xref: /kvm-unit-tests/x86/vmx_tests.c (revision badc98cafea47463b8151eefbc6d4954f6aec6a9)
1 /*
2  * All test cases of nested virtualization should be in this file
3  *
4  * Author : Arthur Chunqi Li <yzt356@gmail.com>
5  */
6 
7 #include <asm/debugreg.h>
8 
9 #include "vmx.h"
10 #include "msr.h"
11 #include "processor.h"
12 #include "vm.h"
13 #include "pci.h"
14 #include "fwcfg.h"
15 #include "isr.h"
16 #include "desc.h"
17 #include "apic.h"
18 #include "types.h"
19 #include "vmalloc.h"
20 #include "alloc_page.h"
21 #include "smp.h"
22 #include "delay.h"
23 
24 #define NONCANONICAL            0xaaaaaaaaaaaaaaaaull
25 
26 #define VPID_CAP_INVVPID_TYPES_SHIFT 40
27 
28 u64 ia32_pat;
29 u64 ia32_efer;
30 void *io_bitmap_a, *io_bitmap_b;
31 u16 ioport;
32 
33 unsigned long *pml4;
34 u64 eptp;
35 void *data_page1, *data_page2;
36 
37 phys_addr_t pci_physaddr;
38 
39 void *pml_log;
40 #define PML_INDEX 512
41 
42 static inline unsigned ffs(unsigned x)
43 {
44 	int pos = -1;
45 
46 	__asm__ __volatile__("bsf %1, %%eax; cmovnz %%eax, %0"
47 			     : "+r"(pos) : "rm"(x) : "eax");
48 	return pos + 1;
49 }
50 
51 static inline void vmcall(void)
52 {
53 	asm volatile("vmcall");
54 }
55 
56 static void basic_guest_main(void)
57 {
58 	report("Basic VMX test", 1);
59 }
60 
61 static int basic_exit_handler(void)
62 {
63 	report("Basic VMX test", 0);
64 	print_vmexit_info();
65 	return VMX_TEST_EXIT;
66 }
67 
68 static void vmenter_main(void)
69 {
70 	u64 rax;
71 	u64 rsp, resume_rsp;
72 
73 	report("test vmlaunch", 1);
74 
75 	asm volatile(
76 		"mov %%rsp, %0\n\t"
77 		"mov %3, %%rax\n\t"
78 		"vmcall\n\t"
79 		"mov %%rax, %1\n\t"
80 		"mov %%rsp, %2\n\t"
81 		: "=r"(rsp), "=r"(rax), "=r"(resume_rsp)
82 		: "g"(0xABCD));
83 	report("test vmresume", (rax == 0xFFFF) && (rsp == resume_rsp));
84 }
85 
86 static int vmenter_exit_handler(void)
87 {
88 	u64 guest_rip;
89 	ulong reason;
90 
91 	guest_rip = vmcs_read(GUEST_RIP);
92 	reason = vmcs_read(EXI_REASON) & 0xff;
93 	switch (reason) {
94 	case VMX_VMCALL:
95 		if (regs.rax != 0xABCD) {
96 			report("test vmresume", 0);
97 			return VMX_TEST_VMEXIT;
98 		}
99 		regs.rax = 0xFFFF;
100 		vmcs_write(GUEST_RIP, guest_rip + 3);
101 		return VMX_TEST_RESUME;
102 	default:
103 		report("test vmresume", 0);
104 		print_vmexit_info();
105 	}
106 	return VMX_TEST_VMEXIT;
107 }
108 
109 u32 preempt_scale;
110 volatile unsigned long long tsc_val;
111 volatile u32 preempt_val;
112 u64 saved_rip;
113 
114 static int preemption_timer_init(struct vmcs *vmcs)
115 {
116 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
117 		printf("\tPreemption timer is not supported\n");
118 		return VMX_TEST_EXIT;
119 	}
120 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
121 	preempt_val = 10000000;
122 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
123 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
124 
125 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
126 		printf("\tSave preemption value is not supported\n");
127 
128 	return VMX_TEST_START;
129 }
130 
131 static void preemption_timer_main(void)
132 {
133 	tsc_val = rdtsc();
134 	if (ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) {
135 		vmx_set_test_stage(0);
136 		vmcall();
137 		if (vmx_get_test_stage() == 1)
138 			vmcall();
139 	}
140 	vmx_set_test_stage(1);
141 	while (vmx_get_test_stage() == 1) {
142 		if (((rdtsc() - tsc_val) >> preempt_scale)
143 				> 10 * preempt_val) {
144 			vmx_set_test_stage(2);
145 			vmcall();
146 		}
147 	}
148 	tsc_val = rdtsc();
149 	asm volatile ("hlt");
150 	vmcall();
151 	vmx_set_test_stage(5);
152 	vmcall();
153 }
154 
155 static int preemption_timer_exit_handler(void)
156 {
157 	bool guest_halted;
158 	u64 guest_rip;
159 	ulong reason;
160 	u32 insn_len;
161 	u32 ctrl_exit;
162 
163 	guest_rip = vmcs_read(GUEST_RIP);
164 	reason = vmcs_read(EXI_REASON) & 0xff;
165 	insn_len = vmcs_read(EXI_INST_LEN);
166 	switch (reason) {
167 	case VMX_PREEMPT:
168 		switch (vmx_get_test_stage()) {
169 		case 1:
170 		case 2:
171 			report("busy-wait for preemption timer",
172 			       ((rdtsc() - tsc_val) >> preempt_scale) >=
173 			       preempt_val);
174 			vmx_set_test_stage(3);
175 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
176 			return VMX_TEST_RESUME;
177 		case 3:
178 			guest_halted =
179 				(vmcs_read(GUEST_ACTV_STATE) == ACTV_HLT);
180 			report("preemption timer during hlt",
181 			       ((rdtsc() - tsc_val) >> preempt_scale) >=
182 			       preempt_val && guest_halted);
183 			vmx_set_test_stage(4);
184 			vmcs_write(PIN_CONTROLS,
185 				   vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
186 			vmcs_write(EXI_CONTROLS,
187 				   vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_PREEMPT);
188 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
189 			return VMX_TEST_RESUME;
190 		case 4:
191 			report("preemption timer with 0 value",
192 			       saved_rip == guest_rip);
193 			break;
194 		default:
195 			report("Invalid stage.", false);
196 			print_vmexit_info();
197 			break;
198 		}
199 		break;
200 	case VMX_VMCALL:
201 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
202 		switch (vmx_get_test_stage()) {
203 		case 0:
204 			report("Keep preemption value",
205 			       vmcs_read(PREEMPT_TIMER_VALUE) == preempt_val);
206 			vmx_set_test_stage(1);
207 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
208 			ctrl_exit = (vmcs_read(EXI_CONTROLS) |
209 				EXI_SAVE_PREEMPT) & ctrl_exit_rev.clr;
210 			vmcs_write(EXI_CONTROLS, ctrl_exit);
211 			return VMX_TEST_RESUME;
212 		case 1:
213 			report("Save preemption value",
214 			       vmcs_read(PREEMPT_TIMER_VALUE) < preempt_val);
215 			return VMX_TEST_RESUME;
216 		case 2:
217 			report("busy-wait for preemption timer", 0);
218 			vmx_set_test_stage(3);
219 			vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
220 			return VMX_TEST_RESUME;
221 		case 3:
222 			report("preemption timer during hlt", 0);
223 			vmx_set_test_stage(4);
224 			/* fall through */
225 		case 4:
226 			vmcs_write(PIN_CONTROLS,
227 				   vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
228 			vmcs_write(PREEMPT_TIMER_VALUE, 0);
229 			saved_rip = guest_rip + insn_len;
230 			return VMX_TEST_RESUME;
231 		case 5:
232 			report("preemption timer with 0 value (vmcall stage 5)", 0);
233 			break;
234 		default:
235 			// Should not reach here
236 			report("unexpected stage, %d", false,
237 			       vmx_get_test_stage());
238 			print_vmexit_info();
239 			return VMX_TEST_VMEXIT;
240 		}
241 		break;
242 	default:
243 		report("Unknown exit reason, %ld", false, reason);
244 		print_vmexit_info();
245 	}
246 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_PREEMPT);
247 	return VMX_TEST_VMEXIT;
248 }
249 
250 static void msr_bmp_init(void)
251 {
252 	void *msr_bitmap;
253 	u32 ctrl_cpu0;
254 
255 	msr_bitmap = alloc_page();
256 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
257 	ctrl_cpu0 |= CPU_MSR_BITMAP;
258 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
259 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
260 }
261 
262 static void *get_msr_bitmap(void)
263 {
264 	void *msr_bitmap;
265 
266 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
267 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
268 	} else {
269 		msr_bitmap = alloc_page();
270 		memset(msr_bitmap, 0xff, PAGE_SIZE);
271 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
272 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
273 	}
274 
275 	return msr_bitmap;
276 }
277 
278 static void disable_intercept_for_x2apic_msrs(void)
279 {
280 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
281 	u32 msr;
282 
283 	for (msr = APIC_BASE_MSR;
284 		 msr < (APIC_BASE_MSR+0xff);
285 		 msr += BITS_PER_LONG) {
286 		unsigned int word = msr / BITS_PER_LONG;
287 
288 		msr_bitmap[word] = 0;
289 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
290 	}
291 }
292 
293 static int test_ctrl_pat_init(struct vmcs *vmcs)
294 {
295 	u64 ctrl_ent;
296 	u64 ctrl_exi;
297 
298 	msr_bmp_init();
299 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
300 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
301 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
302 		printf("\tSave/load PAT is not supported\n");
303 		return 1;
304 	}
305 
306 	ctrl_ent = vmcs_read(ENT_CONTROLS);
307 	ctrl_exi = vmcs_read(EXI_CONTROLS);
308 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
309 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
310 	vmcs_write(ENT_CONTROLS, ctrl_ent);
311 	vmcs_write(EXI_CONTROLS, ctrl_exi);
312 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
313 	vmcs_write(GUEST_PAT, 0x0);
314 	vmcs_write(HOST_PAT, ia32_pat);
315 	return VMX_TEST_START;
316 }
317 
318 static void test_ctrl_pat_main(void)
319 {
320 	u64 guest_ia32_pat;
321 
322 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
323 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
324 		printf("\tENT_LOAD_PAT is not supported.\n");
325 	else {
326 		if (guest_ia32_pat != 0) {
327 			report("Entry load PAT", 0);
328 			return;
329 		}
330 	}
331 	wrmsr(MSR_IA32_CR_PAT, 0x6);
332 	vmcall();
333 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
334 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
335 		report("Entry load PAT", guest_ia32_pat == ia32_pat);
336 }
337 
338 static int test_ctrl_pat_exit_handler(void)
339 {
340 	u64 guest_rip;
341 	ulong reason;
342 	u64 guest_pat;
343 
344 	guest_rip = vmcs_read(GUEST_RIP);
345 	reason = vmcs_read(EXI_REASON) & 0xff;
346 	switch (reason) {
347 	case VMX_VMCALL:
348 		guest_pat = vmcs_read(GUEST_PAT);
349 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
350 			printf("\tEXI_SAVE_PAT is not supported\n");
351 			vmcs_write(GUEST_PAT, 0x6);
352 		} else {
353 			report("Exit save PAT", guest_pat == 0x6);
354 		}
355 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
356 			printf("\tEXI_LOAD_PAT is not supported\n");
357 		else
358 			report("Exit load PAT", rdmsr(MSR_IA32_CR_PAT) == ia32_pat);
359 		vmcs_write(GUEST_PAT, ia32_pat);
360 		vmcs_write(GUEST_RIP, guest_rip + 3);
361 		return VMX_TEST_RESUME;
362 	default:
363 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
364 		break;
365 	}
366 	return VMX_TEST_VMEXIT;
367 }
368 
369 static int test_ctrl_efer_init(struct vmcs *vmcs)
370 {
371 	u64 ctrl_ent;
372 	u64 ctrl_exi;
373 
374 	msr_bmp_init();
375 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
376 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
377 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
378 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
379 	ia32_efer = rdmsr(MSR_EFER);
380 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
381 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
382 	return VMX_TEST_START;
383 }
384 
385 static void test_ctrl_efer_main(void)
386 {
387 	u64 guest_ia32_efer;
388 
389 	guest_ia32_efer = rdmsr(MSR_EFER);
390 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
391 		printf("\tENT_LOAD_EFER is not supported.\n");
392 	else {
393 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
394 			report("Entry load EFER", 0);
395 			return;
396 		}
397 	}
398 	wrmsr(MSR_EFER, ia32_efer);
399 	vmcall();
400 	guest_ia32_efer = rdmsr(MSR_EFER);
401 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
402 		report("Entry load EFER", guest_ia32_efer == ia32_efer);
403 }
404 
405 static int test_ctrl_efer_exit_handler(void)
406 {
407 	u64 guest_rip;
408 	ulong reason;
409 	u64 guest_efer;
410 
411 	guest_rip = vmcs_read(GUEST_RIP);
412 	reason = vmcs_read(EXI_REASON) & 0xff;
413 	switch (reason) {
414 	case VMX_VMCALL:
415 		guest_efer = vmcs_read(GUEST_EFER);
416 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
417 			printf("\tEXI_SAVE_EFER is not supported\n");
418 			vmcs_write(GUEST_EFER, ia32_efer);
419 		} else {
420 			report("Exit save EFER", guest_efer == ia32_efer);
421 		}
422 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
423 			printf("\tEXI_LOAD_EFER is not supported\n");
424 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
425 		} else {
426 			report("Exit load EFER", rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX));
427 		}
428 		vmcs_write(GUEST_PAT, ia32_efer);
429 		vmcs_write(GUEST_RIP, guest_rip + 3);
430 		return VMX_TEST_RESUME;
431 	default:
432 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
433 		break;
434 	}
435 	return VMX_TEST_VMEXIT;
436 }
437 
438 u32 guest_cr0, guest_cr4;
439 
440 static void cr_shadowing_main(void)
441 {
442 	u32 cr0, cr4, tmp;
443 
444 	// Test read through
445 	vmx_set_test_stage(0);
446 	guest_cr0 = read_cr0();
447 	if (vmx_get_test_stage() == 1)
448 		report("Read through CR0", 0);
449 	else
450 		vmcall();
451 	vmx_set_test_stage(1);
452 	guest_cr4 = read_cr4();
453 	if (vmx_get_test_stage() == 2)
454 		report("Read through CR4", 0);
455 	else
456 		vmcall();
457 	// Test write through
458 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
459 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
460 	vmx_set_test_stage(2);
461 	write_cr0(guest_cr0);
462 	if (vmx_get_test_stage() == 3)
463 		report("Write throuth CR0", 0);
464 	else
465 		vmcall();
466 	vmx_set_test_stage(3);
467 	write_cr4(guest_cr4);
468 	if (vmx_get_test_stage() == 4)
469 		report("Write through CR4", 0);
470 	else
471 		vmcall();
472 	// Test read shadow
473 	vmx_set_test_stage(4);
474 	vmcall();
475 	cr0 = read_cr0();
476 	if (vmx_get_test_stage() != 5)
477 		report("Read shadowing CR0", cr0 == guest_cr0);
478 	vmx_set_test_stage(5);
479 	cr4 = read_cr4();
480 	if (vmx_get_test_stage() != 6)
481 		report("Read shadowing CR4", cr4 == guest_cr4);
482 	// Test write shadow (same value with shadow)
483 	vmx_set_test_stage(6);
484 	write_cr0(guest_cr0);
485 	if (vmx_get_test_stage() == 7)
486 		report("Write shadowing CR0 (same value with shadow)", 0);
487 	else
488 		vmcall();
489 	vmx_set_test_stage(7);
490 	write_cr4(guest_cr4);
491 	if (vmx_get_test_stage() == 8)
492 		report("Write shadowing CR4 (same value with shadow)", 0);
493 	else
494 		vmcall();
495 	// Test write shadow (different value)
496 	vmx_set_test_stage(8);
497 	tmp = guest_cr0 ^ X86_CR0_TS;
498 	asm volatile("mov %0, %%rsi\n\t"
499 		"mov %%rsi, %%cr0\n\t"
500 		::"m"(tmp)
501 		:"rsi", "memory", "cc");
502 	report("Write shadowing different X86_CR0_TS", vmx_get_test_stage() == 9);
503 	vmx_set_test_stage(9);
504 	tmp = guest_cr0 ^ X86_CR0_MP;
505 	asm volatile("mov %0, %%rsi\n\t"
506 		"mov %%rsi, %%cr0\n\t"
507 		::"m"(tmp)
508 		:"rsi", "memory", "cc");
509 	report("Write shadowing different X86_CR0_MP", vmx_get_test_stage() == 10);
510 	vmx_set_test_stage(10);
511 	tmp = guest_cr4 ^ X86_CR4_TSD;
512 	asm volatile("mov %0, %%rsi\n\t"
513 		"mov %%rsi, %%cr4\n\t"
514 		::"m"(tmp)
515 		:"rsi", "memory", "cc");
516 	report("Write shadowing different X86_CR4_TSD", vmx_get_test_stage() == 11);
517 	vmx_set_test_stage(11);
518 	tmp = guest_cr4 ^ X86_CR4_DE;
519 	asm volatile("mov %0, %%rsi\n\t"
520 		"mov %%rsi, %%cr4\n\t"
521 		::"m"(tmp)
522 		:"rsi", "memory", "cc");
523 	report("Write shadowing different X86_CR4_DE", vmx_get_test_stage() == 12);
524 }
525 
526 static int cr_shadowing_exit_handler(void)
527 {
528 	u64 guest_rip;
529 	ulong reason;
530 	u32 insn_len;
531 	u32 exit_qual;
532 
533 	guest_rip = vmcs_read(GUEST_RIP);
534 	reason = vmcs_read(EXI_REASON) & 0xff;
535 	insn_len = vmcs_read(EXI_INST_LEN);
536 	exit_qual = vmcs_read(EXI_QUALIFICATION);
537 	switch (reason) {
538 	case VMX_VMCALL:
539 		switch (vmx_get_test_stage()) {
540 		case 0:
541 			report("Read through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
542 			break;
543 		case 1:
544 			report("Read through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
545 			break;
546 		case 2:
547 			report("Write through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
548 			break;
549 		case 3:
550 			report("Write through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
551 			break;
552 		case 4:
553 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
554 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
555 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
556 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
557 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
558 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
559 			break;
560 		case 6:
561 			report("Write shadowing CR0 (same value)",
562 					guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)));
563 			break;
564 		case 7:
565 			report("Write shadowing CR4 (same value)",
566 					guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)));
567 			break;
568 		default:
569 			// Should not reach here
570 			report("unexpected stage, %d", false,
571 			       vmx_get_test_stage());
572 			print_vmexit_info();
573 			return VMX_TEST_VMEXIT;
574 		}
575 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
576 		return VMX_TEST_RESUME;
577 	case VMX_CR:
578 		switch (vmx_get_test_stage()) {
579 		case 4:
580 			report("Read shadowing CR0", 0);
581 			vmx_inc_test_stage();
582 			break;
583 		case 5:
584 			report("Read shadowing CR4", 0);
585 			vmx_inc_test_stage();
586 			break;
587 		case 6:
588 			report("Write shadowing CR0 (same value)", 0);
589 			vmx_inc_test_stage();
590 			break;
591 		case 7:
592 			report("Write shadowing CR4 (same value)", 0);
593 			vmx_inc_test_stage();
594 			break;
595 		case 8:
596 		case 9:
597 			// 0x600 encodes "mov %esi, %cr0"
598 			if (exit_qual == 0x600)
599 				vmx_inc_test_stage();
600 			break;
601 		case 10:
602 		case 11:
603 			// 0x604 encodes "mov %esi, %cr4"
604 			if (exit_qual == 0x604)
605 				vmx_inc_test_stage();
606 			break;
607 		default:
608 			// Should not reach here
609 			report("unexpected stage, %d", false,
610 			       vmx_get_test_stage());
611 			print_vmexit_info();
612 			return VMX_TEST_VMEXIT;
613 		}
614 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
615 		return VMX_TEST_RESUME;
616 	default:
617 		report("Unknown exit reason, %ld", false, reason);
618 		print_vmexit_info();
619 	}
620 	return VMX_TEST_VMEXIT;
621 }
622 
623 static int iobmp_init(struct vmcs *vmcs)
624 {
625 	u32 ctrl_cpu0;
626 
627 	io_bitmap_a = alloc_page();
628 	io_bitmap_b = alloc_page();
629 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
630 	ctrl_cpu0 |= CPU_IO_BITMAP;
631 	ctrl_cpu0 &= (~CPU_IO);
632 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
633 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
634 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
635 	return VMX_TEST_START;
636 }
637 
638 static void iobmp_main(void)
639 {
640 	// stage 0, test IO pass
641 	vmx_set_test_stage(0);
642 	inb(0x5000);
643 	outb(0x0, 0x5000);
644 	report("I/O bitmap - I/O pass", vmx_get_test_stage() == 0);
645 	// test IO width, in/out
646 	((u8 *)io_bitmap_a)[0] = 0xFF;
647 	vmx_set_test_stage(2);
648 	inb(0x0);
649 	report("I/O bitmap - trap in", vmx_get_test_stage() == 3);
650 	vmx_set_test_stage(3);
651 	outw(0x0, 0x0);
652 	report("I/O bitmap - trap out", vmx_get_test_stage() == 4);
653 	vmx_set_test_stage(4);
654 	inl(0x0);
655 	report("I/O bitmap - I/O width, long", vmx_get_test_stage() == 5);
656 	// test low/high IO port
657 	vmx_set_test_stage(5);
658 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
659 	inb(0x5000);
660 	report("I/O bitmap - I/O port, low part", vmx_get_test_stage() == 6);
661 	vmx_set_test_stage(6);
662 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
663 	inb(0x9000);
664 	report("I/O bitmap - I/O port, high part", vmx_get_test_stage() == 7);
665 	// test partial pass
666 	vmx_set_test_stage(7);
667 	inl(0x4FFF);
668 	report("I/O bitmap - partial pass", vmx_get_test_stage() == 8);
669 	// test overrun
670 	vmx_set_test_stage(8);
671 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
672 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
673 	inl(0xFFFF);
674 	report("I/O bitmap - overrun", vmx_get_test_stage() == 9);
675 	vmx_set_test_stage(9);
676 	vmcall();
677 	outb(0x0, 0x0);
678 	report("I/O bitmap - ignore unconditional exiting",
679 	       vmx_get_test_stage() == 9);
680 	vmx_set_test_stage(10);
681 	vmcall();
682 	outb(0x0, 0x0);
683 	report("I/O bitmap - unconditional exiting",
684 	       vmx_get_test_stage() == 11);
685 }
686 
687 static int iobmp_exit_handler(void)
688 {
689 	u64 guest_rip;
690 	ulong reason, exit_qual;
691 	u32 insn_len, ctrl_cpu0;
692 
693 	guest_rip = vmcs_read(GUEST_RIP);
694 	reason = vmcs_read(EXI_REASON) & 0xff;
695 	exit_qual = vmcs_read(EXI_QUALIFICATION);
696 	insn_len = vmcs_read(EXI_INST_LEN);
697 	switch (reason) {
698 	case VMX_IO:
699 		switch (vmx_get_test_stage()) {
700 		case 0:
701 		case 1:
702 			vmx_inc_test_stage();
703 			break;
704 		case 2:
705 			report("I/O bitmap - I/O width, byte",
706 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE);
707 			report("I/O bitmap - I/O direction, in", exit_qual & VMX_IO_IN);
708 			vmx_inc_test_stage();
709 			break;
710 		case 3:
711 			report("I/O bitmap - I/O width, word",
712 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD);
713 			report("I/O bitmap - I/O direction, out",
714 					!(exit_qual & VMX_IO_IN));
715 			vmx_inc_test_stage();
716 			break;
717 		case 4:
718 			report("I/O bitmap - I/O width, long",
719 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG);
720 			vmx_inc_test_stage();
721 			break;
722 		case 5:
723 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
724 				vmx_inc_test_stage();
725 			break;
726 		case 6:
727 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
728 				vmx_inc_test_stage();
729 			break;
730 		case 7:
731 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
732 				vmx_inc_test_stage();
733 			break;
734 		case 8:
735 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
736 				vmx_inc_test_stage();
737 			break;
738 		case 9:
739 		case 10:
740 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
741 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
742 			vmx_inc_test_stage();
743 			break;
744 		default:
745 			// Should not reach here
746 			report("unexpected stage, %d", false,
747 			       vmx_get_test_stage());
748 			print_vmexit_info();
749 			return VMX_TEST_VMEXIT;
750 		}
751 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
752 		return VMX_TEST_RESUME;
753 	case VMX_VMCALL:
754 		switch (vmx_get_test_stage()) {
755 		case 9:
756 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
757 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
758 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
759 			break;
760 		case 10:
761 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
762 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
763 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
764 			break;
765 		default:
766 			// Should not reach here
767 			report("unexpected stage, %d", false,
768 			       vmx_get_test_stage());
769 			print_vmexit_info();
770 			return VMX_TEST_VMEXIT;
771 		}
772 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
773 		return VMX_TEST_RESUME;
774 	default:
775 		printf("guest_rip = %#lx\n", guest_rip);
776 		printf("\tERROR : Undefined exit reason, reason = %ld.\n", reason);
777 		break;
778 	}
779 	return VMX_TEST_VMEXIT;
780 }
781 
782 #define INSN_CPU0		0
783 #define INSN_CPU1		1
784 #define INSN_ALWAYS_TRAP	2
785 
786 #define FIELD_EXIT_QUAL		(1 << 0)
787 #define FIELD_INSN_INFO		(1 << 1)
788 
789 asm(
790 	"insn_hlt: hlt;ret\n\t"
791 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
792 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
793 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
794 	"insn_rdtsc: rdtsc;ret\n\t"
795 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
796 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
797 #ifdef __x86_64__
798 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
799 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
800 #endif
801 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
802 	"insn_pause: pause;ret\n\t"
803 	"insn_wbinvd: wbinvd;ret\n\t"
804 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
805 	"insn_invd: invd;ret\n\t"
806 	"insn_sgdt: sgdt gdt64_desc;ret\n\t"
807 	"insn_lgdt: lgdt gdt64_desc;ret\n\t"
808 	"insn_sidt: sidt idt_descr;ret\n\t"
809 	"insn_lidt: lidt idt_descr;ret\n\t"
810 	"insn_sldt: sldt %ax;ret\n\t"
811 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
812 	"insn_str: str %ax;ret\n\t"
813 	"insn_rdrand: rdrand %rax;ret\n\t"
814 	"insn_rdseed: rdseed %rax;ret\n\t"
815 );
816 extern void insn_hlt(void);
817 extern void insn_invlpg(void);
818 extern void insn_mwait(void);
819 extern void insn_rdpmc(void);
820 extern void insn_rdtsc(void);
821 extern void insn_cr3_load(void);
822 extern void insn_cr3_store(void);
823 #ifdef __x86_64__
824 extern void insn_cr8_load(void);
825 extern void insn_cr8_store(void);
826 #endif
827 extern void insn_monitor(void);
828 extern void insn_pause(void);
829 extern void insn_wbinvd(void);
830 extern void insn_sgdt(void);
831 extern void insn_lgdt(void);
832 extern void insn_sidt(void);
833 extern void insn_lidt(void);
834 extern void insn_sldt(void);
835 extern void insn_lldt(void);
836 extern void insn_str(void);
837 extern void insn_cpuid(void);
838 extern void insn_invd(void);
839 extern void insn_rdrand(void);
840 extern void insn_rdseed(void);
841 
842 u32 cur_insn;
843 u64 cr3;
844 
845 #define X86_FEATURE_MONITOR	(1 << 3)
846 
847 typedef bool (*supported_fn)(void);
848 
849 static bool monitor_supported(void)
850 {
851 	return this_cpu_has(X86_FEATURE_MWAIT);
852 }
853 
854 struct insn_table {
855 	const char *name;
856 	u32 flag;
857 	void (*insn_func)(void);
858 	u32 type;
859 	u32 reason;
860 	ulong exit_qual;
861 	u32 insn_info;
862 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
863 	// which field need to be tested, reason is always tested
864 	u32 test_field;
865 	const supported_fn supported_fn;
866 	u8 disabled;
867 };
868 
869 /*
870  * Add more test cases of instruction intercept here. Elements in this
871  * table is:
872  *	name/control flag/insn function/type/exit reason/exit qulification/
873  *	instruction info/field to test
874  * The last field defines which fields (exit_qual and insn_info) need to be
875  * tested in exit handler. If set to 0, only "reason" is checked.
876  */
877 static struct insn_table insn_table[] = {
878 	// Flags for Primary Processor-Based VM-Execution Controls
879 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
880 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
881 		0x12345678, 0, FIELD_EXIT_QUAL},
882 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0, &monitor_supported},
883 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0},
884 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
885 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
886 		FIELD_EXIT_QUAL},
887 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
888 		FIELD_EXIT_QUAL},
889 #ifdef __x86_64__
890 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
891 		FIELD_EXIT_QUAL},
892 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
893 		FIELD_EXIT_QUAL},
894 #endif
895 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0, &monitor_supported},
896 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
897 	// Flags for Secondary Processor-Based VM-Execution Controls
898 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
899 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
900 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
901 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
902 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
903 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
904 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
905 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
906 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
907 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
908 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
909 	// Instructions always trap
910 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
911 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
912 	// Instructions never trap
913 	{NULL},
914 };
915 
916 static int insn_intercept_init(struct vmcs *vmcs)
917 {
918 	u32 ctrl_cpu, cur_insn;
919 
920 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
921 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
922 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
923 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
924 	cr3 = read_cr3();
925 
926 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
927 		if (insn_table[cur_insn].supported_fn == NULL)
928 			continue;
929 		insn_table[cur_insn].disabled = !insn_table[cur_insn].supported_fn();
930 	}
931 	return VMX_TEST_START;
932 }
933 
934 static void insn_intercept_main(void)
935 {
936 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
937 		vmx_set_test_stage(cur_insn * 2);
938 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
939 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
940 		    (insn_table[cur_insn].type == INSN_CPU1 &&
941 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
942 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
943 			       insn_table[cur_insn].type - INSN_CPU0,
944 			       insn_table[cur_insn].name);
945 			continue;
946 		}
947 
948 		if (insn_table[cur_insn].disabled) {
949 			printf("\tFeature required for %s is not supported.\n",
950 			       insn_table[cur_insn].name);
951 			continue;
952 		}
953 
954 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
955 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
956 		    (insn_table[cur_insn].type == INSN_CPU1 &&
957 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
958 			/* skip hlt, it stalls the guest and is tested below */
959 			if (insn_table[cur_insn].insn_func != insn_hlt)
960 				insn_table[cur_insn].insn_func();
961 			report("execute %s", vmx_get_test_stage() == cur_insn * 2,
962 					insn_table[cur_insn].name);
963 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
964 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
965 			       insn_table[cur_insn].type - INSN_CPU0,
966 			       insn_table[cur_insn].name);
967 
968 		vmcall();
969 
970 		insn_table[cur_insn].insn_func();
971 		report("intercept %s", vmx_get_test_stage() == cur_insn * 2 + 1,
972 				insn_table[cur_insn].name);
973 
974 		vmx_set_test_stage(cur_insn * 2 + 1);
975 		vmcall();
976 	}
977 }
978 
979 static int insn_intercept_exit_handler(void)
980 {
981 	u64 guest_rip;
982 	u32 reason;
983 	ulong exit_qual;
984 	u32 insn_len;
985 	u32 insn_info;
986 	bool pass;
987 
988 	guest_rip = vmcs_read(GUEST_RIP);
989 	reason = vmcs_read(EXI_REASON) & 0xff;
990 	exit_qual = vmcs_read(EXI_QUALIFICATION);
991 	insn_len = vmcs_read(EXI_INST_LEN);
992 	insn_info = vmcs_read(EXI_INST_INFO);
993 
994 	if (reason == VMX_VMCALL) {
995 		u32 val = 0;
996 
997 		if (insn_table[cur_insn].type == INSN_CPU0)
998 			val = vmcs_read(CPU_EXEC_CTRL0);
999 		else if (insn_table[cur_insn].type == INSN_CPU1)
1000 			val = vmcs_read(CPU_EXEC_CTRL1);
1001 
1002 		if (vmx_get_test_stage() & 1)
1003 			val &= ~insn_table[cur_insn].flag;
1004 		else
1005 			val |= insn_table[cur_insn].flag;
1006 
1007 		if (insn_table[cur_insn].type == INSN_CPU0)
1008 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
1009 		else if (insn_table[cur_insn].type == INSN_CPU1)
1010 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
1011 	} else {
1012 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
1013 			insn_table[cur_insn].reason == reason;
1014 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
1015 		    insn_table[cur_insn].exit_qual != exit_qual)
1016 			pass = false;
1017 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
1018 		    insn_table[cur_insn].insn_info != insn_info)
1019 			pass = false;
1020 		if (pass)
1021 			vmx_inc_test_stage();
1022 	}
1023 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1024 	return VMX_TEST_RESUME;
1025 }
1026 
1027 /**
1028  * __setup_ept - Setup the VMCS fields to enable Extended Page Tables (EPT)
1029  * @hpa:	Host physical address of the top-level, a.k.a. root, EPT table
1030  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1031  *
1032  * Returns 0 on success, 1 on failure.
1033  *
1034  * Note that @hpa doesn't need to point at actual memory if VM-Launch is
1035  * expected to fail, e.g. setup_dummy_ept() arbitrarily passes '0' to satisfy
1036  * the various EPTP consistency checks, but doesn't ensure backing for HPA '0'.
1037  */
1038 static int __setup_ept(u64 hpa, bool enable_ad)
1039 {
1040 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1041 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1042 		printf("\tEPT is not supported");
1043 		return 1;
1044 	}
1045 	if (!(ept_vpid.val & EPT_CAP_WB)) {
1046 		printf("WB memtype for EPT walks not supported\n");
1047 		return 1;
1048 	}
1049 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
1050 		printf("\tPWL4 is not supported\n");
1051 		return 1;
1052 	}
1053 
1054 	eptp = EPT_MEM_TYPE_WB;
1055 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1056 	eptp |= hpa;
1057 	if (enable_ad)
1058 		eptp |= EPTP_AD_FLAG;
1059 
1060 	vmcs_write(EPTP, eptp);
1061 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY);
1062 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT);
1063 
1064 	return 0;
1065 }
1066 
1067 /**
1068  * setup_ept - Enable Extended Page Tables (EPT) and setup an identity map
1069  * @enable_ad:	Whether or not to enable Access/Dirty bits for EPT entries
1070  *
1071  * Returns 0 on success, 1 on failure.
1072  *
1073  * This is the "real" function for setting up EPT tables, i.e. use this for
1074  * tests that need to run code in the guest with EPT enabled.
1075  */
1076 static int setup_ept(bool enable_ad)
1077 {
1078 	unsigned long end_of_memory;
1079 
1080 	pml4 = alloc_page();
1081 
1082 	if (__setup_ept(virt_to_phys(pml4), enable_ad))
1083 		return 1;
1084 
1085 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1086 	if (end_of_memory < (1ul << 32))
1087 		end_of_memory = (1ul << 32);
1088 	/* Cannot use large EPT pages if we need to track EPT
1089 	 * accessed/dirty bits at 4K granularity.
1090 	 */
1091 	setup_ept_range(pml4, 0, end_of_memory, 0,
1092 			!enable_ad && ept_2m_supported(),
1093 			EPT_WA | EPT_RA | EPT_EA);
1094 	return 0;
1095 }
1096 
1097 /**
1098  * setup_dummy_ept - Enable Extended Page Tables (EPT) with a dummy root HPA
1099  *
1100  * Setup EPT using a semi-arbitrary dummy root HPA.  This function is intended
1101  * for use by tests that need EPT enabled to verify dependent VMCS controls
1102  * but never expect to fully enter the guest, i.e. don't need setup the actual
1103  * EPT tables.
1104  */
1105 static void setup_dummy_ept(void)
1106 {
1107 	if (__setup_ept(0, false))
1108 		report_abort("EPT setup unexpectedly failed");
1109 }
1110 
1111 static int enable_unrestricted_guest(void)
1112 {
1113 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1114 	    !(ctrl_cpu_rev[1].clr & CPU_URG) ||
1115 	    !(ctrl_cpu_rev[1].clr & CPU_EPT))
1116 		return 1;
1117 
1118 	setup_dummy_ept();
1119 
1120 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
1121 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
1122 
1123 	return 0;
1124 }
1125 
1126 static void ept_enable_ad_bits(void)
1127 {
1128 	eptp |= EPTP_AD_FLAG;
1129 	vmcs_write(EPTP, eptp);
1130 }
1131 
1132 static void ept_disable_ad_bits(void)
1133 {
1134 	eptp &= ~EPTP_AD_FLAG;
1135 	vmcs_write(EPTP, eptp);
1136 }
1137 
1138 static void ept_enable_ad_bits_or_skip_test(void)
1139 {
1140 	if (!ept_ad_bits_supported())
1141 		test_skip("EPT AD bits not supported.");
1142 	ept_enable_ad_bits();
1143 }
1144 
1145 static int apic_version;
1146 
1147 static int ept_init_common(bool have_ad)
1148 {
1149 	int ret;
1150 	struct pci_dev pcidev;
1151 
1152 	if (setup_ept(have_ad))
1153 		return VMX_TEST_EXIT;
1154 	data_page1 = alloc_page();
1155 	data_page2 = alloc_page();
1156 	*((u32 *)data_page1) = MAGIC_VAL_1;
1157 	*((u32 *)data_page2) = MAGIC_VAL_2;
1158 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1159 			EPT_RA | EPT_WA | EPT_EA);
1160 
1161 	apic_version = apic_read(APIC_LVR);
1162 
1163 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1164 	if (ret != PCIDEVADDR_INVALID) {
1165 		pci_dev_init(&pcidev, ret);
1166 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1167 	}
1168 
1169 	return VMX_TEST_START;
1170 }
1171 
1172 static int ept_init(struct vmcs *vmcs)
1173 {
1174 	return ept_init_common(false);
1175 }
1176 
1177 static void ept_common(void)
1178 {
1179 	vmx_set_test_stage(0);
1180 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1181 			*((u32 *)data_page1) != MAGIC_VAL_1)
1182 		report("EPT basic framework - read", 0);
1183 	else {
1184 		*((u32 *)data_page2) = MAGIC_VAL_3;
1185 		vmcall();
1186 		if (vmx_get_test_stage() == 1) {
1187 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1188 					*((u32 *)data_page2) == MAGIC_VAL_2)
1189 				report("EPT basic framework", 1);
1190 			else
1191 				report("EPT basic framework - remap", 1);
1192 		}
1193 	}
1194 	// Test EPT Misconfigurations
1195 	vmx_set_test_stage(1);
1196 	vmcall();
1197 	*((u32 *)data_page1) = MAGIC_VAL_1;
1198 	if (vmx_get_test_stage() != 2) {
1199 		report("EPT misconfigurations", 0);
1200 		goto t1;
1201 	}
1202 	vmx_set_test_stage(2);
1203 	vmcall();
1204 	*((u32 *)data_page1) = MAGIC_VAL_1;
1205 	report("EPT misconfigurations", vmx_get_test_stage() == 3);
1206 t1:
1207 	// Test EPT violation
1208 	vmx_set_test_stage(3);
1209 	vmcall();
1210 	*((u32 *)data_page1) = MAGIC_VAL_1;
1211 	report("EPT violation - page permission", vmx_get_test_stage() == 4);
1212 	// Violation caused by EPT paging structure
1213 	vmx_set_test_stage(4);
1214 	vmcall();
1215 	*((u32 *)data_page1) = MAGIC_VAL_2;
1216 	report("EPT violation - paging structure", vmx_get_test_stage() == 5);
1217 
1218 	// MMIO Read/Write
1219 	vmx_set_test_stage(5);
1220 	vmcall();
1221 
1222 	*(u32 volatile *)pci_physaddr;
1223 	report("MMIO EPT violation - read", vmx_get_test_stage() == 6);
1224 
1225 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1226 	report("MMIO EPT violation - write", vmx_get_test_stage() == 7);
1227 }
1228 
1229 static void ept_main(void)
1230 {
1231 	ept_common();
1232 
1233 	// Test EPT access to L1 MMIO
1234 	vmx_set_test_stage(7);
1235 	report("EPT - MMIO access", *((u32 *)0xfee00030UL) == apic_version);
1236 
1237 	// Test invalid operand for INVEPT
1238 	vmcall();
1239 	report("EPT - unsupported INVEPT", vmx_get_test_stage() == 8);
1240 }
1241 
1242 static bool invept_test(int type, u64 eptp)
1243 {
1244 	bool ret, supported;
1245 
1246 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1247 	ret = invept(type, eptp);
1248 
1249 	if (ret == !supported)
1250 		return false;
1251 
1252 	if (!supported)
1253 		printf("WARNING: unsupported invept passed!\n");
1254 	else
1255 		printf("WARNING: invept failed!\n");
1256 
1257 	return true;
1258 }
1259 
1260 static int pml_exit_handler(void)
1261 {
1262 	u16 index, count;
1263 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1264 	u64 *pmlbuf = pml_log;
1265 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1266 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1267 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1268 
1269 	switch (reason) {
1270 	case VMX_VMCALL:
1271 		switch (vmx_get_test_stage()) {
1272 		case 0:
1273 			index = vmcs_read(GUEST_PML_INDEX);
1274 			for (count = index + 1; count < PML_INDEX; count++) {
1275 				if (pmlbuf[count] == (u64)data_page2) {
1276 					vmx_inc_test_stage();
1277 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1278 					break;
1279 				}
1280 			}
1281 			break;
1282 		case 1:
1283 			index = vmcs_read(GUEST_PML_INDEX);
1284 			/* Keep clearing the dirty bit till a overflow */
1285 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1286 			break;
1287 		default:
1288 			report("unexpected stage, %d.", false,
1289 			       vmx_get_test_stage());
1290 			print_vmexit_info();
1291 			return VMX_TEST_VMEXIT;
1292 		}
1293 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1294 		return VMX_TEST_RESUME;
1295 	case VMX_PML_FULL:
1296 		vmx_inc_test_stage();
1297 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1298 		return VMX_TEST_RESUME;
1299 	default:
1300 		report("Unknown exit reason, %ld", false, reason);
1301 		print_vmexit_info();
1302 	}
1303 	return VMX_TEST_VMEXIT;
1304 }
1305 
1306 static int ept_exit_handler_common(bool have_ad)
1307 {
1308 	u64 guest_rip;
1309 	u64 guest_cr3;
1310 	ulong reason;
1311 	u32 insn_len;
1312 	u32 exit_qual;
1313 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte;
1314 
1315 	guest_rip = vmcs_read(GUEST_RIP);
1316 	guest_cr3 = vmcs_read(GUEST_CR3);
1317 	reason = vmcs_read(EXI_REASON) & 0xff;
1318 	insn_len = vmcs_read(EXI_INST_LEN);
1319 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1320 	switch (reason) {
1321 	case VMX_VMCALL:
1322 		switch (vmx_get_test_stage()) {
1323 		case 0:
1324 			check_ept_ad(pml4, guest_cr3,
1325 				     (unsigned long)data_page1,
1326 				     have_ad ? EPT_ACCESS_FLAG : 0,
1327 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1328 			check_ept_ad(pml4, guest_cr3,
1329 				     (unsigned long)data_page2,
1330 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1331 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1332 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1333 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1334 			if (have_ad)
1335 				ept_sync(INVEPT_SINGLE, eptp);;
1336 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1337 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1338 				vmx_inc_test_stage();
1339 				install_ept(pml4, (unsigned long)data_page2,
1340 						(unsigned long)data_page2,
1341 						EPT_RA | EPT_WA | EPT_EA);
1342 			} else
1343 				report("EPT basic framework - write", 0);
1344 			break;
1345 		case 1:
1346 			install_ept(pml4, (unsigned long)data_page1,
1347  				(unsigned long)data_page1, EPT_WA);
1348 			ept_sync(INVEPT_SINGLE, eptp);
1349 			break;
1350 		case 2:
1351 			install_ept(pml4, (unsigned long)data_page1,
1352  				(unsigned long)data_page1,
1353  				EPT_RA | EPT_WA | EPT_EA |
1354  				(2 << EPT_MEM_TYPE_SHIFT));
1355 			ept_sync(INVEPT_SINGLE, eptp);
1356 			break;
1357 		case 3:
1358 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1359 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1360 						1, &data_page1_pte));
1361 			set_ept_pte(pml4, (unsigned long)data_page1,
1362 				1, data_page1_pte & ~EPT_PRESENT);
1363 			ept_sync(INVEPT_SINGLE, eptp);
1364 			break;
1365 		case 4:
1366 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1367 						2, &data_page1_pte));
1368 			data_page1_pte &= PAGE_MASK;
1369 			TEST_ASSERT(get_ept_pte(pml4, data_page1_pte,
1370 						2, &data_page1_pte_pte));
1371 			set_ept_pte(pml4, data_page1_pte, 2,
1372 				data_page1_pte_pte & ~EPT_PRESENT);
1373 			ept_sync(INVEPT_SINGLE, eptp);
1374 			break;
1375 		case 5:
1376 			install_ept(pml4, (unsigned long)pci_physaddr,
1377 				(unsigned long)pci_physaddr, 0);
1378 			ept_sync(INVEPT_SINGLE, eptp);
1379 			break;
1380 		case 7:
1381 			if (!invept_test(0, eptp))
1382 				vmx_inc_test_stage();
1383 			break;
1384 		// Should not reach here
1385 		default:
1386 			report("ERROR - unexpected stage, %d.", false,
1387 			       vmx_get_test_stage());
1388 			print_vmexit_info();
1389 			return VMX_TEST_VMEXIT;
1390 		}
1391 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1392 		return VMX_TEST_RESUME;
1393 	case VMX_EPT_MISCONFIG:
1394 		switch (vmx_get_test_stage()) {
1395 		case 1:
1396 		case 2:
1397 			vmx_inc_test_stage();
1398 			install_ept(pml4, (unsigned long)data_page1,
1399  				(unsigned long)data_page1,
1400  				EPT_RA | EPT_WA | EPT_EA);
1401 			ept_sync(INVEPT_SINGLE, eptp);
1402 			break;
1403 		// Should not reach here
1404 		default:
1405 			report("ERROR - unexpected stage, %d.", false,
1406 			       vmx_get_test_stage());
1407 			print_vmexit_info();
1408 			return VMX_TEST_VMEXIT;
1409 		}
1410 		return VMX_TEST_RESUME;
1411 	case VMX_EPT_VIOLATION:
1412 		switch(vmx_get_test_stage()) {
1413 		case 3:
1414 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1415 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1416 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1417 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1418 					EPT_VLT_PADDR))
1419 				vmx_inc_test_stage();
1420 			set_ept_pte(pml4, (unsigned long)data_page1,
1421 				1, data_page1_pte | (EPT_PRESENT));
1422 			ept_sync(INVEPT_SINGLE, eptp);
1423 			break;
1424 		case 4:
1425 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1426 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1427 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1428 			if (exit_qual == (EPT_VLT_RD |
1429 					  (have_ad ? EPT_VLT_WR : 0) |
1430 					  EPT_VLT_LADDR_VLD))
1431 				vmx_inc_test_stage();
1432 			set_ept_pte(pml4, data_page1_pte, 2,
1433 				data_page1_pte_pte | (EPT_PRESENT));
1434 			ept_sync(INVEPT_SINGLE, eptp);
1435 			break;
1436 		case 5:
1437 			if (exit_qual & EPT_VLT_RD)
1438 				vmx_inc_test_stage();
1439 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1440 						1, &memaddr_pte));
1441 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1442 			ept_sync(INVEPT_SINGLE, eptp);
1443 			break;
1444 		case 6:
1445 			if (exit_qual & EPT_VLT_WR)
1446 				vmx_inc_test_stage();
1447 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1448 						1, &memaddr_pte));
1449 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1450 			ept_sync(INVEPT_SINGLE, eptp);
1451 			break;
1452 		default:
1453 			// Should not reach here
1454 			report("ERROR : unexpected stage, %d", false,
1455 			       vmx_get_test_stage());
1456 			print_vmexit_info();
1457 			return VMX_TEST_VMEXIT;
1458 		}
1459 		return VMX_TEST_RESUME;
1460 	default:
1461 		report("Unknown exit reason, %ld", false, reason);
1462 		print_vmexit_info();
1463 	}
1464 	return VMX_TEST_VMEXIT;
1465 }
1466 
1467 static int ept_exit_handler(void)
1468 {
1469 	return ept_exit_handler_common(false);
1470 }
1471 
1472 static int eptad_init(struct vmcs *vmcs)
1473 {
1474 	int r = ept_init_common(true);
1475 
1476 	if (r == VMX_TEST_EXIT)
1477 		return r;
1478 
1479 	if ((rdmsr(MSR_IA32_VMX_EPT_VPID_CAP) & EPT_CAP_AD_FLAG) == 0) {
1480 		printf("\tEPT A/D bits are not supported");
1481 		return VMX_TEST_EXIT;
1482 	}
1483 
1484 	return r;
1485 }
1486 
1487 static int pml_init(struct vmcs *vmcs)
1488 {
1489 	u32 ctrl_cpu;
1490 	int r = eptad_init(vmcs);
1491 
1492 	if (r == VMX_TEST_EXIT)
1493 		return r;
1494 
1495 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1496 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1497 		printf("\tPML is not supported");
1498 		return VMX_TEST_EXIT;
1499 	}
1500 
1501 	pml_log = alloc_page();
1502 	vmcs_write(PMLADDR, (u64)pml_log);
1503 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1504 
1505 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1506 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1507 
1508 	return VMX_TEST_START;
1509 }
1510 
1511 static void pml_main(void)
1512 {
1513 	int count = 0;
1514 
1515 	vmx_set_test_stage(0);
1516 	*((u32 *)data_page2) = 0x1;
1517 	vmcall();
1518 	report("PML - Dirty GPA Logging", vmx_get_test_stage() == 1);
1519 
1520 	while (vmx_get_test_stage() == 1) {
1521 		vmcall();
1522 		*((u32 *)data_page2) = 0x1;
1523 		if (count++ > PML_INDEX)
1524 			break;
1525 	}
1526 	report("PML Full Event", vmx_get_test_stage() == 2);
1527 }
1528 
1529 static void eptad_main(void)
1530 {
1531 	ept_common();
1532 }
1533 
1534 static int eptad_exit_handler(void)
1535 {
1536 	return ept_exit_handler_common(true);
1537 }
1538 
1539 static bool invvpid_test(int type, u16 vpid)
1540 {
1541 	bool ret, supported;
1542 
1543 	supported = ept_vpid.val &
1544 		(VPID_CAP_INVVPID_ADDR >> INVVPID_ADDR << type);
1545 	ret = invvpid(type, vpid, 0);
1546 
1547 	if (ret == !supported)
1548 		return false;
1549 
1550 	if (!supported)
1551 		printf("WARNING: unsupported invvpid passed!\n");
1552 	else
1553 		printf("WARNING: invvpid failed!\n");
1554 
1555 	return true;
1556 }
1557 
1558 static int vpid_init(struct vmcs *vmcs)
1559 {
1560 	u32 ctrl_cpu1;
1561 
1562 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1563 		!(ctrl_cpu_rev[1].clr & CPU_VPID)) {
1564 		printf("\tVPID is not supported");
1565 		return VMX_TEST_EXIT;
1566 	}
1567 
1568 	ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
1569 	ctrl_cpu1 |= CPU_VPID;
1570 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
1571 	return VMX_TEST_START;
1572 }
1573 
1574 static void vpid_main(void)
1575 {
1576 	vmx_set_test_stage(0);
1577 	vmcall();
1578 	report("INVVPID SINGLE ADDRESS", vmx_get_test_stage() == 1);
1579 	vmx_set_test_stage(2);
1580 	vmcall();
1581 	report("INVVPID SINGLE", vmx_get_test_stage() == 3);
1582 	vmx_set_test_stage(4);
1583 	vmcall();
1584 	report("INVVPID ALL", vmx_get_test_stage() == 5);
1585 }
1586 
1587 static int vpid_exit_handler(void)
1588 {
1589 	u64 guest_rip;
1590 	ulong reason;
1591 	u32 insn_len;
1592 
1593 	guest_rip = vmcs_read(GUEST_RIP);
1594 	reason = vmcs_read(EXI_REASON) & 0xff;
1595 	insn_len = vmcs_read(EXI_INST_LEN);
1596 
1597 	switch (reason) {
1598 	case VMX_VMCALL:
1599 		switch(vmx_get_test_stage()) {
1600 		case 0:
1601 			if (!invvpid_test(INVVPID_ADDR, 1))
1602 				vmx_inc_test_stage();
1603 			break;
1604 		case 2:
1605 			if (!invvpid_test(INVVPID_CONTEXT_GLOBAL, 1))
1606 				vmx_inc_test_stage();
1607 			break;
1608 		case 4:
1609 			if (!invvpid_test(INVVPID_ALL, 1))
1610 				vmx_inc_test_stage();
1611 			break;
1612 		default:
1613 			report("ERROR: unexpected stage, %d", false,
1614 					vmx_get_test_stage());
1615 			print_vmexit_info();
1616 			return VMX_TEST_VMEXIT;
1617 		}
1618 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1619 		return VMX_TEST_RESUME;
1620 	default:
1621 		report("Unknown exit reason, %ld", false, reason);
1622 		print_vmexit_info();
1623 	}
1624 	return VMX_TEST_VMEXIT;
1625 }
1626 
1627 #define TIMER_VECTOR	222
1628 
1629 static volatile bool timer_fired;
1630 
1631 static void timer_isr(isr_regs_t *regs)
1632 {
1633 	timer_fired = true;
1634 	apic_write(APIC_EOI, 0);
1635 }
1636 
1637 static int interrupt_init(struct vmcs *vmcs)
1638 {
1639 	msr_bmp_init();
1640 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1641 	handle_irq(TIMER_VECTOR, timer_isr);
1642 	return VMX_TEST_START;
1643 }
1644 
1645 static void interrupt_main(void)
1646 {
1647 	long long start, loops;
1648 
1649 	vmx_set_test_stage(0);
1650 
1651 	apic_write(APIC_LVTT, TIMER_VECTOR);
1652 	irq_enable();
1653 
1654 	apic_write(APIC_TMICT, 1);
1655 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1656 		asm volatile ("nop");
1657 	report("direct interrupt while running guest", timer_fired);
1658 
1659 	apic_write(APIC_TMICT, 0);
1660 	irq_disable();
1661 	vmcall();
1662 	timer_fired = false;
1663 	apic_write(APIC_TMICT, 1);
1664 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1665 		asm volatile ("nop");
1666 	report("intercepted interrupt while running guest", timer_fired);
1667 
1668 	irq_enable();
1669 	apic_write(APIC_TMICT, 0);
1670 	irq_disable();
1671 	vmcall();
1672 	timer_fired = false;
1673 	start = rdtsc();
1674 	apic_write(APIC_TMICT, 1000000);
1675 
1676 	asm volatile ("sti; hlt");
1677 
1678 	report("direct interrupt + hlt",
1679 	       rdtsc() - start > 1000000 && timer_fired);
1680 
1681 	apic_write(APIC_TMICT, 0);
1682 	irq_disable();
1683 	vmcall();
1684 	timer_fired = false;
1685 	start = rdtsc();
1686 	apic_write(APIC_TMICT, 1000000);
1687 
1688 	asm volatile ("sti; hlt");
1689 
1690 	report("intercepted interrupt + hlt",
1691 	       rdtsc() - start > 10000 && timer_fired);
1692 
1693 	apic_write(APIC_TMICT, 0);
1694 	irq_disable();
1695 	vmcall();
1696 	timer_fired = false;
1697 	start = rdtsc();
1698 	apic_write(APIC_TMICT, 1000000);
1699 
1700 	irq_enable();
1701 	asm volatile ("nop");
1702 	vmcall();
1703 
1704 	report("direct interrupt + activity state hlt",
1705 	       rdtsc() - start > 10000 && timer_fired);
1706 
1707 	apic_write(APIC_TMICT, 0);
1708 	irq_disable();
1709 	vmcall();
1710 	timer_fired = false;
1711 	start = rdtsc();
1712 	apic_write(APIC_TMICT, 1000000);
1713 
1714 	irq_enable();
1715 	asm volatile ("nop");
1716 	vmcall();
1717 
1718 	report("intercepted interrupt + activity state hlt",
1719 	       rdtsc() - start > 10000 && timer_fired);
1720 
1721 	apic_write(APIC_TMICT, 0);
1722 	irq_disable();
1723 	vmx_set_test_stage(7);
1724 	vmcall();
1725 	timer_fired = false;
1726 	apic_write(APIC_TMICT, 1);
1727 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1728 		asm volatile ("nop");
1729 	report("running a guest with interrupt acknowledgement set", timer_fired);
1730 
1731 	apic_write(APIC_TMICT, 0);
1732 	irq_enable();
1733 	timer_fired = false;
1734 	vmcall();
1735 	report("Inject an event to a halted guest", timer_fired);
1736 }
1737 
1738 static int interrupt_exit_handler(void)
1739 {
1740 	u64 guest_rip = vmcs_read(GUEST_RIP);
1741 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1742 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1743 
1744 	switch (reason) {
1745 	case VMX_VMCALL:
1746 		switch (vmx_get_test_stage()) {
1747 		case 0:
1748 		case 2:
1749 		case 5:
1750 			vmcs_write(PIN_CONTROLS,
1751 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1752 			break;
1753 		case 7:
1754 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1755 			vmcs_write(PIN_CONTROLS,
1756 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1757 			break;
1758 		case 1:
1759 		case 3:
1760 			vmcs_write(PIN_CONTROLS,
1761 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1762 			break;
1763 		case 4:
1764 		case 6:
1765 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1766 			break;
1767 
1768 		case 8:
1769 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1770 			vmcs_write(ENT_INTR_INFO,
1771 				   TIMER_VECTOR |
1772 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1773 				   INTR_INFO_VALID_MASK);
1774 			break;
1775 		}
1776 		vmx_inc_test_stage();
1777 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1778 		return VMX_TEST_RESUME;
1779 	case VMX_EXTINT:
1780 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1781 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1782 			handle_external_interrupt(vector);
1783 		} else {
1784 			irq_enable();
1785 			asm volatile ("nop");
1786 			irq_disable();
1787 		}
1788 		if (vmx_get_test_stage() >= 2)
1789 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1790 		return VMX_TEST_RESUME;
1791 	default:
1792 		report("Unknown exit reason, %ld", false, reason);
1793 		print_vmexit_info();
1794 	}
1795 
1796 	return VMX_TEST_VMEXIT;
1797 }
1798 
1799 static int dbgctls_init(struct vmcs *vmcs)
1800 {
1801 	u64 dr7 = 0x402;
1802 	u64 zero = 0;
1803 
1804 	msr_bmp_init();
1805 	asm volatile(
1806 		"mov %0,%%dr0\n\t"
1807 		"mov %0,%%dr1\n\t"
1808 		"mov %0,%%dr2\n\t"
1809 		"mov %1,%%dr7\n\t"
1810 		: : "r" (zero), "r" (dr7));
1811 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1812 	vmcs_write(GUEST_DR7, 0x404);
1813 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1814 
1815 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1816 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1817 
1818 	return VMX_TEST_START;
1819 }
1820 
1821 static void dbgctls_main(void)
1822 {
1823 	u64 dr7, debugctl;
1824 
1825 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1826 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1827 	/* Commented out: KVM does not support DEBUGCTL so far */
1828 	(void)debugctl;
1829 	report("Load debug controls", dr7 == 0x404 /* && debugctl == 0x2 */);
1830 
1831 	dr7 = 0x408;
1832 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1833 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1834 
1835 	vmx_set_test_stage(0);
1836 	vmcall();
1837 	report("Save debug controls", vmx_get_test_stage() == 1);
1838 
1839 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1840 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1841 		printf("\tDebug controls are always loaded/saved\n");
1842 		return;
1843 	}
1844 	vmx_set_test_stage(2);
1845 	vmcall();
1846 
1847 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1848 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1849 	/* Commented out: KVM does not support DEBUGCTL so far */
1850 	(void)debugctl;
1851 	report("Guest=host debug controls", dr7 == 0x402 /* && debugctl == 0x1 */);
1852 
1853 	dr7 = 0x408;
1854 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1855 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1856 
1857 	vmx_set_test_stage(3);
1858 	vmcall();
1859 	report("Don't save debug controls", vmx_get_test_stage() == 4);
1860 }
1861 
1862 static int dbgctls_exit_handler(void)
1863 {
1864 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
1865 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1866 	u64 guest_rip = vmcs_read(GUEST_RIP);
1867 	u64 dr7, debugctl;
1868 
1869 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1870 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1871 
1872 	switch (reason) {
1873 	case VMX_VMCALL:
1874 		switch (vmx_get_test_stage()) {
1875 		case 0:
1876 			if (dr7 == 0x400 && debugctl == 0 &&
1877 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1878 			    Commented out: KVM does not support DEBUGCTL so far
1879 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1880 				vmx_inc_test_stage();
1881 			break;
1882 		case 2:
1883 			dr7 = 0x402;
1884 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1885 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1886 			vmcs_write(GUEST_DR7, 0x404);
1887 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1888 
1889 			vmcs_write(ENT_CONTROLS,
1890 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1891 			vmcs_write(EXI_CONTROLS,
1892 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1893 			break;
1894 		case 3:
1895 			if (dr7 == 0x400 && debugctl == 0 &&
1896 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1897 			    Commented out: KVM does not support DEBUGCTL so far
1898 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1899 				vmx_inc_test_stage();
1900 			break;
1901 		}
1902 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1903 		return VMX_TEST_RESUME;
1904 	default:
1905 		report("Unknown exit reason, %d", false, reason);
1906 		print_vmexit_info();
1907 	}
1908 	return VMX_TEST_VMEXIT;
1909 }
1910 
1911 struct vmx_msr_entry {
1912 	u32 index;
1913 	u32 reserved;
1914 	u64 value;
1915 } __attribute__((packed));
1916 
1917 #define MSR_MAGIC 0x31415926
1918 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1919 
1920 static int msr_switch_init(struct vmcs *vmcs)
1921 {
1922 	msr_bmp_init();
1923 	exit_msr_store = alloc_page();
1924 	exit_msr_load = alloc_page();
1925 	entry_msr_load = alloc_page();
1926 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1927 	entry_msr_load[0].value = MSR_MAGIC;
1928 
1929 	vmx_set_test_stage(1);
1930 	vmcs_write(ENT_MSR_LD_CNT, 1);
1931 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1932 	vmcs_write(EXI_MSR_ST_CNT, 1);
1933 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1934 	vmcs_write(EXI_MSR_LD_CNT, 1);
1935 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1936 	return VMX_TEST_START;
1937 }
1938 
1939 static void msr_switch_main(void)
1940 {
1941 	if (vmx_get_test_stage() == 1) {
1942 		report("VM entry MSR load",
1943 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC);
1944 		vmx_set_test_stage(2);
1945 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
1946 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
1947 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
1948 		exit_msr_load[0].value = MSR_MAGIC + 2;
1949 	}
1950 	vmcall();
1951 }
1952 
1953 static int msr_switch_exit_handler(void)
1954 {
1955 	ulong reason;
1956 
1957 	reason = vmcs_read(EXI_REASON);
1958 	if (reason == VMX_VMCALL && vmx_get_test_stage() == 2) {
1959 		report("VM exit MSR store",
1960 			exit_msr_store[0].value == MSR_MAGIC + 1);
1961 		report("VM exit MSR load",
1962 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2);
1963 		vmx_set_test_stage(3);
1964 		entry_msr_load[0].index = MSR_FS_BASE;
1965 		return VMX_TEST_RESUME;
1966 	}
1967 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1968 		__func__, vmx_get_test_stage(), reason);
1969 	return VMX_TEST_EXIT;
1970 }
1971 
1972 static int msr_switch_entry_failure(struct vmentry_failure *failure)
1973 {
1974 	ulong reason;
1975 
1976 	if (failure->early) {
1977 		printf("ERROR %s: early exit\n", __func__);
1978 		return VMX_TEST_EXIT;
1979 	}
1980 
1981 	reason = vmcs_read(EXI_REASON);
1982 	if (reason == (VMX_ENTRY_FAILURE | VMX_FAIL_MSR) &&
1983 	    vmx_get_test_stage() == 3) {
1984 		report("VM entry MSR load: try to load FS_BASE",
1985 			vmcs_read(EXI_QUALIFICATION) == 1);
1986 		return VMX_TEST_VMEXIT;
1987 	}
1988 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1989 		__func__, vmx_get_test_stage(), reason);
1990 	return VMX_TEST_EXIT;
1991 }
1992 
1993 static int vmmcall_init(struct vmcs *vmcs)
1994 {
1995 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
1996 	return VMX_TEST_START;
1997 }
1998 
1999 static void vmmcall_main(void)
2000 {
2001 	asm volatile(
2002 		"mov $0xABCD, %%rax\n\t"
2003 		"vmmcall\n\t"
2004 		::: "rax");
2005 
2006 	report("VMMCALL", 0);
2007 }
2008 
2009 static int vmmcall_exit_handler(void)
2010 {
2011 	ulong reason;
2012 
2013 	reason = vmcs_read(EXI_REASON);
2014 	switch (reason) {
2015 	case VMX_VMCALL:
2016 		printf("here\n");
2017 		report("VMMCALL triggers #UD", 0);
2018 		break;
2019 	case VMX_EXC_NMI:
2020 		report("VMMCALL triggers #UD",
2021 		       (vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR);
2022 		break;
2023 	default:
2024 		report("Unknown exit reason, %ld", false, reason);
2025 		print_vmexit_info();
2026 	}
2027 
2028 	return VMX_TEST_VMEXIT;
2029 }
2030 
2031 static int disable_rdtscp_init(struct vmcs *vmcs)
2032 {
2033 	u32 ctrl_cpu1;
2034 
2035 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
2036 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
2037 		ctrl_cpu1 &= ~CPU_RDTSCP;
2038 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
2039 	}
2040 
2041 	return VMX_TEST_START;
2042 }
2043 
2044 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
2045 {
2046 	switch (vmx_get_test_stage()) {
2047 	case 0:
2048 		report("RDTSCP triggers #UD", true);
2049 		vmx_inc_test_stage();
2050 		regs->rip += 3;
2051 		break;
2052 	case 2:
2053 		report("RDPID triggers #UD", true);
2054 		vmx_inc_test_stage();
2055 		regs->rip += 4;
2056 		break;
2057 	}
2058 	return;
2059 
2060 }
2061 
2062 static void disable_rdtscp_main(void)
2063 {
2064 	/* Test that #UD is properly injected in L2.  */
2065 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2066 
2067 	vmx_set_test_stage(0);
2068 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2069 	vmcall();
2070 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2071 
2072 	handle_exception(UD_VECTOR, 0);
2073 	vmcall();
2074 }
2075 
2076 static int disable_rdtscp_exit_handler(void)
2077 {
2078 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
2079 
2080 	switch (reason) {
2081 	case VMX_VMCALL:
2082 		switch (vmx_get_test_stage()) {
2083 		case 0:
2084 			report("RDTSCP triggers #UD", false);
2085 			vmx_inc_test_stage();
2086 			/* fallthrough */
2087 		case 1:
2088 			vmx_inc_test_stage();
2089 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2090 			return VMX_TEST_RESUME;
2091 		case 2:
2092 			report("RDPID triggers #UD", false);
2093 			break;
2094 		}
2095 		break;
2096 
2097 	default:
2098 		report("Unknown exit reason, %d", false, reason);
2099 		print_vmexit_info();
2100 	}
2101 	return VMX_TEST_VMEXIT;
2102 }
2103 
2104 static int int3_init(struct vmcs *vmcs)
2105 {
2106 	vmcs_write(EXC_BITMAP, ~0u);
2107 	return VMX_TEST_START;
2108 }
2109 
2110 static void int3_guest_main(void)
2111 {
2112 	asm volatile ("int3");
2113 }
2114 
2115 static int int3_exit_handler(void)
2116 {
2117 	u32 reason = vmcs_read(EXI_REASON);
2118 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2119 
2120 	report("L1 intercepts #BP", reason == VMX_EXC_NMI &&
2121 	       (intr_info & INTR_INFO_VALID_MASK) &&
2122 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2123 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2124 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2125 
2126 	return VMX_TEST_VMEXIT;
2127 }
2128 
2129 static int into_init(struct vmcs *vmcs)
2130 {
2131 	vmcs_write(EXC_BITMAP, ~0u);
2132 	return VMX_TEST_START;
2133 }
2134 
2135 static void into_guest_main(void)
2136 {
2137 	struct far_pointer32 fp = {
2138 		.offset = (uintptr_t)&&into,
2139 		.selector = KERNEL_CS32,
2140 	};
2141 	register uintptr_t rsp asm("rsp");
2142 
2143 	if (fp.offset != (uintptr_t)&&into) {
2144 		printf("Code address too high.\n");
2145 		return;
2146 	}
2147 	if ((u32)rsp != rsp) {
2148 		printf("Stack address too high.\n");
2149 		return;
2150 	}
2151 
2152 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2153 	return;
2154 into:
2155 	asm volatile (".code32;"
2156 		      "movl $0x7fffffff, %eax;"
2157 		      "addl %eax, %eax;"
2158 		      "into;"
2159 		      "lret;"
2160 		      ".code64");
2161 	__builtin_unreachable();
2162 }
2163 
2164 static int into_exit_handler(void)
2165 {
2166 	u32 reason = vmcs_read(EXI_REASON);
2167 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2168 
2169 	report("L1 intercepts #OF", reason == VMX_EXC_NMI &&
2170 	       (intr_info & INTR_INFO_VALID_MASK) &&
2171 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2172 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2173 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2174 
2175 	return VMX_TEST_VMEXIT;
2176 }
2177 
2178 static void exit_monitor_from_l2_main(void)
2179 {
2180 	printf("Calling exit(0) from l2...\n");
2181 	exit(0);
2182 }
2183 
2184 static int exit_monitor_from_l2_handler(void)
2185 {
2186 	report("The guest should have killed the VMM", false);
2187 	return VMX_TEST_EXIT;
2188 }
2189 
2190 static void assert_exit_reason(u64 expected)
2191 {
2192 	u64 actual = vmcs_read(EXI_REASON);
2193 
2194 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2195 			   exit_reason_description(expected),
2196 			   exit_reason_description(actual));
2197 }
2198 
2199 static void skip_exit_insn(void)
2200 {
2201 	u64 guest_rip = vmcs_read(GUEST_RIP);
2202 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2203 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2204 }
2205 
2206 static void skip_exit_vmcall(void)
2207 {
2208 	assert_exit_reason(VMX_VMCALL);
2209 	skip_exit_insn();
2210 }
2211 
2212 static void v2_null_test_guest(void)
2213 {
2214 }
2215 
2216 static void v2_null_test(void)
2217 {
2218 	test_set_guest(v2_null_test_guest);
2219 	enter_guest();
2220 	report(__func__, 1);
2221 }
2222 
2223 static void v2_multiple_entries_test_guest(void)
2224 {
2225 	vmx_set_test_stage(1);
2226 	vmcall();
2227 	vmx_set_test_stage(2);
2228 }
2229 
2230 static void v2_multiple_entries_test(void)
2231 {
2232 	test_set_guest(v2_multiple_entries_test_guest);
2233 	enter_guest();
2234 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2235 	skip_exit_vmcall();
2236 	enter_guest();
2237 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2238 	report(__func__, 1);
2239 }
2240 
2241 static int fixture_test_data = 1;
2242 
2243 static void fixture_test_teardown(void *data)
2244 {
2245 	*((int *) data) = 1;
2246 }
2247 
2248 static void fixture_test_guest(void)
2249 {
2250 	fixture_test_data++;
2251 }
2252 
2253 
2254 static void fixture_test_setup(void)
2255 {
2256 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2257 			   "fixture_test_teardown didn't run?!");
2258 	fixture_test_data = 2;
2259 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2260 	test_set_guest(fixture_test_guest);
2261 }
2262 
2263 static void fixture_test_case1(void)
2264 {
2265 	fixture_test_setup();
2266 	TEST_ASSERT_EQ(2, fixture_test_data);
2267 	enter_guest();
2268 	TEST_ASSERT_EQ(3, fixture_test_data);
2269 	report(__func__, 1);
2270 }
2271 
2272 static void fixture_test_case2(void)
2273 {
2274 	fixture_test_setup();
2275 	TEST_ASSERT_EQ(2, fixture_test_data);
2276 	enter_guest();
2277 	TEST_ASSERT_EQ(3, fixture_test_data);
2278 	report(__func__, 1);
2279 }
2280 
2281 enum ept_access_op {
2282 	OP_READ,
2283 	OP_WRITE,
2284 	OP_EXEC,
2285 	OP_FLUSH_TLB,
2286 	OP_EXIT,
2287 };
2288 
2289 static struct ept_access_test_data {
2290 	unsigned long gpa;
2291 	unsigned long *gva;
2292 	unsigned long hpa;
2293 	unsigned long *hva;
2294 	enum ept_access_op op;
2295 } ept_access_test_data;
2296 
2297 extern unsigned char ret42_start;
2298 extern unsigned char ret42_end;
2299 
2300 /* Returns 42. */
2301 asm(
2302 	".align 64\n"
2303 	"ret42_start:\n"
2304 	"mov $42, %eax\n"
2305 	"ret\n"
2306 	"ret42_end:\n"
2307 );
2308 
2309 static void
2310 diagnose_ept_violation_qual(u64 expected, u64 actual)
2311 {
2312 
2313 #define DIAGNOSE(flag)							\
2314 do {									\
2315 	if ((expected & flag) != (actual & flag))			\
2316 		printf(#flag " %sexpected\n",				\
2317 		       (expected & flag) ? "" : "un");			\
2318 } while (0)
2319 
2320 	DIAGNOSE(EPT_VLT_RD);
2321 	DIAGNOSE(EPT_VLT_WR);
2322 	DIAGNOSE(EPT_VLT_FETCH);
2323 	DIAGNOSE(EPT_VLT_PERM_RD);
2324 	DIAGNOSE(EPT_VLT_PERM_WR);
2325 	DIAGNOSE(EPT_VLT_PERM_EX);
2326 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2327 	DIAGNOSE(EPT_VLT_PADDR);
2328 
2329 #undef DIAGNOSE
2330 }
2331 
2332 static void do_ept_access_op(enum ept_access_op op)
2333 {
2334 	ept_access_test_data.op = op;
2335 	enter_guest();
2336 }
2337 
2338 /*
2339  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2340  * needed by tests that modify guest PTEs.
2341  */
2342 static void ept_access_test_guest_flush_tlb(void)
2343 {
2344 	do_ept_access_op(OP_FLUSH_TLB);
2345 	skip_exit_vmcall();
2346 }
2347 
2348 /*
2349  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2350  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2351  * a huge page.
2352  */
2353 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2354 				 unsigned long clear, unsigned long set)
2355 {
2356 	struct ept_access_test_data *data = &ept_access_test_data;
2357 	unsigned long orig_pte;
2358 	unsigned long pte;
2359 
2360 	/* Screw with the mapping at the requested level. */
2361 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2362 	pte = orig_pte;
2363 	if (mkhuge)
2364 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2365 	else
2366 		pte = orig_pte;
2367 	pte = (pte & ~clear) | set;
2368 	set_ept_pte(pml4, gpa, level, pte);
2369 	ept_sync(INVEPT_SINGLE, eptp);
2370 
2371 	return orig_pte;
2372 }
2373 
2374 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2375 {
2376 	set_ept_pte(pml4, gpa, level, orig_pte);
2377 }
2378 
2379 static void do_ept_violation(bool leaf, enum ept_access_op op,
2380 			     u64 expected_qual, u64 expected_paddr)
2381 {
2382 	u64 qual;
2383 
2384 	/* Try the access and observe the violation. */
2385 	do_ept_access_op(op);
2386 
2387 	assert_exit_reason(VMX_EPT_VIOLATION);
2388 
2389 	qual = vmcs_read(EXI_QUALIFICATION);
2390 
2391 	/* Mask undefined bits (which may later be defined in certain cases). */
2392 	qual &= ~(EPT_VLT_GUEST_USER | EPT_VLT_GUEST_RW | EPT_VLT_GUEST_EX |
2393 		 EPT_VLT_PERM_USER_EX);
2394 
2395 	diagnose_ept_violation_qual(expected_qual, qual);
2396 	TEST_EXPECT_EQ(expected_qual, qual);
2397 
2398 	#if 0
2399 	/* Disable for now otherwise every test will fail */
2400 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2401 		       (unsigned long) (
2402 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2403 	#endif
2404 	/*
2405 	 * TODO: tests that probe expected_paddr in pages other than the one at
2406 	 * the beginning of the 1g region.
2407 	 */
2408 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2409 }
2410 
2411 static void
2412 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2413 			      unsigned long set, enum ept_access_op op,
2414 			      u64 expected_qual)
2415 {
2416 	struct ept_access_test_data *data = &ept_access_test_data;
2417 	unsigned long orig_pte;
2418 
2419 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2420 
2421 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2422 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2423 					 data->gpa);
2424 
2425 	/* Fix the violation and resume the op loop. */
2426 	ept_untwiddle(data->gpa, level, orig_pte);
2427 	enter_guest();
2428 	skip_exit_vmcall();
2429 }
2430 
2431 static void
2432 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2433 		       enum ept_access_op op, u64 expected_qual)
2434 {
2435 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2436 				      expected_qual);
2437 	if (ept_huge_pages_supported(level))
2438 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2439 					      expected_qual);
2440 }
2441 
2442 static void ept_violation(unsigned long clear, unsigned long set,
2443 			  enum ept_access_op op, u64 expected_qual)
2444 {
2445 	ept_violation_at_level(1, clear, set, op, expected_qual);
2446 	ept_violation_at_level(2, clear, set, op, expected_qual);
2447 	ept_violation_at_level(3, clear, set, op, expected_qual);
2448 	ept_violation_at_level(4, clear, set, op, expected_qual);
2449 }
2450 
2451 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2452 				       u64 expected_qual)
2453 {
2454 	ept_violation(EPT_PRESENT, access, op,
2455 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2456 }
2457 
2458 /*
2459  * For translations that don't involve a GVA, that is physical address (paddr)
2460  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2461  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2462  * translations don't involve GVAs, such as when the hardware does the guest
2463  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2464  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2465  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2466  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2467  *
2468  * Note that paddr violations can also be triggered by loading PAE page tables
2469  * with wonky addresses. We don't test that yet.
2470  *
2471  * This function modifies the EPT entry that maps the GPA that the guest page
2472  * table entry mapping ept_access_data.gva resides on.
2473  *
2474  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2475  *
2476  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2477  *
2478  *	@op		Guest operation to perform with ept_access_data.gva.
2479  *
2480  *	@expect_violation
2481  *			Is a violation expected during the paddr access?
2482  *
2483  *	@expected_qual	Expected qualification for the EPT violation.
2484  *			EPT_VLT_PADDR should be clear.
2485  */
2486 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2487 			     enum ept_access_op op, bool expect_violation,
2488 			     u64 expected_qual)
2489 {
2490 	struct ept_access_test_data *data = &ept_access_test_data;
2491 	unsigned long *ptep;
2492 	unsigned long gpa;
2493 	unsigned long orig_epte;
2494 
2495 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2496 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2497 	TEST_ASSERT(ptep);
2498 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2499 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2500 	ept_access_test_guest_flush_tlb();
2501 
2502 	/*
2503 	 * Now modify the access bits on the EPT entry for the GPA that the
2504 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2505 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2506 	 * constructed our test such that those other 511 PTEs aren't used by
2507 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2508 	 * PTE we're modifying is at the beginning of a 4K page and the
2509 	 * following 511 entires are also under our control (and not touched by
2510 	 * the guest).
2511 	 */
2512 	gpa = virt_to_phys(ptep);
2513 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2514 	/*
2515 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2516 	 * otherwise our level=1 twiddling below will fail. We use the
2517 	 * identity map (gpa = gpa) since page tables are shared with the host.
2518 	 */
2519 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2520 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2521 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2522 
2523 	if (expect_violation) {
2524 		do_ept_violation(/*leaf=*/true, op,
2525 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2526 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2527 		do_ept_access_op(op);
2528 	} else {
2529 		do_ept_access_op(op);
2530 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2531 	}
2532 
2533 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2534 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2535 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2536 
2537 	skip_exit_vmcall();
2538 }
2539 
2540 static void ept_access_allowed_paddr(unsigned long ept_access,
2541 				     unsigned long pte_ad,
2542 				     enum ept_access_op op)
2543 {
2544 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2545 			 /*expected_qual=*/-1);
2546 }
2547 
2548 static void ept_access_violation_paddr(unsigned long ept_access,
2549 				       unsigned long pte_ad,
2550 				       enum ept_access_op op,
2551 				       u64 expected_qual)
2552 {
2553 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2554 			 expected_qual);
2555 }
2556 
2557 
2558 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2559 					unsigned long clear,
2560 					unsigned long set,
2561 					enum ept_access_op op)
2562 {
2563 	struct ept_access_test_data *data = &ept_access_test_data;
2564 	unsigned long orig_pte;
2565 
2566 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2567 
2568 	/* No violation. Should proceed to vmcall. */
2569 	do_ept_access_op(op);
2570 	skip_exit_vmcall();
2571 
2572 	ept_untwiddle(data->gpa, level, orig_pte);
2573 }
2574 
2575 static void ept_allowed_at_level(int level, unsigned long clear,
2576 				 unsigned long set, enum ept_access_op op)
2577 {
2578 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2579 	if (ept_huge_pages_supported(level))
2580 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2581 }
2582 
2583 static void ept_allowed(unsigned long clear, unsigned long set,
2584 			enum ept_access_op op)
2585 {
2586 	ept_allowed_at_level(1, clear, set, op);
2587 	ept_allowed_at_level(2, clear, set, op);
2588 	ept_allowed_at_level(3, clear, set, op);
2589 	ept_allowed_at_level(4, clear, set, op);
2590 }
2591 
2592 static void ept_ignored_bit(int bit)
2593 {
2594 	/* Set the bit. */
2595 	ept_allowed(0, 1ul << bit, OP_READ);
2596 	ept_allowed(0, 1ul << bit, OP_WRITE);
2597 	ept_allowed(0, 1ul << bit, OP_EXEC);
2598 
2599 	/* Clear the bit. */
2600 	ept_allowed(1ul << bit, 0, OP_READ);
2601 	ept_allowed(1ul << bit, 0, OP_WRITE);
2602 	ept_allowed(1ul << bit, 0, OP_EXEC);
2603 }
2604 
2605 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2606 {
2607 	ept_allowed(EPT_PRESENT, access, op);
2608 }
2609 
2610 
2611 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2612 					     unsigned long clear,
2613 					     unsigned long set,
2614 					     enum ept_access_op op)
2615 {
2616 	struct ept_access_test_data *data = &ept_access_test_data;
2617 	unsigned long orig_pte;
2618 
2619 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2620 
2621 	do_ept_access_op(op);
2622 	assert_exit_reason(VMX_EPT_MISCONFIG);
2623 
2624 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2625 	#if 0
2626 	/* broken: */
2627 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2628 	#endif
2629 	#if 0
2630 	/*
2631 	 * broken:
2632 	 * According to description of exit qual for EPT violation,
2633 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2634 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2635 	 * to be set for msiconfig.
2636 	 */
2637 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2638 		       (unsigned long) (
2639 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2640 	#endif
2641 
2642 	/* Fix the violation and resume the op loop. */
2643 	ept_untwiddle(data->gpa, level, orig_pte);
2644 	enter_guest();
2645 	skip_exit_vmcall();
2646 }
2647 
2648 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2649 					  unsigned long clear,
2650 					  unsigned long set)
2651 {
2652 	/* The op shouldn't matter (read, write, exec), so try them all! */
2653 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2654 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2655 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2656 }
2657 
2658 static void ept_misconfig_at_level(int level, unsigned long clear,
2659 				   unsigned long set)
2660 {
2661 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2662 	if (ept_huge_pages_supported(level))
2663 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2664 }
2665 
2666 static void ept_misconfig(unsigned long clear, unsigned long set)
2667 {
2668 	ept_misconfig_at_level(1, clear, set);
2669 	ept_misconfig_at_level(2, clear, set);
2670 	ept_misconfig_at_level(3, clear, set);
2671 	ept_misconfig_at_level(4, clear, set);
2672 }
2673 
2674 static void ept_access_misconfig(unsigned long access)
2675 {
2676 	ept_misconfig(EPT_PRESENT, access);
2677 }
2678 
2679 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2680 {
2681 	/* Setting the bit causes a misconfig. */
2682 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2683 
2684 	/* Making the entry non-present turns reserved bits into ignored. */
2685 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2686 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2687 }
2688 
2689 static void ept_reserved_bit_at_level_huge(int level, int bit)
2690 {
2691 	/* Setting the bit causes a misconfig. */
2692 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2693 
2694 	/* Making the entry non-present turns reserved bits into ignored. */
2695 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2696 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2697 }
2698 
2699 static void ept_reserved_bit_at_level(int level, int bit)
2700 {
2701 	/* Setting the bit causes a misconfig. */
2702 	ept_misconfig_at_level(level, 0, 1ul << bit);
2703 
2704 	/* Making the entry non-present turns reserved bits into ignored. */
2705 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2706 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2707 }
2708 
2709 static void ept_reserved_bit(int bit)
2710 {
2711 	ept_reserved_bit_at_level(1, bit);
2712 	ept_reserved_bit_at_level(2, bit);
2713 	ept_reserved_bit_at_level(3, bit);
2714 	ept_reserved_bit_at_level(4, bit);
2715 }
2716 
2717 #define PAGE_2M_ORDER 9
2718 #define PAGE_1G_ORDER 18
2719 
2720 static void *get_1g_page(void)
2721 {
2722 	static void *alloc;
2723 
2724 	if (!alloc)
2725 		alloc = alloc_pages(PAGE_1G_ORDER);
2726 	return alloc;
2727 }
2728 
2729 static void ept_access_test_teardown(void *unused)
2730 {
2731 	/* Exit the guest cleanly. */
2732 	do_ept_access_op(OP_EXIT);
2733 }
2734 
2735 static void ept_access_test_guest(void)
2736 {
2737 	struct ept_access_test_data *data = &ept_access_test_data;
2738 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2739 
2740 	while (true) {
2741 		switch (data->op) {
2742 		case OP_READ:
2743 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2744 			break;
2745 		case OP_WRITE:
2746 			*data->gva = MAGIC_VAL_2;
2747 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2748 			*data->gva = MAGIC_VAL_1;
2749 			break;
2750 		case OP_EXEC:
2751 			TEST_ASSERT_EQ(42, code());
2752 			break;
2753 		case OP_FLUSH_TLB:
2754 			write_cr3(read_cr3());
2755 			break;
2756 		case OP_EXIT:
2757 			return;
2758 		default:
2759 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2760 		}
2761 		vmcall();
2762 	}
2763 }
2764 
2765 static void ept_access_test_setup(void)
2766 {
2767 	struct ept_access_test_data *data = &ept_access_test_data;
2768 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2769 	unsigned long size = npages * PAGE_SIZE;
2770 	unsigned long *page_table = current_page_table();
2771 	unsigned long pte;
2772 
2773 	if (setup_ept(false))
2774 		test_skip("EPT not supported");
2775 
2776 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2777 	if (cpuid_maxphyaddr() < 40)
2778 		test_skip("Test needs MAXPHYADDR >= 40");
2779 
2780 	test_set_guest(ept_access_test_guest);
2781 	test_add_teardown(ept_access_test_teardown, NULL);
2782 
2783 	data->hva = get_1g_page();
2784 	TEST_ASSERT(data->hva);
2785 	data->hpa = virt_to_phys(data->hva);
2786 
2787 	data->gpa = 1ul << 39;
2788 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2789 				   size);
2790 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2791 	install_pages(page_table, data->gpa, size, data->gva);
2792 
2793 	/*
2794 	 * Make sure nothing's mapped here so the tests that screw with the
2795 	 * pml4 entry don't inadvertently break something.
2796 	 */
2797 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2798 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2799 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2800 
2801 	data->hva[0] = MAGIC_VAL_1;
2802 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2803 }
2804 
2805 static void ept_access_test_not_present(void)
2806 {
2807 	ept_access_test_setup();
2808 	/* --- */
2809 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2810 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2811 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2812 }
2813 
2814 static void ept_access_test_read_only(void)
2815 {
2816 	ept_access_test_setup();
2817 
2818 	/* r-- */
2819 	ept_access_allowed(EPT_RA, OP_READ);
2820 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2821 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2822 }
2823 
2824 static void ept_access_test_write_only(void)
2825 {
2826 	ept_access_test_setup();
2827 	/* -w- */
2828 	ept_access_misconfig(EPT_WA);
2829 }
2830 
2831 static void ept_access_test_read_write(void)
2832 {
2833 	ept_access_test_setup();
2834 	/* rw- */
2835 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2836 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2837 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2838 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2839 }
2840 
2841 
2842 static void ept_access_test_execute_only(void)
2843 {
2844 	ept_access_test_setup();
2845 	/* --x */
2846 	if (ept_execute_only_supported()) {
2847 		ept_access_violation(EPT_EA, OP_READ,
2848 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2849 		ept_access_violation(EPT_EA, OP_WRITE,
2850 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2851 		ept_access_allowed(EPT_EA, OP_EXEC);
2852 	} else {
2853 		ept_access_misconfig(EPT_EA);
2854 	}
2855 }
2856 
2857 static void ept_access_test_read_execute(void)
2858 {
2859 	ept_access_test_setup();
2860 	/* r-x */
2861 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2862 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2863 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2864 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2865 }
2866 
2867 static void ept_access_test_write_execute(void)
2868 {
2869 	ept_access_test_setup();
2870 	/* -wx */
2871 	ept_access_misconfig(EPT_WA | EPT_EA);
2872 }
2873 
2874 static void ept_access_test_read_write_execute(void)
2875 {
2876 	ept_access_test_setup();
2877 	/* rwx */
2878 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2879 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2880 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2881 }
2882 
2883 static void ept_access_test_reserved_bits(void)
2884 {
2885 	int i;
2886 	int maxphyaddr;
2887 
2888 	ept_access_test_setup();
2889 
2890 	/* Reserved bits above maxphyaddr. */
2891 	maxphyaddr = cpuid_maxphyaddr();
2892 	for (i = maxphyaddr; i <= 51; i++) {
2893 		report_prefix_pushf("reserved_bit=%d", i);
2894 		ept_reserved_bit(i);
2895 		report_prefix_pop();
2896 	}
2897 
2898 	/* Level-specific reserved bits. */
2899 	ept_reserved_bit_at_level_nohuge(2, 3);
2900 	ept_reserved_bit_at_level_nohuge(2, 4);
2901 	ept_reserved_bit_at_level_nohuge(2, 5);
2902 	ept_reserved_bit_at_level_nohuge(2, 6);
2903 	/* 2M alignment. */
2904 	for (i = 12; i < 20; i++) {
2905 		report_prefix_pushf("reserved_bit=%d", i);
2906 		ept_reserved_bit_at_level_huge(2, i);
2907 		report_prefix_pop();
2908 	}
2909 	ept_reserved_bit_at_level_nohuge(3, 3);
2910 	ept_reserved_bit_at_level_nohuge(3, 4);
2911 	ept_reserved_bit_at_level_nohuge(3, 5);
2912 	ept_reserved_bit_at_level_nohuge(3, 6);
2913 	/* 1G alignment. */
2914 	for (i = 12; i < 29; i++) {
2915 		report_prefix_pushf("reserved_bit=%d", i);
2916 		ept_reserved_bit_at_level_huge(3, i);
2917 		report_prefix_pop();
2918 	}
2919 	ept_reserved_bit_at_level(4, 3);
2920 	ept_reserved_bit_at_level(4, 4);
2921 	ept_reserved_bit_at_level(4, 5);
2922 	ept_reserved_bit_at_level(4, 6);
2923 	ept_reserved_bit_at_level(4, 7);
2924 }
2925 
2926 static void ept_access_test_ignored_bits(void)
2927 {
2928 	ept_access_test_setup();
2929 	/*
2930 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2931 	 * far as translation is concerned even if AD bits are enabled in the
2932 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2933 	 * control is 0.
2934 	 */
2935 	ept_ignored_bit(8);
2936 	ept_ignored_bit(9);
2937 	ept_ignored_bit(10);
2938 	ept_ignored_bit(11);
2939 	ept_ignored_bit(52);
2940 	ept_ignored_bit(53);
2941 	ept_ignored_bit(54);
2942 	ept_ignored_bit(55);
2943 	ept_ignored_bit(56);
2944 	ept_ignored_bit(57);
2945 	ept_ignored_bit(58);
2946 	ept_ignored_bit(59);
2947 	ept_ignored_bit(60);
2948 	ept_ignored_bit(61);
2949 	ept_ignored_bit(62);
2950 	ept_ignored_bit(63);
2951 }
2952 
2953 static void ept_access_test_paddr_not_present_ad_disabled(void)
2954 {
2955 	ept_access_test_setup();
2956 	ept_disable_ad_bits();
2957 
2958 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
2959 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
2960 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
2961 }
2962 
2963 static void ept_access_test_paddr_not_present_ad_enabled(void)
2964 {
2965 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
2966 
2967 	ept_access_test_setup();
2968 	ept_enable_ad_bits_or_skip_test();
2969 
2970 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
2971 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
2972 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
2973 }
2974 
2975 static void ept_access_test_paddr_read_only_ad_disabled(void)
2976 {
2977 	/*
2978 	 * When EPT AD bits are disabled, all accesses to guest paging
2979 	 * structures are reported separately as a read and (after
2980 	 * translation of the GPA to host physical address) a read+write
2981 	 * if the A/D bits have to be set.
2982 	 */
2983 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2984 
2985 	ept_access_test_setup();
2986 	ept_disable_ad_bits();
2987 
2988 	/* Can't update A bit, so all accesses fail. */
2989 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2990 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
2991 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
2992 	/* AD bits disabled, so only writes try to update the D bit. */
2993 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
2994 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
2995 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
2996 	/* Both A and D already set, so read-only is OK. */
2997 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
2998 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
2999 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
3000 }
3001 
3002 static void ept_access_test_paddr_read_only_ad_enabled(void)
3003 {
3004 	/*
3005 	 * When EPT AD bits are enabled, all accesses to guest paging
3006 	 * structures are considered writes as far as EPT translation
3007 	 * is concerned.
3008 	 */
3009 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
3010 
3011 	ept_access_test_setup();
3012 	ept_enable_ad_bits_or_skip_test();
3013 
3014 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
3015 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
3016 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
3017 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
3018 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
3019 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
3020 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
3021 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
3022 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
3023 }
3024 
3025 static void ept_access_test_paddr_read_write(void)
3026 {
3027 	ept_access_test_setup();
3028 	/* Read-write access to paging structure. */
3029 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
3030 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
3031 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
3032 }
3033 
3034 static void ept_access_test_paddr_read_write_execute(void)
3035 {
3036 	ept_access_test_setup();
3037 	/* RWX access to paging structure. */
3038 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
3039 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
3040 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
3041 }
3042 
3043 static void ept_access_test_paddr_read_execute_ad_disabled(void)
3044 {
3045   	/*
3046 	 * When EPT AD bits are disabled, all accesses to guest paging
3047 	 * structures are reported separately as a read and (after
3048 	 * translation of the GPA to host physical address) a read+write
3049 	 * if the A/D bits have to be set.
3050 	 */
3051 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3052 
3053 	ept_access_test_setup();
3054 	ept_disable_ad_bits();
3055 
3056 	/* Can't update A bit, so all accesses fail. */
3057 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3058 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3059 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3060 	/* AD bits disabled, so only writes try to update the D bit. */
3061 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
3062 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3063 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3064 	/* Both A and D already set, so read-only is OK. */
3065 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3066 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3067 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3068 }
3069 
3070 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3071 {
3072 	/*
3073 	 * When EPT AD bits are enabled, all accesses to guest paging
3074 	 * structures are considered writes as far as EPT translation
3075 	 * is concerned.
3076 	 */
3077 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3078 
3079 	ept_access_test_setup();
3080 	ept_enable_ad_bits_or_skip_test();
3081 
3082 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3083 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3084 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3085 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3086 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3087 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3088 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3089 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3090 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3091 }
3092 
3093 static void ept_access_test_paddr_not_present_page_fault(void)
3094 {
3095 	ept_access_test_setup();
3096 	/*
3097 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3098 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3099 	 * Thus guest page fault before host takes EPT violation for trying to
3100 	 * update A bit.
3101 	 */
3102 }
3103 
3104 static void ept_access_test_force_2m_page(void)
3105 {
3106 	ept_access_test_setup();
3107 
3108 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3109 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3110 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3111 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3112 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3113 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3114 }
3115 
3116 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3117 {
3118 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3119 
3120 	TEST_ASSERT(msr & VPID_CAP_INVVPID);
3121 
3122 	if (type < INVVPID_ADDR || type > INVVPID_CONTEXT_LOCAL)
3123 		return false;
3124 
3125 	if (!(msr & (1ull << (type + VPID_CAP_INVVPID_TYPES_SHIFT))))
3126 		return false;
3127 
3128 	if (vpid >> 16)
3129 		return false;
3130 
3131 	if (type != INVVPID_ALL && !vpid)
3132 		return false;
3133 
3134 	if (type == INVVPID_ADDR && !is_canonical(gla))
3135 		return false;
3136 
3137 	return true;
3138 }
3139 
3140 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3141 {
3142 	int rc;
3143 	bool valid = invvpid_valid(type, vpid, gla);
3144 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3145 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3146 	/*
3147 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3148 	 * that we can tell if it is updated by INVVPID.
3149 	 */
3150 	vmcs_read(~0);
3151 	rc = invvpid(type, vpid, gla);
3152 	report("INVVPID type %ld VPID %lx GLA %lx %s",
3153 	       !rc == valid, type, vpid, gla,
3154 	       valid ? "passes" : "fails");
3155 	report("After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3156 	       vmcs_read(VMX_INST_ERROR) == expected,
3157 	       rc ? "failed" : "successful",
3158 	       expected, vmcs_read(VMX_INST_ERROR));
3159 }
3160 
3161 static void ds_invvpid(void *data)
3162 {
3163 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3164 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3165 
3166 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3167 	asm volatile("invvpid %0, %1"
3168 		     :
3169 		     : "m"(*(struct invvpid_operand *)data),
3170 		       "r"(type));
3171 }
3172 
3173 /*
3174  * The SS override is ignored in 64-bit mode, so we use an addressing
3175  * mode with %rsp as the base register to generate an implicit SS
3176  * reference.
3177  */
3178 static void ss_invvpid(void *data)
3179 {
3180 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3181 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3182 
3183 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3184 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3185 		     : "+r"(data)
3186 		     : "r"(type));
3187 }
3188 
3189 static void invvpid_test_gp(void)
3190 {
3191 	bool fault;
3192 
3193 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3194 				   (void *)NONCANONICAL);
3195 	report("INVVPID with non-canonical DS operand raises #GP", fault);
3196 }
3197 
3198 static void invvpid_test_ss(void)
3199 {
3200 	bool fault;
3201 
3202 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3203 				   (void *)NONCANONICAL);
3204 	report("INVVPID with non-canonical SS operand raises #SS", fault);
3205 }
3206 
3207 static void invvpid_test_pf(void)
3208 {
3209 	void *vpage = alloc_vpage();
3210 	bool fault;
3211 
3212 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3213 	report("INVVPID with unmapped operand raises #PF", fault);
3214 }
3215 
3216 static void try_compat_invvpid(void *unused)
3217 {
3218 	struct far_pointer32 fp = {
3219 		.offset = (uintptr_t)&&invvpid,
3220 		.selector = KERNEL_CS32,
3221 	};
3222 	register uintptr_t rsp asm("rsp");
3223 
3224 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3225 			"Code address too high.");
3226 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3227 
3228 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3229 	return;
3230 invvpid:
3231 	asm volatile (".code32;"
3232 		      "invvpid (%eax), %eax;"
3233 		      "lret;"
3234 		      ".code64");
3235 	__builtin_unreachable();
3236 }
3237 
3238 static void invvpid_test_compatibility_mode(void)
3239 {
3240 	bool fault;
3241 
3242 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3243 	report("Compatibility mode INVVPID raises #UD", fault);
3244 }
3245 
3246 static void invvpid_test_not_in_vmx_operation(void)
3247 {
3248 	bool fault;
3249 
3250 	TEST_ASSERT(!vmx_off());
3251 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3252 	report("INVVPID outside of VMX operation raises #UD", fault);
3253 	TEST_ASSERT(!vmx_on());
3254 }
3255 
3256 /*
3257  * This does not test real-address mode, virtual-8086 mode, protected mode,
3258  * or CPL > 0.
3259  */
3260 static void invvpid_test_v2(void)
3261 {
3262 	u64 msr;
3263 	int i;
3264 	unsigned types = 0;
3265 	unsigned type;
3266 
3267 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
3268 	    !(ctrl_cpu_rev[1].clr & CPU_VPID))
3269 		test_skip("VPID not supported");
3270 
3271 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3272 
3273 	if (!(msr & VPID_CAP_INVVPID))
3274 		test_skip("INVVPID not supported.\n");
3275 
3276 	if (msr & VPID_CAP_INVVPID_ADDR)
3277 		types |= 1u << INVVPID_ADDR;
3278 	if (msr & VPID_CAP_INVVPID_CXTGLB)
3279 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3280 	if (msr & VPID_CAP_INVVPID_ALL)
3281 		types |= 1u << INVVPID_ALL;
3282 	if (msr & VPID_CAP_INVVPID_CXTLOC)
3283 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3284 
3285 	if (!types)
3286 		test_skip("No INVVPID types supported.\n");
3287 
3288 	for (i = -127; i < 128; i++)
3289 		try_invvpid(i, 0xffff, 0);
3290 
3291 	/*
3292 	 * VPID must not be more than 16 bits.
3293 	 */
3294 	for (i = 0; i < 64; i++)
3295 		for (type = 0; type < 4; type++)
3296 			if (types & (1u << type))
3297 				try_invvpid(type, 1ul << i, 0);
3298 
3299 	/*
3300 	 * VPID must not be zero, except for "all contexts."
3301 	 */
3302 	for (type = 0; type < 4; type++)
3303 		if (types & (1u << type))
3304 			try_invvpid(type, 0, 0);
3305 
3306 	/*
3307 	 * The gla operand is only validated for single-address INVVPID.
3308 	 */
3309 	if (types & (1u << INVVPID_ADDR))
3310 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3311 
3312 	invvpid_test_gp();
3313 	invvpid_test_ss();
3314 	invvpid_test_pf();
3315 	invvpid_test_compatibility_mode();
3316 	invvpid_test_not_in_vmx_operation();
3317 }
3318 
3319 /*
3320  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3321  * at least as far as the guest-state checks. Returns false if the
3322  * VMLAUNCH fails early and execution falls through to the next
3323  * instruction.
3324  */
3325 static bool vmlaunch_succeeds(void)
3326 {
3327 	u32 exit_reason;
3328 
3329 	/*
3330 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3331 	 * unsupported VMCS component"). The caller can then check
3332 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3333 	 */
3334 	vmcs_write(~0u, 0);
3335 
3336 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3337 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3338 				   :
3339 				   : "r" ((u64)HOST_RSP)
3340 				   : "cc", "memory"
3341 				   : success);
3342 	return false;
3343 success:
3344 	exit_reason = vmcs_read(EXI_REASON);
3345 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3346 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3347 	return true;
3348 }
3349 
3350 /*
3351  * Try to launch the current VMCS.
3352  */
3353 static void test_vmx_vmlaunch(u32 xerror, bool xfail)
3354 {
3355 	bool success = vmlaunch_succeeds();
3356 	u32 vmx_inst_err;
3357 
3358 	report_xfail("vmlaunch %s", xfail, success == !xerror,
3359 		     !xerror ? "succeeds" : "fails");
3360 	if (!success && xerror) {
3361 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3362 		report("VMX inst error is %d (actual %d)",
3363 		       vmx_inst_err == xerror, xerror, vmx_inst_err);
3364 	}
3365 }
3366 
3367 static void test_vmx_invalid_controls(bool xfail)
3368 {
3369 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_CONTROL_FIELD, xfail);
3370 }
3371 
3372 static void test_vmx_valid_controls(bool xfail)
3373 {
3374 	test_vmx_vmlaunch(0, xfail);
3375 }
3376 
3377 /*
3378  * Test a particular value of a VM-execution control bit, if the value
3379  * is required or if the value is zero.
3380  */
3381 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3382 				    enum Encoding encoding, unsigned bit,
3383 				    unsigned val)
3384 {
3385 	u32 mask = 1u << bit;
3386 	bool expected;
3387 	u32 controls;
3388 
3389 	if (msr.set & mask)
3390 		TEST_ASSERT(msr.clr & mask);
3391 
3392 	/*
3393 	 * We can't arbitrarily turn on a control bit, because it may
3394 	 * introduce dependencies on other VMCS fields. So, we only
3395 	 * test turning on bits that have a required setting.
3396 	 */
3397 	if (val && (msr.clr & mask) && !(msr.set & mask))
3398 		return;
3399 
3400 	report_prefix_pushf("%s %s bit %d",
3401 			    val ? "Set" : "Clear", name, bit);
3402 
3403 	controls = vmcs_read(encoding);
3404 	if (val) {
3405 		vmcs_write(encoding, msr.set | mask);
3406 		expected = (msr.clr & mask);
3407 	} else {
3408 		vmcs_write(encoding, msr.set & ~mask);
3409 		expected = !(msr.set & mask);
3410 	}
3411 	if (expected)
3412 		test_vmx_valid_controls(false);
3413 	else
3414 		test_vmx_invalid_controls(false);
3415 	vmcs_write(encoding, controls);
3416 	report_prefix_pop();
3417 }
3418 
3419 /*
3420  * Test reserved values of a VM-execution control bit, based on the
3421  * allowed bit settings from the corresponding VMX capability MSR.
3422  */
3423 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3424 			      enum Encoding encoding, unsigned bit)
3425 {
3426 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3427 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3428 }
3429 
3430 /*
3431  * Reserved bits in the pin-based VM-execution controls must be set
3432  * properly. Software may consult the VMX capability MSRs to determine
3433  * the proper settings.
3434  * [Intel SDM]
3435  */
3436 static void test_pin_based_ctls(void)
3437 {
3438 	unsigned bit;
3439 
3440 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3441 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3442 	for (bit = 0; bit < 32; bit++)
3443 		test_rsvd_ctl_bit("pin-based controls",
3444 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3445 }
3446 
3447 /*
3448  * Reserved bits in the primary processor-based VM-execution controls
3449  * must be set properly. Software may consult the VMX capability MSRs
3450  * to determine the proper settings.
3451  * [Intel SDM]
3452  */
3453 static void test_primary_processor_based_ctls(void)
3454 {
3455 	unsigned bit;
3456 
3457 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3458 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3459 	for (bit = 0; bit < 32; bit++)
3460 		test_rsvd_ctl_bit("primary processor-based controls",
3461 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3462 }
3463 
3464 /*
3465  * If the "activate secondary controls" primary processor-based
3466  * VM-execution control is 1, reserved bits in the secondary
3467  * processor-based VM-execution controls must be cleared. Software may
3468  * consult the VMX capability MSRs to determine which bits are
3469  * reserved.
3470  * If the "activate secondary controls" primary processor-based
3471  * VM-execution control is 0 (or if the processor does not support the
3472  * 1-setting of that control), no checks are performed on the
3473  * secondary processor-based VM-execution controls.
3474  * [Intel SDM]
3475  */
3476 static void test_secondary_processor_based_ctls(void)
3477 {
3478 	u32 primary;
3479 	u32 secondary;
3480 	unsigned bit;
3481 
3482 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3483 		return;
3484 
3485 	primary = vmcs_read(CPU_EXEC_CTRL0);
3486 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3487 
3488 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3489 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3490 	for (bit = 0; bit < 32; bit++)
3491 		test_rsvd_ctl_bit("secondary processor-based controls",
3492 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3493 
3494 	/*
3495 	 * When the "activate secondary controls" VM-execution control
3496 	 * is clear, there are no checks on the secondary controls.
3497 	 */
3498 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3499 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3500 	report("Secondary processor-based controls ignored",
3501 	       vmlaunch_succeeds());
3502 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3503 	vmcs_write(CPU_EXEC_CTRL0, primary);
3504 }
3505 
3506 static void try_cr3_target_count(unsigned i, unsigned max)
3507 {
3508 	report_prefix_pushf("CR3 target count 0x%x", i);
3509 	vmcs_write(CR3_TARGET_COUNT, i);
3510 	if (i <= max)
3511 		test_vmx_valid_controls(false);
3512 	else
3513 		test_vmx_invalid_controls(false);
3514 	report_prefix_pop();
3515 }
3516 
3517 /*
3518  * The CR3-target count must not be greater than 4. Future processors
3519  * may support a different number of CR3-target values. Software
3520  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3521  * number of values supported.
3522  * [Intel SDM]
3523  */
3524 static void test_cr3_targets(void)
3525 {
3526 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3527 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3528 	unsigned i;
3529 
3530 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3531 	TEST_ASSERT(supported_targets <= 256);
3532 
3533 	try_cr3_target_count(-1u, supported_targets);
3534 	try_cr3_target_count(0x80000000, supported_targets);
3535 	try_cr3_target_count(0x7fffffff, supported_targets);
3536 	for (i = 0; i <= supported_targets + 1; i++)
3537 		try_cr3_target_count(i, supported_targets);
3538 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3539 }
3540 
3541 /*
3542  * Test a particular address setting in the VMCS
3543  */
3544 static void test_vmcs_addr(const char *name,
3545 			   enum Encoding encoding,
3546 			   u64 align,
3547 			   bool ignored,
3548 			   bool xfail_beyond_mapped_ram,
3549 			   u64 addr)
3550 {
3551 	bool xfail =
3552 		(xfail_beyond_mapped_ram &&
3553 		 addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3554 		 addr < (1ul << cpuid_maxphyaddr()));
3555 
3556 	report_prefix_pushf("%s = %lx", name, addr);
3557 	vmcs_write(encoding, addr);
3558 	if (ignored || (IS_ALIGNED(addr, align) &&
3559 	    addr < (1ul << cpuid_maxphyaddr())))
3560 		test_vmx_valid_controls(xfail);
3561 	else
3562 		test_vmx_invalid_controls(xfail);
3563 	report_prefix_pop();
3564 	xfail = false;
3565 }
3566 
3567 /*
3568  * Test interesting values for a VMCS address
3569  */
3570 static void test_vmcs_addr_values(const char *name,
3571 				  enum Encoding encoding,
3572 				  u64 align,
3573 				  bool ignored,
3574 				  bool xfail_beyond_mapped_ram,
3575 				  u32 bit_start, u32 bit_end)
3576 {
3577 	unsigned i;
3578 	u64 orig_val = vmcs_read(encoding);
3579 
3580 	for (i = bit_start; i <= bit_end; i++)
3581 		test_vmcs_addr(name, encoding, align, ignored,
3582 			       xfail_beyond_mapped_ram, 1ul << i);
3583 
3584 	test_vmcs_addr(name, encoding, align, ignored,
3585 		       xfail_beyond_mapped_ram, PAGE_SIZE - 1);
3586 	test_vmcs_addr(name, encoding, align, ignored,
3587 		       xfail_beyond_mapped_ram, PAGE_SIZE);
3588 	test_vmcs_addr(name, encoding, align, ignored,
3589 		       xfail_beyond_mapped_ram,
3590 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3591 	test_vmcs_addr(name, encoding, align, ignored,
3592 		       xfail_beyond_mapped_ram, -1ul);
3593 
3594 	vmcs_write(encoding, orig_val);
3595 }
3596 
3597 /*
3598  * Test a physical address reference in the VMCS, when the corresponding
3599  * feature is enabled and when the corresponding feature is disabled.
3600  */
3601 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3602 				     const char *field_name,
3603 				     const char *control_name, u64 align,
3604 				     bool xfail_beyond_mapped_ram,
3605 				     bool control_primary)
3606 {
3607 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3608 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3609 	u64 page_addr;
3610 
3611 	if (control_primary) {
3612 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3613 			return;
3614 	} else {
3615 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3616 			return;
3617 	}
3618 
3619 	page_addr = vmcs_read(field);
3620 
3621 	report_prefix_pushf("%s enabled", control_name);
3622 	if (control_primary) {
3623 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3624 	} else {
3625 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3626 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3627 	}
3628 
3629 	test_vmcs_addr_values(field_name, field, align, false,
3630 			      xfail_beyond_mapped_ram, 0, 63);
3631 	report_prefix_pop();
3632 
3633 	report_prefix_pushf("%s disabled", control_name);
3634 	if (control_primary) {
3635 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3636 	} else {
3637 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3638 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3639 	}
3640 
3641 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3642 	report_prefix_pop();
3643 
3644 	vmcs_write(field, page_addr);
3645 	vmcs_write(CPU_EXEC_CTRL0, primary);
3646 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3647 }
3648 
3649 /*
3650  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3651  * each I/O-bitmap address must be 0. Neither address should set any
3652  * bits beyond the processor's physical-address width.
3653  * [Intel SDM]
3654  */
3655 static void test_io_bitmaps(void)
3656 {
3657 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3658 				 "I/O bitmap A", "Use I/O bitmaps",
3659 				 PAGE_SIZE, false, true);
3660 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3661 				 "I/O bitmap B", "Use I/O bitmaps",
3662 				 PAGE_SIZE, false, true);
3663 }
3664 
3665 /*
3666  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3667  * the MSR-bitmap address must be 0. The address should not set any
3668  * bits beyond the processor's physical-address width.
3669  * [Intel SDM]
3670  */
3671 static void test_msr_bitmap(void)
3672 {
3673 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3674 				 "MSR bitmap", "Use MSR bitmaps",
3675 				 PAGE_SIZE, false, true);
3676 }
3677 
3678 /*
3679  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3680  * address must satisfy the following checks:
3681  * - Bits 11:0 of the address must be 0.
3682  * - The address should not set any bits beyond the processor's
3683  *   physical-address width.
3684  * [Intel SDM]
3685  */
3686 static void test_apic_virt_addr(void)
3687 {
3688 	/*
3689 	 * Ensure the processor will never use the virtual-APIC page, since
3690 	 * we will point it to invalid RAM.  Otherwise KVM is puzzled about
3691 	 * what we're trying to achieve and fails vmentry.
3692 	 */
3693 	u32 cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
3694 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0 | CPU_CR8_LOAD | CPU_CR8_STORE);
3695 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3696 				 "virtual-APIC address", "Use TPR shadow",
3697 				 PAGE_SIZE, false, true);
3698 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
3699 }
3700 
3701 /*
3702  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3703  * APIC-access address must satisfy the following checks:
3704  *  - Bits 11:0 of the address must be 0.
3705  *  - The address should not set any bits beyond the processor's
3706  *    physical-address width.
3707  * [Intel SDM]
3708  */
3709 static void test_apic_access_addr(void)
3710 {
3711 	void *apic_access_page = alloc_page();
3712 
3713 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3714 
3715 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3716 				 "APIC-access address",
3717 				 "virtualize APIC-accesses", PAGE_SIZE,
3718 				 false, false);
3719 }
3720 
3721 static bool set_bit_pattern(u8 mask, u32 *secondary)
3722 {
3723 	u8 i;
3724 	bool flag = false;
3725 	u32 test_bits[3] = {
3726 		CPU_VIRT_X2APIC,
3727 		CPU_APIC_REG_VIRT,
3728 		CPU_VINTD
3729 	};
3730 
3731         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3732 		if ((mask & (1u << i)) &&
3733 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3734 			*secondary |= test_bits[i];
3735 			flag = true;
3736 		}
3737 	}
3738 
3739 	return (flag);
3740 }
3741 
3742 /*
3743  * If the "use TPR shadow" VM-execution control is 0, the following
3744  * VM-execution controls must also be 0:
3745  * 	- virtualize x2APIC mode
3746  *	- APIC-register virtualization
3747  *	- virtual-interrupt delivery
3748  *    [Intel SDM]
3749  *
3750  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3751  *    "virtualize APIC accesses" VM-execution control must be 0.
3752  *    [Intel SDM]
3753  */
3754 static void test_apic_virtual_ctls(void)
3755 {
3756 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3757 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3758 	u32 primary = saved_primary;
3759 	u32 secondary = saved_secondary;
3760 	bool ctrl = false;
3761 	char str[10] = "disabled";
3762 	u8 i = 0, j;
3763 
3764 	/*
3765 	 * First test
3766 	 */
3767 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3768 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3769 		return;
3770 
3771 	primary |= CPU_SECONDARY;
3772 	primary &= ~CPU_TPR_SHADOW;
3773 	vmcs_write(CPU_EXEC_CTRL0, primary);
3774 
3775 	while (1) {
3776 		for (j = 1; j < 8; j++) {
3777 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3778 			if (primary & CPU_TPR_SHADOW) {
3779 				ctrl = true;
3780 			} else {
3781 				if (! set_bit_pattern(j, &secondary))
3782 					ctrl = true;
3783 				else
3784 					ctrl = false;
3785 			}
3786 
3787 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3788 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3789 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3790 			if (ctrl)
3791 				test_vmx_valid_controls(false);
3792 			else
3793 				test_vmx_invalid_controls(false);
3794 			report_prefix_pop();
3795 		}
3796 
3797 		if (i == 1)
3798 			break;
3799 		i++;
3800 
3801 		primary |= CPU_TPR_SHADOW;
3802 		vmcs_write(CPU_EXEC_CTRL0, primary);
3803 		strcpy(str, "enabled");
3804 	}
3805 
3806 	/*
3807 	 * Second test
3808 	 */
3809 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3810 
3811 	primary = saved_primary;
3812 	secondary = saved_secondary;
3813 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3814 		return;
3815 
3816 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3817 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3818 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3819 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3820 	test_vmx_valid_controls(false);
3821 	report_prefix_pop();
3822 
3823 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3824 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3825 	test_vmx_valid_controls(false);
3826 	report_prefix_pop();
3827 
3828 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3829 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3830 	test_vmx_invalid_controls(false);
3831 	report_prefix_pop();
3832 
3833 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3834 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3835 	test_vmx_valid_controls(false);
3836 	report_prefix_pop();
3837 
3838 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3839 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3840 }
3841 
3842 /*
3843  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3844  * "external-interrupt exiting" VM-execution control must be 1.
3845  * [Intel SDM]
3846  */
3847 static void test_virtual_intr_ctls(void)
3848 {
3849 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3850 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3851 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3852 	u32 primary = saved_primary;
3853 	u32 secondary = saved_secondary;
3854 	u32 pin = saved_pin;
3855 
3856 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3857 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
3858 		return;
3859 
3860 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3861 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
3862 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3863 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
3864 	test_vmx_valid_controls(false);
3865 	report_prefix_pop();
3866 
3867 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
3868 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3869 	test_vmx_invalid_controls(false);
3870 	report_prefix_pop();
3871 
3872 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
3873 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
3874 	test_vmx_valid_controls(false);
3875 	report_prefix_pop();
3876 
3877 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3878 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3879 	test_vmx_invalid_controls(false);
3880 	report_prefix_pop();
3881 
3882 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3883 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3884 	vmcs_write(PIN_CONTROLS, saved_pin);
3885 }
3886 
3887 static void test_pi_desc_addr(u64 addr, bool ctrl)
3888 {
3889 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
3890 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
3891 	if (ctrl)
3892 		test_vmx_valid_controls(false);
3893 	else
3894 		test_vmx_invalid_controls(false);
3895 	report_prefix_pop();
3896 }
3897 
3898 /*
3899  * If the “process posted interrupts†VM-execution control is 1, the
3900  * following must be true:
3901  *
3902  *	- The “virtual-interrupt delivery†VM-execution control is 1.
3903  *	- The “acknowledge interrupt on exit†VM-exit control is 1.
3904  *	- The posted-interrupt notification vector has a value in the
3905  *	- range 0–255 (bits 15:8 are all 0).
3906  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
3907  *	- The posted-interrupt descriptor address does not set any bits
3908  *	  beyond the processor's physical-address width.
3909  * [Intel SDM]
3910  */
3911 static void test_posted_intr(void)
3912 {
3913 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3914 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3915 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3916 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
3917 	u32 primary = saved_primary;
3918 	u32 secondary = saved_secondary;
3919 	u32 pin = saved_pin;
3920 	u32 exit_ctl = exit_ctl_saved;
3921 	u16 vec;
3922 	int i;
3923 
3924 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
3925 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3926 	    (ctrl_exit_rev.clr & EXI_INTA)))
3927 		return;
3928 
3929 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3930 
3931 	/*
3932 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
3933 	 */
3934 	pin |= PIN_POST_INTR;
3935 	vmcs_write(PIN_CONTROLS, pin);
3936 	secondary &= ~CPU_VINTD;
3937 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3938 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
3939 	test_vmx_invalid_controls(false);
3940 	report_prefix_pop();
3941 
3942 	secondary |= CPU_VINTD;
3943 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3944 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
3945 	test_vmx_invalid_controls(false);
3946 	report_prefix_pop();
3947 
3948 	exit_ctl &= ~EXI_INTA;
3949 	vmcs_write(EXI_CONTROLS, exit_ctl);
3950 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
3951 	test_vmx_invalid_controls(false);
3952 	report_prefix_pop();
3953 
3954 	exit_ctl |= EXI_INTA;
3955 	vmcs_write(EXI_CONTROLS, exit_ctl);
3956 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3957 	test_vmx_valid_controls(false);
3958 	report_prefix_pop();
3959 
3960 	secondary &= ~CPU_VINTD;
3961 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3962 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
3963 	test_vmx_invalid_controls(false);
3964 	report_prefix_pop();
3965 
3966 	secondary |= CPU_VINTD;
3967 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3968 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3969 	test_vmx_valid_controls(false);
3970 	report_prefix_pop();
3971 
3972 	/*
3973 	 * Test posted-interrupt notification vector
3974 	 */
3975 	for (i = 0; i < 8; i++) {
3976 		vec = (1ul << i);
3977 		vmcs_write(PINV, vec);
3978 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3979 		test_vmx_valid_controls(false);
3980 		report_prefix_pop();
3981 	}
3982 	for (i = 8; i < 16; i++) {
3983 		vec = (1ul << i);
3984 		vmcs_write(PINV, vec);
3985 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3986 		test_vmx_invalid_controls(false);
3987 		report_prefix_pop();
3988 	}
3989 
3990 	vec &= ~(0xff << 8);
3991 	vmcs_write(PINV, vec);
3992 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3993 	test_vmx_valid_controls(false);
3994 	report_prefix_pop();
3995 
3996 	/*
3997 	 * Test posted-interrupt descriptor addresss
3998 	 */
3999 	for (i = 0; i < 6; i++) {
4000 		test_pi_desc_addr(1ul << i, false);
4001 	}
4002 
4003 	test_pi_desc_addr(0xf0, false);
4004 	test_pi_desc_addr(0xff, false);
4005 	test_pi_desc_addr(0x0f, false);
4006 	test_pi_desc_addr(0x8000, true);
4007 	test_pi_desc_addr(0x00, true);
4008 	test_pi_desc_addr(0xc000, true);
4009 
4010 	test_vmcs_addr_values("process-posted interrupts",
4011 			       POSTED_INTR_DESC_ADDR, 64,
4012 			       false, false, 0, 63);
4013 
4014 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4015 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4016 	vmcs_write(PIN_CONTROLS, saved_pin);
4017 }
4018 
4019 static void test_apic_ctls(void)
4020 {
4021 	test_apic_virt_addr();
4022 	test_apic_access_addr();
4023 	test_apic_virtual_ctls();
4024 	test_virtual_intr_ctls();
4025 	test_posted_intr();
4026 }
4027 
4028 /*
4029  * If the “enable VPID†VM-execution control is 1, the value of the
4030  * of the VPID VM-execution control field must not be 0000H.
4031  * [Intel SDM]
4032  */
4033 static void test_vpid(void)
4034 {
4035 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
4036 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
4037 	u16 vpid = 0x0000;
4038 	int i;
4039 
4040 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4041 	    (ctrl_cpu_rev[1].clr & CPU_VPID))) {
4042 		test_skip("Secondary controls and/or VPID not supported");
4043 		return;
4044 	}
4045 
4046 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
4047 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
4048 	vmcs_write(VPID, vpid);
4049 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
4050 	test_vmx_valid_controls(false);
4051 	report_prefix_pop();
4052 
4053 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
4054 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4055 	test_vmx_invalid_controls(false);
4056 	report_prefix_pop();
4057 
4058 	for (i = 0; i < 16; i++) {
4059 		vpid = (short)1 << i;;
4060 		vmcs_write(VPID, vpid);
4061 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
4062 		test_vmx_valid_controls(false);
4063 		report_prefix_pop();
4064 	}
4065 
4066 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
4067 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
4068 }
4069 
4070 static void set_vtpr(unsigned vtpr)
4071 {
4072 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
4073 }
4074 
4075 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
4076 {
4077 	bool valid = true;
4078 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4079 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4080 
4081 	if ((primary & CPU_TPR_SHADOW) &&
4082 	    (!(primary & CPU_SECONDARY) ||
4083 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
4084 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
4085 
4086 	set_vtpr(vtpr);
4087 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
4088 	    threshold, (vtpr >> 4) & 0xf);
4089 	if (valid)
4090 		test_vmx_valid_controls(false);
4091 	else
4092 		test_vmx_invalid_controls(false);
4093 	report_prefix_pop();
4094 }
4095 
4096 static void test_invalid_event_injection(void)
4097 {
4098 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
4099 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
4100 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
4101 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
4102 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
4103 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4104 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4105 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4106 	u32 cnt;
4107 
4108 	/* Setup */
4109 	report_prefix_push("invalid event injection");
4110 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4111 	vmcs_write(ENT_INST_LEN, 0x00000001);
4112 
4113 	/* The field’s interruption type is not set to a reserved value. */
4114 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4115 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4116 			    "RESERVED interruption type invalid [-]",
4117 			    ent_intr_info);
4118 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4119 	test_vmx_invalid_controls(false);
4120 	report_prefix_pop();
4121 
4122 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4123 			DE_VECTOR;
4124 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4125 			    "RESERVED interruption type invalid [+]",
4126 			    ent_intr_info);
4127 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4128 	test_vmx_valid_controls(false);
4129 	report_prefix_pop();
4130 
4131 	/* If the interruption type is other event, the vector is 0. */
4132 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4133 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4134 			    "(OTHER EVENT && vector != 0) invalid [-]",
4135 			    ent_intr_info);
4136 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4137 	test_vmx_invalid_controls(false);
4138 	report_prefix_pop();
4139 
4140 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4141 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4142 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4143 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4144 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4145 	test_vmx_invalid_controls(false);
4146 	report_prefix_pop();
4147 
4148 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4149 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4150 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4151 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4152 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4153 	test_vmx_valid_controls(false);
4154 	report_prefix_pop();
4155 
4156 	/*
4157 	 * If the interruption type
4158 	 * is HW exception, the vector is at most 31.
4159 	 */
4160 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4161 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4162 			    "(HW exception && vector > 31) invalid [-]",
4163 			    ent_intr_info);
4164 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4165 	test_vmx_invalid_controls(false);
4166 	report_prefix_pop();
4167 
4168 	/*
4169 	 * deliver-error-code is 1 iff either
4170 	 * (a) the "unrestricted guest" VM-execution control is 0
4171 	 * (b) CR0.PE is set.
4172 	 */
4173 
4174 	/* Assert that unrestricted guest is disabled or unsupported */
4175 	assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
4176 	       !(secondary_save & CPU_URG));
4177 
4178 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4179 			GP_VECTOR;
4180 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4181 			    "error code <-> (!URG || prot_mode) [-]",
4182 			    ent_intr_info);
4183 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4184 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4185 	test_vmx_invalid_controls(false);
4186 	report_prefix_pop();
4187 
4188 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4189 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4190 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4191 			    "error code <-> (!URG || prot_mode) [+]",
4192 			    ent_intr_info);
4193 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4194 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4195 	test_vmx_valid_controls(false);
4196 	report_prefix_pop();
4197 
4198 	if (enable_unrestricted_guest())
4199 		goto skip_unrestricted_guest;
4200 
4201 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4202 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4203 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4204 			    "error code <-> (!URG || prot_mode) [-]",
4205 			    ent_intr_info);
4206 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4207 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4208 	test_vmx_invalid_controls(false);
4209 	report_prefix_pop();
4210 
4211 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4212 			GP_VECTOR;
4213 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4214 			    "error code <-> (!URG || prot_mode) [-]",
4215 			    ent_intr_info);
4216 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4217 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4218 	test_vmx_invalid_controls(false);
4219 	report_prefix_pop();
4220 
4221 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4222 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4223 
4224 skip_unrestricted_guest:
4225 	vmcs_write(GUEST_CR0, guest_cr0_save);
4226 
4227 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4228 	report_prefix_push("error code <-> HW exception");
4229 	for (cnt = 0; cnt < 8; cnt++) {
4230 		u32 exception_type_mask = cnt << 8;
4231 		u32 deliver_error_code_mask =
4232 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4233 			INTR_INFO_DELIVER_CODE_MASK : 0;
4234 
4235 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4236 				exception_type_mask | GP_VECTOR;
4237 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4238 				    ent_intr_info);
4239 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4240 		test_vmx_invalid_controls(false);
4241 		report_prefix_pop();
4242 	}
4243 	report_prefix_pop();
4244 
4245 	/*
4246 	 * deliver-error-code is 1 iff the the vector
4247 	 * indicates an exception that would normally deliver an error code
4248 	 */
4249 	report_prefix_push("error code <-> vector delivers error code");
4250 	for (cnt = 0; cnt < 32; cnt++) {
4251 		bool has_error_code = false;
4252 		u32 deliver_error_code_mask;
4253 
4254 		switch (cnt) {
4255 		case DF_VECTOR:
4256 		case TS_VECTOR:
4257 		case NP_VECTOR:
4258 		case SS_VECTOR:
4259 		case GP_VECTOR:
4260 		case PF_VECTOR:
4261 		case AC_VECTOR:
4262 			has_error_code = true;
4263 		}
4264 
4265 		/* Negative case */
4266 		deliver_error_code_mask = has_error_code ?
4267 						0 :
4268 						INTR_INFO_DELIVER_CODE_MASK;
4269 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4270 				INTR_TYPE_HARD_EXCEPTION | cnt;
4271 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4272 				    ent_intr_info);
4273 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4274 		test_vmx_invalid_controls(false);
4275 		report_prefix_pop();
4276 
4277 		/* Positive case */
4278 		deliver_error_code_mask = has_error_code ?
4279 						INTR_INFO_DELIVER_CODE_MASK :
4280 						0;
4281 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4282 				INTR_TYPE_HARD_EXCEPTION | cnt;
4283 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4284 				    ent_intr_info);
4285 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4286 		test_vmx_valid_controls(false);
4287 		report_prefix_pop();
4288 	}
4289 	report_prefix_pop();
4290 
4291 	/* Reserved bits in the field (30:12) are 0. */
4292 	report_prefix_push("reserved bits clear");
4293 	for (cnt = 12; cnt <= 30; cnt++) {
4294 		ent_intr_info = ent_intr_info_base |
4295 				INTR_INFO_DELIVER_CODE_MASK |
4296 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4297 				(1U << cnt);
4298 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4299 				    ent_intr_info);
4300 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4301 		test_vmx_invalid_controls(false);
4302 		report_prefix_pop();
4303 	}
4304 	report_prefix_pop();
4305 
4306 	/*
4307 	 * If deliver-error-code is 1
4308 	 * bits 31:15 of the VM-entry exception error-code field are 0.
4309 	 */
4310 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4311 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4312 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4313 			    "VM-entry exception error code[31:15] clear",
4314 			    ent_intr_info);
4315 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4316 	for (cnt = 15; cnt <= 31; cnt++) {
4317 		ent_intr_err = 1U << cnt;
4318 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4319 				    ent_intr_err);
4320 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4321 		test_vmx_invalid_controls(false);
4322 		report_prefix_pop();
4323 	}
4324 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4325 	report_prefix_pop();
4326 
4327 	/*
4328 	 * If the interruption type is software interrupt, software exception,
4329 	 * or privileged software exception, the VM-entry instruction-length
4330 	 * field is in the range 0–15.
4331 	 */
4332 
4333 	for (cnt = 0; cnt < 3; cnt++) {
4334 		switch (cnt) {
4335 		case 0:
4336 			ent_intr_info = ent_intr_info_base |
4337 					INTR_TYPE_SOFT_INTR;
4338 			break;
4339 		case 1:
4340 			ent_intr_info = ent_intr_info_base |
4341 					INTR_TYPE_SOFT_EXCEPTION;
4342 			break;
4343 		case 2:
4344 			ent_intr_info = ent_intr_info_base |
4345 					INTR_TYPE_PRIV_SW_EXCEPTION;
4346 			break;
4347 		}
4348 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4349 				    "VM-entry instruction-length check",
4350 				    ent_intr_info);
4351 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4352 
4353 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4354 		ent_intr_len = -1;
4355 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4356 				    ent_intr_len);
4357 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4358 		test_vmx_invalid_controls(false);
4359 		report_prefix_pop();
4360 
4361 		/* Instruction length set to 16 should fail */
4362 		ent_intr_len = 0x00000010;
4363 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4364 				    ent_intr_len);
4365 		vmcs_write(ENT_INST_LEN, 0x00000010);
4366 		test_vmx_invalid_controls(false);
4367 		report_prefix_pop();
4368 
4369 		report_prefix_pop();
4370 	}
4371 
4372 	/* Cleanup */
4373 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4374 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4375 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4376 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4377 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4378 	vmcs_write(GUEST_CR0, guest_cr0_save);
4379 	report_prefix_pop();
4380 }
4381 
4382 /*
4383  * Test interesting vTPR values for a given TPR threshold.
4384  */
4385 static void test_vtpr_values(unsigned threshold)
4386 {
4387 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4388 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4389 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4390 }
4391 
4392 static void try_tpr_threshold(unsigned threshold)
4393 {
4394 	bool valid = true;
4395 
4396 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4397 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4398 
4399 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4400 	    (secondary & CPU_VINTD)))
4401 		valid = !(threshold >> 4);
4402 
4403 	set_vtpr(-1);
4404 	vmcs_write(TPR_THRESHOLD, threshold);
4405 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4406 	if (valid)
4407 		test_vmx_valid_controls(false);
4408 	else
4409 		test_vmx_invalid_controls(false);
4410 	report_prefix_pop();
4411 
4412 	if (valid)
4413 		test_vtpr_values(threshold);
4414 }
4415 
4416 /*
4417  * Test interesting TPR threshold values.
4418  */
4419 static void test_tpr_threshold_values(void)
4420 {
4421 	unsigned i;
4422 
4423 	for (i = 0; i < 0x10; i++)
4424 		try_tpr_threshold(i);
4425 	for (i = 4; i < 32; i++)
4426 		try_tpr_threshold(1u << i);
4427 	try_tpr_threshold(-1u);
4428 	try_tpr_threshold(0x7fffffff);
4429 }
4430 
4431 /*
4432  * This test covers the following two VM entry checks:
4433  *
4434  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4435  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4436  *         31:4 of the TPR threshold VM-execution control field must
4437 	   be 0.
4438  *         [Intel SDM]
4439  *
4440  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4441  *          "virtual-interrupt delivery" VM-execution control is 0
4442  *          and the "virtualize APIC accesses" VM-execution control
4443  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4444  *          control field must not be greater than the value of bits
4445  *          7:4 of VTPR.
4446  *          [Intel SDM]
4447  */
4448 static void test_tpr_threshold(void)
4449 {
4450 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4451 	u64 apic_virt_addr = vmcs_read(APIC_VIRT_ADDR);
4452 	u64 threshold = vmcs_read(TPR_THRESHOLD);
4453 	void *virtual_apic_page;
4454 
4455 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4456 		return;
4457 
4458 	virtual_apic_page = alloc_page();
4459 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4460 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4461 
4462 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4463 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4464 	test_tpr_threshold_values();
4465 	report_prefix_pop();
4466 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4467 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4468 	test_tpr_threshold_values();
4469 	report_prefix_pop();
4470 
4471 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4472 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES))))
4473 		goto out;
4474 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4475 
4476 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4477 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4478 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4479 		test_tpr_threshold_values();
4480 		report_prefix_pop();
4481 
4482 		vmcs_write(CPU_EXEC_CTRL0,
4483 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4484 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4485 		test_tpr_threshold_values();
4486 		report_prefix_pop();
4487 	}
4488 
4489 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4490 		vmcs_write(CPU_EXEC_CTRL0,
4491 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4492 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4493 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4494 		test_tpr_threshold_values();
4495 		report_prefix_pop();
4496 
4497 		vmcs_write(CPU_EXEC_CTRL0,
4498 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4499 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4500 		test_tpr_threshold_values();
4501 		report_prefix_pop();
4502 	}
4503 
4504 	if ((ctrl_cpu_rev[1].clr &
4505 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4506 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4507 		vmcs_write(CPU_EXEC_CTRL0,
4508 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4509 		vmcs_write(CPU_EXEC_CTRL1,
4510 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4511 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4512 		test_tpr_threshold_values();
4513 		report_prefix_pop();
4514 
4515 		vmcs_write(CPU_EXEC_CTRL0,
4516 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4517 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4518 		test_tpr_threshold_values();
4519 		report_prefix_pop();
4520 	}
4521 
4522 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4523 out:
4524 	vmcs_write(TPR_THRESHOLD, threshold);
4525 	vmcs_write(APIC_VIRT_ADDR, apic_virt_addr);
4526 	vmcs_write(CPU_EXEC_CTRL0, primary);
4527 }
4528 
4529 /*
4530  * This test verifies the following two vmentry checks:
4531  *
4532  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4533  *  VM-execution control must be 0.
4534  *  [Intel SDM]
4535  *
4536  *  If the “virtual NMIs” VM-execution control is 0, the “NMI-window
4537  *  exiting” VM-execution control must be 0.
4538  *  [Intel SDM]
4539  */
4540 static void test_nmi_ctrls(void)
4541 {
4542 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4543 
4544 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4545 	    (PIN_NMI | PIN_VIRT_NMI)) {
4546 		test_skip("NMI exiting and Virtual NMIs are not supported !");
4547 		return;
4548 	}
4549 
4550 	/* Save the controls so that we can restore them after our tests */
4551 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4552 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4553 
4554 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4555 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4556 
4557 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4558 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4559 	test_vmx_valid_controls(false);
4560 	report_prefix_pop();
4561 
4562 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4563 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4564 	test_vmx_invalid_controls(false);
4565 	report_prefix_pop();
4566 
4567 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4568 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4569 	test_vmx_valid_controls(false);
4570 	report_prefix_pop();
4571 
4572 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4573 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4574 	test_vmx_valid_controls(false);
4575 	report_prefix_pop();
4576 
4577 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4578 		report_info("NMI-window exiting is not supported, skipping...");
4579 		goto done;
4580 	}
4581 
4582 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4583 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4584 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4585 	test_vmx_invalid_controls(false);
4586 	report_prefix_pop();
4587 
4588 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4589 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4590 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4591 	test_vmx_valid_controls(false);
4592 	report_prefix_pop();
4593 
4594 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4595 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4596 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4597 	test_vmx_valid_controls(false);
4598 	report_prefix_pop();
4599 
4600 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4601 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4602 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4603 	test_vmx_valid_controls(false);
4604 	report_prefix_pop();
4605 
4606 	/* Restore the controls to their original values */
4607 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4608 done:
4609 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4610 }
4611 
4612 static void test_eptp_ad_bit(u64 eptp, bool ctrl)
4613 {
4614 	vmcs_write(EPTP, eptp);
4615 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4616 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4617 	if (ctrl)
4618 		test_vmx_valid_controls(false);
4619 	else
4620 		test_vmx_invalid_controls(false);
4621 	report_prefix_pop();
4622 
4623 }
4624 
4625 /*
4626  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4627  *    control field must satisfy the following checks:
4628  *
4629  *     - The EPT memory type (bits 2:0) must be a value supported by the
4630  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4631  *     - Bits 5:3 (1 less than the EPT page-walk length) must be 3,
4632  *	 indicating an EPT page-walk length of 4.
4633  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4634  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4635  *	 indicating that the processor does not support accessed and dirty
4636  *	 dirty flags for EPT.
4637  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4638  *	 physical-address width) must all be 0.
4639  *
4640  * 2. If the "unrestricted guest" VM-execution control is 1, the
4641  *    "enable EPT" VM-execution control must also be 1.
4642  */
4643 static void test_ept_eptp(void)
4644 {
4645 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4646 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4647 	u64 eptp_saved = vmcs_read(EPTP);
4648 	u32 primary = primary_saved;
4649 	u32 secondary = secondary_saved;
4650 	u64 msr, eptp = eptp_saved;
4651 	bool un_cache = false;
4652 	bool wr_bk = false;
4653 	bool ctrl;
4654 	u32 i, maxphysaddr;
4655 	u64 j, resv_bits_mask = 0;
4656 
4657 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4658 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4659 		test_skip("\"CPU secondary\" and/or \"enable EPT\" execution controls are not supported !");
4660 		return;
4661 	}
4662 
4663 	/*
4664 	 * Memory type (bits 2:0)
4665 	 */
4666 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
4667 	if (msr & EPT_CAP_UC)
4668 		un_cache = true;
4669 	if (msr & EPT_CAP_WB)
4670 		wr_bk = true;
4671 
4672 	primary |= CPU_SECONDARY;
4673 	vmcs_write(CPU_EXEC_CTRL0, primary);
4674 	secondary |= CPU_EPT;
4675 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4676 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4677 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4678 	vmcs_write(EPTP, eptp);
4679 
4680 	for (i = 0; i < 8; i++) {
4681 		if (i == 0) {
4682 			if (un_cache) {
4683 				report_info("EPT paging structure memory-type is Un-cacheable\n");
4684 				ctrl = true;
4685 			} else {
4686 				ctrl = false;
4687 			}
4688 		} else if (i == 6) {
4689 			if (wr_bk) {
4690 				report_info("EPT paging structure memory-type is Write-back\n");
4691 				ctrl = true;
4692 			} else {
4693 				ctrl = false;
4694 			}
4695 		} else {
4696 			ctrl = false;
4697 		}
4698 
4699 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4700 		vmcs_write(EPTP, eptp);
4701 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4702 		    eptp & EPT_MEM_TYPE_MASK);
4703 		if (ctrl)
4704 			test_vmx_valid_controls(false);
4705 		else
4706 			test_vmx_invalid_controls(false);
4707 		report_prefix_pop();
4708 	}
4709 
4710 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4711 
4712 	/*
4713 	 * Page walk length (bits 5:3)
4714 	 */
4715 	for (i = 0; i < 8; i++) {
4716 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4717 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4718 		if (i == 3)
4719 			ctrl = true;
4720 		else
4721 			ctrl = false;
4722 
4723 		vmcs_write(EPTP, eptp);
4724 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4725 		    eptp & EPTP_PG_WALK_LEN_MASK);
4726 		if (ctrl)
4727 			test_vmx_valid_controls(false);
4728 		else
4729 			test_vmx_invalid_controls(false);
4730 		report_prefix_pop();
4731 	}
4732 
4733 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4734 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4735 
4736 	/*
4737 	 * Accessed and dirty flag (bit 6)
4738 	 */
4739 	if (msr & EPT_CAP_AD_FLAG) {
4740 		report_info("Processor supports accessed and dirty flag");
4741 		eptp &= ~EPTP_AD_FLAG;
4742 		test_eptp_ad_bit(eptp, true);
4743 
4744 		eptp |= EPTP_AD_FLAG;
4745 		test_eptp_ad_bit(eptp, true);
4746 	} else {
4747 		report_info("Processor does not supports accessed and dirty flag");
4748 		eptp &= ~EPTP_AD_FLAG;
4749 		test_eptp_ad_bit(eptp, true);
4750 
4751 		eptp |= EPTP_AD_FLAG;
4752 		test_eptp_ad_bit(eptp, false);
4753 	}
4754 
4755 	/*
4756 	 * Reserved bits [11:7] and [63:N]
4757 	 */
4758 	for (i = 0; i < 32; i++) {
4759 		eptp = (eptp &
4760 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4761 		    (i << EPTP_RESERV_BITS_SHIFT);
4762 		vmcs_write(EPTP, eptp);
4763 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4764 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4765 		    EPTP_RESERV_BITS_MASK);
4766 		if (i == 0)
4767 			test_vmx_valid_controls(false);
4768 		else
4769 			test_vmx_invalid_controls(false);
4770 		report_prefix_pop();
4771 	}
4772 
4773 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4774 
4775 	maxphysaddr = cpuid_maxphyaddr();
4776 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4777 		resv_bits_mask |= 1ul << i;
4778 	}
4779 
4780 	for (j = maxphysaddr - 1; j <= 63; j++) {
4781 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4782 		    (j < maxphysaddr ? 0 : 1ul << j);
4783 		vmcs_write(EPTP, eptp);
4784 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4785 		    (eptp >> maxphysaddr) & resv_bits_mask);
4786 		if (j < maxphysaddr)
4787 			test_vmx_valid_controls(false);
4788 		else
4789 			test_vmx_invalid_controls(false);
4790 		report_prefix_pop();
4791 	}
4792 
4793 	secondary &= ~(CPU_EPT | CPU_URG);
4794 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4795 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4796 	test_vmx_valid_controls(false);
4797 	report_prefix_pop();
4798 
4799 	secondary |= CPU_URG;
4800 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4801 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4802 	test_vmx_invalid_controls(false);
4803 	report_prefix_pop();
4804 
4805 	secondary |= CPU_EPT;
4806 	setup_dummy_ept();
4807 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4808 	test_vmx_valid_controls(false);
4809 	report_prefix_pop();
4810 
4811 	secondary &= ~CPU_URG;
4812 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4813 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4814 	test_vmx_valid_controls(false);
4815 	report_prefix_pop();
4816 
4817 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4818 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4819 	vmcs_write(EPTP, eptp_saved);
4820 }
4821 
4822 /*
4823  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4824  * VM-execution control must also be 1. In addition, the PML address
4825  * must satisfy the following checks:
4826  *
4827  *    * Bits 11:0 of the address must be 0.
4828  *    * The address should not set any bits beyond the processor's
4829  *	physical-address width.
4830  *
4831  *  [Intel SDM]
4832  */
4833 static void test_pml(void)
4834 {
4835 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4836 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4837 	u32 primary = primary_saved;
4838 	u32 secondary = secondary_saved;
4839 
4840 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4841 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
4842 		test_skip("\"Secondary execution\" control or \"enable EPT\" control or \"enable PML\" control is not supported !");
4843 		return;
4844 	}
4845 
4846 	primary |= CPU_SECONDARY;
4847 	vmcs_write(CPU_EXEC_CTRL0, primary);
4848 	secondary &= ~(CPU_PML | CPU_EPT);
4849 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4850 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
4851 	test_vmx_valid_controls(false);
4852 	report_prefix_pop();
4853 
4854 	secondary |= CPU_PML;
4855 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4856 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
4857 	test_vmx_invalid_controls(false);
4858 	report_prefix_pop();
4859 
4860 	secondary |= CPU_EPT;
4861 	setup_dummy_ept();
4862 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
4863 	test_vmx_valid_controls(false);
4864 	report_prefix_pop();
4865 
4866 	secondary &= ~CPU_PML;
4867 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4868 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
4869 	test_vmx_valid_controls(false);
4870 	report_prefix_pop();
4871 
4872 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
4873 				 PAGE_SIZE, false, false);
4874 
4875 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4876 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4877 }
4878 
4879  /*
4880  * If the "activate VMX-preemption timer" VM-execution control is 0, the
4881  * the "save VMX-preemption timer value" VM-exit control must also be 0.
4882  *
4883  *  [Intel SDM]
4884  */
4885 static void test_vmx_preemption_timer(void)
4886 {
4887 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4888 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
4889 	u32 pin = saved_pin;
4890 	u32 exit = saved_exit;
4891 
4892 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
4893 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
4894 		printf("\"Save-VMX-preemption-timer\" control and/or \"Enable-VMX-preemption-timer\" control is not supported\n");
4895 		return;
4896 	}
4897 
4898 	pin |= PIN_PREEMPT;
4899 	vmcs_write(PIN_CONTROLS, pin);
4900 	exit &= ~EXI_SAVE_PREEMPT;
4901 	vmcs_write(EXI_CONTROLS, exit);
4902 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
4903 	test_vmx_valid_controls(false);
4904 	report_prefix_pop();
4905 
4906 	exit |= EXI_SAVE_PREEMPT;
4907 	vmcs_write(EXI_CONTROLS, exit);
4908 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
4909 	test_vmx_valid_controls(false);
4910 	report_prefix_pop();
4911 
4912 	pin &= ~PIN_PREEMPT;
4913 	vmcs_write(PIN_CONTROLS, pin);
4914 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
4915 	test_vmx_invalid_controls(false);
4916 	report_prefix_pop();
4917 
4918 	exit &= ~EXI_SAVE_PREEMPT;
4919 	vmcs_write(EXI_CONTROLS, exit);
4920 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
4921 	test_vmx_valid_controls(false);
4922 	report_prefix_pop();
4923 
4924 	vmcs_write(PIN_CONTROLS, saved_pin);
4925 	vmcs_write(EXI_CONTROLS, saved_exit);
4926 }
4927 
4928 /*
4929  * Tests for VM-execution control fields
4930  */
4931 static void test_vm_execution_ctls(void)
4932 {
4933 	test_pin_based_ctls();
4934 	test_primary_processor_based_ctls();
4935 	test_secondary_processor_based_ctls();
4936 	test_cr3_targets();
4937 	test_io_bitmaps();
4938 	test_msr_bitmap();
4939 	test_apic_ctls();
4940 	test_tpr_threshold();
4941 	test_nmi_ctrls();
4942 	test_pml();
4943 	test_vpid();
4944 	test_ept_eptp();
4945 	test_vmx_preemption_timer();
4946 }
4947 
4948  /*
4949   * The following checks are performed for the VM-entry MSR-load address if
4950   * the VM-entry MSR-load count field is non-zero:
4951   *
4952   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
4953   *      The address should not set any bits beyond the processor’s
4954   *      physical-address width.
4955   *
4956   *    - The address of the last byte in the VM-entry MSR-load area
4957   *      should not set any bits beyond the processor’s physical-address
4958   *      width. The address of this last byte is VM-entry MSR-load address
4959   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
4960   *      uses more bits than the processor’s physical-address width.)
4961   *
4962   *
4963   *  [Intel SDM]
4964   */
4965 static void test_entry_msr_load(void)
4966 {
4967 	entry_msr_load = alloc_page();
4968 	u64 tmp;
4969 	u32 entry_msr_ld_cnt = 1;
4970 	int i;
4971 	u32 addr_len = 64;
4972 
4973 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
4974 
4975 	/* Check first 4 bits of VM-entry MSR-load address */
4976 	for (i = 0; i < 4; i++) {
4977 		tmp = (u64)entry_msr_load | 1ull << i;
4978 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
4979 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
4980 				    tmp & 0xf);
4981 		test_vmx_invalid_controls(false);
4982 		report_prefix_pop();
4983 	}
4984 
4985 	if (basic.val & (1ul << 48))
4986 		addr_len = 32;
4987 
4988 	test_vmcs_addr_values("VM-entry-MSR-load address",
4989 				ENTER_MSR_LD_ADDR, 16, false, false,
4990 				4, addr_len - 1);
4991 
4992 	/*
4993 	 * Check last byte of VM-entry MSR-load address
4994 	 */
4995 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
4996 
4997 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
4998 							i < 64; i++) {
4999 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
5000 			1ul << i;
5001 		vmcs_write(ENTER_MSR_LD_ADDR,
5002 			   tmp - (entry_msr_ld_cnt * 16 - 1));
5003 		test_vmx_invalid_controls(false);
5004 	}
5005 
5006 	vmcs_write(ENT_MSR_LD_CNT, 2);
5007 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5008 	test_vmx_invalid_controls(false);
5009 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5010 	test_vmx_valid_controls(false);
5011 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5012 	test_vmx_valid_controls(false);
5013 }
5014 
5015 static void guest_state_test_main(void)
5016 {
5017 	while (1) {
5018 		if (vmx_get_test_stage() != 2)
5019 			vmcall();
5020 		else
5021 			break;
5022 	}
5023 
5024 	asm volatile("fnop");
5025 }
5026 
5027 static void report_guest_state_test(const char *test, u32 xreason,
5028 				    u64 field, const char * field_name)
5029 {
5030 	u32 reason = vmcs_read(EXI_REASON);
5031 	u64 guest_rip;
5032 	u32 insn_len;
5033 
5034 	report("%s, %s %lx", reason == xreason, test, field_name, field);
5035 
5036 	guest_rip = vmcs_read(GUEST_RIP);
5037 	insn_len = vmcs_read(EXI_INST_LEN);
5038 	if (! (reason & 0x80000021))
5039 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
5040 }
5041 
5042 /*
5043  * Tests for VM-entry control fields
5044  */
5045 static void test_vm_entry_ctls(void)
5046 {
5047 	test_invalid_event_injection();
5048 	test_entry_msr_load();
5049 }
5050 
5051 /*
5052  * The following checks are performed for the VM-exit MSR-store address if
5053  * the VM-exit MSR-store count field is non-zero:
5054  *
5055  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
5056  *      The address should not set any bits beyond the processor’s
5057  *      physical-address width.
5058  *
5059  *    - The address of the last byte in the VM-exit MSR-store area
5060  *      should not set any bits beyond the processor’s physical-address
5061  *      width. The address of this last byte is VM-exit MSR-store address
5062  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
5063  *      uses more bits than the processor’s physical-address width.)
5064  *
5065  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
5066  * in the range 63:32.
5067  *
5068  *  [Intel SDM]
5069  */
5070 static void test_exit_msr_store(void)
5071 {
5072 	exit_msr_store = alloc_page();
5073 	u64 tmp;
5074 	u32 exit_msr_st_cnt = 1;
5075 	int i;
5076 	u32 addr_len = 64;
5077 
5078 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
5079 
5080 	/* Check first 4 bits of VM-exit MSR-store address */
5081 	for (i = 0; i < 4; i++) {
5082 		tmp = (u64)exit_msr_store | 1ull << i;
5083 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
5084 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
5085 				    tmp & 0xf);
5086 		test_vmx_invalid_controls(false);
5087 		report_prefix_pop();
5088 	}
5089 
5090 	if (basic.val & (1ul << 48))
5091 		addr_len = 32;
5092 
5093 	test_vmcs_addr_values("VM-exit-MSR-store address",
5094 				EXIT_MSR_ST_ADDR, 16, false, false,
5095 				4, addr_len - 1);
5096 
5097 	/*
5098 	 * Check last byte of VM-exit MSR-store address
5099 	 */
5100 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
5101 
5102 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
5103 							i < 64; i++) {
5104 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
5105 			1ul << i;
5106 		vmcs_write(EXIT_MSR_ST_ADDR,
5107 			   tmp - (exit_msr_st_cnt * 16 - 1));
5108 		test_vmx_invalid_controls(false);
5109 	}
5110 
5111 	vmcs_write(EXI_MSR_ST_CNT, 2);
5112 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5113 	test_vmx_invalid_controls(false);
5114 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5115 	test_vmx_valid_controls(false);
5116 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5117 	test_vmx_valid_controls(false);
5118 }
5119 
5120 /*
5121  * Tests for VM-exit controls
5122  */
5123 static void test_vm_exit_ctls(void)
5124 {
5125 	test_exit_msr_store();
5126 }
5127 
5128 /*
5129  * Check that the virtual CPU checks all of the VMX controls as
5130  * documented in the Intel SDM.
5131  */
5132 static void vmx_controls_test(void)
5133 {
5134 	/*
5135 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
5136 	 * fail due to invalid guest state, should we make it that
5137 	 * far.
5138 	 */
5139 	vmcs_write(GUEST_RFLAGS, 0);
5140 
5141 	test_vm_execution_ctls();
5142 	test_vm_exit_ctls();
5143 	test_vm_entry_ctls();
5144 }
5145 
5146 struct apic_reg_virt_config {
5147 	bool apic_register_virtualization;
5148 	bool use_tpr_shadow;
5149 	bool virtualize_apic_accesses;
5150 	bool virtualize_x2apic_mode;
5151 	bool activate_secondary_controls;
5152 };
5153 
5154 struct apic_reg_test {
5155 	const char *name;
5156 	struct apic_reg_virt_config apic_reg_virt_config;
5157 };
5158 
5159 struct apic_reg_virt_expectation {
5160 	enum Reason rd_exit_reason;
5161 	enum Reason wr_exit_reason;
5162 	u32 val;
5163 	u32 (*virt_fn)(u32);
5164 
5165 	/*
5166 	 * If false, accessing the APIC access address from L2 is treated as a
5167 	 * normal memory operation, rather than triggering virtualization.
5168 	 */
5169 	bool virtualize_apic_accesses;
5170 };
5171 
5172 static u32 apic_virt_identity(u32 val)
5173 {
5174 	return val;
5175 }
5176 
5177 static u32 apic_virt_nibble1(u32 val)
5178 {
5179 	return val & 0xf0;
5180 }
5181 
5182 static u32 apic_virt_byte3(u32 val)
5183 {
5184 	return val & (0xff << 24);
5185 }
5186 
5187 static bool apic_reg_virt_exit_expectation(
5188 	u32 reg, struct apic_reg_virt_config *config,
5189 	struct apic_reg_virt_expectation *expectation)
5190 {
5191 	/* Good configs, where some L2 APIC accesses are virtualized. */
5192 	bool virtualize_apic_accesses_only =
5193 		config->virtualize_apic_accesses &&
5194 		!config->use_tpr_shadow &&
5195 		!config->apic_register_virtualization &&
5196 		!config->virtualize_x2apic_mode &&
5197 		config->activate_secondary_controls;
5198 	bool virtualize_apic_accesses_and_use_tpr_shadow =
5199 		config->virtualize_apic_accesses &&
5200 		config->use_tpr_shadow &&
5201 		!config->apic_register_virtualization &&
5202 		!config->virtualize_x2apic_mode &&
5203 		config->activate_secondary_controls;
5204 	bool apic_register_virtualization =
5205 		config->virtualize_apic_accesses &&
5206 		config->use_tpr_shadow &&
5207 		config->apic_register_virtualization &&
5208 		!config->virtualize_x2apic_mode &&
5209 		config->activate_secondary_controls;
5210 
5211 	expectation->val = MAGIC_VAL_1;
5212 	expectation->virt_fn = apic_virt_identity;
5213 	expectation->virtualize_apic_accesses =
5214 		config->virtualize_apic_accesses &&
5215 		config->activate_secondary_controls;
5216 	if (virtualize_apic_accesses_only) {
5217 		expectation->rd_exit_reason = VMX_APIC_ACCESS;
5218 		expectation->wr_exit_reason = VMX_APIC_ACCESS;
5219 	} else if (virtualize_apic_accesses_and_use_tpr_shadow) {
5220 		switch (reg) {
5221 		case APIC_TASKPRI:
5222 			expectation->rd_exit_reason = VMX_VMCALL;
5223 			expectation->wr_exit_reason = VMX_VMCALL;
5224 			expectation->virt_fn = apic_virt_nibble1;
5225 			break;
5226 		default:
5227 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5228 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5229 		}
5230 	} else if (apic_register_virtualization) {
5231 		expectation->rd_exit_reason = VMX_VMCALL;
5232 
5233 		switch (reg) {
5234 		case APIC_ID:
5235 		case APIC_EOI:
5236 		case APIC_LDR:
5237 		case APIC_DFR:
5238 		case APIC_SPIV:
5239 		case APIC_ESR:
5240 		case APIC_ICR:
5241 		case APIC_LVTT:
5242 		case APIC_LVTTHMR:
5243 		case APIC_LVTPC:
5244 		case APIC_LVT0:
5245 		case APIC_LVT1:
5246 		case APIC_LVTERR:
5247 		case APIC_TMICT:
5248 		case APIC_TDCR:
5249 			expectation->wr_exit_reason = VMX_APIC_WRITE;
5250 			break;
5251 		case APIC_LVR:
5252 		case APIC_ISR ... APIC_ISR + 0x70:
5253 		case APIC_TMR ... APIC_TMR + 0x70:
5254 		case APIC_IRR ... APIC_IRR + 0x70:
5255 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5256 			break;
5257 		case APIC_TASKPRI:
5258 			expectation->wr_exit_reason = VMX_VMCALL;
5259 			expectation->virt_fn = apic_virt_nibble1;
5260 			break;
5261 		case APIC_ICR2:
5262 			expectation->wr_exit_reason = VMX_VMCALL;
5263 			expectation->virt_fn = apic_virt_byte3;
5264 			break;
5265 		default:
5266 			expectation->rd_exit_reason = VMX_APIC_ACCESS;
5267 			expectation->wr_exit_reason = VMX_APIC_ACCESS;
5268 		}
5269 	} else if (!expectation->virtualize_apic_accesses) {
5270 		/*
5271 		 * No APIC registers are directly virtualized. This includes
5272 		 * VTPR, which can be virtualized through MOV to/from CR8 via
5273 		 * the use TPR shadow control, but not through directly
5274 		 * accessing VTPR.
5275 		 */
5276 		expectation->rd_exit_reason = VMX_VMCALL;
5277 		expectation->wr_exit_reason = VMX_VMCALL;
5278 	} else {
5279 		printf("Cannot parse APIC register virtualization config:\n"
5280 		       "\tvirtualize_apic_accesses: %d\n"
5281 		       "\tuse_tpr_shadow: %d\n"
5282 		       "\tapic_register_virtualization: %d\n"
5283 		       "\tvirtualize_x2apic_mode: %d\n"
5284 		       "\tactivate_secondary_controls: %d\n",
5285 		       config->virtualize_apic_accesses,
5286 		       config->use_tpr_shadow,
5287 		       config->apic_register_virtualization,
5288 		       config->virtualize_x2apic_mode,
5289 		       config->activate_secondary_controls);
5290 
5291 		return false;
5292 	}
5293 
5294 	return true;
5295 }
5296 
5297 struct apic_reg_test apic_reg_tests[] = {
5298 	/* Good configs, where some L2 APIC accesses are virtualized. */
5299 	{
5300 		.name = "Virtualize APIC accesses",
5301 		.apic_reg_virt_config = {
5302 			.virtualize_apic_accesses = true,
5303 			.use_tpr_shadow = false,
5304 			.apic_register_virtualization = false,
5305 			.virtualize_x2apic_mode = false,
5306 			.activate_secondary_controls = true,
5307 		},
5308 	},
5309 	{
5310 		.name = "Virtualize APIC accesses + Use TPR shadow",
5311 		.apic_reg_virt_config = {
5312 			.virtualize_apic_accesses = true,
5313 			.use_tpr_shadow = true,
5314 			.apic_register_virtualization = false,
5315 			.virtualize_x2apic_mode = false,
5316 			.activate_secondary_controls = true,
5317 		},
5318 	},
5319 	{
5320 		.name = "APIC-register virtualization",
5321 		.apic_reg_virt_config = {
5322 			.virtualize_apic_accesses = true,
5323 			.use_tpr_shadow = true,
5324 			.apic_register_virtualization = true,
5325 			.virtualize_x2apic_mode = false,
5326 			.activate_secondary_controls = true,
5327 		},
5328 	},
5329 
5330 	/*
5331 	 * Test that the secondary processor-based VM-execution controls are
5332 	 * correctly ignored when "activate secondary controls" is disabled.
5333 	 */
5334 	{
5335 		.name = "Activate secondary controls off",
5336 		.apic_reg_virt_config = {
5337 			.virtualize_apic_accesses = true,
5338 			.use_tpr_shadow = false,
5339 			.apic_register_virtualization = true,
5340 			.virtualize_x2apic_mode = true,
5341 			.activate_secondary_controls = false,
5342 		},
5343 	},
5344 	{
5345 		.name = "Activate secondary controls off + Use TPR shadow",
5346 		.apic_reg_virt_config = {
5347 			.virtualize_apic_accesses = true,
5348 			.use_tpr_shadow = true,
5349 			.apic_register_virtualization = true,
5350 			.virtualize_x2apic_mode = true,
5351 			.activate_secondary_controls = false,
5352 		},
5353 	},
5354 
5355 	/*
5356 	 * Test that the APIC access address is treated like an arbitrary memory
5357 	 * address when "virtualize APIC accesses" is disabled.
5358 	 */
5359 	{
5360 		.name = "Virtualize APIC accesses off + Use TPR shadow",
5361 		.apic_reg_virt_config = {
5362 			.virtualize_apic_accesses = false,
5363 			.use_tpr_shadow = true,
5364 			.apic_register_virtualization = true,
5365 			.virtualize_x2apic_mode = true,
5366 			.activate_secondary_controls = true,
5367 		},
5368 	},
5369 
5370 	/*
5371 	 * Test that VM entry fails due to invalid controls when
5372 	 * "APIC-register virtualization" is enabled while "use TPR shadow" is
5373 	 * disabled.
5374 	 */
5375 	{
5376 		.name = "APIC-register virtualization + Use TPR shadow off",
5377 		.apic_reg_virt_config = {
5378 			.virtualize_apic_accesses = true,
5379 			.use_tpr_shadow = false,
5380 			.apic_register_virtualization = true,
5381 			.virtualize_x2apic_mode = false,
5382 			.activate_secondary_controls = true,
5383 		},
5384 	},
5385 
5386 	/*
5387 	 * Test that VM entry fails due to invalid controls when
5388 	 * "Virtualize x2APIC mode" is enabled while "use TPR shadow" is
5389 	 * disabled.
5390 	 */
5391 	{
5392 		.name = "Virtualize x2APIC mode + Use TPR shadow off",
5393 		.apic_reg_virt_config = {
5394 			.virtualize_apic_accesses = false,
5395 			.use_tpr_shadow = false,
5396 			.apic_register_virtualization = false,
5397 			.virtualize_x2apic_mode = true,
5398 			.activate_secondary_controls = true,
5399 		},
5400 	},
5401 	{
5402 		.name = "Virtualize x2APIC mode + Use TPR shadow off v2",
5403 		.apic_reg_virt_config = {
5404 			.virtualize_apic_accesses = false,
5405 			.use_tpr_shadow = false,
5406 			.apic_register_virtualization = true,
5407 			.virtualize_x2apic_mode = true,
5408 			.activate_secondary_controls = true,
5409 		},
5410 	},
5411 
5412 	/*
5413 	 * Test that VM entry fails due to invalid controls when
5414 	 * "virtualize x2APIC mode" is enabled while "virtualize APIC accesses"
5415 	 * is enabled.
5416 	 */
5417 	{
5418 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses",
5419 		.apic_reg_virt_config = {
5420 			.virtualize_apic_accesses = true,
5421 			.use_tpr_shadow = true,
5422 			.apic_register_virtualization = false,
5423 			.virtualize_x2apic_mode = true,
5424 			.activate_secondary_controls = true,
5425 		},
5426 	},
5427 	{
5428 		.name = "Virtualize x2APIC mode + Virtualize APIC accesses v2",
5429 		.apic_reg_virt_config = {
5430 			.virtualize_apic_accesses = true,
5431 			.use_tpr_shadow = true,
5432 			.apic_register_virtualization = true,
5433 			.virtualize_x2apic_mode = true,
5434 			.activate_secondary_controls = true,
5435 		},
5436 	},
5437 };
5438 
5439 enum Apic_op {
5440 	APIC_OP_XAPIC_RD,
5441 	APIC_OP_XAPIC_WR,
5442 	TERMINATE,
5443 };
5444 
5445 static u32 vmx_xapic_read(u32 *apic_access_address, u32 reg)
5446 {
5447 	return *(volatile u32 *)((uintptr_t)apic_access_address + reg);
5448 }
5449 
5450 static void vmx_xapic_write(u32 *apic_access_address, u32 reg, u32 val)
5451 {
5452 	*(volatile u32 *)((uintptr_t)apic_access_address + reg) = val;
5453 }
5454 
5455 struct apic_reg_virt_guest_args {
5456 	enum Apic_op op;
5457 	u32 *apic_access_address;
5458 	u32 reg;
5459 	u32 val;
5460 	bool check_rd;
5461 	u32 (*virt_fn)(u32);
5462 } apic_reg_virt_guest_args;
5463 
5464 static void apic_reg_virt_guest(void)
5465 {
5466 	volatile struct apic_reg_virt_guest_args *args =
5467 		&apic_reg_virt_guest_args;
5468 
5469 	for (;;) {
5470 		enum Apic_op op = args->op;
5471 		u32 *apic_access_address = args->apic_access_address;
5472 		u32 reg = args->reg;
5473 		u32 val = args->val;
5474 		bool check_rd = args->check_rd;
5475 		u32 (*virt_fn)(u32) = args->virt_fn;
5476 
5477 		if (op == TERMINATE)
5478 			break;
5479 
5480 		if (op == APIC_OP_XAPIC_RD) {
5481 			u32 ret = vmx_xapic_read(apic_access_address, reg);
5482 
5483 			if (check_rd) {
5484 				u32 want = virt_fn(val);
5485 				u32 got = virt_fn(ret);
5486 
5487 				report("read 0x%x, expected 0x%x.",
5488 				       got == want, got, want);
5489 			}
5490 		} else if (op == APIC_OP_XAPIC_WR) {
5491 			vmx_xapic_write(apic_access_address, reg, val);
5492 		}
5493 
5494 		/*
5495 		 * The L1 should always execute a vmcall after it's done testing
5496 		 * an individual APIC operation. This helps to validate that the
5497 		 * L1 and L2 are in sync with each other, as expected.
5498 		 */
5499 		vmcall();
5500 	}
5501 }
5502 
5503 static void test_xapic_rd(
5504 	u32 reg, struct apic_reg_virt_expectation *expectation,
5505 	u32 *apic_access_address, u32 *virtual_apic_page)
5506 {
5507 	u32 val = expectation->val;
5508 	u32 exit_reason_want = expectation->rd_exit_reason;
5509 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5510 
5511 	report_prefix_pushf("xapic - reading 0x%03x", reg);
5512 
5513 	/* Configure guest to do an xapic read */
5514 	args->op = APIC_OP_XAPIC_RD;
5515 	args->apic_access_address = apic_access_address;
5516 	args->reg = reg;
5517 	args->val = val;
5518 	args->check_rd = exit_reason_want == VMX_VMCALL;
5519 	args->virt_fn = expectation->virt_fn;
5520 
5521 	/* Setup virtual APIC page */
5522 	if (!expectation->virtualize_apic_accesses) {
5523 		apic_access_address[apic_reg_index(reg)] = val;
5524 		virtual_apic_page[apic_reg_index(reg)] = 0;
5525 	} else if (exit_reason_want == VMX_VMCALL) {
5526 		apic_access_address[apic_reg_index(reg)] = 0;
5527 		virtual_apic_page[apic_reg_index(reg)] = val;
5528 	}
5529 
5530 	/* Enter guest */
5531 	enter_guest();
5532 
5533 	/*
5534 	 * Validate the behavior and
5535 	 * pass a magic value back to the guest.
5536 	 */
5537 	if (exit_reason_want == VMX_APIC_ACCESS) {
5538 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5539 
5540 		assert_exit_reason(exit_reason_want);
5541 		report("got APIC access exit @ page offset 0x%03x, want 0x%03x",
5542 		       apic_page_offset == reg, apic_page_offset, reg);
5543 		skip_exit_insn();
5544 
5545 		/* Reenter guest so it can consume/check rcx and exit again. */
5546 		enter_guest();
5547 	} else if (exit_reason_want != VMX_VMCALL) {
5548 		report("Oops, bad exit expectation: %u.", false,
5549 		       exit_reason_want);
5550 	}
5551 
5552 	skip_exit_vmcall();
5553 	report_prefix_pop();
5554 }
5555 
5556 static void test_xapic_wr(
5557 	u32 reg, struct apic_reg_virt_expectation *expectation,
5558 	u32 *apic_access_address, u32 *virtual_apic_page)
5559 {
5560 	u32 val = expectation->val;
5561 	u32 exit_reason_want = expectation->wr_exit_reason;
5562 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5563 	bool virtualized =
5564 		expectation->virtualize_apic_accesses &&
5565 		(exit_reason_want == VMX_APIC_WRITE ||
5566 		 exit_reason_want == VMX_VMCALL);
5567 	bool checked = false;
5568 
5569 	report_prefix_pushf("xapic - writing 0x%x to 0x%03x", val, reg);
5570 
5571 	/* Configure guest to do an xapic read */
5572 	args->op = APIC_OP_XAPIC_WR;
5573 	args->apic_access_address = apic_access_address;
5574 	args->reg = reg;
5575 	args->val = val;
5576 
5577 	/* Setup virtual APIC page */
5578 	if (virtualized || !expectation->virtualize_apic_accesses) {
5579 		apic_access_address[apic_reg_index(reg)] = 0;
5580 		virtual_apic_page[apic_reg_index(reg)] = 0;
5581 	}
5582 
5583 	/* Enter guest */
5584 	enter_guest();
5585 
5586 	/*
5587 	 * Validate the behavior and
5588 	 * pass a magic value back to the guest.
5589 	 */
5590 	if (exit_reason_want == VMX_APIC_ACCESS) {
5591 		u32 apic_page_offset = vmcs_read(EXI_QUALIFICATION) & 0xfff;
5592 
5593 		assert_exit_reason(exit_reason_want);
5594 		report("got APIC access exit @ page offset 0x%03x, want 0x%03x",
5595 		       apic_page_offset == reg, apic_page_offset, reg);
5596 		skip_exit_insn();
5597 
5598 		/* Reenter guest so it can consume/check rcx and exit again. */
5599 		enter_guest();
5600 	} else if (exit_reason_want == VMX_APIC_WRITE) {
5601 		assert_exit_reason(exit_reason_want);
5602 		report("got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%x",
5603 		       virtual_apic_page[apic_reg_index(reg)] == val,
5604 		       apic_reg_index(reg),
5605 		       virtual_apic_page[apic_reg_index(reg)], val);
5606 		checked = true;
5607 
5608 		/* Reenter guest so it can consume/check rcx and exit again. */
5609 		enter_guest();
5610 	} else if (exit_reason_want != VMX_VMCALL) {
5611 		report("Oops, bad exit expectation: %u.", false,
5612 		       exit_reason_want);
5613 	}
5614 
5615 	assert_exit_reason(VMX_VMCALL);
5616 	if (virtualized && !checked) {
5617 		u32 want = expectation->virt_fn(val);
5618 		u32 got = virtual_apic_page[apic_reg_index(reg)];
5619 		got = expectation->virt_fn(got);
5620 
5621 		report("exitless write; val is 0x%x, want 0x%x",
5622 		       got == want, got, want);
5623 	} else if (!expectation->virtualize_apic_accesses && !checked) {
5624 		u32 got = apic_access_address[apic_reg_index(reg)];
5625 
5626 		report("non-virtualized write; val is 0x%x, want 0x%x",
5627 		       got == val, got, val);
5628 	} else if (!expectation->virtualize_apic_accesses && checked) {
5629 		report("Non-virtualized write was prematurely checked!", false);
5630 	}
5631 
5632 	skip_exit_vmcall();
5633 	report_prefix_pop();
5634 }
5635 
5636 enum Config_type {
5637 	CONFIG_TYPE_GOOD,
5638 	CONFIG_TYPE_UNSUPPORTED,
5639 	CONFIG_TYPE_VMENTRY_FAILS_EARLY,
5640 };
5641 
5642 static enum Config_type configure_apic_reg_virt_test(
5643 	struct apic_reg_virt_config *apic_reg_virt_config)
5644 {
5645 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5646 	u32 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
5647 	/* Configs where L2 entry fails early, due to invalid controls. */
5648 	bool use_tpr_shadow_incorrectly_off =
5649 		!apic_reg_virt_config->use_tpr_shadow &&
5650 		(apic_reg_virt_config->apic_register_virtualization ||
5651 		 apic_reg_virt_config->virtualize_x2apic_mode) &&
5652 		apic_reg_virt_config->activate_secondary_controls;
5653 	bool virtualize_apic_accesses_incorrectly_on =
5654 		apic_reg_virt_config->virtualize_apic_accesses &&
5655 		apic_reg_virt_config->virtualize_x2apic_mode &&
5656 		apic_reg_virt_config->activate_secondary_controls;
5657 	bool vmentry_fails_early =
5658 		use_tpr_shadow_incorrectly_off ||
5659 		virtualize_apic_accesses_incorrectly_on;
5660 
5661 	if (apic_reg_virt_config->activate_secondary_controls) {
5662 		if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
5663 			printf("VM-execution control \"activate secondary controls\" NOT supported.\n");
5664 			return CONFIG_TYPE_UNSUPPORTED;
5665 		}
5666 		cpu_exec_ctrl0 |= CPU_SECONDARY;
5667 	} else {
5668 		cpu_exec_ctrl0 &= ~CPU_SECONDARY;
5669 	}
5670 
5671 	if (apic_reg_virt_config->virtualize_apic_accesses) {
5672 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES)) {
5673 			printf("VM-execution control \"virtualize APIC accesses\" NOT supported.\n");
5674 			return CONFIG_TYPE_UNSUPPORTED;
5675 		}
5676 		cpu_exec_ctrl1 |= CPU_VIRT_APIC_ACCESSES;
5677 	} else {
5678 		cpu_exec_ctrl1 &= ~CPU_VIRT_APIC_ACCESSES;
5679 	}
5680 
5681 	if (apic_reg_virt_config->use_tpr_shadow) {
5682 		if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
5683 			printf("VM-execution control \"use TPR shadow\" NOT supported.\n");
5684 			return CONFIG_TYPE_UNSUPPORTED;
5685 		}
5686 		cpu_exec_ctrl0 |= CPU_TPR_SHADOW;
5687 	} else {
5688 		cpu_exec_ctrl0 &= ~CPU_TPR_SHADOW;
5689 	}
5690 
5691 	if (apic_reg_virt_config->apic_register_virtualization) {
5692 		if (!(ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT)) {
5693 			printf("VM-execution control \"APIC-register virtualization\" NOT supported.\n");
5694 			return CONFIG_TYPE_UNSUPPORTED;
5695 		}
5696 		cpu_exec_ctrl1 |= CPU_APIC_REG_VIRT;
5697 	} else {
5698 		cpu_exec_ctrl1 &= ~CPU_APIC_REG_VIRT;
5699 	}
5700 
5701 	if (apic_reg_virt_config->virtualize_x2apic_mode) {
5702 		if (!(ctrl_cpu_rev[1].clr & CPU_VIRT_X2APIC)) {
5703 			printf("VM-execution control \"virtualize x2APIC mode\" NOT supported.\n");
5704 			return CONFIG_TYPE_UNSUPPORTED;
5705 		}
5706 		cpu_exec_ctrl1 |= CPU_VIRT_X2APIC;
5707 	} else {
5708 		cpu_exec_ctrl1 &= ~CPU_VIRT_X2APIC;
5709 	}
5710 
5711 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
5712 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
5713 
5714 	if (vmentry_fails_early)
5715 		return CONFIG_TYPE_VMENTRY_FAILS_EARLY;
5716 
5717 	return CONFIG_TYPE_GOOD;
5718 }
5719 
5720 static bool cpu_has_apicv(void)
5721 {
5722 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
5723 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
5724 		(ctrl_pin_rev.clr & PIN_POST_INTR));
5725 }
5726 
5727 /* Validates APIC register access across valid virtualization configurations. */
5728 static void apic_reg_virt_test(void)
5729 {
5730 	u32 *apic_access_address;
5731 	u32 *virtual_apic_page;
5732 	u64 control;
5733 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
5734 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
5735 	int i;
5736 	struct apic_reg_virt_guest_args *args = &apic_reg_virt_guest_args;
5737 
5738 	if (!cpu_has_apicv()) {
5739 		report_skip(__func__);
5740 		return;
5741 	}
5742 
5743 	control = cpu_exec_ctrl1;
5744 	control &= ~CPU_VINTD;
5745 	vmcs_write(CPU_EXEC_CTRL1, control);
5746 
5747 	test_set_guest(apic_reg_virt_guest);
5748 
5749 	/*
5750 	 * From the SDM: The 1-setting of the "virtualize APIC accesses"
5751 	 * VM-execution is guaranteed to apply only if translations to the
5752 	 * APIC-access address use a 4-KByte page.
5753 	 */
5754 	apic_access_address = alloc_page();
5755 	force_4k_page(apic_access_address);
5756 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_address));
5757 
5758 	virtual_apic_page = alloc_page();
5759 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
5760 
5761 	for (i = 0; i < ARRAY_SIZE(apic_reg_tests); i++) {
5762 		struct apic_reg_test *apic_reg_test = &apic_reg_tests[i];
5763 		struct apic_reg_virt_config *apic_reg_virt_config =
5764 				&apic_reg_test->apic_reg_virt_config;
5765 		enum Config_type config_type;
5766 		u32 reg;
5767 
5768 		printf("--- %s test ---\n", apic_reg_test->name);
5769 		config_type =
5770 			configure_apic_reg_virt_test(apic_reg_virt_config);
5771 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
5772 			printf("Skip because of missing features.\n");
5773 			continue;
5774 		}
5775 
5776 		if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
5777 			enter_guest_with_bad_controls();
5778 			continue;
5779 		}
5780 
5781 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
5782 			struct apic_reg_virt_expectation expectation = {};
5783 			bool ok;
5784 
5785 			ok = apic_reg_virt_exit_expectation(
5786 				reg, apic_reg_virt_config, &expectation);
5787 			if (!ok) {
5788 				report("Malformed test.", false);
5789 				break;
5790 			}
5791 
5792 			test_xapic_rd(reg, &expectation, apic_access_address,
5793 				      virtual_apic_page);
5794 			test_xapic_wr(reg, &expectation, apic_access_address,
5795 				      virtual_apic_page);
5796 		}
5797 	}
5798 
5799 	/* Terminate the guest */
5800 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
5801 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
5802 	args->op = TERMINATE;
5803 	enter_guest();
5804 	assert_exit_reason(VMX_VMCALL);
5805 }
5806 
5807 struct virt_x2apic_mode_config {
5808 	struct apic_reg_virt_config apic_reg_virt_config;
5809 	bool virtual_interrupt_delivery;
5810 	bool use_msr_bitmaps;
5811 	bool disable_x2apic_msr_intercepts;
5812 	bool disable_x2apic;
5813 };
5814 
5815 struct virt_x2apic_mode_test_case {
5816 	const char *name;
5817 	struct virt_x2apic_mode_config virt_x2apic_mode_config;
5818 };
5819 
5820 enum Virt_x2apic_mode_behavior_type {
5821 	X2APIC_ACCESS_VIRTUALIZED,
5822 	X2APIC_ACCESS_PASSED_THROUGH,
5823 	X2APIC_ACCESS_TRIGGERS_GP,
5824 };
5825 
5826 struct virt_x2apic_mode_expectation {
5827 	enum Reason rd_exit_reason;
5828 	enum Reason wr_exit_reason;
5829 
5830 	/*
5831 	 * RDMSR and WRMSR handle 64-bit values. However, except for ICR, all of
5832 	 * the x2APIC registers are 32 bits. Notice:
5833 	 *   1. vmx_x2apic_read() clears the upper 32 bits for 32-bit registers.
5834 	 *   2. vmx_x2apic_write() expects the val arg to be well-formed.
5835 	 */
5836 	u64 rd_val;
5837 	u64 wr_val;
5838 
5839 	/*
5840 	 * Compares input to virtualized output;
5841 	 * 1st arg is pointer to return expected virtualization output.
5842 	 */
5843 	u64 (*virt_fn)(u64);
5844 
5845 	enum Virt_x2apic_mode_behavior_type rd_behavior;
5846 	enum Virt_x2apic_mode_behavior_type wr_behavior;
5847 	bool wr_only;
5848 };
5849 
5850 static u64 virt_x2apic_mode_identity(u64 val)
5851 {
5852 	return val;
5853 }
5854 
5855 static u64 virt_x2apic_mode_nibble1(u64 val)
5856 {
5857 	return val & 0xf0;
5858 }
5859 
5860 static bool is_cmci_enabled(void)
5861 {
5862 	return rdmsr(MSR_IA32_MCG_CAP) & BIT_ULL(10);
5863 }
5864 
5865 static void virt_x2apic_mode_rd_expectation(
5866 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
5867 	bool apic_register_virtualization, bool virtual_interrupt_delivery,
5868 	struct virt_x2apic_mode_expectation *expectation)
5869 {
5870 	bool readable =
5871 		!x2apic_reg_reserved(reg) &&
5872 		reg != APIC_EOI;
5873 
5874 	if (reg == APIC_CMCI && !is_cmci_enabled())
5875 		readable = false;
5876 
5877 	expectation->rd_exit_reason = VMX_VMCALL;
5878 	expectation->virt_fn = virt_x2apic_mode_identity;
5879 	if (virt_x2apic_mode_on && apic_register_virtualization) {
5880 		expectation->rd_val = MAGIC_VAL_1;
5881 		if (reg == APIC_PROCPRI && virtual_interrupt_delivery)
5882 			expectation->virt_fn = virt_x2apic_mode_nibble1;
5883 		else if (reg == APIC_TASKPRI)
5884 			expectation->virt_fn = virt_x2apic_mode_nibble1;
5885 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
5886 	} else if (virt_x2apic_mode_on && !apic_register_virtualization &&
5887 		   reg == APIC_TASKPRI) {
5888 		expectation->rd_val = MAGIC_VAL_1;
5889 		expectation->virt_fn = virt_x2apic_mode_nibble1;
5890 		expectation->rd_behavior = X2APIC_ACCESS_VIRTUALIZED;
5891 	} else if (!disable_x2apic && readable) {
5892 		expectation->rd_val = apic_read(reg);
5893 		expectation->rd_behavior = X2APIC_ACCESS_PASSED_THROUGH;
5894 	} else {
5895 		expectation->rd_behavior = X2APIC_ACCESS_TRIGGERS_GP;
5896 	}
5897 }
5898 
5899 /*
5900  * get_x2apic_wr_val() creates an innocuous write value for an x2APIC register.
5901  *
5902  * For writable registers, get_x2apic_wr_val() deposits the write value into the
5903  * val pointer arg and returns true. For non-writable registers, val is not
5904  * modified and get_x2apic_wr_val() returns false.
5905  */
5906 static bool get_x2apic_wr_val(u32 reg, u64 *val)
5907 {
5908 	switch (reg) {
5909 	case APIC_TASKPRI:
5910 		/* Bits 31:8 are reserved. */
5911 		*val &= 0xff;
5912 		break;
5913 	case APIC_EOI:
5914 	case APIC_ESR:
5915 	case APIC_TMICT:
5916 		/*
5917 		 * EOI, ESR: WRMSR of a non-zero value causes #GP(0).
5918 		 * TMICT: A write of 0 to the initial-count register effectively
5919 		 *        stops the local APIC timer, in both one-shot and
5920 		 *        periodic mode.
5921 		 */
5922 		*val = 0;
5923 		break;
5924 	case APIC_SPIV:
5925 	case APIC_LVTT:
5926 	case APIC_LVTTHMR:
5927 	case APIC_LVTPC:
5928 	case APIC_LVT0:
5929 	case APIC_LVT1:
5930 	case APIC_LVTERR:
5931 	case APIC_TDCR:
5932 		/*
5933 		 * To avoid writing a 1 to a reserved bit or causing some other
5934 		 * unintended side effect, read the current value and use it as
5935 		 * the write value.
5936 		 */
5937 		*val = apic_read(reg);
5938 		break;
5939 	case APIC_CMCI:
5940 		if (!is_cmci_enabled())
5941 			return false;
5942 		*val = apic_read(reg);
5943 		break;
5944 	case APIC_ICR:
5945 		*val = 0x40000 | 0xf1;
5946 		break;
5947 	case APIC_SELF_IPI:
5948 		/*
5949 		 * With special processing (i.e., virtualize x2APIC mode +
5950 		 * virtual interrupt delivery), writing zero causes an
5951 		 * APIC-write VM exit. We plan to add a test for enabling
5952 		 * "virtual-interrupt delivery" in VMCS12, and that's where we
5953 		 * will test a self IPI with special processing.
5954 		 */
5955 		*val = 0x0;
5956 		break;
5957 	default:
5958 		return false;
5959 	}
5960 
5961 	return true;
5962 }
5963 
5964 static bool special_processing_applies(u32 reg, u64 *val,
5965 				       bool virt_int_delivery)
5966 {
5967 	bool special_processing =
5968 		(reg == APIC_TASKPRI) ||
5969 		(virt_int_delivery &&
5970 		 (reg == APIC_EOI || reg == APIC_SELF_IPI));
5971 
5972 	if (special_processing) {
5973 		TEST_ASSERT(get_x2apic_wr_val(reg, val));
5974 		return true;
5975 	}
5976 
5977 	return false;
5978 }
5979 
5980 static void virt_x2apic_mode_wr_expectation(
5981 	u32 reg, bool virt_x2apic_mode_on, bool disable_x2apic,
5982 	bool virt_int_delivery,
5983 	struct virt_x2apic_mode_expectation *expectation)
5984 {
5985 	expectation->wr_exit_reason = VMX_VMCALL;
5986 	expectation->wr_val = MAGIC_VAL_1;
5987 	expectation->wr_only = false;
5988 
5989 	if (virt_x2apic_mode_on &&
5990 	    special_processing_applies(reg, &expectation->wr_val,
5991 				       virt_int_delivery)) {
5992 		expectation->wr_behavior = X2APIC_ACCESS_VIRTUALIZED;
5993 		if (reg == APIC_SELF_IPI)
5994 			expectation->wr_exit_reason = VMX_APIC_WRITE;
5995 	} else if (!disable_x2apic &&
5996 		   get_x2apic_wr_val(reg, &expectation->wr_val)) {
5997 		expectation->wr_behavior = X2APIC_ACCESS_PASSED_THROUGH;
5998 		if (reg == APIC_EOI || reg == APIC_SELF_IPI)
5999 			expectation->wr_only = true;
6000 		if (reg == APIC_ICR)
6001 			expectation->wr_exit_reason = VMX_EXTINT;
6002 	} else {
6003 		expectation->wr_behavior = X2APIC_ACCESS_TRIGGERS_GP;
6004 		/*
6005 		 * Writing 1 to a reserved bit triggers a #GP.
6006 		 * Thus, set the write value to 0, which seems
6007 		 * the most likely to detect a missed #GP.
6008 		 */
6009 		expectation->wr_val = 0;
6010 	}
6011 }
6012 
6013 static void virt_x2apic_mode_exit_expectation(
6014 	u32 reg, struct virt_x2apic_mode_config *config,
6015 	struct virt_x2apic_mode_expectation *expectation)
6016 {
6017 	struct apic_reg_virt_config *base_config =
6018 		&config->apic_reg_virt_config;
6019 	bool virt_x2apic_mode_on =
6020 		base_config->virtualize_x2apic_mode &&
6021 		config->use_msr_bitmaps &&
6022 		config->disable_x2apic_msr_intercepts &&
6023 		base_config->activate_secondary_controls;
6024 
6025 	virt_x2apic_mode_wr_expectation(
6026 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6027 		config->virtual_interrupt_delivery, expectation);
6028 	virt_x2apic_mode_rd_expectation(
6029 		reg, virt_x2apic_mode_on, config->disable_x2apic,
6030 		base_config->apic_register_virtualization,
6031 		config->virtual_interrupt_delivery, expectation);
6032 }
6033 
6034 struct virt_x2apic_mode_test_case virt_x2apic_mode_tests[] = {
6035 	/*
6036 	 * Baseline "virtualize x2APIC mode" configuration:
6037 	 *   - virtualize x2APIC mode
6038 	 *   - virtual-interrupt delivery
6039 	 *   - APIC-register virtualization
6040 	 *   - x2APIC MSR intercepts disabled
6041 	 *
6042 	 * Reads come from virtual APIC page, special processing applies to
6043 	 * VTPR, EOI, and SELF IPI, and all other writes pass through to L1
6044 	 * APIC.
6045 	 */
6046 	{
6047 		.name = "Baseline",
6048 		.virt_x2apic_mode_config = {
6049 			.virtual_interrupt_delivery = true,
6050 			.use_msr_bitmaps = true,
6051 			.disable_x2apic_msr_intercepts = true,
6052 			.disable_x2apic = false,
6053 			.apic_reg_virt_config = {
6054 				.apic_register_virtualization = true,
6055 				.use_tpr_shadow = true,
6056 				.virtualize_apic_accesses = false,
6057 				.virtualize_x2apic_mode = true,
6058 				.activate_secondary_controls = true,
6059 			},
6060 		},
6061 	},
6062 	{
6063 		.name = "Baseline w/ x2apic disabled",
6064 		.virt_x2apic_mode_config = {
6065 			.virtual_interrupt_delivery = true,
6066 			.use_msr_bitmaps = true,
6067 			.disable_x2apic_msr_intercepts = true,
6068 			.disable_x2apic = true,
6069 			.apic_reg_virt_config = {
6070 				.apic_register_virtualization = true,
6071 				.use_tpr_shadow = true,
6072 				.virtualize_apic_accesses = false,
6073 				.virtualize_x2apic_mode = true,
6074 				.activate_secondary_controls = true,
6075 			},
6076 		},
6077 	},
6078 
6079 	/*
6080 	 * Baseline, minus virtual-interrupt delivery. Reads come from virtual
6081 	 * APIC page, special processing applies to VTPR, and all other writes
6082 	 * pass through to L1 APIC.
6083 	 */
6084 	{
6085 		.name = "Baseline - virtual interrupt delivery",
6086 		.virt_x2apic_mode_config = {
6087 			.virtual_interrupt_delivery = false,
6088 			.use_msr_bitmaps = true,
6089 			.disable_x2apic_msr_intercepts = true,
6090 			.disable_x2apic = false,
6091 			.apic_reg_virt_config = {
6092 				.apic_register_virtualization = true,
6093 				.use_tpr_shadow = true,
6094 				.virtualize_apic_accesses = false,
6095 				.virtualize_x2apic_mode = true,
6096 				.activate_secondary_controls = true,
6097 			},
6098 		},
6099 	},
6100 
6101 	/*
6102 	 * Baseline, minus APIC-register virtualization. x2APIC reads pass
6103 	 * through to L1's APIC, unless reading VTPR
6104 	 */
6105 	{
6106 		.name = "Virtualize x2APIC mode, no APIC reg virt",
6107 		.virt_x2apic_mode_config = {
6108 			.virtual_interrupt_delivery = true,
6109 			.use_msr_bitmaps = true,
6110 			.disable_x2apic_msr_intercepts = true,
6111 			.disable_x2apic = false,
6112 			.apic_reg_virt_config = {
6113 				.apic_register_virtualization = false,
6114 				.use_tpr_shadow = true,
6115 				.virtualize_apic_accesses = false,
6116 				.virtualize_x2apic_mode = true,
6117 				.activate_secondary_controls = true,
6118 			},
6119 		},
6120 	},
6121 	{
6122 		.name = "Virtualize x2APIC mode, no APIC reg virt, x2APIC off",
6123 		.virt_x2apic_mode_config = {
6124 			.virtual_interrupt_delivery = true,
6125 			.use_msr_bitmaps = true,
6126 			.disable_x2apic_msr_intercepts = true,
6127 			.disable_x2apic = true,
6128 			.apic_reg_virt_config = {
6129 				.apic_register_virtualization = false,
6130 				.use_tpr_shadow = true,
6131 				.virtualize_apic_accesses = false,
6132 				.virtualize_x2apic_mode = true,
6133 				.activate_secondary_controls = true,
6134 			},
6135 		},
6136 	},
6137 
6138 	/*
6139 	 * Enable "virtualize x2APIC mode" and "APIC-register virtualization",
6140 	 * and disable intercepts for the x2APIC MSRs, but fail to enable
6141 	 * "activate secondary controls" (i.e. L2 gets access to L1's x2APIC
6142 	 * MSRs).
6143 	 */
6144 	{
6145 		.name = "Fail to enable activate secondary controls",
6146 		.virt_x2apic_mode_config = {
6147 			.virtual_interrupt_delivery = true,
6148 			.use_msr_bitmaps = true,
6149 			.disable_x2apic_msr_intercepts = true,
6150 			.disable_x2apic = false,
6151 			.apic_reg_virt_config = {
6152 				.apic_register_virtualization = true,
6153 				.use_tpr_shadow = true,
6154 				.virtualize_apic_accesses = false,
6155 				.virtualize_x2apic_mode = true,
6156 				.activate_secondary_controls = false,
6157 			},
6158 		},
6159 	},
6160 
6161 	/*
6162 	 * Enable "APIC-register virtualization" and enable "activate secondary
6163 	 * controls" and disable intercepts for the x2APIC MSRs, but do not
6164 	 * enable the "virtualize x2APIC mode" VM-execution control (i.e. L2
6165 	 * gets access to L1's x2APIC MSRs).
6166 	 */
6167 	{
6168 		.name = "Fail to enable virtualize x2APIC mode",
6169 		.virt_x2apic_mode_config = {
6170 			.virtual_interrupt_delivery = true,
6171 			.use_msr_bitmaps = true,
6172 			.disable_x2apic_msr_intercepts = true,
6173 			.disable_x2apic = false,
6174 			.apic_reg_virt_config = {
6175 				.apic_register_virtualization = true,
6176 				.use_tpr_shadow = true,
6177 				.virtualize_apic_accesses = false,
6178 				.virtualize_x2apic_mode = false,
6179 				.activate_secondary_controls = true,
6180 			},
6181 		},
6182 	},
6183 
6184 	/*
6185 	 * Disable "Virtualize x2APIC mode", disable x2APIC MSR intercepts, and
6186 	 * enable "APIC-register virtualization" --> L2 gets L1's x2APIC MSRs.
6187 	 */
6188 	{
6189 		.name = "Baseline",
6190 		.virt_x2apic_mode_config = {
6191 			.virtual_interrupt_delivery = true,
6192 			.use_msr_bitmaps = true,
6193 			.disable_x2apic_msr_intercepts = true,
6194 			.disable_x2apic = false,
6195 			.apic_reg_virt_config = {
6196 				.apic_register_virtualization = true,
6197 				.use_tpr_shadow = true,
6198 				.virtualize_apic_accesses = false,
6199 				.virtualize_x2apic_mode = false,
6200 				.activate_secondary_controls = true,
6201 			},
6202 		},
6203 	},
6204 };
6205 
6206 enum X2apic_op {
6207 	X2APIC_OP_RD,
6208 	X2APIC_OP_WR,
6209 	X2APIC_TERMINATE,
6210 };
6211 
6212 static u64 vmx_x2apic_read(u32 reg)
6213 {
6214 	u32 msr_addr = x2apic_msr(reg);
6215 	u64 val;
6216 
6217 	val = rdmsr(msr_addr);
6218 
6219 	return val;
6220 }
6221 
6222 static void vmx_x2apic_write(u32 reg, u64 val)
6223 {
6224 	u32 msr_addr = x2apic_msr(reg);
6225 
6226 	wrmsr(msr_addr, val);
6227 }
6228 
6229 struct virt_x2apic_mode_guest_args {
6230 	enum X2apic_op op;
6231 	u32 reg;
6232 	u64 val;
6233 	bool should_gp;
6234 	u64 (*virt_fn)(u64);
6235 } virt_x2apic_mode_guest_args;
6236 
6237 static volatile bool handle_x2apic_gp_ran;
6238 static volatile u32 handle_x2apic_gp_insn_len;
6239 static void handle_x2apic_gp(struct ex_regs *regs)
6240 {
6241 	handle_x2apic_gp_ran = true;
6242 	regs->rip += handle_x2apic_gp_insn_len;
6243 }
6244 
6245 static handler setup_x2apic_gp_handler(void)
6246 {
6247 	handler old_handler;
6248 
6249 	old_handler = handle_exception(GP_VECTOR, handle_x2apic_gp);
6250 	/* RDMSR and WRMSR are both 2 bytes, assuming no prefixes. */
6251 	handle_x2apic_gp_insn_len = 2;
6252 
6253 	return old_handler;
6254 }
6255 
6256 static void teardown_x2apic_gp_handler(handler old_handler)
6257 {
6258 	handle_exception(GP_VECTOR, old_handler);
6259 
6260 	/*
6261 	 * Defensively reset instruction length, so that if the handler is
6262 	 * incorrectly used, it will loop infinitely, rather than run off into
6263 	 * la la land.
6264 	 */
6265 	handle_x2apic_gp_insn_len = 0;
6266 	handle_x2apic_gp_ran = false;
6267 }
6268 
6269 static void virt_x2apic_mode_guest(void)
6270 {
6271 	volatile struct virt_x2apic_mode_guest_args *args =
6272 		&virt_x2apic_mode_guest_args;
6273 
6274 	for (;;) {
6275 		enum X2apic_op op = args->op;
6276 		u32 reg = args->reg;
6277 		u64 val = args->val;
6278 		bool should_gp = args->should_gp;
6279 		u64 (*virt_fn)(u64) = args->virt_fn;
6280 		handler old_handler;
6281 
6282 		if (op == X2APIC_TERMINATE)
6283 			break;
6284 
6285 		if (should_gp) {
6286 			TEST_ASSERT(!handle_x2apic_gp_ran);
6287 			old_handler = setup_x2apic_gp_handler();
6288 		}
6289 
6290 		if (op == X2APIC_OP_RD) {
6291 			u64 ret = vmx_x2apic_read(reg);
6292 
6293 			if (!should_gp) {
6294 				u64 want = virt_fn(val);
6295 				u64 got = virt_fn(ret);
6296 
6297 				report("APIC read; got 0x%lx, want 0x%lx.",
6298 				       got == want, got, want);
6299 			}
6300 		} else if (op == X2APIC_OP_WR) {
6301 			vmx_x2apic_write(reg, val);
6302 		}
6303 
6304 		if (should_gp) {
6305 			report("x2APIC op triggered GP.",
6306 			       handle_x2apic_gp_ran);
6307 			teardown_x2apic_gp_handler(old_handler);
6308 		}
6309 
6310 		/*
6311 		 * The L1 should always execute a vmcall after it's done testing
6312 		 * an individual APIC operation. This helps to validate that the
6313 		 * L1 and L2 are in sync with each other, as expected.
6314 		 */
6315 		vmcall();
6316 	}
6317 }
6318 
6319 static void test_x2apic_rd(
6320 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6321 	u32 *virtual_apic_page)
6322 {
6323 	u64 val = expectation->rd_val;
6324 	u32 exit_reason_want = expectation->rd_exit_reason;
6325 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6326 
6327 	report_prefix_pushf("x2apic - reading 0x%03x", reg);
6328 
6329 	/* Configure guest to do an x2apic read */
6330 	args->op = X2APIC_OP_RD;
6331 	args->reg = reg;
6332 	args->val = val;
6333 	args->should_gp = expectation->rd_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6334 	args->virt_fn = expectation->virt_fn;
6335 
6336 	/* Setup virtual APIC page */
6337 	if (expectation->rd_behavior == X2APIC_ACCESS_VIRTUALIZED)
6338 		virtual_apic_page[apic_reg_index(reg)] = (u32)val;
6339 
6340 	/* Enter guest */
6341 	enter_guest();
6342 
6343 	if (exit_reason_want != VMX_VMCALL) {
6344 		report("Oops, bad exit expectation: %u.", false,
6345 		       exit_reason_want);
6346 	}
6347 
6348 	skip_exit_vmcall();
6349 	report_prefix_pop();
6350 }
6351 
6352 static volatile bool handle_x2apic_ipi_ran;
6353 static void handle_x2apic_ipi(isr_regs_t *regs)
6354 {
6355 	handle_x2apic_ipi_ran = true;
6356 	eoi();
6357 }
6358 
6359 static void test_x2apic_wr(
6360 	u32 reg, struct virt_x2apic_mode_expectation *expectation,
6361 	u32 *virtual_apic_page)
6362 {
6363 	u64 val = expectation->wr_val;
6364 	u32 exit_reason_want = expectation->wr_exit_reason;
6365 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6366 	int ipi_vector = 0xf1;
6367 	u32 restore_val = 0;
6368 
6369 	report_prefix_pushf("x2apic - writing 0x%lx to 0x%03x", val, reg);
6370 
6371 	/* Configure guest to do an x2apic read */
6372 	args->op = X2APIC_OP_WR;
6373 	args->reg = reg;
6374 	args->val = val;
6375 	args->should_gp = expectation->wr_behavior == X2APIC_ACCESS_TRIGGERS_GP;
6376 
6377 	/* Setup virtual APIC page */
6378 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED)
6379 		virtual_apic_page[apic_reg_index(reg)] = 0;
6380 	if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH && !expectation->wr_only)
6381 		restore_val = apic_read(reg);
6382 
6383 	/* Setup IPI handler */
6384 	handle_x2apic_ipi_ran = false;
6385 	handle_irq(ipi_vector, handle_x2apic_ipi);
6386 
6387 	/* Enter guest */
6388 	enter_guest();
6389 
6390 	/*
6391 	 * Validate the behavior and
6392 	 * pass a magic value back to the guest.
6393 	 */
6394 	if (exit_reason_want == VMX_EXTINT) {
6395 		assert_exit_reason(exit_reason_want);
6396 
6397 		/* Clear the external interrupt. */
6398 		irq_enable();
6399 		asm volatile ("nop");
6400 		irq_disable();
6401 		report("Got pending interrupt after IRQ enabled.",
6402 		       handle_x2apic_ipi_ran);
6403 
6404 		enter_guest();
6405 	} else if (exit_reason_want == VMX_APIC_WRITE) {
6406 		assert_exit_reason(exit_reason_want);
6407 		report("got APIC write exit @ page offset 0x%03x; val is 0x%x, want 0x%lx",
6408 		       virtual_apic_page[apic_reg_index(reg)] == val,
6409 		       apic_reg_index(reg),
6410 		       virtual_apic_page[apic_reg_index(reg)], val);
6411 
6412 		/* Reenter guest so it can consume/check rcx and exit again. */
6413 		enter_guest();
6414 	} else if (exit_reason_want != VMX_VMCALL) {
6415 		report("Oops, bad exit expectation: %u.", false,
6416 		       exit_reason_want);
6417 	}
6418 
6419 	assert_exit_reason(VMX_VMCALL);
6420 	if (expectation->wr_behavior == X2APIC_ACCESS_VIRTUALIZED) {
6421 		u64 want = val;
6422 		u32 got = virtual_apic_page[apic_reg_index(reg)];
6423 
6424 		report("x2APIC write; got 0x%x, want 0x%lx",
6425 		       got == want, got, want);
6426 	} else if (expectation->wr_behavior == X2APIC_ACCESS_PASSED_THROUGH) {
6427 		if (!expectation->wr_only) {
6428 			u32 got = apic_read(reg);
6429 			bool ok;
6430 
6431 			/*
6432 			 * When L1's TPR is passed through to L2, the lower
6433 			 * nibble can be lost. For example, if L2 executes
6434 			 * WRMSR(0x808, 0x78), then, L1 might read 0x70.
6435 			 *
6436 			 * Here's how the lower nibble can get lost:
6437 			 *   1. L2 executes WRMSR(0x808, 0x78).
6438 			 *   2. L2 exits to L0 with a WRMSR exit.
6439 			 *   3. L0 emulates WRMSR, by writing L1's TPR.
6440 			 *   4. L0 re-enters L2.
6441 			 *   5. L2 exits to L0 (reason doesn't matter).
6442 			 *   6. L0 reflects L2's exit to L1.
6443 			 *   7. Before entering L1, L0 exits to user-space
6444 			 *      (e.g., to satisfy TPR access reporting).
6445 			 *   8. User-space executes KVM_SET_REGS ioctl, which
6446 			 *      clears the lower nibble of L1's TPR.
6447 			 */
6448 			if (reg == APIC_TASKPRI) {
6449 				got = apic_virt_nibble1(got);
6450 				val = apic_virt_nibble1(val);
6451 			}
6452 
6453 			ok = got == val;
6454 			report("non-virtualized write; val is 0x%x, want 0x%lx",
6455 			       ok, got, val);
6456 			apic_write(reg, restore_val);
6457 		} else {
6458 			report("non-virtualized and write-only OK", true);
6459 		}
6460 	}
6461 	skip_exit_insn();
6462 
6463 	report_prefix_pop();
6464 }
6465 
6466 static enum Config_type configure_virt_x2apic_mode_test(
6467 	struct virt_x2apic_mode_config *virt_x2apic_mode_config,
6468 	u8 *msr_bitmap_page)
6469 {
6470 	int msr;
6471 	u32 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6472 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6473 
6474 	/* x2apic-specific VMCS config */
6475 	if (virt_x2apic_mode_config->use_msr_bitmaps) {
6476 		/* virt_x2apic_mode_test() checks for MSR bitmaps support */
6477 		cpu_exec_ctrl0 |= CPU_MSR_BITMAP;
6478 	} else {
6479 		cpu_exec_ctrl0 &= ~CPU_MSR_BITMAP;
6480 	}
6481 
6482 	if (virt_x2apic_mode_config->virtual_interrupt_delivery) {
6483 		if (!(ctrl_cpu_rev[1].clr & CPU_VINTD)) {
6484 			report_skip("VM-execution control \"virtual-interrupt delivery\" NOT supported.\n");
6485 			return CONFIG_TYPE_UNSUPPORTED;
6486 		}
6487 		cpu_exec_ctrl1 |= CPU_VINTD;
6488 	} else {
6489 		cpu_exec_ctrl1 &= ~CPU_VINTD;
6490 	}
6491 
6492 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6493 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6494 
6495 	/* x2APIC MSR intercepts are usually off for "Virtualize x2APIC mode" */
6496 	for (msr = 0x800; msr <= 0x8ff; msr++) {
6497 		if (virt_x2apic_mode_config->disable_x2apic_msr_intercepts) {
6498 			clear_bit(msr, msr_bitmap_page + 0x000);
6499 			clear_bit(msr, msr_bitmap_page + 0x800);
6500 		} else {
6501 			set_bit(msr, msr_bitmap_page + 0x000);
6502 			set_bit(msr, msr_bitmap_page + 0x800);
6503 		}
6504 	}
6505 
6506 	/* x2APIC mode can impact virtualization */
6507 	reset_apic();
6508 	if (!virt_x2apic_mode_config->disable_x2apic)
6509 		enable_x2apic();
6510 
6511 	return configure_apic_reg_virt_test(
6512 		&virt_x2apic_mode_config->apic_reg_virt_config);
6513 }
6514 
6515 static void virt_x2apic_mode_test(void)
6516 {
6517 	u32 *virtual_apic_page;
6518 	u8 *msr_bitmap_page;
6519 	u64 cpu_exec_ctrl0 = vmcs_read(CPU_EXEC_CTRL0);
6520 	u64 cpu_exec_ctrl1 = vmcs_read(CPU_EXEC_CTRL1);
6521 	int i;
6522 	struct virt_x2apic_mode_guest_args *args = &virt_x2apic_mode_guest_args;
6523 
6524 	if (!cpu_has_apicv()) {
6525 		report_skip(__func__);
6526 		return;
6527 	}
6528 
6529 	/*
6530 	 * This is to exercise an issue in KVM's logic to merge L0's and L1's
6531 	 * MSR bitmaps. Previously, an L1 could get at L0's x2APIC MSRs by
6532 	 * writing the IA32_SPEC_CTRL MSR or the IA32_PRED_CMD MSRs. KVM would
6533 	 * then proceed to manipulate the MSR bitmaps, as if VMCS12 had the
6534 	 * "Virtualize x2APIC mod" control set, even when it didn't.
6535 	 */
6536 	if (has_spec_ctrl())
6537 		wrmsr(MSR_IA32_SPEC_CTRL, 1);
6538 
6539 	/*
6540 	 * Check that VMCS12 supports:
6541 	 *   - "Virtual-APIC address", indicated by "use TPR shadow"
6542 	 *   - "MSR-bitmap address", indicated by "use MSR bitmaps"
6543 	 */
6544 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW)) {
6545 		report_skip("VM-execution control \"use TPR shadow\" NOT supported.\n");
6546 		return;
6547 	} else if (!(ctrl_cpu_rev[0].clr & CPU_MSR_BITMAP)) {
6548 		report_skip("VM-execution control \"use MSR bitmaps\" NOT supported.\n");
6549 		return;
6550 	}
6551 
6552 	test_set_guest(virt_x2apic_mode_guest);
6553 
6554 	virtual_apic_page = alloc_page();
6555 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
6556 
6557 	msr_bitmap_page = alloc_page();
6558 	memset(msr_bitmap_page, 0xff, PAGE_SIZE);
6559 	vmcs_write(MSR_BITMAP, virt_to_phys(msr_bitmap_page));
6560 
6561 	for (i = 0; i < ARRAY_SIZE(virt_x2apic_mode_tests); i++) {
6562 		struct virt_x2apic_mode_test_case *virt_x2apic_mode_test_case =
6563 			&virt_x2apic_mode_tests[i];
6564 		struct virt_x2apic_mode_config *virt_x2apic_mode_config =
6565 			&virt_x2apic_mode_test_case->virt_x2apic_mode_config;
6566 		enum Config_type config_type;
6567 		u32 reg;
6568 
6569 		printf("--- %s test ---\n", virt_x2apic_mode_test_case->name);
6570 		config_type =
6571 			configure_virt_x2apic_mode_test(virt_x2apic_mode_config,
6572 							msr_bitmap_page);
6573 		if (config_type == CONFIG_TYPE_UNSUPPORTED) {
6574 			report_skip("Skip because of missing features.\n");
6575 			continue;
6576 		} else if (config_type == CONFIG_TYPE_VMENTRY_FAILS_EARLY) {
6577 			enter_guest_with_bad_controls();
6578 			continue;
6579 		}
6580 
6581 		for (reg = 0; reg < PAGE_SIZE / sizeof(u32); reg += 0x10) {
6582 			struct virt_x2apic_mode_expectation expectation;
6583 
6584 			virt_x2apic_mode_exit_expectation(
6585 				reg, virt_x2apic_mode_config, &expectation);
6586 
6587 			test_x2apic_rd(reg, &expectation, virtual_apic_page);
6588 			test_x2apic_wr(reg, &expectation, virtual_apic_page);
6589 		}
6590 	}
6591 
6592 
6593 	/* Terminate the guest */
6594 	vmcs_write(CPU_EXEC_CTRL0, cpu_exec_ctrl0);
6595 	vmcs_write(CPU_EXEC_CTRL1, cpu_exec_ctrl1);
6596 	args->op = X2APIC_TERMINATE;
6597 	enter_guest();
6598 	assert_exit_reason(VMX_VMCALL);
6599 }
6600 
6601 /*
6602  * On processors that support Intel 64 architecture, the IA32_SYSENTER_ESP
6603  * field and the IA32_SYSENTER_EIP field must each contain a canonical
6604  * address.
6605  *
6606  *  [Intel SDM]
6607  */
6608 static void test_sysenter_field(u32 field, const char *name)
6609 {
6610 	u64 addr_saved = vmcs_read(field);
6611 
6612 	vmcs_write(field, NONCANONICAL);
6613 	report_prefix_pushf("%s non-canonical", name);
6614 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD, false);
6615 	report_prefix_pop();
6616 
6617 	vmcs_write(field, 0xffffffff);
6618 	report_prefix_pushf("%s canonical", name);
6619 	test_vmx_vmlaunch(0, false);
6620 	report_prefix_pop();
6621 
6622 	vmcs_write(field, addr_saved);
6623 }
6624 
6625 static void test_ctl_reg(const char *cr_name, u64 cr, u64 fixed0, u64 fixed1)
6626 {
6627 	u64 val;
6628 	u64 cr_saved = vmcs_read(cr);
6629 	int i;
6630 
6631 	val = fixed0 & fixed1;
6632 	if (cr == HOST_CR4)
6633 		vmcs_write(cr, val | X86_CR4_PAE);
6634 	else
6635 		vmcs_write(cr, val);
6636 	report_prefix_pushf("%s %lx", cr_name, val);
6637 	if (val == fixed0)
6638 		test_vmx_vmlaunch(0, false);
6639 	else
6640 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
6641 				  false);
6642 	report_prefix_pop();
6643 
6644 	for (i = 0; i < 64; i++) {
6645 
6646 		/* Set a bit when the corresponding bit in fixed1 is 0 */
6647 		if ((fixed1 & (1ull << i)) == 0) {
6648 			if (cr == HOST_CR4 && ((1ull << i) & X86_CR4_SMEP ||
6649 					       (1ull << i) & X86_CR4_SMAP))
6650 				continue;
6651 
6652 			vmcs_write(cr, cr_saved | (1ull << i));
6653 			report_prefix_pushf("%s %llx", cr_name,
6654 						cr_saved | (1ull << i));
6655 			test_vmx_vmlaunch(
6656 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
6657 					false);
6658 			report_prefix_pop();
6659 		}
6660 
6661 		/* Unset a bit when the corresponding bit in fixed0 is 1 */
6662 		if (fixed0 & (1ull << i)) {
6663 			vmcs_write(cr, cr_saved & ~(1ull << i));
6664 			report_prefix_pushf("%s %llx", cr_name,
6665 						cr_saved & ~(1ull << i));
6666 			test_vmx_vmlaunch(
6667 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
6668 					false);
6669 			report_prefix_pop();
6670 		}
6671 	}
6672 
6673 	vmcs_write(cr, cr_saved);
6674 }
6675 
6676 /*
6677  * 1. The CR0 field must not set any bit to a value not supported in VMX
6678  *    operation.
6679  * 2. The CR4 field must not set any bit to a value not supported in VMX
6680  *    operation.
6681  * 3. On processors that support Intel 64 architecture, the CR3 field must
6682  *    be such that bits 63:52 and bits in the range 51:32 beyond the
6683  *    processor’s physical-address width must be 0.
6684  *
6685  *  [Intel SDM]
6686  */
6687 static void test_host_ctl_regs(void)
6688 {
6689 	u64 fixed0, fixed1, cr3, cr3_saved;
6690 	int i;
6691 
6692 	/* Test CR0 */
6693 	fixed0 = rdmsr(MSR_IA32_VMX_CR0_FIXED0);
6694 	fixed1 = rdmsr(MSR_IA32_VMX_CR0_FIXED1);
6695 	test_ctl_reg("HOST_CR0", HOST_CR0, fixed0, fixed1);
6696 
6697 	/* Test CR4 */
6698 	fixed0 = rdmsr(MSR_IA32_VMX_CR4_FIXED0);
6699 	fixed1 = rdmsr(MSR_IA32_VMX_CR4_FIXED1) &
6700 		 ~(X86_CR4_SMEP | X86_CR4_SMAP);
6701 	test_ctl_reg("HOST_CR4", HOST_CR4, fixed0, fixed1);
6702 
6703 	/* Test CR3 */
6704 	cr3_saved = vmcs_read(HOST_CR3);
6705 	for (i = cpuid_maxphyaddr(); i < 64; i++) {
6706 		cr3 = cr3_saved | (1ul << i);
6707 		vmcs_write(HOST_CR3, cr3);
6708 		report_prefix_pushf("HOST_CR3 %lx", cr3);
6709 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
6710 				  false);
6711 		report_prefix_pop();
6712 	}
6713 
6714 	vmcs_write(HOST_CR3, cr3_saved);
6715 }
6716 
6717 static void test_efer_bit(u32 fld, const char * fld_name, u32 ctrl_fld,
6718 			   u64 ctrl_bit, u64 efer_bit,
6719 			   const char *efer_bit_name)
6720 {
6721 	u64 efer_saved = vmcs_read(fld);
6722 	u32 ctrl_saved = vmcs_read(ctrl_fld);
6723 	u64 host_addr_size = ctrl_saved & EXI_HOST_64;
6724 	u64 efer;
6725 
6726 	vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit);
6727 	efer = efer_saved & ~efer_bit;
6728 	vmcs_write(fld, efer);
6729 	report_prefix_pushf("%s bit turned off, %s %lx", efer_bit_name,
6730 			    fld_name, efer);
6731 	test_vmx_vmlaunch(0, false);
6732 	report_prefix_pop();
6733 
6734 	efer = efer_saved | efer_bit;
6735 	vmcs_write(fld, efer);
6736 	report_prefix_pushf("%s bit turned on, %s %lx", efer_bit_name,
6737 			    fld_name, efer);
6738 	test_vmx_vmlaunch(0, false);
6739 	report_prefix_pop();
6740 
6741 	vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit);
6742 	efer = efer_saved & ~efer_bit;
6743 	vmcs_write(fld, efer);
6744 	report_prefix_pushf("%s bit turned off, %s %lx", efer_bit_name,
6745 			    fld_name, efer);
6746 	if (host_addr_size)
6747 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
6748 				  false);
6749 	else
6750 		test_vmx_vmlaunch(0, false);
6751 	report_prefix_pop();
6752 
6753 	efer = efer_saved | efer_bit;
6754 	vmcs_write(fld, efer);
6755 	report_prefix_pushf("%s bit turned on, %s %lx", efer_bit_name,
6756 			    fld_name, efer);
6757 	if (host_addr_size)
6758 		test_vmx_vmlaunch(0, false);
6759 	else
6760 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
6761 				  false);
6762 	report_prefix_pop();
6763 
6764 	vmcs_write(ctrl_fld, ctrl_saved);
6765 	vmcs_write(fld, efer_saved);
6766 }
6767 
6768 static void test_efer(u32 fld, const char * fld_name, u32 ctrl_fld,
6769 		      u64 ctrl_bit)
6770 {
6771 	u64 efer_saved = vmcs_read(fld);
6772 	u32 ctrl_saved = vmcs_read(ctrl_fld);
6773 	u64 efer_reserved_bits =  ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
6774 	u64 i;
6775 	u64 efer;
6776 
6777 	if (cpu_has_efer_nx())
6778 		efer_reserved_bits &= ~EFER_NX;
6779 
6780 	/*
6781 	 * Check reserved bits
6782 	 */
6783 	vmcs_write(ctrl_fld, ctrl_saved & ~ctrl_bit);
6784 	for (i = 0; i < 64; i++) {
6785 		if ((1ull << i) & efer_reserved_bits) {
6786 			efer = efer_saved | (1ull << i);
6787 			vmcs_write(fld, efer);
6788 			report_prefix_pushf("%s %lx", fld_name, efer);
6789 			test_vmx_vmlaunch(0, false);
6790 			report_prefix_pop();
6791 		}
6792 	}
6793 
6794 	vmcs_write(ctrl_fld, ctrl_saved | ctrl_bit);
6795 	for (i = 0; i < 64; i++) {
6796 		if ((1ull << i) & efer_reserved_bits) {
6797 			efer = efer_saved | (1ull << i);
6798 			vmcs_write(fld, efer);
6799 			report_prefix_pushf("%s %lx", fld_name, efer);
6800 			test_vmx_vmlaunch(
6801 				VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
6802 				false);
6803 			report_prefix_pop();
6804 		}
6805 	}
6806 
6807 	vmcs_write(ctrl_fld, ctrl_saved);
6808 	vmcs_write(fld, efer_saved);
6809 
6810 	/*
6811 	 * Check LMA and LME bits
6812 	 */
6813 	test_efer_bit(fld, fld_name, ctrl_fld, ctrl_bit, EFER_LMA,
6814 		      "EFER_LMA");
6815 	test_efer_bit(fld, fld_name, ctrl_fld, ctrl_bit, EFER_LME,
6816 		      "EFER_LME");
6817 }
6818 
6819 /*
6820  * If the “load IA32_EFER†VM-exit control is 1, bits reserved in the
6821  * IA32_EFER MSR must be 0 in the field for that register. In addition,
6822  * the values of the LMA and LME bits in the field must each be that of
6823  * the “host address-space size†VM-exit control.
6824  *
6825  *  [Intel SDM]
6826  */
6827 static void test_host_efer(void)
6828 {
6829 	if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
6830 		printf("\"Load-IA32-EFER\" exit control not supported\n");
6831 		return;
6832 	}
6833 
6834 	test_efer(HOST_EFER, "HOST_EFER", EXI_CONTROLS, EXI_LOAD_EFER);
6835 }
6836 
6837 /*
6838  * PAT values higher than 8 are uninteresting since they're likely lumped
6839  * in with "8". We only test values above 8 one bit at a time,
6840  * in order to reduce the number of VM-Entries and keep the runtime reasonable.
6841  */
6842 #define	PAT_VAL_LIMIT	8
6843 
6844 static void test_pat(u32 field, const char * field_name, u32 ctrl_field,
6845 		     u64 ctrl_bit)
6846 {
6847 	u32 ctrl_saved = vmcs_read(ctrl_field);
6848 	u64 pat_saved = vmcs_read(field);
6849 	u64 i, val;
6850 	u32 j;
6851 	int error;
6852 
6853 	vmcs_clear_bits(ctrl_field, ctrl_bit);
6854 	if (field == GUEST_PAT) {
6855 		vmx_set_test_stage(1);
6856 		test_set_guest(guest_state_test_main);
6857 	}
6858 
6859 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
6860 		/* Test PAT0..PAT7 fields */
6861 		for (j = 0; j < (i ? 8 : 1); j++) {
6862 			val = i << j * 8;
6863 			vmcs_write(field, val);
6864 			if (field == HOST_PAT) {
6865 				report_prefix_pushf("%s %lx", field_name, val);
6866 				test_vmx_vmlaunch(0, false);
6867 				report_prefix_pop();
6868 
6869 			} else {	// GUEST_PAT
6870 				enter_guest();
6871 				report_guest_state_test("ENT_LOAD_PAT enabled",
6872 							VMX_VMCALL, val,
6873 							"GUEST_PAT");
6874 			}
6875 		}
6876 	}
6877 
6878 	vmcs_set_bits(ctrl_field, ctrl_bit);
6879 	for (i = 0; i < 256; i = (i < PAT_VAL_LIMIT) ? i + 1 : i * 2) {
6880 		/* Test PAT0..PAT7 fields */
6881 		for (j = 0; j < (i ? 8 : 1); j++) {
6882 			val = i << j * 8;
6883 			vmcs_write(field, val);
6884 
6885 			if (field == HOST_PAT) {
6886 				report_prefix_pushf("%s %lx", field_name, val);
6887 				if (i == 0x2 || i == 0x3 || i >= 0x8)
6888 					error =
6889 					VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
6890 				else
6891 					error = 0;
6892 
6893 				test_vmx_vmlaunch(error, false);
6894 				report_prefix_pop();
6895 
6896 			} else {	// GUEST_PAT
6897 				if (i == 0x2 || i == 0x3 || i >= 0x8) {
6898 					enter_guest_with_invalid_guest_state();
6899 					report_guest_state_test("ENT_LOAD_PAT "
6900 							        "enabled",
6901 							        VMX_FAIL_STATE | VMX_ENTRY_FAILURE,
6902 							        val,
6903 							        "GUEST_PAT");
6904 				} else {
6905 					enter_guest();
6906 					report_guest_state_test("ENT_LOAD_PAT "
6907 							        "enabled",
6908 							        VMX_VMCALL,
6909 							        val,
6910 							        "GUEST_PAT");
6911 				}
6912 			}
6913 
6914 		}
6915 	}
6916 
6917 	if (field == GUEST_PAT) {
6918 		/*
6919 		 * Let the guest finish execution
6920 		 */
6921 		vmx_set_test_stage(2);
6922 		vmcs_write(field, pat_saved);
6923 		enter_guest();
6924 	}
6925 
6926 	vmcs_write(ctrl_field, ctrl_saved);
6927 	vmcs_write(field, pat_saved);
6928 }
6929 
6930 /*
6931  *  If the "load IA32_PAT" VM-exit control is 1, the value of the field
6932  *  for the IA32_PAT MSR must be one that could be written by WRMSR
6933  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
6934  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
6935  *  6 (WB), or 7 (UC-).
6936  *
6937  *  [Intel SDM]
6938  */
6939 static void test_load_host_pat(void)
6940 {
6941 	/*
6942 	 * "load IA32_PAT" VM-exit control
6943 	 */
6944 	if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT)) {
6945 		printf("\"Load-IA32-PAT\" exit control not supported\n");
6946 		return;
6947 	}
6948 
6949 	test_pat(HOST_PAT, "HOST_PAT", EXI_CONTROLS, EXI_LOAD_PAT);
6950 }
6951 
6952 /*
6953  * Test a value for the given VMCS field.
6954  *
6955  *  "field" - VMCS field
6956  *  "field_name" - string name of VMCS field
6957  *  "bit_start" - starting bit
6958  *  "bit_end" - ending bit
6959  *  "val" - value that the bit range must or must not contain
6960  *  "valid_val" - whether value given in 'val' must be valid or not
6961  *  "error" - expected VMCS error when vmentry fails for an invalid value
6962  */
6963 static void test_vmcs_field(u64 field, const char *field_name, u32 bit_start,
6964 			    u32 bit_end, u64 val, bool valid_val, u32 error)
6965 {
6966 	u64 field_saved = vmcs_read(field);
6967 	u32 i;
6968 	u64 tmp;
6969 	u32 bit_on;
6970 	u64 mask = ~0ull;
6971 
6972 	mask = (mask >> bit_end) << bit_end;
6973 	mask = mask | ((1 << bit_start) - 1);
6974 	tmp = (field_saved & mask) | (val << bit_start);
6975 
6976 	vmcs_write(field, tmp);
6977 	report_prefix_pushf("%s %lx", field_name, tmp);
6978 	if (valid_val)
6979 		test_vmx_vmlaunch(0, false);
6980 	else
6981 		test_vmx_vmlaunch(error, false);
6982 	report_prefix_pop();
6983 
6984 	for (i = bit_start; i <= bit_end; i = i + 2) {
6985 		bit_on = ((1ull < i) & (val << bit_start)) ? 0 : 1;
6986 		if (bit_on)
6987 			tmp = field_saved | (1ull << i);
6988 		else
6989 			tmp = field_saved & ~(1ull << i);
6990 		vmcs_write(field, tmp);
6991 		report_prefix_pushf("%s %lx", field_name, tmp);
6992 		if (valid_val)
6993 			test_vmx_vmlaunch(error, false);
6994 		else
6995 			test_vmx_vmlaunch(0, false);
6996 		report_prefix_pop();
6997 	}
6998 
6999 	vmcs_write(field, field_saved);
7000 }
7001 
7002 static void test_canonical(u64 field, const char * field_name)
7003 {
7004 	u64 addr_saved = vmcs_read(field);
7005 	u64 addr = addr_saved;
7006 
7007 	report_prefix_pushf("%s %lx", field_name, addr);
7008 	if (is_canonical(addr)) {
7009 		test_vmx_vmlaunch(0, false);
7010 		report_prefix_pop();
7011 
7012 		addr = make_non_canonical(addr);
7013 		vmcs_write(field, addr);
7014 		report_prefix_pushf("%s %lx", field_name, addr);
7015 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
7016 				  false);
7017 
7018 		vmcs_write(field, addr_saved);
7019 	} else {
7020 		test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD,
7021 				  false);
7022 	}
7023 	report_prefix_pop();
7024 }
7025 
7026 /*
7027  * 1. In the selector field for each of CS, SS, DS, ES, FS, GS and TR, the
7028  *    RPL (bits 1:0) and the TI flag (bit 2) must be 0.
7029  * 2. The selector fields for CS and TR cannot be 0000H.
7030  * 3. The selector field for SS cannot be 0000H if the "host address-space
7031  *    size" VM-exit control is 0.
7032  * 4. On processors that support Intel 64 architecture, the base-address
7033  *    fields for FS, GS and TR must contain canonical addresses.
7034  */
7035 static void test_host_segment_regs(void)
7036 {
7037 	u32 exit_ctrl_saved = vmcs_read(EXI_CONTROLS);
7038 	u16 selector_saved;
7039 
7040 	/*
7041 	 * Test RPL and TI flags
7042 	 */
7043 	test_vmcs_field(HOST_SEL_CS, "HOST_SEL_CS", 0, 2, 0x0, true,
7044 		     VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7045 	test_vmcs_field(HOST_SEL_SS, "HOST_SEL_SS", 0, 2, 0x0, true,
7046 		     VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7047 	test_vmcs_field(HOST_SEL_DS, "HOST_SEL_DS", 0, 2, 0x0, true,
7048 		     VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7049 	test_vmcs_field(HOST_SEL_ES, "HOST_SEL_ES", 0, 2, 0x0, true,
7050 		     VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7051 	test_vmcs_field(HOST_SEL_FS, "HOST_SEL_FS", 0, 2, 0x0, true,
7052 		     VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7053 	test_vmcs_field(HOST_SEL_GS, "HOST_SEL_GS", 0, 2, 0x0, true,
7054 		     VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7055 	test_vmcs_field(HOST_SEL_TR, "HOST_SEL_TR", 0, 2, 0x0, true,
7056 		     VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7057 
7058 	/*
7059 	 * Test that CS and TR fields can not be 0x0000
7060 	 */
7061 	test_vmcs_field(HOST_SEL_CS, "HOST_SEL_CS", 3, 15, 0x0000, false,
7062 			     VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7063 	test_vmcs_field(HOST_SEL_TR, "HOST_SEL_TR", 3, 15, 0x0000, false,
7064 			     VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7065 
7066 	/*
7067 	 * SS field can not be 0x0000 if "host address-space size" VM-exit
7068 	 * control is 0
7069 	 */
7070 	selector_saved = vmcs_read(HOST_SEL_SS);
7071 	vmcs_write(HOST_SEL_SS, 0);
7072 	if (exit_ctrl_saved & EXI_HOST_64) {
7073 		report_prefix_pushf("HOST_SEL_SS 0");
7074 		test_vmx_vmlaunch(0, false);
7075 		report_prefix_pop();
7076 
7077 		vmcs_write(EXI_CONTROLS, exit_ctrl_saved & ~EXI_HOST_64);
7078 	}
7079 
7080 	report_prefix_pushf("HOST_SEL_SS 0");
7081 	test_vmx_vmlaunch(VMXERR_ENTRY_INVALID_HOST_STATE_FIELD, false);
7082 	report_prefix_pop();
7083 
7084 	vmcs_write(HOST_SEL_SS, selector_saved);
7085 	vmcs_write(EXI_CONTROLS, exit_ctrl_saved);
7086 
7087 #ifdef __x86_64__
7088 	/*
7089 	 * Base address for FS, GS and TR must be canonical
7090 	 */
7091 	test_canonical(HOST_BASE_FS, "HOST_BASE_FS");
7092 	test_canonical(HOST_BASE_GS, "HOST_BASE_GS");
7093 	test_canonical(HOST_BASE_TR, "HOST_BASE_TR");
7094 #endif
7095 }
7096 
7097 /*
7098  *  On processors that support Intel 64 architecture, the base-address
7099  *  fields for GDTR and IDTR must contain canonical addresses.
7100  */
7101 static void test_host_desc_tables(void)
7102 {
7103 #ifdef __x86_64__
7104 	test_canonical(HOST_BASE_GDTR, "HOST_BASE_GDTR");
7105 	test_canonical(HOST_BASE_IDTR, "HOST_BASE_IDTR");
7106 #endif
7107 }
7108 
7109 /*
7110  * Check that the virtual CPU checks the VMX Host State Area as
7111  * documented in the Intel SDM.
7112  */
7113 static void vmx_host_state_area_test(void)
7114 {
7115 	/*
7116 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
7117 	 * fail due to invalid guest state, should we make it that
7118 	 * far.
7119 	 */
7120 	vmcs_write(GUEST_RFLAGS, 0);
7121 
7122 	test_host_ctl_regs();
7123 
7124 	test_sysenter_field(HOST_SYSENTER_ESP, "HOST_SYSENTER_ESP");
7125 	test_sysenter_field(HOST_SYSENTER_EIP, "HOST_SYSENTER_EIP");
7126 
7127 	test_host_efer();
7128 	test_load_host_pat();
7129 	test_host_segment_regs();
7130 	test_host_desc_tables();
7131 }
7132 
7133 /*
7134  *  If the "load IA32_PAT" VM-entry control is 1, the value of the field
7135  *  for the IA32_PAT MSR must be one that could be written by WRMSR
7136  *  without fault at CPL 0. Specifically, each of the 8 bytes in the
7137  *  field must have one of the values 0 (UC), 1 (WC), 4 (WT), 5 (WP),
7138  *  6 (WB), or 7 (UC-).
7139  *
7140  *  [Intel SDM]
7141  */
7142 static void test_load_guest_pat(void)
7143 {
7144 	/*
7145 	 * "load IA32_PAT" VM-entry control
7146 	 */
7147 	if (!(ctrl_exit_rev.clr & ENT_LOAD_PAT)) {
7148 		printf("\"Load-IA32-PAT\" entry control not supported\n");
7149 		return;
7150 	}
7151 
7152 	test_pat(GUEST_PAT, "GUEST_PAT", ENT_CONTROLS, ENT_LOAD_PAT);
7153 }
7154 
7155 /*
7156  * Check that the virtual CPU checks the VMX Guest State Area as
7157  * documented in the Intel SDM.
7158  */
7159 static void vmx_guest_state_area_test(void)
7160 {
7161 	test_load_guest_pat();
7162 }
7163 
7164 static bool valid_vmcs_for_vmentry(void)
7165 {
7166 	struct vmcs *current_vmcs = NULL;
7167 
7168 	if (vmcs_save(&current_vmcs))
7169 		return false;
7170 
7171 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
7172 }
7173 
7174 static void try_vmentry_in_movss_shadow(void)
7175 {
7176 	u32 vm_inst_err;
7177 	u32 flags;
7178 	bool early_failure = false;
7179 	u32 expected_flags = X86_EFLAGS_FIXED;
7180 	bool valid_vmcs = valid_vmcs_for_vmentry();
7181 
7182 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
7183 
7184 	/*
7185 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
7186 	 * unsupported VMCS component").
7187 	 */
7188 	vmcs_write(~0u, 0);
7189 
7190 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
7191 			      "vmwrite %%rsp, %%rdx;"
7192 			      "mov 0f, %%rax;"
7193 			      "mov %[host_rip], %%edx;"
7194 			      "vmwrite %%rax, %%rdx;"
7195 			      "mov $-1, %%ah;"
7196 			      "sahf;"
7197 			      "mov %%ss, %%ax;"
7198 			      "mov %%ax, %%ss;"
7199 			      "vmlaunch;"
7200 			      "mov $1, %[early_failure];"
7201 			      "0: lahf;"
7202 			      "movzbl %%ah, %[flags]"
7203 			      : [early_failure] "+r" (early_failure),
7204 				[flags] "=&a" (flags)
7205 			      : [host_rsp] "i" (HOST_RSP),
7206 				[host_rip] "i" (HOST_RIP)
7207 			      : "rdx", "cc", "memory");
7208 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
7209 
7210 	report("Early VM-entry failure", early_failure);
7211 	report("RFLAGS[8:0] is %x (actual %x)", flags == expected_flags,
7212 	       expected_flags, flags);
7213 	if (valid_vmcs)
7214 		report("VM-instruction error is %d (actual %d)",
7215 		       vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
7216 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
7217 }
7218 
7219 static void vmentry_movss_shadow_test(void)
7220 {
7221 	struct vmcs *orig_vmcs;
7222 
7223 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
7224 
7225 	/*
7226 	 * Set the launched flag on the current VMCS to verify the correct
7227 	 * error priority, below.
7228 	 */
7229 	test_set_guest(v2_null_test_guest);
7230 	enter_guest();
7231 
7232 	/*
7233 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
7234 	 * fail due to invalid guest state (if we make it that far).
7235 	 */
7236 	vmcs_write(GUEST_RFLAGS, 0);
7237 
7238 	/*
7239 	 * "VM entry with events blocked by MOV SS" takes precedence over
7240 	 * "VMLAUNCH with non-clear VMCS."
7241 	 */
7242 	report_prefix_push("valid current-VMCS");
7243 	try_vmentry_in_movss_shadow();
7244 	report_prefix_pop();
7245 
7246 	/*
7247 	 * VMfailInvalid takes precedence over "VM entry with events
7248 	 * blocked by MOV SS."
7249 	 */
7250 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
7251 	report_prefix_push("no current-VMCS");
7252 	try_vmentry_in_movss_shadow();
7253 	report_prefix_pop();
7254 
7255 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
7256 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
7257 }
7258 
7259 static void vmx_cr_load_test(void)
7260 {
7261 	unsigned long cr3, cr4, orig_cr3, orig_cr4;
7262 
7263 	orig_cr4 = read_cr4();
7264 	orig_cr3 = read_cr3();
7265 
7266 	if (!this_cpu_has(X86_FEATURE_PCID)) {
7267 		report_skip("PCID not detected");
7268 		return;
7269 	}
7270 	if (!this_cpu_has(X86_FEATURE_MCE)) {
7271 		report_skip("MCE not detected");
7272 		return;
7273 	}
7274 
7275 	TEST_ASSERT(!(orig_cr3 & X86_CR3_PCID_MASK));
7276 
7277 	/* Enable PCID for L1. */
7278 	cr4 = orig_cr4 | X86_CR4_PCIDE;
7279 	cr3 = orig_cr3 | 0x1;
7280 	TEST_ASSERT(!write_cr4_checking(cr4));
7281 	write_cr3(cr3);
7282 
7283 	test_set_guest(v2_null_test_guest);
7284 	vmcs_write(HOST_CR4, cr4);
7285 	vmcs_write(HOST_CR3, cr3);
7286 	enter_guest();
7287 
7288 	/*
7289 	 * No exception is expected.
7290 	 *
7291 	 * NB. KVM loads the last guest write to CR4 into CR4 read
7292 	 *     shadow. In order to trigger an exit to KVM, we can toggle a
7293 	 *     bit that is owned by KVM. We use CR4.MCE, which shall
7294 	 *     have no side effect because normally no guest MCE (e.g., as the
7295 	 *     result of bad memory) would happen during this test.
7296 	 */
7297 	TEST_ASSERT(!write_cr4_checking(cr4 ^ X86_CR4_MCE));
7298 
7299 	/* Cleanup L1 state. */
7300 	write_cr3(orig_cr3);
7301 	TEST_ASSERT(!write_cr4_checking(orig_cr4));
7302 }
7303 
7304 static void vmx_nm_test_guest(void)
7305 {
7306 	write_cr0(read_cr0() | X86_CR0_TS);
7307 	asm volatile("fnop");
7308 }
7309 
7310 static void check_nm_exit(const char *test)
7311 {
7312 	u32 reason = vmcs_read(EXI_REASON);
7313 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
7314 	const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
7315 		NM_VECTOR;
7316 
7317 	report("%s", reason == VMX_EXC_NMI && intr_info == expected, test);
7318 }
7319 
7320 /*
7321  * This test checks that:
7322  *
7323  * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then
7324  *     a subsequent #NM VM-exit is reflected to L1.
7325  *
7326  * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a
7327  *     subsequent #NM VM-exit is reflected to L1.
7328  */
7329 static void vmx_nm_test(void)
7330 {
7331 	unsigned long cr0 = read_cr0();
7332 
7333 	test_set_guest(vmx_nm_test_guest);
7334 
7335 	/*
7336 	 * L1 wants to intercept #NM exceptions encountered in L2.
7337 	 */
7338 	vmcs_write(EXC_BITMAP, 1 << NM_VECTOR);
7339 
7340 	/*
7341 	 * Launch L2 with CR0.TS clear, but don't claim host ownership of
7342 	 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop,
7343 	 * which will raise #NM. L0 should reflect the #NM VM-exit to L1.
7344 	 */
7345 	vmcs_write(CR0_MASK, 0);
7346 	vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS);
7347 	enter_guest();
7348 	check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1");
7349 
7350 	/*
7351 	 * Re-enter L2 at the fnop instruction, with CR0.TS clear but
7352 	 * CR0.EM set. The fnop will still raise #NM, and L0 should
7353 	 * reflect the #NM VM-exit to L1.
7354 	 */
7355 	vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM);
7356 	enter_guest();
7357 	check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1");
7358 
7359 	/*
7360 	 * Re-enter L2 at the fnop instruction, with both CR0.TS and
7361 	 * CR0.EM clear. There will be no #NM, and the L2 guest should
7362 	 * exit normally.
7363 	 */
7364 	vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM));
7365 	enter_guest();
7366 }
7367 
7368 bool vmx_pending_event_ipi_fired;
7369 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
7370 {
7371 	vmx_pending_event_ipi_fired = true;
7372 	eoi();
7373 }
7374 
7375 bool vmx_pending_event_guest_run;
7376 static void vmx_pending_event_guest(void)
7377 {
7378 	vmcall();
7379 	vmx_pending_event_guest_run = true;
7380 }
7381 
7382 static void vmx_pending_event_test_core(bool guest_hlt)
7383 {
7384 	int ipi_vector = 0xf1;
7385 
7386 	vmx_pending_event_ipi_fired = false;
7387 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
7388 
7389 	vmx_pending_event_guest_run = false;
7390 	test_set_guest(vmx_pending_event_guest);
7391 
7392 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
7393 
7394 	enter_guest();
7395 	skip_exit_vmcall();
7396 
7397 	if (guest_hlt)
7398 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7399 
7400 	irq_disable();
7401 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
7402 				   APIC_DM_FIXED | ipi_vector,
7403 				   0);
7404 
7405 	enter_guest();
7406 
7407 	assert_exit_reason(VMX_EXTINT);
7408 	report("Guest did not run before host received IPI",
7409 		   !vmx_pending_event_guest_run);
7410 
7411 	irq_enable();
7412 	asm volatile ("nop");
7413 	irq_disable();
7414 	report("Got pending interrupt after IRQ enabled",
7415 		   vmx_pending_event_ipi_fired);
7416 
7417 	if (guest_hlt)
7418 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
7419 
7420 	enter_guest();
7421 	report("Guest finished running when no interrupt",
7422 		   vmx_pending_event_guest_run);
7423 }
7424 
7425 static void vmx_pending_event_test(void)
7426 {
7427 	vmx_pending_event_test_core(false);
7428 }
7429 
7430 static void vmx_pending_event_hlt_test(void)
7431 {
7432 	vmx_pending_event_test_core(true);
7433 }
7434 
7435 static int vmx_window_test_db_count;
7436 
7437 static void vmx_window_test_db_handler(struct ex_regs *regs)
7438 {
7439 	vmx_window_test_db_count++;
7440 }
7441 
7442 static void vmx_nmi_window_test_guest(void)
7443 {
7444 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
7445 
7446 	asm volatile("vmcall\n\t"
7447 		     "nop\n\t");
7448 
7449 	handle_exception(DB_VECTOR, NULL);
7450 }
7451 
7452 static void verify_nmi_window_exit(u64 rip)
7453 {
7454 	u32 exit_reason = vmcs_read(EXI_REASON);
7455 
7456 	report("Exit reason (%d) is 'NMI window'",
7457 	       exit_reason == VMX_NMI_WINDOW, exit_reason);
7458 	report("RIP (%#lx) is %#lx", vmcs_read(GUEST_RIP) == rip,
7459 	       vmcs_read(GUEST_RIP), rip);
7460 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
7461 }
7462 
7463 static void vmx_nmi_window_test(void)
7464 {
7465 	u64 nop_addr;
7466 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
7467 
7468 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
7469 		report_skip("CPU does not support the \"Virtual NMIs\" VM-execution control.");
7470 		return;
7471 	}
7472 
7473 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
7474 		report_skip("CPU does not support the \"NMI-window exiting\" VM-execution control.");
7475 		return;
7476 	}
7477 
7478 	vmx_window_test_db_count = 0;
7479 
7480 	report_prefix_push("NMI-window");
7481 	test_set_guest(vmx_nmi_window_test_guest);
7482 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
7483 	enter_guest();
7484 	skip_exit_vmcall();
7485 	nop_addr = vmcs_read(GUEST_RIP);
7486 
7487 	/*
7488 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
7489 	 * RIP will not advance.
7490 	 */
7491 	report_prefix_push("active, no blocking");
7492 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
7493 	enter_guest();
7494 	verify_nmi_window_exit(nop_addr);
7495 	report_prefix_pop();
7496 
7497 	/*
7498 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
7499 	 * a VM-exit on the next instruction after the nop. (The nop
7500 	 * is one byte.)
7501 	 */
7502 	report_prefix_push("active, blocking by MOV-SS");
7503 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
7504 	enter_guest();
7505 	verify_nmi_window_exit(nop_addr + 1);
7506 	report_prefix_pop();
7507 
7508 	/*
7509 	 * Ask for "NMI-window exiting" (with event injection), and
7510 	 * expect a VM-exit after the event is injected. (RIP should
7511 	 * be at the address specified in the IDT entry for #DB.)
7512 	 */
7513 	report_prefix_push("active, no blocking, injecting #DB");
7514 	vmcs_write(ENT_INTR_INFO,
7515 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
7516 	enter_guest();
7517 	verify_nmi_window_exit((u64)db_fault_addr);
7518 	report_prefix_pop();
7519 
7520 	/*
7521 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
7522 	 * a VM-exit after the next IRET (i.e. after the #DB handler
7523 	 * returns). So, RIP should be back at one byte past the nop.
7524 	 */
7525 	report_prefix_push("active, blocking by NMI");
7526 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
7527 	enter_guest();
7528 	verify_nmi_window_exit(nop_addr + 1);
7529 	report("#DB handler executed once (actual %d times)",
7530 	       vmx_window_test_db_count == 1,
7531 	       vmx_window_test_db_count);
7532 	report_prefix_pop();
7533 
7534 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
7535 		report_skip("CPU does not support activity state HLT.");
7536 	} else {
7537 		/*
7538 		 * Ask for "NMI-window exiting" when entering activity
7539 		 * state HLT, and expect an immediate VM-exit. RIP is
7540 		 * still one byte past the nop.
7541 		 */
7542 		report_prefix_push("halted, no blocking");
7543 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7544 		enter_guest();
7545 		verify_nmi_window_exit(nop_addr + 1);
7546 		report_prefix_pop();
7547 
7548 		/*
7549 		 * Ask for "NMI-window exiting" when entering activity
7550 		 * state HLT (with event injection), and expect a
7551 		 * VM-exit after the event is injected. (RIP should be
7552 		 * at the address specified in the IDT entry for #DB.)
7553 		 */
7554 		report_prefix_push("halted, no blocking, injecting #DB");
7555 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7556 		vmcs_write(ENT_INTR_INFO,
7557 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
7558 			   DB_VECTOR);
7559 		enter_guest();
7560 		verify_nmi_window_exit((u64)db_fault_addr);
7561 		report_prefix_pop();
7562 	}
7563 
7564 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
7565 	enter_guest();
7566 	report_prefix_pop();
7567 }
7568 
7569 static void vmx_intr_window_test_guest(void)
7570 {
7571 	handle_exception(DB_VECTOR, vmx_window_test_db_handler);
7572 
7573 	/*
7574 	 * The two consecutive STIs are to ensure that only the first
7575 	 * one has a shadow. Note that NOP and STI are one byte
7576 	 * instructions.
7577 	 */
7578 	asm volatile("vmcall\n\t"
7579 		     "nop\n\t"
7580 		     "sti\n\t"
7581 		     "sti\n\t");
7582 
7583 	handle_exception(DB_VECTOR, NULL);
7584 }
7585 
7586 static void verify_intr_window_exit(u64 rip)
7587 {
7588 	u32 exit_reason = vmcs_read(EXI_REASON);
7589 
7590 	report("Exit reason (%d) is 'interrupt window'",
7591 	       exit_reason == VMX_INTR_WINDOW, exit_reason);
7592 	report("RIP (%#lx) is %#lx", vmcs_read(GUEST_RIP) == rip,
7593 	       vmcs_read(GUEST_RIP), rip);
7594 	vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
7595 }
7596 
7597 static void vmx_intr_window_test(void)
7598 {
7599 	u64 vmcall_addr;
7600 	u64 nop_addr;
7601 	unsigned int orig_db_gate_type;
7602 	void *db_fault_addr = get_idt_addr(&boot_idt[DB_VECTOR]);
7603 
7604 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
7605 		report_skip("CPU does not support the \"interrupt-window exiting\" VM-execution control.");
7606 		return;
7607 	}
7608 
7609 	/*
7610 	 * Change the IDT entry for #DB from interrupt gate to trap gate,
7611 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
7612 	 * be disabled after vectoring a #DB.
7613 	 */
7614 	orig_db_gate_type = boot_idt[DB_VECTOR].type;
7615 	boot_idt[DB_VECTOR].type = 15;
7616 
7617 	report_prefix_push("interrupt-window");
7618 	test_set_guest(vmx_intr_window_test_guest);
7619 	enter_guest();
7620 	assert_exit_reason(VMX_VMCALL);
7621 	vmcall_addr = vmcs_read(GUEST_RIP);
7622 
7623 	/*
7624 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
7625 	 * no blocking; expect an immediate VM-exit. Note that we have
7626 	 * not advanced past the vmcall instruction yet, so RIP should
7627 	 * point to the vmcall instruction.
7628 	 */
7629 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
7630 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
7631 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
7632 	enter_guest();
7633 	verify_intr_window_exit(vmcall_addr);
7634 	report_prefix_pop();
7635 
7636 	/*
7637 	 * Ask for "interrupt-window exiting" (with event injection)
7638 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
7639 	 * the event is injected. That is, RIP should should be at the
7640 	 * address specified in the IDT entry for #DB.
7641 	 */
7642 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #DB");
7643 	vmcs_write(ENT_INTR_INFO,
7644 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | DB_VECTOR);
7645 	vmcall_addr = vmcs_read(GUEST_RIP);
7646 	enter_guest();
7647 	verify_intr_window_exit((u64)db_fault_addr);
7648 	report_prefix_pop();
7649 
7650 	/*
7651 	 * Let the L2 guest run through the IRET, back to the VMCALL.
7652 	 * We have to clear the "interrupt-window exiting"
7653 	 * VM-execution control, or it would just keep causing
7654 	 * VM-exits. Then, advance past the VMCALL and set the
7655 	 * "interrupt-window exiting" VM-execution control again.
7656 	 */
7657 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
7658 	enter_guest();
7659 	skip_exit_vmcall();
7660 	nop_addr = vmcs_read(GUEST_RIP);
7661 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
7662 
7663 	/*
7664 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
7665 	 * RFLAGS.IF set, and expect a VM-exit on the next
7666 	 * instruction. (NOP is one byte.)
7667 	 */
7668 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
7669 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
7670 	enter_guest();
7671 	verify_intr_window_exit(nop_addr + 1);
7672 	report_prefix_pop();
7673 
7674 	/*
7675 	 * Back up to the NOP and ask for "interrupt-window exiting"
7676 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
7677 	 * on the next instruction. (NOP is one byte.)
7678 	 */
7679 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
7680 	vmcs_write(GUEST_RIP, nop_addr);
7681 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
7682 	enter_guest();
7683 	verify_intr_window_exit(nop_addr + 1);
7684 	report_prefix_pop();
7685 
7686 	/*
7687 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
7688 	 * and expect a VM-exit on the instruction following the STI
7689 	 * shadow. Only the first STI (which is one byte past the NOP)
7690 	 * should have a shadow. The second STI (which is two bytes
7691 	 * past the NOP) has no shadow. Therefore, the interrupt
7692 	 * window opens at three bytes past the NOP.
7693 	 */
7694 	report_prefix_push("active, RFLAGS.IF = 0");
7695 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
7696 	enter_guest();
7697 	verify_intr_window_exit(nop_addr + 3);
7698 	report_prefix_pop();
7699 
7700 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
7701 		report_skip("CPU does not support activity state HLT.");
7702 	} else {
7703 		/*
7704 		 * Ask for "interrupt-window exiting" when entering
7705 		 * activity state HLT, and expect an immediate
7706 		 * VM-exit. RIP is still three bytes past the nop.
7707 		 */
7708 		report_prefix_push("halted, no blocking");
7709 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7710 		enter_guest();
7711 		verify_intr_window_exit(nop_addr + 3);
7712 		report_prefix_pop();
7713 
7714 		/*
7715 		 * Ask for "interrupt-window exiting" when entering
7716 		 * activity state HLT (with event injection), and
7717 		 * expect a VM-exit after the event is injected. That
7718 		 * is, RIP should should be at the address specified
7719 		 * in the IDT entry for #DB.
7720 		 */
7721 		report_prefix_push("halted, no blocking, injecting #DB");
7722 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
7723 		vmcs_write(ENT_INTR_INFO,
7724 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
7725 			   DB_VECTOR);
7726 		enter_guest();
7727 		verify_intr_window_exit((u64)db_fault_addr);
7728 		report_prefix_pop();
7729 	}
7730 
7731 	boot_idt[DB_VECTOR].type = orig_db_gate_type;
7732 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
7733 	enter_guest();
7734 	report_prefix_pop();
7735 }
7736 
7737 #define GUEST_TSC_OFFSET (1u << 30)
7738 
7739 static u64 guest_tsc;
7740 
7741 static void vmx_store_tsc_test_guest(void)
7742 {
7743 	guest_tsc = rdtsc();
7744 }
7745 
7746 /*
7747  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
7748  * list, the value saved is not subject to the TSC offset that is
7749  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
7750  */
7751 static void vmx_store_tsc_test(void)
7752 {
7753 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
7754 	u64 low, high;
7755 
7756 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
7757 		report_skip("'Use TSC offsetting' not supported");
7758 		return;
7759 	}
7760 
7761 	test_set_guest(vmx_store_tsc_test_guest);
7762 
7763 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
7764 	vmcs_write(EXI_MSR_ST_CNT, 1);
7765 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
7766 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
7767 
7768 	low = rdtsc();
7769 	enter_guest();
7770 	high = rdtsc();
7771 
7772 	report("RDTSC value in the guest (%lu) is in range [%lu, %lu]",
7773 	       low + GUEST_TSC_OFFSET <= guest_tsc &&
7774 	       guest_tsc <= high + GUEST_TSC_OFFSET,
7775 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
7776 	report("IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
7777 	       low <= msr_entry.value && msr_entry.value <= high,
7778 	       msr_entry.value, low, high);
7779 }
7780 
7781 static void vmx_db_test_guest(void)
7782 {
7783 	/*
7784 	 * For a hardware generated single-step #DB.
7785 	 */
7786 	asm volatile("vmcall;"
7787 		     "nop;"
7788 		     ".Lpost_nop:");
7789 	/*
7790 	 * ...in a MOVSS shadow, with pending debug exceptions.
7791 	 */
7792 	asm volatile("vmcall;"
7793 		     "nop;"
7794 		     ".Lpost_movss_nop:");
7795 	/*
7796 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
7797 	 * emulates it in software.)
7798 	 */
7799 	asm volatile("vmcall;"
7800 		     "wbinvd;"
7801 		     ".Lpost_wbinvd:");
7802 	/*
7803 	 * ...in a MOVSS shadow, with pending debug exceptions.
7804 	 */
7805 	asm volatile("vmcall;"
7806 		     "wbinvd;"
7807 		     ".Lpost_movss_wbinvd:");
7808 	/*
7809 	 * For a hardware generated single-step #DB in a transactional region.
7810 	 */
7811 	asm volatile("vmcall;"
7812 		     ".Lxbegin: xbegin .Lskip_rtm;"
7813 		     "xend;"
7814 		     ".Lskip_rtm:");
7815 }
7816 
7817 /*
7818  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
7819  * L2. No #DB is delivered and L2 continues to the next point of
7820  * interest.
7821  */
7822 static void dismiss_db(void)
7823 {
7824 	vmcs_write(GUEST_PENDING_DEBUG, 0);
7825 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
7826 	enter_guest();
7827 }
7828 
7829 /*
7830  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
7831  * Then throw away the #DB exception and resume L2.
7832  */
7833 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
7834 			  void *expected_rip, u64 expected_exit_qual,
7835 			  u64 expected_dr6)
7836 {
7837 	u32 reason = vmcs_read(EXI_REASON);
7838 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
7839 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
7840 	u64 guest_rip = vmcs_read(GUEST_RIP);
7841 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
7842 	u64 dr6 = read_dr6();
7843 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
7844 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
7845 
7846 	report("Expected #DB VM-exit",
7847 	       reason == VMX_EXC_NMI && intr_info == expected_intr_info);
7848 	report("Expected RIP %p (actual %lx)", (u64)expected_rip == guest_rip,
7849 	       expected_rip, guest_rip);
7850 	report_xfail("Expected pending debug exceptions 0 (actual %lx)",
7851 		     xfail_pdbg, 0 == guest_pending_dbg, guest_pending_dbg);
7852 	report_xfail("Expected exit qualification %lx (actual %lx)", xfail_qual,
7853 		     expected_exit_qual == exit_qual,
7854 		     expected_exit_qual, exit_qual);
7855 	report_xfail("Expected DR6 %lx (actual %lx)", xfail_dr6,
7856 		     expected_dr6 == dr6, expected_dr6, dr6);
7857 	dismiss_db();
7858 }
7859 
7860 /*
7861  * Assuming the guest has just exited on a VMCALL instruction, skip
7862  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
7863  * pending debug exceptions are non-zero, set the VMCS up as if the
7864  * previous instruction was a MOVSS that generated the indicated
7865  * pending debug exceptions. Then enter L2.
7866  */
7867 static void single_step_guest(const char *test_name, u64 starting_dr6,
7868 			      u64 pending_debug_exceptions)
7869 {
7870 	printf("\n%s\n", test_name);
7871 	skip_exit_vmcall();
7872 	write_dr6(starting_dr6);
7873 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
7874 	if (pending_debug_exceptions) {
7875 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
7876 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
7877 	}
7878 	enter_guest();
7879 }
7880 
7881 /*
7882  * When L1 intercepts #DB, verify that a single-step trap clears
7883  * pending debug exceptions, populates the exit qualification field
7884  * properly, and that DR6 is not prematurely clobbered. In a
7885  * (simulated) MOVSS shadow, make sure that the pending debug
7886  * exception bits are properly accumulated into the exit qualification
7887  * field.
7888  */
7889 static void vmx_db_test(void)
7890 {
7891 	/*
7892 	 * We are going to set a few arbitrary bits in DR6 to verify that
7893 	 * (a) DR6 is not modified by an intercepted #DB, and
7894 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
7895          *     the exit qualification field for a subsequent #DB exception.
7896 	 */
7897 	const u64 starting_dr6 = DR6_RESERVED | BIT(13) | DR_TRAP3 | DR_TRAP1;
7898 	extern char post_nop asm(".Lpost_nop");
7899 	extern char post_movss_nop asm(".Lpost_movss_nop");
7900 	extern char post_wbinvd asm(".Lpost_wbinvd");
7901 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
7902 	extern char xbegin asm(".Lxbegin");
7903 	extern char skip_rtm asm(".Lskip_rtm");
7904 
7905 	/*
7906 	 * L1 wants to intercept #DB exceptions encountered in L2.
7907 	 */
7908 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
7909 
7910 	/*
7911 	 * Start L2 and run it up to the first point of interest.
7912 	 */
7913 	test_set_guest(vmx_db_test_guest);
7914 	enter_guest();
7915 
7916 	/*
7917 	 * Hardware-delivered #DB trap for single-step sets the
7918 	 * standard that L0 has to follow for emulated instructions.
7919 	 */
7920 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
7921 	check_db_exit(false, false, false, &post_nop, DR_STEP, starting_dr6);
7922 
7923 	/*
7924 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
7925 	 * also sets the standard that L0 has to follow for emulated
7926 	 * instructions. Here, we establish the VMCS pending debug
7927 	 * exceptions to indicate that the simulated MOVSS triggered a
7928 	 * data breakpoint as well as the single-step trap.
7929 	 */
7930 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
7931 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0 );
7932 	check_db_exit(false, false, false, &post_movss_nop, DR_STEP | DR_TRAP0,
7933 		      starting_dr6);
7934 
7935 	/*
7936 	 * L0 synthesized #DB trap for single-step is buggy, because
7937 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
7938 	 * reconstitute the exit qualification from the prematurely
7939 	 * modified DR6, but fails miserably.
7940 	 */
7941 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
7942 	check_db_exit(true, true, false, &post_wbinvd, DR_STEP, starting_dr6);
7943 
7944 	/*
7945 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
7946 	 * even worse, because L0 also leaves the pending debug
7947 	 * exceptions in the VMCS instead of accumulating them into
7948 	 * the exit qualification field for the #DB exception.
7949 	 */
7950 	single_step_guest("Software synthesized single-step in MOVSS shadow",
7951 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0);
7952 	check_db_exit(true, true, true, &post_movss_wbinvd, DR_STEP | DR_TRAP0,
7953 		      starting_dr6);
7954 
7955 	/*
7956 	 * Optional RTM test for hardware that supports RTM, to
7957 	 * demonstrate that the current volume 3 of the SDM
7958 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
7959 	 * qualification for debug exceptions is not reserved. It is
7960 	 * set to 1 if a debug exception (#DB) or a breakpoint
7961 	 * exception (#BP) occurs inside an RTM region while advanced
7962 	 * debugging of RTM transactional regions is enabled.
7963 	 */
7964 	if (this_cpu_has(X86_FEATURE_RTM)) {
7965 		vmcs_write(ENT_CONTROLS,
7966 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
7967 		/*
7968 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
7969 		 * in the guest to enable advanced debugging of RTM
7970 		 * transactional regions.
7971 		 */
7972 		vmcs_write(GUEST_DR7, BIT(11));
7973 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
7974 		single_step_guest("Hardware delivered single-step in "
7975 				  "transactional region", starting_dr6, 0);
7976 		check_db_exit(false, false, false, &xbegin, BIT(16),
7977 			      starting_dr6);
7978 	} else {
7979 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
7980 		enter_guest();
7981 	}
7982 }
7983 
7984 static void enable_vid(void)
7985 {
7986 	void *virtual_apic_page;
7987 
7988 	assert(cpu_has_apicv());
7989 
7990 	disable_intercept_for_x2apic_msrs();
7991 
7992 	virtual_apic_page = alloc_page();
7993 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
7994 
7995 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
7996 
7997 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
7998 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
7999 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
8000 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
8001 
8002 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
8003 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
8004 }
8005 
8006 static void trigger_ioapic_scan_thread(void *data)
8007 {
8008 	/* Wait until other CPU entered L2 */
8009 	while (vmx_get_test_stage() != 1)
8010 		;
8011 
8012 	/* Trigger ioapic scan */
8013 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
8014 	vmx_set_test_stage(2);
8015 }
8016 
8017 static void irq_79_handler_guest(isr_regs_t *regs)
8018 {
8019 	eoi();
8020 
8021 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
8022 	vmcall();
8023 }
8024 
8025 /*
8026  * Constant for num of busy-loop iterations after which
8027  * a timer interrupt should have happened in host
8028  */
8029 #define TIMER_INTERRUPT_DELAY 100000000
8030 
8031 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
8032 {
8033 	handle_irq(0x79, irq_79_handler_guest);
8034 	irq_enable();
8035 
8036 	/* Signal to L1 CPU to trigger ioapic scan */
8037 	vmx_set_test_stage(1);
8038 	/* Wait until L1 CPU to trigger ioapic scan */
8039 	while (vmx_get_test_stage() != 2)
8040 		;
8041 
8042 	/*
8043 	 * Wait for L0 timer interrupt to be raised while we run in L2
8044 	 * such that L0 will process the IOAPIC scan request before
8045 	 * resuming L2
8046 	 */
8047 	delay(TIMER_INTERRUPT_DELAY);
8048 
8049 	asm volatile ("int $0x79");
8050 }
8051 
8052 static void vmx_eoi_bitmap_ioapic_scan_test(void)
8053 {
8054 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
8055 		report_skip(__func__);
8056 		return;
8057 	}
8058 
8059 	enable_vid();
8060 
8061 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
8062 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
8063 
8064 	/*
8065 	 * Launch L2.
8066 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
8067 	 * In case the reason isn't VMX_VMCALL, the asserion inside
8068 	 * skip_exit_vmcall() will fail.
8069 	 */
8070 	enter_guest();
8071 	skip_exit_vmcall();
8072 
8073 	/* Let L2 finish */
8074 	enter_guest();
8075 	report(__func__, 1);
8076 }
8077 
8078 #define HLT_WITH_RVI_VECTOR		(0xf1)
8079 
8080 bool vmx_hlt_with_rvi_guest_isr_fired;
8081 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
8082 {
8083 	vmx_hlt_with_rvi_guest_isr_fired = true;
8084 	eoi();
8085 }
8086 
8087 static void vmx_hlt_with_rvi_guest(void)
8088 {
8089 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
8090 
8091 	irq_enable();
8092 	asm volatile ("nop");
8093 
8094 	vmcall();
8095 }
8096 
8097 static void vmx_hlt_with_rvi_test(void)
8098 {
8099 	if (!cpu_has_apicv()) {
8100 		report_skip(__func__);
8101 		return;
8102 	}
8103 
8104 	enable_vid();
8105 
8106 	vmx_hlt_with_rvi_guest_isr_fired = false;
8107 	test_set_guest(vmx_hlt_with_rvi_guest);
8108 
8109 	enter_guest();
8110 	skip_exit_vmcall();
8111 
8112 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
8113 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
8114 	enter_guest();
8115 
8116 	report("Interrupt raised in guest", vmx_hlt_with_rvi_guest_isr_fired);
8117 }
8118 
8119 static void set_irq_line_thread(void *data)
8120 {
8121 	/* Wait until other CPU entered L2 */
8122 	while (vmx_get_test_stage() != 1)
8123 		;
8124 
8125 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
8126 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
8127 	vmx_set_test_stage(2);
8128 }
8129 
8130 static bool irq_78_handler_vmcall_before_eoi;
8131 static void irq_78_handler_guest(isr_regs_t *regs)
8132 {
8133 	set_irq_line(0xf, 0);
8134 	if (irq_78_handler_vmcall_before_eoi)
8135 		vmcall();
8136 	eoi();
8137 	vmcall();
8138 }
8139 
8140 static void vmx_apic_passthrough_guest(void)
8141 {
8142 	handle_irq(0x78, irq_78_handler_guest);
8143 	irq_enable();
8144 
8145 	/* If requested, wait for other CPU to trigger ioapic scan */
8146 	if (vmx_get_test_stage() < 1) {
8147 		vmx_set_test_stage(1);
8148 		while (vmx_get_test_stage() != 2)
8149 			;
8150 	}
8151 
8152 	set_irq_line(0xf, 1);
8153 }
8154 
8155 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
8156 {
8157 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
8158 		report_skip(__func__);
8159 		return;
8160 	}
8161 
8162 	u64 cpu_ctrl_0 = CPU_SECONDARY;
8163 	u64 cpu_ctrl_1 = 0;
8164 
8165 	disable_intercept_for_x2apic_msrs();
8166 
8167 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
8168 
8169 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
8170 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
8171 
8172 	if (set_irq_line_from_thread) {
8173 		irq_78_handler_vmcall_before_eoi = false;
8174 		on_cpu_async(1, set_irq_line_thread, NULL);
8175 	} else {
8176 		irq_78_handler_vmcall_before_eoi = true;
8177 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
8178 		vmx_set_test_stage(2);
8179 	}
8180 	test_set_guest(vmx_apic_passthrough_guest);
8181 
8182 	if (irq_78_handler_vmcall_before_eoi) {
8183 		/* Before EOI remote_irr should still be set */
8184 		enter_guest();
8185 		skip_exit_vmcall();
8186 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
8187 			"IOAPIC pass-through: remote_irr=1 before EOI");
8188 	}
8189 
8190 	/* After EOI remote_irr should be cleared */
8191 	enter_guest();
8192 	skip_exit_vmcall();
8193 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
8194 		"IOAPIC pass-through: remote_irr=0 after EOI");
8195 
8196 	/* Let L2 finish */
8197 	enter_guest();
8198 	report(__func__, 1);
8199 }
8200 
8201 static void vmx_apic_passthrough_test(void)
8202 {
8203 	vmx_apic_passthrough(false);
8204 }
8205 
8206 static void vmx_apic_passthrough_thread_test(void)
8207 {
8208 	vmx_apic_passthrough(true);
8209 }
8210 
8211 enum vmcs_access {
8212 	ACCESS_VMREAD,
8213 	ACCESS_VMWRITE,
8214 	ACCESS_NONE,
8215 };
8216 
8217 struct vmcs_shadow_test_common {
8218 	enum vmcs_access op;
8219 	enum Reason reason;
8220 	u64 field;
8221 	u64 value;
8222 	u64 flags;
8223 	u64 time;
8224 } l1_l2_common;
8225 
8226 static inline u64 vmread_flags(u64 field, u64 *val)
8227 {
8228 	u64 flags;
8229 
8230 	asm volatile ("vmread %2, %1; pushf; pop %0"
8231 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
8232 	return flags & X86_EFLAGS_ALU;
8233 }
8234 
8235 static inline u64 vmwrite_flags(u64 field, u64 val)
8236 {
8237 	u64 flags;
8238 
8239 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
8240 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
8241 	return flags & X86_EFLAGS_ALU;
8242 }
8243 
8244 static void vmx_vmcs_shadow_test_guest(void)
8245 {
8246 	struct vmcs_shadow_test_common *c = &l1_l2_common;
8247 	u64 start;
8248 
8249 	while (c->op != ACCESS_NONE) {
8250 		start = rdtsc();
8251 		switch (c->op) {
8252 		default:
8253 			c->flags = -1ull;
8254 			break;
8255 		case ACCESS_VMREAD:
8256 			c->flags = vmread_flags(c->field, &c->value);
8257 			break;
8258 		case ACCESS_VMWRITE:
8259 			c->flags = vmwrite_flags(c->field, 0);
8260 			break;
8261 		}
8262 		c->time = rdtsc() - start;
8263 		vmcall();
8264 	}
8265 }
8266 
8267 static u64 vmread_from_shadow(u64 field)
8268 {
8269 	struct vmcs *primary;
8270 	struct vmcs *shadow;
8271 	u64 value;
8272 
8273 	TEST_ASSERT(!vmcs_save(&primary));
8274 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
8275 	TEST_ASSERT(!make_vmcs_current(shadow));
8276 	value = vmcs_read(field);
8277 	TEST_ASSERT(!make_vmcs_current(primary));
8278 	return value;
8279 }
8280 
8281 static u64 vmwrite_to_shadow(u64 field, u64 value)
8282 {
8283 	struct vmcs *primary;
8284 	struct vmcs *shadow;
8285 
8286 	TEST_ASSERT(!vmcs_save(&primary));
8287 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
8288 	TEST_ASSERT(!make_vmcs_current(shadow));
8289 	vmcs_write(field, value);
8290 	value = vmcs_read(field);
8291 	TEST_ASSERT(!make_vmcs_current(primary));
8292 	return value;
8293 }
8294 
8295 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
8296 {
8297 	struct vmcs_shadow_test_common *c = &l1_l2_common;
8298 
8299 	c->op = access;
8300 	vmcs_write(VMX_INST_ERROR, 0);
8301 	enter_guest();
8302 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
8303 	if (c->reason != VMX_VMCALL) {
8304 		skip_exit_insn();
8305 		enter_guest();
8306 	}
8307 	skip_exit_vmcall();
8308 }
8309 
8310 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
8311 {
8312 	struct vmcs_shadow_test_common *c = &l1_l2_common;
8313 	struct vmcs *shadow;
8314 	u64 value;
8315 	uintptr_t flags[2];
8316 	bool good_shadow;
8317 	u32 vmx_inst_error;
8318 
8319 	report_prefix_pushf("field %lx", field);
8320 	c->field = field;
8321 
8322 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
8323 	if (shadow != (struct vmcs *)-1ull) {
8324 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
8325 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
8326 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
8327 	} else {
8328 		/*
8329 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
8330 		 * shadowed-fields should fail with setting RFLAGS.CF.
8331 		 */
8332 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
8333 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
8334 		good_shadow = false;
8335 	}
8336 
8337 	/* Intercept both VMREAD and VMWRITE. */
8338 	report_prefix_push("no VMREAD/VMWRITE permission");
8339 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
8340 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
8341 		set_bit(field, bitmap[ACCESS_VMREAD]);
8342 		set_bit(field, bitmap[ACCESS_VMWRITE]);
8343 	}
8344 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
8345 	report("not shadowed for VMWRITE", c->reason == VMX_VMWRITE);
8346 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
8347 	report("not shadowed for VMREAD", c->reason == VMX_VMREAD);
8348 	report_prefix_pop();
8349 
8350 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
8351 		goto out;
8352 
8353 	/* Permit shadowed VMREAD. */
8354 	report_prefix_push("VMREAD permission only");
8355 	clear_bit(field, bitmap[ACCESS_VMREAD]);
8356 	set_bit(field, bitmap[ACCESS_VMWRITE]);
8357 	if (good_shadow)
8358 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
8359 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
8360 	report("not shadowed for VMWRITE", c->reason == VMX_VMWRITE);
8361 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
8362 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
8363 	report("shadowed for VMREAD (in %ld cycles)", c->reason == VMX_VMCALL,
8364 	       c->time);
8365 	report("ALU flags after VMREAD (%lx) are as expected (%lx)",
8366 	       c->flags == flags[ACCESS_VMREAD],
8367 	       c->flags, flags[ACCESS_VMREAD]);
8368 	if (good_shadow)
8369 		report("value read from shadow (%lx) is as expected (%lx)",
8370 		       c->value == value, c->value, value);
8371 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
8372 		report("VMX_INST_ERROR (%d) is as expected (%d)",
8373 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
8374 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8375 	report_prefix_pop();
8376 
8377 	/* Permit shadowed VMWRITE. */
8378 	report_prefix_push("VMWRITE permission only");
8379 	set_bit(field, bitmap[ACCESS_VMREAD]);
8380 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
8381 	if (good_shadow)
8382 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
8383 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
8384 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
8385 	report("shadowed for VMWRITE (in %ld cycles)", c->reason == VMX_VMCALL,
8386 		c->time);
8387 	report("ALU flags after VMWRITE (%lx) are as expected (%lx)",
8388 	       c->flags == flags[ACCESS_VMREAD],
8389 	       c->flags, flags[ACCESS_VMREAD]);
8390 	if (good_shadow) {
8391 		value = vmread_from_shadow(field);
8392 		report("shadow VMCS value (%lx) is as expected (%lx)",
8393 		       value == 0, value, 0ul);
8394 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
8395 		report("VMX_INST_ERROR (%d) is as expected (%d)",
8396 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
8397 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8398 	}
8399 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
8400 	report("not shadowed for VMREAD", c->reason == VMX_VMREAD);
8401 	report_prefix_pop();
8402 
8403 	/* Permit shadowed VMREAD and VMWRITE. */
8404 	report_prefix_push("VMREAD and VMWRITE permission");
8405 	clear_bit(field, bitmap[ACCESS_VMREAD]);
8406 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
8407 	if (good_shadow)
8408 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
8409 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
8410 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
8411 	report("shadowed for VMWRITE (in %ld cycles)", c->reason == VMX_VMCALL,
8412 		c->time);
8413 	report("ALU flags after VMWRITE (%lx) are as expected (%lx)",
8414 	       c->flags == flags[ACCESS_VMREAD],
8415 	       c->flags, flags[ACCESS_VMREAD]);
8416 	if (good_shadow) {
8417 		value = vmread_from_shadow(field);
8418 		report("shadow VMCS value (%lx) is as expected (%lx)",
8419 		       value == 0, value, 0ul);
8420 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
8421 		report("VMX_INST_ERROR (%d) is as expected (%d)",
8422 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
8423 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8424 	}
8425 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
8426 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
8427 	report("shadowed for VMREAD (in %ld cycles)", c->reason == VMX_VMCALL,
8428 	       c->time);
8429 	report("ALU flags after VMREAD (%lx) are as expected (%lx)",
8430 	       c->flags == flags[ACCESS_VMREAD],
8431 	       c->flags, flags[ACCESS_VMREAD]);
8432 	if (good_shadow)
8433 		report("value read from shadow (%lx) is as expected (%lx)",
8434 		       c->value == 0, c->value, 0ul);
8435 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
8436 		report("VMX_INST_ERROR (%d) is as expected (%d)",
8437 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
8438 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
8439 	report_prefix_pop();
8440 
8441 out:
8442 	report_prefix_pop();
8443 }
8444 
8445 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
8446 {
8447 	unsigned base;
8448 	unsigned index;
8449 	unsigned bit;
8450 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
8451 
8452 	/* Run test on all possible valid VMCS fields */
8453 	for (base = 0;
8454 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
8455 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
8456 		for (index = 0; index <= highest_index; index++)
8457 			vmcs_shadow_test_field(bitmap, base + index);
8458 
8459 	/*
8460 	 * Run tests on some invalid VMCS fields
8461 	 * (Have reserved bit set).
8462 	 */
8463 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
8464 		vmcs_shadow_test_field(bitmap, (1ull << bit));
8465 }
8466 
8467 static void vmx_vmcs_shadow_test(void)
8468 {
8469 	u8 *bitmap[2];
8470 	struct vmcs *shadow;
8471 
8472 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
8473 		printf("\t'Activate secondary controls' not supported.\n");
8474 		return;
8475 	}
8476 
8477 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
8478 		printf("\t'VMCS shadowing' not supported.\n");
8479 		return;
8480 	}
8481 
8482 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
8483 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
8484 		printf("\tVMWRITE can't modify VM-exit information fields.\n");
8485 		return;
8486 	}
8487 
8488 	test_set_guest(vmx_vmcs_shadow_test_guest);
8489 
8490 	bitmap[ACCESS_VMREAD] = alloc_page();
8491 	bitmap[ACCESS_VMWRITE] = alloc_page();
8492 
8493 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
8494 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
8495 
8496 	shadow = alloc_page();
8497 	shadow->hdr.revision_id = basic.revision;
8498 	shadow->hdr.shadow_vmcs = 1;
8499 	TEST_ASSERT(!vmcs_clear(shadow));
8500 
8501 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
8502 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
8503 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
8504 
8505 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
8506 	report_prefix_push("valid link pointer");
8507 	vmx_vmcs_shadow_test_body(bitmap);
8508 	report_prefix_pop();
8509 
8510 	vmcs_write(VMCS_LINK_PTR, -1ull);
8511 	report_prefix_push("invalid link pointer");
8512 	vmx_vmcs_shadow_test_body(bitmap);
8513 	report_prefix_pop();
8514 
8515 	l1_l2_common.op = ACCESS_NONE;
8516 	enter_guest();
8517 }
8518 
8519 
8520 
8521 static int invalid_msr_init(struct vmcs *vmcs)
8522 {
8523 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
8524 		printf("\tPreemption timer is not supported\n");
8525 		return VMX_TEST_EXIT;
8526 	}
8527 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
8528 	preempt_val = 10000000;
8529 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
8530 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
8531 
8532 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
8533 		printf("\tSave preemption value is not supported\n");
8534 
8535 	vmcs_write(ENT_MSR_LD_CNT, 1);
8536 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
8537 
8538 	return VMX_TEST_START;
8539 }
8540 
8541 
8542 static void invalid_msr_main(void)
8543 {
8544 	report("Invalid MSR load", 0);
8545 }
8546 
8547 static int invalid_msr_exit_handler(void)
8548 {
8549 	report("Invalid MSR load", 0);
8550 	print_vmexit_info();
8551 	return VMX_TEST_EXIT;
8552 }
8553 
8554 static int invalid_msr_entry_failure(struct vmentry_failure *failure)
8555 {
8556 	ulong reason;
8557 
8558 	reason = vmcs_read(EXI_REASON);
8559 	report("Invalid MSR load", reason == (0x80000000u | VMX_FAIL_MSR));
8560 	return VMX_TEST_VMEXIT;
8561 }
8562 
8563 
8564 #define TEST(name) { #name, .v2 = name }
8565 
8566 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
8567 struct vmx_test vmx_tests[] = {
8568 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
8569 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
8570 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
8571 		preemption_timer_exit_handler, NULL, {0} },
8572 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
8573 		test_ctrl_pat_exit_handler, NULL, {0} },
8574 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
8575 		test_ctrl_efer_exit_handler, NULL, {0} },
8576 	{ "CR shadowing", NULL, cr_shadowing_main,
8577 		cr_shadowing_exit_handler, NULL, {0} },
8578 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
8579 		NULL, {0} },
8580 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
8581 		insn_intercept_exit_handler, NULL, {0} },
8582 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
8583 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
8584 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
8585 	{ "VPID", vpid_init, vpid_main, vpid_exit_handler, NULL, {0} },
8586 	{ "interrupt", interrupt_init, interrupt_main,
8587 		interrupt_exit_handler, NULL, {0} },
8588 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
8589 		NULL, {0} },
8590 	{ "MSR switch", msr_switch_init, msr_switch_main,
8591 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
8592 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
8593 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
8594 		disable_rdtscp_exit_handler, NULL, {0} },
8595 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
8596 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
8597 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
8598 		exit_monitor_from_l2_handler, NULL, {0} },
8599 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
8600 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
8601 	/* Basic V2 tests. */
8602 	TEST(v2_null_test),
8603 	TEST(v2_multiple_entries_test),
8604 	TEST(fixture_test_case1),
8605 	TEST(fixture_test_case2),
8606 	/* Opcode tests. */
8607 	TEST(invvpid_test_v2),
8608 	/* VM-entry tests */
8609 	TEST(vmx_controls_test),
8610 	TEST(vmx_host_state_area_test),
8611 	TEST(vmx_guest_state_area_test),
8612 	TEST(vmentry_movss_shadow_test),
8613 	/* APICv tests */
8614 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
8615 	TEST(vmx_hlt_with_rvi_test),
8616 	TEST(apic_reg_virt_test),
8617 	TEST(virt_x2apic_mode_test),
8618 	/* APIC pass-through tests */
8619 	TEST(vmx_apic_passthrough_test),
8620 	TEST(vmx_apic_passthrough_thread_test),
8621 	/* VMCS Shadowing tests */
8622 	TEST(vmx_vmcs_shadow_test),
8623 	/* Regression tests */
8624 	TEST(vmx_cr_load_test),
8625 	TEST(vmx_nm_test),
8626 	TEST(vmx_db_test),
8627 	TEST(vmx_nmi_window_test),
8628 	TEST(vmx_intr_window_test),
8629 	TEST(vmx_pending_event_test),
8630 	TEST(vmx_pending_event_hlt_test),
8631 	TEST(vmx_store_tsc_test),
8632 	/* EPT access tests. */
8633 	TEST(ept_access_test_not_present),
8634 	TEST(ept_access_test_read_only),
8635 	TEST(ept_access_test_write_only),
8636 	TEST(ept_access_test_read_write),
8637 	TEST(ept_access_test_execute_only),
8638 	TEST(ept_access_test_read_execute),
8639 	TEST(ept_access_test_write_execute),
8640 	TEST(ept_access_test_read_write_execute),
8641 	TEST(ept_access_test_reserved_bits),
8642 	TEST(ept_access_test_ignored_bits),
8643 	TEST(ept_access_test_paddr_not_present_ad_disabled),
8644 	TEST(ept_access_test_paddr_not_present_ad_enabled),
8645 	TEST(ept_access_test_paddr_read_only_ad_disabled),
8646 	TEST(ept_access_test_paddr_read_only_ad_enabled),
8647 	TEST(ept_access_test_paddr_read_write),
8648 	TEST(ept_access_test_paddr_read_write_execute),
8649 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
8650 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
8651 	TEST(ept_access_test_paddr_not_present_page_fault),
8652 	TEST(ept_access_test_force_2m_page),
8653 	{ NULL, NULL, NULL, NULL, NULL, {0} },
8654 };
8655