4, orig_cr3, orig_cr4;
8294 	u32 ctrls[2] = {0};
8295 	pgd_t *pml5;
8296 
8297 	orig_cr4 = read_cr4();
8298 	orig_cr3 = read_cr3();
8299 
8300 	if (!this_cpu_has(X86_FEATURE_PCID)) {
8301 		report_skip("PCID not detected");
8302 		return;
8303 	}
8304 	if (!this_cpu_has(X86_FEATURE_MCE)) {
8305 		report_skip("MCE not detected");
8306 		return;
8307 	}
8308 
8309 	TEST_ASSERT(!(orig_cr3 & X86_CR3_PCID_MASK));
8310 
8311 	/* Enable PCID for L1. */
8312 	cr4 = orig_cr4 | X86_CR4_PCIDE;
8313 	cr3 = orig_cr3 | 0x1;
8314 	TEST_ASSERT(!write_cr4_safe(cr4));
8315 	write_cr3(cr3);
8316 
8317 	test_set_guest(vmx_single_vmcall_guest);
8318 	vmcs_write(HOST_CR4, cr4);
8319 	vmcs_write(HOST_CR3, cr3);
8320 	enter_guest();
8321 
8322 	/*
8323 	 * No exception is expected.
8324 	 *
8325 	 * NB. KVM loads the last guest write to CR4 into CR4 read
8326 	 *     shadow. In order to trigger an exit to KVM, we can toggle a
8327 	 *     bit that is owned by KVM. We use CR4.MCE, which shall
8328 	 *     have no side effect because normally no guest MCE (e.g., as the
8329 	 *     result of bad memory) would happen during this test.
8330 	 */
8331 	TEST_ASSERT(!write_cr4_safe(cr4 ^ X86_CR4_MCE));
8332 
8333 	/* Cleanup L1 state. */
8334 	write_cr3(orig_cr3);
8335 	TEST_ASSERT(!write_cr4_safe(orig_cr4));
8336 
8337 	if (!this_cpu_has(X86_FEATURE_LA57))
8338 		goto done;
8339 
8340 	/*
8341 	 * Allocate a full page for PML5 to guarantee alignment, though only
8342 	 * the first entry needs to be filled (the test's virtual addresses
8343 	 * most definitely do not have any of bits 56:48 set).
8344 	 */
8345 	pml5 = alloc_page();
8346 	*pml5 = orig_cr3 | PT_PRESENT_MASK | PT_WRITABLE_MASK;
8347 
8348 	/*
8349 	 * Transition to/from 5-level paging in the host via VM-Exit.  CR4.LA57
8350 	 * can't be toggled while long is active via MOV CR4, but there are no
8351 	 * such restrictions on VM-Exit.
8352 	 */
8353 lol_5level:
8354 	vmcs_write(HOST_CR4, orig_cr4 | X86_CR4_LA57);
8355 	vmcs_write(HOST_CR3, virt_to_phys(pml5));
8356 	enter_guest();
8357 
8358 	/*
8359 	 * VMREAD with a memory operand to verify KVM detects the LA57 change,
8360 	 * e.g. uses the correct guest root level in gva_to_gpa().
8361 	 */
8362 	TEST_ASSERT(vmcs_readm(HOST_CR3) == virt_to_phys(pml5));
8363 	TEST_ASSERT(vmcs_readm(HOST_CR4) == (orig_cr4 | X86_CR4_LA57));
8364 
8365 	vmcs_write(HOST_CR4, orig_cr4);
8366 	vmcs_write(HOST_CR3, orig_cr3);
8367 	enter_guest();
8368 
8369 	TEST_ASSERT(vmcs_readm(HOST_CR3) == orig_cr3);
8370 	TEST_ASSERT(vmcs_readm(HOST_CR4) == orig_cr4);
8371 
8372 	/*
8373 	 * And now do the same LA57 shenanigans with EPT enabled.  KVM uses
8374 	 * two separate MMUs when L1 uses TDP, whereas the above shadow paging
8375 	 * version shares an MMU between L1 and L2.
8376 	 *
8377 	 * If the saved execution controls are non-zero then the EPT version
8378 	 * has already run.  In that case, restore the old controls.  If EPT
8379 	 * setup fails, e.g. EPT isn't supported, fall through and finish up.
8380 	 */
8381 	if (ctrls[0]) {
8382 		vmcs_write(CPU_EXEC_CTRL0, ctrls[0]);
8383 		vmcs_write(CPU_EXEC_CTRL1, ctrls[1]);
8384 	} else if (!setup_ept(false)) {
8385 		ctrls[0] = vmcs_read(CPU_EXEC_CTRL0);
8386 		ctrls[1]  = vmcs_read(CPU_EXEC_CTRL1);
8387 		goto lol_5level;
8388 	}
8389 
8390 	free_page(pml5);
8391 
8392 done:
8393 	skip_exit_vmcall();
8394 	enter_guest();
8395 }
8396 
8397 static void vmx_cr4_osxsave_test_guest(void)
8398 {
8399 	write_cr4(read_cr4() & ~X86_CR4_OSXSAVE);
8400 }
8401 
8402 /*
8403  * Ensure that kvm recalculates the L1 guest's CPUID.01H:ECX.OSXSAVE
8404  * after VM-exit from an L2 guest that sets CR4.OSXSAVE to a different
8405  * value than in L1.
8406  */
8407 static void vmx_cr4_osxsave_test(void)
8408 {
8409 	if (!this_cpu_has(X86_FEATURE_XSAVE)) {
8410 		report_skip("XSAVE not detected");
8411 		return;
8412 	}
8413 
8414 	if (!(read_cr4() & X86_CR4_OSXSAVE)) {
8415 		unsigned long cr4 = read_cr4() | X86_CR4_OSXSAVE;
8416 
8417 		write_cr4(cr4);
8418 		vmcs_write(GUEST_CR4, cr4);
8419 		vmcs_write(HOST_CR4, cr4);
8420 	}
8421 
8422 	TEST_ASSERT(cpuid_osxsave());
8423 
8424 	test_set_guest(vmx_cr4_osxsave_test_guest);
8425 	enter_guest();
8426 
8427 	TEST_ASSERT(cpuid_osxsave());
8428 }
8429 
8430 static void vmx_nm_test_guest(void)
8431 {
8432 	write_cr0(read_cr0() | X86_CR0_TS);
8433 	asm volatile("fnop");
8434 }
8435 
8436 static void check_nm_exit(const char *test)
8437 {
8438 	u32 reason = vmcs_read(EXI_REASON);
8439 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
8440 	const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8441 		NM_VECTOR;
8442 
8443 	report(reason == VMX_EXC_NMI && intr_info == expected, "%s", test);
8444 }
8445 
8446 /*
8447  * This test checks that:
8448  *
8449  * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then
8450  *     a subsequent #NM VM-exit is reflected to L1.
8451  *
8452  * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a
8453  *     subsequent #NM VM-exit is reflected to L1.
8454  */
8455 static void vmx_nm_test(void)
8456 {
8457 	unsigned long cr0 = read_cr0();
8458 
8459 	test_set_guest(vmx_nm_test_guest);
8460 
8461 	/*
8462 	 * L1 wants to intercept #NM exceptions encountered in L2.
8463 	 */
8464 	vmcs_write(EXC_BITMAP, 1 << NM_VECTOR);
8465 
8466 	/*
8467 	 * Launch L2 with CR0.TS clear, but don't claim host ownership of
8468 	 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop,
8469 	 * which will raise #NM. L0 should reflect the #NM VM-exit to L1.
8470 	 */
8471 	vmcs_write(CR0_MASK, 0);
8472 	vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS);
8473 	enter_guest();
8474 	check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1");
8475 
8476 	/*
8477 	 * Re-enter L2 at the fnop instruction, with CR0.TS clear but
8478 	 * CR0.EM set. The fnop will still raise #NM, and L0 should
8479 	 * reflect the #NM VM-exit to L1.
8480 	 */
8481 	vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM);
8482 	enter_guest();
8483 	check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1");
8484 
8485 	/*
8486 	 * Re-enter L2 at the fnop instruction, with both CR0.TS and
8487 	 * CR0.EM clear. There will be no #NM, and the L2 guest should
8488 	 * exit normally.
8489 	 */
8490 	vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM));
8491 	enter_guest();
8492 }
8493 
8494 bool vmx_pending_event_ipi_fired;
8495 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
8496 {
8497 	vmx_pending_event_ipi_fired = true;
8498 	eoi();
8499 }
8500 
8501 bool vmx_pending_event_guest_run;
8502 static void vmx_pending_event_guest(void)
8503 {
8504 	vmcall();
8505 	vmx_pending_event_guest_run = true;
8506 }
8507 
8508 static void vmx_pending_event_test_core(bool guest_hlt)
8509 {
8510 	int ipi_vector = 0xf1;
8511 
8512 	vmx_pending_event_ipi_fired = false;
8513 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
8514 
8515 	vmx_pending_event_guest_run = false;
8516 	test_set_guest(vmx_pending_event_guest);
8517 
8518 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
8519 
8520 	enter_guest();
8521 	skip_exit_vmcall();
8522 
8523 	if (guest_hlt)
8524 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8525 
8526 	irq_disable();
8527 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
8528 				   APIC_DM_FIXED | ipi_vector,
8529 				   0);
8530 
8531 	enter_guest();
8532 
8533 	assert_exit_reason(VMX_EXTINT);
8534 	report(!vmx_pending_event_guest_run,
8535 	       "Guest did not run before host received IPI");
8536 
8537 	irq_enable();
8538 	asm volatile ("nop");
8539 	irq_disable();
8540 	report(vmx_pending_event_ipi_fired,
8541 	       "Got pending interrupt after IRQ enabled");
8542 
8543 	if (guest_hlt)
8544 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8545 
8546 	enter_guest();
8547 	report(vmx_pending_event_guest_run,
8548 	       "Guest finished running when no interrupt");
8549 }
8550 
8551 static void vmx_pending_event_test(void)
8552 {
8553 	vmx_pending_event_test_core(false);
8554 }
8555 
8556 static void vmx_pending_event_hlt_test(void)
8557 {
8558 	vmx_pending_event_test_core(true);
8559 }
8560 
8561 static int vmx_window_test_db_count;
8562 
8563 static void vmx_window_test_db_handler(struct ex_regs *regs)
8564 {
8565 	vmx_window_test_db_count++;
8566 }
8567 
8568 static void vmx_nmi_window_test_guest(void)
8569 {
8570 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8571 
8572 	asm volatile("vmcall\n\t"
8573 		     "nop\n\t");
8574 
8575 	handle_exception(DB_VECTOR, NULL);
8576 }
8577 
8578 static void verify_nmi_window_exit(u64 rip)
8579 {
8580 	u32 exit_reason = vmcs_read(EXI_REASON);
8581 
8582 	report(exit_reason == VMX_NMI_WINDOW,
8583 	       "Exit reason (%d) is 'NMI window'", exit_reason);
8584 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8585 	       vmcs_read(GUEST_RIP), rip);
8586 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8587 }
8588 
8589 static void vmx_nmi_window_test(void)
8590 {
8591 	u64 nop_addr;
8592 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8593 
8594 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
8595 		report_skip("CPU does not support the \"Virtual NMIs\" VM-execution control.");
8596 		return;
8597 	}
8598 
8599 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
8600 		report_skip("CPU does not support the \"NMI-window exiting\" VM-execution control.");
8601 		return;
8602 	}
8603 
8604 	vmx_window_test_db_count = 0;
8605 
8606 	report_prefix_push("NMI-window");
8607 	test_set_guest(vmx_nmi_window_test_guest);
8608 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
8609 	enter_guest();
8610 	skip_exit_vmcall();
8611 	nop_addr = vmcs_read(GUEST_RIP);
8612 
8613 	/*
8614 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
8615 	 * RIP will not advance.
8616 	 */
8617 	report_prefix_push("active, no blocking");
8618 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8619 	enter_guest();
8620 	verify_nmi_window_exit(nop_addr);
8621 	report_prefix_pop();
8622 
8623 	/*
8624 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
8625 	 * a VM-exit on the next instruction after the nop. (The nop
8626 	 * is one byte.)
8627 	 */
8628 	report_prefix_push("active, blocking by MOV-SS");
8629 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8630 	enter_guest();
8631 	verify_nmi_window_exit(nop_addr + 1);
8632 	report_prefix_pop();
8633 
8634 	/*
8635 	 * Ask for "NMI-window exiting" (with event injection), and
8636 	 * expect a VM-exit after the event is injected. (RIP should
8637 	 * be at the address specified in the IDT entry for #DB.)
8638 	 */
8639 	report_prefix_push("active, no blocking, injecting #DB");
8640 	vmcs_write(ENT_INTR_INFO,
8641 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8642 	enter_guest();
8643 	verify_nmi_window_exit((u64)db_fault_addr);
8644 	report_prefix_pop();
8645 
8646 	/*
8647 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
8648 	 * a VM-exit after the next IRET (i.e. after the #DB handler
8649 	 * returns). So, RIP should be back at one byte past the nop.
8650 	 */
8651 	report_prefix_push("active, blocking by NMI");
8652 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
8653 	enter_guest();
8654 	verify_nmi_window_exit(nop_addr + 1);
8655 	report(vmx_window_test_db_count == 1,
8656 	       "#DB handler executed once (actual %d times)",
8657 	       vmx_window_test_db_count);
8658 	report_prefix_pop();
8659 
8660 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8661 		report_skip("CPU does not support activity state HLT.");
8662 	} else {
8663 		/*
8664 		 * Ask for "NMI-window exiting" when entering activity
8665 		 * state HLT, and expect an immediate VM-exit. RIP is
8666 		 * still one byte past the nop.
8667 		 */
8668 		report_prefix_push("halted, no blocking");
8669 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8670 		enter_guest();
8671 		verify_nmi_window_exit(nop_addr + 1);
8672 		report_prefix_pop();
8673 
8674 		/*
8675 		 * Ask for "NMI-window exiting" when entering activity
8676 		 * state HLT (with event injection), and expect a
8677 		 * VM-exit after the event is injected. (RIP should be
8678 		 * at the address specified in the IDT entry for #DB.)
8679 		 */
8680 		report_prefix_push("halted, no blocking, injecting #DB");
8681 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8682 		vmcs_write(ENT_INTR_INFO,
8683 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8684 			   DB_VECTOR);
8685 		enter_guest();
8686 		verify_nmi_window_exit((u64)db_fault_addr);
8687 		report_prefix_pop();
8688 	}
8689 
8690 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
8691 	enter_guest();
8692 	report_prefix_pop();
8693 }
8694 
8695 static void vmx_intr_window_test_guest(void)
8696 {
8697 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
8698 
8699 	/*
8700 	 * The two consecutive STIs are to ensure that only the first
8701 	 * one has a shadow. Note that NOP and STI are one byte
8702 	 * instructions.
8703 	 */
8704 	asm volatile("vmcall\n\t"
8705 		     "nop\n\t"
8706 		     "sti\n\t"
8707 		     "sti\n\t");
8708 
8709 	handle_exception(DB_VECTOR, NULL);
8710 }
8711 
8712 static void verify_intr_window_exit(u64 rip)
8713 {
8714 	u32 exit_reason = vmcs_read(EXI_REASON);
8715 
8716 	report(exit_reason == VMX_INTR_WINDOW,
8717 	       "Exit reason (%d) is 'interrupt window'", exit_reason);
8718 	report(vmcs_read(GUEST_RIP) == rip, "RIP (%#lx) is %#lx",
8719 	       vmcs_read(GUEST_RIP), rip);
8720 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
8721 }
8722 
8723 static void vmx_intr_window_test(void)
8724 {
8725 	u64 vmcall_addr;
8726 	u64 nop_addr;
8727 	unsigned int orig_db_gate_type;
8728 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
8729 
8730 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
8731 		report_skip("CPU does not support the \"interrupt-window exiting\" VM-execution control.");
8732 		return;
8733 	}
8734 
8735 	/*
8736 	 * Change the IDT entry for #DB from interrupt gate to trap gate,
8737 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
8738 	 * be disabled after vectoring a #DB.
8739 	 */
8740 	orig_db_gate_type = boot_idt[DB_VECTOR].type;
8741 	boot_idt[DB_VECTOR].type = 15;
8742 
8743 	report_prefix_push("interrupt-window");
8744 	test_set_guest(vmx_intr_window_test_guest);
8745 	enter_guest();
8746 	assert_exit_reason(VMX_VMCALL);
8747 	vmcall_addr = vmcs_read(GUEST_RIP);
8748 
8749 	/*
8750 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
8751 	 * no blocking; expect an immediate VM-exit. Note that we have
8752 	 * not advanced past the vmcall instruction yet, so RIP should
8753 	 * point to the vmcall instruction.
8754 	 */
8755 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
8756 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8757 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
8758 	enter_guest();
8759 	verify_intr_window_exit(vmcall_addr);
8760 	report_prefix_pop();
8761 
8762 	/*
8763 	 * Ask for "interrupt-window exiting" (with event injection)
8764 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
8765 	 * the event is injected. That is, RIP should should be at the
8766 	 * address specified in the IDT entry for #DB.
8767 	 */
8768 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #DB");
8769 	vmcs_write(ENT_INTR_INFO,
8770 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8771 	vmcall_addr = vmcs_read(GUEST_RIP);
8772 	enter_guest();
8773 	verify_intr_window_exit((u64)db_fault_addr);
8774 	report_prefix_pop();
8775 
8776 	/*
8777 	 * Let the L2 guest run through the IRET, back to the VMCALL.
8778 	 * We have to clear the "interrupt-window exiting"
8779 	 * VM-execution control, or it would just keep causing
8780 	 * VM-exits. Then, advance past the VMCALL and set the
8781 	 * "interrupt-window exiting" VM-execution control again.
8782 	 */
8783 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8784 	enter_guest();
8785 	skip_exit_vmcall();
8786 	nop_addr = vmcs_read(GUEST_RIP);
8787 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8788 
8789 	/*
8790 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
8791 	 * RFLAGS.IF set, and expect a VM-exit on the next
8792 	 * instruction. (NOP is one byte.)
8793 	 */
8794 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
8795 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
8796 	enter_guest();
8797 	verify_intr_window_exit(nop_addr + 1);
8798 	report_prefix_pop();
8799 
8800 	/*
8801 	 * Back up to the NOP and ask for "interrupt-window exiting"
8802 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
8803 	 * on the next instruction. (NOP is one byte.)
8804 	 */
8805 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
8806 	vmcs_write(GUEST_RIP, nop_addr);
8807 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
8808 	enter_guest();
8809 	verify_intr_window_exit(nop_addr + 1);
8810 	report_prefix_pop();
8811 
8812 	/*
8813 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
8814 	 * and expect a VM-exit on the instruction following the STI
8815 	 * shadow. Only the first STI (which is one byte past the NOP)
8816 	 * should have a shadow. The second STI (which is two bytes
8817 	 * past the NOP) has no shadow. Therefore, the interrupt
8818 	 * window opens at three bytes past the NOP.
8819 	 */
8820 	report_prefix_push("active, RFLAGS.IF = 0");
8821 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
8822 	enter_guest();
8823 	verify_intr_window_exit(nop_addr + 3);
8824 	report_prefix_pop();
8825 
8826 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
8827 		report_skip("CPU does not support activity state HLT.");
8828 	} else {
8829 		/*
8830 		 * Ask for "interrupt-window exiting" when entering
8831 		 * activity state HLT, and expect an immediate
8832 		 * VM-exit. RIP is still three bytes past the nop.
8833 		 */
8834 		report_prefix_push("halted, no blocking");
8835 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8836 		enter_guest();
8837 		verify_intr_window_exit(nop_addr + 3);
8838 		report_prefix_pop();
8839 
8840 		/*
8841 		 * Ask for "interrupt-window exiting" when entering
8842 		 * activity state HLT (with event injection), and
8843 		 * expect a VM-exit after the event is injected. That
8844 		 * is, RIP should should be at the address specified
8845 		 * in the IDT entry for #DB.
8846 		 */
8847 		report_prefix_push("halted, no blocking, injecting #DB");
8848 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8849 		vmcs_write(ENT_INTR_INFO,
8850 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
8851 			   DB_VECTOR);
8852 		enter_guest();
8853 		verify_intr_window_exit((u64)db_fault_addr);
8854 		report_prefix_pop();
8855 	}
8856 
8857 	boot_idt[DB_VECTOR].type = orig_db_gate_type;
8858 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
8859 	enter_guest();
8860 	report_prefix_pop();
8861 }
8862 
8863 #define GUEST_TSC_OFFSET (1u << 30)
8864 
8865 static u64 guest_tsc;
8866 
8867 static void vmx_store_tsc_test_guest(void)
8868 {
8869 	guest_tsc = rdtsc();
8870 }
8871 
8872 /*
8873  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
8874  * list, the value saved is not subject to the TSC offset that is
8875  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
8876  */
8877 static void vmx_store_tsc_test(void)
8878 {
8879 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
8880 	u64 low, high;
8881 
8882 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
8883 		report_skip("'Use TSC offsetting' not supported");
8884 		return;
8885 	}
8886 
8887 	test_set_guest(vmx_store_tsc_test_guest);
8888 
8889 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
8890 	vmcs_write(EXI_MSR_ST_CNT, 1);
8891 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
8892 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
8893 
8894 	low = rdtsc();
8895 	enter_guest();
8896 	high = rdtsc();
8897 
8898 	report(low + GUEST_TSC_OFFSET <= guest_tsc &&
8899 	       guest_tsc <= high + GUEST_TSC_OFFSET,
8900 	       "RDTSC value in the guest (%lu) is in range [%lu, %lu]",
8901 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
8902 	report(low <= msr_entry.value && msr_entry.value <= high,
8903 	       "IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
8904 	       msr_entry.value, low, high);
8905 }
8906 
8907 static void vmx_preemption_timer_zero_test_db_handler(struct ex_regs *regs)
8908 {
8909 }
8910 
8911 static void vmx_preemption_timer_zero_test_guest(void)
8912 {
8913 	while (vmx_get_test_stage() < 3)
8914 		vmcall();
8915 }
8916 
8917 static void vmx_preemption_timer_zero_activate_preemption_timer(void)
8918 {
8919 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
8920 	vmcs_write(PREEMPT_TIMER_VALUE, 0);
8921 }
8922 
8923 static void vmx_preemption_timer_zero_advance_past_vmcall(void)
8924 {
8925 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
8926 	enter_guest();
8927 	skip_exit_vmcall();
8928 }
8929 
8930 static void vmx_preemption_timer_zero_inject_db(bool intercept_db)
8931 {
8932 	vmx_preemption_timer_zero_activate_preemption_timer();
8933 	vmcs_write(ENT_INTR_INFO, INTR_INFO_VALID_MASK |
8934 		   INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
8935 	vmcs_write(EXC_BITMAP, intercept_db ? 1 << DB_VECTOR : 0);
8936 	enter_guest();
8937 }
8938 
8939 static void vmx_preemption_timer_zero_set_pending_dbg(u32 exception_bitmap)
8940 {
8941 	vmx_preemption_timer_zero_activate_preemption_timer();
8942 	vmcs_write(GUEST_PENDING_DEBUG, PENDING_DBG_TRAP | DR6_TRAP1);
8943 	vmcs_write(EXC_BITMAP, exception_bitmap);
8944 	enter_guest();
8945 }
8946 
8947 static void vmx_preemption_timer_zero_expect_preempt_at_rip(u64 expected_rip)
8948 {
8949 	u32 reason = (u32)vmcs_read(EXI_REASON);
8950 	u64 guest_rip = vmcs_read(GUEST_RIP);
8951 
8952 	report(reason == VMX_PREEMPT && guest_rip == expected_rip,
8953 	       "Exit reason is 0x%x (expected 0x%x) and guest RIP is %lx (0x%lx expected).",
8954 	       reason, VMX_PREEMPT, guest_rip, expected_rip);
8955 }
8956 
8957 /*
8958  * This test ensures that when the VMX preemption timer is zero at
8959  * VM-entry, a VM-exit occurs after any event injection and after any
8960  * pending debug exceptions are raised, but before execution of any
8961  * guest instructions.
8962  */
8963 static void vmx_preemption_timer_zero_test(void)
8964 {
8965 	u64 db_fault_address = (u64)get_idt_addr(&boot_idt[DB_VECTOR]);
8966 	handler old_db;
8967 	u32 reason;
8968 
8969 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
8970 		report_skip("'Activate VMX-preemption timer' not supported");
8971 		return;
8972 	}
8973 
8974 	/*
8975 	 * Install a custom #DB handler that doesn't abort.
8976 	 */
8977 	old_db = handle_exception(DB_VECTOR,
8978 				  vmx_preemption_timer_zero_test_db_handler);
8979 
8980 	test_set_guest(vmx_preemption_timer_zero_test_guest);
8981 
8982 	/*
8983 	 * VMX-preemption timer should fire after event injection.
8984 	 */
8985 	vmx_set_test_stage(0);
8986 	vmx_preemption_timer_zero_inject_db(0);
8987 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8988 	vmx_preemption_timer_zero_advance_past_vmcall();
8989 
8990 	/*
8991 	 * VMX-preemption timer should fire after event injection.
8992 	 * Exception bitmap is irrelevant, since you can't intercept
8993 	 * an event that you injected.
8994 	 */
8995 	vmx_set_test_stage(1);
8996 	vmx_preemption_timer_zero_inject_db(true);
8997 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
8998 	vmx_preemption_timer_zero_advance_past_vmcall();
8999 
9000 	/*
9001 	 * VMX-preemption timer should fire after pending debug exceptions
9002 	 * have delivered a #DB trap.
9003 	 */
9004 	vmx_set_test_stage(2);
9005 	vmx_preemption_timer_zero_set_pending_dbg(0);
9006 	vmx_preemption_timer_zero_expect_preempt_at_rip(db_fault_address);
9007 	vmx_preemption_timer_zero_advance_past_vmcall();
9008 
9009 	/*
9010 	 * VMX-preemption timer would fire after pending debug exceptions
9011 	 * have delivered a #DB trap, but in this case, the #DB trap is
9012 	 * intercepted.
9013 	 */
9014 	vmx_set_test_stage(3);
9015 	vmx_preemption_timer_zero_set_pending_dbg(1 << DB_VECTOR);
9016 	reason = (u32)vmcs_read(EXI_REASON);
9017 	report(reason == VMX_EXC_NMI, "Exit reason is 0x%x (expected 0x%x)",
9018 	       reason, VMX_EXC_NMI);
9019 
9020 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9021 	enter_guest();
9022 
9023 	handle_exception(DB_VECTOR, old_db);
9024 }
9025 
9026 static u64 vmx_preemption_timer_tf_test_prev_rip;
9027 
9028 static void vmx_preemption_timer_tf_test_db_handler(struct ex_regs *regs)
9029 {
9030 	extern char vmx_preemption_timer_tf_test_endloop;
9031 
9032 	if (vmx_get_test_stage() == 2) {
9033 		/*
9034 		 * Stage 2 means that we're done, one way or another.
9035 		 * Arrange for the iret to drop us out of the wbinvd
9036 		 * loop and stop single-stepping.
9037 		 */
9038 		regs->rip = (u64)&vmx_preemption_timer_tf_test_endloop;
9039 		regs->rflags &= ~X86_EFLAGS_TF;
9040 	} else if (regs->rip == vmx_preemption_timer_tf_test_prev_rip) {
9041 		/*
9042 		 * The RIP should alternate between the wbinvd and the
9043 		 * jmp instruction in the code below. If we ever see
9044 		 * the same instruction twice in a row, that means a
9045 		 * single-step trap has been dropped. Let the
9046 		 * hypervisor know about the failure by executing a
9047 		 * VMCALL.
9048 		 */
9049 		vmcall();
9050 	}
9051 	vmx_preemption_timer_tf_test_prev_rip = regs->rip;
9052 }
9053 
9054 static void vmx_preemption_timer_tf_test_guest(void)
9055 {
9056 	/*
9057 	 * The hypervisor doesn't intercept WBINVD, so the loop below
9058 	 * shouldn't be a problem--it's just two instructions
9059 	 * executing in VMX non-root mode. However, when the
9060 	 * hypervisor is running in a virtual environment, the parent
9061 	 * hypervisor might intercept WBINVD and emulate it. If the
9062 	 * parent hypervisor is broken, the single-step trap after the
9063 	 * WBINVD might be lost.
9064 	 */
9065 	asm volatile("vmcall\n\t"
9066 		     "0: wbinvd\n\t"
9067 		     "1: jmp 0b\n\t"
9068 		     "vmx_preemption_timer_tf_test_endloop:");
9069 }
9070 
9071 /*
9072  * Ensure that the delivery of a "VMX-preemption timer expired"
9073  * VM-exit doesn't disrupt single-stepping in the guest. Note that
9074  * passing this test doesn't ensure correctness, because the test will
9075  * only fail if the VMX-preemtion timer fires at the right time (or
9076  * the wrong time, as it were).
9077  */
9078 static void vmx_preemption_timer_tf_test(void)
9079 {
9080 	handler old_db;
9081 	u32 reason;
9082 	int i;
9083 
9084 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9085 		report_skip("'Activate VMX-preemption timer' not supported");
9086 		return;
9087 	}
9088 
9089 	old_db = handle_exception(DB_VECTOR,
9090 				  vmx_preemption_timer_tf_test_db_handler);
9091 
9092 	test_set_guest(vmx_preemption_timer_tf_test_guest);
9093 
9094 	enter_guest();
9095 	skip_exit_vmcall();
9096 
9097 	vmx_set_test_stage(1);
9098 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9099 	vmcs_write(PREEMPT_TIMER_VALUE, 50000);
9100 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9101 
9102 	/*
9103 	 * The only exit we should see is "VMX-preemption timer
9104 	 * expired."  If we get a VMCALL exit, that means the #DB
9105 	 * handler has detected a missing single-step trap. It doesn't
9106 	 * matter where the guest RIP is when the VMX-preemption timer
9107 	 * expires (whether it's in the WBINVD loop or in the #DB
9108 	 * handler)--a single-step trap should never be discarded.
9109 	 */
9110 	for (i = 0; i < 10000; i++) {
9111 		enter_guest();
9112 		reason = (u32)vmcs_read(EXI_REASON);
9113 		if (reason == VMX_PREEMPT)
9114 			continue;
9115 		TEST_ASSERT(reason == VMX_VMCALL);
9116 		skip_exit_insn();
9117 		break;
9118 	}
9119 
9120 	report(reason == VMX_PREEMPT, "No single-step traps skipped");
9121 
9122 	vmx_set_test_stage(2);
9123 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9124 	enter_guest();
9125 
9126 	handle_exception(DB_VECTOR, old_db);
9127 }
9128 
9129 #define VMX_PREEMPTION_TIMER_EXPIRY_CYCLES 1000000
9130 
9131 static u64 vmx_preemption_timer_expiry_start;
9132 static u64 vmx_preemption_timer_expiry_finish;
9133 
9134 static void vmx_preemption_timer_expiry_test_guest(void)
9135 {
9136 	vmcall();
9137 	vmx_preemption_timer_expiry_start = fenced_rdtsc();
9138 
9139 	while (vmx_get_test_stage() == 0)
9140 		vmx_preemption_timer_expiry_finish = fenced_rdtsc();
9141 }
9142 
9143 /*
9144  * Test that the VMX-preemption timer is not excessively delayed.
9145  *
9146  * Per the SDM, volume 3, VM-entry starts the VMX-preemption timer
9147  * with the unsigned value in the VMX-preemption timer-value field,
9148  * and the VMX-preemption timer counts down by 1 every time bit X in
9149  * the TSC changes due to a TSC increment (where X is
9150  * IA32_VMX_MISC[4:0]). If the timer counts down to zero in any state
9151  * other than the wait-for-SIPI state, the logical processor
9152  * transitions to the C0 C-state and causes a VM-exit.
9153  *
9154  * The guest code above reads the starting TSC after VM-entry. At this
9155  * point, the VMX-preemption timer has already been activated. Next,
9156  * the guest code reads the current TSC in a loop, storing the value
9157  * read to memory.
9158  *
9159  * If the RDTSC in the loop reads a value past the VMX-preemption
9160  * timer deadline, then the VMX-preemption timer VM-exit must be
9161  * delivered before the next instruction retires. Even if a higher
9162  * priority SMI is delivered first, the VMX-preemption timer VM-exit
9163  * must be delivered before the next instruction retires. Hence, a TSC
9164  * value past the VMX-preemption timer deadline might be read, but it
9165  * cannot be stored. If a TSC value past the deadline *is* stored,
9166  * then the architectural specification has been violated.
9167  */
9168 static void vmx_preemption_timer_expiry_test(void)
9169 {
9170 	u32 preemption_timer_value;
9171 	union vmx_misc misc;
9172 	u64 tsc_deadline;
9173 	u32 reason;
9174 
9175 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
9176 		report_skip("'Activate VMX-preemption timer' not supported");
9177 		return;
9178 	}
9179 
9180 	test_set_guest(vmx_preemption_timer_expiry_test_guest);
9181 
9182 	enter_guest();
9183 	skip_exit_vmcall();
9184 
9185 	misc.val = rdmsr(MSR_IA32_VMX_MISC);
9186 	preemption_timer_value =
9187 		VMX_PREEMPTION_TIMER_EXPIRY_CYCLES >> misc.pt_bit;
9188 
9189 	vmcs_set_bits(PIN_CONTROLS, PIN_PREEMPT);
9190 	vmcs_write(PREEMPT_TIMER_VALUE, preemption_timer_value);
9191 	vmx_set_test_stage(0);
9192 
9193 	enter_guest();
9194 	reason = (u32)vmcs_read(EXI_REASON);
9195 	TEST_ASSERT(reason == VMX_PREEMPT);
9196 
9197 	tsc_deadline = ((vmx_preemption_timer_expiry_start >> misc.pt_bit) <<
9198 			misc.pt_bit) + (preemption_timer_value << misc.pt_bit);
9199 
9200 	report(vmx_preemption_timer_expiry_finish < tsc_deadline,
9201 	       "Last stored guest TSC (%lu) < TSC deadline (%lu)",
9202 	       vmx_preemption_timer_expiry_finish, tsc_deadline);
9203 
9204 	vmcs_clear_bits(PIN_CONTROLS, PIN_PREEMPT);
9205 	vmx_set_test_stage(1);
9206 	enter_guest();
9207 }
9208 
9209 static void vmx_db_test_guest(void)
9210 {
9211 	/*
9212 	 * For a hardware generated single-step #DB.
9213 	 */
9214 	asm volatile("vmcall;"
9215 		     "nop;"
9216 		     ".Lpost_nop:");
9217 	/*
9218 	 * ...in a MOVSS shadow, with pending debug exceptions.
9219 	 */
9220 	asm volatile("vmcall;"
9221 		     "nop;"
9222 		     ".Lpost_movss_nop:");
9223 	/*
9224 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
9225 	 * emulates it in software.)
9226 	 */
9227 	asm volatile("vmcall;"
9228 		     "wbinvd;"
9229 		     ".Lpost_wbinvd:");
9230 	/*
9231 	 * ...in a MOVSS shadow, with pending debug exceptions.
9232 	 */
9233 	asm volatile("vmcall;"
9234 		     "wbinvd;"
9235 		     ".Lpost_movss_wbinvd:");
9236 	/*
9237 	 * For a hardware generated single-step #DB in a transactional region.
9238 	 */
9239 	asm volatile("vmcall;"
9240 		     ".Lxbegin: xbegin .Lskip_rtm;"
9241 		     "xend;"
9242 		     ".Lskip_rtm:");
9243 }
9244 
9245 /*
9246  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
9247  * L2. No #DB is delivered and L2 continues to the next point of
9248  * interest.
9249  */
9250 static void dismiss_db(void)
9251 {
9252 	vmcs_write(GUEST_PENDING_DEBUG, 0);
9253 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
9254 	enter_guest();
9255 }
9256 
9257 /*
9258  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
9259  * Then throw away the #DB exception and resume L2.
9260  */
9261 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
9262 			  void *expected_rip, u64 expected_exit_qual,
9263 			  u64 expected_dr6)
9264 {
9265 	u32 reason = vmcs_read(EXI_REASON);
9266 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
9267 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
9268 	u64 guest_rip = vmcs_read(GUEST_RIP);
9269 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
9270 	u64 dr6 = read_dr6();
9271 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
9272 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
9273 
9274 	report(reason == VMX_EXC_NMI && intr_info == expected_intr_info,
9275 	       "Expected #DB VM-exit");
9276 	report((u64)expected_rip == guest_rip, "Expected RIP %p (actual %lx)",
9277 	       expected_rip, guest_rip);
9278 	report_xfail(xfail_pdbg, 0 == guest_pending_dbg,
9279 		     "Expected pending debug exceptions 0 (actual %lx)",
9280 		     guest_pending_dbg);
9281 	report_xfail(xfail_qual, expected_exit_qual == exit_qual,
9282 		     "Expected exit qualification %lx (actual %lx)",
9283 		     expected_exit_qual, exit_qual);
9284 	report_xfail(xfail_dr6, expected_dr6 == dr6,
9285 		     "Expected DR6 %lx (actual %lx)", expected_dr6, dr6);
9286 	dismiss_db();
9287 }
9288 
9289 /*
9290  * Assuming the guest has just exited on a VMCALL instruction, skip
9291  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
9292  * pending debug exceptions are non-zero, set the VMCS up as if the
9293  * previous instruction was a MOVSS that generated the indicated
9294  * pending debug exceptions. Then enter L2.
9295  */
9296 static void single_step_guest(const char *test_name, u64 starting_dr6,
9297 			      u64 pending_debug_exceptions)
9298 {
9299 	printf("\n%s\n", test_name);
9300 	skip_exit_vmcall();
9301 	write_dr6(starting_dr6);
9302 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
9303 	if (pending_debug_exceptions) {
9304 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
9305 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
9306 	}
9307 	enter_guest();
9308 }
9309 
9310 /*
9311  * When L1 intercepts #DB, verify that a single-step trap clears
9312  * pending debug exceptions, populates the exit qualification field
9313  * properly, and that DR6 is not prematurely clobbered. In a
9314  * (simulated) MOVSS shadow, make sure that the pending debug
9315  * exception bits are properly accumulated into the exit qualification
9316  * field.
9317  */
9318 static void vmx_db_test(void)
9319 {
9320 	/*
9321 	 * We are going to set a few arbitrary bits in DR6 to verify that
9322 	 * (a) DR6 is not modified by an intercepted #DB, and
9323 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
9324          *     the exit qualification field for a subsequent #DB exception.
9325 	 */
9326 	const u64 starting_dr6 = DR6_ACTIVE_LOW | DR6_BS | DR6_TRAP3 | DR6_TRAP1;
9327 	extern char post_nop asm(".Lpost_nop");
9328 	extern char post_movss_nop asm(".Lpost_movss_nop");
9329 	extern char post_wbinvd asm(".Lpost_wbinvd");
9330 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
9331 	extern char xbegin asm(".Lxbegin");
9332 	extern char skip_rtm asm(".Lskip_rtm");
9333 
9334 	/*
9335 	 * L1 wants to intercept #DB exceptions encountered in L2.
9336 	 */
9337 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
9338 
9339 	/*
9340 	 * Start L2 and run it up to the first point of interest.
9341 	 */
9342 	test_set_guest(vmx_db_test_guest);
9343 	enter_guest();
9344 
9345 	/*
9346 	 * Hardware-delivered #DB trap for single-step sets the
9347 	 * standard that L0 has to follow for emulated instructions.
9348 	 */
9349 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
9350 	check_db_exit(false, false, false, &post_nop, DR6_BS, starting_dr6);
9351 
9352 	/*
9353 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
9354 	 * also sets the standard that L0 has to follow for emulated
9355 	 * instructions. Here, we establish the VMCS pending debug
9356 	 * exceptions to indicate that the simulated MOVSS triggered a
9357 	 * data breakpoint as well as the single-step trap.
9358 	 */
9359 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
9360 			  starting_dr6, DR6_BS | PENDING_DBG_TRAP | DR6_TRAP0);
9361 	check_db_exit(false, false, false, &post_movss_nop, DR6_BS | DR6_TRAP0,
9362 		      starting_dr6);
9363 
9364 	/*
9365 	 * L0 synthesized #DB trap for single-step is buggy, because
9366 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
9367 	 * reconstitute the exit qualification from the prematurely
9368 	 * modified DR6, but fails miserably.
9369 	 */
9370 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
9371 	check_db_exit(false, false, false, &post_wbinvd, DR6_BS, starting_dr6);
9372 
9373 	/*
9374 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
9375 	 * even worse, because L0 also leaves the pending debug
9376 	 * exceptions in the VMCS instead of accumulating them into
9377 	 * the exit qualification field for the #DB exception.
9378 	 */
9379 	single_step_guest("Software synthesized single-step in MOVSS shadow",
9380 			  starting_dr6, DR6_BS | PENDING_DBG_TRAP | DR6_TRAP0);
9381 	check_db_exit(true, false, true, &post_movss_wbinvd, DR6_BS | DR6_TRAP0,
9382 		      starting_dr6);
9383 
9384 	/*
9385 	 * Optional RTM test for hardware that supports RTM, to
9386 	 * demonstrate that the current volume 3 of the SDM
9387 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
9388 	 * qualification for debug exceptions is not reserved. It is
9389 	 * set to 1 if a debug exception (#DB) or a breakpoint
9390 	 * exception (#BP) occurs inside an RTM region while advanced
9391 	 * debugging of RTM transactional regions is enabled.
9392 	 */
9393 	if (this_cpu_has(X86_FEATURE_RTM)) {
9394 		vmcs_write(ENT_CONTROLS,
9395 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
9396 		/*
9397 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
9398 		 * in the guest to enable advanced debugging of RTM
9399 		 * transactional regions.
9400 		 */
9401 		vmcs_write(GUEST_DR7, BIT(11));
9402 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
9403 		single_step_guest("Hardware delivered single-step in "
9404 				  "transactional region", starting_dr6, 0);
9405 		check_db_exit(false, false, false, &xbegin, BIT(16),
9406 			      starting_dr6);
9407 	} else {
9408 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
9409 		enter_guest();
9410 	}
9411 }
9412 
9413 static void enable_vid(void)
9414 {
9415 	void *virtual_apic_page;
9416 
9417 	assert(cpu_has_apicv());
9418 
9419 	disable_intercept_for_x2apic_msrs();
9420 
9421 	virtual_apic_page = alloc_page();
9422 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
9423 
9424 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
9425 
9426 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
9427 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
9428 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
9429 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
9430 
9431 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
9432 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
9433 }
9434 
9435 static void trigger_ioapic_scan_thread(void *data)
9436 {
9437 	/* Wait until other CPU entered L2 */
9438 	while (vmx_get_test_stage() != 1)
9439 		;
9440 
9441 	/* Trigger ioapic scan */
9442 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
9443 	vmx_set_test_stage(2);
9444 }
9445 
9446 static void irq_79_handler_guest(isr_regs_t *regs)
9447 {
9448 	eoi();
9449 
9450 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
9451 	vmcall();
9452 }
9453 
9454 /*
9455  * Constant for num of busy-loop iterations after which
9456  * a timer interrupt should have happened in host
9457  */
9458 #define TIMER_INTERRUPT_DELAY 100000000
9459 
9460 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
9461 {
9462 	handle_irq(0x79, irq_79_handler_guest);
9463 	irq_enable();
9464 
9465 	/* Signal to L1 CPU to trigger ioapic scan */
9466 	vmx_set_test_stage(1);
9467 	/* Wait until L1 CPU to trigger ioapic scan */
9468 	while (vmx_get_test_stage() != 2)
9469 		;
9470 
9471 	/*
9472 	 * Wait for L0 timer interrupt to be raised while we run in L2
9473 	 * such that L0 will process the IOAPIC scan request before
9474 	 * resuming L2
9475 	 */
9476 	delay(TIMER_INTERRUPT_DELAY);
9477 
9478 	asm volatile ("int $0x79");
9479 }
9480 
9481 static void vmx_eoi_bitmap_ioapic_scan_test(void)
9482 {
9483 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
9484 		report_skip(__func__);
9485 		return;
9486 	}
9487 
9488 	enable_vid();
9489 
9490 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
9491 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
9492 
9493 	/*
9494 	 * Launch L2.
9495 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
9496 	 * In case the reason isn't VMX_VMCALL, the asserion inside
9497 	 * skip_exit_vmcall() will fail.
9498 	 */
9499 	enter_guest();
9500 	skip_exit_vmcall();
9501 
9502 	/* Let L2 finish */
9503 	enter_guest();
9504 	report_pass(__func__);
9505 }
9506 
9507 #define HLT_WITH_RVI_VECTOR		(0xf1)
9508 
9509 bool vmx_hlt_with_rvi_guest_isr_fired;
9510 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
9511 {
9512 	vmx_hlt_with_rvi_guest_isr_fired = true;
9513 	eoi();
9514 }
9515 
9516 static void vmx_hlt_with_rvi_guest(void)
9517 {
9518 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
9519 
9520 	irq_enable();
9521 	asm volatile ("nop");
9522 
9523 	vmcall();
9524 }
9525 
9526 static void vmx_hlt_with_rvi_test(void)
9527 {
9528 	if (!cpu_has_apicv()) {
9529 		report_skip(__func__);
9530 		return;
9531 	}
9532 
9533 	enable_vid();
9534 
9535 	vmx_hlt_with_rvi_guest_isr_fired = false;
9536 	test_set_guest(vmx_hlt_with_rvi_guest);
9537 
9538 	enter_guest();
9539 	skip_exit_vmcall();
9540 
9541 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
9542 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
9543 	enter_guest();
9544 
9545 	report(vmx_hlt_with_rvi_guest_isr_fired, "Interrupt raised in guest");
9546 }
9547 
9548 static void set_irq_line_thread(void *data)
9549 {
9550 	/* Wait until other CPU entered L2 */
9551 	while (vmx_get_test_stage() != 1)
9552 		;
9553 
9554 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
9555 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9556 	vmx_set_test_stage(2);
9557 }
9558 
9559 static bool irq_78_handler_vmcall_before_eoi;
9560 static void irq_78_handler_guest(isr_regs_t *regs)
9561 {
9562 	set_irq_line(0xf, 0);
9563 	if (irq_78_handler_vmcall_before_eoi)
9564 		vmcall();
9565 	eoi();
9566 	vmcall();
9567 }
9568 
9569 static void vmx_apic_passthrough_guest(void)
9570 {
9571 	handle_irq(0x78, irq_78_handler_guest);
9572 	irq_enable();
9573 
9574 	/* If requested, wait for other CPU to trigger ioapic scan */
9575 	if (vmx_get_test_stage() < 1) {
9576 		vmx_set_test_stage(1);
9577 		while (vmx_get_test_stage() != 2)
9578 			;
9579 	}
9580 
9581 	set_irq_line(0xf, 1);
9582 }
9583 
9584 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
9585 {
9586 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
9587 		report_skip(__func__);
9588 		return;
9589 	}
9590 
9591 	/* Test device is required for generating IRQs */
9592 	if (!test_device_enabled()) {
9593 		report_skip(__func__);
9594 		return;
9595 	}
9596 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9597 	u64 cpu_ctrl_1 = 0;
9598 
9599 	disable_intercept_for_x2apic_msrs();
9600 
9601 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9602 
9603 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9604 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9605 
9606 	if (set_irq_line_from_thread) {
9607 		irq_78_handler_vmcall_before_eoi = false;
9608 		on_cpu_async(1, set_irq_line_thread, NULL);
9609 	} else {
9610 		irq_78_handler_vmcall_before_eoi = true;
9611 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
9612 		vmx_set_test_stage(2);
9613 	}
9614 	test_set_guest(vmx_apic_passthrough_guest);
9615 
9616 	if (irq_78_handler_vmcall_before_eoi) {
9617 		/* Before EOI remote_irr should still be set */
9618 		enter_guest();
9619 		skip_exit_vmcall();
9620 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
9621 			"IOAPIC pass-through: remote_irr=1 before EOI");
9622 	}
9623 
9624 	/* After EOI remote_irr should be cleared */
9625 	enter_guest();
9626 	skip_exit_vmcall();
9627 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
9628 		"IOAPIC pass-through: remote_irr=0 after EOI");
9629 
9630 	/* Let L2 finish */
9631 	enter_guest();
9632 	report_pass(__func__);
9633 }
9634 
9635 static void vmx_apic_passthrough_test(void)
9636 {
9637 	vmx_apic_passthrough(false);
9638 }
9639 
9640 static void vmx_apic_passthrough_thread_test(void)
9641 {
9642 	vmx_apic_passthrough(true);
9643 }
9644 
9645 static void vmx_apic_passthrough_tpr_threshold_guest(void)
9646 {
9647 	cli();
9648 	apic_set_tpr(0);
9649 }
9650 
9651 static bool vmx_apic_passthrough_tpr_threshold_ipi_isr_fired;
9652 static void vmx_apic_passthrough_tpr_threshold_ipi_isr(isr_regs_t *regs)
9653 {
9654 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = true;
9655 	eoi();
9656 }
9657 
9658 static void vmx_apic_passthrough_tpr_threshold_test(void)
9659 {
9660 	int ipi_vector = 0xe1;
9661 
9662 	disable_intercept_for_x2apic_msrs();
9663 	vmcs_clear_bits(PIN_CONTROLS, PIN_EXTINT);
9664 
9665 	/* Raise L0 TPR-threshold by queueing vector in LAPIC IRR */
9666 	cli();
9667 	apic_set_tpr((ipi_vector >> 4) + 1);
9668 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
9669 			APIC_DM_FIXED | ipi_vector,
9670 			0);
9671 
9672 	test_set_guest(vmx_apic_passthrough_tpr_threshold_guest);
9673 	enter_guest();
9674 
9675 	report(apic_get_tpr() == 0, "TPR was zero by guest");
9676 
9677 	/* Clean pending self-IPI */
9678 	vmx_apic_passthrough_tpr_threshold_ipi_isr_fired = false;
9679 	handle_irq(ipi_vector, vmx_apic_passthrough_tpr_threshold_ipi_isr);
9680 	sti();
9681 	asm volatile ("nop");
9682 	report(vmx_apic_passthrough_tpr_threshold_ipi_isr_fired, "self-IPI fired");
9683 
9684 	report_pass(__func__);
9685 }
9686 
9687 static u64 init_signal_test_exit_reason;
9688 static bool init_signal_test_thread_continued;
9689 
9690 static void init_signal_test_thread(void *data)
9691 {
9692 	struct vmcs *test_vmcs = data;
9693 
9694 	/* Enter VMX operation (i.e. exec VMXON) */
9695 	u64 *ap_vmxon_region = alloc_page();
9696 	enable_vmx();
9697 	init_vmx(ap_vmxon_region);
9698 	_vmx_on(ap_vmxon_region);
9699 
9700 	/* Signal CPU have entered VMX operation */
9701 	vmx_set_test_stage(1);
9702 
9703 	/* Wait for BSP CPU to send INIT signal */
9704 	while (vmx_get_test_stage() != 2)
9705 		;
9706 
9707 	/*
9708 	 * Signal that we continue as usual as INIT signal
9709 	 * should be blocked while CPU is in VMX operation
9710 	 */
9711 	vmx_set_test_stage(3);
9712 
9713 	/* Wait for signal to enter VMX non-root mode */
9714 	while (vmx_get_test_stage() != 4)
9715 		;
9716 
9717 	/* Enter VMX non-root mode */
9718 	test_set_guest(v2_null_test_guest);
9719 	make_vmcs_current(test_vmcs);
9720 	enter_guest();
9721 	/* Save exit reason for BSP CPU to compare to expected result */
9722 	init_signal_test_exit_reason = vmcs_read(EXI_REASON);
9723 	/* VMCLEAR test-vmcs so it could be loaded by BSP CPU */
9724 	vmcs_clear(test_vmcs);
9725 	launched = false;
9726 	/* Signal that CPU exited to VMX root mode */
9727 	vmx_set_test_stage(5);
9728 
9729 	/* Wait for BSP CPU to signal to exit VMX operation */
9730 	while (vmx_get_test_stage() != 6)
9731 		;
9732 
9733 	/* Exit VMX operation (i.e. exec VMXOFF) */
9734 	vmx_off();
9735 
9736 	/*
9737 	 * Signal to BSP CPU that we continue as usual as INIT signal
9738 	 * should have been consumed by VMX_INIT exit from guest
9739 	 */
9740 	vmx_set_test_stage(7);
9741 
9742 	/* Wait for BSP CPU to signal to enter VMX operation */
9743 	while (vmx_get_test_stage() != 8)
9744 		;
9745 	/* Enter VMX operation (i.e. exec VMXON) */
9746 	_vmx_on(ap_vmxon_region);
9747 	/* Signal to BSP we are in VMX operation */
9748 	vmx_set_test_stage(9);
9749 
9750 	/* Wait for BSP CPU to send INIT signal */
9751 	while (vmx_get_test_stage() != 10)
9752 		;
9753 
9754 	/* Exit VMX operation (i.e. exec VMXOFF) */
9755 	vmx_off();
9756 
9757 	/*
9758 	 * Exiting VMX operation should result in latched
9759 	 * INIT signal being processed. Therefore, we should
9760 	 * never reach the below code. Thus, signal to BSP
9761 	 * CPU if we have reached here so it is able to
9762 	 * report an issue if it happens.
9763 	 */
9764 	init_signal_test_thread_continued = true;
9765 }
9766 
9767 #define INIT_SIGNAL_TEST_DELAY	100000000ULL
9768 
9769 static void vmx_init_signal_test(void)
9770 {
9771 	struct vmcs *test_vmcs;
9772 
9773 	if (cpu_count() < 2) {
9774 		report_skip(__func__);
9775 		return;
9776 	}
9777 
9778 	/* VMCLEAR test-vmcs so it could be loaded by other CPU */
9779 	vmcs_save(&test_vmcs);
9780 	vmcs_clear(test_vmcs);
9781 
9782 	vmx_set_test_stage(0);
9783 	on_cpu_async(1, init_signal_test_thread, test_vmcs);
9784 
9785 	/* Wait for other CPU to enter VMX operation */
9786 	while (vmx_get_test_stage() != 1)
9787 		;
9788 
9789 	/* Send INIT signal to other CPU */
9790 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9791 				   id_map[1]);
9792 	/* Signal other CPU we have sent INIT signal */
9793 	vmx_set_test_stage(2);
9794 
9795 	/*
9796 	 * Wait reasonable amount of time for INIT signal to
9797 	 * be received on other CPU and verify that other CPU
9798 	 * have proceed as usual to next test stage as INIT
9799 	 * signal should be blocked while other CPU in
9800 	 * VMX operation
9801 	 */
9802 	delay(INIT_SIGNAL_TEST_DELAY);
9803 	report(vmx_get_test_stage() == 3,
9804 	       "INIT signal blocked when CPU in VMX operation");
9805 	/* No point to continue if we failed at this point */
9806 	if (vmx_get_test_stage() != 3)
9807 		return;
9808 
9809 	/* Signal other CPU to enter VMX non-root mode */
9810 	init_signal_test_exit_reason = -1ull;
9811 	vmx_set_test_stage(4);
9812 	/*
9813 	 * Wait reasonable amont of time for other CPU
9814 	 * to exit to VMX root mode
9815 	 */
9816 	delay(INIT_SIGNAL_TEST_DELAY);
9817 	if (vmx_get_test_stage() != 5) {
9818 		report_fail("Pending INIT signal didn't result in VMX exit");
9819 		return;
9820 	}
9821 	report(init_signal_test_exit_reason == VMX_INIT,
9822 			"INIT signal during VMX non-root mode result in exit-reason %s (%lu)",
9823 			exit_reason_description(init_signal_test_exit_reason),
9824 			init_signal_test_exit_reason);
9825 
9826 	/* Run guest to completion */
9827 	make_vmcs_current(test_vmcs);
9828 	enter_guest();
9829 
9830 	/* Signal other CPU to exit VMX operation */
9831 	init_signal_test_thread_continued = false;
9832 	vmx_set_test_stage(6);
9833 
9834 	/* Wait reasonable amount of time for other CPU to exit VMX operation */
9835 	delay(INIT_SIGNAL_TEST_DELAY);
9836 	report(vmx_get_test_stage() == 7,
9837 	       "INIT signal consumed on VMX_INIT exit");
9838 	/* No point to continue if we failed at this point */
9839 	if (vmx_get_test_stage() != 7)
9840 		return;
9841 
9842 	/* Signal other CPU to enter VMX operation */
9843 	vmx_set_test_stage(8);
9844 	/* Wait for other CPU to enter VMX operation */
9845 	while (vmx_get_test_stage() != 9)
9846 		;
9847 
9848 	/* Send INIT signal to other CPU */
9849 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
9850 				   id_map[1]);
9851 	/* Signal other CPU we have sent INIT signal */
9852 	vmx_set_test_stage(10);
9853 
9854 	/*
9855 	 * Wait reasonable amount of time for other CPU
9856 	 * to exit VMX operation and process INIT signal
9857 	 */
9858 	delay(INIT_SIGNAL_TEST_DELAY);
9859 	report(!init_signal_test_thread_continued,
9860 	       "INIT signal processed after exit VMX operation");
9861 
9862 	/*
9863 	 * TODO: Send SIPI to other CPU to sipi_entry (See x86/cstart64.S)
9864 	 * to re-init it to kvm-unit-tests standard environment.
9865 	 * Somehow (?) verify that SIPI was indeed received.
9866 	 */
9867 }
9868 
9869 #define SIPI_SIGNAL_TEST_DELAY	100000000ULL
9870 
9871 static void vmx_sipi_test_guest(void)
9872 {
9873 	if (apic_id() == 0) {
9874 		/* wait AP enter guest with activity=WAIT_SIPI */
9875 		while (vmx_get_test_stage() != 1)
9876 			;
9877 		delay(SIPI_SIGNAL_TEST_DELAY);
9878 
9879 		/* First SIPI signal */
9880 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9881 		report_pass("BSP(L2): Send first SIPI to cpu[%d]", id_map[1]);
9882 
9883 		/* wait AP enter guest */
9884 		while (vmx_get_test_stage() != 2)
9885 			;
9886 		delay(SIPI_SIGNAL_TEST_DELAY);
9887 
9888 		/* Second SIPI signal should be ignored since AP is not in WAIT_SIPI state */
9889 		apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP | APIC_INT_ASSERT, id_map[1]);
9890 		report_pass("BSP(L2): Send second SIPI to cpu[%d]", id_map[1]);
9891 
9892 		/* Delay a while to check whether second SIPI would cause VMExit */
9893 		delay(SIPI_SIGNAL_TEST_DELAY);
9894 
9895 		/* Test is done, notify AP to exit test */
9896 		vmx_set_test_stage(3);
9897 
9898 		/* wait AP exit non-root mode */
9899 		while (vmx_get_test_stage() != 5)
9900 			;
9901 	} else {
9902 		/* wait BSP notify test is done */
9903 		while (vmx_get_test_stage() != 3)
9904 			;
9905 
9906 		/* AP exit guest */
9907 		vmx_set_test_stage(4);
9908 	}
9909 }
9910 
9911 static void sipi_test_ap_thread(void *data)
9912 {
9913 	struct vmcs *ap_vmcs;
9914 	u64 *ap_vmxon_region;
9915 	void *ap_stack, *ap_syscall_stack;
9916 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9917 	u64 cpu_ctrl_1 = 0;
9918 
9919 	/* Enter VMX operation (i.e. exec VMXON) */
9920 	ap_vmxon_region = alloc_page();
9921 	enable_vmx();
9922 	init_vmx(ap_vmxon_region);
9923 	_vmx_on(ap_vmxon_region);
9924 	init_vmcs(&ap_vmcs);
9925 	make_vmcs_current(ap_vmcs);
9926 
9927 	/* Set stack for AP */
9928 	ap_stack = alloc_page();
9929 	ap_syscall_stack = alloc_page();
9930 	vmcs_write(GUEST_RSP, (u64)(ap_stack + PAGE_SIZE - 1));
9931 	vmcs_write(GUEST_SYSENTER_ESP, (u64)(ap_syscall_stack + PAGE_SIZE - 1));
9932 
9933 	/* passthrough lapic to L2 */
9934 	disable_intercept_for_x2apic_msrs();
9935 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9936 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9937 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9938 
9939 	/* Set guest activity state to wait-for-SIPI state */
9940 	vmcs_write(GUEST_ACTV_STATE, ACTV_WAIT_SIPI);
9941 
9942 	vmx_set_test_stage(1);
9943 
9944 	/* AP enter guest */
9945 	enter_guest();
9946 
9947 	if (vmcs_read(EXI_REASON) == VMX_SIPI) {
9948 		report_pass("AP: Handle SIPI VMExit");
9949 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
9950 		vmx_set_test_stage(2);
9951 	} else {
9952 		report_fail("AP: Unexpected VMExit, reason=%ld", vmcs_read(EXI_REASON));
9953 		vmx_off();
9954 		return;
9955 	}
9956 
9957 	/* AP enter guest */
9958 	enter_guest();
9959 
9960 	report(vmcs_read(EXI_REASON) != VMX_SIPI,
9961 		"AP: should no SIPI VMExit since activity is not in WAIT_SIPI state");
9962 
9963 	/* notify BSP that AP is already exit from non-root mode */
9964 	vmx_set_test_stage(5);
9965 
9966 	/* Leave VMX operation */
9967 	vmx_off();
9968 }
9969 
9970 static void vmx_sipi_signal_test(void)
9971 {
9972 	if (!(rdmsr(MSR_IA32_VMX_MISC) & MSR_IA32_VMX_MISC_ACTIVITY_WAIT_SIPI)) {
9973 		printf("\tACTIVITY_WAIT_SIPI state is not supported.\n");
9974 		return;
9975 	}
9976 
9977 	if (cpu_count() < 2) {
9978 		report_skip(__func__);
9979 		return;
9980 	}
9981 
9982 	u64 cpu_ctrl_0 = CPU_SECONDARY;
9983 	u64 cpu_ctrl_1 = 0;
9984 
9985 	/* passthrough lapic to L2 */
9986 	disable_intercept_for_x2apic_msrs();
9987 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
9988 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
9989 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
9990 
9991 	test_set_guest(vmx_sipi_test_guest);
9992 
9993 	/* update CR3 on AP */
9994 	on_cpu(1, update_cr3, (void *)read_cr3());
9995 
9996 	/* start AP */
9997 	on_cpu_async(1, sipi_test_ap_thread, NULL);
9998 
9999 	vmx_set_test_stage(0);
10000 
10001 	/* BSP enter guest */
10002 	enter_guest();
10003 }
10004 
10005 
10006 enum vmcs_access {
10007 	ACCESS_VMREAD,
10008 	ACCESS_VMWRITE,
10009 	ACCESS_NONE,
10010 };
10011 
10012 struct vmcs_shadow_test_common {
10013 	enum vmcs_access op;
10014 	enum Reason reason;
10015 	u64 field;
10016 	u64 value;
10017 	u64 flags;
10018 	u64 time;
10019 } l1_l2_common;
10020 
10021 static inline u64 vmread_flags(u64 field, u64 *val)
10022 {
10023 	u64 flags;
10024 
10025 	asm volatile ("vmread %2, %1; pushf; pop %0"
10026 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
10027 	return flags & X86_EFLAGS_ALU;
10028 }
10029 
10030 static inline u64 vmwrite_flags(u64 field, u64 val)
10031 {
10032 	u64 flags;
10033 
10034 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
10035 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
10036 	return flags & X86_EFLAGS_ALU;
10037 }
10038 
10039 static void vmx_vmcs_shadow_test_guest(void)
10040 {
10041 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10042 	u64 start;
10043 
10044 	while (c->op != ACCESS_NONE) {
10045 		start = rdtsc();
10046 		switch (c->op) {
10047 		default:
10048 			c->flags = -1ull;
10049 			break;
10050 		case ACCESS_VMREAD:
10051 			c->flags = vmread_flags(c->field, &c->value);
10052 			break;
10053 		case ACCESS_VMWRITE:
10054 			c->flags = vmwrite_flags(c->field, 0);
10055 			break;
10056 		}
10057 		c->time = rdtsc() - start;
10058 		vmcall();
10059 	}
10060 }
10061 
10062 static u64 vmread_from_shadow(u64 field)
10063 {
10064 	struct vmcs *primary;
10065 	struct vmcs *shadow;
10066 	u64 value;
10067 
10068 	TEST_ASSERT(!vmcs_save(&primary));
10069 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10070 	TEST_ASSERT(!make_vmcs_current(shadow));
10071 	value = vmcs_read(field);
10072 	TEST_ASSERT(!make_vmcs_current(primary));
10073 	return value;
10074 }
10075 
10076 static u64 vmwrite_to_shadow(u64 field, u64 value)
10077 {
10078 	struct vmcs *primary;
10079 	struct vmcs *shadow;
10080 
10081 	TEST_ASSERT(!vmcs_save(&primary));
10082 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10083 	TEST_ASSERT(!make_vmcs_current(shadow));
10084 	vmcs_write(field, value);
10085 	value = vmcs_read(field);
10086 	TEST_ASSERT(!make_vmcs_current(primary));
10087 	return value;
10088 }
10089 
10090 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
10091 {
10092 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10093 
10094 	c->op = access;
10095 	vmcs_write(VMX_INST_ERROR, 0);
10096 	enter_guest();
10097 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
10098 	if (c->reason != VMX_VMCALL) {
10099 		skip_exit_insn();
10100 		enter_guest();
10101 	}
10102 	skip_exit_vmcall();
10103 }
10104 
10105 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
10106 {
10107 	struct vmcs_shadow_test_common *c = &l1_l2_common;
10108 	struct vmcs *shadow;
10109 	u64 value;
10110 	uintptr_t flags[2];
10111 	bool good_shadow;
10112 	u32 vmx_inst_error;
10113 
10114 	report_prefix_pushf("field %lx", field);
10115 	c->field = field;
10116 
10117 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
10118 	if (shadow != (struct vmcs *)-1ull) {
10119 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
10120 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
10121 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
10122 	} else {
10123 		/*
10124 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
10125 		 * shadowed-fields should fail with setting RFLAGS.CF.
10126 		 */
10127 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
10128 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
10129 		good_shadow = false;
10130 	}
10131 
10132 	/* Intercept both VMREAD and VMWRITE. */
10133 	report_prefix_push("no VMREAD/VMWRITE permission");
10134 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
10135 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
10136 		set_bit(field, bitmap[ACCESS_VMREAD]);
10137 		set_bit(field, bitmap[ACCESS_VMWRITE]);
10138 	}
10139 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10140 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10141 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10142 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10143 	report_prefix_pop();
10144 
10145 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
10146 		goto out;
10147 
10148 	/* Permit shadowed VMREAD. */
10149 	report_prefix_push("VMREAD permission only");
10150 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10151 	set_bit(field, bitmap[ACCESS_VMWRITE]);
10152 	if (good_shadow)
10153 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10154 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10155 	report(c->reason == VMX_VMWRITE, "not shadowed for VMWRITE");
10156 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10157 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10158 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10159 	       c->time);
10160 	report(c->flags == flags[ACCESS_VMREAD],
10161 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10162 	       c->flags, flags[ACCESS_VMREAD]);
10163 	if (good_shadow)
10164 		report(c->value == value,
10165 		       "value read from shadow (%lx) is as expected (%lx)",
10166 		       c->value, value);
10167 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10168 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10169 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10170 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10171 	report_prefix_pop();
10172 
10173 	/* Permit shadowed VMWRITE. */
10174 	report_prefix_push("VMWRITE permission only");
10175 	set_bit(field, bitmap[ACCESS_VMREAD]);
10176 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10177 	if (good_shadow)
10178 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10179 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10180 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10181 	report(c->reason == VMX_VMCALL,
10182 		"shadowed for VMWRITE (in %ld cycles)",
10183 		c->time);
10184 	report(c->flags == flags[ACCESS_VMREAD],
10185 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10186 	       c->flags, flags[ACCESS_VMREAD]);
10187 	if (good_shadow) {
10188 		value = vmread_from_shadow(field);
10189 		report(value == 0,
10190 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10191 		       0ul);
10192 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10193 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10194 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10195 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10196 	}
10197 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10198 	report(c->reason == VMX_VMREAD, "not shadowed for VMREAD");
10199 	report_prefix_pop();
10200 
10201 	/* Permit shadowed VMREAD and VMWRITE. */
10202 	report_prefix_push("VMREAD and VMWRITE permission");
10203 	clear_bit(field, bitmap[ACCESS_VMREAD]);
10204 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
10205 	if (good_shadow)
10206 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
10207 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
10208 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10209 	report(c->reason == VMX_VMCALL,
10210 		"shadowed for VMWRITE (in %ld cycles)",
10211 		c->time);
10212 	report(c->flags == flags[ACCESS_VMREAD],
10213 	       "ALU flags after VMWRITE (%lx) are as expected (%lx)",
10214 	       c->flags, flags[ACCESS_VMREAD]);
10215 	if (good_shadow) {
10216 		value = vmread_from_shadow(field);
10217 		report(value == 0,
10218 		       "shadow VMCS value (%lx) is as expected (%lx)", value,
10219 		       0ul);
10220 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
10221 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10222 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10223 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10224 	}
10225 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
10226 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
10227 	report(c->reason == VMX_VMCALL, "shadowed for VMREAD (in %ld cycles)",
10228 	       c->time);
10229 	report(c->flags == flags[ACCESS_VMREAD],
10230 	       "ALU flags after VMREAD (%lx) are as expected (%lx)",
10231 	       c->flags, flags[ACCESS_VMREAD]);
10232 	if (good_shadow)
10233 		report(c->value == 0,
10234 		       "value read from shadow (%lx) is as expected (%lx)",
10235 		       c->value, 0ul);
10236 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
10237 		report(vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
10238 		       "VMX_INST_ERROR (%d) is as expected (%d)",
10239 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
10240 	report_prefix_pop();
10241 
10242 out:
10243 	report_prefix_pop();
10244 }
10245 
10246 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
10247 {
10248 	unsigned base;
10249 	unsigned index;
10250 	unsigned bit;
10251 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
10252 
10253 	/* Run test on all possible valid VMCS fields */
10254 	for (base = 0;
10255 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
10256 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
10257 		for (index = 0; index <= highest_index; index++)
10258 			vmcs_shadow_test_field(bitmap, base + index);
10259 
10260 	/*
10261 	 * Run tests on some invalid VMCS fields
10262 	 * (Have reserved bit set).
10263 	 */
10264 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
10265 		vmcs_shadow_test_field(bitmap, (1ull << bit));
10266 }
10267 
10268 static void vmx_vmcs_shadow_test(void)
10269 {
10270 	u8 *bitmap[2];
10271 	struct vmcs *shadow;
10272 
10273 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
10274 		printf("\t'Activate secondary controls' not supported.\n");
10275 		return;
10276 	}
10277 
10278 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
10279 		printf("\t'VMCS shadowing' not supported.\n");
10280 		return;
10281 	}
10282 
10283 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
10284 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
10285 		printf("\tVMWRITE can't modify VM-exit information fields.\n");
10286 		return;
10287 	}
10288 
10289 	test_set_guest(vmx_vmcs_shadow_test_guest);
10290 
10291 	bitmap[ACCESS_VMREAD] = alloc_page();
10292 	bitmap[ACCESS_VMWRITE] = alloc_page();
10293 
10294 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
10295 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
10296 
10297 	shadow = alloc_page();
10298 	shadow->hdr.revision_id = basic.revision;
10299 	shadow->hdr.shadow_vmcs = 1;
10300 	TEST_ASSERT(!vmcs_clear(shadow));
10301 
10302 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
10303 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
10304 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
10305 
10306 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
10307 	report_prefix_push("valid link pointer");
10308 	vmx_vmcs_shadow_test_body(bitmap);
10309 	report_prefix_pop();
10310 
10311 	vmcs_write(VMCS_LINK_PTR, -1ull);
10312 	report_prefix_push("invalid link pointer");
10313 	vmx_vmcs_shadow_test_body(bitmap);
10314 	report_prefix_pop();
10315 
10316 	l1_l2_common.op = ACCESS_NONE;
10317 	enter_guest();
10318 }
10319 
10320 /*
10321  * This test monitors the difference between a guest RDTSC instruction
10322  * and the IA32_TIME_STAMP_COUNTER MSR value stored in the VMCS12
10323  * VM-exit MSR-store list when taking a VM-exit on the instruction
10324  * following RDTSC.
10325  */
10326 #define RDTSC_DIFF_ITERS 100000
10327 #define RDTSC_DIFF_FAILS 100
10328 #define HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD 750
10329 
10330 /*
10331  * Set 'use TSC offsetting' and set the guest offset to the
10332  * inverse of the host's current TSC value, so that the guest starts running
10333  * with an effective TSC value of 0.
10334  */
10335 static void reset_guest_tsc_to_zero(void)
10336 {
10337 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
10338 	vmcs_write(TSC_OFFSET, -rdtsc());
10339 }
10340 
10341 static void rdtsc_vmexit_diff_test_guest(void)
10342 {
10343 	int i;
10344 
10345 	for (i = 0; i < RDTSC_DIFF_ITERS; i++)
10346 		/* Ensure rdtsc is the last instruction before the vmcall. */
10347 		asm volatile("rdtsc; vmcall" : : : "eax", "edx");
10348 }
10349 
10350 /*
10351  * This function only considers the "use TSC offsetting" VM-execution
10352  * control.  It does not handle "use TSC scaling" (because the latter
10353  * isn't available to the host today.)
10354  */
10355 static unsigned long long host_time_to_guest_time(unsigned long long t)
10356 {
10357 	TEST_ASSERT(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
10358 		    !(vmcs_read(CPU_EXEC_CTRL1) & CPU_USE_TSC_SCALING));
10359 
10360 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_USE_TSC_OFFSET)
10361 		t += vmcs_read(TSC_OFFSET);
10362 
10363 	return t;
10364 }
10365 
10366 static unsigned long long rdtsc_vmexit_diff_test_iteration(void)
10367 {
10368 	unsigned long long guest_tsc, host_to_guest_tsc;
10369 
10370 	enter_guest();
10371 	skip_exit_vmcall();
10372 	guest_tsc = (u32) regs.rax + (regs.rdx << 32);
10373 	host_to_guest_tsc = host_time_to_guest_time(exit_msr_store[0].value);
10374 
10375 	return host_to_guest_tsc - guest_tsc;
10376 }
10377 
10378 static void rdtsc_vmexit_diff_test(void)
10379 {
10380 	unsigned long long delta;
10381 	int fail = 0;
10382 	int i;
10383 
10384 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET))
10385 		test_skip("CPU doesn't support the 'use TSC offsetting' processor-based VM-execution control.\n");
10386 
10387 	test_set_guest(rdtsc_vmexit_diff_test_guest);
10388 
10389 	reset_guest_tsc_to_zero();
10390 
10391 	/*
10392 	 * Set up the VMCS12 VM-exit MSR-store list to store just one
10393 	 * MSR: IA32_TIME_STAMP_COUNTER. Note that the value stored is
10394 	 * in the host time domain (i.e., it is not adjusted according
10395 	 * to the TSC multiplier and TSC offset fields in the VMCS12,
10396 	 * as a guest RDTSC would be.)
10397 	 */
10398 	exit_msr_store = alloc_page();
10399 	exit_msr_store[0].index = MSR_IA32_TSC;
10400 	vmcs_write(EXI_MSR_ST_CNT, 1);
10401 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(exit_msr_store));
10402 
10403 	for (i = 0; i < RDTSC_DIFF_ITERS && fail < RDTSC_DIFF_FAILS; i++) {
10404 		delta = rdtsc_vmexit_diff_test_iteration();
10405 		if (delta >= HOST_CAPTURED_GUEST_TSC_DIFF_THRESHOLD)
10406 			fail++;
10407 	}
10408 
10409 	enter_guest();
10410 
10411 	report(fail < RDTSC_DIFF_FAILS,
10412 	       "RDTSC to VM-exit delta too high in %d of %d iterations, last = %llu",
10413 	       fail, i, delta);
10414 }
10415 
10416 static int invalid_msr_init(struct vmcs *vmcs)
10417 {
10418 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
10419 		printf("\tPreemption timer is not supported\n");
10420 		return VMX_TEST_EXIT;
10421 	}
10422 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
10423 	preempt_val = 10000000;
10424 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
10425 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
10426 
10427 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
10428 		printf("\tSave preemption value is not supported\n");
10429 
10430 	vmcs_write(ENT_MSR_LD_CNT, 1);
10431 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
10432 
10433 	return VMX_TEST_START;
10434 }
10435 
10436 
10437 static void invalid_msr_main(void)
10438 {
10439 	report_fail("Invalid MSR load");
10440 }
10441 
10442 static int invalid_msr_exit_handler(union exit_reason exit_reason)
10443 {
10444 	report_fail("Invalid MSR load");
10445 	print_vmexit_info(exit_reason);
10446 	return VMX_TEST_EXIT;
10447 }
10448 
10449 static int invalid_msr_entry_failure(struct vmentry_result *result)
10450 {
10451 	report(result->exit_reason.failed_vmentry &&
10452 	       result->exit_reason.basic == VMX_FAIL_MSR, "Invalid MSR load");
10453 	return VMX_TEST_VMEXIT;
10454 }
10455 
10456 /*
10457  * The max number of MSRs in an atomic switch MSR list is:
10458  * (111B + 1) * 512 = 4096
10459  *
10460  * Each list entry consumes:
10461  * 4-byte MSR index + 4 bytes reserved + 8-byte data = 16 bytes
10462  *
10463  * Allocate 128 kB to cover max_msr_list_size (i.e., 64 kB) and then some.
10464  */
10465 static const u32 msr_list_page_order = 5;
10466 
10467 static void atomic_switch_msr_limit_test_guest(void)
10468 {
10469 	vmcall();
10470 }
10471 
10472 static void populate_msr_list(struct vmx_msr_entry *msr_list,
10473 			      size_t byte_capacity, int count)
10474 {
10475 	int i;
10476 
10477 	for (i = 0; i < count; i++) {
10478 		msr_list[i].index = MSR_IA32_TSC;
10479 		msr_list[i].reserved = 0;
10480 		msr_list[i].value = 0x1234567890abcdef;
10481 	}
10482 
10483 	memset(msr_list + count, 0xff,
10484 	       byte_capacity - count * sizeof(*msr_list));
10485 }
10486 
10487 static int max_msr_list_size(void)
10488 {
10489 	u32 vmx_misc = rdmsr(MSR_IA32_VMX_MISC);
10490 	u32 factor = ((vmx_misc & GENMASK(27, 25)) >> 25) + 1;
10491 
10492 	return factor * 512;
10493 }
10494 
10495 static void atomic_switch_msrs_test(int count)
10496 {
10497 	struct vmx_msr_entry *vm_enter_load;
10498         struct vmx_msr_entry *vm_exit_load;
10499         struct vmx_msr_entry *vm_exit_store;
10500 	int max_allowed = max_msr_list_size();
10501 	int byte_capacity = 1ul << (msr_list_page_order + PAGE_SHIFT);
10502 	/* Exceeding the max MSR list size at exit triggers KVM to abort. */
10503 	int exit_count = count > max_allowed ? max_allowed : count;
10504 	int cleanup_count = count > max_allowed ? 2 : 1;
10505 	int i;
10506 
10507 	/*
10508 	 * Check for the IA32_TSC MSR,
10509 	 * available with the "TSC flag" and used to populate the MSR lists.
10510 	 */
10511 	if (!(cpuid(1).d & (1 << 4))) {
10512 		report_skip(__func__);
10513 		return;
10514 	}
10515 
10516 	/* Set L2 guest. */
10517 	test_set_guest(atomic_switch_msr_limit_test_guest);
10518 
10519 	/* Setup atomic MSR switch lists. */
10520 	vm_enter_load = alloc_pages(msr_list_page_order);
10521 	vm_exit_load = alloc_pages(msr_list_page_order);
10522 	vm_exit_store = alloc_pages(msr_list_page_order);
10523 
10524 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)vm_enter_load);
10525 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)vm_exit_load);
10526 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)vm_exit_store);
10527 
10528 	/*
10529 	 * VM-Enter should succeed up to the max number of MSRs per list, and
10530 	 * should not consume junk beyond the last entry.
10531 	 */
10532 	populate_msr_list(vm_enter_load, byte_capacity, count);
10533 	populate_msr_list(vm_exit_load, byte_capacity, exit_count);
10534 	populate_msr_list(vm_exit_store, byte_capacity, exit_count);
10535 
10536 	vmcs_write(ENT_MSR_LD_CNT, count);
10537 	vmcs_write(EXI_MSR_LD_CNT, exit_count);
10538 	vmcs_write(EXI_MSR_ST_CNT, exit_count);
10539 
10540 	if (count <= max_allowed) {
10541 		enter_guest();
10542 		assert_exit_reason(VMX_VMCALL);
10543 		skip_exit_vmcall();
10544 	} else {
10545 		u32 exit_qual;
10546 
10547 		test_guest_state("Invalid MSR Load Count", true, count,
10548 				 "ENT_MSR_LD_CNT");
10549 
10550 		exit_qual = vmcs_read(EXI_QUALIFICATION);
10551 		report(exit_qual == max_allowed + 1, "exit_qual, %u, is %u.",
10552 		       exit_qual, max_allowed + 1);
10553 	}
10554 
10555 	/* Cleanup. */
10556 	vmcs_write(ENT_MSR_LD_CNT, 0);
10557 	vmcs_write(EXI_MSR_LD_CNT, 0);
10558 	vmcs_write(EXI_MSR_ST_CNT, 0);
10559 	for (i = 0; i < cleanup_count; i++) {
10560 		enter_guest();
10561 		skip_exit_vmcall();
10562 	}
10563 	free_pages_by_order(vm_enter_load, msr_list_page_order);
10564 	free_pages_by_order(vm_exit_load, msr_list_page_order);
10565 	free_pages_by_order(vm_exit_store, msr_list_page_order);
10566 }
10567 
10568 static void atomic_switch_max_msrs_test(void)
10569 {
10570 	atomic_switch_msrs_test(max_msr_list_size());
10571 }
10572 
10573 static void atomic_switch_overflow_msrs_test(void)
10574 {
10575 	if (test_device_enabled())
10576 		atomic_switch_msrs_test(max_msr_list_size() + 1);
10577 	else
10578 		test_skip("Test is only supported on KVM");
10579 }
10580 
10581 static void vmx_pf_exception_test_guest(void)
10582 {
10583 	ac_test_run(PT_LEVEL_PML4);
10584 }
10585 
10586 typedef void (*invalidate_tlb_t)(void *data);
10587 
10588 static void __vmx_pf_exception_test(invalidate_tlb_t inv_fn, void *data)
10589 {
10590 	u64 efer;
10591 	struct cpuid cpuid;
10592 
10593 	test_set_guest(vmx_pf_exception_test_guest);
10594 
10595 	/* Intercept INVLPG when to perform TLB invalidation from L1 (this). */
10596 	if (inv_fn)
10597 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INVLPG);
10598 	else
10599 		vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INVLPG);
10600 
10601 	enter_guest();
10602 
10603 	while (vmcs_read(EXI_REASON) != VMX_VMCALL) {
10604 		switch (vmcs_read(EXI_REASON)) {
10605 		case VMX_RDMSR:
10606 			assert(regs.rcx == MSR_EFER);
10607 			efer = vmcs_read(GUEST_EFER);
10608 			regs.rdx = efer >> 32;
10609 			regs.rax = efer & 0xffffffff;
10610 			break;
10611 		case VMX_WRMSR:
10612 			assert(regs.rcx == MSR_EFER);
10613 			efer = regs.rdx << 32 | (regs.rax & 0xffffffff);
10614 			vmcs_write(GUEST_EFER, efer);
10615 			break;
10616 		case VMX_CPUID:
10617 			cpuid = (struct cpuid) {0, 0, 0, 0};
10618 			cpuid = raw_cpuid(regs.rax, regs.rcx);
10619 			regs.rax = cpuid.a;
10620 			regs.rbx = cpuid.b;
10621 			regs.rcx = cpuid.c;
10622 			regs.rdx = cpuid.d;
10623 			break;
10624 		case VMX_INVLPG:
10625 			inv_fn(data);
10626 			break;
10627 		default:
10628 			assert_msg(false,
10629 				"Unexpected exit to L1, exit_reason: %s (0x%lx)",
10630 				exit_reason_description(vmcs_read(EXI_REASON)),
10631 				vmcs_read(EXI_REASON));
10632 		}
10633 		skip_exit_insn();
10634 		enter_guest();
10635 	}
10636 
10637 	assert_exit_reason(VMX_VMCALL);
10638 }
10639 
10640 static void vmx_pf_exception_test(void)
10641 {
10642 	__vmx_pf_exception_test(NULL, NULL);
10643 }
10644 
10645 static void invalidate_tlb_no_vpid(void *data)
10646 {
10647 	/* If VPID is disabled, the TLB is flushed on VM-Enter and VM-Exit. */
10648 }
10649 
10650 static void vmx_pf_no_vpid_test(void)
10651 {
10652 	if (is_vpid_supported())
10653 		vmcs_clear_bits(CPU_EXEC_CTRL1, CPU_VPID);
10654 
10655 	__vmx_pf_exception_test(invalidate_tlb_no_vpid, NULL);
10656 }
10657 
10658 static void invalidate_tlb_invvpid_addr(void *data)
10659 {
10660 	invvpid(INVVPID_ALL, *(u16 *)data, vmcs_read(EXI_QUALIFICATION));
10661 }
10662 
10663 static void invalidate_tlb_new_vpid(void *data)
10664 {
10665 	u16 *vpid = data;
10666 
10667 	/*
10668 	 * Bump VPID to effectively flush L2's TLB from L0's perspective.
10669 	 * Invalidate all VPIDs when the VPID wraps to zero as hardware/KVM is
10670 	 * architecturally allowed to keep TLB entries indefinitely.
10671 	 */
10672 	++(*vpid);
10673 	if (*vpid == 0) {
10674 		++(*vpid);
10675 		invvpid(INVVPID_ALL, 0, 0);
10676 	}
10677 	vmcs_write(VPID, *vpid);
10678 }
10679 
10680 static void __vmx_pf_vpid_test(invalidate_tlb_t inv_fn, u16 vpid)
10681 {
10682 	if (!is_vpid_supported())
10683 		test_skip("VPID unsupported");
10684 
10685 	if (!is_invvpid_supported())
10686 		test_skip("INVVPID unsupported");
10687 
10688 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
10689 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VPID);
10690 	vmcs_write(VPID, vpid);
10691 
10692 	__vmx_pf_exception_test(inv_fn, &vpid);
10693 }
10694 
10695 static void vmx_pf_invvpid_test(void)
10696 {
10697 	if (!is_invvpid_type_supported(INVVPID_ADDR))
10698 		test_skip("INVVPID ADDR unsupported");
10699 
10700 	__vmx_pf_vpid_test(invalidate_tlb_invvpid_addr, 0xaaaa);
10701 }
10702 
10703 static void vmx_pf_vpid_test(void)
10704 {
10705 	/* Need INVVPID(ALL) to flush VPIDs upon wrap/reuse. */
10706 	if (!is_invvpid_type_supported(INVVPID_ALL))
10707 		test_skip("INVVPID ALL unsupported");
10708 
10709 	__vmx_pf_vpid_test(invalidate_tlb_new_vpid, 1);
10710 }
10711 
10712 static void vmx_l2_gp_test(void)
10713 {
10714 	*(volatile u64 *)NONCANONICAL = 0;
10715 }
10716 
10717 static void vmx_l2_ud_test(void)
10718 {
10719 	asm volatile ("ud2");
10720 }
10721 
10722 static void vmx_l2_de_test(void)
10723 {
10724 	asm volatile (
10725 		"xor %%eax, %%eax\n\t"
10726 		"xor %%ebx, %%ebx\n\t"
10727 		"xor %%edx, %%edx\n\t"
10728 		"idiv %%ebx\n\t"
10729 		::: "eax", "ebx", "edx");
10730 }
10731 
10732 static void vmx_l2_bp_test(void)
10733 {
10734 	asm volatile ("int3");
10735 }
10736 
10737 static void vmx_l2_db_test(void)
10738 {
10739 	write_rflags(read_rflags() | X86_EFLAGS_TF);
10740 }
10741 
10742 static uint64_t usermode_callback(void)
10743 {
10744 	/*
10745 	 * Trigger an #AC by writing 8 bytes to a 4-byte aligned address.
10746 	 * Disclaimer: It is assumed that the stack pointer is aligned
10747 	 * on a 16-byte boundary as x86_64 stacks should be.
10748 	 */
10749 	asm volatile("movq $0, -0x4(%rsp)");
10750 
10751 	return 0;
10752 }
10753 
10754 static void vmx_l2_ac_test(void)
10755 {
10756 	bool hit_ac = false;
10757 
10758 	write_cr0(read_cr0() | X86_CR0_AM);
10759 	write_rflags(read_rflags() | X86_EFLAGS_AC);
10760 
10761 	run_in_user(usermode_callback, AC_VECTOR, 0, 0, 0, 0, &hit_ac);
10762 	report(hit_ac, "Usermode #AC handled in L2");
10763 	vmcall();
10764 }
10765 
10766 struct vmx_exception_test {
10767 	u8 vector;
10768 	void (*guest_code)(void);
10769 };
10770 
10771 struct vmx_exception_test vmx_exception_tests[] = {
10772 	{ GP_VECTOR, vmx_l2_gp_test },
10773 	{ UD_VECTOR, vmx_l2_ud_test },
10774 	{ DE_VECTOR, vmx_l2_de_test },
10775 	{ DB_VECTOR, vmx_l2_db_test },
10776 	{ BP_VECTOR, vmx_l2_bp_test },
10777 	{ AC_VECTOR, vmx_l2_ac_test },
10778 };
10779 
10780 static u8 vmx_exception_test_vector;
10781 
10782 static void vmx_exception_handler(struct ex_regs *regs)
10783 {
10784 	report(regs->vector == vmx_exception_test_vector,
10785 	       "Handling %s in L2's exception handler",
10786 	       exception_mnemonic(vmx_exception_test_vector));
10787 	vmcall();
10788 }
10789 
10790 static void handle_exception_in_l2(u8 vector)
10791 {
10792 	handler old_handler = handle_exception(vector, vmx_exception_handler);
10793 
10794 	vmx_exception_test_vector = vector;
10795 
10796 	enter_guest();
10797 	report(vmcs_read(EXI_REASON) == VMX_VMCALL,
10798 	       "%s handled by L2", exception_mnemonic(vector));
10799 
10800 	handle_exception(vector, old_handler);
10801 }
10802 
10803 static void handle_exception_in_l1(u32 vector)
10804 {
10805 	u32 old_eb = vmcs_read(EXC_BITMAP);
10806 
10807 	vmcs_write(EXC_BITMAP, old_eb | (1u << vector));
10808 
10809 	enter_guest();
10810 
10811 	report((vmcs_read(EXI_REASON) == VMX_EXC_NMI) &&
10812 	       ((vmcs_read(EXI_INTR_INFO) & 0xff) == vector),
10813 	       "%s handled by L1", exception_mnemonic(vector));
10814 
10815 	vmcs_write(EXC_BITMAP, old_eb);
10816 }
10817 
10818 static void vmx_exception_test(void)
10819 {
10820 	struct vmx_exception_test *t;
10821 	int i;
10822 
10823 	for (i = 0; i < ARRAY_SIZE(vmx_exception_tests); i++) {
10824 		t = &vmx_exception_tests[i];
10825 
10826 		/*
10827 		 * Override the guest code before each run even though it's the
10828 		 * same code, the VMCS guest state needs to be reinitialized.
10829 		 */
10830 		test_override_guest(t->guest_code);
10831 		handle_exception_in_l2(t->vector);
10832 
10833 		test_override_guest(t->guest_code);
10834 		handle_exception_in_l1(t->vector);
10835 	}
10836 
10837 	test_set_guest_finished();
10838 }
10839 
10840 #define TEST(name) { #name, .v2 = name }
10841 
10842 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
10843 struct vmx_test vmx_tests[] = {
10844 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
10845 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
10846 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
10847 		preemption_timer_exit_handler, NULL, {0} },
10848 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
10849 		test_ctrl_pat_exit_handler, NULL, {0} },
10850 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
10851 		test_ctrl_efer_exit_handler, NULL, {0} },
10852 	{ "CR shadowing", NULL, cr_shadowing_main,
10853 		cr_shadowing_exit_handler, NULL, {0} },
10854 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
10855 		NULL, {0} },
10856 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
10857 		insn_intercept_exit_handler, NULL, {0} },
10858 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
10859 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
10860 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
10861 	{ "interrupt", interrupt_init, interrupt_main,
10862 		interrupt_exit_handler, NULL, {0} },
10863 	{ "nmi_hlt", nmi_hlt_init, nmi_hlt_main,
10864 		nmi_hlt_exit_handler, NULL, {0} },
10865 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
10866 		NULL, {0} },
10867 	{ "MSR switch", msr_switch_init, msr_switch_main,
10868 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
10869 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
10870 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
10871 		disable_rdtscp_exit_handler, NULL, {0} },
10872 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
10873 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
10874 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
10875 		exit_monitor_from_l2_handler, NULL, {0} },
10876 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
10877 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
10878 	/* Basic V2 tests. */
10879 	TEST(v2_null_test),
10880 	TEST(v2_multiple_entries_test),
10881 	TEST(fixture_test_case1),
10882 	TEST(fixture_test_case2),
10883 	/* Opcode tests. */
10884 	TEST(invvpid_test),
10885 	/* VM-entry tests */
10886 	TEST(vmx_controls_test),
10887 	TEST(vmx_host_state_area_test),
10888 	TEST(vmx_guest_state_area_test),
10889 	TEST(vmentry_movss_shadow_test),
10890 	TEST(vmentry_unrestricted_guest_test),
10891 	/* APICv tests */
10892 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
10893 	TEST(vmx_hlt_with_rvi_test),
10894 	TEST(apic_reg_virt_test),
10895 	TEST(virt_x2apic_mode_test),
10896 	/* APIC pass-through tests */
10897 	TEST(vmx_apic_passthrough_test),
10898 	TEST(vmx_apic_passthrough_thread_test),
10899 	TEST(vmx_apic_passthrough_tpr_threshold_test),
10900 	TEST(vmx_init_signal_test),
10901 	TEST(vmx_sipi_signal_test),
10902 	/* VMCS Shadowing tests */
10903 	TEST(vmx_vmcs_shadow_test),
10904 	/* Regression tests */
10905 	TEST(vmx_ldtr_test),
10906 	TEST(vmx_cr_load_test),
10907 	TEST(vmx_cr4_osxsave_test),
10908 	TEST(vmx_nm_test),
10909 	TEST(vmx_db_test),
10910 	TEST(vmx_nmi_window_test),
10911 	TEST(vmx_intr_window_test),
10912 	TEST(vmx_pending_event_test),
10913 	TEST(vmx_pending_event_hlt_test),
10914 	TEST(vmx_store_tsc_test),
10915 	TEST(vmx_preemption_timer_zero_test),
10916 	TEST(vmx_preemption_timer_tf_test),
10917 	TEST(vmx_preemption_timer_expiry_test),
10918 	/* EPT access tests. */
10919 	TEST(ept_access_test_not_present),
10920 	TEST(ept_access_test_read_only),
10921 	TEST(ept_access_test_write_only),
10922 	TEST(ept_access_test_read_write),
10923 	TEST(ept_access_test_execute_only),
10924 	TEST(ept_access_test_read_execute),
10925 	TEST(ept_access_test_write_execute),
10926 	TEST(ept_access_test_read_write_execute),
10927 	TEST(ept_access_test_reserved_bits),
10928 	TEST(ept_access_test_ignored_bits),
10929 	TEST(ept_access_test_paddr_not_present_ad_disabled),
10930 	TEST(ept_access_test_paddr_not_present_ad_enabled),
10931 	TEST(ept_access_test_paddr_read_only_ad_disabled),
10932 	TEST(ept_access_test_paddr_read_only_ad_enabled),
10933 	TEST(ept_access_test_paddr_read_write),
10934 	TEST(ept_access_test_paddr_read_write_execute),
10935 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
10936 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
10937 	TEST(ept_access_test_paddr_not_present_page_fault),
10938 	TEST(ept_access_test_force_2m_page),
10939 	/* Atomic MSR switch tests. */
10940 	TEST(atomic_switch_max_msrs_test),
10941 	TEST(atomic_switch_overflow_msrs_test),
10942 	TEST(rdtsc_vmexit_diff_test),
10943 	TEST(vmx_mtf_test),
10944 	TEST(vmx_mtf_pdpte_test),
10945 	TEST(vmx_pf_exception_test),
10946 	TEST(vmx_pf_no_vpid_test),
10947 	TEST(vmx_pf_invvpid_test),
10948 	TEST(vmx_pf_vpid_test),
10949 	TEST(vmx_exception_test),
10950 	{ NULL, NULL, NULL, NULL, NULL, {0} },
10951 };
10952