xref: /kvm-unit-tests/x86/vmx_tests.c (revision a668a77af88c23b7e6fc12b985bef560a2c3548c)
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 	memset(msr_bitmap, 0x0, PAGE_SIZE);
257 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
258 	ctrl_cpu0 |= CPU_MSR_BITMAP;
259 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
260 	vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
261 }
262 
263 static void *get_msr_bitmap(void)
264 {
265 	void *msr_bitmap;
266 
267 	if (vmcs_read(CPU_EXEC_CTRL0) & CPU_MSR_BITMAP) {
268 		msr_bitmap = (void *)vmcs_read(MSR_BITMAP);
269 	} else {
270 		msr_bitmap = alloc_page();
271 		memset(msr_bitmap, 0xff, PAGE_SIZE);
272 		vmcs_write(MSR_BITMAP, (u64)msr_bitmap);
273 		vmcs_set_bits(CPU_EXEC_CTRL0, CPU_MSR_BITMAP);
274 	}
275 
276 	return msr_bitmap;
277 }
278 
279 static void disable_intercept_for_x2apic_msrs(void)
280 {
281 	unsigned long *msr_bitmap = (unsigned long *)get_msr_bitmap();
282 	u32 msr;
283 
284 	for (msr = APIC_BASE_MSR;
285 		 msr < (APIC_BASE_MSR+0xff);
286 		 msr += BITS_PER_LONG) {
287 		unsigned int word = msr / BITS_PER_LONG;
288 
289 		msr_bitmap[word] = 0;
290 		msr_bitmap[word + (0x800 / sizeof(long))] = 0;
291 	}
292 }
293 
294 static int test_ctrl_pat_init(struct vmcs *vmcs)
295 {
296 	u64 ctrl_ent;
297 	u64 ctrl_exi;
298 
299 	msr_bmp_init();
300 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT) &&
301 	    !(ctrl_exit_rev.clr & EXI_LOAD_PAT) &&
302 	    !(ctrl_enter_rev.clr & ENT_LOAD_PAT)) {
303 		printf("\tSave/load PAT is not supported\n");
304 		return 1;
305 	}
306 
307 	ctrl_ent = vmcs_read(ENT_CONTROLS);
308 	ctrl_exi = vmcs_read(EXI_CONTROLS);
309 	ctrl_ent |= ctrl_enter_rev.clr & ENT_LOAD_PAT;
310 	ctrl_exi |= ctrl_exit_rev.clr & (EXI_SAVE_PAT | EXI_LOAD_PAT);
311 	vmcs_write(ENT_CONTROLS, ctrl_ent);
312 	vmcs_write(EXI_CONTROLS, ctrl_exi);
313 	ia32_pat = rdmsr(MSR_IA32_CR_PAT);
314 	vmcs_write(GUEST_PAT, 0x0);
315 	vmcs_write(HOST_PAT, ia32_pat);
316 	return VMX_TEST_START;
317 }
318 
319 static void test_ctrl_pat_main(void)
320 {
321 	u64 guest_ia32_pat;
322 
323 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
324 	if (!(ctrl_enter_rev.clr & ENT_LOAD_PAT))
325 		printf("\tENT_LOAD_PAT is not supported.\n");
326 	else {
327 		if (guest_ia32_pat != 0) {
328 			report("Entry load PAT", 0);
329 			return;
330 		}
331 	}
332 	wrmsr(MSR_IA32_CR_PAT, 0x6);
333 	vmcall();
334 	guest_ia32_pat = rdmsr(MSR_IA32_CR_PAT);
335 	if (ctrl_enter_rev.clr & ENT_LOAD_PAT)
336 		report("Entry load PAT", guest_ia32_pat == ia32_pat);
337 }
338 
339 static int test_ctrl_pat_exit_handler(void)
340 {
341 	u64 guest_rip;
342 	ulong reason;
343 	u64 guest_pat;
344 
345 	guest_rip = vmcs_read(GUEST_RIP);
346 	reason = vmcs_read(EXI_REASON) & 0xff;
347 	switch (reason) {
348 	case VMX_VMCALL:
349 		guest_pat = vmcs_read(GUEST_PAT);
350 		if (!(ctrl_exit_rev.clr & EXI_SAVE_PAT)) {
351 			printf("\tEXI_SAVE_PAT is not supported\n");
352 			vmcs_write(GUEST_PAT, 0x6);
353 		} else {
354 			report("Exit save PAT", guest_pat == 0x6);
355 		}
356 		if (!(ctrl_exit_rev.clr & EXI_LOAD_PAT))
357 			printf("\tEXI_LOAD_PAT is not supported\n");
358 		else
359 			report("Exit load PAT", rdmsr(MSR_IA32_CR_PAT) == ia32_pat);
360 		vmcs_write(GUEST_PAT, ia32_pat);
361 		vmcs_write(GUEST_RIP, guest_rip + 3);
362 		return VMX_TEST_RESUME;
363 	default:
364 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
365 		break;
366 	}
367 	return VMX_TEST_VMEXIT;
368 }
369 
370 static int test_ctrl_efer_init(struct vmcs *vmcs)
371 {
372 	u64 ctrl_ent;
373 	u64 ctrl_exi;
374 
375 	msr_bmp_init();
376 	ctrl_ent = vmcs_read(ENT_CONTROLS) | ENT_LOAD_EFER;
377 	ctrl_exi = vmcs_read(EXI_CONTROLS) | EXI_SAVE_EFER | EXI_LOAD_EFER;
378 	vmcs_write(ENT_CONTROLS, ctrl_ent & ctrl_enter_rev.clr);
379 	vmcs_write(EXI_CONTROLS, ctrl_exi & ctrl_exit_rev.clr);
380 	ia32_efer = rdmsr(MSR_EFER);
381 	vmcs_write(GUEST_EFER, ia32_efer ^ EFER_NX);
382 	vmcs_write(HOST_EFER, ia32_efer ^ EFER_NX);
383 	return VMX_TEST_START;
384 }
385 
386 static void test_ctrl_efer_main(void)
387 {
388 	u64 guest_ia32_efer;
389 
390 	guest_ia32_efer = rdmsr(MSR_EFER);
391 	if (!(ctrl_enter_rev.clr & ENT_LOAD_EFER))
392 		printf("\tENT_LOAD_EFER is not supported.\n");
393 	else {
394 		if (guest_ia32_efer != (ia32_efer ^ EFER_NX)) {
395 			report("Entry load EFER", 0);
396 			return;
397 		}
398 	}
399 	wrmsr(MSR_EFER, ia32_efer);
400 	vmcall();
401 	guest_ia32_efer = rdmsr(MSR_EFER);
402 	if (ctrl_enter_rev.clr & ENT_LOAD_EFER)
403 		report("Entry load EFER", guest_ia32_efer == ia32_efer);
404 }
405 
406 static int test_ctrl_efer_exit_handler(void)
407 {
408 	u64 guest_rip;
409 	ulong reason;
410 	u64 guest_efer;
411 
412 	guest_rip = vmcs_read(GUEST_RIP);
413 	reason = vmcs_read(EXI_REASON) & 0xff;
414 	switch (reason) {
415 	case VMX_VMCALL:
416 		guest_efer = vmcs_read(GUEST_EFER);
417 		if (!(ctrl_exit_rev.clr & EXI_SAVE_EFER)) {
418 			printf("\tEXI_SAVE_EFER is not supported\n");
419 			vmcs_write(GUEST_EFER, ia32_efer);
420 		} else {
421 			report("Exit save EFER", guest_efer == ia32_efer);
422 		}
423 		if (!(ctrl_exit_rev.clr & EXI_LOAD_EFER)) {
424 			printf("\tEXI_LOAD_EFER is not supported\n");
425 			wrmsr(MSR_EFER, ia32_efer ^ EFER_NX);
426 		} else {
427 			report("Exit load EFER", rdmsr(MSR_EFER) == (ia32_efer ^ EFER_NX));
428 		}
429 		vmcs_write(GUEST_PAT, ia32_efer);
430 		vmcs_write(GUEST_RIP, guest_rip + 3);
431 		return VMX_TEST_RESUME;
432 	default:
433 		printf("ERROR : Undefined exit reason, reason = %ld.\n", reason);
434 		break;
435 	}
436 	return VMX_TEST_VMEXIT;
437 }
438 
439 u32 guest_cr0, guest_cr4;
440 
441 static void cr_shadowing_main(void)
442 {
443 	u32 cr0, cr4, tmp;
444 
445 	// Test read through
446 	vmx_set_test_stage(0);
447 	guest_cr0 = read_cr0();
448 	if (vmx_get_test_stage() == 1)
449 		report("Read through CR0", 0);
450 	else
451 		vmcall();
452 	vmx_set_test_stage(1);
453 	guest_cr4 = read_cr4();
454 	if (vmx_get_test_stage() == 2)
455 		report("Read through CR4", 0);
456 	else
457 		vmcall();
458 	// Test write through
459 	guest_cr0 = guest_cr0 ^ (X86_CR0_TS | X86_CR0_MP);
460 	guest_cr4 = guest_cr4 ^ (X86_CR4_TSD | X86_CR4_DE);
461 	vmx_set_test_stage(2);
462 	write_cr0(guest_cr0);
463 	if (vmx_get_test_stage() == 3)
464 		report("Write throuth CR0", 0);
465 	else
466 		vmcall();
467 	vmx_set_test_stage(3);
468 	write_cr4(guest_cr4);
469 	if (vmx_get_test_stage() == 4)
470 		report("Write through CR4", 0);
471 	else
472 		vmcall();
473 	// Test read shadow
474 	vmx_set_test_stage(4);
475 	vmcall();
476 	cr0 = read_cr0();
477 	if (vmx_get_test_stage() != 5)
478 		report("Read shadowing CR0", cr0 == guest_cr0);
479 	vmx_set_test_stage(5);
480 	cr4 = read_cr4();
481 	if (vmx_get_test_stage() != 6)
482 		report("Read shadowing CR4", cr4 == guest_cr4);
483 	// Test write shadow (same value with shadow)
484 	vmx_set_test_stage(6);
485 	write_cr0(guest_cr0);
486 	if (vmx_get_test_stage() == 7)
487 		report("Write shadowing CR0 (same value with shadow)", 0);
488 	else
489 		vmcall();
490 	vmx_set_test_stage(7);
491 	write_cr4(guest_cr4);
492 	if (vmx_get_test_stage() == 8)
493 		report("Write shadowing CR4 (same value with shadow)", 0);
494 	else
495 		vmcall();
496 	// Test write shadow (different value)
497 	vmx_set_test_stage(8);
498 	tmp = guest_cr0 ^ X86_CR0_TS;
499 	asm volatile("mov %0, %%rsi\n\t"
500 		"mov %%rsi, %%cr0\n\t"
501 		::"m"(tmp)
502 		:"rsi", "memory", "cc");
503 	report("Write shadowing different X86_CR0_TS", vmx_get_test_stage() == 9);
504 	vmx_set_test_stage(9);
505 	tmp = guest_cr0 ^ X86_CR0_MP;
506 	asm volatile("mov %0, %%rsi\n\t"
507 		"mov %%rsi, %%cr0\n\t"
508 		::"m"(tmp)
509 		:"rsi", "memory", "cc");
510 	report("Write shadowing different X86_CR0_MP", vmx_get_test_stage() == 10);
511 	vmx_set_test_stage(10);
512 	tmp = guest_cr4 ^ X86_CR4_TSD;
513 	asm volatile("mov %0, %%rsi\n\t"
514 		"mov %%rsi, %%cr4\n\t"
515 		::"m"(tmp)
516 		:"rsi", "memory", "cc");
517 	report("Write shadowing different X86_CR4_TSD", vmx_get_test_stage() == 11);
518 	vmx_set_test_stage(11);
519 	tmp = guest_cr4 ^ X86_CR4_DE;
520 	asm volatile("mov %0, %%rsi\n\t"
521 		"mov %%rsi, %%cr4\n\t"
522 		::"m"(tmp)
523 		:"rsi", "memory", "cc");
524 	report("Write shadowing different X86_CR4_DE", vmx_get_test_stage() == 12);
525 }
526 
527 static int cr_shadowing_exit_handler(void)
528 {
529 	u64 guest_rip;
530 	ulong reason;
531 	u32 insn_len;
532 	u32 exit_qual;
533 
534 	guest_rip = vmcs_read(GUEST_RIP);
535 	reason = vmcs_read(EXI_REASON) & 0xff;
536 	insn_len = vmcs_read(EXI_INST_LEN);
537 	exit_qual = vmcs_read(EXI_QUALIFICATION);
538 	switch (reason) {
539 	case VMX_VMCALL:
540 		switch (vmx_get_test_stage()) {
541 		case 0:
542 			report("Read through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
543 			break;
544 		case 1:
545 			report("Read through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
546 			break;
547 		case 2:
548 			report("Write through CR0", guest_cr0 == vmcs_read(GUEST_CR0));
549 			break;
550 		case 3:
551 			report("Write through CR4", guest_cr4 == vmcs_read(GUEST_CR4));
552 			break;
553 		case 4:
554 			guest_cr0 = vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP);
555 			guest_cr4 = vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE);
556 			vmcs_write(CR0_MASK, X86_CR0_TS | X86_CR0_MP);
557 			vmcs_write(CR0_READ_SHADOW, guest_cr0 & (X86_CR0_TS | X86_CR0_MP));
558 			vmcs_write(CR4_MASK, X86_CR4_TSD | X86_CR4_DE);
559 			vmcs_write(CR4_READ_SHADOW, guest_cr4 & (X86_CR4_TSD | X86_CR4_DE));
560 			break;
561 		case 6:
562 			report("Write shadowing CR0 (same value)",
563 					guest_cr0 == (vmcs_read(GUEST_CR0) ^ (X86_CR0_TS | X86_CR0_MP)));
564 			break;
565 		case 7:
566 			report("Write shadowing CR4 (same value)",
567 					guest_cr4 == (vmcs_read(GUEST_CR4) ^ (X86_CR4_TSD | X86_CR4_DE)));
568 			break;
569 		default:
570 			// Should not reach here
571 			report("unexpected stage, %d", false,
572 			       vmx_get_test_stage());
573 			print_vmexit_info();
574 			return VMX_TEST_VMEXIT;
575 		}
576 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
577 		return VMX_TEST_RESUME;
578 	case VMX_CR:
579 		switch (vmx_get_test_stage()) {
580 		case 4:
581 			report("Read shadowing CR0", 0);
582 			vmx_inc_test_stage();
583 			break;
584 		case 5:
585 			report("Read shadowing CR4", 0);
586 			vmx_inc_test_stage();
587 			break;
588 		case 6:
589 			report("Write shadowing CR0 (same value)", 0);
590 			vmx_inc_test_stage();
591 			break;
592 		case 7:
593 			report("Write shadowing CR4 (same value)", 0);
594 			vmx_inc_test_stage();
595 			break;
596 		case 8:
597 		case 9:
598 			// 0x600 encodes "mov %esi, %cr0"
599 			if (exit_qual == 0x600)
600 				vmx_inc_test_stage();
601 			break;
602 		case 10:
603 		case 11:
604 			// 0x604 encodes "mov %esi, %cr4"
605 			if (exit_qual == 0x604)
606 				vmx_inc_test_stage();
607 			break;
608 		default:
609 			// Should not reach here
610 			report("unexpected stage, %d", false,
611 			       vmx_get_test_stage());
612 			print_vmexit_info();
613 			return VMX_TEST_VMEXIT;
614 		}
615 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
616 		return VMX_TEST_RESUME;
617 	default:
618 		report("Unknown exit reason, %ld", false, reason);
619 		print_vmexit_info();
620 	}
621 	return VMX_TEST_VMEXIT;
622 }
623 
624 static int iobmp_init(struct vmcs *vmcs)
625 {
626 	u32 ctrl_cpu0;
627 
628 	io_bitmap_a = alloc_page();
629 	io_bitmap_b = alloc_page();
630 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
631 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
632 	ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
633 	ctrl_cpu0 |= CPU_IO_BITMAP;
634 	ctrl_cpu0 &= (~CPU_IO);
635 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
636 	vmcs_write(IO_BITMAP_A, (u64)io_bitmap_a);
637 	vmcs_write(IO_BITMAP_B, (u64)io_bitmap_b);
638 	return VMX_TEST_START;
639 }
640 
641 static void iobmp_main(void)
642 {
643 	// stage 0, test IO pass
644 	vmx_set_test_stage(0);
645 	inb(0x5000);
646 	outb(0x0, 0x5000);
647 	report("I/O bitmap - I/O pass", vmx_get_test_stage() == 0);
648 	// test IO width, in/out
649 	((u8 *)io_bitmap_a)[0] = 0xFF;
650 	vmx_set_test_stage(2);
651 	inb(0x0);
652 	report("I/O bitmap - trap in", vmx_get_test_stage() == 3);
653 	vmx_set_test_stage(3);
654 	outw(0x0, 0x0);
655 	report("I/O bitmap - trap out", vmx_get_test_stage() == 4);
656 	vmx_set_test_stage(4);
657 	inl(0x0);
658 	report("I/O bitmap - I/O width, long", vmx_get_test_stage() == 5);
659 	// test low/high IO port
660 	vmx_set_test_stage(5);
661 	((u8 *)io_bitmap_a)[0x5000 / 8] = (1 << (0x5000 % 8));
662 	inb(0x5000);
663 	report("I/O bitmap - I/O port, low part", vmx_get_test_stage() == 6);
664 	vmx_set_test_stage(6);
665 	((u8 *)io_bitmap_b)[0x1000 / 8] = (1 << (0x1000 % 8));
666 	inb(0x9000);
667 	report("I/O bitmap - I/O port, high part", vmx_get_test_stage() == 7);
668 	// test partial pass
669 	vmx_set_test_stage(7);
670 	inl(0x4FFF);
671 	report("I/O bitmap - partial pass", vmx_get_test_stage() == 8);
672 	// test overrun
673 	vmx_set_test_stage(8);
674 	memset(io_bitmap_a, 0x0, PAGE_SIZE);
675 	memset(io_bitmap_b, 0x0, PAGE_SIZE);
676 	inl(0xFFFF);
677 	report("I/O bitmap - overrun", vmx_get_test_stage() == 9);
678 	vmx_set_test_stage(9);
679 	vmcall();
680 	outb(0x0, 0x0);
681 	report("I/O bitmap - ignore unconditional exiting",
682 	       vmx_get_test_stage() == 9);
683 	vmx_set_test_stage(10);
684 	vmcall();
685 	outb(0x0, 0x0);
686 	report("I/O bitmap - unconditional exiting",
687 	       vmx_get_test_stage() == 11);
688 }
689 
690 static int iobmp_exit_handler(void)
691 {
692 	u64 guest_rip;
693 	ulong reason, exit_qual;
694 	u32 insn_len, ctrl_cpu0;
695 
696 	guest_rip = vmcs_read(GUEST_RIP);
697 	reason = vmcs_read(EXI_REASON) & 0xff;
698 	exit_qual = vmcs_read(EXI_QUALIFICATION);
699 	insn_len = vmcs_read(EXI_INST_LEN);
700 	switch (reason) {
701 	case VMX_IO:
702 		switch (vmx_get_test_stage()) {
703 		case 0:
704 		case 1:
705 			vmx_inc_test_stage();
706 			break;
707 		case 2:
708 			report("I/O bitmap - I/O width, byte",
709 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_BYTE);
710 			report("I/O bitmap - I/O direction, in", exit_qual & VMX_IO_IN);
711 			vmx_inc_test_stage();
712 			break;
713 		case 3:
714 			report("I/O bitmap - I/O width, word",
715 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_WORD);
716 			report("I/O bitmap - I/O direction, out",
717 					!(exit_qual & VMX_IO_IN));
718 			vmx_inc_test_stage();
719 			break;
720 		case 4:
721 			report("I/O bitmap - I/O width, long",
722 					(exit_qual & VMX_IO_SIZE_MASK) == _VMX_IO_LONG);
723 			vmx_inc_test_stage();
724 			break;
725 		case 5:
726 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x5000)
727 				vmx_inc_test_stage();
728 			break;
729 		case 6:
730 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x9000)
731 				vmx_inc_test_stage();
732 			break;
733 		case 7:
734 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0x4FFF)
735 				vmx_inc_test_stage();
736 			break;
737 		case 8:
738 			if (((exit_qual & VMX_IO_PORT_MASK) >> VMX_IO_PORT_SHIFT) == 0xFFFF)
739 				vmx_inc_test_stage();
740 			break;
741 		case 9:
742 		case 10:
743 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
744 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0 & ~CPU_IO);
745 			vmx_inc_test_stage();
746 			break;
747 		default:
748 			// Should not reach here
749 			report("unexpected stage, %d", false,
750 			       vmx_get_test_stage());
751 			print_vmexit_info();
752 			return VMX_TEST_VMEXIT;
753 		}
754 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
755 		return VMX_TEST_RESUME;
756 	case VMX_VMCALL:
757 		switch (vmx_get_test_stage()) {
758 		case 9:
759 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
760 			ctrl_cpu0 |= CPU_IO | CPU_IO_BITMAP;
761 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
762 			break;
763 		case 10:
764 			ctrl_cpu0 = vmcs_read(CPU_EXEC_CTRL0);
765 			ctrl_cpu0 = (ctrl_cpu0 & ~CPU_IO_BITMAP) | CPU_IO;
766 			vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu0);
767 			break;
768 		default:
769 			// Should not reach here
770 			report("unexpected stage, %d", false,
771 			       vmx_get_test_stage());
772 			print_vmexit_info();
773 			return VMX_TEST_VMEXIT;
774 		}
775 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
776 		return VMX_TEST_RESUME;
777 	default:
778 		printf("guest_rip = %#lx\n", guest_rip);
779 		printf("\tERROR : Undefined exit reason, reason = %ld.\n", reason);
780 		break;
781 	}
782 	return VMX_TEST_VMEXIT;
783 }
784 
785 #define INSN_CPU0		0
786 #define INSN_CPU1		1
787 #define INSN_ALWAYS_TRAP	2
788 
789 #define FIELD_EXIT_QUAL		(1 << 0)
790 #define FIELD_INSN_INFO		(1 << 1)
791 
792 asm(
793 	"insn_hlt: hlt;ret\n\t"
794 	"insn_invlpg: invlpg 0x12345678;ret\n\t"
795 	"insn_mwait: xor %eax, %eax; xor %ecx, %ecx; mwait;ret\n\t"
796 	"insn_rdpmc: xor %ecx, %ecx; rdpmc;ret\n\t"
797 	"insn_rdtsc: rdtsc;ret\n\t"
798 	"insn_cr3_load: mov cr3,%rax; mov %rax,%cr3;ret\n\t"
799 	"insn_cr3_store: mov %cr3,%rax;ret\n\t"
800 #ifdef __x86_64__
801 	"insn_cr8_load: xor %eax, %eax; mov %rax,%cr8;ret\n\t"
802 	"insn_cr8_store: mov %cr8,%rax;ret\n\t"
803 #endif
804 	"insn_monitor: xor %eax, %eax; xor %ecx, %ecx; xor %edx, %edx; monitor;ret\n\t"
805 	"insn_pause: pause;ret\n\t"
806 	"insn_wbinvd: wbinvd;ret\n\t"
807 	"insn_cpuid: mov $10, %eax; cpuid;ret\n\t"
808 	"insn_invd: invd;ret\n\t"
809 	"insn_sgdt: sgdt gdt64_desc;ret\n\t"
810 	"insn_lgdt: lgdt gdt64_desc;ret\n\t"
811 	"insn_sidt: sidt idt_descr;ret\n\t"
812 	"insn_lidt: lidt idt_descr;ret\n\t"
813 	"insn_sldt: sldt %ax;ret\n\t"
814 	"insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t"
815 	"insn_str: str %ax;ret\n\t"
816 	"insn_rdrand: rdrand %rax;ret\n\t"
817 	"insn_rdseed: rdseed %rax;ret\n\t"
818 );
819 extern void insn_hlt(void);
820 extern void insn_invlpg(void);
821 extern void insn_mwait(void);
822 extern void insn_rdpmc(void);
823 extern void insn_rdtsc(void);
824 extern void insn_cr3_load(void);
825 extern void insn_cr3_store(void);
826 #ifdef __x86_64__
827 extern void insn_cr8_load(void);
828 extern void insn_cr8_store(void);
829 #endif
830 extern void insn_monitor(void);
831 extern void insn_pause(void);
832 extern void insn_wbinvd(void);
833 extern void insn_sgdt(void);
834 extern void insn_lgdt(void);
835 extern void insn_sidt(void);
836 extern void insn_lidt(void);
837 extern void insn_sldt(void);
838 extern void insn_lldt(void);
839 extern void insn_str(void);
840 extern void insn_cpuid(void);
841 extern void insn_invd(void);
842 extern void insn_rdrand(void);
843 extern void insn_rdseed(void);
844 
845 u32 cur_insn;
846 u64 cr3;
847 
848 struct insn_table {
849 	const char *name;
850 	u32 flag;
851 	void (*insn_func)(void);
852 	u32 type;
853 	u32 reason;
854 	ulong exit_qual;
855 	u32 insn_info;
856 	// Use FIELD_EXIT_QUAL and FIELD_INSN_INFO to define
857 	// which field need to be tested, reason is always tested
858 	u32 test_field;
859 };
860 
861 /*
862  * Add more test cases of instruction intercept here. Elements in this
863  * table is:
864  *	name/control flag/insn function/type/exit reason/exit qulification/
865  *	instruction info/field to test
866  * The last field defines which fields (exit_qual and insn_info) need to be
867  * tested in exit handler. If set to 0, only "reason" is checked.
868  */
869 static struct insn_table insn_table[] = {
870 	// Flags for Primary Processor-Based VM-Execution Controls
871 	{"HLT",  CPU_HLT, insn_hlt, INSN_CPU0, 12, 0, 0, 0},
872 	{"INVLPG", CPU_INVLPG, insn_invlpg, INSN_CPU0, 14,
873 		0x12345678, 0, FIELD_EXIT_QUAL},
874 	{"MWAIT", CPU_MWAIT, insn_mwait, INSN_CPU0, 36, 0, 0, 0},
875 	{"RDPMC", CPU_RDPMC, insn_rdpmc, INSN_CPU0, 15, 0, 0, 0},
876 	{"RDTSC", CPU_RDTSC, insn_rdtsc, INSN_CPU0, 16, 0, 0, 0},
877 	{"CR3 load", CPU_CR3_LOAD, insn_cr3_load, INSN_CPU0, 28, 0x3, 0,
878 		FIELD_EXIT_QUAL},
879 	{"CR3 store", CPU_CR3_STORE, insn_cr3_store, INSN_CPU0, 28, 0x13, 0,
880 		FIELD_EXIT_QUAL},
881 #ifdef __x86_64__
882 	{"CR8 load", CPU_CR8_LOAD, insn_cr8_load, INSN_CPU0, 28, 0x8, 0,
883 		FIELD_EXIT_QUAL},
884 	{"CR8 store", CPU_CR8_STORE, insn_cr8_store, INSN_CPU0, 28, 0x18, 0,
885 		FIELD_EXIT_QUAL},
886 #endif
887 	{"MONITOR", CPU_MONITOR, insn_monitor, INSN_CPU0, 39, 0, 0, 0},
888 	{"PAUSE", CPU_PAUSE, insn_pause, INSN_CPU0, 40, 0, 0, 0},
889 	// Flags for Secondary Processor-Based VM-Execution Controls
890 	{"WBINVD", CPU_WBINVD, insn_wbinvd, INSN_CPU1, 54, 0, 0, 0},
891 	{"DESC_TABLE (SGDT)", CPU_DESC_TABLE, insn_sgdt, INSN_CPU1, 46, 0, 0, 0},
892 	{"DESC_TABLE (LGDT)", CPU_DESC_TABLE, insn_lgdt, INSN_CPU1, 46, 0, 0, 0},
893 	{"DESC_TABLE (SIDT)", CPU_DESC_TABLE, insn_sidt, INSN_CPU1, 46, 0, 0, 0},
894 	{"DESC_TABLE (LIDT)", CPU_DESC_TABLE, insn_lidt, INSN_CPU1, 46, 0, 0, 0},
895 	{"DESC_TABLE (SLDT)", CPU_DESC_TABLE, insn_sldt, INSN_CPU1, 47, 0, 0, 0},
896 	{"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0},
897 	{"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0},
898 	/* LTR causes a #GP if done with a busy selector, so it is not tested.  */
899 	{"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0},
900 	{"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0},
901 	// Instructions always trap
902 	{"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0},
903 	{"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0},
904 	// Instructions never trap
905 	{NULL},
906 };
907 
908 static int insn_intercept_init(struct vmcs *vmcs)
909 {
910 	u32 ctrl_cpu;
911 
912 	ctrl_cpu = ctrl_cpu_rev[0].set | CPU_SECONDARY;
913 	ctrl_cpu &= ctrl_cpu_rev[0].clr;
914 	vmcs_write(CPU_EXEC_CTRL0, ctrl_cpu);
915 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu_rev[1].set);
916 	cr3 = read_cr3();
917 	return VMX_TEST_START;
918 }
919 
920 static void insn_intercept_main(void)
921 {
922 	for (cur_insn = 0; insn_table[cur_insn].name != NULL; cur_insn++) {
923 		vmx_set_test_stage(cur_insn * 2);
924 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
925 		     !(ctrl_cpu_rev[0].clr & insn_table[cur_insn].flag)) ||
926 		    (insn_table[cur_insn].type == INSN_CPU1 &&
927 		     !(ctrl_cpu_rev[1].clr & insn_table[cur_insn].flag))) {
928 			printf("\tCPU_CTRL%d.CPU_%s is not supported.\n",
929 			       insn_table[cur_insn].type - INSN_CPU0,
930 			       insn_table[cur_insn].name);
931 			continue;
932 		}
933 
934 		if ((insn_table[cur_insn].type == INSN_CPU0 &&
935 		     !(ctrl_cpu_rev[0].set & insn_table[cur_insn].flag)) ||
936 		    (insn_table[cur_insn].type == INSN_CPU1 &&
937 		     !(ctrl_cpu_rev[1].set & insn_table[cur_insn].flag))) {
938 			/* skip hlt, it stalls the guest and is tested below */
939 			if (insn_table[cur_insn].insn_func != insn_hlt)
940 				insn_table[cur_insn].insn_func();
941 			report("execute %s", vmx_get_test_stage() == cur_insn * 2,
942 					insn_table[cur_insn].name);
943 		} else if (insn_table[cur_insn].type != INSN_ALWAYS_TRAP)
944 			printf("\tCPU_CTRL%d.CPU_%s always traps.\n",
945 			       insn_table[cur_insn].type - INSN_CPU0,
946 			       insn_table[cur_insn].name);
947 
948 		vmcall();
949 
950 		insn_table[cur_insn].insn_func();
951 		report("intercept %s", vmx_get_test_stage() == cur_insn * 2 + 1,
952 				insn_table[cur_insn].name);
953 
954 		vmx_set_test_stage(cur_insn * 2 + 1);
955 		vmcall();
956 	}
957 }
958 
959 static int insn_intercept_exit_handler(void)
960 {
961 	u64 guest_rip;
962 	u32 reason;
963 	ulong exit_qual;
964 	u32 insn_len;
965 	u32 insn_info;
966 	bool pass;
967 
968 	guest_rip = vmcs_read(GUEST_RIP);
969 	reason = vmcs_read(EXI_REASON) & 0xff;
970 	exit_qual = vmcs_read(EXI_QUALIFICATION);
971 	insn_len = vmcs_read(EXI_INST_LEN);
972 	insn_info = vmcs_read(EXI_INST_INFO);
973 
974 	if (reason == VMX_VMCALL) {
975 		u32 val = 0;
976 
977 		if (insn_table[cur_insn].type == INSN_CPU0)
978 			val = vmcs_read(CPU_EXEC_CTRL0);
979 		else if (insn_table[cur_insn].type == INSN_CPU1)
980 			val = vmcs_read(CPU_EXEC_CTRL1);
981 
982 		if (vmx_get_test_stage() & 1)
983 			val &= ~insn_table[cur_insn].flag;
984 		else
985 			val |= insn_table[cur_insn].flag;
986 
987 		if (insn_table[cur_insn].type == INSN_CPU0)
988 			vmcs_write(CPU_EXEC_CTRL0, val | ctrl_cpu_rev[0].set);
989 		else if (insn_table[cur_insn].type == INSN_CPU1)
990 			vmcs_write(CPU_EXEC_CTRL1, val | ctrl_cpu_rev[1].set);
991 	} else {
992 		pass = (cur_insn * 2 == vmx_get_test_stage()) &&
993 			insn_table[cur_insn].reason == reason;
994 		if (insn_table[cur_insn].test_field & FIELD_EXIT_QUAL &&
995 		    insn_table[cur_insn].exit_qual != exit_qual)
996 			pass = false;
997 		if (insn_table[cur_insn].test_field & FIELD_INSN_INFO &&
998 		    insn_table[cur_insn].insn_info != insn_info)
999 			pass = false;
1000 		if (pass)
1001 			vmx_inc_test_stage();
1002 	}
1003 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
1004 	return VMX_TEST_RESUME;
1005 }
1006 
1007 static int setup_eptp(u64 hpa, bool enable_ad)
1008 {
1009 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1010 	    !(ctrl_cpu_rev[1].clr & CPU_EPT)) {
1011 		printf("\tEPT is not supported");
1012 		return 1;
1013 	}
1014 
1015 	if (!(ept_vpid.val & EPT_CAP_UC) &&
1016 			!(ept_vpid.val & EPT_CAP_WB)) {
1017 		printf("\tEPT paging-structure memory type "
1018 				"UC&WB are not supported\n");
1019 		return 1;
1020 	}
1021 	if (!(ept_vpid.val & EPT_CAP_PWL4)) {
1022 		printf("\tPWL4 is not supported\n");
1023 		return 1;
1024 	}
1025 
1026 	if (ept_vpid.val & EPT_CAP_UC)
1027 		eptp = EPT_MEM_TYPE_UC;
1028 	else
1029 		eptp = EPT_MEM_TYPE_WB;
1030 	eptp |= (3 << EPTP_PG_WALK_LEN_SHIFT);
1031 	eptp |= hpa;
1032 	if (enable_ad)
1033 		eptp |= EPTP_AD_FLAG;
1034 
1035 	vmcs_write(EPTP, eptp);
1036 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0)| CPU_SECONDARY);
1037 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1)| CPU_EPT);
1038 
1039 	return 0;
1040 }
1041 
1042 /* Enables EPT and sets up the identity map. */
1043 static int setup_ept(bool enable_ad)
1044 {
1045 	unsigned long end_of_memory;
1046 
1047 	pml4 = alloc_page();
1048 
1049 	if (setup_eptp(virt_to_phys(pml4), enable_ad))
1050 		return 1;
1051 
1052 	memset(pml4, 0, PAGE_SIZE);
1053 
1054 	end_of_memory = fwcfg_get_u64(FW_CFG_RAM_SIZE);
1055 	if (end_of_memory < (1ul << 32))
1056 		end_of_memory = (1ul << 32);
1057 	/* Cannot use large EPT pages if we need to track EPT
1058 	 * accessed/dirty bits at 4K granularity.
1059 	 */
1060 	setup_ept_range(pml4, 0, end_of_memory, 0,
1061 			!enable_ad && ept_2m_supported(),
1062 			EPT_WA | EPT_RA | EPT_EA);
1063 	return 0;
1064 }
1065 
1066 static void enable_ept(void)
1067 {
1068 	if (setup_eptp(0, false))
1069 		report_abort("EPT setup unexpectedly failed");
1070 }
1071 
1072 static int enable_unrestricted_guest(void)
1073 {
1074 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1075 	    !(ctrl_cpu_rev[1].clr & CPU_URG) ||
1076 	    !(ctrl_cpu_rev[1].clr & CPU_EPT))
1077 		return 1;
1078 
1079 	enable_ept();
1080 
1081 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
1082 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | CPU_URG);
1083 
1084 	return 0;
1085 }
1086 
1087 static void ept_enable_ad_bits(void)
1088 {
1089 	eptp |= EPTP_AD_FLAG;
1090 	vmcs_write(EPTP, eptp);
1091 }
1092 
1093 static void ept_disable_ad_bits(void)
1094 {
1095 	eptp &= ~EPTP_AD_FLAG;
1096 	vmcs_write(EPTP, eptp);
1097 }
1098 
1099 static void ept_enable_ad_bits_or_skip_test(void)
1100 {
1101 	if (!ept_ad_bits_supported())
1102 		test_skip("EPT AD bits not supported.");
1103 	ept_enable_ad_bits();
1104 }
1105 
1106 static int apic_version;
1107 
1108 static int ept_init_common(bool have_ad)
1109 {
1110 	int ret;
1111 	struct pci_dev pcidev;
1112 
1113 	if (setup_ept(have_ad))
1114 		return VMX_TEST_EXIT;
1115 	data_page1 = alloc_page();
1116 	data_page2 = alloc_page();
1117 	memset(data_page1, 0x0, PAGE_SIZE);
1118 	memset(data_page2, 0x0, PAGE_SIZE);
1119 	*((u32 *)data_page1) = MAGIC_VAL_1;
1120 	*((u32 *)data_page2) = MAGIC_VAL_2;
1121 	install_ept(pml4, (unsigned long)data_page1, (unsigned long)data_page2,
1122 			EPT_RA | EPT_WA | EPT_EA);
1123 
1124 	apic_version = apic_read(APIC_LVR);
1125 
1126 	ret = pci_find_dev(PCI_VENDOR_ID_REDHAT, PCI_DEVICE_ID_REDHAT_TEST);
1127 	if (ret != PCIDEVADDR_INVALID) {
1128 		pci_dev_init(&pcidev, ret);
1129 		pci_physaddr = pcidev.resource[PCI_TESTDEV_BAR_MEM];
1130 	}
1131 
1132 	return VMX_TEST_START;
1133 }
1134 
1135 static int ept_init(struct vmcs *vmcs)
1136 {
1137 	return ept_init_common(false);
1138 }
1139 
1140 static void ept_common(void)
1141 {
1142 	vmx_set_test_stage(0);
1143 	if (*((u32 *)data_page2) != MAGIC_VAL_1 ||
1144 			*((u32 *)data_page1) != MAGIC_VAL_1)
1145 		report("EPT basic framework - read", 0);
1146 	else {
1147 		*((u32 *)data_page2) = MAGIC_VAL_3;
1148 		vmcall();
1149 		if (vmx_get_test_stage() == 1) {
1150 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1151 					*((u32 *)data_page2) == MAGIC_VAL_2)
1152 				report("EPT basic framework", 1);
1153 			else
1154 				report("EPT basic framework - remap", 1);
1155 		}
1156 	}
1157 	// Test EPT Misconfigurations
1158 	vmx_set_test_stage(1);
1159 	vmcall();
1160 	*((u32 *)data_page1) = MAGIC_VAL_1;
1161 	if (vmx_get_test_stage() != 2) {
1162 		report("EPT misconfigurations", 0);
1163 		goto t1;
1164 	}
1165 	vmx_set_test_stage(2);
1166 	vmcall();
1167 	*((u32 *)data_page1) = MAGIC_VAL_1;
1168 	report("EPT misconfigurations", vmx_get_test_stage() == 3);
1169 t1:
1170 	// Test EPT violation
1171 	vmx_set_test_stage(3);
1172 	vmcall();
1173 	*((u32 *)data_page1) = MAGIC_VAL_1;
1174 	report("EPT violation - page permission", vmx_get_test_stage() == 4);
1175 	// Violation caused by EPT paging structure
1176 	vmx_set_test_stage(4);
1177 	vmcall();
1178 	*((u32 *)data_page1) = MAGIC_VAL_2;
1179 	report("EPT violation - paging structure", vmx_get_test_stage() == 5);
1180 
1181 	// MMIO Read/Write
1182 	vmx_set_test_stage(5);
1183 	vmcall();
1184 
1185 	*(u32 volatile *)pci_physaddr;
1186 	report("MMIO EPT violation - read", vmx_get_test_stage() == 6);
1187 
1188 	*(u32 volatile *)pci_physaddr = MAGIC_VAL_1;
1189 	report("MMIO EPT violation - write", vmx_get_test_stage() == 7);
1190 }
1191 
1192 static void ept_main(void)
1193 {
1194 	ept_common();
1195 
1196 	// Test EPT access to L1 MMIO
1197 	vmx_set_test_stage(7);
1198 	report("EPT - MMIO access", *((u32 *)0xfee00030UL) == apic_version);
1199 
1200 	// Test invalid operand for INVEPT
1201 	vmcall();
1202 	report("EPT - unsupported INVEPT", vmx_get_test_stage() == 8);
1203 }
1204 
1205 static bool invept_test(int type, u64 eptp)
1206 {
1207 	bool ret, supported;
1208 
1209 	supported = ept_vpid.val & (EPT_CAP_INVEPT_SINGLE >> INVEPT_SINGLE << type);
1210 	ret = invept(type, eptp);
1211 
1212 	if (ret == !supported)
1213 		return false;
1214 
1215 	if (!supported)
1216 		printf("WARNING: unsupported invept passed!\n");
1217 	else
1218 		printf("WARNING: invept failed!\n");
1219 
1220 	return true;
1221 }
1222 
1223 static int pml_exit_handler(void)
1224 {
1225 	u16 index, count;
1226 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1227 	u64 *pmlbuf = pml_log;
1228 	u64 guest_rip = vmcs_read(GUEST_RIP);;
1229 	u64 guest_cr3 = vmcs_read(GUEST_CR3);
1230 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1231 
1232 	switch (reason) {
1233 	case VMX_VMCALL:
1234 		switch (vmx_get_test_stage()) {
1235 		case 0:
1236 			index = vmcs_read(GUEST_PML_INDEX);
1237 			for (count = index + 1; count < PML_INDEX; count++) {
1238 				if (pmlbuf[count] == (u64)data_page2) {
1239 					vmx_inc_test_stage();
1240 					clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1241 					break;
1242 				}
1243 			}
1244 			break;
1245 		case 1:
1246 			index = vmcs_read(GUEST_PML_INDEX);
1247 			/* Keep clearing the dirty bit till a overflow */
1248 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1249 			break;
1250 		default:
1251 			report("unexpected stage, %d.", false,
1252 			       vmx_get_test_stage());
1253 			print_vmexit_info();
1254 			return VMX_TEST_VMEXIT;
1255 		}
1256 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1257 		return VMX_TEST_RESUME;
1258 	case VMX_PML_FULL:
1259 		vmx_inc_test_stage();
1260 		vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1261 		return VMX_TEST_RESUME;
1262 	default:
1263 		report("Unknown exit reason, %ld", false, reason);
1264 		print_vmexit_info();
1265 	}
1266 	return VMX_TEST_VMEXIT;
1267 }
1268 
1269 static int ept_exit_handler_common(bool have_ad)
1270 {
1271 	u64 guest_rip;
1272 	u64 guest_cr3;
1273 	ulong reason;
1274 	u32 insn_len;
1275 	u32 exit_qual;
1276 	static unsigned long data_page1_pte, data_page1_pte_pte, memaddr_pte;
1277 
1278 	guest_rip = vmcs_read(GUEST_RIP);
1279 	guest_cr3 = vmcs_read(GUEST_CR3);
1280 	reason = vmcs_read(EXI_REASON) & 0xff;
1281 	insn_len = vmcs_read(EXI_INST_LEN);
1282 	exit_qual = vmcs_read(EXI_QUALIFICATION);
1283 	switch (reason) {
1284 	case VMX_VMCALL:
1285 		switch (vmx_get_test_stage()) {
1286 		case 0:
1287 			check_ept_ad(pml4, guest_cr3,
1288 				     (unsigned long)data_page1,
1289 				     have_ad ? EPT_ACCESS_FLAG : 0,
1290 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1291 			check_ept_ad(pml4, guest_cr3,
1292 				     (unsigned long)data_page2,
1293 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0,
1294 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1295 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1296 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page2);
1297 			if (have_ad)
1298 				ept_sync(INVEPT_SINGLE, eptp);;
1299 			if (*((u32 *)data_page1) == MAGIC_VAL_3 &&
1300 					*((u32 *)data_page2) == MAGIC_VAL_2) {
1301 				vmx_inc_test_stage();
1302 				install_ept(pml4, (unsigned long)data_page2,
1303 						(unsigned long)data_page2,
1304 						EPT_RA | EPT_WA | EPT_EA);
1305 			} else
1306 				report("EPT basic framework - write", 0);
1307 			break;
1308 		case 1:
1309 			install_ept(pml4, (unsigned long)data_page1,
1310  				(unsigned long)data_page1, EPT_WA);
1311 			ept_sync(INVEPT_SINGLE, eptp);
1312 			break;
1313 		case 2:
1314 			install_ept(pml4, (unsigned long)data_page1,
1315  				(unsigned long)data_page1,
1316  				EPT_RA | EPT_WA | EPT_EA |
1317  				(2 << EPT_MEM_TYPE_SHIFT));
1318 			ept_sync(INVEPT_SINGLE, eptp);
1319 			break;
1320 		case 3:
1321 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1322 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1323 						1, &data_page1_pte));
1324 			set_ept_pte(pml4, (unsigned long)data_page1,
1325 				1, data_page1_pte & ~EPT_PRESENT);
1326 			ept_sync(INVEPT_SINGLE, eptp);
1327 			break;
1328 		case 4:
1329 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)data_page1,
1330 						2, &data_page1_pte));
1331 			data_page1_pte &= PAGE_MASK;
1332 			TEST_ASSERT(get_ept_pte(pml4, data_page1_pte,
1333 						2, &data_page1_pte_pte));
1334 			set_ept_pte(pml4, data_page1_pte, 2,
1335 				data_page1_pte_pte & ~EPT_PRESENT);
1336 			ept_sync(INVEPT_SINGLE, eptp);
1337 			break;
1338 		case 5:
1339 			install_ept(pml4, (unsigned long)pci_physaddr,
1340 				(unsigned long)pci_physaddr, 0);
1341 			ept_sync(INVEPT_SINGLE, eptp);
1342 			break;
1343 		case 7:
1344 			if (!invept_test(0, eptp))
1345 				vmx_inc_test_stage();
1346 			break;
1347 		// Should not reach here
1348 		default:
1349 			report("ERROR - unexpected stage, %d.", false,
1350 			       vmx_get_test_stage());
1351 			print_vmexit_info();
1352 			return VMX_TEST_VMEXIT;
1353 		}
1354 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1355 		return VMX_TEST_RESUME;
1356 	case VMX_EPT_MISCONFIG:
1357 		switch (vmx_get_test_stage()) {
1358 		case 1:
1359 		case 2:
1360 			vmx_inc_test_stage();
1361 			install_ept(pml4, (unsigned long)data_page1,
1362  				(unsigned long)data_page1,
1363  				EPT_RA | EPT_WA | EPT_EA);
1364 			ept_sync(INVEPT_SINGLE, eptp);
1365 			break;
1366 		// Should not reach here
1367 		default:
1368 			report("ERROR - unexpected stage, %d.", false,
1369 			       vmx_get_test_stage());
1370 			print_vmexit_info();
1371 			return VMX_TEST_VMEXIT;
1372 		}
1373 		return VMX_TEST_RESUME;
1374 	case VMX_EPT_VIOLATION:
1375 		switch(vmx_get_test_stage()) {
1376 		case 3:
1377 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1378 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1379 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1380 			if (exit_qual == (EPT_VLT_WR | EPT_VLT_LADDR_VLD |
1381 					EPT_VLT_PADDR))
1382 				vmx_inc_test_stage();
1383 			set_ept_pte(pml4, (unsigned long)data_page1,
1384 				1, data_page1_pte | (EPT_PRESENT));
1385 			ept_sync(INVEPT_SINGLE, eptp);
1386 			break;
1387 		case 4:
1388 			check_ept_ad(pml4, guest_cr3, (unsigned long)data_page1, 0,
1389 				     have_ad ? EPT_ACCESS_FLAG | EPT_DIRTY_FLAG : 0);
1390 			clear_ept_ad(pml4, guest_cr3, (unsigned long)data_page1);
1391 			if (exit_qual == (EPT_VLT_RD |
1392 					  (have_ad ? EPT_VLT_WR : 0) |
1393 					  EPT_VLT_LADDR_VLD))
1394 				vmx_inc_test_stage();
1395 			set_ept_pte(pml4, data_page1_pte, 2,
1396 				data_page1_pte_pte | (EPT_PRESENT));
1397 			ept_sync(INVEPT_SINGLE, eptp);
1398 			break;
1399 		case 5:
1400 			if (exit_qual & EPT_VLT_RD)
1401 				vmx_inc_test_stage();
1402 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1403 						1, &memaddr_pte));
1404 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA);
1405 			ept_sync(INVEPT_SINGLE, eptp);
1406 			break;
1407 		case 6:
1408 			if (exit_qual & EPT_VLT_WR)
1409 				vmx_inc_test_stage();
1410 			TEST_ASSERT(get_ept_pte(pml4, (unsigned long)pci_physaddr,
1411 						1, &memaddr_pte));
1412 			set_ept_pte(pml4, memaddr_pte, 1, memaddr_pte | EPT_RA | EPT_WA);
1413 			ept_sync(INVEPT_SINGLE, eptp);
1414 			break;
1415 		default:
1416 			// Should not reach here
1417 			report("ERROR : unexpected stage, %d", false,
1418 			       vmx_get_test_stage());
1419 			print_vmexit_info();
1420 			return VMX_TEST_VMEXIT;
1421 		}
1422 		return VMX_TEST_RESUME;
1423 	default:
1424 		report("Unknown exit reason, %ld", false, reason);
1425 		print_vmexit_info();
1426 	}
1427 	return VMX_TEST_VMEXIT;
1428 }
1429 
1430 static int ept_exit_handler(void)
1431 {
1432 	return ept_exit_handler_common(false);
1433 }
1434 
1435 static int eptad_init(struct vmcs *vmcs)
1436 {
1437 	int r = ept_init_common(true);
1438 
1439 	if (r == VMX_TEST_EXIT)
1440 		return r;
1441 
1442 	if ((rdmsr(MSR_IA32_VMX_EPT_VPID_CAP) & EPT_CAP_AD_FLAG) == 0) {
1443 		printf("\tEPT A/D bits are not supported");
1444 		return VMX_TEST_EXIT;
1445 	}
1446 
1447 	return r;
1448 }
1449 
1450 static int pml_init(struct vmcs *vmcs)
1451 {
1452 	u32 ctrl_cpu;
1453 	int r = eptad_init(vmcs);
1454 
1455 	if (r == VMX_TEST_EXIT)
1456 		return r;
1457 
1458 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1459 		!(ctrl_cpu_rev[1].clr & CPU_PML)) {
1460 		printf("\tPML is not supported");
1461 		return VMX_TEST_EXIT;
1462 	}
1463 
1464 	pml_log = alloc_page();
1465 	memset(pml_log, 0x0, PAGE_SIZE);
1466 	vmcs_write(PMLADDR, (u64)pml_log);
1467 	vmcs_write(GUEST_PML_INDEX, PML_INDEX - 1);
1468 
1469 	ctrl_cpu = vmcs_read(CPU_EXEC_CTRL1) | CPU_PML;
1470 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu);
1471 
1472 	return VMX_TEST_START;
1473 }
1474 
1475 static void pml_main(void)
1476 {
1477 	int count = 0;
1478 
1479 	vmx_set_test_stage(0);
1480 	*((u32 *)data_page2) = 0x1;
1481 	vmcall();
1482 	report("PML - Dirty GPA Logging", vmx_get_test_stage() == 1);
1483 
1484 	while (vmx_get_test_stage() == 1) {
1485 		vmcall();
1486 		*((u32 *)data_page2) = 0x1;
1487 		if (count++ > PML_INDEX)
1488 			break;
1489 	}
1490 	report("PML Full Event", vmx_get_test_stage() == 2);
1491 }
1492 
1493 static void eptad_main(void)
1494 {
1495 	ept_common();
1496 }
1497 
1498 static int eptad_exit_handler(void)
1499 {
1500 	return ept_exit_handler_common(true);
1501 }
1502 
1503 static bool invvpid_test(int type, u16 vpid)
1504 {
1505 	bool ret, supported;
1506 
1507 	supported = ept_vpid.val &
1508 		(VPID_CAP_INVVPID_ADDR >> INVVPID_ADDR << type);
1509 	ret = invvpid(type, vpid, 0);
1510 
1511 	if (ret == !supported)
1512 		return false;
1513 
1514 	if (!supported)
1515 		printf("WARNING: unsupported invvpid passed!\n");
1516 	else
1517 		printf("WARNING: invvpid failed!\n");
1518 
1519 	return true;
1520 }
1521 
1522 static int vpid_init(struct vmcs *vmcs)
1523 {
1524 	u32 ctrl_cpu1;
1525 
1526 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
1527 		!(ctrl_cpu_rev[1].clr & CPU_VPID)) {
1528 		printf("\tVPID is not supported");
1529 		return VMX_TEST_EXIT;
1530 	}
1531 
1532 	ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
1533 	ctrl_cpu1 |= CPU_VPID;
1534 	vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
1535 	return VMX_TEST_START;
1536 }
1537 
1538 static void vpid_main(void)
1539 {
1540 	vmx_set_test_stage(0);
1541 	vmcall();
1542 	report("INVVPID SINGLE ADDRESS", vmx_get_test_stage() == 1);
1543 	vmx_set_test_stage(2);
1544 	vmcall();
1545 	report("INVVPID SINGLE", vmx_get_test_stage() == 3);
1546 	vmx_set_test_stage(4);
1547 	vmcall();
1548 	report("INVVPID ALL", vmx_get_test_stage() == 5);
1549 }
1550 
1551 static int vpid_exit_handler(void)
1552 {
1553 	u64 guest_rip;
1554 	ulong reason;
1555 	u32 insn_len;
1556 
1557 	guest_rip = vmcs_read(GUEST_RIP);
1558 	reason = vmcs_read(EXI_REASON) & 0xff;
1559 	insn_len = vmcs_read(EXI_INST_LEN);
1560 
1561 	switch (reason) {
1562 	case VMX_VMCALL:
1563 		switch(vmx_get_test_stage()) {
1564 		case 0:
1565 			if (!invvpid_test(INVVPID_ADDR, 1))
1566 				vmx_inc_test_stage();
1567 			break;
1568 		case 2:
1569 			if (!invvpid_test(INVVPID_CONTEXT_GLOBAL, 1))
1570 				vmx_inc_test_stage();
1571 			break;
1572 		case 4:
1573 			if (!invvpid_test(INVVPID_ALL, 1))
1574 				vmx_inc_test_stage();
1575 			break;
1576 		default:
1577 			report("ERROR: unexpected stage, %d", false,
1578 					vmx_get_test_stage());
1579 			print_vmexit_info();
1580 			return VMX_TEST_VMEXIT;
1581 		}
1582 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1583 		return VMX_TEST_RESUME;
1584 	default:
1585 		report("Unknown exit reason, %ld", false, reason);
1586 		print_vmexit_info();
1587 	}
1588 	return VMX_TEST_VMEXIT;
1589 }
1590 
1591 #define TIMER_VECTOR	222
1592 
1593 static volatile bool timer_fired;
1594 
1595 static void timer_isr(isr_regs_t *regs)
1596 {
1597 	timer_fired = true;
1598 	apic_write(APIC_EOI, 0);
1599 }
1600 
1601 static int interrupt_init(struct vmcs *vmcs)
1602 {
1603 	msr_bmp_init();
1604 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1605 	handle_irq(TIMER_VECTOR, timer_isr);
1606 	return VMX_TEST_START;
1607 }
1608 
1609 static void interrupt_main(void)
1610 {
1611 	long long start, loops;
1612 
1613 	vmx_set_test_stage(0);
1614 
1615 	apic_write(APIC_LVTT, TIMER_VECTOR);
1616 	irq_enable();
1617 
1618 	apic_write(APIC_TMICT, 1);
1619 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1620 		asm volatile ("nop");
1621 	report("direct interrupt while running guest", timer_fired);
1622 
1623 	apic_write(APIC_TMICT, 0);
1624 	irq_disable();
1625 	vmcall();
1626 	timer_fired = false;
1627 	apic_write(APIC_TMICT, 1);
1628 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1629 		asm volatile ("nop");
1630 	report("intercepted interrupt while running guest", timer_fired);
1631 
1632 	irq_enable();
1633 	apic_write(APIC_TMICT, 0);
1634 	irq_disable();
1635 	vmcall();
1636 	timer_fired = false;
1637 	start = rdtsc();
1638 	apic_write(APIC_TMICT, 1000000);
1639 
1640 	asm volatile ("sti; hlt");
1641 
1642 	report("direct interrupt + hlt",
1643 	       rdtsc() - start > 1000000 && timer_fired);
1644 
1645 	apic_write(APIC_TMICT, 0);
1646 	irq_disable();
1647 	vmcall();
1648 	timer_fired = false;
1649 	start = rdtsc();
1650 	apic_write(APIC_TMICT, 1000000);
1651 
1652 	asm volatile ("sti; hlt");
1653 
1654 	report("intercepted interrupt + hlt",
1655 	       rdtsc() - start > 10000 && timer_fired);
1656 
1657 	apic_write(APIC_TMICT, 0);
1658 	irq_disable();
1659 	vmcall();
1660 	timer_fired = false;
1661 	start = rdtsc();
1662 	apic_write(APIC_TMICT, 1000000);
1663 
1664 	irq_enable();
1665 	asm volatile ("nop");
1666 	vmcall();
1667 
1668 	report("direct interrupt + activity state hlt",
1669 	       rdtsc() - start > 10000 && timer_fired);
1670 
1671 	apic_write(APIC_TMICT, 0);
1672 	irq_disable();
1673 	vmcall();
1674 	timer_fired = false;
1675 	start = rdtsc();
1676 	apic_write(APIC_TMICT, 1000000);
1677 
1678 	irq_enable();
1679 	asm volatile ("nop");
1680 	vmcall();
1681 
1682 	report("intercepted interrupt + activity state hlt",
1683 	       rdtsc() - start > 10000 && timer_fired);
1684 
1685 	apic_write(APIC_TMICT, 0);
1686 	irq_disable();
1687 	vmx_set_test_stage(7);
1688 	vmcall();
1689 	timer_fired = false;
1690 	apic_write(APIC_TMICT, 1);
1691 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1692 		asm volatile ("nop");
1693 	report("running a guest with interrupt acknowledgement set", timer_fired);
1694 
1695 	apic_write(APIC_TMICT, 0);
1696 	irq_enable();
1697 	timer_fired = false;
1698 	vmcall();
1699 	report("Inject an event to a halted guest", timer_fired);
1700 }
1701 
1702 static int interrupt_exit_handler(void)
1703 {
1704 	u64 guest_rip = vmcs_read(GUEST_RIP);
1705 	ulong reason = vmcs_read(EXI_REASON) & 0xff;
1706 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1707 
1708 	switch (reason) {
1709 	case VMX_VMCALL:
1710 		switch (vmx_get_test_stage()) {
1711 		case 0:
1712 		case 2:
1713 		case 5:
1714 			vmcs_write(PIN_CONTROLS,
1715 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1716 			break;
1717 		case 7:
1718 			vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_INTA);
1719 			vmcs_write(PIN_CONTROLS,
1720 				   vmcs_read(PIN_CONTROLS) | PIN_EXTINT);
1721 			break;
1722 		case 1:
1723 		case 3:
1724 			vmcs_write(PIN_CONTROLS,
1725 				   vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
1726 			break;
1727 		case 4:
1728 		case 6:
1729 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1730 			break;
1731 
1732 		case 8:
1733 			vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
1734 			vmcs_write(ENT_INTR_INFO,
1735 				   TIMER_VECTOR |
1736 				   (VMX_INTR_TYPE_EXT_INTR << INTR_INFO_INTR_TYPE_SHIFT) |
1737 				   INTR_INFO_VALID_MASK);
1738 			break;
1739 		}
1740 		vmx_inc_test_stage();
1741 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1742 		return VMX_TEST_RESUME;
1743 	case VMX_EXTINT:
1744 		if (vmcs_read(EXI_CONTROLS) & EXI_INTA) {
1745 			int vector = vmcs_read(EXI_INTR_INFO) & 0xff;
1746 			handle_external_interrupt(vector);
1747 		} else {
1748 			irq_enable();
1749 			asm volatile ("nop");
1750 			irq_disable();
1751 		}
1752 		if (vmx_get_test_stage() >= 2)
1753 			vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
1754 		return VMX_TEST_RESUME;
1755 	default:
1756 		report("Unknown exit reason, %ld", false, reason);
1757 		print_vmexit_info();
1758 	}
1759 
1760 	return VMX_TEST_VMEXIT;
1761 }
1762 
1763 static int dbgctls_init(struct vmcs *vmcs)
1764 {
1765 	u64 dr7 = 0x402;
1766 	u64 zero = 0;
1767 
1768 	msr_bmp_init();
1769 	asm volatile(
1770 		"mov %0,%%dr0\n\t"
1771 		"mov %0,%%dr1\n\t"
1772 		"mov %0,%%dr2\n\t"
1773 		"mov %1,%%dr7\n\t"
1774 		: : "r" (zero), "r" (dr7));
1775 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1776 	vmcs_write(GUEST_DR7, 0x404);
1777 	vmcs_write(GUEST_DEBUGCTL, 0x2);
1778 
1779 	vmcs_write(ENT_CONTROLS, vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
1780 	vmcs_write(EXI_CONTROLS, vmcs_read(EXI_CONTROLS) | EXI_SAVE_DBGCTLS);
1781 
1782 	return VMX_TEST_START;
1783 }
1784 
1785 static void dbgctls_main(void)
1786 {
1787 	u64 dr7, debugctl;
1788 
1789 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1790 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1791 	/* Commented out: KVM does not support DEBUGCTL so far */
1792 	(void)debugctl;
1793 	report("Load debug controls", dr7 == 0x404 /* && debugctl == 0x2 */);
1794 
1795 	dr7 = 0x408;
1796 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1797 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1798 
1799 	vmx_set_test_stage(0);
1800 	vmcall();
1801 	report("Save debug controls", vmx_get_test_stage() == 1);
1802 
1803 	if (ctrl_enter_rev.set & ENT_LOAD_DBGCTLS ||
1804 	    ctrl_exit_rev.set & EXI_SAVE_DBGCTLS) {
1805 		printf("\tDebug controls are always loaded/saved\n");
1806 		return;
1807 	}
1808 	vmx_set_test_stage(2);
1809 	vmcall();
1810 
1811 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1812 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1813 	/* Commented out: KVM does not support DEBUGCTL so far */
1814 	(void)debugctl;
1815 	report("Guest=host debug controls", dr7 == 0x402 /* && debugctl == 0x1 */);
1816 
1817 	dr7 = 0x408;
1818 	asm volatile("mov %0,%%dr7" : : "r" (dr7));
1819 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0x3);
1820 
1821 	vmx_set_test_stage(3);
1822 	vmcall();
1823 	report("Don't save debug controls", vmx_get_test_stage() == 4);
1824 }
1825 
1826 static int dbgctls_exit_handler(void)
1827 {
1828 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
1829 	u32 insn_len = vmcs_read(EXI_INST_LEN);
1830 	u64 guest_rip = vmcs_read(GUEST_RIP);
1831 	u64 dr7, debugctl;
1832 
1833 	asm volatile("mov %%dr7,%0" : "=r" (dr7));
1834 	debugctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
1835 
1836 	switch (reason) {
1837 	case VMX_VMCALL:
1838 		switch (vmx_get_test_stage()) {
1839 		case 0:
1840 			if (dr7 == 0x400 && debugctl == 0 &&
1841 			    vmcs_read(GUEST_DR7) == 0x408 /* &&
1842 			    Commented out: KVM does not support DEBUGCTL so far
1843 			    vmcs_read(GUEST_DEBUGCTL) == 0x3 */)
1844 				vmx_inc_test_stage();
1845 			break;
1846 		case 2:
1847 			dr7 = 0x402;
1848 			asm volatile("mov %0,%%dr7" : : "r" (dr7));
1849 			wrmsr(MSR_IA32_DEBUGCTLMSR, 0x1);
1850 			vmcs_write(GUEST_DR7, 0x404);
1851 			vmcs_write(GUEST_DEBUGCTL, 0x2);
1852 
1853 			vmcs_write(ENT_CONTROLS,
1854 				vmcs_read(ENT_CONTROLS) & ~ENT_LOAD_DBGCTLS);
1855 			vmcs_write(EXI_CONTROLS,
1856 				vmcs_read(EXI_CONTROLS) & ~EXI_SAVE_DBGCTLS);
1857 			break;
1858 		case 3:
1859 			if (dr7 == 0x400 && debugctl == 0 &&
1860 			    vmcs_read(GUEST_DR7) == 0x404 /* &&
1861 			    Commented out: KVM does not support DEBUGCTL so far
1862 			    vmcs_read(GUEST_DEBUGCTL) == 0x2 */)
1863 				vmx_inc_test_stage();
1864 			break;
1865 		}
1866 		vmcs_write(GUEST_RIP, guest_rip + insn_len);
1867 		return VMX_TEST_RESUME;
1868 	default:
1869 		report("Unknown exit reason, %d", false, reason);
1870 		print_vmexit_info();
1871 	}
1872 	return VMX_TEST_VMEXIT;
1873 }
1874 
1875 struct vmx_msr_entry {
1876 	u32 index;
1877 	u32 reserved;
1878 	u64 value;
1879 } __attribute__((packed));
1880 
1881 #define MSR_MAGIC 0x31415926
1882 struct vmx_msr_entry *exit_msr_store, *entry_msr_load, *exit_msr_load;
1883 
1884 static int msr_switch_init(struct vmcs *vmcs)
1885 {
1886 	msr_bmp_init();
1887 	exit_msr_store = alloc_page();
1888 	exit_msr_load = alloc_page();
1889 	entry_msr_load = alloc_page();
1890 	memset(exit_msr_store, 0, PAGE_SIZE);
1891 	memset(exit_msr_load, 0, PAGE_SIZE);
1892 	memset(entry_msr_load, 0, PAGE_SIZE);
1893 	entry_msr_load[0].index = MSR_KERNEL_GS_BASE;
1894 	entry_msr_load[0].value = MSR_MAGIC;
1895 
1896 	vmx_set_test_stage(1);
1897 	vmcs_write(ENT_MSR_LD_CNT, 1);
1898 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)entry_msr_load);
1899 	vmcs_write(EXI_MSR_ST_CNT, 1);
1900 	vmcs_write(EXIT_MSR_ST_ADDR, (u64)exit_msr_store);
1901 	vmcs_write(EXI_MSR_LD_CNT, 1);
1902 	vmcs_write(EXIT_MSR_LD_ADDR, (u64)exit_msr_load);
1903 	return VMX_TEST_START;
1904 }
1905 
1906 static void msr_switch_main(void)
1907 {
1908 	if (vmx_get_test_stage() == 1) {
1909 		report("VM entry MSR load",
1910 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC);
1911 		vmx_set_test_stage(2);
1912 		wrmsr(MSR_KERNEL_GS_BASE, MSR_MAGIC + 1);
1913 		exit_msr_store[0].index = MSR_KERNEL_GS_BASE;
1914 		exit_msr_load[0].index = MSR_KERNEL_GS_BASE;
1915 		exit_msr_load[0].value = MSR_MAGIC + 2;
1916 	}
1917 	vmcall();
1918 }
1919 
1920 static int msr_switch_exit_handler(void)
1921 {
1922 	ulong reason;
1923 
1924 	reason = vmcs_read(EXI_REASON);
1925 	if (reason == VMX_VMCALL && vmx_get_test_stage() == 2) {
1926 		report("VM exit MSR store",
1927 			exit_msr_store[0].value == MSR_MAGIC + 1);
1928 		report("VM exit MSR load",
1929 			rdmsr(MSR_KERNEL_GS_BASE) == MSR_MAGIC + 2);
1930 		vmx_set_test_stage(3);
1931 		entry_msr_load[0].index = MSR_FS_BASE;
1932 		return VMX_TEST_RESUME;
1933 	}
1934 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1935 		__func__, vmx_get_test_stage(), reason);
1936 	return VMX_TEST_EXIT;
1937 }
1938 
1939 static int msr_switch_entry_failure(struct vmentry_failure *failure)
1940 {
1941 	ulong reason;
1942 
1943 	if (failure->early) {
1944 		printf("ERROR %s: early exit\n", __func__);
1945 		return VMX_TEST_EXIT;
1946 	}
1947 
1948 	reason = vmcs_read(EXI_REASON);
1949 	if (reason == (VMX_ENTRY_FAILURE | VMX_FAIL_MSR) &&
1950 	    vmx_get_test_stage() == 3) {
1951 		report("VM entry MSR load: try to load FS_BASE",
1952 			vmcs_read(EXI_QUALIFICATION) == 1);
1953 		return VMX_TEST_VMEXIT;
1954 	}
1955 	printf("ERROR %s: unexpected stage=%u or reason=%lu\n",
1956 		__func__, vmx_get_test_stage(), reason);
1957 	return VMX_TEST_EXIT;
1958 }
1959 
1960 static int vmmcall_init(struct vmcs *vmcs)
1961 {
1962 	vmcs_write(EXC_BITMAP, 1 << UD_VECTOR);
1963 	return VMX_TEST_START;
1964 }
1965 
1966 static void vmmcall_main(void)
1967 {
1968 	asm volatile(
1969 		"mov $0xABCD, %%rax\n\t"
1970 		"vmmcall\n\t"
1971 		::: "rax");
1972 
1973 	report("VMMCALL", 0);
1974 }
1975 
1976 static int vmmcall_exit_handler(void)
1977 {
1978 	ulong reason;
1979 
1980 	reason = vmcs_read(EXI_REASON);
1981 	switch (reason) {
1982 	case VMX_VMCALL:
1983 		printf("here\n");
1984 		report("VMMCALL triggers #UD", 0);
1985 		break;
1986 	case VMX_EXC_NMI:
1987 		report("VMMCALL triggers #UD",
1988 		       (vmcs_read(EXI_INTR_INFO) & 0xff) == UD_VECTOR);
1989 		break;
1990 	default:
1991 		report("Unknown exit reason, %ld", false, reason);
1992 		print_vmexit_info();
1993 	}
1994 
1995 	return VMX_TEST_VMEXIT;
1996 }
1997 
1998 static int disable_rdtscp_init(struct vmcs *vmcs)
1999 {
2000 	u32 ctrl_cpu1;
2001 
2002 	if (ctrl_cpu_rev[0].clr & CPU_SECONDARY) {
2003 		ctrl_cpu1 = vmcs_read(CPU_EXEC_CTRL1);
2004 		ctrl_cpu1 &= ~CPU_RDTSCP;
2005 		vmcs_write(CPU_EXEC_CTRL1, ctrl_cpu1);
2006 	}
2007 
2008 	return VMX_TEST_START;
2009 }
2010 
2011 static void disable_rdtscp_ud_handler(struct ex_regs *regs)
2012 {
2013 	switch (vmx_get_test_stage()) {
2014 	case 0:
2015 		report("RDTSCP triggers #UD", true);
2016 		vmx_inc_test_stage();
2017 		regs->rip += 3;
2018 		break;
2019 	case 2:
2020 		report("RDPID triggers #UD", true);
2021 		vmx_inc_test_stage();
2022 		regs->rip += 4;
2023 		break;
2024 	}
2025 	return;
2026 
2027 }
2028 
2029 static void disable_rdtscp_main(void)
2030 {
2031 	/* Test that #UD is properly injected in L2.  */
2032 	handle_exception(UD_VECTOR, disable_rdtscp_ud_handler);
2033 
2034 	vmx_set_test_stage(0);
2035 	asm volatile("rdtscp" : : : "eax", "ecx", "edx");
2036 	vmcall();
2037 	asm volatile(".byte 0xf3, 0x0f, 0xc7, 0xf8" : : : "eax");
2038 
2039 	handle_exception(UD_VECTOR, 0);
2040 	vmcall();
2041 }
2042 
2043 static int disable_rdtscp_exit_handler(void)
2044 {
2045 	unsigned int reason = vmcs_read(EXI_REASON) & 0xff;
2046 
2047 	switch (reason) {
2048 	case VMX_VMCALL:
2049 		switch (vmx_get_test_stage()) {
2050 		case 0:
2051 			report("RDTSCP triggers #UD", false);
2052 			vmx_inc_test_stage();
2053 			/* fallthrough */
2054 		case 1:
2055 			vmx_inc_test_stage();
2056 			vmcs_write(GUEST_RIP, vmcs_read(GUEST_RIP) + 3);
2057 			return VMX_TEST_RESUME;
2058 		case 2:
2059 			report("RDPID triggers #UD", false);
2060 			break;
2061 		}
2062 		break;
2063 
2064 	default:
2065 		report("Unknown exit reason, %d", false, reason);
2066 		print_vmexit_info();
2067 	}
2068 	return VMX_TEST_VMEXIT;
2069 }
2070 
2071 static int int3_init(struct vmcs *vmcs)
2072 {
2073 	vmcs_write(EXC_BITMAP, ~0u);
2074 	return VMX_TEST_START;
2075 }
2076 
2077 static void int3_guest_main(void)
2078 {
2079 	asm volatile ("int3");
2080 }
2081 
2082 static int int3_exit_handler(void)
2083 {
2084 	u32 reason = vmcs_read(EXI_REASON);
2085 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2086 
2087 	report("L1 intercepts #BP", reason == VMX_EXC_NMI &&
2088 	       (intr_info & INTR_INFO_VALID_MASK) &&
2089 	       (intr_info & INTR_INFO_VECTOR_MASK) == BP_VECTOR &&
2090 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2091 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2092 
2093 	return VMX_TEST_VMEXIT;
2094 }
2095 
2096 static int into_init(struct vmcs *vmcs)
2097 {
2098 	vmcs_write(EXC_BITMAP, ~0u);
2099 	return VMX_TEST_START;
2100 }
2101 
2102 static void into_guest_main(void)
2103 {
2104 	struct far_pointer32 fp = {
2105 		.offset = (uintptr_t)&&into,
2106 		.selector = KERNEL_CS32,
2107 	};
2108 	register uintptr_t rsp asm("rsp");
2109 
2110 	if (fp.offset != (uintptr_t)&&into) {
2111 		printf("Code address too high.\n");
2112 		return;
2113 	}
2114 	if ((u32)rsp != rsp) {
2115 		printf("Stack address too high.\n");
2116 		return;
2117 	}
2118 
2119 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : into);
2120 	return;
2121 into:
2122 	asm volatile (".code32;"
2123 		      "movl $0x7fffffff, %eax;"
2124 		      "addl %eax, %eax;"
2125 		      "into;"
2126 		      "lret;"
2127 		      ".code64");
2128 	__builtin_unreachable();
2129 }
2130 
2131 static int into_exit_handler(void)
2132 {
2133 	u32 reason = vmcs_read(EXI_REASON);
2134 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
2135 
2136 	report("L1 intercepts #OF", reason == VMX_EXC_NMI &&
2137 	       (intr_info & INTR_INFO_VALID_MASK) &&
2138 	       (intr_info & INTR_INFO_VECTOR_MASK) == OF_VECTOR &&
2139 	       ((intr_info & INTR_INFO_INTR_TYPE_MASK) >>
2140 		INTR_INFO_INTR_TYPE_SHIFT) == VMX_INTR_TYPE_SOFT_EXCEPTION);
2141 
2142 	return VMX_TEST_VMEXIT;
2143 }
2144 
2145 static void exit_monitor_from_l2_main(void)
2146 {
2147 	printf("Calling exit(0) from l2...\n");
2148 	exit(0);
2149 }
2150 
2151 static int exit_monitor_from_l2_handler(void)
2152 {
2153 	report("The guest should have killed the VMM", false);
2154 	return VMX_TEST_EXIT;
2155 }
2156 
2157 static void assert_exit_reason(u64 expected)
2158 {
2159 	u64 actual = vmcs_read(EXI_REASON);
2160 
2161 	TEST_ASSERT_EQ_MSG(expected, actual, "Expected %s, got %s.",
2162 			   exit_reason_description(expected),
2163 			   exit_reason_description(actual));
2164 }
2165 
2166 static void skip_exit_insn(void)
2167 {
2168 	u64 guest_rip = vmcs_read(GUEST_RIP);
2169 	u32 insn_len = vmcs_read(EXI_INST_LEN);
2170 	vmcs_write(GUEST_RIP, guest_rip + insn_len);
2171 }
2172 
2173 static void skip_exit_vmcall(void)
2174 {
2175 	assert_exit_reason(VMX_VMCALL);
2176 	skip_exit_insn();
2177 }
2178 
2179 static void v2_null_test_guest(void)
2180 {
2181 }
2182 
2183 static void v2_null_test(void)
2184 {
2185 	test_set_guest(v2_null_test_guest);
2186 	enter_guest();
2187 	report(__func__, 1);
2188 }
2189 
2190 static void v2_multiple_entries_test_guest(void)
2191 {
2192 	vmx_set_test_stage(1);
2193 	vmcall();
2194 	vmx_set_test_stage(2);
2195 }
2196 
2197 static void v2_multiple_entries_test(void)
2198 {
2199 	test_set_guest(v2_multiple_entries_test_guest);
2200 	enter_guest();
2201 	TEST_ASSERT_EQ(vmx_get_test_stage(), 1);
2202 	skip_exit_vmcall();
2203 	enter_guest();
2204 	TEST_ASSERT_EQ(vmx_get_test_stage(), 2);
2205 	report(__func__, 1);
2206 }
2207 
2208 static int fixture_test_data = 1;
2209 
2210 static void fixture_test_teardown(void *data)
2211 {
2212 	*((int *) data) = 1;
2213 }
2214 
2215 static void fixture_test_guest(void)
2216 {
2217 	fixture_test_data++;
2218 }
2219 
2220 
2221 static void fixture_test_setup(void)
2222 {
2223 	TEST_ASSERT_EQ_MSG(1, fixture_test_data,
2224 			   "fixture_test_teardown didn't run?!");
2225 	fixture_test_data = 2;
2226 	test_add_teardown(fixture_test_teardown, &fixture_test_data);
2227 	test_set_guest(fixture_test_guest);
2228 }
2229 
2230 static void fixture_test_case1(void)
2231 {
2232 	fixture_test_setup();
2233 	TEST_ASSERT_EQ(2, fixture_test_data);
2234 	enter_guest();
2235 	TEST_ASSERT_EQ(3, fixture_test_data);
2236 	report(__func__, 1);
2237 }
2238 
2239 static void fixture_test_case2(void)
2240 {
2241 	fixture_test_setup();
2242 	TEST_ASSERT_EQ(2, fixture_test_data);
2243 	enter_guest();
2244 	TEST_ASSERT_EQ(3, fixture_test_data);
2245 	report(__func__, 1);
2246 }
2247 
2248 enum ept_access_op {
2249 	OP_READ,
2250 	OP_WRITE,
2251 	OP_EXEC,
2252 	OP_FLUSH_TLB,
2253 	OP_EXIT,
2254 };
2255 
2256 static struct ept_access_test_data {
2257 	unsigned long gpa;
2258 	unsigned long *gva;
2259 	unsigned long hpa;
2260 	unsigned long *hva;
2261 	enum ept_access_op op;
2262 } ept_access_test_data;
2263 
2264 extern unsigned char ret42_start;
2265 extern unsigned char ret42_end;
2266 
2267 /* Returns 42. */
2268 asm(
2269 	".align 64\n"
2270 	"ret42_start:\n"
2271 	"mov $42, %eax\n"
2272 	"ret\n"
2273 	"ret42_end:\n"
2274 );
2275 
2276 static void
2277 diagnose_ept_violation_qual(u64 expected, u64 actual)
2278 {
2279 
2280 #define DIAGNOSE(flag)							\
2281 do {									\
2282 	if ((expected & flag) != (actual & flag))			\
2283 		printf(#flag " %sexpected\n",				\
2284 		       (expected & flag) ? "" : "un");			\
2285 } while (0)
2286 
2287 	DIAGNOSE(EPT_VLT_RD);
2288 	DIAGNOSE(EPT_VLT_WR);
2289 	DIAGNOSE(EPT_VLT_FETCH);
2290 	DIAGNOSE(EPT_VLT_PERM_RD);
2291 	DIAGNOSE(EPT_VLT_PERM_WR);
2292 	DIAGNOSE(EPT_VLT_PERM_EX);
2293 	DIAGNOSE(EPT_VLT_LADDR_VLD);
2294 	DIAGNOSE(EPT_VLT_PADDR);
2295 
2296 #undef DIAGNOSE
2297 }
2298 
2299 static void do_ept_access_op(enum ept_access_op op)
2300 {
2301 	ept_access_test_data.op = op;
2302 	enter_guest();
2303 }
2304 
2305 /*
2306  * Force the guest to flush its TLB (i.e., flush gva -> gpa mappings). Only
2307  * needed by tests that modify guest PTEs.
2308  */
2309 static void ept_access_test_guest_flush_tlb(void)
2310 {
2311 	do_ept_access_op(OP_FLUSH_TLB);
2312 	skip_exit_vmcall();
2313 }
2314 
2315 /*
2316  * Modifies the EPT entry at @level in the mapping of @gpa. First clears the
2317  * bits in @clear then sets the bits in @set. @mkhuge transforms the entry into
2318  * a huge page.
2319  */
2320 static unsigned long ept_twiddle(unsigned long gpa, bool mkhuge, int level,
2321 				 unsigned long clear, unsigned long set)
2322 {
2323 	struct ept_access_test_data *data = &ept_access_test_data;
2324 	unsigned long orig_pte;
2325 	unsigned long pte;
2326 
2327 	/* Screw with the mapping at the requested level. */
2328 	TEST_ASSERT(get_ept_pte(pml4, gpa, level, &orig_pte));
2329 	pte = orig_pte;
2330 	if (mkhuge)
2331 		pte = (orig_pte & ~EPT_ADDR_MASK) | data->hpa | EPT_LARGE_PAGE;
2332 	else
2333 		pte = orig_pte;
2334 	pte = (pte & ~clear) | set;
2335 	set_ept_pte(pml4, gpa, level, pte);
2336 	ept_sync(INVEPT_SINGLE, eptp);
2337 
2338 	return orig_pte;
2339 }
2340 
2341 static void ept_untwiddle(unsigned long gpa, int level, unsigned long orig_pte)
2342 {
2343 	set_ept_pte(pml4, gpa, level, orig_pte);
2344 }
2345 
2346 static void do_ept_violation(bool leaf, enum ept_access_op op,
2347 			     u64 expected_qual, u64 expected_paddr)
2348 {
2349 	u64 qual;
2350 
2351 	/* Try the access and observe the violation. */
2352 	do_ept_access_op(op);
2353 
2354 	assert_exit_reason(VMX_EPT_VIOLATION);
2355 
2356 	qual = vmcs_read(EXI_QUALIFICATION);
2357 
2358 	diagnose_ept_violation_qual(expected_qual, qual);
2359 	TEST_EXPECT_EQ(expected_qual, qual);
2360 
2361 	#if 0
2362 	/* Disable for now otherwise every test will fail */
2363 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2364 		       (unsigned long) (
2365 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2366 	#endif
2367 	/*
2368 	 * TODO: tests that probe expected_paddr in pages other than the one at
2369 	 * the beginning of the 1g region.
2370 	 */
2371 	TEST_EXPECT_EQ(vmcs_read(INFO_PHYS_ADDR), expected_paddr);
2372 }
2373 
2374 static void
2375 ept_violation_at_level_mkhuge(bool mkhuge, int level, unsigned long clear,
2376 			      unsigned long set, enum ept_access_op op,
2377 			      u64 expected_qual)
2378 {
2379 	struct ept_access_test_data *data = &ept_access_test_data;
2380 	unsigned long orig_pte;
2381 
2382 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2383 
2384 	do_ept_violation(level == 1 || mkhuge, op, expected_qual,
2385 			 op == OP_EXEC ? data->gpa + sizeof(unsigned long) :
2386 					 data->gpa);
2387 
2388 	/* Fix the violation and resume the op loop. */
2389 	ept_untwiddle(data->gpa, level, orig_pte);
2390 	enter_guest();
2391 	skip_exit_vmcall();
2392 }
2393 
2394 static void
2395 ept_violation_at_level(int level, unsigned long clear, unsigned long set,
2396 		       enum ept_access_op op, u64 expected_qual)
2397 {
2398 	ept_violation_at_level_mkhuge(false, level, clear, set, op,
2399 				      expected_qual);
2400 	if (ept_huge_pages_supported(level))
2401 		ept_violation_at_level_mkhuge(true, level, clear, set, op,
2402 					      expected_qual);
2403 }
2404 
2405 static void ept_violation(unsigned long clear, unsigned long set,
2406 			  enum ept_access_op op, u64 expected_qual)
2407 {
2408 	ept_violation_at_level(1, clear, set, op, expected_qual);
2409 	ept_violation_at_level(2, clear, set, op, expected_qual);
2410 	ept_violation_at_level(3, clear, set, op, expected_qual);
2411 	ept_violation_at_level(4, clear, set, op, expected_qual);
2412 }
2413 
2414 static void ept_access_violation(unsigned long access, enum ept_access_op op,
2415 				       u64 expected_qual)
2416 {
2417 	ept_violation(EPT_PRESENT, access, op,
2418 		      expected_qual | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2419 }
2420 
2421 /*
2422  * For translations that don't involve a GVA, that is physical address (paddr)
2423  * accesses, EPT violations don't set the flag EPT_VLT_PADDR.  For a typical
2424  * guest memory access, the hardware does GVA -> GPA -> HPA.  However, certain
2425  * translations don't involve GVAs, such as when the hardware does the guest
2426  * page table walk. For example, in translating GVA_1 -> GPA_1, the guest MMU
2427  * might try to set an A bit on a guest PTE. If the GPA_2 that the PTE resides
2428  * on isn't present in the EPT, then the EPT violation will be for GPA_2 and
2429  * the EPT_VLT_PADDR bit will be clear in the exit qualification.
2430  *
2431  * Note that paddr violations can also be triggered by loading PAE page tables
2432  * with wonky addresses. We don't test that yet.
2433  *
2434  * This function modifies the EPT entry that maps the GPA that the guest page
2435  * table entry mapping ept_access_data.gva resides on.
2436  *
2437  *	@ept_access	EPT permissions to set. Other permissions are cleared.
2438  *
2439  *	@pte_ad		Set the A/D bits on the guest PTE accordingly.
2440  *
2441  *	@op		Guest operation to perform with ept_access_data.gva.
2442  *
2443  *	@expect_violation
2444  *			Is a violation expected during the paddr access?
2445  *
2446  *	@expected_qual	Expected qualification for the EPT violation.
2447  *			EPT_VLT_PADDR should be clear.
2448  */
2449 static void ept_access_paddr(unsigned long ept_access, unsigned long pte_ad,
2450 			     enum ept_access_op op, bool expect_violation,
2451 			     u64 expected_qual)
2452 {
2453 	struct ept_access_test_data *data = &ept_access_test_data;
2454 	unsigned long *ptep;
2455 	unsigned long gpa;
2456 	unsigned long orig_epte;
2457 
2458 	/* Modify the guest PTE mapping data->gva according to @pte_ad.  */
2459 	ptep = get_pte_level(current_page_table(), data->gva, /*level=*/1);
2460 	TEST_ASSERT(ptep);
2461 	TEST_ASSERT_EQ(*ptep & PT_ADDR_MASK, data->gpa);
2462 	*ptep = (*ptep & ~PT_AD_MASK) | pte_ad;
2463 	ept_access_test_guest_flush_tlb();
2464 
2465 	/*
2466 	 * Now modify the access bits on the EPT entry for the GPA that the
2467 	 * guest PTE resides on. Note that by modifying a single EPT entry,
2468 	 * we're potentially affecting 512 guest PTEs. However, we've carefully
2469 	 * constructed our test such that those other 511 PTEs aren't used by
2470 	 * the guest: data->gva is at the beginning of a 1G huge page, thus the
2471 	 * PTE we're modifying is at the beginning of a 4K page and the
2472 	 * following 511 entires are also under our control (and not touched by
2473 	 * the guest).
2474 	 */
2475 	gpa = virt_to_phys(ptep);
2476 	TEST_ASSERT_EQ(gpa & ~PAGE_MASK, 0);
2477 	/*
2478 	 * Make sure the guest page table page is mapped with a 4K EPT entry,
2479 	 * otherwise our level=1 twiddling below will fail. We use the
2480 	 * identity map (gpa = gpa) since page tables are shared with the host.
2481 	 */
2482 	install_ept(pml4, gpa, gpa, EPT_PRESENT);
2483 	orig_epte = ept_twiddle(gpa, /*mkhuge=*/0, /*level=*/1,
2484 				/*clear=*/EPT_PRESENT, /*set=*/ept_access);
2485 
2486 	if (expect_violation) {
2487 		do_ept_violation(/*leaf=*/true, op,
2488 				 expected_qual | EPT_VLT_LADDR_VLD, gpa);
2489 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2490 		do_ept_access_op(op);
2491 	} else {
2492 		do_ept_access_op(op);
2493 		ept_untwiddle(gpa, /*level=*/1, orig_epte);
2494 	}
2495 
2496 	TEST_ASSERT(*ptep & PT_ACCESSED_MASK);
2497 	if ((pte_ad & PT_DIRTY_MASK) || op == OP_WRITE)
2498 		TEST_ASSERT(*ptep & PT_DIRTY_MASK);
2499 
2500 	skip_exit_vmcall();
2501 }
2502 
2503 static void ept_access_allowed_paddr(unsigned long ept_access,
2504 				     unsigned long pte_ad,
2505 				     enum ept_access_op op)
2506 {
2507 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/false,
2508 			 /*expected_qual=*/-1);
2509 }
2510 
2511 static void ept_access_violation_paddr(unsigned long ept_access,
2512 				       unsigned long pte_ad,
2513 				       enum ept_access_op op,
2514 				       u64 expected_qual)
2515 {
2516 	ept_access_paddr(ept_access, pte_ad, op, /*expect_violation=*/true,
2517 			 expected_qual);
2518 }
2519 
2520 
2521 static void ept_allowed_at_level_mkhuge(bool mkhuge, int level,
2522 					unsigned long clear,
2523 					unsigned long set,
2524 					enum ept_access_op op)
2525 {
2526 	struct ept_access_test_data *data = &ept_access_test_data;
2527 	unsigned long orig_pte;
2528 
2529 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2530 
2531 	/* No violation. Should proceed to vmcall. */
2532 	do_ept_access_op(op);
2533 	skip_exit_vmcall();
2534 
2535 	ept_untwiddle(data->gpa, level, orig_pte);
2536 }
2537 
2538 static void ept_allowed_at_level(int level, unsigned long clear,
2539 				 unsigned long set, enum ept_access_op op)
2540 {
2541 	ept_allowed_at_level_mkhuge(false, level, clear, set, op);
2542 	if (ept_huge_pages_supported(level))
2543 		ept_allowed_at_level_mkhuge(true, level, clear, set, op);
2544 }
2545 
2546 static void ept_allowed(unsigned long clear, unsigned long set,
2547 			enum ept_access_op op)
2548 {
2549 	ept_allowed_at_level(1, clear, set, op);
2550 	ept_allowed_at_level(2, clear, set, op);
2551 	ept_allowed_at_level(3, clear, set, op);
2552 	ept_allowed_at_level(4, clear, set, op);
2553 }
2554 
2555 static void ept_ignored_bit(int bit)
2556 {
2557 	/* Set the bit. */
2558 	ept_allowed(0, 1ul << bit, OP_READ);
2559 	ept_allowed(0, 1ul << bit, OP_WRITE);
2560 	ept_allowed(0, 1ul << bit, OP_EXEC);
2561 
2562 	/* Clear the bit. */
2563 	ept_allowed(1ul << bit, 0, OP_READ);
2564 	ept_allowed(1ul << bit, 0, OP_WRITE);
2565 	ept_allowed(1ul << bit, 0, OP_EXEC);
2566 }
2567 
2568 static void ept_access_allowed(unsigned long access, enum ept_access_op op)
2569 {
2570 	ept_allowed(EPT_PRESENT, access, op);
2571 }
2572 
2573 
2574 static void ept_misconfig_at_level_mkhuge_op(bool mkhuge, int level,
2575 					     unsigned long clear,
2576 					     unsigned long set,
2577 					     enum ept_access_op op)
2578 {
2579 	struct ept_access_test_data *data = &ept_access_test_data;
2580 	unsigned long orig_pte;
2581 
2582 	orig_pte = ept_twiddle(data->gpa, mkhuge, level, clear, set);
2583 
2584 	do_ept_access_op(op);
2585 	assert_exit_reason(VMX_EPT_MISCONFIG);
2586 
2587 	/* Intel 27.2.1, "For all other VM exits, this field is cleared." */
2588 	#if 0
2589 	/* broken: */
2590 	TEST_EXPECT_EQ_MSG(vmcs_read(EXI_QUALIFICATION), 0);
2591 	#endif
2592 	#if 0
2593 	/*
2594 	 * broken:
2595 	 * According to description of exit qual for EPT violation,
2596 	 * EPT_VLT_LADDR_VLD indicates if GUEST_LINEAR_ADDRESS is valid.
2597 	 * However, I can't find anything that says GUEST_LINEAR_ADDRESS ought
2598 	 * to be set for msiconfig.
2599 	 */
2600 	TEST_EXPECT_EQ(vmcs_read(GUEST_LINEAR_ADDRESS),
2601 		       (unsigned long) (
2602 			       op == OP_EXEC ? data->gva + 1 : data->gva));
2603 	#endif
2604 
2605 	/* Fix the violation and resume the op loop. */
2606 	ept_untwiddle(data->gpa, level, orig_pte);
2607 	enter_guest();
2608 	skip_exit_vmcall();
2609 }
2610 
2611 static void ept_misconfig_at_level_mkhuge(bool mkhuge, int level,
2612 					  unsigned long clear,
2613 					  unsigned long set)
2614 {
2615 	/* The op shouldn't matter (read, write, exec), so try them all! */
2616 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_READ);
2617 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_WRITE);
2618 	ept_misconfig_at_level_mkhuge_op(mkhuge, level, clear, set, OP_EXEC);
2619 }
2620 
2621 static void ept_misconfig_at_level(int level, unsigned long clear,
2622 				   unsigned long set)
2623 {
2624 	ept_misconfig_at_level_mkhuge(false, level, clear, set);
2625 	if (ept_huge_pages_supported(level))
2626 		ept_misconfig_at_level_mkhuge(true, level, clear, set);
2627 }
2628 
2629 static void ept_misconfig(unsigned long clear, unsigned long set)
2630 {
2631 	ept_misconfig_at_level(1, clear, set);
2632 	ept_misconfig_at_level(2, clear, set);
2633 	ept_misconfig_at_level(3, clear, set);
2634 	ept_misconfig_at_level(4, clear, set);
2635 }
2636 
2637 static void ept_access_misconfig(unsigned long access)
2638 {
2639 	ept_misconfig(EPT_PRESENT, access);
2640 }
2641 
2642 static void ept_reserved_bit_at_level_nohuge(int level, int bit)
2643 {
2644 	/* Setting the bit causes a misconfig. */
2645 	ept_misconfig_at_level_mkhuge(false, level, 0, 1ul << bit);
2646 
2647 	/* Making the entry non-present turns reserved bits into ignored. */
2648 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2649 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2650 }
2651 
2652 static void ept_reserved_bit_at_level_huge(int level, int bit)
2653 {
2654 	/* Setting the bit causes a misconfig. */
2655 	ept_misconfig_at_level_mkhuge(true, level, 0, 1ul << bit);
2656 
2657 	/* Making the entry non-present turns reserved bits into ignored. */
2658 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2659 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2660 }
2661 
2662 static void ept_reserved_bit_at_level(int level, int bit)
2663 {
2664 	/* Setting the bit causes a misconfig. */
2665 	ept_misconfig_at_level(level, 0, 1ul << bit);
2666 
2667 	/* Making the entry non-present turns reserved bits into ignored. */
2668 	ept_violation_at_level(level, EPT_PRESENT, 1ul << bit, OP_READ,
2669 			       EPT_VLT_RD | EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
2670 }
2671 
2672 static void ept_reserved_bit(int bit)
2673 {
2674 	ept_reserved_bit_at_level(1, bit);
2675 	ept_reserved_bit_at_level(2, bit);
2676 	ept_reserved_bit_at_level(3, bit);
2677 	ept_reserved_bit_at_level(4, bit);
2678 }
2679 
2680 #define PAGE_2M_ORDER 9
2681 #define PAGE_1G_ORDER 18
2682 
2683 static void *get_1g_page(void)
2684 {
2685 	static void *alloc;
2686 
2687 	if (!alloc)
2688 		alloc = alloc_pages(PAGE_1G_ORDER);
2689 	return alloc;
2690 }
2691 
2692 static void ept_access_test_teardown(void *unused)
2693 {
2694 	/* Exit the guest cleanly. */
2695 	do_ept_access_op(OP_EXIT);
2696 }
2697 
2698 static void ept_access_test_guest(void)
2699 {
2700 	struct ept_access_test_data *data = &ept_access_test_data;
2701 	int (*code)(void) = (int (*)(void)) &data->gva[1];
2702 
2703 	while (true) {
2704 		switch (data->op) {
2705 		case OP_READ:
2706 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_1);
2707 			break;
2708 		case OP_WRITE:
2709 			*data->gva = MAGIC_VAL_2;
2710 			TEST_ASSERT_EQ(*data->gva, MAGIC_VAL_2);
2711 			*data->gva = MAGIC_VAL_1;
2712 			break;
2713 		case OP_EXEC:
2714 			TEST_ASSERT_EQ(42, code());
2715 			break;
2716 		case OP_FLUSH_TLB:
2717 			write_cr3(read_cr3());
2718 			break;
2719 		case OP_EXIT:
2720 			return;
2721 		default:
2722 			TEST_ASSERT_MSG(false, "Unknown op %d", data->op);
2723 		}
2724 		vmcall();
2725 	}
2726 }
2727 
2728 static void ept_access_test_setup(void)
2729 {
2730 	struct ept_access_test_data *data = &ept_access_test_data;
2731 	unsigned long npages = 1ul << PAGE_1G_ORDER;
2732 	unsigned long size = npages * PAGE_SIZE;
2733 	unsigned long *page_table = current_page_table();
2734 	unsigned long pte;
2735 
2736 	if (setup_ept(false))
2737 		test_skip("EPT not supported");
2738 
2739 	/* We use data->gpa = 1 << 39 so that test data has a separate pml4 entry */
2740 	if (cpuid_maxphyaddr() < 40)
2741 		test_skip("Test needs MAXPHYADDR >= 40");
2742 
2743 	test_set_guest(ept_access_test_guest);
2744 	test_add_teardown(ept_access_test_teardown, NULL);
2745 
2746 	data->hva = get_1g_page();
2747 	TEST_ASSERT(data->hva);
2748 	data->hpa = virt_to_phys(data->hva);
2749 
2750 	data->gpa = 1ul << 39;
2751 	data->gva = (void *) ALIGN((unsigned long) alloc_vpages(npages * 2),
2752 				   size);
2753 	TEST_ASSERT(!any_present_pages(page_table, data->gva, size));
2754 	install_pages(page_table, data->gpa, size, data->gva);
2755 
2756 	/*
2757 	 * Make sure nothing's mapped here so the tests that screw with the
2758 	 * pml4 entry don't inadvertently break something.
2759 	 */
2760 	TEST_ASSERT(get_ept_pte(pml4, data->gpa, 4, &pte) && pte == 0);
2761 	TEST_ASSERT(get_ept_pte(pml4, data->gpa + size - 1, 4, &pte) && pte == 0);
2762 	install_ept(pml4, data->hpa, data->gpa, EPT_PRESENT);
2763 
2764 	data->hva[0] = MAGIC_VAL_1;
2765 	memcpy(&data->hva[1], &ret42_start, &ret42_end - &ret42_start);
2766 }
2767 
2768 static void ept_access_test_not_present(void)
2769 {
2770 	ept_access_test_setup();
2771 	/* --- */
2772 	ept_access_violation(0, OP_READ, EPT_VLT_RD);
2773 	ept_access_violation(0, OP_WRITE, EPT_VLT_WR);
2774 	ept_access_violation(0, OP_EXEC, EPT_VLT_FETCH);
2775 }
2776 
2777 static void ept_access_test_read_only(void)
2778 {
2779 	ept_access_test_setup();
2780 
2781 	/* r-- */
2782 	ept_access_allowed(EPT_RA, OP_READ);
2783 	ept_access_violation(EPT_RA, OP_WRITE, EPT_VLT_WR | EPT_VLT_PERM_RD);
2784 	ept_access_violation(EPT_RA, OP_EXEC, EPT_VLT_FETCH | EPT_VLT_PERM_RD);
2785 }
2786 
2787 static void ept_access_test_write_only(void)
2788 {
2789 	ept_access_test_setup();
2790 	/* -w- */
2791 	ept_access_misconfig(EPT_WA);
2792 }
2793 
2794 static void ept_access_test_read_write(void)
2795 {
2796 	ept_access_test_setup();
2797 	/* rw- */
2798 	ept_access_allowed(EPT_RA | EPT_WA, OP_READ);
2799 	ept_access_allowed(EPT_RA | EPT_WA, OP_WRITE);
2800 	ept_access_violation(EPT_RA | EPT_WA, OP_EXEC,
2801 			   EPT_VLT_FETCH | EPT_VLT_PERM_RD | EPT_VLT_PERM_WR);
2802 }
2803 
2804 
2805 static void ept_access_test_execute_only(void)
2806 {
2807 	ept_access_test_setup();
2808 	/* --x */
2809 	if (ept_execute_only_supported()) {
2810 		ept_access_violation(EPT_EA, OP_READ,
2811 				     EPT_VLT_RD | EPT_VLT_PERM_EX);
2812 		ept_access_violation(EPT_EA, OP_WRITE,
2813 				     EPT_VLT_WR | EPT_VLT_PERM_EX);
2814 		ept_access_allowed(EPT_EA, OP_EXEC);
2815 	} else {
2816 		ept_access_misconfig(EPT_EA);
2817 	}
2818 }
2819 
2820 static void ept_access_test_read_execute(void)
2821 {
2822 	ept_access_test_setup();
2823 	/* r-x */
2824 	ept_access_allowed(EPT_RA | EPT_EA, OP_READ);
2825 	ept_access_violation(EPT_RA | EPT_EA, OP_WRITE,
2826 			   EPT_VLT_WR | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX);
2827 	ept_access_allowed(EPT_RA | EPT_EA, OP_EXEC);
2828 }
2829 
2830 static void ept_access_test_write_execute(void)
2831 {
2832 	ept_access_test_setup();
2833 	/* -wx */
2834 	ept_access_misconfig(EPT_WA | EPT_EA);
2835 }
2836 
2837 static void ept_access_test_read_write_execute(void)
2838 {
2839 	ept_access_test_setup();
2840 	/* rwx */
2841 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_READ);
2842 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_WRITE);
2843 	ept_access_allowed(EPT_RA | EPT_WA | EPT_EA, OP_EXEC);
2844 }
2845 
2846 static void ept_access_test_reserved_bits(void)
2847 {
2848 	int i;
2849 	int maxphyaddr;
2850 
2851 	ept_access_test_setup();
2852 
2853 	/* Reserved bits above maxphyaddr. */
2854 	maxphyaddr = cpuid_maxphyaddr();
2855 	for (i = maxphyaddr; i <= 51; i++) {
2856 		report_prefix_pushf("reserved_bit=%d", i);
2857 		ept_reserved_bit(i);
2858 		report_prefix_pop();
2859 	}
2860 
2861 	/* Level-specific reserved bits. */
2862 	ept_reserved_bit_at_level_nohuge(2, 3);
2863 	ept_reserved_bit_at_level_nohuge(2, 4);
2864 	ept_reserved_bit_at_level_nohuge(2, 5);
2865 	ept_reserved_bit_at_level_nohuge(2, 6);
2866 	/* 2M alignment. */
2867 	for (i = 12; i < 20; i++) {
2868 		report_prefix_pushf("reserved_bit=%d", i);
2869 		ept_reserved_bit_at_level_huge(2, i);
2870 		report_prefix_pop();
2871 	}
2872 	ept_reserved_bit_at_level_nohuge(3, 3);
2873 	ept_reserved_bit_at_level_nohuge(3, 4);
2874 	ept_reserved_bit_at_level_nohuge(3, 5);
2875 	ept_reserved_bit_at_level_nohuge(3, 6);
2876 	/* 1G alignment. */
2877 	for (i = 12; i < 29; i++) {
2878 		report_prefix_pushf("reserved_bit=%d", i);
2879 		ept_reserved_bit_at_level_huge(3, i);
2880 		report_prefix_pop();
2881 	}
2882 	ept_reserved_bit_at_level(4, 3);
2883 	ept_reserved_bit_at_level(4, 4);
2884 	ept_reserved_bit_at_level(4, 5);
2885 	ept_reserved_bit_at_level(4, 6);
2886 	ept_reserved_bit_at_level(4, 7);
2887 }
2888 
2889 static void ept_access_test_ignored_bits(void)
2890 {
2891 	ept_access_test_setup();
2892 	/*
2893 	 * Bits ignored at every level. Bits 8 and 9 (A and D) are ignored as
2894 	 * far as translation is concerned even if AD bits are enabled in the
2895 	 * EPTP. Bit 63 is ignored because "EPT-violation #VE" VM-execution
2896 	 * control is 0.
2897 	 */
2898 	ept_ignored_bit(8);
2899 	ept_ignored_bit(9);
2900 	ept_ignored_bit(10);
2901 	ept_ignored_bit(11);
2902 	ept_ignored_bit(52);
2903 	ept_ignored_bit(53);
2904 	ept_ignored_bit(54);
2905 	ept_ignored_bit(55);
2906 	ept_ignored_bit(56);
2907 	ept_ignored_bit(57);
2908 	ept_ignored_bit(58);
2909 	ept_ignored_bit(59);
2910 	ept_ignored_bit(60);
2911 	ept_ignored_bit(61);
2912 	ept_ignored_bit(62);
2913 	ept_ignored_bit(63);
2914 }
2915 
2916 static void ept_access_test_paddr_not_present_ad_disabled(void)
2917 {
2918 	ept_access_test_setup();
2919 	ept_disable_ad_bits();
2920 
2921 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, EPT_VLT_RD);
2922 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, EPT_VLT_RD);
2923 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, EPT_VLT_RD);
2924 }
2925 
2926 static void ept_access_test_paddr_not_present_ad_enabled(void)
2927 {
2928 	u64 qual = EPT_VLT_RD | EPT_VLT_WR;
2929 
2930 	ept_access_test_setup();
2931 	ept_enable_ad_bits_or_skip_test();
2932 
2933 	ept_access_violation_paddr(0, PT_AD_MASK, OP_READ, qual);
2934 	ept_access_violation_paddr(0, PT_AD_MASK, OP_WRITE, qual);
2935 	ept_access_violation_paddr(0, PT_AD_MASK, OP_EXEC, qual);
2936 }
2937 
2938 static void ept_access_test_paddr_read_only_ad_disabled(void)
2939 {
2940 	/*
2941 	 * When EPT AD bits are disabled, all accesses to guest paging
2942 	 * structures are reported separately as a read and (after
2943 	 * translation of the GPA to host physical address) a read+write
2944 	 * if the A/D bits have to be set.
2945 	 */
2946 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2947 
2948 	ept_access_test_setup();
2949 	ept_disable_ad_bits();
2950 
2951 	/* Can't update A bit, so all accesses fail. */
2952 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2953 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
2954 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
2955 	/* AD bits disabled, so only writes try to update the D bit. */
2956 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ);
2957 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
2958 	ept_access_allowed_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC);
2959 	/* Both A and D already set, so read-only is OK. */
2960 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_READ);
2961 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_WRITE);
2962 	ept_access_allowed_paddr(EPT_RA, PT_AD_MASK, OP_EXEC);
2963 }
2964 
2965 static void ept_access_test_paddr_read_only_ad_enabled(void)
2966 {
2967 	/*
2968 	 * When EPT AD bits are enabled, all accesses to guest paging
2969 	 * structures are considered writes as far as EPT translation
2970 	 * is concerned.
2971 	 */
2972 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD;
2973 
2974 	ept_access_test_setup();
2975 	ept_enable_ad_bits_or_skip_test();
2976 
2977 	ept_access_violation_paddr(EPT_RA, 0, OP_READ, qual);
2978 	ept_access_violation_paddr(EPT_RA, 0, OP_WRITE, qual);
2979 	ept_access_violation_paddr(EPT_RA, 0, OP_EXEC, qual);
2980 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_READ, qual);
2981 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_WRITE, qual);
2982 	ept_access_violation_paddr(EPT_RA, PT_ACCESSED_MASK, OP_EXEC, qual);
2983 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_READ, qual);
2984 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_WRITE, qual);
2985 	ept_access_violation_paddr(EPT_RA, PT_AD_MASK, OP_EXEC, qual);
2986 }
2987 
2988 static void ept_access_test_paddr_read_write(void)
2989 {
2990 	ept_access_test_setup();
2991 	/* Read-write access to paging structure. */
2992 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_READ);
2993 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_WRITE);
2994 	ept_access_allowed_paddr(EPT_RA | EPT_WA, 0, OP_EXEC);
2995 }
2996 
2997 static void ept_access_test_paddr_read_write_execute(void)
2998 {
2999 	ept_access_test_setup();
3000 	/* RWX access to paging structure. */
3001 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_READ);
3002 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_WRITE);
3003 	ept_access_allowed_paddr(EPT_PRESENT, 0, OP_EXEC);
3004 }
3005 
3006 static void ept_access_test_paddr_read_execute_ad_disabled(void)
3007 {
3008   	/*
3009 	 * When EPT AD bits are disabled, all accesses to guest paging
3010 	 * structures are reported separately as a read and (after
3011 	 * translation of the GPA to host physical address) a read+write
3012 	 * if the A/D bits have to be set.
3013 	 */
3014 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3015 
3016 	ept_access_test_setup();
3017 	ept_disable_ad_bits();
3018 
3019 	/* Can't update A bit, so all accesses fail. */
3020 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3021 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3022 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3023 	/* AD bits disabled, so only writes try to update the D bit. */
3024 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ);
3025 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3026 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC);
3027 	/* Both A and D already set, so read-only is OK. */
3028 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ);
3029 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE);
3030 	ept_access_allowed_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC);
3031 }
3032 
3033 static void ept_access_test_paddr_read_execute_ad_enabled(void)
3034 {
3035 	/*
3036 	 * When EPT AD bits are enabled, all accesses to guest paging
3037 	 * structures are considered writes as far as EPT translation
3038 	 * is concerned.
3039 	 */
3040 	u64 qual = EPT_VLT_WR | EPT_VLT_RD | EPT_VLT_PERM_RD | EPT_VLT_PERM_EX;
3041 
3042 	ept_access_test_setup();
3043 	ept_enable_ad_bits_or_skip_test();
3044 
3045 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_READ, qual);
3046 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_WRITE, qual);
3047 	ept_access_violation_paddr(EPT_RA | EPT_EA, 0, OP_EXEC, qual);
3048 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_READ, qual);
3049 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_WRITE, qual);
3050 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_ACCESSED_MASK, OP_EXEC, qual);
3051 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_READ, qual);
3052 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_WRITE, qual);
3053 	ept_access_violation_paddr(EPT_RA | EPT_EA, PT_AD_MASK, OP_EXEC, qual);
3054 }
3055 
3056 static void ept_access_test_paddr_not_present_page_fault(void)
3057 {
3058 	ept_access_test_setup();
3059 	/*
3060 	 * TODO: test no EPT violation as long as guest PF occurs. e.g., GPA is
3061 	 * page is read-only in EPT but GVA is also mapped read only in PT.
3062 	 * Thus guest page fault before host takes EPT violation for trying to
3063 	 * update A bit.
3064 	 */
3065 }
3066 
3067 static void ept_access_test_force_2m_page(void)
3068 {
3069 	ept_access_test_setup();
3070 
3071 	TEST_ASSERT_EQ(ept_2m_supported(), true);
3072 	ept_allowed_at_level_mkhuge(true, 2, 0, 0, OP_READ);
3073 	ept_violation_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_RA, OP_WRITE,
3074 				      EPT_VLT_WR | EPT_VLT_PERM_RD |
3075 				      EPT_VLT_LADDR_VLD | EPT_VLT_PADDR);
3076 	ept_misconfig_at_level_mkhuge(true, 2, EPT_PRESENT, EPT_WA);
3077 }
3078 
3079 static bool invvpid_valid(u64 type, u64 vpid, u64 gla)
3080 {
3081 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3082 
3083 	TEST_ASSERT(msr & VPID_CAP_INVVPID);
3084 
3085 	if (type < INVVPID_ADDR || type > INVVPID_CONTEXT_LOCAL)
3086 		return false;
3087 
3088 	if (!(msr & (1ull << (type + VPID_CAP_INVVPID_TYPES_SHIFT))))
3089 		return false;
3090 
3091 	if (vpid >> 16)
3092 		return false;
3093 
3094 	if (type != INVVPID_ALL && !vpid)
3095 		return false;
3096 
3097 	if (type == INVVPID_ADDR && !is_canonical(gla))
3098 		return false;
3099 
3100 	return true;
3101 }
3102 
3103 static void try_invvpid(u64 type, u64 vpid, u64 gla)
3104 {
3105 	int rc;
3106 	bool valid = invvpid_valid(type, vpid, gla);
3107 	u64 expected = valid ? VMXERR_UNSUPPORTED_VMCS_COMPONENT
3108 		: VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID;
3109 	/*
3110 	 * Set VMX_INST_ERROR to VMXERR_UNVALID_VMCS_COMPONENT, so
3111 	 * that we can tell if it is updated by INVVPID.
3112 	 */
3113 	vmcs_read(~0);
3114 	rc = invvpid(type, vpid, gla);
3115 	report("INVVPID type %ld VPID %lx GLA %lx %s",
3116 	       !rc == valid, type, vpid, gla,
3117 	       valid ? "passes" : "fails");
3118 	report("After %s INVVPID, VMX_INST_ERR is %ld (actual %ld)",
3119 	       vmcs_read(VMX_INST_ERROR) == expected,
3120 	       rc ? "failed" : "successful",
3121 	       expected, vmcs_read(VMX_INST_ERROR));
3122 }
3123 
3124 static void ds_invvpid(void *data)
3125 {
3126 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3127 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3128 
3129 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3130 	asm volatile("invvpid %0, %1"
3131 		     :
3132 		     : "m"(*(struct invvpid_operand *)data),
3133 		       "r"(type));
3134 }
3135 
3136 /*
3137  * The SS override is ignored in 64-bit mode, so we use an addressing
3138  * mode with %rsp as the base register to generate an implicit SS
3139  * reference.
3140  */
3141 static void ss_invvpid(void *data)
3142 {
3143 	u64 msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3144 	u64 type = ffs(msr >> VPID_CAP_INVVPID_TYPES_SHIFT) - 1;
3145 
3146 	TEST_ASSERT(type >= INVVPID_ADDR && type <= INVVPID_CONTEXT_LOCAL);
3147 	asm volatile("sub %%rsp,%0; invvpid (%%rsp,%0,1), %1"
3148 		     : "+r"(data)
3149 		     : "r"(type));
3150 }
3151 
3152 static void invvpid_test_gp(void)
3153 {
3154 	bool fault;
3155 
3156 	fault = test_for_exception(GP_VECTOR, &ds_invvpid,
3157 				   (void *)NONCANONICAL);
3158 	report("INVVPID with non-canonical DS operand raises #GP", fault);
3159 }
3160 
3161 static void invvpid_test_ss(void)
3162 {
3163 	bool fault;
3164 
3165 	fault = test_for_exception(SS_VECTOR, &ss_invvpid,
3166 				   (void *)NONCANONICAL);
3167 	report("INVVPID with non-canonical SS operand raises #SS", fault);
3168 }
3169 
3170 static void invvpid_test_pf(void)
3171 {
3172 	void *vpage = alloc_vpage();
3173 	bool fault;
3174 
3175 	fault = test_for_exception(PF_VECTOR, &ds_invvpid, vpage);
3176 	report("INVVPID with unmapped operand raises #PF", fault);
3177 }
3178 
3179 static void try_compat_invvpid(void *unused)
3180 {
3181 	struct far_pointer32 fp = {
3182 		.offset = (uintptr_t)&&invvpid,
3183 		.selector = KERNEL_CS32,
3184 	};
3185 	register uintptr_t rsp asm("rsp");
3186 
3187 	TEST_ASSERT_MSG(fp.offset == (uintptr_t)&&invvpid,
3188 			"Code address too high.");
3189 	TEST_ASSERT_MSG(rsp == (u32)rsp, "Stack address too high.");
3190 
3191 	asm goto ("lcall *%0" : : "m" (fp) : "rax" : invvpid);
3192 	return;
3193 invvpid:
3194 	asm volatile (".code32;"
3195 		      "invvpid (%eax), %eax;"
3196 		      "lret;"
3197 		      ".code64");
3198 	__builtin_unreachable();
3199 }
3200 
3201 static void invvpid_test_compatibility_mode(void)
3202 {
3203 	bool fault;
3204 
3205 	fault = test_for_exception(UD_VECTOR, &try_compat_invvpid, NULL);
3206 	report("Compatibility mode INVVPID raises #UD", fault);
3207 }
3208 
3209 static void invvpid_test_not_in_vmx_operation(void)
3210 {
3211 	bool fault;
3212 
3213 	TEST_ASSERT(!vmx_off());
3214 	fault = test_for_exception(UD_VECTOR, &ds_invvpid, NULL);
3215 	report("INVVPID outside of VMX operation raises #UD", fault);
3216 	TEST_ASSERT(!vmx_on());
3217 }
3218 
3219 /*
3220  * This does not test real-address mode, virtual-8086 mode, protected mode,
3221  * or CPL > 0.
3222  */
3223 static void invvpid_test_v2(void)
3224 {
3225 	u64 msr;
3226 	int i;
3227 	unsigned types = 0;
3228 	unsigned type;
3229 
3230 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
3231 	    !(ctrl_cpu_rev[1].clr & CPU_VPID))
3232 		test_skip("VPID not supported");
3233 
3234 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
3235 
3236 	if (!(msr & VPID_CAP_INVVPID))
3237 		test_skip("INVVPID not supported.\n");
3238 
3239 	if (msr & VPID_CAP_INVVPID_ADDR)
3240 		types |= 1u << INVVPID_ADDR;
3241 	if (msr & VPID_CAP_INVVPID_CXTGLB)
3242 		types |= 1u << INVVPID_CONTEXT_GLOBAL;
3243 	if (msr & VPID_CAP_INVVPID_ALL)
3244 		types |= 1u << INVVPID_ALL;
3245 	if (msr & VPID_CAP_INVVPID_CXTLOC)
3246 		types |= 1u << INVVPID_CONTEXT_LOCAL;
3247 
3248 	if (!types)
3249 		test_skip("No INVVPID types supported.\n");
3250 
3251 	for (i = -127; i < 128; i++)
3252 		try_invvpid(i, 0xffff, 0);
3253 
3254 	/*
3255 	 * VPID must not be more than 16 bits.
3256 	 */
3257 	for (i = 0; i < 64; i++)
3258 		for (type = 0; type < 4; type++)
3259 			if (types & (1u << type))
3260 				try_invvpid(type, 1ul << i, 0);
3261 
3262 	/*
3263 	 * VPID must not be zero, except for "all contexts."
3264 	 */
3265 	for (type = 0; type < 4; type++)
3266 		if (types & (1u << type))
3267 			try_invvpid(type, 0, 0);
3268 
3269 	/*
3270 	 * The gla operand is only validated for single-address INVVPID.
3271 	 */
3272 	if (types & (1u << INVVPID_ADDR))
3273 		try_invvpid(INVVPID_ADDR, 0xffff, NONCANONICAL);
3274 
3275 	invvpid_test_gp();
3276 	invvpid_test_ss();
3277 	invvpid_test_pf();
3278 	invvpid_test_compatibility_mode();
3279 	invvpid_test_not_in_vmx_operation();
3280 }
3281 
3282 /*
3283  * Test for early VMLAUNCH failure. Returns true if VMLAUNCH makes it
3284  * at least as far as the guest-state checks. Returns false if the
3285  * VMLAUNCH fails early and execution falls through to the next
3286  * instruction.
3287  */
3288 static bool vmlaunch_succeeds(void)
3289 {
3290 	u32 exit_reason;
3291 
3292 	/*
3293 	 * Indirectly set VMX_INST_ERR to 12 ("VMREAD/VMWRITE from/to
3294 	 * unsupported VMCS component"). The caller can then check
3295 	 * to see if a failed VM-entry sets VMX_INST_ERR as expected.
3296 	 */
3297 	vmcs_write(~0u, 0);
3298 
3299 	vmcs_write(HOST_RIP, (uintptr_t)&&success);
3300 	__asm__ __volatile__ goto ("vmwrite %%rsp, %0; vmlaunch"
3301 				   :
3302 				   : "r" ((u64)HOST_RSP)
3303 				   : "cc", "memory"
3304 				   : success);
3305 	return false;
3306 success:
3307 	exit_reason = vmcs_read(EXI_REASON);
3308 	TEST_ASSERT(exit_reason == (VMX_FAIL_STATE | VMX_ENTRY_FAILURE) ||
3309 		    exit_reason == (VMX_FAIL_MSR | VMX_ENTRY_FAILURE));
3310 	return true;
3311 }
3312 
3313 /*
3314  * Try to launch the current VMCS.
3315  */
3316 static void test_vmx_controls(bool controls_valid, bool xfail)
3317 {
3318 	bool success = vmlaunch_succeeds();
3319 	u32 vmx_inst_err;
3320 
3321 	report_xfail("vmlaunch %s", xfail, success == controls_valid,
3322 		     controls_valid ? "succeeds" : "fails");
3323 	if (!success) {
3324 		vmx_inst_err = vmcs_read(VMX_INST_ERROR);
3325 		report("VMX inst error is %d (actual %d)",
3326 		       vmx_inst_err == VMXERR_ENTRY_INVALID_CONTROL_FIELD,
3327 		       VMXERR_ENTRY_INVALID_CONTROL_FIELD, vmx_inst_err);
3328 	}
3329 }
3330 
3331 /*
3332  * Test a particular value of a VM-execution control bit, if the value
3333  * is required or if the value is zero.
3334  */
3335 static void test_rsvd_ctl_bit_value(const char *name, union vmx_ctrl_msr msr,
3336 				    enum Encoding encoding, unsigned bit,
3337 				    unsigned val)
3338 {
3339 	u32 mask = 1u << bit;
3340 	bool expected;
3341 	u32 controls;
3342 
3343 	if (msr.set & mask)
3344 		TEST_ASSERT(msr.clr & mask);
3345 
3346 	/*
3347 	 * We can't arbitrarily turn on a control bit, because it may
3348 	 * introduce dependencies on other VMCS fields. So, we only
3349 	 * test turning on bits that have a required setting.
3350 	 */
3351 	if (val && (msr.clr & mask) && !(msr.set & mask))
3352 		return;
3353 
3354 	report_prefix_pushf("%s %s bit %d",
3355 			    val ? "Set" : "Clear", name, bit);
3356 
3357 	controls = vmcs_read(encoding);
3358 	if (val) {
3359 		vmcs_write(encoding, msr.set | mask);
3360 		expected = (msr.clr & mask);
3361 	} else {
3362 		vmcs_write(encoding, msr.set & ~mask);
3363 		expected = !(msr.set & mask);
3364 	}
3365 	test_vmx_controls(expected, false);
3366 	vmcs_write(encoding, controls);
3367 	report_prefix_pop();
3368 }
3369 
3370 /*
3371  * Test reserved values of a VM-execution control bit, based on the
3372  * allowed bit settings from the corresponding VMX capability MSR.
3373  */
3374 static void test_rsvd_ctl_bit(const char *name, union vmx_ctrl_msr msr,
3375 			      enum Encoding encoding, unsigned bit)
3376 {
3377 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 0);
3378 	test_rsvd_ctl_bit_value(name, msr, encoding, bit, 1);
3379 }
3380 
3381 /*
3382  * Reserved bits in the pin-based VM-execution controls must be set
3383  * properly. Software may consult the VMX capability MSRs to determine
3384  * the proper settings.
3385  * [Intel SDM]
3386  */
3387 static void test_pin_based_ctls(void)
3388 {
3389 	unsigned bit;
3390 
3391 	printf("%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PIN" :
3392 	       "MSR_IA32_VMX_PINBASED_CTLS", ctrl_pin_rev.val);
3393 	for (bit = 0; bit < 32; bit++)
3394 		test_rsvd_ctl_bit("pin-based controls",
3395 				  ctrl_pin_rev, PIN_CONTROLS, bit);
3396 }
3397 
3398 /*
3399  * Reserved bits in the primary processor-based VM-execution controls
3400  * must be set properly. Software may consult the VMX capability MSRs
3401  * to determine the proper settings.
3402  * [Intel SDM]
3403  */
3404 static void test_primary_processor_based_ctls(void)
3405 {
3406 	unsigned bit;
3407 
3408 	printf("\n%s: %lx\n", basic.ctrl ? "MSR_IA32_VMX_TRUE_PROC" :
3409 	       "MSR_IA32_VMX_PROCBASED_CTLS", ctrl_cpu_rev[0].val);
3410 	for (bit = 0; bit < 32; bit++)
3411 		test_rsvd_ctl_bit("primary processor-based controls",
3412 				  ctrl_cpu_rev[0], CPU_EXEC_CTRL0, bit);
3413 }
3414 
3415 /*
3416  * If the "activate secondary controls" primary processor-based
3417  * VM-execution control is 1, reserved bits in the secondary
3418  * processor-based VM-execution controls must be cleared. Software may
3419  * consult the VMX capability MSRs to determine which bits are
3420  * reserved.
3421  * If the "activate secondary controls" primary processor-based
3422  * VM-execution control is 0 (or if the processor does not support the
3423  * 1-setting of that control), no checks are performed on the
3424  * secondary processor-based VM-execution controls.
3425  * [Intel SDM]
3426  */
3427 static void test_secondary_processor_based_ctls(void)
3428 {
3429 	u32 primary;
3430 	u32 secondary;
3431 	unsigned bit;
3432 
3433 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY))
3434 		return;
3435 
3436 	primary = vmcs_read(CPU_EXEC_CTRL0);
3437 	secondary = vmcs_read(CPU_EXEC_CTRL1);
3438 
3439 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3440 	printf("\nMSR_IA32_VMX_PROCBASED_CTLS2: %lx\n", ctrl_cpu_rev[1].val);
3441 	for (bit = 0; bit < 32; bit++)
3442 		test_rsvd_ctl_bit("secondary processor-based controls",
3443 				  ctrl_cpu_rev[1], CPU_EXEC_CTRL1, bit);
3444 
3445 	/*
3446 	 * When the "activate secondary controls" VM-execution control
3447 	 * is clear, there are no checks on the secondary controls.
3448 	 */
3449 	vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3450 	vmcs_write(CPU_EXEC_CTRL1, ~0);
3451 	report("Secondary processor-based controls ignored",
3452 	       vmlaunch_succeeds());
3453 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3454 	vmcs_write(CPU_EXEC_CTRL0, primary);
3455 }
3456 
3457 static void try_cr3_target_count(unsigned i, unsigned max)
3458 {
3459 	report_prefix_pushf("CR3 target count 0x%x", i);
3460 	vmcs_write(CR3_TARGET_COUNT, i);
3461 	test_vmx_controls(i <= max, false);
3462 	report_prefix_pop();
3463 }
3464 
3465 /*
3466  * The CR3-target count must not be greater than 4. Future processors
3467  * may support a different number of CR3-target values. Software
3468  * should read the VMX capability MSR IA32_VMX_MISC to determine the
3469  * number of values supported.
3470  * [Intel SDM]
3471  */
3472 static void test_cr3_targets(void)
3473 {
3474 	unsigned supported_targets = (rdmsr(MSR_IA32_VMX_MISC) >> 16) & 0x1ff;
3475 	u32 cr3_targets = vmcs_read(CR3_TARGET_COUNT);
3476 	unsigned i;
3477 
3478 	printf("\nSupported CR3 targets: %d\n", supported_targets);
3479 	TEST_ASSERT(supported_targets <= 256);
3480 
3481 	try_cr3_target_count(-1u, supported_targets);
3482 	try_cr3_target_count(0x80000000, supported_targets);
3483 	try_cr3_target_count(0x7fffffff, supported_targets);
3484 	for (i = 0; i <= supported_targets + 1; i++)
3485 		try_cr3_target_count(i, supported_targets);
3486 	vmcs_write(CR3_TARGET_COUNT, cr3_targets);
3487 }
3488 
3489 /*
3490  * Test a particular address setting in the VMCS
3491  */
3492 static void test_vmcs_addr(const char *name,
3493 			   enum Encoding encoding,
3494 			   u64 align,
3495 			   bool ignored,
3496 			   bool xfail_beyond_mapped_ram,
3497 			   u64 addr)
3498 {
3499 	bool xfail =
3500 		(xfail_beyond_mapped_ram &&
3501 		 addr > fwcfg_get_u64(FW_CFG_RAM_SIZE) - align &&
3502 		 addr < (1ul << cpuid_maxphyaddr()));
3503 
3504 	report_prefix_pushf("%s = %lx", name, addr);
3505 	vmcs_write(encoding, addr);
3506 	test_vmx_controls(ignored || (IS_ALIGNED(addr, align) &&
3507 				  addr < (1ul << cpuid_maxphyaddr())),
3508 			  xfail);
3509 	report_prefix_pop();
3510 	xfail = false;
3511 }
3512 
3513 /*
3514  * Test interesting values for a VMCS address
3515  */
3516 static void test_vmcs_addr_values(const char *name,
3517 				  enum Encoding encoding,
3518 				  u64 align,
3519 				  bool ignored,
3520 				  bool xfail_beyond_mapped_ram,
3521 				  u32 bit_start, u32 bit_end)
3522 {
3523 	unsigned i;
3524 	u64 orig_val = vmcs_read(encoding);
3525 
3526 	for (i = bit_start; i <= bit_end; i++)
3527 		test_vmcs_addr(name, encoding, align, ignored,
3528 			       xfail_beyond_mapped_ram, 1ul << i);
3529 
3530 	test_vmcs_addr(name, encoding, align, ignored,
3531 		       xfail_beyond_mapped_ram, PAGE_SIZE - 1);
3532 	test_vmcs_addr(name, encoding, align, ignored,
3533 		       xfail_beyond_mapped_ram, PAGE_SIZE);
3534 	test_vmcs_addr(name, encoding, align, ignored,
3535 		       xfail_beyond_mapped_ram,
3536 		      (1ul << cpuid_maxphyaddr()) - PAGE_SIZE);
3537 	test_vmcs_addr(name, encoding, align, ignored,
3538 		       xfail_beyond_mapped_ram, -1ul);
3539 
3540 	vmcs_write(encoding, orig_val);
3541 }
3542 
3543 /*
3544  * Test a physical address reference in the VMCS, when the corresponding
3545  * feature is enabled and when the corresponding feature is disabled.
3546  */
3547 static void test_vmcs_addr_reference(u32 control_bit, enum Encoding field,
3548 				     const char *field_name,
3549 				     const char *control_name, u64 align,
3550 				     bool xfail_beyond_mapped_ram,
3551 				     bool control_primary)
3552 {
3553 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
3554 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
3555 	u64 page_addr;
3556 
3557 	if (control_primary) {
3558 		if (!(ctrl_cpu_rev[0].clr & control_bit))
3559 			return;
3560 	} else {
3561 		if (!(ctrl_cpu_rev[1].clr & control_bit))
3562 			return;
3563 	}
3564 
3565 	page_addr = vmcs_read(field);
3566 
3567 	report_prefix_pushf("%s enabled", control_name);
3568 	if (control_primary) {
3569 		vmcs_write(CPU_EXEC_CTRL0, primary | control_bit);
3570 	} else {
3571 		vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3572 		vmcs_write(CPU_EXEC_CTRL1, secondary | control_bit);
3573 	}
3574 
3575 	test_vmcs_addr_values(field_name, field, align, false,
3576 			      xfail_beyond_mapped_ram, 0, 63);
3577 	report_prefix_pop();
3578 
3579 	report_prefix_pushf("%s disabled", control_name);
3580 	if (control_primary) {
3581 		vmcs_write(CPU_EXEC_CTRL0, primary & ~control_bit);
3582 	} else {
3583 		vmcs_write(CPU_EXEC_CTRL0, primary & ~CPU_SECONDARY);
3584 		vmcs_write(CPU_EXEC_CTRL1, secondary & ~control_bit);
3585 	}
3586 
3587 	test_vmcs_addr_values(field_name, field, align, true, false, 0, 63);
3588 	report_prefix_pop();
3589 
3590 	vmcs_write(field, page_addr);
3591 	vmcs_write(CPU_EXEC_CTRL0, primary);
3592 }
3593 
3594 /*
3595  * If the "use I/O bitmaps" VM-execution control is 1, bits 11:0 of
3596  * each I/O-bitmap address must be 0. Neither address should set any
3597  * bits beyond the processor's physical-address width.
3598  * [Intel SDM]
3599  */
3600 static void test_io_bitmaps(void)
3601 {
3602 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_A,
3603 				 "I/O bitmap A", "Use I/O bitmaps",
3604 				 PAGE_SIZE, false, true);
3605 	test_vmcs_addr_reference(CPU_IO_BITMAP, IO_BITMAP_B,
3606 				 "I/O bitmap B", "Use I/O bitmaps",
3607 				 PAGE_SIZE, false, true);
3608 }
3609 
3610 /*
3611  * If the "use MSR bitmaps" VM-execution control is 1, bits 11:0 of
3612  * the MSR-bitmap address must be 0. The address should not set any
3613  * bits beyond the processor's physical-address width.
3614  * [Intel SDM]
3615  */
3616 static void test_msr_bitmap(void)
3617 {
3618 	test_vmcs_addr_reference(CPU_MSR_BITMAP, MSR_BITMAP,
3619 				 "MSR bitmap", "Use MSR bitmaps",
3620 				 PAGE_SIZE, false, true);
3621 }
3622 
3623 /*
3624  * If the "use TPR shadow" VM-execution control is 1, the virtual-APIC
3625  * address must satisfy the following checks:
3626  * - Bits 11:0 of the address must be 0.
3627  * - The address should not set any bits beyond the processor's
3628  *   physical-address width.
3629  * [Intel SDM]
3630  */
3631 static void test_apic_virt_addr(void)
3632 {
3633 	test_vmcs_addr_reference(CPU_TPR_SHADOW, APIC_VIRT_ADDR,
3634 				 "virtual-APIC address", "Use TPR shadow",
3635 				 PAGE_SIZE, true, true);
3636 }
3637 
3638 /*
3639  * If the "virtualize APIC-accesses" VM-execution control is 1, the
3640  * APIC-access address must satisfy the following checks:
3641  *  - Bits 11:0 of the address must be 0.
3642  *  - The address should not set any bits beyond the processor's
3643  *    physical-address width.
3644  * [Intel SDM]
3645  */
3646 static void test_apic_access_addr(void)
3647 {
3648 	void *apic_access_page = alloc_page();
3649 
3650 	vmcs_write(APIC_ACCS_ADDR, virt_to_phys(apic_access_page));
3651 
3652 	test_vmcs_addr_reference(CPU_VIRT_APIC_ACCESSES, APIC_ACCS_ADDR,
3653 				 "APIC-access address",
3654 				 "virtualize APIC-accesses", PAGE_SIZE,
3655 				 false, false);
3656 }
3657 
3658 static bool set_bit_pattern(u8 mask, u32 *secondary)
3659 {
3660 	u8 i;
3661 	bool flag = false;
3662 	u32 test_bits[3] = {
3663 		CPU_VIRT_X2APIC,
3664 		CPU_APIC_REG_VIRT,
3665 		CPU_VINTD
3666 	};
3667 
3668         for (i = 0; i < ARRAY_SIZE(test_bits); i++) {
3669 		if ((mask & (1u << i)) &&
3670 		    (ctrl_cpu_rev[1].clr & test_bits[i])) {
3671 			*secondary |= test_bits[i];
3672 			flag = true;
3673 		}
3674 	}
3675 
3676 	return (flag);
3677 }
3678 
3679 /*
3680  * If the "use TPR shadow" VM-execution control is 0, the following
3681  * VM-execution controls must also be 0:
3682  * 	- virtualize x2APIC mode
3683  *	- APIC-register virtualization
3684  *	- virtual-interrupt delivery
3685  *    [Intel SDM]
3686  *
3687  * 2. If the "virtualize x2APIC mode" VM-execution control is 1, the
3688  *    "virtualize APIC accesses" VM-execution control must be 0.
3689  *    [Intel SDM]
3690  */
3691 static void test_apic_virtual_ctls(void)
3692 {
3693 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3694 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3695 	u32 primary = saved_primary;
3696 	u32 secondary = saved_secondary;
3697 	bool ctrl = false;
3698 	char str[10] = "disabled";
3699 	u8 i = 0, j;
3700 
3701 	/*
3702 	 * First test
3703 	 */
3704 	if (!((ctrl_cpu_rev[0].clr & (CPU_SECONDARY | CPU_TPR_SHADOW)) ==
3705 	    (CPU_SECONDARY | CPU_TPR_SHADOW)))
3706 		return;
3707 
3708 	primary |= CPU_SECONDARY;
3709 	primary &= ~CPU_TPR_SHADOW;
3710 	vmcs_write(CPU_EXEC_CTRL0, primary);
3711 
3712 	while (1) {
3713 		for (j = 1; j < 8; j++) {
3714 			secondary &= ~(CPU_VIRT_X2APIC | CPU_APIC_REG_VIRT | CPU_VINTD);
3715 			if (primary & CPU_TPR_SHADOW) {
3716 				ctrl = true;
3717 			} else {
3718 				if (! set_bit_pattern(j, &secondary))
3719 					ctrl = true;
3720 				else
3721 					ctrl = false;
3722 			}
3723 
3724 			vmcs_write(CPU_EXEC_CTRL1, secondary);
3725 			report_prefix_pushf("Use TPR shadow %s, virtualize x2APIC mode %s, APIC-register virtualization %s, virtual-interrupt delivery %s",
3726 				str, (secondary & CPU_VIRT_X2APIC) ? "enabled" : "disabled", (secondary & CPU_APIC_REG_VIRT) ? "enabled" : "disabled", (secondary & CPU_VINTD) ? "enabled" : "disabled");
3727 			test_vmx_controls(ctrl, false);
3728 			report_prefix_pop();
3729 		}
3730 
3731 		if (i == 1)
3732 			break;
3733 		i++;
3734 
3735 		primary |= CPU_TPR_SHADOW;
3736 		vmcs_write(CPU_EXEC_CTRL0, primary);
3737 		strcpy(str, "enabled");
3738 	}
3739 
3740 	/*
3741 	 * Second test
3742 	 */
3743 	u32 apic_virt_ctls = (CPU_VIRT_X2APIC | CPU_VIRT_APIC_ACCESSES);
3744 
3745 	primary = saved_primary;
3746 	secondary = saved_secondary;
3747 	if (!((ctrl_cpu_rev[1].clr & apic_virt_ctls) == apic_virt_ctls))
3748 		return;
3749 
3750 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY);
3751 	secondary &= ~CPU_VIRT_APIC_ACCESSES;
3752 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_X2APIC);
3753 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access disabled");
3754 	test_vmx_controls(true, false);
3755 	report_prefix_pop();
3756 
3757 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_APIC_ACCESSES);
3758 	report_prefix_pushf("Virtualize x2APIC mode disabled; virtualize APIC access enabled");
3759 	test_vmx_controls(true, false);
3760 	report_prefix_pop();
3761 
3762 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VIRT_X2APIC);
3763 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access enabled");
3764 	test_vmx_controls(false, false);
3765 	report_prefix_pop();
3766 
3767 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VIRT_APIC_ACCESSES);
3768 	report_prefix_pushf("Virtualize x2APIC mode enabled; virtualize APIC access disabled");
3769 	test_vmx_controls(true, false);
3770 	report_prefix_pop();
3771 
3772 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3773 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3774 }
3775 
3776 /*
3777  * If the "virtual-interrupt delivery" VM-execution control is 1, the
3778  * "external-interrupt exiting" VM-execution control must be 1.
3779  * [Intel SDM]
3780  */
3781 static void test_virtual_intr_ctls(void)
3782 {
3783 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3784 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3785 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3786 	u32 primary = saved_primary;
3787 	u32 secondary = saved_secondary;
3788 	u32 pin = saved_pin;
3789 
3790 	if (!((ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3791 	    (ctrl_pin_rev.clr & PIN_EXTINT)))
3792 		return;
3793 
3794 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3795 	vmcs_write(CPU_EXEC_CTRL1, secondary & ~CPU_VINTD);
3796 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3797 	report_prefix_pushf("Virtualize interrupt-delivery disabled; external-interrupt exiting disabled");
3798 	test_vmx_controls(true, false);
3799 	report_prefix_pop();
3800 
3801 	vmcs_write(CPU_EXEC_CTRL1, secondary | CPU_VINTD);
3802 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3803 	test_vmx_controls(false, false);
3804 	report_prefix_pop();
3805 
3806 	vmcs_write(PIN_CONTROLS, pin | PIN_EXTINT);
3807 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting enabled");
3808 	test_vmx_controls(true, false);
3809 	report_prefix_pop();
3810 
3811 	vmcs_write(PIN_CONTROLS, pin & ~PIN_EXTINT);
3812 	report_prefix_pushf("Virtualize interrupt-delivery enabled; external-interrupt exiting disabled");
3813 	test_vmx_controls(false, false);
3814 	report_prefix_pop();
3815 
3816 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3817 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3818 	vmcs_write(PIN_CONTROLS, saved_pin);
3819 }
3820 
3821 static void test_pi_desc_addr(u64 addr, bool ctrl)
3822 {
3823 	vmcs_write(POSTED_INTR_DESC_ADDR, addr);
3824 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-descriptor-address 0x%lx", addr);
3825 	test_vmx_controls(ctrl, false);
3826 	report_prefix_pop();
3827 }
3828 
3829 /*
3830  * If the “process posted interrupts†VM-execution control is 1, the
3831  * following must be true:
3832  *
3833  *	- The “virtual-interrupt delivery†VM-execution control is 1.
3834  *	- The “acknowledge interrupt on exit†VM-exit control is 1.
3835  *	- The posted-interrupt notification vector has a value in the
3836  *	- range 0–255 (bits 15:8 are all 0).
3837  *	- Bits 5:0 of the posted-interrupt descriptor address are all 0.
3838  *	- The posted-interrupt descriptor address does not set any bits
3839  *	  beyond the processor's physical-address width.
3840  * [Intel SDM]
3841  */
3842 static void test_posted_intr(void)
3843 {
3844 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3845 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3846 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
3847 	u32 exit_ctl_saved = vmcs_read(EXI_CONTROLS);
3848 	u32 primary = saved_primary;
3849 	u32 secondary = saved_secondary;
3850 	u32 pin = saved_pin;
3851 	u32 exit_ctl = exit_ctl_saved;
3852 	u16 vec;
3853 	int i;
3854 
3855 	if (!((ctrl_pin_rev.clr & PIN_POST_INTR) &&
3856 	    (ctrl_cpu_rev[1].clr & CPU_VINTD) &&
3857 	    (ctrl_exit_rev.clr & EXI_INTA)))
3858 		return;
3859 
3860 	vmcs_write(CPU_EXEC_CTRL0, primary | CPU_SECONDARY | CPU_TPR_SHADOW);
3861 
3862 	/*
3863 	 * Test virtual-interrupt-delivery and acknowledge-interrupt-on-exit
3864 	 */
3865 	pin |= PIN_POST_INTR;
3866 	vmcs_write(PIN_CONTROLS, pin);
3867 	secondary &= ~CPU_VINTD;
3868 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3869 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled");
3870 	test_vmx_controls(false, false);
3871 	report_prefix_pop();
3872 
3873 	secondary |= CPU_VINTD;
3874 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3875 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled");
3876 	test_vmx_controls(false, false);
3877 	report_prefix_pop();
3878 
3879 	exit_ctl &= ~EXI_INTA;
3880 	vmcs_write(EXI_CONTROLS, exit_ctl);
3881 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit disabled");
3882 	test_vmx_controls(false, false);
3883 	report_prefix_pop();
3884 
3885 	exit_ctl |= EXI_INTA;
3886 	vmcs_write(EXI_CONTROLS, exit_ctl);
3887 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3888 	test_vmx_controls(true, false);
3889 	report_prefix_pop();
3890 
3891 	secondary &= ~CPU_VINTD;
3892 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3893 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery disabled; acknowledge-interrupt-on-exit enabled");
3894 	test_vmx_controls(false, false);
3895 	report_prefix_pop();
3896 
3897 	secondary |= CPU_VINTD;
3898 	vmcs_write(CPU_EXEC_CTRL1, secondary);
3899 	report_prefix_pushf("Process-posted-interrupts enabled; virtual-interrupt-delivery enabled; acknowledge-interrupt-on-exit enabled");
3900 	test_vmx_controls(true, false);
3901 	report_prefix_pop();
3902 
3903 	/*
3904 	 * Test posted-interrupt notification vector
3905 	 */
3906 	for (i = 0; i < 8; i++) {
3907 		vec = (1ul << i);
3908 		vmcs_write(PINV, vec);
3909 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3910 		test_vmx_controls(true, false);
3911 		report_prefix_pop();
3912 	}
3913 	for (i = 8; i < 16; i++) {
3914 		vec = (1ul << i);
3915 		vmcs_write(PINV, vec);
3916 		report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3917 		test_vmx_controls(false, false);
3918 		report_prefix_pop();
3919 	}
3920 
3921 	vec &= ~(0xff << 8);
3922 	vmcs_write(PINV, vec);
3923 	report_prefix_pushf("Process-posted-interrupts enabled; posted-interrupt-notification-vector %u", vec);
3924 	test_vmx_controls(true, false);
3925 	report_prefix_pop();
3926 
3927 	/*
3928 	 * Test posted-interrupt descriptor addresss
3929 	 */
3930 	for (i = 0; i < 6; i++) {
3931 		test_pi_desc_addr(1ul << i, false);
3932 	}
3933 
3934 	test_pi_desc_addr(0xf0, false);
3935 	test_pi_desc_addr(0xff, false);
3936 	test_pi_desc_addr(0x0f, false);
3937 	test_pi_desc_addr(0x8000, true);
3938 	test_pi_desc_addr(0x00, true);
3939 	test_pi_desc_addr(0xc000, true);
3940 
3941 	test_vmcs_addr_values("process-posted interrupts",
3942 			       POSTED_INTR_DESC_ADDR, 64,
3943 			       false, false, 0, 63);
3944 
3945 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3946 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3947 	vmcs_write(PIN_CONTROLS, saved_pin);
3948 }
3949 
3950 static void test_apic_ctls(void)
3951 {
3952 	test_apic_virt_addr();
3953 	test_apic_access_addr();
3954 	test_apic_virtual_ctls();
3955 	test_virtual_intr_ctls();
3956 	test_posted_intr();
3957 }
3958 
3959 /*
3960  * If the “enable VPID†VM-execution control is 1, the value of the
3961  * of the VPID VM-execution control field must not be 0000H.
3962  * [Intel SDM]
3963  */
3964 static void test_vpid(void)
3965 {
3966 	u32 saved_primary = vmcs_read(CPU_EXEC_CTRL0);
3967 	u32 saved_secondary = vmcs_read(CPU_EXEC_CTRL1);
3968 	u16 vpid = 0x0000;
3969 	int i;
3970 
3971 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
3972 	    (ctrl_cpu_rev[1].clr & CPU_VPID))) {
3973 		test_skip("Secondary controls and/or VPID not supported");
3974 		return;
3975 	}
3976 
3977 	vmcs_write(CPU_EXEC_CTRL0, saved_primary | CPU_SECONDARY);
3978 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary & ~CPU_VPID);
3979 	vmcs_write(VPID, vpid);
3980 	report_prefix_pushf("VPID disabled; VPID value %x", vpid);
3981 	test_vmx_controls(true, false);
3982 	report_prefix_pop();
3983 
3984 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary | CPU_VPID);
3985 	report_prefix_pushf("VPID enabled; VPID value %x", vpid);
3986 	test_vmx_controls(false, false);
3987 	report_prefix_pop();
3988 
3989 	for (i = 0; i < 16; i++) {
3990 		vpid = (short)1 << i;;
3991 		vmcs_write(VPID, vpid);
3992 		report_prefix_pushf("VPID enabled; VPID value %x", vpid);
3993 		test_vmx_controls(true, false);
3994 		report_prefix_pop();
3995 	}
3996 
3997 	vmcs_write(CPU_EXEC_CTRL0, saved_primary);
3998 	vmcs_write(CPU_EXEC_CTRL1, saved_secondary);
3999 }
4000 
4001 static void set_vtpr(unsigned vtpr)
4002 {
4003 	*(u32 *)phys_to_virt(vmcs_read(APIC_VIRT_ADDR) + APIC_TASKPRI) = vtpr;
4004 }
4005 
4006 static void try_tpr_threshold_and_vtpr(unsigned threshold, unsigned vtpr)
4007 {
4008 	bool valid = true;
4009 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4010 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4011 
4012 	if ((primary & CPU_TPR_SHADOW) &&
4013 	    (!(primary & CPU_SECONDARY) ||
4014 	     !(secondary & (CPU_VINTD | CPU_VIRT_APIC_ACCESSES))))
4015 		valid = (threshold & 0xf) <= ((vtpr >> 4) & 0xf);
4016 
4017 	set_vtpr(vtpr);
4018 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0x%x",
4019 	    threshold, (vtpr >> 4) & 0xf);
4020 	test_vmx_controls(valid, false);
4021 	report_prefix_pop();
4022 }
4023 
4024 static void test_invalid_event_injection(void)
4025 {
4026 	u32 ent_intr_info_save = vmcs_read(ENT_INTR_INFO);
4027 	u32 ent_intr_error_save = vmcs_read(ENT_INTR_ERROR);
4028 	u32 ent_inst_len_save = vmcs_read(ENT_INST_LEN);
4029 	u32 primary_save = vmcs_read(CPU_EXEC_CTRL0);
4030 	u32 secondary_save = vmcs_read(CPU_EXEC_CTRL1);
4031 	u64 guest_cr0_save = vmcs_read(GUEST_CR0);
4032 	u32 ent_intr_info_base = INTR_INFO_VALID_MASK;
4033 	u32 ent_intr_info, ent_intr_err, ent_intr_len;
4034 	u32 cnt;
4035 
4036 	/* Setup */
4037 	report_prefix_push("invalid event injection");
4038 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4039 	vmcs_write(ENT_INST_LEN, 0x00000001);
4040 
4041 	/* The field’s interruption type is not set to a reserved value. */
4042 	ent_intr_info = ent_intr_info_base | INTR_TYPE_RESERVED | DE_VECTOR;
4043 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4044 			    "RESERVED interruption type invalid [-]",
4045 			    ent_intr_info);
4046 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4047 	test_vmx_controls(false, false);
4048 	report_prefix_pop();
4049 
4050 	ent_intr_info = ent_intr_info_base | INTR_TYPE_EXT_INTR |
4051 			DE_VECTOR;
4052 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4053 			    "RESERVED interruption type invalid [+]",
4054 			    ent_intr_info);
4055 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4056 	test_vmx_controls(true, false);
4057 	report_prefix_pop();
4058 
4059 	/* If the interruption type is other event, the vector is 0. */
4060 	ent_intr_info = ent_intr_info_base | INTR_TYPE_OTHER_EVENT | DB_VECTOR;
4061 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4062 			    "(OTHER EVENT && vector != 0) invalid [-]",
4063 			    ent_intr_info);
4064 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4065 	test_vmx_controls(false, false);
4066 	report_prefix_pop();
4067 
4068 	/* If the interruption type is NMI, the vector is 2 (negative case). */
4069 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | DE_VECTOR;
4070 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4071 			    "(NMI && vector != 2) invalid [-]", ent_intr_info);
4072 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4073 	test_vmx_controls(false, false);
4074 	report_prefix_pop();
4075 
4076 	/* If the interruption type is NMI, the vector is 2 (positive case). */
4077 	ent_intr_info = ent_intr_info_base | INTR_TYPE_NMI_INTR | NMI_VECTOR;
4078 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4079 			    "(NMI && vector == 2) valid [+]", ent_intr_info);
4080 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4081 	test_vmx_controls(true, false);
4082 	report_prefix_pop();
4083 
4084 	/*
4085 	 * If the interruption type
4086 	 * is HW exception, the vector is at most 31.
4087 	 */
4088 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION | 0x20;
4089 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4090 			    "(HW exception && vector > 31) invalid [-]",
4091 			    ent_intr_info);
4092 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4093 	test_vmx_controls(false, false);
4094 	report_prefix_pop();
4095 
4096 	/*
4097 	 * deliver-error-code is 1 iff either
4098 	 * (a) the "unrestricted guest" VM-execution control is 0
4099 	 * (b) CR0.PE is set.
4100 	 */
4101 
4102 	/* Assert that unrestricted guest is disabled or unsupported */
4103 	assert(!(ctrl_cpu_rev[0].clr & CPU_SECONDARY) ||
4104 	       !(secondary_save & CPU_URG));
4105 
4106 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4107 			GP_VECTOR;
4108 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4109 			    "error code <-> (!URG || prot_mode) [-]",
4110 			    ent_intr_info);
4111 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4112 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4113 	test_vmx_controls(false, false);
4114 	report_prefix_pop();
4115 
4116 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4117 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4118 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4119 			    "error code <-> (!URG || prot_mode) [+]",
4120 			    ent_intr_info);
4121 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4122 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4123 	test_vmx_controls(true, false);
4124 	report_prefix_pop();
4125 
4126 	if (enable_unrestricted_guest())
4127 		goto skip_unrestricted_guest;
4128 
4129 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4130 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4131 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4132 			    "error code <-> (!URG || prot_mode) [-]",
4133 			    ent_intr_info);
4134 	vmcs_write(GUEST_CR0, guest_cr0_save & ~X86_CR0_PE & ~X86_CR0_PG);
4135 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4136 	test_vmx_controls(false, false);
4137 	report_prefix_pop();
4138 
4139 	ent_intr_info = ent_intr_info_base | INTR_TYPE_HARD_EXCEPTION |
4140 			GP_VECTOR;
4141 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4142 			    "error code <-> (!URG || prot_mode) [-]",
4143 			    ent_intr_info);
4144 	vmcs_write(GUEST_CR0, guest_cr0_save | X86_CR0_PE);
4145 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4146 	test_vmx_controls(false, false);
4147 	report_prefix_pop();
4148 
4149 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4150 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4151 
4152 skip_unrestricted_guest:
4153 	vmcs_write(GUEST_CR0, guest_cr0_save);
4154 
4155 	/* deliver-error-code is 1 iff the interruption type is HW exception */
4156 	report_prefix_push("error code <-> HW exception");
4157 	for (cnt = 0; cnt < 8; cnt++) {
4158 		u32 exception_type_mask = cnt << 8;
4159 		u32 deliver_error_code_mask =
4160 			exception_type_mask != INTR_TYPE_HARD_EXCEPTION ?
4161 			INTR_INFO_DELIVER_CODE_MASK : 0;
4162 
4163 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4164 				exception_type_mask | GP_VECTOR;
4165 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4166 				    ent_intr_info);
4167 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4168 		test_vmx_controls(false, false);
4169 		report_prefix_pop();
4170 	}
4171 	report_prefix_pop();
4172 
4173 	/*
4174 	 * deliver-error-code is 1 iff the the vector
4175 	 * indicates an exception that would normally deliver an error code
4176 	 */
4177 	report_prefix_push("error code <-> vector delivers error code");
4178 	for (cnt = 0; cnt < 32; cnt++) {
4179 		bool has_error_code = false;
4180 		u32 deliver_error_code_mask;
4181 
4182 		switch (cnt) {
4183 		case DF_VECTOR:
4184 		case TS_VECTOR:
4185 		case NP_VECTOR:
4186 		case SS_VECTOR:
4187 		case GP_VECTOR:
4188 		case PF_VECTOR:
4189 		case AC_VECTOR:
4190 			has_error_code = true;
4191 		}
4192 
4193 		/* Negative case */
4194 		deliver_error_code_mask = has_error_code ?
4195 						0 :
4196 						INTR_INFO_DELIVER_CODE_MASK;
4197 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4198 				INTR_TYPE_HARD_EXCEPTION | cnt;
4199 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4200 				    ent_intr_info);
4201 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4202 		test_vmx_controls(false, false);
4203 		report_prefix_pop();
4204 
4205 		/* Positive case */
4206 		deliver_error_code_mask = has_error_code ?
4207 						INTR_INFO_DELIVER_CODE_MASK :
4208 						0;
4209 		ent_intr_info = ent_intr_info_base | deliver_error_code_mask |
4210 				INTR_TYPE_HARD_EXCEPTION | cnt;
4211 		report_prefix_pushf("VM-entry intr info=0x%x [+]",
4212 				    ent_intr_info);
4213 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4214 		test_vmx_controls(true, false);
4215 		report_prefix_pop();
4216 	}
4217 	report_prefix_pop();
4218 
4219 	/* Reserved bits in the field (30:12) are 0. */
4220 	report_prefix_push("reserved bits clear");
4221 	for (cnt = 12; cnt <= 30; cnt++) {
4222 		ent_intr_info = ent_intr_info_base |
4223 				INTR_INFO_DELIVER_CODE_MASK |
4224 				INTR_TYPE_HARD_EXCEPTION | GP_VECTOR |
4225 				(1U << cnt);
4226 		report_prefix_pushf("VM-entry intr info=0x%x [-]",
4227 				    ent_intr_info);
4228 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4229 		test_vmx_controls(false, false);
4230 		report_prefix_pop();
4231 	}
4232 	report_prefix_pop();
4233 
4234 	/*
4235 	 * If deliver-error-code is 1
4236 	 * bits 31:15 of the VM-entry exception error-code field are 0.
4237 	 */
4238 	ent_intr_info = ent_intr_info_base | INTR_INFO_DELIVER_CODE_MASK |
4239 			INTR_TYPE_HARD_EXCEPTION | GP_VECTOR;
4240 	report_prefix_pushf("%s, VM-entry intr info=0x%x",
4241 			    "VM-entry exception error code[31:15] clear",
4242 			    ent_intr_info);
4243 	vmcs_write(ENT_INTR_INFO, ent_intr_info);
4244 	for (cnt = 15; cnt <= 31; cnt++) {
4245 		ent_intr_err = 1U << cnt;
4246 		report_prefix_pushf("VM-entry intr error=0x%x [-]",
4247 				    ent_intr_err);
4248 		vmcs_write(ENT_INTR_ERROR, ent_intr_err);
4249 		test_vmx_controls(false, false);
4250 		report_prefix_pop();
4251 	}
4252 	vmcs_write(ENT_INTR_ERROR, 0x00000000);
4253 	report_prefix_pop();
4254 
4255 	/*
4256 	 * If the interruption type is software interrupt, software exception,
4257 	 * or privileged software exception, the VM-entry instruction-length
4258 	 * field is in the range 0–15.
4259 	 */
4260 
4261 	for (cnt = 0; cnt < 3; cnt++) {
4262 		switch (cnt) {
4263 		case 0:
4264 			ent_intr_info = ent_intr_info_base |
4265 					INTR_TYPE_SOFT_INTR;
4266 			break;
4267 		case 1:
4268 			ent_intr_info = ent_intr_info_base |
4269 					INTR_TYPE_SOFT_EXCEPTION;
4270 			break;
4271 		case 2:
4272 			ent_intr_info = ent_intr_info_base |
4273 					INTR_TYPE_PRIV_SW_EXCEPTION;
4274 			break;
4275 		}
4276 		report_prefix_pushf("%s, VM-entry intr info=0x%x",
4277 				    "VM-entry instruction-length check",
4278 				    ent_intr_info);
4279 		vmcs_write(ENT_INTR_INFO, ent_intr_info);
4280 
4281 		/* Instruction length set to -1 (0xFFFFFFFF) should fail */
4282 		ent_intr_len = -1;
4283 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4284 				    ent_intr_len);
4285 		vmcs_write(ENT_INST_LEN, ent_intr_len);
4286 		test_vmx_controls(false, false);
4287 		report_prefix_pop();
4288 
4289 		/* Instruction length set to 16 should fail */
4290 		ent_intr_len = 0x00000010;
4291 		report_prefix_pushf("VM-entry intr length = 0x%x [-]",
4292 				    ent_intr_len);
4293 		vmcs_write(ENT_INST_LEN, 0x00000010);
4294 		test_vmx_controls(false, false);
4295 		report_prefix_pop();
4296 
4297 		report_prefix_pop();
4298 	}
4299 
4300 	/* Cleanup */
4301 	vmcs_write(ENT_INTR_INFO, ent_intr_info_save);
4302 	vmcs_write(ENT_INTR_ERROR, ent_intr_error_save);
4303 	vmcs_write(ENT_INST_LEN, ent_inst_len_save);
4304 	vmcs_write(CPU_EXEC_CTRL0, primary_save);
4305 	vmcs_write(CPU_EXEC_CTRL1, secondary_save);
4306 	vmcs_write(GUEST_CR0, guest_cr0_save);
4307 	report_prefix_pop();
4308 }
4309 
4310 /*
4311  * Test interesting vTPR values for a given TPR threshold.
4312  */
4313 static void test_vtpr_values(unsigned threshold)
4314 {
4315 	try_tpr_threshold_and_vtpr(threshold, (threshold - 1) << 4);
4316 	try_tpr_threshold_and_vtpr(threshold, threshold << 4);
4317 	try_tpr_threshold_and_vtpr(threshold, (threshold + 1) << 4);
4318 }
4319 
4320 static void try_tpr_threshold(unsigned threshold)
4321 {
4322 	bool valid = true;
4323 
4324 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4325 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4326 
4327 	if ((primary & CPU_TPR_SHADOW) && !((primary & CPU_SECONDARY) &&
4328 	    (secondary & CPU_VINTD)))
4329 		valid = !(threshold >> 4);
4330 
4331 	set_vtpr(-1);
4332 	vmcs_write(TPR_THRESHOLD, threshold);
4333 	report_prefix_pushf("TPR threshold 0x%x, VTPR.class 0xf", threshold);
4334 	test_vmx_controls(valid, false);
4335 	report_prefix_pop();
4336 
4337 	if (valid)
4338 		test_vtpr_values(threshold);
4339 }
4340 
4341 /*
4342  * Test interesting TPR threshold values.
4343  */
4344 static void test_tpr_threshold_values(void)
4345 {
4346 	unsigned i;
4347 
4348 	for (i = 0; i < 0x10; i++)
4349 		try_tpr_threshold(i);
4350 	for (i = 4; i < 32; i++)
4351 		try_tpr_threshold(1u << i);
4352 	try_tpr_threshold(-1u);
4353 	try_tpr_threshold(0x7fffffff);
4354 }
4355 
4356 /*
4357  * This test covers the following two VM entry checks:
4358  *
4359  *      i) If the "use TPR shadow" VM-execution control is 1 and the
4360  *         "virtual-interrupt delivery" VM-execution control is 0, bits
4361  *         31:4 of the TPR threshold VM-execution control field must
4362 	   be 0.
4363  *         [Intel SDM]
4364  *
4365  *      ii) If the "use TPR shadow" VM-execution control is 1, the
4366  *          "virtual-interrupt delivery" VM-execution control is 0
4367  *          and the "virtualize APIC accesses" VM-execution control
4368  *          is 0, the value of bits 3:0 of the TPR threshold VM-execution
4369  *          control field must not be greater than the value of bits
4370  *          7:4 of VTPR.
4371  *          [Intel SDM]
4372  */
4373 static void test_tpr_threshold(void)
4374 {
4375 	u32 primary = vmcs_read(CPU_EXEC_CTRL0);
4376 	void *virtual_apic_page;
4377 
4378 	if (!(ctrl_cpu_rev[0].clr & CPU_TPR_SHADOW))
4379 		return;
4380 
4381 	virtual_apic_page = alloc_page();
4382 	memset(virtual_apic_page, 0xff, PAGE_SIZE);
4383 	vmcs_write(APIC_VIRT_ADDR, virt_to_phys(virtual_apic_page));
4384 
4385 	vmcs_write(CPU_EXEC_CTRL0, primary & ~(CPU_TPR_SHADOW | CPU_SECONDARY));
4386 	report_prefix_pushf("Use TPR shadow disabled, secondary controls disabled");
4387 	test_tpr_threshold_values();
4388 	report_prefix_pop();
4389 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | CPU_TPR_SHADOW);
4390 	report_prefix_pushf("Use TPR shadow enabled, secondary controls disabled");
4391 	test_tpr_threshold_values();
4392 	report_prefix_pop();
4393 
4394 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4395 	    (ctrl_cpu_rev[1].clr & (CPU_VINTD  | CPU_VIRT_APIC_ACCESSES)))) {
4396 		vmcs_write(CPU_EXEC_CTRL0, primary);
4397 		return;
4398 	}
4399 
4400 	u32 secondary = vmcs_read(CPU_EXEC_CTRL1);
4401 
4402 	if (ctrl_cpu_rev[1].clr & CPU_VINTD) {
4403 		vmcs_write(CPU_EXEC_CTRL1, CPU_VINTD);
4404 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4405 		test_tpr_threshold_values();
4406 		report_prefix_pop();
4407 
4408 		vmcs_write(CPU_EXEC_CTRL0,
4409 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4410 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses disabled");
4411 		test_tpr_threshold_values();
4412 		report_prefix_pop();
4413 	}
4414 
4415 	if (ctrl_cpu_rev[1].clr & CPU_VIRT_APIC_ACCESSES) {
4416 		vmcs_write(CPU_EXEC_CTRL0,
4417 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4418 		vmcs_write(CPU_EXEC_CTRL1, CPU_VIRT_APIC_ACCESSES);
4419 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4420 		test_tpr_threshold_values();
4421 		report_prefix_pop();
4422 
4423 		vmcs_write(CPU_EXEC_CTRL0,
4424 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4425 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4426 		test_tpr_threshold_values();
4427 		report_prefix_pop();
4428 	}
4429 
4430 	if ((ctrl_cpu_rev[1].clr &
4431 	     (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) ==
4432 	    (CPU_VINTD | CPU_VIRT_APIC_ACCESSES)) {
4433 		vmcs_write(CPU_EXEC_CTRL0,
4434 			   vmcs_read(CPU_EXEC_CTRL0) & ~CPU_SECONDARY);
4435 		vmcs_write(CPU_EXEC_CTRL1,
4436 			   CPU_VINTD | CPU_VIRT_APIC_ACCESSES);
4437 		report_prefix_pushf("Use TPR shadow enabled; secondary controls disabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4438 		test_tpr_threshold_values();
4439 		report_prefix_pop();
4440 
4441 		vmcs_write(CPU_EXEC_CTRL0,
4442 			   vmcs_read(CPU_EXEC_CTRL0) | CPU_SECONDARY);
4443 		report_prefix_pushf("Use TPR shadow enabled; secondary controls enabled; virtual-interrupt delivery enabled; virtualize APIC accesses enabled");
4444 		test_tpr_threshold_values();
4445 		report_prefix_pop();
4446 	}
4447 
4448 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4449 	vmcs_write(CPU_EXEC_CTRL0, primary);
4450 }
4451 
4452 /*
4453  * This test verifies the following two vmentry checks:
4454  *
4455  *  If the "NMI exiting" VM-execution control is 0, "Virtual NMIs"
4456  *  VM-execution control must be 0.
4457  *  [Intel SDM]
4458  *
4459  *  If the “virtual NMIs” VM-execution control is 0, the “NMI-window
4460  *  exiting” VM-execution control must be 0.
4461  *  [Intel SDM]
4462  */
4463 static void test_nmi_ctrls(void)
4464 {
4465 	u32 pin_ctrls, cpu_ctrls0, test_pin_ctrls, test_cpu_ctrls0;
4466 
4467 	if ((ctrl_pin_rev.clr & (PIN_NMI | PIN_VIRT_NMI)) !=
4468 	    (PIN_NMI | PIN_VIRT_NMI)) {
4469 		test_skip("NMI exiting and Virtual NMIs are not supported !");
4470 		return;
4471 	}
4472 
4473 	/* Save the controls so that we can restore them after our tests */
4474 	pin_ctrls = vmcs_read(PIN_CONTROLS);
4475 	cpu_ctrls0 = vmcs_read(CPU_EXEC_CTRL0);
4476 
4477 	test_pin_ctrls = pin_ctrls & ~(PIN_NMI | PIN_VIRT_NMI);
4478 	test_cpu_ctrls0 = cpu_ctrls0 & ~CPU_NMI_WINDOW;
4479 
4480 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4481 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs disabled");
4482 	test_vmx_controls(true, false);
4483 	report_prefix_pop();
4484 
4485 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_VIRT_NMI);
4486 	report_prefix_pushf("NMI-exiting disabled, virtual-NMIs enabled");
4487 	test_vmx_controls(false, false);
4488 	report_prefix_pop();
4489 
4490 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4491 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs enabled");
4492 	test_vmx_controls(true, false);
4493 	report_prefix_pop();
4494 
4495 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | PIN_NMI);
4496 	report_prefix_pushf("NMI-exiting enabled, virtual-NMIs disabled");
4497 	test_vmx_controls(true, false);
4498 	report_prefix_pop();
4499 
4500 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
4501 		report_info("NMI-window exiting is not supported, skipping...");
4502 		goto done;
4503 	}
4504 
4505 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4506 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4507 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting enabled");
4508 	test_vmx_controls(false, false);
4509 	report_prefix_pop();
4510 
4511 	vmcs_write(PIN_CONTROLS, test_pin_ctrls);
4512 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4513 	report_prefix_pushf("Virtual-NMIs disabled, NMI-window-exiting disabled");
4514 	test_vmx_controls(true, false);
4515 	report_prefix_pop();
4516 
4517 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4518 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0 | CPU_NMI_WINDOW);
4519 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting enabled");
4520 	test_vmx_controls(true, false);
4521 	report_prefix_pop();
4522 
4523 	vmcs_write(PIN_CONTROLS, test_pin_ctrls | (PIN_NMI | PIN_VIRT_NMI));
4524 	vmcs_write(CPU_EXEC_CTRL0, test_cpu_ctrls0);
4525 	report_prefix_pushf("Virtual-NMIs enabled, NMI-window-exiting disabled");
4526 	test_vmx_controls(true, false);
4527 	report_prefix_pop();
4528 
4529 	/* Restore the controls to their original values */
4530 	vmcs_write(CPU_EXEC_CTRL0, cpu_ctrls0);
4531 done:
4532 	vmcs_write(PIN_CONTROLS, pin_ctrls);
4533 }
4534 
4535 static void test_eptp_ad_bit(u64 eptp, bool ctrl)
4536 {
4537 	vmcs_write(EPTP, eptp);
4538 	report_prefix_pushf("Enable-EPT enabled; EPT accessed and dirty flag %s",
4539 	    (eptp & EPTP_AD_FLAG) ? "1": "0");
4540 	test_vmx_controls(ctrl, false);
4541 	report_prefix_pop();
4542 
4543 }
4544 
4545 /*
4546  * 1. If the "enable EPT" VM-execution control is 1, the "EPTP VM-execution"
4547  *    control field must satisfy the following checks:
4548  *
4549  *     - The EPT memory type (bits 2:0) must be a value supported by the
4550  *	 processor as indicated in the IA32_VMX_EPT_VPID_CAP MSR.
4551  *     - Bits 5:3 (1 less than the EPT page-walk length) must be 3,
4552  *	 indicating an EPT page-walk length of 4.
4553  *     - Bit 6 (enable bit for accessed and dirty flags for EPT) must be
4554  *	 0 if bit 21 of the IA32_VMX_EPT_VPID_CAP MSR is read as 0,
4555  *	 indicating that the processor does not support accessed and dirty
4556  *	 dirty flags for EPT.
4557  *     - Reserved bits 11:7 and 63:N (where N is the processor's
4558  *	 physical-address width) must all be 0.
4559  *
4560  * 2. If the "unrestricted guest" VM-execution control is 1, the
4561  *    "enable EPT" VM-execution control must also be 1.
4562  */
4563 static void test_ept_eptp(void)
4564 {
4565 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4566 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4567 	u64 eptp_saved = vmcs_read(EPTP);
4568 	u32 primary = primary_saved;
4569 	u32 secondary = secondary_saved;
4570 	u64 msr, eptp = eptp_saved;
4571 	bool un_cache = false;
4572 	bool wr_bk = false;
4573 	bool ctrl;
4574 	u32 i, maxphysaddr;
4575 	u64 j, resv_bits_mask = 0;
4576 
4577 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4578 	    (ctrl_cpu_rev[1].clr & CPU_EPT))) {
4579 		test_skip("\"CPU secondary\" and/or \"enable EPT\" execution controls are not supported !");
4580 		return;
4581 	}
4582 
4583 	/*
4584 	 * Memory type (bits 2:0)
4585 	 */
4586 	msr = rdmsr(MSR_IA32_VMX_EPT_VPID_CAP);
4587 	if (msr & EPT_CAP_UC)
4588 		un_cache = true;
4589 	if (msr & EPT_CAP_WB)
4590 		wr_bk = true;
4591 
4592 	primary |= CPU_SECONDARY;
4593 	vmcs_write(CPU_EXEC_CTRL0, primary);
4594 	secondary |= CPU_EPT;
4595 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4596 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4597 	    (3ul << EPTP_PG_WALK_LEN_SHIFT);
4598 	vmcs_write(EPTP, eptp);
4599 
4600 	for (i = 0; i < 8; i++) {
4601 		if (i == 0) {
4602 			if (un_cache) {
4603 				report_info("EPT paging structure memory-type is Un-cacheable\n");
4604 				ctrl = true;
4605 			} else {
4606 				ctrl = false;
4607 			}
4608 		} else if (i == 6) {
4609 			if (wr_bk) {
4610 				report_info("EPT paging structure memory-type is Write-back\n");
4611 				ctrl = true;
4612 			} else {
4613 				ctrl = false;
4614 			}
4615 		} else {
4616 			ctrl = false;
4617 		}
4618 
4619 		eptp = (eptp & ~EPT_MEM_TYPE_MASK) | i;
4620 		vmcs_write(EPTP, eptp);
4621 		report_prefix_pushf("Enable-EPT enabled; EPT memory type %lu",
4622 		    eptp & EPT_MEM_TYPE_MASK);
4623 		test_vmx_controls(ctrl, false);
4624 		report_prefix_pop();
4625 	}
4626 
4627 	eptp = (eptp & ~EPT_MEM_TYPE_MASK) | 6ul;
4628 
4629 	/*
4630 	 * Page walk length (bits 5:3)
4631 	 */
4632 	for (i = 0; i < 8; i++) {
4633 		eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4634 		    (i << EPTP_PG_WALK_LEN_SHIFT);
4635 		if (i == 3)
4636 			ctrl = true;
4637 		else
4638 			ctrl = false;
4639 
4640 		vmcs_write(EPTP, eptp);
4641 		report_prefix_pushf("Enable-EPT enabled; EPT page walk length %lu",
4642 		    eptp & EPTP_PG_WALK_LEN_MASK);
4643 		test_vmx_controls(ctrl, false);
4644 		report_prefix_pop();
4645 	}
4646 
4647 	eptp = (eptp & ~EPTP_PG_WALK_LEN_MASK) |
4648 	    3ul << EPTP_PG_WALK_LEN_SHIFT;
4649 
4650 	/*
4651 	 * Accessed and dirty flag (bit 6)
4652 	 */
4653 	if (msr & EPT_CAP_AD_FLAG) {
4654 		report_info("Processor supports accessed and dirty flag");
4655 		eptp &= ~EPTP_AD_FLAG;
4656 		test_eptp_ad_bit(eptp, true);
4657 
4658 		eptp |= EPTP_AD_FLAG;
4659 		test_eptp_ad_bit(eptp, true);
4660 	} else {
4661 		report_info("Processor does not supports accessed and dirty flag");
4662 		eptp &= ~EPTP_AD_FLAG;
4663 		test_eptp_ad_bit(eptp, true);
4664 
4665 		eptp |= EPTP_AD_FLAG;
4666 		test_eptp_ad_bit(eptp, false);
4667 	}
4668 
4669 	/*
4670 	 * Reserved bits [11:7] and [63:N]
4671 	 */
4672 	for (i = 0; i < 32; i++) {
4673 		if (i  == 0)
4674 			ctrl = true;
4675 		else
4676 			ctrl = false;
4677 
4678 		eptp = (eptp &
4679 		    ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT)) |
4680 		    (i << EPTP_RESERV_BITS_SHIFT);
4681 		vmcs_write(EPTP, eptp);
4682 		report_prefix_pushf("Enable-EPT enabled; reserved bits [11:7] %lu",
4683 		    (eptp >> EPTP_RESERV_BITS_SHIFT) &
4684 		    EPTP_RESERV_BITS_MASK);
4685 		test_vmx_controls(ctrl, false);
4686 		report_prefix_pop();
4687 	}
4688 
4689 	eptp = (eptp & ~(EPTP_RESERV_BITS_MASK << EPTP_RESERV_BITS_SHIFT));
4690 
4691 	maxphysaddr = cpuid_maxphyaddr();
4692 	for (i = 0; i < (63 - maxphysaddr + 1); i++) {
4693 		resv_bits_mask |= 1ul << i;
4694 	}
4695 
4696 	for (j = 0; j < (63 - maxphysaddr + 1); j++) {
4697 		if (j  == 0)
4698 			ctrl = true;
4699 		else
4700 			ctrl = false;
4701 
4702 		eptp = (eptp & ~(resv_bits_mask << maxphysaddr)) |
4703 		    (j << maxphysaddr);
4704 		vmcs_write(EPTP, eptp);
4705 		report_prefix_pushf("Enable-EPT enabled; reserved bits [63:N] %lu",
4706 		    (eptp >> maxphysaddr) & resv_bits_mask);
4707 		test_vmx_controls(ctrl, false);
4708 		report_prefix_pop();
4709 	}
4710 
4711 	secondary &= ~(CPU_EPT | CPU_URG);
4712 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4713 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest disabled");
4714 	test_vmx_controls(true, false);
4715 	report_prefix_pop();
4716 
4717 	secondary |= CPU_URG;
4718 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4719 	report_prefix_pushf("Enable-EPT disabled, unrestricted-guest enabled");
4720 	test_vmx_controls(false, false);
4721 	report_prefix_pop();
4722 
4723 	secondary |= CPU_EPT;
4724 	enable_ept();
4725 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest enabled");
4726 	test_vmx_controls(true, false);
4727 	report_prefix_pop();
4728 
4729 	secondary &= ~CPU_URG;
4730 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4731 	report_prefix_pushf("Enable-EPT enabled, unrestricted-guest disabled");
4732 	test_vmx_controls(true, false);
4733 	report_prefix_pop();
4734 
4735 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4736 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4737 	vmcs_write(EPTP, eptp_saved);
4738 }
4739 
4740 /*
4741  * If the 'enable PML' VM-execution control is 1, the 'enable EPT'
4742  * VM-execution control must also be 1. In addition, the PML address
4743  * must satisfy the following checks:
4744  *
4745  *    * Bits 11:0 of the address must be 0.
4746  *    * The address should not set any bits beyond the processor's
4747  *	physical-address width.
4748  *
4749  *  [Intel SDM]
4750  */
4751 static void test_pml(void)
4752 {
4753 	u32 primary_saved = vmcs_read(CPU_EXEC_CTRL0);
4754 	u32 secondary_saved = vmcs_read(CPU_EXEC_CTRL1);
4755 	u32 primary = primary_saved;
4756 	u32 secondary = secondary_saved;
4757 
4758 	if (!((ctrl_cpu_rev[0].clr & CPU_SECONDARY) &&
4759 	    (ctrl_cpu_rev[1].clr & CPU_EPT) && (ctrl_cpu_rev[1].clr & CPU_PML))) {
4760 		test_skip("\"Secondary execution\" control or \"enable EPT\" control or \"enable PML\" control is not supported !");
4761 		return;
4762 	}
4763 
4764 	primary |= CPU_SECONDARY;
4765 	vmcs_write(CPU_EXEC_CTRL0, primary);
4766 	secondary &= ~(CPU_PML | CPU_EPT);
4767 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4768 	report_prefix_pushf("enable-PML disabled, enable-EPT disabled");
4769 	test_vmx_controls(true, false);
4770 	report_prefix_pop();
4771 
4772 	secondary |= CPU_PML;
4773 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4774 	report_prefix_pushf("enable-PML enabled, enable-EPT disabled");
4775 	test_vmx_controls(false, false);
4776 	report_prefix_pop();
4777 
4778 	secondary |= CPU_EPT;
4779 	enable_ept();
4780 	report_prefix_pushf("enable-PML enabled, enable-EPT enabled");
4781 	test_vmx_controls(true, false);
4782 	report_prefix_pop();
4783 
4784 	secondary &= ~CPU_PML;
4785 	vmcs_write(CPU_EXEC_CTRL1, secondary);
4786 	report_prefix_pushf("enable-PML disabled, enable EPT enabled");
4787 	test_vmx_controls(true, false);
4788 	report_prefix_pop();
4789 
4790 	test_vmcs_addr_reference(CPU_PML, PMLADDR, "PML address", "PML",
4791 				 PAGE_SIZE, false, false);
4792 
4793 	vmcs_write(CPU_EXEC_CTRL0, primary_saved);
4794 	vmcs_write(CPU_EXEC_CTRL1, secondary_saved);
4795 }
4796 
4797  /*
4798  * If the "activate VMX-preemption timer" VM-execution control is 0, the
4799  * the "save VMX-preemption timer value" VM-exit control must also be 0.
4800  *
4801  *  [Intel SDM]
4802  */
4803 static void test_vmx_preemption_timer(void)
4804 {
4805 	u32 saved_pin = vmcs_read(PIN_CONTROLS);
4806 	u32 saved_exit = vmcs_read(EXI_CONTROLS);
4807 	u32 pin = saved_pin;
4808 	u32 exit = saved_exit;
4809 
4810 	if (!((ctrl_exit_rev.clr & EXI_SAVE_PREEMPT) ||
4811 	    (ctrl_pin_rev.clr & PIN_PREEMPT))) {
4812 		printf("\"Save-VMX-preemption-timer\" control and/or \"Enable-VMX-preemption-timer\" control is not supported\n");
4813 		return;
4814 	}
4815 
4816 	pin |= PIN_PREEMPT;
4817 	vmcs_write(PIN_CONTROLS, pin);
4818 	exit &= ~EXI_SAVE_PREEMPT;
4819 	vmcs_write(EXI_CONTROLS, exit);
4820 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer disabled");
4821 	test_vmx_controls(true, false);
4822 	report_prefix_pop();
4823 
4824 	exit |= EXI_SAVE_PREEMPT;
4825 	vmcs_write(EXI_CONTROLS, exit);
4826 	report_prefix_pushf("enable-VMX-preemption-timer enabled, save-VMX-preemption-timer enabled");
4827 	test_vmx_controls(true, false);
4828 	report_prefix_pop();
4829 
4830 	pin &= ~PIN_PREEMPT;
4831 	vmcs_write(PIN_CONTROLS, pin);
4832 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer enabled");
4833 	test_vmx_controls(false, false);
4834 	report_prefix_pop();
4835 
4836 	exit &= ~EXI_SAVE_PREEMPT;
4837 	vmcs_write(EXI_CONTROLS, exit);
4838 	report_prefix_pushf("enable-VMX-preemption-timer disabled, save-VMX-preemption-timer disabled");
4839 	test_vmx_controls(true, false);
4840 	report_prefix_pop();
4841 
4842 	vmcs_write(PIN_CONTROLS, saved_pin);
4843 	vmcs_write(EXI_CONTROLS, saved_exit);
4844 }
4845 
4846 /*
4847  * Tests for VM-execution control fields
4848  */
4849 static void test_vm_execution_ctls(void)
4850 {
4851 	test_pin_based_ctls();
4852 	test_primary_processor_based_ctls();
4853 	test_secondary_processor_based_ctls();
4854 	test_cr3_targets();
4855 	test_io_bitmaps();
4856 	test_msr_bitmap();
4857 	test_apic_ctls();
4858 	test_tpr_threshold();
4859 	test_nmi_ctrls();
4860 	test_pml();
4861 	test_vpid();
4862 	test_ept_eptp();
4863 	test_vmx_preemption_timer();
4864 }
4865 
4866  /*
4867   * The following checks are performed for the VM-entry MSR-load address if
4868   * the VM-entry MSR-load count field is non-zero:
4869   *
4870   *    - The lower 4 bits of the VM-entry MSR-load address must be 0.
4871   *      The address should not set any bits beyond the processor’s
4872   *      physical-address width.
4873   *
4874   *    - The address of the last byte in the VM-entry MSR-load area
4875   *      should not set any bits beyond the processor’s physical-address
4876   *      width. The address of this last byte is VM-entry MSR-load address
4877   *      + (MSR count * 16) - 1. (The arithmetic used for the computation
4878   *      uses more bits than the processor’s physical-address width.)
4879   *
4880   *
4881   *  [Intel SDM]
4882   */
4883 static void test_entry_msr_load(void)
4884 {
4885 	entry_msr_load = alloc_page();
4886 	u64 tmp;
4887 	u32 entry_msr_ld_cnt = 1;
4888 	int i;
4889 	u32 addr_len = 64;
4890 
4891 	vmcs_write(ENT_MSR_LD_CNT, entry_msr_ld_cnt);
4892 
4893 	/* Check first 4 bits of VM-entry MSR-load address */
4894 	for (i = 0; i < 4; i++) {
4895 		tmp = (u64)entry_msr_load | 1ull << i;
4896 		vmcs_write(ENTER_MSR_LD_ADDR, tmp);
4897 		report_prefix_pushf("VM-entry MSR-load addr [4:0] %lx",
4898 				    tmp & 0xf);
4899 		test_vmx_controls(false, false);
4900 		report_prefix_pop();
4901 	}
4902 
4903 	if (basic.val & (1ul << 48))
4904 		addr_len = 32;
4905 
4906 	test_vmcs_addr_values("VM-entry-MSR-load address",
4907 				ENTER_MSR_LD_ADDR, 16, false, false,
4908 				4, addr_len - 1);
4909 
4910 	/*
4911 	 * Check last byte of VM-entry MSR-load address
4912 	 */
4913 	entry_msr_load = (struct vmx_msr_entry *)((u64)entry_msr_load & ~0xf);
4914 
4915 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
4916 							i < 64; i++) {
4917 		tmp = ((u64)entry_msr_load + entry_msr_ld_cnt * 16 - 1) |
4918 			1ul << i;
4919 		vmcs_write(ENTER_MSR_LD_ADDR,
4920 			   tmp - (entry_msr_ld_cnt * 16 - 1));
4921 		test_vmx_controls(false, false);
4922 	}
4923 
4924 	vmcs_write(ENT_MSR_LD_CNT, 2);
4925 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
4926 	test_vmx_controls(false, false);
4927 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
4928 	test_vmx_controls(true, false);
4929 	vmcs_write(ENTER_MSR_LD_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
4930 	test_vmx_controls(true, false);
4931 }
4932 
4933 /*
4934  * Tests for VM-entry control fields
4935  */
4936 static void test_vm_entry_ctls(void)
4937 {
4938 	test_invalid_event_injection();
4939 	test_entry_msr_load();
4940 }
4941 
4942 /*
4943  * The following checks are performed for the VM-exit MSR-store address if
4944  * the VM-exit MSR-store count field is non-zero:
4945  *
4946  *    - The lower 4 bits of the VM-exit MSR-store address must be 0.
4947  *      The address should not set any bits beyond the processor’s
4948  *      physical-address width.
4949  *
4950  *    - The address of the last byte in the VM-exit MSR-store area
4951  *      should not set any bits beyond the processor’s physical-address
4952  *      width. The address of this last byte is VM-exit MSR-store address
4953  *      + (MSR count * 16) - 1. (The arithmetic used for the computation
4954  *      uses more bits than the processor’s physical-address width.)
4955  *
4956  * If IA32_VMX_BASIC[48] is read as 1, neither address should set any bits
4957  * in the range 63:32.
4958  *
4959  *  [Intel SDM]
4960  */
4961 static void test_exit_msr_store(void)
4962 {
4963 	exit_msr_store = alloc_page();
4964 	u64 tmp;
4965 	u32 exit_msr_st_cnt = 1;
4966 	int i;
4967 	u32 addr_len = 64;
4968 
4969 	vmcs_write(EXI_MSR_ST_CNT, exit_msr_st_cnt);
4970 
4971 	/* Check first 4 bits of VM-exit MSR-store address */
4972 	for (i = 0; i < 4; i++) {
4973 		tmp = (u64)exit_msr_store | 1ull << i;
4974 		vmcs_write(EXIT_MSR_ST_ADDR, tmp);
4975 		report_prefix_pushf("VM-exit MSR-store addr [4:0] %lx",
4976 				    tmp & 0xf);
4977 		test_vmx_controls(false, false);
4978 		report_prefix_pop();
4979 	}
4980 
4981 	if (basic.val & (1ul << 48))
4982 		addr_len = 32;
4983 
4984 	test_vmcs_addr_values("VM-exit-MSR-store address",
4985 				EXIT_MSR_ST_ADDR, 16, false, false,
4986 				4, addr_len - 1);
4987 
4988 	/*
4989 	 * Check last byte of VM-exit MSR-store address
4990 	 */
4991 	exit_msr_store = (struct vmx_msr_entry *)((u64)exit_msr_store & ~0xf);
4992 
4993 	for (i = (addr_len == 64 ? cpuid_maxphyaddr(): addr_len);
4994 							i < 64; i++) {
4995 		tmp = ((u64)exit_msr_store + exit_msr_st_cnt * 16 - 1) |
4996 			1ul << i;
4997 		vmcs_write(EXIT_MSR_ST_ADDR,
4998 			   tmp - (exit_msr_st_cnt * 16 - 1));
4999                 test_vmx_controls(false, false);
5000 	}
5001 
5002 	vmcs_write(EXI_MSR_ST_CNT, 2);
5003 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 16);
5004 	test_vmx_controls(false, false);
5005 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 32);
5006 	test_vmx_controls(true, false);
5007 	vmcs_write(EXIT_MSR_ST_ADDR, (1ULL << cpuid_maxphyaddr()) - 48);
5008 	test_vmx_controls(true, false);
5009 }
5010 
5011 /*
5012  * Tests for VM-exit controls
5013  */
5014 static void test_vm_exit_ctls(void)
5015 {
5016 	test_exit_msr_store();
5017 }
5018 
5019 /*
5020  * Check that the virtual CPU checks all of the VMX controls as
5021  * documented in the Intel SDM.
5022  */
5023 static void vmx_controls_test(void)
5024 {
5025 	/*
5026 	 * Bit 1 of the guest's RFLAGS must be 1, or VM-entry will
5027 	 * fail due to invalid guest state, should we make it that
5028 	 * far.
5029 	 */
5030 	vmcs_write(GUEST_RFLAGS, 0);
5031 
5032 	test_vm_execution_ctls();
5033 	test_vm_entry_ctls();
5034 	test_vm_exit_ctls();
5035 }
5036 
5037 static bool valid_vmcs_for_vmentry(void)
5038 {
5039 	struct vmcs *current_vmcs = NULL;
5040 
5041 	if (vmcs_save(&current_vmcs))
5042 		return false;
5043 
5044 	return current_vmcs && !current_vmcs->hdr.shadow_vmcs;
5045 }
5046 
5047 static void try_vmentry_in_movss_shadow(void)
5048 {
5049 	u32 vm_inst_err;
5050 	u32 flags;
5051 	bool early_failure = false;
5052 	u32 expected_flags = X86_EFLAGS_FIXED;
5053 	bool valid_vmcs = valid_vmcs_for_vmentry();
5054 
5055 	expected_flags |= valid_vmcs ? X86_EFLAGS_ZF : X86_EFLAGS_CF;
5056 
5057 	/*
5058 	 * Indirectly set VM_INST_ERR to 12 ("VMREAD/VMWRITE from/to
5059 	 * unsupported VMCS component").
5060 	 */
5061 	vmcs_write(~0u, 0);
5062 
5063 	__asm__ __volatile__ ("mov %[host_rsp], %%edx;"
5064 			      "vmwrite %%rsp, %%rdx;"
5065 			      "mov 0f, %%rax;"
5066 			      "mov %[host_rip], %%edx;"
5067 			      "vmwrite %%rax, %%rdx;"
5068 			      "mov $-1, %%ah;"
5069 			      "sahf;"
5070 			      "mov %%ss, %%ax;"
5071 			      "mov %%ax, %%ss;"
5072 			      "vmlaunch;"
5073 			      "mov $1, %[early_failure];"
5074 			      "0: lahf;"
5075 			      "movzbl %%ah, %[flags]"
5076 			      : [early_failure] "+r" (early_failure),
5077 				[flags] "=&a" (flags)
5078 			      : [host_rsp] "i" (HOST_RSP),
5079 				[host_rip] "i" (HOST_RIP)
5080 			      : "rdx", "cc", "memory");
5081 	vm_inst_err = vmcs_read(VMX_INST_ERROR);
5082 
5083 	report("Early VM-entry failure", early_failure);
5084 	report("RFLAGS[8:0] is %x (actual %x)", flags == expected_flags,
5085 	       expected_flags, flags);
5086 	if (valid_vmcs)
5087 		report("VM-instruction error is %d (actual %d)",
5088 		       vm_inst_err == VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS,
5089 		       VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS, vm_inst_err);
5090 }
5091 
5092 static void vmentry_movss_shadow_test(void)
5093 {
5094 	struct vmcs *orig_vmcs;
5095 
5096 	TEST_ASSERT(!vmcs_save(&orig_vmcs));
5097 
5098 	/*
5099 	 * Set the launched flag on the current VMCS to verify the correct
5100 	 * error priority, below.
5101 	 */
5102 	test_set_guest(v2_null_test_guest);
5103 	enter_guest();
5104 
5105 	/*
5106 	 * With bit 1 of the guest's RFLAGS clear, VM-entry should
5107 	 * fail due to invalid guest state (if we make it that far).
5108 	 */
5109 	vmcs_write(GUEST_RFLAGS, 0);
5110 
5111 	/*
5112 	 * "VM entry with events blocked by MOV SS" takes precedence over
5113 	 * "VMLAUNCH with non-clear VMCS."
5114 	 */
5115 	report_prefix_push("valid current-VMCS");
5116 	try_vmentry_in_movss_shadow();
5117 	report_prefix_pop();
5118 
5119 	/*
5120 	 * VMfailInvalid takes precedence over "VM entry with events
5121 	 * blocked by MOV SS."
5122 	 */
5123 	TEST_ASSERT(!vmcs_clear(orig_vmcs));
5124 	report_prefix_push("no current-VMCS");
5125 	try_vmentry_in_movss_shadow();
5126 	report_prefix_pop();
5127 
5128 	TEST_ASSERT(!make_vmcs_current(orig_vmcs));
5129 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
5130 }
5131 
5132 #define X86_FEATURE_PCID       (1 << 17)
5133 #define X86_FEATURE_MCE        (1 << 7)
5134 
5135 static int write_cr4_checking(unsigned long val)
5136 {
5137 	asm volatile(ASM_TRY("1f")
5138 		     "mov %0, %%cr4\n\t"
5139 		     "1:": : "r" (val));
5140 	return exception_vector();
5141 }
5142 
5143 static void vmx_cr_load_test(void)
5144 {
5145 	struct cpuid _cpuid = cpuid(1);
5146 	unsigned long cr4 = read_cr4(), cr3 = read_cr3();
5147 
5148 	if (!(_cpuid.c & X86_FEATURE_PCID)) {
5149 		report_skip("PCID not detected");
5150 		return;
5151 	}
5152 	if (!(_cpuid.d & X86_FEATURE_MCE)) {
5153 		report_skip("MCE not detected");
5154 		return;
5155 	}
5156 
5157 	TEST_ASSERT(!(cr4 & (X86_CR4_PCIDE | X86_CR4_MCE)));
5158 	TEST_ASSERT(!(cr3 & X86_CR3_PCID_MASK));
5159 
5160 	/* Enable PCID for L1. */
5161 	cr4 |= X86_CR4_PCIDE;
5162 	cr3 |= 0x1;
5163 	TEST_ASSERT(!write_cr4_checking(cr4));
5164 	write_cr3(cr3);
5165 
5166 	test_set_guest(v2_null_test_guest);
5167 	vmcs_write(HOST_CR4, cr4);
5168 	vmcs_write(HOST_CR3, cr3);
5169 	enter_guest();
5170 
5171 	/*
5172 	 * No exception is expected.
5173 	 *
5174 	 * NB. KVM loads the last guest write to CR4 into CR4 read
5175 	 *     shadow. In order to trigger an exit to KVM, we can set a
5176 	 *     bit that was zero in the above CR4 write and is owned by
5177 	 *     KVM. We choose to set CR4.MCE, which shall have no side
5178 	 *     effect because normally no guest MCE (e.g., as the result
5179 	 *     of bad memory) would happen during this test.
5180 	 */
5181 	TEST_ASSERT(!write_cr4_checking(cr4 | X86_CR4_MCE));
5182 
5183 	/* Cleanup L1 state: disable PCID. */
5184 	write_cr3(cr3 & ~X86_CR3_PCID_MASK);
5185 	TEST_ASSERT(!write_cr4_checking(cr4 & ~X86_CR4_PCIDE));
5186 }
5187 
5188 static void vmx_nm_test_guest(void)
5189 {
5190 	write_cr0(read_cr0() | X86_CR0_TS);
5191 	asm volatile("fnop");
5192 }
5193 
5194 static void check_nm_exit(const char *test)
5195 {
5196 	u32 reason = vmcs_read(EXI_REASON);
5197 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
5198 	const u32 expected = INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
5199 		NM_VECTOR;
5200 
5201 	report("%s", reason == VMX_EXC_NMI && intr_info == expected, test);
5202 }
5203 
5204 /*
5205  * This test checks that:
5206  *
5207  * (a) If L2 launches with CR0.TS clear, but later sets CR0.TS, then
5208  *     a subsequent #NM VM-exit is reflected to L1.
5209  *
5210  * (b) If L2 launches with CR0.TS clear and CR0.EM set, then a
5211  *     subsequent #NM VM-exit is reflected to L1.
5212  */
5213 static void vmx_nm_test(void)
5214 {
5215 	unsigned long cr0 = read_cr0();
5216 
5217 	test_set_guest(vmx_nm_test_guest);
5218 
5219 	/*
5220 	 * L1 wants to intercept #NM exceptions encountered in L2.
5221 	 */
5222 	vmcs_write(EXC_BITMAP, 1 << NM_VECTOR);
5223 
5224 	/*
5225 	 * Launch L2 with CR0.TS clear, but don't claim host ownership of
5226 	 * any CR0 bits. L2 will set CR0.TS and then try to execute fnop,
5227 	 * which will raise #NM. L0 should reflect the #NM VM-exit to L1.
5228 	 */
5229 	vmcs_write(CR0_MASK, 0);
5230 	vmcs_write(GUEST_CR0, cr0 & ~X86_CR0_TS);
5231 	enter_guest();
5232 	check_nm_exit("fnop with CR0.TS set in L2 triggers #NM VM-exit to L1");
5233 
5234 	/*
5235 	 * Re-enter L2 at the fnop instruction, with CR0.TS clear but
5236 	 * CR0.EM set. The fnop will still raise #NM, and L0 should
5237 	 * reflect the #NM VM-exit to L1.
5238 	 */
5239 	vmcs_write(GUEST_CR0, (cr0 & ~X86_CR0_TS) | X86_CR0_EM);
5240 	enter_guest();
5241 	check_nm_exit("fnop with CR0.EM set in L2 triggers #NM VM-exit to L1");
5242 
5243 	/*
5244 	 * Re-enter L2 at the fnop instruction, with both CR0.TS and
5245 	 * CR0.EM clear. There will be no #NM, and the L2 guest should
5246 	 * exit normally.
5247 	 */
5248 	vmcs_write(GUEST_CR0, cr0 & ~(X86_CR0_TS | X86_CR0_EM));
5249 	enter_guest();
5250 }
5251 
5252 bool vmx_pending_event_ipi_fired;
5253 static void vmx_pending_event_ipi_isr(isr_regs_t *regs)
5254 {
5255 	vmx_pending_event_ipi_fired = true;
5256 	eoi();
5257 }
5258 
5259 bool vmx_pending_event_guest_run;
5260 static void vmx_pending_event_guest(void)
5261 {
5262 	vmcall();
5263 	vmx_pending_event_guest_run = true;
5264 }
5265 
5266 static void vmx_pending_event_test_core(bool guest_hlt)
5267 {
5268 	int ipi_vector = 0xf1;
5269 
5270 	vmx_pending_event_ipi_fired = false;
5271 	handle_irq(ipi_vector, vmx_pending_event_ipi_isr);
5272 
5273 	vmx_pending_event_guest_run = false;
5274 	test_set_guest(vmx_pending_event_guest);
5275 
5276 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
5277 
5278 	enter_guest();
5279 	skip_exit_vmcall();
5280 
5281 	if (guest_hlt)
5282 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
5283 
5284 	irq_disable();
5285 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
5286 				   APIC_DM_FIXED | ipi_vector,
5287 				   0);
5288 
5289 	enter_guest();
5290 
5291 	assert_exit_reason(VMX_EXTINT);
5292 	report("Guest did not run before host received IPI",
5293 		   !vmx_pending_event_guest_run);
5294 
5295 	irq_enable();
5296 	asm volatile ("nop");
5297 	irq_disable();
5298 	report("Got pending interrupt after IRQ enabled",
5299 		   vmx_pending_event_ipi_fired);
5300 
5301 	if (guest_hlt)
5302 		vmcs_write(GUEST_ACTV_STATE, ACTV_ACTIVE);
5303 
5304 	enter_guest();
5305 	report("Guest finished running when no interrupt",
5306 		   vmx_pending_event_guest_run);
5307 }
5308 
5309 static void vmx_pending_event_test(void)
5310 {
5311 	vmx_pending_event_test_core(false);
5312 }
5313 
5314 static void vmx_pending_event_hlt_test(void)
5315 {
5316 	vmx_pending_event_test_core(true);
5317 }
5318 
5319 static int vmx_window_test_ud_count;
5320 
5321 static void vmx_window_test_ud_handler(struct ex_regs *regs)
5322 {
5323 	vmx_window_test_ud_count++;
5324 }
5325 
5326 static void vmx_nmi_window_test_guest(void)
5327 {
5328 	handle_exception(UD_VECTOR, vmx_window_test_ud_handler);
5329 
5330 	asm volatile("vmcall\n\t"
5331 		     "nop\n\t");
5332 
5333 	handle_exception(UD_VECTOR, NULL);
5334 }
5335 
5336 static void verify_nmi_window_exit(u64 rip)
5337 {
5338 	u32 exit_reason = vmcs_read(EXI_REASON);
5339 
5340 	report("Exit reason (%d) is 'NMI window'",
5341 	       exit_reason == VMX_NMI_WINDOW, exit_reason);
5342 	report("RIP (%#lx) is %#lx", vmcs_read(GUEST_RIP) == rip,
5343 	       vmcs_read(GUEST_RIP), rip);
5344 	report("Activity state (%ld) is 'ACTIVE'",
5345 	       vmcs_read(GUEST_ACTV_STATE) == ACTV_ACTIVE,
5346 	       vmcs_read(GUEST_ACTV_STATE));
5347 }
5348 
5349 static void vmx_nmi_window_test(void)
5350 {
5351 	u64 nop_addr;
5352 	void *ud_fault_addr = get_idt_addr(&boot_idt[UD_VECTOR]);
5353 
5354 	if (!(ctrl_pin_rev.clr & PIN_VIRT_NMI)) {
5355 		report_skip("CPU does not support the \"Virtual NMIs\" VM-execution control.");
5356 		return;
5357 	}
5358 
5359 	if (!(ctrl_cpu_rev[0].clr & CPU_NMI_WINDOW)) {
5360 		report_skip("CPU does not support the \"NMI-window exiting\" VM-execution control.");
5361 		return;
5362 	}
5363 
5364 	vmx_window_test_ud_count = 0;
5365 
5366 	report_prefix_push("NMI-window");
5367 	test_set_guest(vmx_nmi_window_test_guest);
5368 	vmcs_set_bits(PIN_CONTROLS, PIN_VIRT_NMI);
5369 	enter_guest();
5370 	skip_exit_vmcall();
5371 	nop_addr = vmcs_read(GUEST_RIP);
5372 
5373 	/*
5374 	 * Ask for "NMI-window exiting," and expect an immediate VM-exit.
5375 	 * RIP will not advance.
5376 	 */
5377 	report_prefix_push("active, no blocking");
5378 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
5379 	enter_guest();
5380 	verify_nmi_window_exit(nop_addr);
5381 	report_prefix_pop();
5382 
5383 	/*
5384 	 * Ask for "NMI-window exiting" in a MOV-SS shadow, and expect
5385 	 * a VM-exit on the next instruction after the nop. (The nop
5386 	 * is one byte.)
5387 	 */
5388 	report_prefix_push("active, blocking by MOV-SS");
5389 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
5390 	enter_guest();
5391 	verify_nmi_window_exit(nop_addr + 1);
5392 	report_prefix_pop();
5393 
5394 	/*
5395 	 * Ask for "NMI-window exiting" (with event injection), and
5396 	 * expect a VM-exit after the event is injected. (RIP should
5397 	 * be at the address specified in the IDT entry for #UD.)
5398 	 */
5399 	report_prefix_push("active, no blocking, injecting #UD");
5400 	vmcs_write(ENT_INTR_INFO,
5401 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | UD_VECTOR);
5402 	enter_guest();
5403 	verify_nmi_window_exit((u64)ud_fault_addr);
5404 	report_prefix_pop();
5405 
5406 	/*
5407 	 * Ask for "NMI-window exiting" with NMI blocking, and expect
5408 	 * a VM-exit after the next IRET (i.e. after the #UD handler
5409 	 * returns). So, RIP should be back at one byte past the nop.
5410 	 */
5411 	report_prefix_push("active, blocking by NMI");
5412 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_NMI);
5413 	enter_guest();
5414 	verify_nmi_window_exit(nop_addr + 1);
5415 	report("#UD handler executed once (actual %d times)",
5416 	       vmx_window_test_ud_count == 1,
5417 	       vmx_window_test_ud_count);
5418 	report_prefix_pop();
5419 
5420 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
5421 		report_skip("CPU does not support activity state HLT.");
5422 	} else {
5423 		/*
5424 		 * Ask for "NMI-window exiting" when entering activity
5425 		 * state HLT, and expect an immediate VM-exit. RIP is
5426 		 * still one byte past the nop.
5427 		 */
5428 		report_prefix_push("halted, no blocking");
5429 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
5430 		enter_guest();
5431 		verify_nmi_window_exit(nop_addr + 1);
5432 		report_prefix_pop();
5433 
5434 		/*
5435 		 * Ask for "NMI-window exiting" when entering activity
5436 		 * state HLT (with event injection), and expect a
5437 		 * VM-exit after the event is injected. (RIP should be
5438 		 * at the address specified in the IDT entry for #UD.)
5439 		 */
5440 		report_prefix_push("halted, no blocking, injecting #UD");
5441 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
5442 		vmcs_write(ENT_INTR_INFO,
5443 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
5444 			   UD_VECTOR);
5445 		enter_guest();
5446 		verify_nmi_window_exit((u64)ud_fault_addr);
5447 		report_prefix_pop();
5448 	}
5449 
5450 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_NMI_WINDOW);
5451 	enter_guest();
5452 	report_prefix_pop();
5453 }
5454 
5455 static void vmx_intr_window_test_guest(void)
5456 {
5457 	handle_exception(UD_VECTOR, vmx_window_test_ud_handler);
5458 
5459 	/*
5460 	 * The two consecutive STIs are to ensure that only the first
5461 	 * one has a shadow. Note that NOP and STI are one byte
5462 	 * instructions.
5463 	 */
5464 	asm volatile("vmcall\n\t"
5465 		     "nop\n\t"
5466 		     "sti\n\t"
5467 		     "sti\n\t");
5468 
5469 	handle_exception(UD_VECTOR, NULL);
5470 }
5471 
5472 static void verify_intr_window_exit(u64 rip)
5473 {
5474 	u32 exit_reason = vmcs_read(EXI_REASON);
5475 
5476 	report("Exit reason (%d) is 'interrupt window'",
5477 	       exit_reason == VMX_INTR_WINDOW, exit_reason);
5478 	report("RIP (%#lx) is %#lx", vmcs_read(GUEST_RIP) == rip,
5479 	       vmcs_read(GUEST_RIP), rip);
5480 	report("Activity state (%ld) is 'ACTIVE'",
5481 	       vmcs_read(GUEST_ACTV_STATE) == ACTV_ACTIVE,
5482 	       vmcs_read(GUEST_ACTV_STATE));
5483 }
5484 
5485 static void vmx_intr_window_test(void)
5486 {
5487 	u64 vmcall_addr;
5488 	u64 nop_addr;
5489 	unsigned int orig_ud_gate_type;
5490 	void *ud_fault_addr = get_idt_addr(&boot_idt[UD_VECTOR]);
5491 
5492 	if (!(ctrl_cpu_rev[0].clr & CPU_INTR_WINDOW)) {
5493 		report_skip("CPU does not support the \"interrupt-window exiting\" VM-execution control.");
5494 		return;
5495 	}
5496 
5497 	/*
5498 	 * Change the IDT entry for #UD from interrupt gate to trap gate,
5499 	 * so that it won't clear RFLAGS.IF. We don't want interrupts to
5500 	 * be disabled after vectoring a #UD.
5501 	 */
5502 	orig_ud_gate_type = boot_idt[UD_VECTOR].type;
5503 	boot_idt[UD_VECTOR].type = 15;
5504 
5505 	report_prefix_push("interrupt-window");
5506 	test_set_guest(vmx_intr_window_test_guest);
5507 	enter_guest();
5508 	assert_exit_reason(VMX_VMCALL);
5509 	vmcall_addr = vmcs_read(GUEST_RIP);
5510 
5511 	/*
5512 	 * Ask for "interrupt-window exiting" with RFLAGS.IF set and
5513 	 * no blocking; expect an immediate VM-exit. Note that we have
5514 	 * not advanced past the vmcall instruction yet, so RIP should
5515 	 * point to the vmcall instruction.
5516 	 */
5517 	report_prefix_push("active, no blocking, RFLAGS.IF=1");
5518 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
5519 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_IF);
5520 	enter_guest();
5521 	verify_intr_window_exit(vmcall_addr);
5522 	report_prefix_pop();
5523 
5524 	/*
5525 	 * Ask for "interrupt-window exiting" (with event injection)
5526 	 * with RFLAGS.IF set and no blocking; expect a VM-exit after
5527 	 * the event is injected. That is, RIP should should be at the
5528 	 * address specified in the IDT entry for #UD.
5529 	 */
5530 	report_prefix_push("active, no blocking, RFLAGS.IF=1, injecting #UD");
5531 	vmcs_write(ENT_INTR_INFO,
5532 		   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION | UD_VECTOR);
5533 	vmcall_addr = vmcs_read(GUEST_RIP);
5534 	enter_guest();
5535 	verify_intr_window_exit((u64)ud_fault_addr);
5536 	report_prefix_pop();
5537 
5538 	/*
5539 	 * Let the L2 guest run through the IRET, back to the VMCALL.
5540 	 * We have to clear the "interrupt-window exiting"
5541 	 * VM-execution control, or it would just keep causing
5542 	 * VM-exits. Then, advance past the VMCALL and set the
5543 	 * "interrupt-window exiting" VM-execution control again.
5544 	 */
5545 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
5546 	enter_guest();
5547 	skip_exit_vmcall();
5548 	nop_addr = vmcs_read(GUEST_RIP);
5549 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
5550 
5551 	/*
5552 	 * Ask for "interrupt-window exiting" in a MOV-SS shadow with
5553 	 * RFLAGS.IF set, and expect a VM-exit on the next
5554 	 * instruction. (NOP is one byte.)
5555 	 */
5556 	report_prefix_push("active, blocking by MOV-SS, RFLAGS.IF=1");
5557 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
5558 	enter_guest();
5559 	verify_intr_window_exit(nop_addr + 1);
5560 	report_prefix_pop();
5561 
5562 	/*
5563 	 * Back up to the NOP and ask for "interrupt-window exiting"
5564 	 * in an STI shadow with RFLAGS.IF set, and expect a VM-exit
5565 	 * on the next instruction. (NOP is one byte.)
5566 	 */
5567 	report_prefix_push("active, blocking by STI, RFLAGS.IF=1");
5568 	vmcs_write(GUEST_RIP, nop_addr);
5569 	vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_STI);
5570 	enter_guest();
5571 	verify_intr_window_exit(nop_addr + 1);
5572 	report_prefix_pop();
5573 
5574 	/*
5575 	 * Ask for "interrupt-window exiting" with RFLAGS.IF clear,
5576 	 * and expect a VM-exit on the instruction following the STI
5577 	 * shadow. Only the first STI (which is one byte past the NOP)
5578 	 * should have a shadow. The second STI (which is two bytes
5579 	 * past the NOP) has no shadow. Therefore, the interrupt
5580 	 * window opens at three bytes past the NOP.
5581 	 */
5582 	report_prefix_push("active, RFLAGS.IF = 0");
5583 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
5584 	enter_guest();
5585 	verify_intr_window_exit(nop_addr + 3);
5586 	report_prefix_pop();
5587 
5588 	if (!(rdmsr(MSR_IA32_VMX_MISC) & (1 << 6))) {
5589 		report_skip("CPU does not support activity state HLT.");
5590 	} else {
5591 		/*
5592 		 * Ask for "interrupt-window exiting" when entering
5593 		 * activity state HLT, and expect an immediate
5594 		 * VM-exit. RIP is still three bytes past the nop.
5595 		 */
5596 		report_prefix_push("halted, no blocking");
5597 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
5598 		enter_guest();
5599 		verify_intr_window_exit(nop_addr + 3);
5600 		report_prefix_pop();
5601 
5602 		/*
5603 		 * Ask for "interrupt-window exiting" when entering
5604 		 * activity state HLT (with event injection), and
5605 		 * expect a VM-exit after the event is injected. That
5606 		 * is, RIP should should be at the address specified
5607 		 * in the IDT entry for #UD.
5608 		 */
5609 		report_prefix_push("halted, no blocking, injecting #UD");
5610 		vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
5611 		vmcs_write(ENT_INTR_INFO,
5612 			   INTR_INFO_VALID_MASK | INTR_TYPE_HARD_EXCEPTION |
5613 			   UD_VECTOR);
5614 		enter_guest();
5615 		verify_intr_window_exit((u64)ud_fault_addr);
5616 		report_prefix_pop();
5617 	}
5618 
5619 	boot_idt[UD_VECTOR].type = orig_ud_gate_type;
5620 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_INTR_WINDOW);
5621 	enter_guest();
5622 	report_prefix_pop();
5623 }
5624 
5625 #define GUEST_TSC_OFFSET (1u << 30)
5626 
5627 static u64 guest_tsc;
5628 
5629 static void vmx_store_tsc_test_guest(void)
5630 {
5631 	guest_tsc = rdtsc();
5632 }
5633 
5634 /*
5635  * This test ensures that when IA32_TSC is in the VM-exit MSR-store
5636  * list, the value saved is not subject to the TSC offset that is
5637  * applied to RDTSC/RDTSCP/RDMSR(IA32_TSC) in guest execution.
5638  */
5639 static void vmx_store_tsc_test(void)
5640 {
5641 	struct vmx_msr_entry msr_entry = { .index = MSR_IA32_TSC };
5642 	u64 low, high;
5643 
5644 	if (!(ctrl_cpu_rev[0].clr & CPU_USE_TSC_OFFSET)) {
5645 		report_skip("'Use TSC offsetting' not supported");
5646 		return;
5647 	}
5648 
5649 	test_set_guest(vmx_store_tsc_test_guest);
5650 
5651 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_USE_TSC_OFFSET);
5652 	vmcs_write(EXI_MSR_ST_CNT, 1);
5653 	vmcs_write(EXIT_MSR_ST_ADDR, virt_to_phys(&msr_entry));
5654 	vmcs_write(TSC_OFFSET, GUEST_TSC_OFFSET);
5655 
5656 	low = rdtsc();
5657 	enter_guest();
5658 	high = rdtsc();
5659 
5660 	report("RDTSC value in the guest (%lu) is in range [%lu, %lu]",
5661 	       low + GUEST_TSC_OFFSET <= guest_tsc &&
5662 	       guest_tsc <= high + GUEST_TSC_OFFSET,
5663 	       guest_tsc, low + GUEST_TSC_OFFSET, high + GUEST_TSC_OFFSET);
5664 	report("IA32_TSC value saved in the VM-exit MSR-store list (%lu) is in range [%lu, %lu]",
5665 	       low <= msr_entry.value && msr_entry.value <= high,
5666 	       msr_entry.value, low, high);
5667 }
5668 
5669 static void vmx_db_test_guest(void)
5670 {
5671 	/*
5672 	 * For a hardware generated single-step #DB.
5673 	 */
5674 	asm volatile("vmcall;"
5675 		     "nop;"
5676 		     ".Lpost_nop:");
5677 	/*
5678 	 * ...in a MOVSS shadow, with pending debug exceptions.
5679 	 */
5680 	asm volatile("vmcall;"
5681 		     "nop;"
5682 		     ".Lpost_movss_nop:");
5683 	/*
5684 	 * For an L0 synthesized single-step #DB. (L0 intercepts WBINVD and
5685 	 * emulates it in software.)
5686 	 */
5687 	asm volatile("vmcall;"
5688 		     "wbinvd;"
5689 		     ".Lpost_wbinvd:");
5690 	/*
5691 	 * ...in a MOVSS shadow, with pending debug exceptions.
5692 	 */
5693 	asm volatile("vmcall;"
5694 		     "wbinvd;"
5695 		     ".Lpost_movss_wbinvd:");
5696 	/*
5697 	 * For a hardware generated single-step #DB in a transactional region.
5698 	 */
5699 	asm volatile("vmcall;"
5700 		     ".Lxbegin: xbegin .Lskip_rtm;"
5701 		     "xend;"
5702 		     ".Lskip_rtm:");
5703 }
5704 
5705 /*
5706  * Clear the pending debug exceptions and RFLAGS.TF and re-enter
5707  * L2. No #DB is delivered and L2 continues to the next point of
5708  * interest.
5709  */
5710 static void dismiss_db(void)
5711 {
5712 	vmcs_write(GUEST_PENDING_DEBUG, 0);
5713 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED);
5714 	enter_guest();
5715 }
5716 
5717 /*
5718  * Check a variety of VMCS fields relevant to an intercepted #DB exception.
5719  * Then throw away the #DB exception and resume L2.
5720  */
5721 static void check_db_exit(bool xfail_qual, bool xfail_dr6, bool xfail_pdbg,
5722 			  void *expected_rip, u64 expected_exit_qual,
5723 			  u64 expected_dr6)
5724 {
5725 	u32 reason = vmcs_read(EXI_REASON);
5726 	u32 intr_info = vmcs_read(EXI_INTR_INFO);
5727 	u64 exit_qual = vmcs_read(EXI_QUALIFICATION);
5728 	u64 guest_rip = vmcs_read(GUEST_RIP);
5729 	u64 guest_pending_dbg = vmcs_read(GUEST_PENDING_DEBUG);
5730 	u64 dr6 = read_dr6();
5731 	const u32 expected_intr_info = INTR_INFO_VALID_MASK |
5732 		INTR_TYPE_HARD_EXCEPTION | DB_VECTOR;
5733 
5734 	report("Expected #DB VM-exit",
5735 	       reason == VMX_EXC_NMI && intr_info == expected_intr_info);
5736 	report("Expected RIP %p (actual %lx)", (u64)expected_rip == guest_rip,
5737 	       expected_rip, guest_rip);
5738 	report_xfail("Expected pending debug exceptions 0 (actual %lx)",
5739 		     xfail_pdbg, 0 == guest_pending_dbg, guest_pending_dbg);
5740 	report_xfail("Expected exit qualification %lx (actual %lx)", xfail_qual,
5741 		     expected_exit_qual == exit_qual,
5742 		     expected_exit_qual, exit_qual);
5743 	report_xfail("Expected DR6 %lx (actual %lx)", xfail_dr6,
5744 		     expected_dr6 == dr6, expected_dr6, dr6);
5745 	dismiss_db();
5746 }
5747 
5748 /*
5749  * Assuming the guest has just exited on a VMCALL instruction, skip
5750  * over the vmcall, and set the guest's RFLAGS.TF in the VMCS. If
5751  * pending debug exceptions are non-zero, set the VMCS up as if the
5752  * previous instruction was a MOVSS that generated the indicated
5753  * pending debug exceptions. Then enter L2.
5754  */
5755 static void single_step_guest(const char *test_name, u64 starting_dr6,
5756 			      u64 pending_debug_exceptions)
5757 {
5758 	printf("\n%s\n", test_name);
5759 	skip_exit_vmcall();
5760 	write_dr6(starting_dr6);
5761 	vmcs_write(GUEST_RFLAGS, X86_EFLAGS_FIXED | X86_EFLAGS_TF);
5762 	if (pending_debug_exceptions) {
5763 		vmcs_write(GUEST_PENDING_DEBUG, pending_debug_exceptions);
5764 		vmcs_write(GUEST_INTR_STATE, GUEST_INTR_STATE_MOVSS);
5765 	}
5766 	enter_guest();
5767 }
5768 
5769 /*
5770  * When L1 intercepts #DB, verify that a single-step trap clears
5771  * pending debug exceptions, populates the exit qualification field
5772  * properly, and that DR6 is not prematurely clobbered. In a
5773  * (simulated) MOVSS shadow, make sure that the pending debug
5774  * exception bits are properly accumulated into the exit qualification
5775  * field.
5776  */
5777 static void vmx_db_test(void)
5778 {
5779 	/*
5780 	 * We are going to set a few arbitrary bits in DR6 to verify that
5781 	 * (a) DR6 is not modified by an intercepted #DB, and
5782 	 * (b) stale bits in DR6 (DR6.BD, in particular) don't leak into
5783          *     the exit qualification field for a subsequent #DB exception.
5784 	 */
5785 	const u64 starting_dr6 = DR6_RESERVED | BIT(13) | DR_TRAP3 | DR_TRAP1;
5786 	extern char post_nop asm(".Lpost_nop");
5787 	extern char post_movss_nop asm(".Lpost_movss_nop");
5788 	extern char post_wbinvd asm(".Lpost_wbinvd");
5789 	extern char post_movss_wbinvd asm(".Lpost_movss_wbinvd");
5790 	extern char xbegin asm(".Lxbegin");
5791 	extern char skip_rtm asm(".Lskip_rtm");
5792 
5793 	/*
5794 	 * L1 wants to intercept #DB exceptions encountered in L2.
5795 	 */
5796 	vmcs_write(EXC_BITMAP, BIT(DB_VECTOR));
5797 
5798 	/*
5799 	 * Start L2 and run it up to the first point of interest.
5800 	 */
5801 	test_set_guest(vmx_db_test_guest);
5802 	enter_guest();
5803 
5804 	/*
5805 	 * Hardware-delivered #DB trap for single-step sets the
5806 	 * standard that L0 has to follow for emulated instructions.
5807 	 */
5808 	single_step_guest("Hardware delivered single-step", starting_dr6, 0);
5809 	check_db_exit(false, false, false, &post_nop, DR_STEP, starting_dr6);
5810 
5811 	/*
5812 	 * Hardware-delivered #DB trap for single-step in MOVSS shadow
5813 	 * also sets the standard that L0 has to follow for emulated
5814 	 * instructions. Here, we establish the VMCS pending debug
5815 	 * exceptions to indicate that the simulated MOVSS triggered a
5816 	 * data breakpoint as well as the single-step trap.
5817 	 */
5818 	single_step_guest("Hardware delivered single-step in MOVSS shadow",
5819 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0 );
5820 	check_db_exit(false, false, false, &post_movss_nop, DR_STEP | DR_TRAP0,
5821 		      starting_dr6);
5822 
5823 	/*
5824 	 * L0 synthesized #DB trap for single-step is buggy, because
5825 	 * kvm (a) clobbers DR6 too early, and (b) tries its best to
5826 	 * reconstitute the exit qualification from the prematurely
5827 	 * modified DR6, but fails miserably.
5828 	 */
5829 	single_step_guest("Software synthesized single-step", starting_dr6, 0);
5830 	check_db_exit(true, true, false, &post_wbinvd, DR_STEP, starting_dr6);
5831 
5832 	/*
5833 	 * L0 synthesized #DB trap for single-step in MOVSS shadow is
5834 	 * even worse, because L0 also leaves the pending debug
5835 	 * exceptions in the VMCS instead of accumulating them into
5836 	 * the exit qualification field for the #DB exception.
5837 	 */
5838 	single_step_guest("Software synthesized single-step in MOVSS shadow",
5839 			  starting_dr6, BIT(12) | DR_STEP | DR_TRAP0);
5840 	check_db_exit(true, true, true, &post_movss_wbinvd, DR_STEP | DR_TRAP0,
5841 		      starting_dr6);
5842 
5843 	/*
5844 	 * Optional RTM test for hardware that supports RTM, to
5845 	 * demonstrate that the current volume 3 of the SDM
5846 	 * (325384-067US), table 27-1 is incorrect. Bit 16 of the exit
5847 	 * qualification for debug exceptions is not reserved. It is
5848 	 * set to 1 if a debug exception (#DB) or a breakpoint
5849 	 * exception (#BP) occurs inside an RTM region while advanced
5850 	 * debugging of RTM transactional regions is enabled.
5851 	 */
5852 	if (cpuid(7).b & BIT(11)) {
5853 		vmcs_write(ENT_CONTROLS,
5854 			   vmcs_read(ENT_CONTROLS) | ENT_LOAD_DBGCTLS);
5855 		/*
5856 		 * Set DR7.RTM[bit 11] and IA32_DEBUGCTL.RTM[bit 15]
5857 		 * in the guest to enable advanced debugging of RTM
5858 		 * transactional regions.
5859 		 */
5860 		vmcs_write(GUEST_DR7, BIT(11));
5861 		vmcs_write(GUEST_DEBUGCTL, BIT(15));
5862 		single_step_guest("Hardware delivered single-step in "
5863 				  "transactional region", starting_dr6, 0);
5864 		check_db_exit(false, false, false, &xbegin, BIT(16),
5865 			      starting_dr6);
5866 	} else {
5867 		vmcs_write(GUEST_RIP, (u64)&skip_rtm);
5868 		enter_guest();
5869 	}
5870 }
5871 
5872 static bool cpu_has_apicv(void)
5873 {
5874 	return ((ctrl_cpu_rev[1].clr & CPU_APIC_REG_VIRT) &&
5875 		(ctrl_cpu_rev[1].clr & CPU_VINTD) &&
5876 		(ctrl_pin_rev.clr & PIN_POST_INTR));
5877 }
5878 
5879 static void enable_vid(void)
5880 {
5881 	void *virtual_apic_page;
5882 
5883 	assert(cpu_has_apicv());
5884 
5885 	disable_intercept_for_x2apic_msrs();
5886 
5887 	virtual_apic_page = alloc_page();
5888 	vmcs_write(APIC_VIRT_ADDR, (u64)virtual_apic_page);
5889 
5890 	vmcs_set_bits(PIN_CONTROLS, PIN_EXTINT);
5891 
5892 	vmcs_write(EOI_EXIT_BITMAP0, 0x0);
5893 	vmcs_write(EOI_EXIT_BITMAP1, 0x0);
5894 	vmcs_write(EOI_EXIT_BITMAP2, 0x0);
5895 	vmcs_write(EOI_EXIT_BITMAP3, 0x0);
5896 
5897 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY | CPU_TPR_SHADOW);
5898 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_VINTD | CPU_VIRT_X2APIC);
5899 }
5900 
5901 static void trigger_ioapic_scan_thread(void *data)
5902 {
5903 	/* Wait until other CPU entered L2 */
5904 	while (vmx_get_test_stage() != 1)
5905 		;
5906 
5907 	/* Trigger ioapic scan */
5908 	ioapic_set_redir(0xf, 0x79, TRIGGER_LEVEL);
5909 	vmx_set_test_stage(2);
5910 }
5911 
5912 static void irq_79_handler_guest(isr_regs_t *regs)
5913 {
5914 	eoi();
5915 
5916 	/* L1 expects vmexit on VMX_VMCALL and not VMX_EOI_INDUCED */
5917 	vmcall();
5918 }
5919 
5920 /*
5921  * Constant for num of busy-loop iterations after which
5922  * a timer interrupt should have happened in host
5923  */
5924 #define TIMER_INTERRUPT_DELAY 100000000
5925 
5926 static void vmx_eoi_bitmap_ioapic_scan_test_guest(void)
5927 {
5928 	handle_irq(0x79, irq_79_handler_guest);
5929 	irq_enable();
5930 
5931 	/* Signal to L1 CPU to trigger ioapic scan */
5932 	vmx_set_test_stage(1);
5933 	/* Wait until L1 CPU to trigger ioapic scan */
5934 	while (vmx_get_test_stage() != 2)
5935 		;
5936 
5937 	/*
5938 	 * Wait for L0 timer interrupt to be raised while we run in L2
5939 	 * such that L0 will process the IOAPIC scan request before
5940 	 * resuming L2
5941 	 */
5942 	delay(TIMER_INTERRUPT_DELAY);
5943 
5944 	asm volatile ("int $0x79");
5945 }
5946 
5947 static void vmx_eoi_bitmap_ioapic_scan_test(void)
5948 {
5949 	if (!cpu_has_apicv() || (cpu_count() < 2)) {
5950 		report_skip(__func__);
5951 		return;
5952 	}
5953 
5954 	enable_vid();
5955 
5956 	on_cpu_async(1, trigger_ioapic_scan_thread, NULL);
5957 	test_set_guest(vmx_eoi_bitmap_ioapic_scan_test_guest);
5958 
5959 	/*
5960 	 * Launch L2.
5961 	 * We expect the exit reason to be VMX_VMCALL (and not EOI INDUCED).
5962 	 * In case the reason isn't VMX_VMCALL, the asserion inside
5963 	 * skip_exit_vmcall() will fail.
5964 	 */
5965 	enter_guest();
5966 	skip_exit_vmcall();
5967 
5968 	/* Let L2 finish */
5969 	enter_guest();
5970 	report(__func__, 1);
5971 }
5972 
5973 #define HLT_WITH_RVI_VECTOR		(0xf1)
5974 
5975 bool vmx_hlt_with_rvi_guest_isr_fired;
5976 static void vmx_hlt_with_rvi_guest_isr(isr_regs_t *regs)
5977 {
5978 	vmx_hlt_with_rvi_guest_isr_fired = true;
5979 	eoi();
5980 }
5981 
5982 static void vmx_hlt_with_rvi_guest(void)
5983 {
5984 	handle_irq(HLT_WITH_RVI_VECTOR, vmx_hlt_with_rvi_guest_isr);
5985 
5986 	irq_enable();
5987 	asm volatile ("nop");
5988 
5989 	vmcall();
5990 }
5991 
5992 static void vmx_hlt_with_rvi_test(void)
5993 {
5994 	if (!cpu_has_apicv()) {
5995 		report_skip(__func__);
5996 		return;
5997 	}
5998 
5999 	enable_vid();
6000 
6001 	vmx_hlt_with_rvi_guest_isr_fired = false;
6002 	test_set_guest(vmx_hlt_with_rvi_guest);
6003 
6004 	enter_guest();
6005 	skip_exit_vmcall();
6006 
6007 	vmcs_write(GUEST_ACTV_STATE, ACTV_HLT);
6008 	vmcs_write(GUEST_INT_STATUS, HLT_WITH_RVI_VECTOR);
6009 	enter_guest();
6010 
6011 	report("Interrupt raised in guest", vmx_hlt_with_rvi_guest_isr_fired);
6012 }
6013 
6014 static void set_irq_line_thread(void *data)
6015 {
6016 	/* Wait until other CPU entered L2 */
6017 	while (vmx_get_test_stage() != 1)
6018 		;
6019 
6020 	/* Set irq-line 0xf to raise vector 0x78 for vCPU 0 */
6021 	ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
6022 	vmx_set_test_stage(2);
6023 }
6024 
6025 static bool irq_78_handler_vmcall_before_eoi;
6026 static void irq_78_handler_guest(isr_regs_t *regs)
6027 {
6028 	set_irq_line(0xf, 0);
6029 	if (irq_78_handler_vmcall_before_eoi)
6030 		vmcall();
6031 	eoi();
6032 	vmcall();
6033 }
6034 
6035 static void vmx_apic_passthrough_guest(void)
6036 {
6037 	handle_irq(0x78, irq_78_handler_guest);
6038 	irq_enable();
6039 
6040 	/* If requested, wait for other CPU to trigger ioapic scan */
6041 	if (vmx_get_test_stage() < 1) {
6042 		vmx_set_test_stage(1);
6043 		while (vmx_get_test_stage() != 2)
6044 			;
6045 	}
6046 
6047 	set_irq_line(0xf, 1);
6048 }
6049 
6050 static void vmx_apic_passthrough(bool set_irq_line_from_thread)
6051 {
6052 	if (set_irq_line_from_thread && (cpu_count() < 2)) {
6053 		report_skip(__func__);
6054 		return;
6055 	}
6056 
6057 	u64 cpu_ctrl_0 = CPU_SECONDARY;
6058 	u64 cpu_ctrl_1 = 0;
6059 
6060 	disable_intercept_for_x2apic_msrs();
6061 
6062 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
6063 
6064 	vmcs_write(CPU_EXEC_CTRL0, vmcs_read(CPU_EXEC_CTRL0) | cpu_ctrl_0);
6065 	vmcs_write(CPU_EXEC_CTRL1, vmcs_read(CPU_EXEC_CTRL1) | cpu_ctrl_1);
6066 
6067 	if (set_irq_line_from_thread) {
6068 		irq_78_handler_vmcall_before_eoi = false;
6069 		on_cpu_async(1, set_irq_line_thread, NULL);
6070 	} else {
6071 		irq_78_handler_vmcall_before_eoi = true;
6072 		ioapic_set_redir(0xf, 0x78, TRIGGER_LEVEL);
6073 		vmx_set_test_stage(2);
6074 	}
6075 	test_set_guest(vmx_apic_passthrough_guest);
6076 
6077 	if (irq_78_handler_vmcall_before_eoi) {
6078 		/* Before EOI remote_irr should still be set */
6079 		enter_guest();
6080 		skip_exit_vmcall();
6081 		TEST_ASSERT_EQ_MSG(1, (int)ioapic_read_redir(0xf).remote_irr,
6082 			"IOAPIC pass-through: remote_irr=1 before EOI");
6083 	}
6084 
6085 	/* After EOI remote_irr should be cleared */
6086 	enter_guest();
6087 	skip_exit_vmcall();
6088 	TEST_ASSERT_EQ_MSG(0, (int)ioapic_read_redir(0xf).remote_irr,
6089 		"IOAPIC pass-through: remote_irr=0 after EOI");
6090 
6091 	/* Let L2 finish */
6092 	enter_guest();
6093 	report(__func__, 1);
6094 }
6095 
6096 static void vmx_apic_passthrough_test(void)
6097 {
6098 	vmx_apic_passthrough(false);
6099 }
6100 
6101 static void vmx_apic_passthrough_thread_test(void)
6102 {
6103 	vmx_apic_passthrough(true);
6104 }
6105 
6106 enum vmcs_access {
6107 	ACCESS_VMREAD,
6108 	ACCESS_VMWRITE,
6109 	ACCESS_NONE,
6110 };
6111 
6112 struct vmcs_shadow_test_common {
6113 	enum vmcs_access op;
6114 	enum Reason reason;
6115 	u64 field;
6116 	u64 value;
6117 	u64 flags;
6118 	u64 time;
6119 } l1_l2_common;
6120 
6121 static inline u64 vmread_flags(u64 field, u64 *val)
6122 {
6123 	u64 flags;
6124 
6125 	asm volatile ("vmread %2, %1; pushf; pop %0"
6126 		      : "=r" (flags), "=rm" (*val) : "r" (field) : "cc");
6127 	return flags & X86_EFLAGS_ALU;
6128 }
6129 
6130 static inline u64 vmwrite_flags(u64 field, u64 val)
6131 {
6132 	u64 flags;
6133 
6134 	asm volatile ("vmwrite %1, %2; pushf; pop %0"
6135 		      : "=r"(flags) : "rm" (val), "r" (field) : "cc");
6136 	return flags & X86_EFLAGS_ALU;
6137 }
6138 
6139 static void vmx_vmcs_shadow_test_guest(void)
6140 {
6141 	struct vmcs_shadow_test_common *c = &l1_l2_common;
6142 	u64 start;
6143 
6144 	while (c->op != ACCESS_NONE) {
6145 		start = rdtsc();
6146 		switch (c->op) {
6147 		default:
6148 			c->flags = -1ull;
6149 			break;
6150 		case ACCESS_VMREAD:
6151 			c->flags = vmread_flags(c->field, &c->value);
6152 			break;
6153 		case ACCESS_VMWRITE:
6154 			c->flags = vmwrite_flags(c->field, 0);
6155 			break;
6156 		}
6157 		c->time = rdtsc() - start;
6158 		vmcall();
6159 	}
6160 }
6161 
6162 static u64 vmread_from_shadow(u64 field)
6163 {
6164 	struct vmcs *primary;
6165 	struct vmcs *shadow;
6166 	u64 value;
6167 
6168 	TEST_ASSERT(!vmcs_save(&primary));
6169 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
6170 	TEST_ASSERT(!make_vmcs_current(shadow));
6171 	value = vmcs_read(field);
6172 	TEST_ASSERT(!make_vmcs_current(primary));
6173 	return value;
6174 }
6175 
6176 static u64 vmwrite_to_shadow(u64 field, u64 value)
6177 {
6178 	struct vmcs *primary;
6179 	struct vmcs *shadow;
6180 
6181 	TEST_ASSERT(!vmcs_save(&primary));
6182 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
6183 	TEST_ASSERT(!make_vmcs_current(shadow));
6184 	vmcs_write(field, value);
6185 	value = vmcs_read(field);
6186 	TEST_ASSERT(!make_vmcs_current(primary));
6187 	return value;
6188 }
6189 
6190 static void vmcs_shadow_test_access(u8 *bitmap[2], enum vmcs_access access)
6191 {
6192 	struct vmcs_shadow_test_common *c = &l1_l2_common;
6193 
6194 	c->op = access;
6195 	vmcs_write(VMX_INST_ERROR, 0);
6196 	enter_guest();
6197 	c->reason = vmcs_read(EXI_REASON) & 0xffff;
6198 	if (c->reason != VMX_VMCALL) {
6199 		skip_exit_insn();
6200 		enter_guest();
6201 	}
6202 	skip_exit_vmcall();
6203 }
6204 
6205 static void vmcs_shadow_test_field(u8 *bitmap[2], u64 field)
6206 {
6207 	struct vmcs_shadow_test_common *c = &l1_l2_common;
6208 	struct vmcs *shadow;
6209 	u64 value;
6210 	uintptr_t flags[2];
6211 	bool good_shadow;
6212 	u32 vmx_inst_error;
6213 
6214 	report_prefix_pushf("field %lx", field);
6215 	c->field = field;
6216 
6217 	shadow = (struct vmcs *)vmcs_read(VMCS_LINK_PTR);
6218 	if (shadow != (struct vmcs *)-1ull) {
6219 		flags[ACCESS_VMREAD] = vmread_flags(field, &value);
6220 		flags[ACCESS_VMWRITE] = vmwrite_flags(field, value);
6221 		good_shadow = !flags[ACCESS_VMREAD] && !flags[ACCESS_VMWRITE];
6222 	} else {
6223 		/*
6224 		 * When VMCS link pointer is -1ull, VMWRITE/VMREAD on
6225 		 * shadowed-fields should fail with setting RFLAGS.CF.
6226 		 */
6227 		flags[ACCESS_VMREAD] = X86_EFLAGS_CF;
6228 		flags[ACCESS_VMWRITE] = X86_EFLAGS_CF;
6229 		good_shadow = false;
6230 	}
6231 
6232 	/* Intercept both VMREAD and VMWRITE. */
6233 	report_prefix_push("no VMREAD/VMWRITE permission");
6234 	/* VMWRITE/VMREAD done on reserved-bit should always intercept */
6235 	if (!(field >> VMCS_FIELD_RESERVED_SHIFT)) {
6236 		set_bit(field, bitmap[ACCESS_VMREAD]);
6237 		set_bit(field, bitmap[ACCESS_VMWRITE]);
6238 	}
6239 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
6240 	report("not shadowed for VMWRITE", c->reason == VMX_VMWRITE);
6241 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
6242 	report("not shadowed for VMREAD", c->reason == VMX_VMREAD);
6243 	report_prefix_pop();
6244 
6245 	if (field >> VMCS_FIELD_RESERVED_SHIFT)
6246 		goto out;
6247 
6248 	/* Permit shadowed VMREAD. */
6249 	report_prefix_push("VMREAD permission only");
6250 	clear_bit(field, bitmap[ACCESS_VMREAD]);
6251 	set_bit(field, bitmap[ACCESS_VMWRITE]);
6252 	if (good_shadow)
6253 		value = vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
6254 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
6255 	report("not shadowed for VMWRITE", c->reason == VMX_VMWRITE);
6256 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
6257 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
6258 	report("shadowed for VMREAD (in %ld cycles)", c->reason == VMX_VMCALL,
6259 	       c->time);
6260 	report("ALU flags after VMREAD (%lx) are as expected (%lx)",
6261 	       c->flags == flags[ACCESS_VMREAD],
6262 	       c->flags, flags[ACCESS_VMREAD]);
6263 	if (good_shadow)
6264 		report("value read from shadow (%lx) is as expected (%lx)",
6265 		       c->value == value, c->value, value);
6266 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
6267 		report("VMX_INST_ERROR (%d) is as expected (%d)",
6268 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
6269 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6270 	report_prefix_pop();
6271 
6272 	/* Permit shadowed VMWRITE. */
6273 	report_prefix_push("VMWRITE permission only");
6274 	set_bit(field, bitmap[ACCESS_VMREAD]);
6275 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
6276 	if (good_shadow)
6277 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
6278 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
6279 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
6280 	report("shadowed for VMWRITE (in %ld cycles)", c->reason == VMX_VMCALL,
6281 		c->time);
6282 	report("ALU flags after VMWRITE (%lx) are as expected (%lx)",
6283 	       c->flags == flags[ACCESS_VMREAD],
6284 	       c->flags, flags[ACCESS_VMREAD]);
6285 	if (good_shadow) {
6286 		value = vmread_from_shadow(field);
6287 		report("shadow VMCS value (%lx) is as expected (%lx)",
6288 		       value == 0, value, 0ul);
6289 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
6290 		report("VMX_INST_ERROR (%d) is as expected (%d)",
6291 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
6292 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6293 	}
6294 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
6295 	report("not shadowed for VMREAD", c->reason == VMX_VMREAD);
6296 	report_prefix_pop();
6297 
6298 	/* Permit shadowed VMREAD and VMWRITE. */
6299 	report_prefix_push("VMREAD and VMWRITE permission");
6300 	clear_bit(field, bitmap[ACCESS_VMREAD]);
6301 	clear_bit(field, bitmap[ACCESS_VMWRITE]);
6302 	if (good_shadow)
6303 		vmwrite_to_shadow(field, MAGIC_VAL_1 + field);
6304 	vmcs_shadow_test_access(bitmap, ACCESS_VMWRITE);
6305 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
6306 	report("shadowed for VMWRITE (in %ld cycles)", c->reason == VMX_VMCALL,
6307 		c->time);
6308 	report("ALU flags after VMWRITE (%lx) are as expected (%lx)",
6309 	       c->flags == flags[ACCESS_VMREAD],
6310 	       c->flags, flags[ACCESS_VMREAD]);
6311 	if (good_shadow) {
6312 		value = vmread_from_shadow(field);
6313 		report("shadow VMCS value (%lx) is as expected (%lx)",
6314 		       value == 0, value, 0ul);
6315 	} else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMWRITE]) {
6316 		report("VMX_INST_ERROR (%d) is as expected (%d)",
6317 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
6318 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6319 	}
6320 	vmcs_shadow_test_access(bitmap, ACCESS_VMREAD);
6321 	vmx_inst_error = vmcs_read(VMX_INST_ERROR);
6322 	report("shadowed for VMREAD (in %ld cycles)", c->reason == VMX_VMCALL,
6323 	       c->time);
6324 	report("ALU flags after VMREAD (%lx) are as expected (%lx)",
6325 	       c->flags == flags[ACCESS_VMREAD],
6326 	       c->flags, flags[ACCESS_VMREAD]);
6327 	if (good_shadow)
6328 		report("value read from shadow (%lx) is as expected (%lx)",
6329 		       c->value == 0, c->value, 0ul);
6330 	else if (shadow != (struct vmcs *)-1ull && flags[ACCESS_VMREAD])
6331 		report("VMX_INST_ERROR (%d) is as expected (%d)",
6332 		       vmx_inst_error == VMXERR_UNSUPPORTED_VMCS_COMPONENT,
6333 		       vmx_inst_error, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6334 	report_prefix_pop();
6335 
6336 out:
6337 	report_prefix_pop();
6338 }
6339 
6340 static void vmx_vmcs_shadow_test_body(u8 *bitmap[2])
6341 {
6342 	unsigned base;
6343 	unsigned index;
6344 	unsigned bit;
6345 	unsigned highest_index = rdmsr(MSR_IA32_VMX_VMCS_ENUM);
6346 
6347 	/* Run test on all possible valid VMCS fields */
6348 	for (base = 0;
6349 	     base < (1 << VMCS_FIELD_RESERVED_SHIFT);
6350 	     base += (1 << VMCS_FIELD_TYPE_SHIFT))
6351 		for (index = 0; index <= highest_index; index++)
6352 			vmcs_shadow_test_field(bitmap, base + index);
6353 
6354 	/*
6355 	 * Run tests on some invalid VMCS fields
6356 	 * (Have reserved bit set).
6357 	 */
6358 	for (bit = VMCS_FIELD_RESERVED_SHIFT; bit < VMCS_FIELD_BIT_SIZE; bit++)
6359 		vmcs_shadow_test_field(bitmap, (1ull << bit));
6360 }
6361 
6362 static void vmx_vmcs_shadow_test(void)
6363 {
6364 	u8 *bitmap[2];
6365 	struct vmcs *shadow;
6366 
6367 	if (!(ctrl_cpu_rev[0].clr & CPU_SECONDARY)) {
6368 		printf("\t'Activate secondary controls' not supported.\n");
6369 		return;
6370 	}
6371 
6372 	if (!(ctrl_cpu_rev[1].clr & CPU_SHADOW_VMCS)) {
6373 		printf("\t'VMCS shadowing' not supported.\n");
6374 		return;
6375 	}
6376 
6377 	if (!(rdmsr(MSR_IA32_VMX_MISC) &
6378 	      MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS)) {
6379 		printf("\tVMWRITE can't modify VM-exit information fields.\n");
6380 		return;
6381 	}
6382 
6383 	test_set_guest(vmx_vmcs_shadow_test_guest);
6384 
6385 	bitmap[ACCESS_VMREAD] = alloc_page();
6386 	bitmap[ACCESS_VMWRITE] = alloc_page();
6387 
6388 	vmcs_write(VMREAD_BITMAP, virt_to_phys(bitmap[ACCESS_VMREAD]));
6389 	vmcs_write(VMWRITE_BITMAP, virt_to_phys(bitmap[ACCESS_VMWRITE]));
6390 
6391 	shadow = alloc_page();
6392 	shadow->hdr.revision_id = basic.revision;
6393 	shadow->hdr.shadow_vmcs = 1;
6394 	TEST_ASSERT(!vmcs_clear(shadow));
6395 
6396 	vmcs_clear_bits(CPU_EXEC_CTRL0, CPU_RDTSC);
6397 	vmcs_set_bits(CPU_EXEC_CTRL0, CPU_SECONDARY);
6398 	vmcs_set_bits(CPU_EXEC_CTRL1, CPU_SHADOW_VMCS);
6399 
6400 	vmcs_write(VMCS_LINK_PTR, virt_to_phys(shadow));
6401 	report_prefix_push("valid link pointer");
6402 	vmx_vmcs_shadow_test_body(bitmap);
6403 	report_prefix_pop();
6404 
6405 	vmcs_write(VMCS_LINK_PTR, -1ull);
6406 	report_prefix_push("invalid link pointer");
6407 	vmx_vmcs_shadow_test_body(bitmap);
6408 	report_prefix_pop();
6409 
6410 	l1_l2_common.op = ACCESS_NONE;
6411 	enter_guest();
6412 }
6413 
6414 
6415 
6416 static int invalid_msr_init(struct vmcs *vmcs)
6417 {
6418 	if (!(ctrl_pin_rev.clr & PIN_PREEMPT)) {
6419 		printf("\tPreemption timer is not supported\n");
6420 		return VMX_TEST_EXIT;
6421 	}
6422 	vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) | PIN_PREEMPT);
6423 	preempt_val = 10000000;
6424 	vmcs_write(PREEMPT_TIMER_VALUE, preempt_val);
6425 	preempt_scale = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
6426 
6427 	if (!(ctrl_exit_rev.clr & EXI_SAVE_PREEMPT))
6428 		printf("\tSave preemption value is not supported\n");
6429 
6430 	vmcs_write(ENT_MSR_LD_CNT, 1);
6431 	vmcs_write(ENTER_MSR_LD_ADDR, (u64)0x13370000);
6432 
6433 	return VMX_TEST_START;
6434 }
6435 
6436 
6437 static void invalid_msr_main(void)
6438 {
6439 	report("Invalid MSR load", 0);
6440 }
6441 
6442 static int invalid_msr_exit_handler(void)
6443 {
6444 	report("Invalid MSR load", 0);
6445 	print_vmexit_info();
6446 	return VMX_TEST_EXIT;
6447 }
6448 
6449 static int invalid_msr_entry_failure(struct vmentry_failure *failure)
6450 {
6451 	ulong reason;
6452 
6453 	reason = vmcs_read(EXI_REASON);
6454 	report("Invalid MSR load", reason == (0x80000000u | VMX_FAIL_MSR));
6455 	return VMX_TEST_VMEXIT;
6456 }
6457 
6458 
6459 #define TEST(name) { #name, .v2 = name }
6460 
6461 /* name/init/guest_main/exit_handler/syscall_handler/guest_regs */
6462 struct vmx_test vmx_tests[] = {
6463 	{ "null", NULL, basic_guest_main, basic_exit_handler, NULL, {0} },
6464 	{ "vmenter", NULL, vmenter_main, vmenter_exit_handler, NULL, {0} },
6465 	{ "preemption timer", preemption_timer_init, preemption_timer_main,
6466 		preemption_timer_exit_handler, NULL, {0} },
6467 	{ "control field PAT", test_ctrl_pat_init, test_ctrl_pat_main,
6468 		test_ctrl_pat_exit_handler, NULL, {0} },
6469 	{ "control field EFER", test_ctrl_efer_init, test_ctrl_efer_main,
6470 		test_ctrl_efer_exit_handler, NULL, {0} },
6471 	{ "CR shadowing", NULL, cr_shadowing_main,
6472 		cr_shadowing_exit_handler, NULL, {0} },
6473 	{ "I/O bitmap", iobmp_init, iobmp_main, iobmp_exit_handler,
6474 		NULL, {0} },
6475 	{ "instruction intercept", insn_intercept_init, insn_intercept_main,
6476 		insn_intercept_exit_handler, NULL, {0} },
6477 	{ "EPT A/D disabled", ept_init, ept_main, ept_exit_handler, NULL, {0} },
6478 	{ "EPT A/D enabled", eptad_init, eptad_main, eptad_exit_handler, NULL, {0} },
6479 	{ "PML", pml_init, pml_main, pml_exit_handler, NULL, {0} },
6480 	{ "VPID", vpid_init, vpid_main, vpid_exit_handler, NULL, {0} },
6481 	{ "interrupt", interrupt_init, interrupt_main,
6482 		interrupt_exit_handler, NULL, {0} },
6483 	{ "debug controls", dbgctls_init, dbgctls_main, dbgctls_exit_handler,
6484 		NULL, {0} },
6485 	{ "MSR switch", msr_switch_init, msr_switch_main,
6486 		msr_switch_exit_handler, NULL, {0}, msr_switch_entry_failure },
6487 	{ "vmmcall", vmmcall_init, vmmcall_main, vmmcall_exit_handler, NULL, {0} },
6488 	{ "disable RDTSCP", disable_rdtscp_init, disable_rdtscp_main,
6489 		disable_rdtscp_exit_handler, NULL, {0} },
6490 	{ "int3", int3_init, int3_guest_main, int3_exit_handler, NULL, {0} },
6491 	{ "into", into_init, into_guest_main, into_exit_handler, NULL, {0} },
6492 	{ "exit_monitor_from_l2_test", NULL, exit_monitor_from_l2_main,
6493 		exit_monitor_from_l2_handler, NULL, {0} },
6494 	{ "invalid_msr", invalid_msr_init, invalid_msr_main,
6495 		invalid_msr_exit_handler, NULL, {0}, invalid_msr_entry_failure},
6496 	/* Basic V2 tests. */
6497 	TEST(v2_null_test),
6498 	TEST(v2_multiple_entries_test),
6499 	TEST(fixture_test_case1),
6500 	TEST(fixture_test_case2),
6501 	/* Opcode tests. */
6502 	TEST(invvpid_test_v2),
6503 	/* VM-entry tests */
6504 	TEST(vmx_controls_test),
6505 	TEST(vmentry_movss_shadow_test),
6506 	/* APICv tests */
6507 	TEST(vmx_eoi_bitmap_ioapic_scan_test),
6508 	TEST(vmx_hlt_with_rvi_test),
6509 	/* APIC pass-through tests */
6510 	TEST(vmx_apic_passthrough_test),
6511 	TEST(vmx_apic_passthrough_thread_test),
6512 	/* VMCS Shadowing tests */
6513 	TEST(vmx_vmcs_shadow_test),
6514 	/* Regression tests */
6515 	TEST(vmx_cr_load_test),
6516 	TEST(vmx_nm_test),
6517 	TEST(vmx_db_test),
6518 	TEST(vmx_nmi_window_test),
6519 	TEST(vmx_intr_window_test),
6520 	TEST(vmx_pending_event_test),
6521 	TEST(vmx_pending_event_hlt_test),
6522 	TEST(vmx_store_tsc_test),
6523 	/* EPT access tests. */
6524 	TEST(ept_access_test_not_present),
6525 	TEST(ept_access_test_read_only),
6526 	TEST(ept_access_test_write_only),
6527 	TEST(ept_access_test_read_write),
6528 	TEST(ept_access_test_execute_only),
6529 	TEST(ept_access_test_read_execute),
6530 	TEST(ept_access_test_write_execute),
6531 	TEST(ept_access_test_read_write_execute),
6532 	TEST(ept_access_test_reserved_bits),
6533 	TEST(ept_access_test_ignored_bits),
6534 	TEST(ept_access_test_paddr_not_present_ad_disabled),
6535 	TEST(ept_access_test_paddr_not_present_ad_enabled),
6536 	TEST(ept_access_test_paddr_read_only_ad_disabled),
6537 	TEST(ept_access_test_paddr_read_only_ad_enabled),
6538 	TEST(ept_access_test_paddr_read_write),
6539 	TEST(ept_access_test_paddr_read_write_execute),
6540 	TEST(ept_access_test_paddr_read_execute_ad_disabled),
6541 	TEST(ept_access_test_paddr_read_execute_ad_enabled),
6542 	TEST(ept_access_test_paddr_not_present_page_fault),
6543 	TEST(ept_access_test_force_2m_page),
6544 	{ NULL, NULL, NULL, NULL, NULL, {0} },
6545 };
6546