xref: /kvm-unit-tests/x86/svm_tests.c (revision 92098120f8579fcbe892b74444e109a232f3f526)
1 #include "svm.h"
2 #include "libcflat.h"
3 #include "processor.h"
4 #include "desc.h"
5 #include "msr.h"
6 #include "vm.h"
7 #include "smp.h"
8 #include "types.h"
9 #include "alloc_page.h"
10 #include "isr.h"
11 #include "apic.h"
12 #include "delay.h"
13 #include "util.h"
14 #include "x86/usermode.h"
15 
16 #define SVM_EXIT_MAX_DR_INTERCEPT 0x3f
17 
18 #define LATENCY_RUNS 1000000
19 
20 u64 tsc_start;
21 u64 tsc_end;
22 
23 u64 vmrun_sum, vmexit_sum;
24 u64 vmsave_sum, vmload_sum;
25 u64 stgi_sum, clgi_sum;
26 u64 latvmrun_max;
27 u64 latvmrun_min;
28 u64 latvmexit_max;
29 u64 latvmexit_min;
30 u64 latvmload_max;
31 u64 latvmload_min;
32 u64 latvmsave_max;
33 u64 latvmsave_min;
34 u64 latstgi_max;
35 u64 latstgi_min;
36 u64 latclgi_max;
37 u64 latclgi_min;
38 u64 runs;
39 
40 static void null_test(struct svm_test *test)
41 {
42 }
43 
44 static bool null_check(struct svm_test *test)
45 {
46 	return vmcb->control.exit_code == SVM_EXIT_VMMCALL;
47 }
48 
49 static void prepare_no_vmrun_int(struct svm_test *test)
50 {
51 	vmcb->control.intercept &= ~(1ULL << INTERCEPT_VMRUN);
52 }
53 
54 static bool check_no_vmrun_int(struct svm_test *test)
55 {
56 	return vmcb->control.exit_code == SVM_EXIT_ERR;
57 }
58 
59 static void test_vmrun(struct svm_test *test)
60 {
61 	asm volatile ("vmrun %0" : : "a"(virt_to_phys(vmcb)));
62 }
63 
64 static bool check_vmrun(struct svm_test *test)
65 {
66 	return vmcb->control.exit_code == SVM_EXIT_VMRUN;
67 }
68 
69 static void prepare_rsm_intercept(struct svm_test *test)
70 {
71 	default_prepare(test);
72 	vmcb->control.intercept |= 1 << INTERCEPT_RSM;
73 	vmcb->control.intercept_exceptions |= (1ULL << UD_VECTOR);
74 }
75 
76 static void test_rsm_intercept(struct svm_test *test)
77 {
78 	asm volatile ("rsm" : : : "memory");
79 }
80 
81 static bool check_rsm_intercept(struct svm_test *test)
82 {
83 	return get_test_stage(test) == 2;
84 }
85 
86 static bool finished_rsm_intercept(struct svm_test *test)
87 {
88 	switch (get_test_stage(test)) {
89 	case 0:
90 		if (vmcb->control.exit_code != SVM_EXIT_RSM) {
91 			report_fail("VMEXIT not due to rsm. Exit reason 0x%x",
92 				    vmcb->control.exit_code);
93 			return true;
94 		}
95 		vmcb->control.intercept &= ~(1 << INTERCEPT_RSM);
96 		inc_test_stage(test);
97 		break;
98 
99 	case 1:
100 		if (vmcb->control.exit_code != SVM_EXIT_EXCP_BASE + UD_VECTOR) {
101 			report_fail("VMEXIT not due to #UD. Exit reason 0x%x",
102 				    vmcb->control.exit_code);
103 			return true;
104 		}
105 		vmcb->save.rip += 2;
106 		inc_test_stage(test);
107 		break;
108 
109 	default:
110 		return true;
111 	}
112 	return get_test_stage(test) == 2;
113 }
114 
115 static void prepare_cr3_intercept(struct svm_test *test)
116 {
117 	default_prepare(test);
118 	vmcb->control.intercept_cr_read |= 1 << 3;
119 }
120 
121 static void test_cr3_intercept(struct svm_test *test)
122 {
123 	asm volatile ("mov %%cr3, %0" : "=r"(test->scratch) : : "memory");
124 }
125 
126 static bool check_cr3_intercept(struct svm_test *test)
127 {
128 	return vmcb->control.exit_code == SVM_EXIT_READ_CR3;
129 }
130 
131 static bool check_cr3_nointercept(struct svm_test *test)
132 {
133 	return null_check(test) && test->scratch == read_cr3();
134 }
135 
136 static void corrupt_cr3_intercept_bypass(void *_test)
137 {
138 	struct svm_test *test = _test;
139 	extern volatile u32 mmio_insn;
140 
141 	while (!__sync_bool_compare_and_swap(&test->scratch, 1, 2))
142 		pause();
143 	pause();
144 	pause();
145 	pause();
146 	mmio_insn = 0x90d8200f;  // mov %cr3, %rax; nop
147 }
148 
149 static void prepare_cr3_intercept_bypass(struct svm_test *test)
150 {
151 	default_prepare(test);
152 	vmcb->control.intercept_cr_read |= 1 << 3;
153 	on_cpu_async(1, corrupt_cr3_intercept_bypass, test);
154 }
155 
156 static void test_cr3_intercept_bypass(struct svm_test *test)
157 {
158 	ulong a = 0xa0000;
159 
160 	test->scratch = 1;
161 	while (test->scratch != 2)
162 		barrier();
163 
164 	asm volatile ("mmio_insn: mov %0, (%0); nop"
165 		      : "+a"(a) : : "memory");
166 	test->scratch = a;
167 }
168 
169 static void prepare_dr_intercept(struct svm_test *test)
170 {
171 	default_prepare(test);
172 	vmcb->control.intercept_dr_read = 0xff;
173 	vmcb->control.intercept_dr_write = 0xff;
174 }
175 
176 static void test_dr_intercept(struct svm_test *test)
177 {
178 	unsigned int i, failcnt = 0;
179 
180 	/* Loop testing debug register reads */
181 	for (i = 0; i < 8; i++) {
182 
183 		switch (i) {
184 		case 0:
185 			asm volatile ("mov %%dr0, %0" : "=r"(test->scratch) : : "memory");
186 			break;
187 		case 1:
188 			asm volatile ("mov %%dr1, %0" : "=r"(test->scratch) : : "memory");
189 			break;
190 		case 2:
191 			asm volatile ("mov %%dr2, %0" : "=r"(test->scratch) : : "memory");
192 			break;
193 		case 3:
194 			asm volatile ("mov %%dr3, %0" : "=r"(test->scratch) : : "memory");
195 			break;
196 		case 4:
197 			asm volatile ("mov %%dr4, %0" : "=r"(test->scratch) : : "memory");
198 			break;
199 		case 5:
200 			asm volatile ("mov %%dr5, %0" : "=r"(test->scratch) : : "memory");
201 			break;
202 		case 6:
203 			asm volatile ("mov %%dr6, %0" : "=r"(test->scratch) : : "memory");
204 			break;
205 		case 7:
206 			asm volatile ("mov %%dr7, %0" : "=r"(test->scratch) : : "memory");
207 			break;
208 		}
209 
210 		if (test->scratch != i) {
211 			report_fail("dr%u read intercept", i);
212 			failcnt++;
213 		}
214 	}
215 
216 	/* Loop testing debug register writes */
217 	for (i = 0; i < 8; i++) {
218 
219 		switch (i) {
220 		case 0:
221 			asm volatile ("mov %0, %%dr0" : : "r"(test->scratch) : "memory");
222 			break;
223 		case 1:
224 			asm volatile ("mov %0, %%dr1" : : "r"(test->scratch) : "memory");
225 			break;
226 		case 2:
227 			asm volatile ("mov %0, %%dr2" : : "r"(test->scratch) : "memory");
228 			break;
229 		case 3:
230 			asm volatile ("mov %0, %%dr3" : : "r"(test->scratch) : "memory");
231 			break;
232 		case 4:
233 			asm volatile ("mov %0, %%dr4" : : "r"(test->scratch) : "memory");
234 			break;
235 		case 5:
236 			asm volatile ("mov %0, %%dr5" : : "r"(test->scratch) : "memory");
237 			break;
238 		case 6:
239 			asm volatile ("mov %0, %%dr6" : : "r"(test->scratch) : "memory");
240 			break;
241 		case 7:
242 			asm volatile ("mov %0, %%dr7" : : "r"(test->scratch) : "memory");
243 			break;
244 		}
245 
246 		if (test->scratch != i) {
247 			report_fail("dr%u write intercept", i);
248 			failcnt++;
249 		}
250 	}
251 
252 	test->scratch = failcnt;
253 }
254 
255 static bool dr_intercept_finished(struct svm_test *test)
256 {
257 	ulong n = (vmcb->control.exit_code - SVM_EXIT_READ_DR0);
258 
259 	/* Only expect DR intercepts */
260 	if (n > (SVM_EXIT_MAX_DR_INTERCEPT - SVM_EXIT_READ_DR0))
261 		return true;
262 
263 	/*
264 	 * Compute debug register number.
265 	 * Per Appendix C "SVM Intercept Exit Codes" of AMD64 Architecture
266 	 * Programmer's Manual Volume 2 - System Programming:
267 	 * http://support.amd.com/TechDocs/24593.pdf
268 	 * there are 16 VMEXIT codes each for DR read and write.
269 	 */
270 	test->scratch = (n % 16);
271 
272 	/* Jump over MOV instruction */
273 	vmcb->save.rip += 3;
274 
275 	return false;
276 }
277 
278 static bool check_dr_intercept(struct svm_test *test)
279 {
280 	return !test->scratch;
281 }
282 
283 static bool next_rip_supported(void)
284 {
285 	return this_cpu_has(X86_FEATURE_NRIPS);
286 }
287 
288 static void prepare_next_rip(struct svm_test *test)
289 {
290 	vmcb->control.intercept |= (1ULL << INTERCEPT_RDTSC);
291 }
292 
293 
294 static void test_next_rip(struct svm_test *test)
295 {
296 	asm volatile ("rdtsc\n\t"
297 		      ".globl exp_next_rip\n\t"
298 		      "exp_next_rip:\n\t" ::: "eax", "edx");
299 }
300 
301 static bool check_next_rip(struct svm_test *test)
302 {
303 	extern char exp_next_rip;
304 	unsigned long address = (unsigned long)&exp_next_rip;
305 
306 	return address == vmcb->control.next_rip;
307 }
308 
309 extern u8 *msr_bitmap;
310 
311 static void prepare_msr_intercept(struct svm_test *test)
312 {
313 	default_prepare(test);
314 	vmcb->control.intercept |= (1ULL << INTERCEPT_MSR_PROT);
315 	vmcb->control.intercept_exceptions |= (1ULL << GP_VECTOR);
316 	memset(msr_bitmap, 0xff, MSR_BITMAP_SIZE);
317 }
318 
319 static void test_msr_intercept(struct svm_test *test)
320 {
321 	unsigned long msr_value = 0xef8056791234abcd; /* Arbitrary value */
322 	unsigned long msr_index;
323 
324 	for (msr_index = 0; msr_index <= 0xc0011fff; msr_index++) {
325 		if (msr_index == 0xC0010131 /* MSR_SEV_STATUS */) {
326 			/*
327 			 * Per section 15.34.10 "SEV_STATUS MSR" of AMD64 Architecture
328 			 * Programmer's Manual volume 2 - System Programming:
329 			 * http://support.amd.com/TechDocs/24593.pdf
330 			 * SEV_STATUS MSR (C001_0131) is a non-interceptable MSR.
331 			 */
332 			continue;
333 		}
334 
335 		/* Skips gaps between supported MSR ranges */
336 		if (msr_index == 0x2000)
337 			msr_index = 0xc0000000;
338 		else if (msr_index == 0xc0002000)
339 			msr_index = 0xc0010000;
340 
341 		test->scratch = -1;
342 
343 		rdmsr(msr_index);
344 
345 		/* Check that a read intercept occurred for MSR at msr_index */
346 		if (test->scratch != msr_index)
347 			report_fail("MSR 0x%lx read intercept", msr_index);
348 
349 		/*
350 		 * Poor man approach to generate a value that
351 		 * seems arbitrary each time around the loop.
352 		 */
353 		msr_value += (msr_value << 1);
354 
355 		wrmsr(msr_index, msr_value);
356 
357 		/* Check that a write intercept occurred for MSR with msr_value */
358 		if (test->scratch != msr_value)
359 			report_fail("MSR 0x%lx write intercept", msr_index);
360 	}
361 
362 	test->scratch = -2;
363 }
364 
365 static bool msr_intercept_finished(struct svm_test *test)
366 {
367 	u32 exit_code = vmcb->control.exit_code;
368 	u64 exit_info_1;
369 	u8 *opcode;
370 
371 	if (exit_code == SVM_EXIT_MSR) {
372 		exit_info_1 = vmcb->control.exit_info_1;
373 	} else {
374 		/*
375 		 * If #GP exception occurs instead, check that it was
376 		 * for RDMSR/WRMSR and set exit_info_1 accordingly.
377 		 */
378 
379 		if (exit_code != (SVM_EXIT_EXCP_BASE + GP_VECTOR))
380 			return true;
381 
382 		opcode = (u8 *)vmcb->save.rip;
383 		if (opcode[0] != 0x0f)
384 			return true;
385 
386 		switch (opcode[1]) {
387 		case 0x30: /* WRMSR */
388 			exit_info_1 = 1;
389 			break;
390 		case 0x32: /* RDMSR */
391 			exit_info_1 = 0;
392 			break;
393 		default:
394 			return true;
395 		}
396 
397 		/*
398 		 * Warn that #GP exception occured instead.
399 		 * RCX holds the MSR index.
400 		 */
401 		printf("%s 0x%lx #GP exception\n",
402 		       exit_info_1 ? "WRMSR" : "RDMSR", get_regs().rcx);
403 	}
404 
405 	/* Jump over RDMSR/WRMSR instruction */
406 	vmcb->save.rip += 2;
407 
408 	/*
409 	 * Test whether the intercept was for RDMSR/WRMSR.
410 	 * For RDMSR, test->scratch is set to the MSR index;
411 	 *      RCX holds the MSR index.
412 	 * For WRMSR, test->scratch is set to the MSR value;
413 	 *      RDX holds the upper 32 bits of the MSR value,
414 	 *      while RAX hold its lower 32 bits.
415 	 */
416 	if (exit_info_1)
417 		test->scratch =
418 			((get_regs().rdx << 32) | (vmcb->save.rax & 0xffffffff));
419 	else
420 		test->scratch = get_regs().rcx;
421 
422 	return false;
423 }
424 
425 static bool check_msr_intercept(struct svm_test *test)
426 {
427 	memset(msr_bitmap, 0, MSR_BITMAP_SIZE);
428 	return (test->scratch == -2);
429 }
430 
431 static void prepare_mode_switch(struct svm_test *test)
432 {
433 	vmcb->control.intercept_exceptions |= (1ULL << GP_VECTOR)
434 		|  (1ULL << UD_VECTOR)
435 		|  (1ULL << DF_VECTOR)
436 		|  (1ULL << PF_VECTOR);
437 	test->scratch = 0;
438 }
439 
440 static void test_mode_switch(struct svm_test *test)
441 {
442 	asm volatile("	cli\n"
443 		     "	ljmp *1f\n" /* jump to 32-bit code segment */
444 		     "1:\n"
445 		     "	.long 2f\n"
446 		     "	.long " xstr(KERNEL_CS32) "\n"
447 		     ".code32\n"
448 		     "2:\n"
449 		     "	movl %%cr0, %%eax\n"
450 		     "	btcl  $31, %%eax\n" /* clear PG */
451 		     "	movl %%eax, %%cr0\n"
452 		     "	movl $0xc0000080, %%ecx\n" /* EFER */
453 		     "	rdmsr\n"
454 		     "	btcl $8, %%eax\n" /* clear LME */
455 		     "	wrmsr\n"
456 		     "	movl %%cr4, %%eax\n"
457 		     "	btcl $5, %%eax\n" /* clear PAE */
458 		     "	movl %%eax, %%cr4\n"
459 		     "	movw %[ds16], %%ax\n"
460 		     "	movw %%ax, %%ds\n"
461 		     "	ljmpl %[cs16], $3f\n" /* jump to 16 bit protected-mode */
462 		     ".code16\n"
463 		     "3:\n"
464 		     "	movl %%cr0, %%eax\n"
465 		     "	btcl $0, %%eax\n" /* clear PE  */
466 		     "	movl %%eax, %%cr0\n"
467 		     "	ljmpl $0, $4f\n"   /* jump to real-mode */
468 		     "4:\n"
469 		     "	vmmcall\n"
470 		     "	movl %%cr0, %%eax\n"
471 		     "	btsl $0, %%eax\n" /* set PE  */
472 		     "	movl %%eax, %%cr0\n"
473 		     "	ljmpl %[cs32], $5f\n" /* back to protected mode */
474 		     ".code32\n"
475 		     "5:\n"
476 		     "	movl %%cr4, %%eax\n"
477 		     "	btsl $5, %%eax\n" /* set PAE */
478 		     "	movl %%eax, %%cr4\n"
479 		     "	movl $0xc0000080, %%ecx\n" /* EFER */
480 		     "	rdmsr\n"
481 		     "	btsl $8, %%eax\n" /* set LME */
482 		     "	wrmsr\n"
483 		     "	movl %%cr0, %%eax\n"
484 		     "	btsl  $31, %%eax\n" /* set PG */
485 		     "	movl %%eax, %%cr0\n"
486 		     "	ljmpl %[cs64], $6f\n"    /* back to long mode */
487 		     ".code64\n\t"
488 		     "6:\n"
489 		     "	vmmcall\n"
490 		     :: [cs16] "i"(KERNEL_CS16), [ds16] "i"(KERNEL_DS16),
491 		      [cs32] "i"(KERNEL_CS32), [cs64] "i"(KERNEL_CS64)
492 		     : "rax", "rbx", "rcx", "rdx", "memory");
493 }
494 
495 static bool mode_switch_finished(struct svm_test *test)
496 {
497 	u64 cr0, cr4, efer;
498 
499 	cr0  = vmcb->save.cr0;
500 	cr4  = vmcb->save.cr4;
501 	efer = vmcb->save.efer;
502 
503 	/* Only expect VMMCALL intercepts */
504 	if (vmcb->control.exit_code != SVM_EXIT_VMMCALL)
505 		return true;
506 
507 	/* Jump over VMMCALL instruction */
508 	vmcb->save.rip += 3;
509 
510 	/* Do sanity checks */
511 	switch (test->scratch) {
512 	case 0:
513 		/* Test should be in real mode now - check for this */
514 		if ((cr0  & 0x80000001) || /* CR0.PG, CR0.PE */
515 		    (cr4  & 0x00000020) || /* CR4.PAE */
516 		    (efer & 0x00000500))   /* EFER.LMA, EFER.LME */
517 			return true;
518 		break;
519 	case 2:
520 		/* Test should be back in long-mode now - check for this */
521 		if (((cr0  & 0x80000001) != 0x80000001) || /* CR0.PG, CR0.PE */
522 		    ((cr4  & 0x00000020) != 0x00000020) || /* CR4.PAE */
523 		    ((efer & 0x00000500) != 0x00000500))   /* EFER.LMA, EFER.LME */
524 			return true;
525 		break;
526 	}
527 
528 	/* one step forward */
529 	test->scratch += 1;
530 
531 	return test->scratch == 2;
532 }
533 
534 static bool check_mode_switch(struct svm_test *test)
535 {
536 	return test->scratch == 2;
537 }
538 
539 extern u8 *io_bitmap;
540 
541 static void prepare_ioio(struct svm_test *test)
542 {
543 	vmcb->control.intercept |= (1ULL << INTERCEPT_IOIO_PROT);
544 	test->scratch = 0;
545 	memset(io_bitmap, 0, 8192);
546 	io_bitmap[8192] = 0xFF;
547 }
548 
549 static void test_ioio(struct svm_test *test)
550 {
551 	// stage 0, test IO pass
552 	inb(0x5000);
553 	outb(0x0, 0x5000);
554 	if (get_test_stage(test) != 0)
555 		goto fail;
556 
557 	// test IO width, in/out
558 	io_bitmap[0] = 0xFF;
559 	inc_test_stage(test);
560 	inb(0x0);
561 	if (get_test_stage(test) != 2)
562 		goto fail;
563 
564 	outw(0x0, 0x0);
565 	if (get_test_stage(test) != 3)
566 		goto fail;
567 
568 	inl(0x0);
569 	if (get_test_stage(test) != 4)
570 		goto fail;
571 
572 	// test low/high IO port
573 	io_bitmap[0x5000 / 8] = (1 << (0x5000 % 8));
574 	inb(0x5000);
575 	if (get_test_stage(test) != 5)
576 		goto fail;
577 
578 	io_bitmap[0x9000 / 8] = (1 << (0x9000 % 8));
579 	inw(0x9000);
580 	if (get_test_stage(test) != 6)
581 		goto fail;
582 
583 	// test partial pass
584 	io_bitmap[0x5000 / 8] = (1 << (0x5000 % 8));
585 	inl(0x4FFF);
586 	if (get_test_stage(test) != 7)
587 		goto fail;
588 
589 	// test across pages
590 	inc_test_stage(test);
591 	inl(0x7FFF);
592 	if (get_test_stage(test) != 8)
593 		goto fail;
594 
595 	inc_test_stage(test);
596 	io_bitmap[0x8000 / 8] = 1 << (0x8000 % 8);
597 	inl(0x7FFF);
598 	if (get_test_stage(test) != 10)
599 		goto fail;
600 
601 	io_bitmap[0] = 0;
602 	inl(0xFFFF);
603 	if (get_test_stage(test) != 11)
604 		goto fail;
605 
606 	io_bitmap[0] = 0xFF;
607 	io_bitmap[8192] = 0;
608 	inl(0xFFFF);
609 	inc_test_stage(test);
610 	if (get_test_stage(test) != 12)
611 		goto fail;
612 
613 	return;
614 
615 fail:
616 	report_fail("stage %d", get_test_stage(test));
617 	test->scratch = -1;
618 }
619 
620 static bool ioio_finished(struct svm_test *test)
621 {
622 	unsigned port, size;
623 
624 	/* Only expect IOIO intercepts */
625 	if (vmcb->control.exit_code == SVM_EXIT_VMMCALL)
626 		return true;
627 
628 	if (vmcb->control.exit_code != SVM_EXIT_IOIO)
629 		return true;
630 
631 	/* one step forward */
632 	test->scratch += 1;
633 
634 	port = vmcb->control.exit_info_1 >> 16;
635 	size = (vmcb->control.exit_info_1 >> SVM_IOIO_SIZE_SHIFT) & 7;
636 
637 	while (size--) {
638 		io_bitmap[port / 8] &= ~(1 << (port & 7));
639 		port++;
640 	}
641 
642 	return false;
643 }
644 
645 static bool check_ioio(struct svm_test *test)
646 {
647 	memset(io_bitmap, 0, 8193);
648 	return test->scratch != -1;
649 }
650 
651 static void prepare_asid_zero(struct svm_test *test)
652 {
653 	vmcb->control.asid = 0;
654 }
655 
656 static void test_asid_zero(struct svm_test *test)
657 {
658 	asm volatile ("vmmcall\n\t");
659 }
660 
661 static bool check_asid_zero(struct svm_test *test)
662 {
663 	return vmcb->control.exit_code == SVM_EXIT_ERR;
664 }
665 
666 static void sel_cr0_bug_prepare(struct svm_test *test)
667 {
668 	vmcb->control.intercept |= (1ULL << INTERCEPT_SELECTIVE_CR0);
669 }
670 
671 static bool sel_cr0_bug_finished(struct svm_test *test)
672 {
673 	return true;
674 }
675 
676 static void sel_cr0_bug_test(struct svm_test *test)
677 {
678 	unsigned long cr0;
679 
680 	/* read cr0, clear CD, and write back */
681 	cr0  = read_cr0();
682 	cr0 |= (1UL << 30);
683 	write_cr0(cr0);
684 
685 	/*
686 	 * If we are here the test failed, not sure what to do now because we
687 	 * are not in guest-mode anymore so we can't trigger an intercept.
688 	 * Trigger a tripple-fault for now.
689 	 */
690 	report_fail("sel_cr0 test. Can not recover from this - exiting");
691 	exit(report_summary());
692 }
693 
694 static bool sel_cr0_bug_check(struct svm_test *test)
695 {
696 	return vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE;
697 }
698 
699 #define TSC_ADJUST_VALUE    (1ll << 32)
700 #define TSC_OFFSET_VALUE    (~0ull << 48)
701 static bool ok;
702 
703 static bool tsc_adjust_supported(void)
704 {
705 	return this_cpu_has(X86_FEATURE_TSC_ADJUST);
706 }
707 
708 static void tsc_adjust_prepare(struct svm_test *test)
709 {
710 	default_prepare(test);
711 	vmcb->control.tsc_offset = TSC_OFFSET_VALUE;
712 
713 	wrmsr(MSR_IA32_TSC_ADJUST, -TSC_ADJUST_VALUE);
714 	int64_t adjust = rdmsr(MSR_IA32_TSC_ADJUST);
715 	ok = adjust == -TSC_ADJUST_VALUE;
716 }
717 
718 static void tsc_adjust_test(struct svm_test *test)
719 {
720 	int64_t adjust = rdmsr(MSR_IA32_TSC_ADJUST);
721 	ok &= adjust == -TSC_ADJUST_VALUE;
722 
723 	uint64_t l1_tsc = rdtsc() - TSC_OFFSET_VALUE;
724 	wrmsr(MSR_IA32_TSC, l1_tsc - TSC_ADJUST_VALUE);
725 
726 	adjust = rdmsr(MSR_IA32_TSC_ADJUST);
727 	ok &= adjust <= -2 * TSC_ADJUST_VALUE;
728 
729 	uint64_t l1_tsc_end = rdtsc() - TSC_OFFSET_VALUE;
730 	ok &= (l1_tsc_end + TSC_ADJUST_VALUE - l1_tsc) < TSC_ADJUST_VALUE;
731 
732 	uint64_t l1_tsc_msr = rdmsr(MSR_IA32_TSC) - TSC_OFFSET_VALUE;
733 	ok &= (l1_tsc_msr + TSC_ADJUST_VALUE - l1_tsc) < TSC_ADJUST_VALUE;
734 }
735 
736 static bool tsc_adjust_check(struct svm_test *test)
737 {
738 	int64_t adjust = rdmsr(MSR_IA32_TSC_ADJUST);
739 
740 	wrmsr(MSR_IA32_TSC_ADJUST, 0);
741 	return ok && adjust <= -2 * TSC_ADJUST_VALUE;
742 }
743 
744 
745 static u64 guest_tsc_delay_value;
746 /* number of bits to shift tsc right for stable result */
747 #define TSC_SHIFT 24
748 #define TSC_SCALE_ITERATIONS 10
749 
750 static void svm_tsc_scale_guest(struct svm_test *test)
751 {
752 	u64 start_tsc = rdtsc();
753 
754 	while (rdtsc() - start_tsc < guest_tsc_delay_value)
755 		cpu_relax();
756 }
757 
758 static void svm_tsc_scale_run_testcase(u64 duration,
759 				       double tsc_scale, u64 tsc_offset)
760 {
761 	u64 start_tsc, actual_duration;
762 
763 	guest_tsc_delay_value = (duration << TSC_SHIFT) * tsc_scale;
764 
765 	test_set_guest(svm_tsc_scale_guest);
766 	vmcb->control.tsc_offset = tsc_offset;
767 	wrmsr(MSR_AMD64_TSC_RATIO, (u64)(tsc_scale * (1ULL << 32)));
768 
769 	start_tsc = rdtsc();
770 
771 	if (svm_vmrun() != SVM_EXIT_VMMCALL)
772 		report_fail("unexpected vm exit code 0x%x", vmcb->control.exit_code);
773 
774 	actual_duration = (rdtsc() - start_tsc) >> TSC_SHIFT;
775 
776 	report(duration == actual_duration, "tsc delay (expected: %lu, actual: %lu)",
777 	       duration, actual_duration);
778 }
779 
780 static void svm_tsc_scale_test(void)
781 {
782 	int i;
783 
784 	if (!tsc_scale_supported()) {
785 		report_skip("TSC scale not supported in the guest");
786 		return;
787 	}
788 
789 	report(rdmsr(MSR_AMD64_TSC_RATIO) == TSC_RATIO_DEFAULT,
790 	       "initial TSC scale ratio");
791 
792 	for (i = 0 ; i < TSC_SCALE_ITERATIONS; i++) {
793 
794 		double tsc_scale = (double)(rdrand() % 100 + 1) / 10;
795 		int duration = rdrand() % 50 + 1;
796 		u64 tsc_offset = rdrand();
797 
798 		report_info("duration=%d, tsc_scale=%d, tsc_offset=%ld",
799 			    duration, (int)(tsc_scale * 100), tsc_offset);
800 
801 		svm_tsc_scale_run_testcase(duration, tsc_scale, tsc_offset);
802 	}
803 
804 	svm_tsc_scale_run_testcase(50, 255, rdrand());
805 	svm_tsc_scale_run_testcase(50, 0.0001, rdrand());
806 }
807 
808 static void latency_prepare(struct svm_test *test)
809 {
810 	default_prepare(test);
811 	runs = LATENCY_RUNS;
812 	latvmrun_min = latvmexit_min = -1ULL;
813 	latvmrun_max = latvmexit_max = 0;
814 	vmrun_sum = vmexit_sum = 0;
815 	tsc_start = rdtsc();
816 }
817 
818 static void latency_test(struct svm_test *test)
819 {
820 	u64 cycles;
821 
822 start:
823 	tsc_end = rdtsc();
824 
825 	cycles = tsc_end - tsc_start;
826 
827 	if (cycles > latvmrun_max)
828 		latvmrun_max = cycles;
829 
830 	if (cycles < latvmrun_min)
831 		latvmrun_min = cycles;
832 
833 	vmrun_sum += cycles;
834 
835 	tsc_start = rdtsc();
836 
837 	asm volatile ("vmmcall" : : : "memory");
838 	goto start;
839 }
840 
841 static bool latency_finished(struct svm_test *test)
842 {
843 	u64 cycles;
844 
845 	tsc_end = rdtsc();
846 
847 	cycles = tsc_end - tsc_start;
848 
849 	if (cycles > latvmexit_max)
850 		latvmexit_max = cycles;
851 
852 	if (cycles < latvmexit_min)
853 		latvmexit_min = cycles;
854 
855 	vmexit_sum += cycles;
856 
857 	vmcb->save.rip += 3;
858 
859 	runs -= 1;
860 
861 	tsc_end = rdtsc();
862 
863 	return runs == 0;
864 }
865 
866 static bool latency_finished_clean(struct svm_test *test)
867 {
868 	vmcb->control.clean = VMCB_CLEAN_ALL;
869 	return latency_finished(test);
870 }
871 
872 static bool latency_check(struct svm_test *test)
873 {
874 	printf("    Latency VMRUN : max: %ld min: %ld avg: %ld\n", latvmrun_max,
875 	       latvmrun_min, vmrun_sum / LATENCY_RUNS);
876 	printf("    Latency VMEXIT: max: %ld min: %ld avg: %ld\n", latvmexit_max,
877 	       latvmexit_min, vmexit_sum / LATENCY_RUNS);
878 	return true;
879 }
880 
881 static void lat_svm_insn_prepare(struct svm_test *test)
882 {
883 	default_prepare(test);
884 	runs = LATENCY_RUNS;
885 	latvmload_min = latvmsave_min = latstgi_min = latclgi_min = -1ULL;
886 	latvmload_max = latvmsave_max = latstgi_max = latclgi_max = 0;
887 	vmload_sum = vmsave_sum = stgi_sum = clgi_sum;
888 }
889 
890 static bool lat_svm_insn_finished(struct svm_test *test)
891 {
892 	u64 vmcb_phys = virt_to_phys(vmcb);
893 	u64 cycles;
894 
895 	for ( ; runs != 0; runs--) {
896 		tsc_start = rdtsc();
897 		asm volatile("vmload %0\n\t" : : "a"(vmcb_phys) : "memory");
898 		cycles = rdtsc() - tsc_start;
899 		if (cycles > latvmload_max)
900 			latvmload_max = cycles;
901 		if (cycles < latvmload_min)
902 			latvmload_min = cycles;
903 		vmload_sum += cycles;
904 
905 		tsc_start = rdtsc();
906 		asm volatile("vmsave %0\n\t" : : "a"(vmcb_phys) : "memory");
907 		cycles = rdtsc() - tsc_start;
908 		if (cycles > latvmsave_max)
909 			latvmsave_max = cycles;
910 		if (cycles < latvmsave_min)
911 			latvmsave_min = cycles;
912 		vmsave_sum += cycles;
913 
914 		tsc_start = rdtsc();
915 		asm volatile("stgi\n\t");
916 		cycles = rdtsc() - tsc_start;
917 		if (cycles > latstgi_max)
918 			latstgi_max = cycles;
919 		if (cycles < latstgi_min)
920 			latstgi_min = cycles;
921 		stgi_sum += cycles;
922 
923 		tsc_start = rdtsc();
924 		asm volatile("clgi\n\t");
925 		cycles = rdtsc() - tsc_start;
926 		if (cycles > latclgi_max)
927 			latclgi_max = cycles;
928 		if (cycles < latclgi_min)
929 			latclgi_min = cycles;
930 		clgi_sum += cycles;
931 	}
932 
933 	tsc_end = rdtsc();
934 
935 	return true;
936 }
937 
938 static bool lat_svm_insn_check(struct svm_test *test)
939 {
940 	printf("    Latency VMLOAD: max: %ld min: %ld avg: %ld\n", latvmload_max,
941 	       latvmload_min, vmload_sum / LATENCY_RUNS);
942 	printf("    Latency VMSAVE: max: %ld min: %ld avg: %ld\n", latvmsave_max,
943 	       latvmsave_min, vmsave_sum / LATENCY_RUNS);
944 	printf("    Latency STGI:   max: %ld min: %ld avg: %ld\n", latstgi_max,
945 	       latstgi_min, stgi_sum / LATENCY_RUNS);
946 	printf("    Latency CLGI:   max: %ld min: %ld avg: %ld\n", latclgi_max,
947 	       latclgi_min, clgi_sum / LATENCY_RUNS);
948 	return true;
949 }
950 
951 /*
952  * Report failures from SVM guest code, and on failure, set the stage to -1 and
953  * do VMMCALL to terminate the test (host side must treat -1 as "finished").
954  * TODO: fix the tests that don't play nice with a straight report, e.g. the
955  * V_TPR test fails if report() is invoked.
956  */
957 #define report_svm_guest(cond, test, fmt, args...)	\
958 do {							\
959 	if (!(cond)) {					\
960 		report_fail(fmt, ##args);		\
961 		set_test_stage(test, -1);		\
962 		vmmcall();				\
963 	}						\
964 } while (0)
965 
966 bool pending_event_ipi_fired;
967 bool pending_event_guest_run;
968 
969 static void pending_event_ipi_isr(isr_regs_t *regs)
970 {
971 	pending_event_ipi_fired = true;
972 	eoi();
973 }
974 
975 static void pending_event_prepare(struct svm_test *test)
976 {
977 	int ipi_vector = 0xf1;
978 
979 	default_prepare(test);
980 
981 	pending_event_ipi_fired = false;
982 
983 	handle_irq(ipi_vector, pending_event_ipi_isr);
984 
985 	pending_event_guest_run = false;
986 
987 	vmcb->control.intercept |= (1ULL << INTERCEPT_INTR);
988 	vmcb->control.int_ctl |= V_INTR_MASKING_MASK;
989 
990 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
991 		       APIC_DM_FIXED | ipi_vector, 0);
992 
993 	set_test_stage(test, 0);
994 }
995 
996 static void pending_event_test(struct svm_test *test)
997 {
998 	pending_event_guest_run = true;
999 }
1000 
1001 static bool pending_event_finished(struct svm_test *test)
1002 {
1003 	switch (get_test_stage(test)) {
1004 	case 0:
1005 		if (vmcb->control.exit_code != SVM_EXIT_INTR) {
1006 			report_fail("VMEXIT not due to pending interrupt. Exit reason 0x%x",
1007 				    vmcb->control.exit_code);
1008 			return true;
1009 		}
1010 
1011 		vmcb->control.intercept &= ~(1ULL << INTERCEPT_INTR);
1012 		vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
1013 
1014 		if (pending_event_guest_run) {
1015 			report_fail("Guest ran before host received IPI\n");
1016 			return true;
1017 		}
1018 
1019 		irq_enable();
1020 		asm volatile ("nop");
1021 		irq_disable();
1022 
1023 		if (!pending_event_ipi_fired) {
1024 			report_fail("Pending interrupt not dispatched after IRQ enabled\n");
1025 			return true;
1026 		}
1027 		break;
1028 
1029 	case 1:
1030 		if (!pending_event_guest_run) {
1031 			report_fail("Guest did not resume when no interrupt\n");
1032 			return true;
1033 		}
1034 		break;
1035 	}
1036 
1037 	inc_test_stage(test);
1038 
1039 	return get_test_stage(test) == 2;
1040 }
1041 
1042 static bool pending_event_check(struct svm_test *test)
1043 {
1044 	return get_test_stage(test) == 2;
1045 }
1046 
1047 static void pending_event_cli_prepare(struct svm_test *test)
1048 {
1049 	default_prepare(test);
1050 
1051 	pending_event_ipi_fired = false;
1052 
1053 	handle_irq(0xf1, pending_event_ipi_isr);
1054 
1055 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
1056 		       APIC_DM_FIXED | 0xf1, 0);
1057 
1058 	set_test_stage(test, 0);
1059 }
1060 
1061 static void pending_event_cli_prepare_gif_clear(struct svm_test *test)
1062 {
1063 	asm("cli");
1064 }
1065 
1066 static void pending_event_cli_test(struct svm_test *test)
1067 {
1068 	report_svm_guest(!pending_event_ipi_fired, test,
1069 			 "IRQ should NOT be delivered while IRQs disabled");
1070 
1071 	/* VINTR_MASKING is zero.  This should cause the IPI to fire.  */
1072 	irq_enable();
1073 	asm volatile ("nop");
1074 	irq_disable();
1075 
1076 	report_svm_guest(pending_event_ipi_fired, test,
1077 			 "IRQ should be delivered after enabling IRQs");
1078 	vmmcall();
1079 
1080 	/*
1081 	 * Now VINTR_MASKING=1, but no interrupt is pending so
1082 	 * the VINTR interception should be clear in VMCB02.  Check
1083 	 * that L0 did not leave a stale VINTR in the VMCB.
1084 	 */
1085 	irq_enable();
1086 	asm volatile ("nop");
1087 	irq_disable();
1088 }
1089 
1090 static bool pending_event_cli_finished(struct svm_test *test)
1091 {
1092 	report_svm_guest(vmcb->control.exit_code == SVM_EXIT_VMMCALL, test,
1093 			 "Wanted VMMCALL VM-Exit, got exit reason 0x%x",
1094 			 vmcb->control.exit_code);
1095 
1096 	switch (get_test_stage(test)) {
1097 	case 0:
1098 		vmcb->save.rip += 3;
1099 
1100 		pending_event_ipi_fired = false;
1101 
1102 		vmcb->control.int_ctl |= V_INTR_MASKING_MASK;
1103 
1104 		/* Now entering again with VINTR_MASKING=1.  */
1105 		apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL |
1106 			       APIC_DM_FIXED | 0xf1, 0);
1107 
1108 		break;
1109 
1110 	case 1:
1111 		if (pending_event_ipi_fired == true) {
1112 			report_fail("Interrupt triggered by guest");
1113 			return true;
1114 		}
1115 
1116 		irq_enable();
1117 		asm volatile ("nop");
1118 		irq_disable();
1119 
1120 		if (pending_event_ipi_fired != true) {
1121 			report_fail("Interrupt not triggered by host");
1122 			return true;
1123 		}
1124 
1125 		break;
1126 
1127 	default:
1128 		return true;
1129 	}
1130 
1131 	inc_test_stage(test);
1132 
1133 	return get_test_stage(test) == 2;
1134 }
1135 
1136 static bool pending_event_cli_check(struct svm_test *test)
1137 {
1138 	return get_test_stage(test) == 2;
1139 }
1140 
1141 #define TIMER_VECTOR    222
1142 
1143 static volatile bool timer_fired;
1144 
1145 static void timer_isr(isr_regs_t *regs)
1146 {
1147 	timer_fired = true;
1148 	apic_write(APIC_EOI, 0);
1149 }
1150 
1151 static void interrupt_prepare(struct svm_test *test)
1152 {
1153 	default_prepare(test);
1154 	handle_irq(TIMER_VECTOR, timer_isr);
1155 	timer_fired = false;
1156 	set_test_stage(test, 0);
1157 }
1158 
1159 static void interrupt_test(struct svm_test *test)
1160 {
1161 	long long start, loops;
1162 
1163 	apic_write(APIC_LVTT, TIMER_VECTOR);
1164 	irq_enable();
1165 	apic_write(APIC_TMICT, 1); //Timer Initial Count Register 0x380 one-shot
1166 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1167 		asm volatile ("nop");
1168 
1169 	report_svm_guest(timer_fired, test,
1170 			 "direct interrupt while running guest");
1171 
1172 	apic_write(APIC_TMICT, 0);
1173 	irq_disable();
1174 	vmmcall();
1175 
1176 	timer_fired = false;
1177 	apic_write(APIC_TMICT, 1);
1178 	for (loops = 0; loops < 10000000 && !timer_fired; loops++)
1179 		asm volatile ("nop");
1180 
1181 	report_svm_guest(timer_fired, test,
1182 			 "intercepted interrupt while running guest");
1183 
1184 	irq_enable();
1185 	apic_write(APIC_TMICT, 0);
1186 	irq_disable();
1187 
1188 	timer_fired = false;
1189 	start = rdtsc();
1190 	apic_write(APIC_TMICT, 1000000);
1191 	safe_halt();
1192 
1193 	report_svm_guest(timer_fired, test, "direct interrupt + hlt");
1194 	report(rdtsc() - start > 10000, "IRQ arrived after expected delay");
1195 
1196 	apic_write(APIC_TMICT, 0);
1197 	irq_disable();
1198 	vmmcall();
1199 
1200 	timer_fired = false;
1201 	start = rdtsc();
1202 	apic_write(APIC_TMICT, 1000000);
1203 	asm volatile ("hlt");
1204 
1205 	report_svm_guest(timer_fired, test, "intercepted interrupt + hlt");
1206 	report(rdtsc() - start > 10000, "IRQ arrived after expected delay");
1207 
1208 	apic_write(APIC_TMICT, 0);
1209 	irq_disable();
1210 }
1211 
1212 static bool interrupt_finished(struct svm_test *test)
1213 {
1214 	switch (get_test_stage(test)) {
1215 	case 0:
1216 	case 2:
1217 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
1218 			report_fail("VMEXIT not due to vmmcall. Exit reason 0x%x",
1219 				    vmcb->control.exit_code);
1220 			return true;
1221 		}
1222 		vmcb->save.rip += 3;
1223 
1224 		vmcb->control.intercept |= (1ULL << INTERCEPT_INTR);
1225 		vmcb->control.int_ctl |= V_INTR_MASKING_MASK;
1226 		break;
1227 
1228 	case 1:
1229 	case 3:
1230 		if (vmcb->control.exit_code != SVM_EXIT_INTR) {
1231 			report_fail("VMEXIT not due to intr intercept. Exit reason 0x%x",
1232 				    vmcb->control.exit_code);
1233 			return true;
1234 		}
1235 
1236 		irq_enable();
1237 		asm volatile ("nop");
1238 		irq_disable();
1239 
1240 		vmcb->control.intercept &= ~(1ULL << INTERCEPT_INTR);
1241 		vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
1242 		break;
1243 
1244 	case 4:
1245 		break;
1246 
1247 	default:
1248 		return true;
1249 	}
1250 
1251 	inc_test_stage(test);
1252 
1253 	return get_test_stage(test) == 5;
1254 }
1255 
1256 static bool interrupt_check(struct svm_test *test)
1257 {
1258 	return get_test_stage(test) == 5;
1259 }
1260 
1261 static volatile bool nmi_fired;
1262 
1263 static void nmi_handler(struct ex_regs *regs)
1264 {
1265 	nmi_fired = true;
1266 }
1267 
1268 static void nmi_prepare(struct svm_test *test)
1269 {
1270 	default_prepare(test);
1271 	nmi_fired = false;
1272 	handle_exception(NMI_VECTOR, nmi_handler);
1273 	set_test_stage(test, 0);
1274 }
1275 
1276 static void nmi_test(struct svm_test *test)
1277 {
1278 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, 0);
1279 
1280 	report_svm_guest(nmi_fired, test, "direct NMI while running guest");
1281 
1282 	vmmcall();
1283 
1284 	nmi_fired = false;
1285 
1286 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, 0);
1287 
1288 	report_svm_guest(nmi_fired, test, "intercepted pending NMI delivered to guest");
1289 }
1290 
1291 static bool nmi_finished(struct svm_test *test)
1292 {
1293 	switch (get_test_stage(test)) {
1294 	case 0:
1295 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
1296 			report_fail("VMEXIT not due to vmmcall. Exit reason 0x%x",
1297 				    vmcb->control.exit_code);
1298 			return true;
1299 		}
1300 		vmcb->save.rip += 3;
1301 
1302 		vmcb->control.intercept |= (1ULL << INTERCEPT_NMI);
1303 		break;
1304 
1305 	case 1:
1306 		if (vmcb->control.exit_code != SVM_EXIT_NMI) {
1307 			report_fail("VMEXIT not due to NMI intercept. Exit reason 0x%x",
1308 				    vmcb->control.exit_code);
1309 			return true;
1310 		}
1311 
1312 		report_pass("NMI intercept while running guest");
1313 		break;
1314 
1315 	case 2:
1316 		break;
1317 
1318 	default:
1319 		return true;
1320 	}
1321 
1322 	inc_test_stage(test);
1323 
1324 	return get_test_stage(test) == 3;
1325 }
1326 
1327 static bool nmi_check(struct svm_test *test)
1328 {
1329 	return get_test_stage(test) == 3;
1330 }
1331 
1332 #define NMI_DELAY 100000000ULL
1333 
1334 static void nmi_message_thread(void *_test)
1335 {
1336 	struct svm_test *test = _test;
1337 
1338 	while (get_test_stage(test) != 1)
1339 		pause();
1340 
1341 	delay(NMI_DELAY);
1342 
1343 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1344 
1345 	while (get_test_stage(test) != 2)
1346 		pause();
1347 
1348 	delay(NMI_DELAY);
1349 
1350 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_NMI | APIC_INT_ASSERT, id_map[0]);
1351 }
1352 
1353 static void nmi_hlt_test(struct svm_test *test)
1354 {
1355 	long long start;
1356 
1357 	on_cpu_async(1, nmi_message_thread, test);
1358 
1359 	start = rdtsc();
1360 
1361 	set_test_stage(test, 1);
1362 
1363 	asm volatile ("hlt");
1364 
1365 	report_svm_guest(nmi_fired, test, "direct NMI + hlt");
1366 	report(rdtsc() - start > NMI_DELAY, "direct NMI after expected delay");
1367 
1368 	nmi_fired = false;
1369 
1370 	vmmcall();
1371 
1372 	start = rdtsc();
1373 
1374 	set_test_stage(test, 2);
1375 
1376 	asm volatile ("hlt");
1377 
1378 	report_svm_guest(nmi_fired, test, "intercepted NMI + hlt");
1379 	report(rdtsc() - start > NMI_DELAY, "intercepted NMI after expected delay");
1380 
1381 	set_test_stage(test, 3);
1382 }
1383 
1384 static bool nmi_hlt_finished(struct svm_test *test)
1385 {
1386 	switch (get_test_stage(test)) {
1387 	case 1:
1388 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
1389 			report_fail("VMEXIT not due to vmmcall. Exit reason 0x%x",
1390 				    vmcb->control.exit_code);
1391 			return true;
1392 		}
1393 		vmcb->save.rip += 3;
1394 
1395 		vmcb->control.intercept |= (1ULL << INTERCEPT_NMI);
1396 		break;
1397 
1398 	case 2:
1399 		if (vmcb->control.exit_code != SVM_EXIT_NMI) {
1400 			report_fail("VMEXIT not due to NMI intercept. Exit reason 0x%x",
1401 				    vmcb->control.exit_code);
1402 			return true;
1403 		}
1404 
1405 		report_pass("NMI intercept while running guest");
1406 		break;
1407 
1408 	case 3:
1409 		break;
1410 
1411 	default:
1412 		return true;
1413 	}
1414 
1415 	return get_test_stage(test) == 3;
1416 }
1417 
1418 static bool nmi_hlt_check(struct svm_test *test)
1419 {
1420 	return get_test_stage(test) == 3;
1421 }
1422 
1423 static void vnmi_prepare(struct svm_test *test)
1424 {
1425 	nmi_prepare(test);
1426 
1427 	/*
1428 	 * Disable NMI interception to start.  Enabling vNMI without
1429 	 * intercepting "real" NMIs should result in an ERR VM-Exit.
1430 	 */
1431 	vmcb->control.intercept &= ~(1ULL << INTERCEPT_NMI);
1432 	vmcb->control.int_ctl = V_NMI_ENABLE_MASK;
1433 	vmcb->control.int_vector = NMI_VECTOR;
1434 }
1435 
1436 static void vnmi_test(struct svm_test *test)
1437 {
1438 	report_svm_guest(!nmi_fired, test, "No vNMI before injection");
1439 	vmmcall();
1440 
1441 	report_svm_guest(nmi_fired, test, "vNMI delivered after injection");
1442 	vmmcall();
1443 }
1444 
1445 static bool vnmi_finished(struct svm_test *test)
1446 {
1447 	switch (get_test_stage(test)) {
1448 	case 0:
1449 		if (vmcb->control.exit_code != SVM_EXIT_ERR) {
1450 			report_fail("Wanted ERR VM-Exit, got 0x%x",
1451 				    vmcb->control.exit_code);
1452 			return true;
1453 		}
1454 		report(!nmi_fired, "vNMI enabled but NMI_INTERCEPT unset!");
1455 		vmcb->control.intercept |= (1ULL << INTERCEPT_NMI);
1456 		vmcb->save.rip += 3;
1457 		break;
1458 
1459 	case 1:
1460 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
1461 			report_fail("Wanted VMMCALL VM-Exit, got 0x%x",
1462 				    vmcb->control.exit_code);
1463 			return true;
1464 		}
1465 		report(!nmi_fired, "vNMI with vector 2 not injected");
1466 		vmcb->control.int_ctl |= V_NMI_PENDING_MASK;
1467 		vmcb->save.rip += 3;
1468 		break;
1469 
1470 	case 2:
1471 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
1472 			report_fail("Wanted VMMCALL VM-Exit, got 0x%x",
1473 				    vmcb->control.exit_code);
1474 			return true;
1475 		}
1476 		if (vmcb->control.int_ctl & V_NMI_BLOCKING_MASK) {
1477 			report_fail("V_NMI_BLOCKING_MASK not cleared on VMEXIT");
1478 			return true;
1479 		}
1480 		report_pass("VNMI serviced");
1481 		vmcb->save.rip += 3;
1482 		break;
1483 
1484 	default:
1485 		return true;
1486 	}
1487 
1488 	inc_test_stage(test);
1489 
1490 	return get_test_stage(test) == 3;
1491 }
1492 
1493 static bool vnmi_check(struct svm_test *test)
1494 {
1495 	return get_test_stage(test) == 3;
1496 }
1497 
1498 static volatile int count_exc = 0;
1499 
1500 static void my_isr(struct ex_regs *r)
1501 {
1502 	count_exc++;
1503 }
1504 
1505 static void exc_inject_prepare(struct svm_test *test)
1506 {
1507 	default_prepare(test);
1508 	handle_exception(DE_VECTOR, my_isr);
1509 	handle_exception(NMI_VECTOR, my_isr);
1510 }
1511 
1512 
1513 static void exc_inject_test(struct svm_test *test)
1514 {
1515 	asm volatile ("vmmcall\n\tvmmcall\n\t");
1516 }
1517 
1518 static bool exc_inject_finished(struct svm_test *test)
1519 {
1520 	switch (get_test_stage(test)) {
1521 	case 0:
1522 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
1523 			report_fail("VMEXIT not due to vmmcall. Exit reason 0x%x",
1524 				    vmcb->control.exit_code);
1525 			return true;
1526 		}
1527 		vmcb->save.rip += 3;
1528 		vmcb->control.event_inj = NMI_VECTOR | SVM_EVTINJ_TYPE_EXEPT | SVM_EVTINJ_VALID;
1529 		break;
1530 
1531 	case 1:
1532 		if (vmcb->control.exit_code != SVM_EXIT_ERR) {
1533 			report_fail("VMEXIT not due to error. Exit reason 0x%x",
1534 				    vmcb->control.exit_code);
1535 			return true;
1536 		}
1537 		report(count_exc == 0, "exception with vector 2 not injected");
1538 		vmcb->control.event_inj = DE_VECTOR | SVM_EVTINJ_TYPE_EXEPT | SVM_EVTINJ_VALID;
1539 		break;
1540 
1541 	case 2:
1542 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
1543 			report_fail("VMEXIT not due to vmmcall. Exit reason 0x%x",
1544 				    vmcb->control.exit_code);
1545 			return true;
1546 		}
1547 		vmcb->save.rip += 3;
1548 		report(count_exc == 1, "divide overflow exception injected");
1549 		report(!(vmcb->control.event_inj & SVM_EVTINJ_VALID), "eventinj.VALID cleared");
1550 		break;
1551 
1552 	default:
1553 		return true;
1554 	}
1555 
1556 	inc_test_stage(test);
1557 
1558 	return get_test_stage(test) == 3;
1559 }
1560 
1561 static bool exc_inject_check(struct svm_test *test)
1562 {
1563 	return count_exc == 1 && get_test_stage(test) == 3;
1564 }
1565 
1566 static volatile bool virq_fired;
1567 
1568 static void virq_isr(isr_regs_t *regs)
1569 {
1570 	virq_fired = true;
1571 }
1572 
1573 static void virq_inject_prepare(struct svm_test *test)
1574 {
1575 	handle_irq(0xf1, virq_isr);
1576 	default_prepare(test);
1577 	vmcb->control.int_ctl = V_INTR_MASKING_MASK | V_IRQ_MASK |
1578 		(0x0f << V_INTR_PRIO_SHIFT); // Set to the highest priority
1579 	vmcb->control.int_vector = 0xf1;
1580 	virq_fired = false;
1581 	set_test_stage(test, 0);
1582 }
1583 
1584 static void virq_inject_test(struct svm_test *test)
1585 {
1586 	report_svm_guest(!virq_fired, test, "virtual IRQ blocked after L2 cli");
1587 
1588 	irq_enable();
1589 	asm volatile ("nop");
1590 	irq_disable();
1591 
1592 	report_svm_guest(virq_fired, test, "virtual IRQ fired after L2 sti");
1593 
1594 	vmmcall();
1595 
1596 	report_svm_guest(!virq_fired, test, "intercepted VINTR blocked after L2 cli");
1597 
1598 	irq_enable();
1599 	asm volatile ("nop");
1600 	irq_disable();
1601 
1602 	report_svm_guest(virq_fired, test, "intercepted VINTR fired after L2 sti");
1603 
1604 	vmmcall();
1605 
1606 	irq_enable();
1607 	asm volatile ("nop");
1608 	irq_disable();
1609 
1610 	report_svm_guest(!virq_fired, test,
1611 			  "virtual IRQ blocked V_IRQ_PRIO less than V_TPR");
1612 
1613 	vmmcall();
1614 	vmmcall();
1615 }
1616 
1617 static bool virq_inject_finished(struct svm_test *test)
1618 {
1619 	vmcb->save.rip += 3;
1620 
1621 	switch (get_test_stage(test)) {
1622 	case 0:
1623 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
1624 			report_fail("VMEXIT not due to vmmcall. Exit reason 0x%x",
1625 				    vmcb->control.exit_code);
1626 			return true;
1627 		}
1628 		if (vmcb->control.int_ctl & V_IRQ_MASK) {
1629 			report_fail("V_IRQ not cleared on VMEXIT after firing");
1630 			return true;
1631 		}
1632 		virq_fired = false;
1633 		vmcb->control.intercept |= (1ULL << INTERCEPT_VINTR);
1634 		vmcb->control.int_ctl = V_INTR_MASKING_MASK | V_IRQ_MASK |
1635 			(0x0f << V_INTR_PRIO_SHIFT);
1636 		break;
1637 
1638 	case 1:
1639 		if (vmcb->control.exit_code != SVM_EXIT_VINTR) {
1640 			report_fail("VMEXIT not due to vintr. Exit reason 0x%x",
1641 				    vmcb->control.exit_code);
1642 			return true;
1643 		}
1644 		if (virq_fired) {
1645 			report_fail("V_IRQ fired before SVM_EXIT_VINTR");
1646 			return true;
1647 		}
1648 		vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1649 		break;
1650 
1651 	case 2:
1652 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
1653 			report_fail("VMEXIT not due to vmmcall. Exit reason 0x%x",
1654 				    vmcb->control.exit_code);
1655 			return true;
1656 		}
1657 		virq_fired = false;
1658 		// Set irq to lower priority
1659 		vmcb->control.int_ctl = V_INTR_MASKING_MASK | V_IRQ_MASK |
1660 			(0x08 << V_INTR_PRIO_SHIFT);
1661 		// Raise guest TPR
1662 		vmcb->control.int_ctl |= 0x0a & V_TPR_MASK;
1663 		break;
1664 
1665 	case 3:
1666 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
1667 			report_fail("VMEXIT not due to vmmcall. Exit reason 0x%x",
1668 				    vmcb->control.exit_code);
1669 			return true;
1670 		}
1671 		vmcb->control.intercept |= (1ULL << INTERCEPT_VINTR);
1672 		break;
1673 
1674 	case 4:
1675 		// INTERCEPT_VINTR should be ignored because V_INTR_PRIO < V_TPR
1676 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
1677 			report_fail("VMEXIT not due to vmmcall. Exit reason 0x%x",
1678 				    vmcb->control.exit_code);
1679 			return true;
1680 		}
1681 		break;
1682 
1683 	default:
1684 		return true;
1685 	}
1686 
1687 	inc_test_stage(test);
1688 
1689 	return get_test_stage(test) == 5;
1690 }
1691 
1692 static bool virq_inject_check(struct svm_test *test)
1693 {
1694 	return get_test_stage(test) == 5;
1695 }
1696 
1697 /*
1698  * Detect nested guest RIP corruption as explained in kernel commit
1699  * b6162e82aef19fee9c32cb3fe9ac30d9116a8c73
1700  *
1701  * In the assembly loop below 'ins' is executed while IO instructions
1702  * are not intercepted; the instruction is emulated by L0.
1703  *
1704  * At the same time we are getting interrupts from the local APIC timer,
1705  * and we do intercept them in L1
1706  *
1707  * If the interrupt happens on the insb instruction, L0 will VMexit, emulate
1708  * the insb instruction and then it will inject the interrupt to L1 through
1709  * a nested VMexit.  Due to a bug, it would leave pre-emulation values of RIP,
1710  * RAX and RSP in the VMCB.
1711  *
1712  * In our intercept handler we detect the bug by checking that RIP is that of
1713  * the insb instruction, but its memory operand has already been written.
1714  * This means that insb was already executed.
1715  */
1716 
1717 static volatile int isr_cnt = 0;
1718 static volatile uint8_t io_port_var = 0xAA;
1719 extern const char insb_instruction_label[];
1720 
1721 static void reg_corruption_isr(isr_regs_t *regs)
1722 {
1723 	isr_cnt++;
1724 	apic_write(APIC_EOI, 0);
1725 }
1726 
1727 static void reg_corruption_prepare(struct svm_test *test)
1728 {
1729 	default_prepare(test);
1730 	set_test_stage(test, 0);
1731 
1732 	vmcb->control.int_ctl = V_INTR_MASKING_MASK;
1733 	vmcb->control.intercept |= (1ULL << INTERCEPT_INTR);
1734 
1735 	handle_irq(TIMER_VECTOR, reg_corruption_isr);
1736 
1737 	/* set local APIC to inject external interrupts */
1738 	apic_write(APIC_TMICT, 0);
1739 	apic_write(APIC_TDCR, 0);
1740 	apic_write(APIC_LVTT, TIMER_VECTOR | APIC_LVT_TIMER_PERIODIC);
1741 	apic_write(APIC_TMICT, 1000);
1742 }
1743 
1744 static void reg_corruption_test(struct svm_test *test)
1745 {
1746 	/* this is endless loop, which is interrupted by the timer interrupt */
1747 	asm volatile (
1748 		      "1:\n\t"
1749 		      "movw $0x4d0, %%dx\n\t" // IO port
1750 		      "lea %[io_port_var], %%rdi\n\t"
1751 		      "movb $0xAA, %[io_port_var]\n\t"
1752 		      "insb_instruction_label:\n\t"
1753 		      "insb\n\t"
1754 		      "jmp 1b\n\t"
1755 
1756 		      : [io_port_var] "=m" (io_port_var)
1757 		      : /* no inputs*/
1758 		      : "rdx", "rdi"
1759 		      );
1760 }
1761 
1762 static bool reg_corruption_finished(struct svm_test *test)
1763 {
1764 	if (isr_cnt == 10000) {
1765 		report_pass("No RIP corruption detected after %d timer interrupts",
1766 			    isr_cnt);
1767 		set_test_stage(test, 1);
1768 		goto cleanup;
1769 	}
1770 
1771 	if (vmcb->control.exit_code == SVM_EXIT_INTR) {
1772 
1773 		void* guest_rip = (void*)vmcb->save.rip;
1774 
1775 		irq_enable();
1776 		asm volatile ("nop");
1777 		irq_disable();
1778 
1779 		if (guest_rip == insb_instruction_label && io_port_var != 0xAA) {
1780 			report_fail("RIP corruption detected after %d timer interrupts",
1781 				    isr_cnt);
1782 			goto cleanup;
1783 		}
1784 
1785 	}
1786 	return false;
1787 cleanup:
1788 	apic_write(APIC_LVTT, APIC_LVT_TIMER_MASK);
1789 	apic_write(APIC_TMICT, 0);
1790 	return true;
1791 
1792 }
1793 
1794 static bool reg_corruption_check(struct svm_test *test)
1795 {
1796 	return get_test_stage(test) == 1;
1797 }
1798 
1799 static void get_tss_entry(void *data)
1800 {
1801 	*((gdt_entry_t **)data) = get_tss_descr();
1802 }
1803 
1804 static int orig_cpu_count;
1805 
1806 static void init_startup_prepare(struct svm_test *test)
1807 {
1808 	gdt_entry_t *tss_entry;
1809 	int i;
1810 
1811 	on_cpu(1, get_tss_entry, &tss_entry);
1812 
1813 	orig_cpu_count = atomic_read(&cpu_online_count);
1814 
1815 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT,
1816 		       id_map[1]);
1817 
1818 	delay(100000000ULL);
1819 
1820 	atomic_dec(&cpu_online_count);
1821 
1822 	tss_entry->type &= ~DESC_BUSY;
1823 
1824 	apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_STARTUP, id_map[1]);
1825 
1826 	for (i = 0; i < 5 && atomic_read(&cpu_online_count) < orig_cpu_count; i++)
1827 		delay(100000000ULL);
1828 }
1829 
1830 static bool init_startup_finished(struct svm_test *test)
1831 {
1832 	return true;
1833 }
1834 
1835 static bool init_startup_check(struct svm_test *test)
1836 {
1837 	return atomic_read(&cpu_online_count) == orig_cpu_count;
1838 }
1839 
1840 static volatile bool init_intercept;
1841 
1842 static void init_intercept_prepare(struct svm_test *test)
1843 {
1844 	init_intercept = false;
1845 	vmcb->control.intercept |= (1ULL << INTERCEPT_INIT);
1846 }
1847 
1848 static void init_intercept_test(struct svm_test *test)
1849 {
1850 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL | APIC_DM_INIT | APIC_INT_ASSERT, 0);
1851 }
1852 
1853 static bool init_intercept_finished(struct svm_test *test)
1854 {
1855 	vmcb->save.rip += 3;
1856 
1857 	if (vmcb->control.exit_code != SVM_EXIT_INIT) {
1858 		report_fail("VMEXIT not due to init intercept. Exit reason 0x%x",
1859 			    vmcb->control.exit_code);
1860 
1861 		return true;
1862 	}
1863 
1864 	init_intercept = true;
1865 
1866 	report_pass("INIT to vcpu intercepted");
1867 
1868 	return true;
1869 }
1870 
1871 static bool init_intercept_check(struct svm_test *test)
1872 {
1873 	return init_intercept;
1874 }
1875 
1876 /*
1877  * Setting host EFLAGS.TF causes a #DB trap after the VMRUN completes on the
1878  * host side (i.e., after the #VMEXIT from the guest).
1879  *
1880  * Setting host EFLAGS.RF suppresses any potential instruction breakpoint
1881  * match on the VMRUN and completion of the VMRUN instruction clears the
1882  * host EFLAGS.RF bit.
1883  *
1884  * [AMD APM]
1885  */
1886 static volatile u8 host_rflags_guest_main_flag = 0;
1887 static volatile u8 host_rflags_db_handler_flag = 0;
1888 static volatile bool host_rflags_ss_on_vmrun = false;
1889 static volatile bool host_rflags_vmrun_reached = false;
1890 static volatile bool host_rflags_set_tf = false;
1891 static volatile bool host_rflags_set_rf = false;
1892 static u64 rip_detected;
1893 
1894 extern u64 *vmrun_rip;
1895 
1896 static void host_rflags_db_handler(struct ex_regs *r)
1897 {
1898 	if (host_rflags_ss_on_vmrun) {
1899 		if (host_rflags_vmrun_reached) {
1900 			if (!host_rflags_set_rf) {
1901 				r->rflags &= ~X86_EFLAGS_TF;
1902 				rip_detected = r->rip;
1903 			} else {
1904 				r->rflags |= X86_EFLAGS_RF;
1905 				++host_rflags_db_handler_flag;
1906 			}
1907 		} else {
1908 			if (r->rip == (u64)&vmrun_rip) {
1909 				host_rflags_vmrun_reached = true;
1910 
1911 				if (host_rflags_set_rf) {
1912 					host_rflags_guest_main_flag = 0;
1913 					rip_detected = r->rip;
1914 					r->rflags &= ~X86_EFLAGS_TF;
1915 
1916 					/* Trigger #DB via debug registers */
1917 					write_dr0((void *)&vmrun_rip);
1918 					write_dr7(0x403);
1919 				}
1920 			}
1921 		}
1922 	} else {
1923 		r->rflags &= ~X86_EFLAGS_TF;
1924 	}
1925 }
1926 
1927 static void host_rflags_prepare(struct svm_test *test)
1928 {
1929 	default_prepare(test);
1930 	handle_exception(DB_VECTOR, host_rflags_db_handler);
1931 	set_test_stage(test, 0);
1932 }
1933 
1934 static void host_rflags_prepare_gif_clear(struct svm_test *test)
1935 {
1936 	if (host_rflags_set_tf)
1937 		write_rflags(read_rflags() | X86_EFLAGS_TF);
1938 }
1939 
1940 static void host_rflags_test(struct svm_test *test)
1941 {
1942 	while (1) {
1943 		if (get_test_stage(test) > 0) {
1944 			if ((host_rflags_set_tf && !host_rflags_ss_on_vmrun && !host_rflags_db_handler_flag) ||
1945 			    (host_rflags_set_rf && host_rflags_db_handler_flag == 1))
1946 				host_rflags_guest_main_flag = 1;
1947 		}
1948 
1949 		if (get_test_stage(test) == 4)
1950 			break;
1951 		vmmcall();
1952 	}
1953 }
1954 
1955 static bool host_rflags_finished(struct svm_test *test)
1956 {
1957 	switch (get_test_stage(test)) {
1958 	case 0:
1959 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
1960 			report_fail("Unexpected VMEXIT. Exit reason 0x%x",
1961 				    vmcb->control.exit_code);
1962 			return true;
1963 		}
1964 		vmcb->save.rip += 3;
1965 		/*
1966 		 * Setting host EFLAGS.TF not immediately before VMRUN, causes
1967 		 * #DB trap before first guest instruction is executed
1968 		 */
1969 		host_rflags_set_tf = true;
1970 		break;
1971 	case 1:
1972 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL ||
1973 		    host_rflags_guest_main_flag != 1) {
1974 			report_fail("Unexpected VMEXIT or #DB handler"
1975 				    " invoked before guest main. Exit reason 0x%x",
1976 				    vmcb->control.exit_code);
1977 			return true;
1978 		}
1979 		vmcb->save.rip += 3;
1980 		/*
1981 		 * Setting host EFLAGS.TF immediately before VMRUN, causes #DB
1982 		 * trap after VMRUN completes on the host side (i.e., after
1983 		 * VMEXIT from guest).
1984 		 */
1985 		host_rflags_ss_on_vmrun = true;
1986 		break;
1987 	case 2:
1988 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL ||
1989 		    rip_detected != (u64)&vmrun_rip + 3) {
1990 			report_fail("Unexpected VMEXIT or RIP mismatch."
1991 				    " Exit reason 0x%x, RIP actual: %lx, RIP expected: "
1992 				    "%lx", vmcb->control.exit_code,
1993 				    (u64)&vmrun_rip + 3, rip_detected);
1994 			return true;
1995 		}
1996 		host_rflags_set_rf = true;
1997 		host_rflags_guest_main_flag = 0;
1998 		host_rflags_vmrun_reached = false;
1999 		vmcb->save.rip += 3;
2000 		break;
2001 	case 3:
2002 		if (vmcb->control.exit_code != SVM_EXIT_VMMCALL ||
2003 		    rip_detected != (u64)&vmrun_rip ||
2004 		    host_rflags_guest_main_flag != 1 ||
2005 		    host_rflags_db_handler_flag > 1 ||
2006 		    read_rflags() & X86_EFLAGS_RF) {
2007 			report_fail("Unexpected VMEXIT or RIP mismatch or "
2008 				    "EFLAGS.RF not cleared."
2009 				    " Exit reason 0x%x, RIP actual: %lx, RIP expected: "
2010 				    "%lx", vmcb->control.exit_code,
2011 				    (u64)&vmrun_rip, rip_detected);
2012 			return true;
2013 		}
2014 		host_rflags_set_tf = false;
2015 		host_rflags_set_rf = false;
2016 		vmcb->save.rip += 3;
2017 		break;
2018 	default:
2019 		return true;
2020 	}
2021 	inc_test_stage(test);
2022 	return get_test_stage(test) == 5;
2023 }
2024 
2025 static bool host_rflags_check(struct svm_test *test)
2026 {
2027 	return get_test_stage(test) == 4;
2028 }
2029 
2030 #define TEST(name) { #name, .v2 = name }
2031 
2032 /*
2033  * v2 tests
2034  */
2035 
2036 /*
2037  * Ensure that kvm recalculates the L1 guest's CPUID.01H:ECX.OSXSAVE
2038  * after VM-exit from an L2 guest that sets CR4.OSXSAVE to a different
2039  * value than in L1.
2040  */
2041 
2042 static void svm_cr4_osxsave_test_guest(struct svm_test *test)
2043 {
2044 	write_cr4(read_cr4() & ~X86_CR4_OSXSAVE);
2045 }
2046 
2047 static void svm_cr4_osxsave_test(void)
2048 {
2049 	if (!this_cpu_has(X86_FEATURE_XSAVE)) {
2050 		report_skip("XSAVE not detected");
2051 		return;
2052 	}
2053 
2054 	if (!(read_cr4() & X86_CR4_OSXSAVE)) {
2055 		unsigned long cr4 = read_cr4() | X86_CR4_OSXSAVE;
2056 
2057 		write_cr4(cr4);
2058 		vmcb->save.cr4 = cr4;
2059 	}
2060 
2061 	report(this_cpu_has(X86_FEATURE_OSXSAVE), "CPUID.01H:ECX.XSAVE set before VMRUN");
2062 
2063 	test_set_guest(svm_cr4_osxsave_test_guest);
2064 	report(svm_vmrun() == SVM_EXIT_VMMCALL,
2065 	       "svm_cr4_osxsave_test_guest finished with VMMCALL");
2066 
2067 	report(this_cpu_has(X86_FEATURE_OSXSAVE), "CPUID.01H:ECX.XSAVE set after VMRUN");
2068 }
2069 
2070 static void basic_guest_main(struct svm_test *test)
2071 {
2072 }
2073 
2074 
2075 #define SVM_TEST_REG_RESERVED_BITS(start, end, inc, str_name, reg, val,	\
2076 				   resv_mask)				\
2077 {									\
2078 	u64 tmp, mask;							\
2079 	int i;								\
2080 									\
2081 	for (i = start; i <= end; i = i + inc) {			\
2082 		mask = 1ull << i;					\
2083 		if (!(mask & resv_mask))				\
2084 			continue;					\
2085 		tmp = val | mask;					\
2086 		reg = tmp;						\
2087 		report(svm_vmrun() == SVM_EXIT_ERR, "Test %s %d:%d: %lx", \
2088 		       str_name, end, start, tmp);			\
2089 	}								\
2090 }
2091 
2092 #define SVM_TEST_CR_RESERVED_BITS(start, end, inc, cr, val, resv_mask,	\
2093 				  exit_code, test_name)			\
2094 {									\
2095 	u64 tmp, mask;							\
2096 	u32 r;								\
2097 	int i;								\
2098 									\
2099 	for (i = start; i <= end; i = i + inc) {			\
2100 		mask = 1ull << i;					\
2101 		if (!(mask & resv_mask))				\
2102 			continue;					\
2103 		tmp = val | mask;					\
2104 		switch (cr) {						\
2105 		case 0:							\
2106 			vmcb->save.cr0 = tmp;				\
2107 			break;						\
2108 		case 3:							\
2109 			vmcb->save.cr3 = tmp;				\
2110 			break;						\
2111 		case 4:							\
2112 			vmcb->save.cr4 = tmp;				\
2113 		}							\
2114 		r = svm_vmrun();					\
2115 		report(r == exit_code, "Test CR%d %s%d:%d: %lx, wanted exit 0x%x, got 0x%x", \
2116 		       cr, test_name, end, start, tmp, exit_code, r);	\
2117 	}								\
2118 }
2119 
2120 static void test_efer(void)
2121 {
2122 	/*
2123 	 * Un-setting EFER.SVME is illegal
2124 	 */
2125 	u64 efer_saved = vmcb->save.efer;
2126 	u64 efer = efer_saved;
2127 
2128 	report (svm_vmrun() == SVM_EXIT_VMMCALL, "EFER.SVME: %lx", efer);
2129 	efer &= ~EFER_SVME;
2130 	vmcb->save.efer = efer;
2131 	report (svm_vmrun() == SVM_EXIT_ERR, "EFER.SVME: %lx", efer);
2132 	vmcb->save.efer = efer_saved;
2133 
2134 	/*
2135 	 * EFER MBZ bits: 63:16, 9
2136 	 */
2137 	efer_saved = vmcb->save.efer;
2138 
2139 	SVM_TEST_REG_RESERVED_BITS(8, 9, 1, "EFER", vmcb->save.efer,
2140 				   efer_saved, SVM_EFER_RESERVED_MASK);
2141 	SVM_TEST_REG_RESERVED_BITS(16, 63, 4, "EFER", vmcb->save.efer,
2142 				   efer_saved, SVM_EFER_RESERVED_MASK);
2143 
2144 	/*
2145 	 * EFER.LME and CR0.PG are both set and CR4.PAE is zero.
2146 	 */
2147 	u64 cr0_saved = vmcb->save.cr0;
2148 	u64 cr0;
2149 	u64 cr4_saved = vmcb->save.cr4;
2150 	u64 cr4;
2151 
2152 	efer = efer_saved | EFER_LME;
2153 	vmcb->save.efer = efer;
2154 	cr0 = cr0_saved | X86_CR0_PG | X86_CR0_PE;
2155 	vmcb->save.cr0 = cr0;
2156 	cr4 = cr4_saved & ~X86_CR4_PAE;
2157 	vmcb->save.cr4 = cr4;
2158 	report(svm_vmrun() == SVM_EXIT_ERR, "EFER.LME=1 (%lx), "
2159 	       "CR0.PG=1 (%lx) and CR4.PAE=0 (%lx)", efer, cr0, cr4);
2160 
2161 	/*
2162 	 * EFER.LME and CR0.PG are both set and CR0.PE is zero.
2163 	 * CR4.PAE needs to be set as we otherwise cannot
2164 	 * determine if CR4.PAE=0 or CR0.PE=0 triggered the
2165 	 * SVM_EXIT_ERR.
2166 	 */
2167 	cr4 = cr4_saved | X86_CR4_PAE;
2168 	vmcb->save.cr4 = cr4;
2169 	cr0 &= ~X86_CR0_PE;
2170 	vmcb->save.cr0 = cr0;
2171 	report(svm_vmrun() == SVM_EXIT_ERR, "EFER.LME=1 (%lx), "
2172 	       "CR0.PG=1 and CR0.PE=0 (%lx)", efer, cr0);
2173 
2174 	/*
2175 	 * EFER.LME, CR0.PG, CR4.PAE, CS.L, and CS.D are all non-zero.
2176 	 */
2177 	u32 cs_attrib_saved = vmcb->save.cs.attrib;
2178 	u32 cs_attrib;
2179 
2180 	cr0 |= X86_CR0_PE;
2181 	vmcb->save.cr0 = cr0;
2182 	cs_attrib = cs_attrib_saved | SVM_SELECTOR_L_MASK |
2183 		SVM_SELECTOR_DB_MASK;
2184 	vmcb->save.cs.attrib = cs_attrib;
2185 	report(svm_vmrun() == SVM_EXIT_ERR, "EFER.LME=1 (%lx), "
2186 	       "CR0.PG=1 (%lx), CR4.PAE=1 (%lx), CS.L=1 and CS.D=1 (%x)",
2187 	       efer, cr0, cr4, cs_attrib);
2188 
2189 	vmcb->save.cr0 = cr0_saved;
2190 	vmcb->save.cr4 = cr4_saved;
2191 	vmcb->save.efer = efer_saved;
2192 	vmcb->save.cs.attrib = cs_attrib_saved;
2193 }
2194 
2195 static void test_cr0(void)
2196 {
2197 	/*
2198 	 * Un-setting CR0.CD and setting CR0.NW is illegal combination
2199 	 */
2200 	u64 cr0_saved = vmcb->save.cr0;
2201 	u64 cr0 = cr0_saved;
2202 
2203 	cr0 |= X86_CR0_CD;
2204 	cr0 &= ~X86_CR0_NW;
2205 	vmcb->save.cr0 = cr0;
2206 	report (svm_vmrun() == SVM_EXIT_VMMCALL, "Test CR0 CD=1,NW=0: %lx",
2207 		cr0);
2208 	cr0 |= X86_CR0_NW;
2209 	vmcb->save.cr0 = cr0;
2210 	report (svm_vmrun() == SVM_EXIT_VMMCALL, "Test CR0 CD=1,NW=1: %lx",
2211 		cr0);
2212 	cr0 &= ~X86_CR0_NW;
2213 	cr0 &= ~X86_CR0_CD;
2214 	vmcb->save.cr0 = cr0;
2215 	report (svm_vmrun() == SVM_EXIT_VMMCALL, "Test CR0 CD=0,NW=0: %lx",
2216 		cr0);
2217 	cr0 |= X86_CR0_NW;
2218 	vmcb->save.cr0 = cr0;
2219 	report (svm_vmrun() == SVM_EXIT_ERR, "Test CR0 CD=0,NW=1: %lx",
2220 		cr0);
2221 	vmcb->save.cr0 = cr0_saved;
2222 
2223 	/*
2224 	 * CR0[63:32] are not zero
2225 	 */
2226 	cr0 = cr0_saved;
2227 
2228 	SVM_TEST_REG_RESERVED_BITS(32, 63, 4, "CR0", vmcb->save.cr0, cr0_saved,
2229 				   SVM_CR0_RESERVED_MASK);
2230 	vmcb->save.cr0 = cr0_saved;
2231 }
2232 
2233 static void test_cr3(void)
2234 {
2235 	/*
2236 	 * CR3 MBZ bits based on different modes:
2237 	 *   [63:52] - long mode
2238 	 */
2239 	u64 cr3_saved = vmcb->save.cr3;
2240 
2241 	SVM_TEST_CR_RESERVED_BITS(0, 63, 1, 3, cr3_saved,
2242 				  SVM_CR3_LONG_MBZ_MASK, SVM_EXIT_ERR, "");
2243 
2244 	vmcb->save.cr3 = cr3_saved & ~SVM_CR3_LONG_MBZ_MASK;
2245 	report(svm_vmrun() == SVM_EXIT_VMMCALL, "Test CR3 63:0: %lx",
2246 	       vmcb->save.cr3);
2247 
2248 	/*
2249 	 * CR3 non-MBZ reserved bits based on different modes:
2250 	 *   [11:5] [2:0] - long mode (PCIDE=0)
2251 	 *          [2:0] - PAE legacy mode
2252 	 */
2253 	u64 cr4_saved = vmcb->save.cr4;
2254 	u64 *pdpe = npt_get_pml4e();
2255 
2256 	/*
2257 	 * Long mode
2258 	 */
2259 	if (this_cpu_has(X86_FEATURE_PCID)) {
2260 		vmcb->save.cr4 = cr4_saved | X86_CR4_PCIDE;
2261 		SVM_TEST_CR_RESERVED_BITS(0, 11, 1, 3, cr3_saved,
2262 					  SVM_CR3_LONG_RESERVED_MASK, SVM_EXIT_VMMCALL, "(PCIDE=1) ");
2263 
2264 		vmcb->save.cr3 = cr3_saved & ~SVM_CR3_LONG_RESERVED_MASK;
2265 		report(svm_vmrun() == SVM_EXIT_VMMCALL, "Test CR3 63:0: %lx",
2266 		       vmcb->save.cr3);
2267 	}
2268 
2269 	vmcb->save.cr4 = cr4_saved & ~X86_CR4_PCIDE;
2270 
2271 	if (!npt_supported())
2272 		goto skip_npt_only;
2273 
2274 	/* Clear P (Present) bit in NPT in order to trigger #NPF */
2275 	pdpe[0] &= ~1ULL;
2276 
2277 	SVM_TEST_CR_RESERVED_BITS(0, 11, 1, 3, cr3_saved,
2278 				  SVM_CR3_LONG_RESERVED_MASK, SVM_EXIT_NPF, "(PCIDE=0) ");
2279 
2280 	pdpe[0] |= 1ULL;
2281 	vmcb->save.cr3 = cr3_saved;
2282 
2283 	/*
2284 	 * PAE legacy
2285 	 */
2286 	pdpe[0] &= ~1ULL;
2287 	vmcb->save.cr4 = cr4_saved | X86_CR4_PAE;
2288 	SVM_TEST_CR_RESERVED_BITS(0, 2, 1, 3, cr3_saved,
2289 				  SVM_CR3_PAE_LEGACY_RESERVED_MASK, SVM_EXIT_NPF, "(PAE) ");
2290 
2291 	pdpe[0] |= 1ULL;
2292 
2293 skip_npt_only:
2294 	vmcb->save.cr3 = cr3_saved;
2295 	vmcb->save.cr4 = cr4_saved;
2296 }
2297 
2298 /* Test CR4 MBZ bits based on legacy or long modes */
2299 static void test_cr4(void)
2300 {
2301 	u64 cr4_saved = vmcb->save.cr4;
2302 	u64 efer_saved = vmcb->save.efer;
2303 	u64 efer = efer_saved;
2304 
2305 	efer &= ~EFER_LME;
2306 	vmcb->save.efer = efer;
2307 	SVM_TEST_CR_RESERVED_BITS(12, 31, 1, 4, cr4_saved,
2308 				  SVM_CR4_LEGACY_RESERVED_MASK, SVM_EXIT_ERR, "");
2309 
2310 	efer |= EFER_LME;
2311 	vmcb->save.efer = efer;
2312 	SVM_TEST_CR_RESERVED_BITS(12, 31, 1, 4, cr4_saved,
2313 				  SVM_CR4_RESERVED_MASK, SVM_EXIT_ERR, "");
2314 	SVM_TEST_CR_RESERVED_BITS(32, 63, 4, 4, cr4_saved,
2315 				  SVM_CR4_RESERVED_MASK, SVM_EXIT_ERR, "");
2316 
2317 	vmcb->save.cr4 = cr4_saved;
2318 	vmcb->save.efer = efer_saved;
2319 }
2320 
2321 static void test_dr(void)
2322 {
2323 	/*
2324 	 * DR6[63:32] and DR7[63:32] are MBZ
2325 	 */
2326 	u64 dr_saved = vmcb->save.dr6;
2327 
2328 	SVM_TEST_REG_RESERVED_BITS(32, 63, 4, "DR6", vmcb->save.dr6, dr_saved,
2329 				   SVM_DR6_RESERVED_MASK);
2330 	vmcb->save.dr6 = dr_saved;
2331 
2332 	dr_saved = vmcb->save.dr7;
2333 	SVM_TEST_REG_RESERVED_BITS(32, 63, 4, "DR7", vmcb->save.dr7, dr_saved,
2334 				   SVM_DR7_RESERVED_MASK);
2335 
2336 	vmcb->save.dr7 = dr_saved;
2337 }
2338 
2339 /* TODO: verify if high 32-bits are sign- or zero-extended on bare metal */
2340 #define	TEST_BITMAP_ADDR(save_intercept, type, addr, exit_code,		\
2341 			 msg) {						\
2342 		vmcb->control.intercept = saved_intercept | 1ULL << type; \
2343 		if (type == INTERCEPT_MSR_PROT)				\
2344 			vmcb->control.msrpm_base_pa = addr;		\
2345 		else							\
2346 			vmcb->control.iopm_base_pa = addr;		\
2347 		report(svm_vmrun() == exit_code,			\
2348 		       "Test %s address: %lx", msg, addr);		\
2349 	}
2350 
2351 /*
2352  * If the MSR or IOIO intercept table extends to a physical address that
2353  * is greater than or equal to the maximum supported physical address, the
2354  * guest state is illegal.
2355  *
2356  * The VMRUN instruction ignores the lower 12 bits of the address specified
2357  * in the VMCB.
2358  *
2359  * MSRPM spans 2 contiguous 4KB pages while IOPM spans 2 contiguous 4KB
2360  * pages + 1 byte.
2361  *
2362  * [APM vol 2]
2363  *
2364  * Note: Unallocated MSRPM addresses conforming to consistency checks, generate
2365  * #NPF.
2366  */
2367 static void test_msrpm_iopm_bitmap_addrs(void)
2368 {
2369 	u64 saved_intercept = vmcb->control.intercept;
2370 	u64 addr_beyond_limit = 1ull << cpuid_maxphyaddr();
2371 	u64 addr = virt_to_phys(msr_bitmap) & (~((1ull << 12) - 1));
2372 
2373 	TEST_BITMAP_ADDR(saved_intercept, INTERCEPT_MSR_PROT,
2374 			 addr_beyond_limit - 2 * PAGE_SIZE, SVM_EXIT_ERR,
2375 			 "MSRPM");
2376 	TEST_BITMAP_ADDR(saved_intercept, INTERCEPT_MSR_PROT,
2377 			 addr_beyond_limit - 2 * PAGE_SIZE + 1, SVM_EXIT_ERR,
2378 			 "MSRPM");
2379 	TEST_BITMAP_ADDR(saved_intercept, INTERCEPT_MSR_PROT,
2380 			 addr_beyond_limit - PAGE_SIZE, SVM_EXIT_ERR,
2381 			 "MSRPM");
2382 	TEST_BITMAP_ADDR(saved_intercept, INTERCEPT_MSR_PROT, addr,
2383 			 SVM_EXIT_VMMCALL, "MSRPM");
2384 	addr |= (1ull << 12) - 1;
2385 	TEST_BITMAP_ADDR(saved_intercept, INTERCEPT_MSR_PROT, addr,
2386 			 SVM_EXIT_VMMCALL, "MSRPM");
2387 
2388 	TEST_BITMAP_ADDR(saved_intercept, INTERCEPT_IOIO_PROT,
2389 			 addr_beyond_limit - 4 * PAGE_SIZE, SVM_EXIT_VMMCALL,
2390 			 "IOPM");
2391 	TEST_BITMAP_ADDR(saved_intercept, INTERCEPT_IOIO_PROT,
2392 			 addr_beyond_limit - 3 * PAGE_SIZE, SVM_EXIT_VMMCALL,
2393 			 "IOPM");
2394 	TEST_BITMAP_ADDR(saved_intercept, INTERCEPT_IOIO_PROT,
2395 			 addr_beyond_limit - 2 * PAGE_SIZE - 2, SVM_EXIT_VMMCALL,
2396 			 "IOPM");
2397 	TEST_BITMAP_ADDR(saved_intercept, INTERCEPT_IOIO_PROT,
2398 			 addr_beyond_limit - 2 * PAGE_SIZE, SVM_EXIT_ERR,
2399 			 "IOPM");
2400 	TEST_BITMAP_ADDR(saved_intercept, INTERCEPT_IOIO_PROT,
2401 			 addr_beyond_limit - PAGE_SIZE, SVM_EXIT_ERR,
2402 			 "IOPM");
2403 	addr = virt_to_phys(io_bitmap) & (~((1ull << 11) - 1));
2404 	TEST_BITMAP_ADDR(saved_intercept, INTERCEPT_IOIO_PROT, addr,
2405 			 SVM_EXIT_VMMCALL, "IOPM");
2406 	addr |= (1ull << 12) - 1;
2407 	TEST_BITMAP_ADDR(saved_intercept, INTERCEPT_IOIO_PROT, addr,
2408 			 SVM_EXIT_VMMCALL, "IOPM");
2409 
2410 	vmcb->control.intercept = saved_intercept;
2411 }
2412 
2413 /*
2414  * Unlike VMSAVE, VMRUN seems not to update the value of noncanonical
2415  * segment bases in the VMCB.  However, VMENTRY succeeds as documented.
2416  */
2417 #define TEST_CANONICAL_VMRUN(seg_base, msg)				\
2418 	saved_addr = seg_base;						\
2419 	seg_base = (seg_base & ((1ul << addr_limit) - 1)) | noncanonical_mask; \
2420 	return_value = svm_vmrun();					\
2421 	report(return_value == SVM_EXIT_VMMCALL,			\
2422 	       "Successful VMRUN with noncanonical %s.base", msg);	\
2423 	seg_base = saved_addr;
2424 
2425 
2426 #define TEST_CANONICAL_VMLOAD(seg_base, msg)				\
2427 	saved_addr = seg_base;						\
2428 	seg_base = (seg_base & ((1ul << addr_limit) - 1)) | noncanonical_mask; \
2429 	asm volatile ("vmload %0" : : "a"(vmcb_phys) : "memory");	\
2430 	asm volatile ("vmsave %0" : : "a"(vmcb_phys) : "memory");	\
2431 	report(is_canonical(seg_base),					\
2432 	       "Test %s.base for canonical form: %lx", msg, seg_base);	\
2433 	seg_base = saved_addr;
2434 
2435 static void test_canonicalization(void)
2436 {
2437 	u64 saved_addr;
2438 	u64 return_value;
2439 	u64 addr_limit;
2440 	u64 vmcb_phys = virt_to_phys(vmcb);
2441 
2442 	addr_limit = (this_cpu_has(X86_FEATURE_LA57)) ? 57 : 48;
2443 	u64 noncanonical_mask = NONCANONICAL & ~((1ul << addr_limit) - 1);
2444 
2445 	TEST_CANONICAL_VMLOAD(vmcb->save.fs.base, "FS");
2446 	TEST_CANONICAL_VMLOAD(vmcb->save.gs.base, "GS");
2447 	TEST_CANONICAL_VMLOAD(vmcb->save.ldtr.base, "LDTR");
2448 	TEST_CANONICAL_VMLOAD(vmcb->save.tr.base, "TR");
2449 	TEST_CANONICAL_VMLOAD(vmcb->save.kernel_gs_base, "KERNEL GS");
2450 	TEST_CANONICAL_VMRUN(vmcb->save.es.base, "ES");
2451 	TEST_CANONICAL_VMRUN(vmcb->save.cs.base, "CS");
2452 	TEST_CANONICAL_VMRUN(vmcb->save.ss.base, "SS");
2453 	TEST_CANONICAL_VMRUN(vmcb->save.ds.base, "DS");
2454 	TEST_CANONICAL_VMRUN(vmcb->save.gdtr.base, "GDTR");
2455 	TEST_CANONICAL_VMRUN(vmcb->save.idtr.base, "IDTR");
2456 }
2457 
2458 /*
2459  * When VMRUN loads a guest value of 1 in EFLAGS.TF, that value does not
2460  * cause a trace trap between the VMRUN and the first guest instruction, but
2461  * rather after completion of the first guest instruction.
2462  *
2463  * [APM vol 2]
2464  */
2465 u64 guest_rflags_test_trap_rip;
2466 
2467 static void guest_rflags_test_db_handler(struct ex_regs *r)
2468 {
2469 	guest_rflags_test_trap_rip = r->rip;
2470 	r->rflags &= ~X86_EFLAGS_TF;
2471 }
2472 
2473 static void svm_guest_state_test(void)
2474 {
2475 	test_set_guest(basic_guest_main);
2476 	test_efer();
2477 	test_cr0();
2478 	test_cr3();
2479 	test_cr4();
2480 	test_dr();
2481 	test_msrpm_iopm_bitmap_addrs();
2482 	test_canonicalization();
2483 }
2484 
2485 extern void guest_rflags_test_guest(struct svm_test *test);
2486 extern u64 *insn2;
2487 extern u64 *guest_end;
2488 
2489 asm("guest_rflags_test_guest:\n\t"
2490     "push %rbp\n\t"
2491     ".global insn2\n\t"
2492     "insn2:\n\t"
2493     "mov %rsp,%rbp\n\t"
2494     "vmmcall\n\t"
2495     "vmmcall\n\t"
2496     ".global guest_end\n\t"
2497     "guest_end:\n\t"
2498     "vmmcall\n\t"
2499     "pop %rbp\n\t"
2500     "ret");
2501 
2502 static void svm_test_singlestep(void)
2503 {
2504 	handle_exception(DB_VECTOR, guest_rflags_test_db_handler);
2505 
2506 	/*
2507 	 * Trap expected after completion of first guest instruction
2508 	 */
2509 	vmcb->save.rflags |= X86_EFLAGS_TF;
2510 	report (__svm_vmrun((u64)guest_rflags_test_guest) == SVM_EXIT_VMMCALL &&
2511 		guest_rflags_test_trap_rip == (u64)&insn2,
2512 		"Test EFLAGS.TF on VMRUN: trap expected  after completion of first guest instruction");
2513 	/*
2514 	 * No trap expected
2515 	 */
2516 	guest_rflags_test_trap_rip = 0;
2517 	vmcb->save.rip += 3;
2518 	vmcb->save.rflags |= X86_EFLAGS_TF;
2519 	report (__svm_vmrun(vmcb->save.rip) == SVM_EXIT_VMMCALL &&
2520 		guest_rflags_test_trap_rip == 0, "Test EFLAGS.TF on VMRUN: trap not expected");
2521 
2522 	/*
2523 	 * Let guest finish execution
2524 	 */
2525 	vmcb->save.rip += 3;
2526 	report (__svm_vmrun(vmcb->save.rip) == SVM_EXIT_VMMCALL &&
2527 		vmcb->save.rip == (u64)&guest_end, "Test EFLAGS.TF on VMRUN: guest execution completion");
2528 }
2529 
2530 static bool volatile svm_errata_reproduced = false;
2531 static unsigned long volatile physical = 0;
2532 
2533 
2534 /*
2535  *
2536  * Test the following errata:
2537  * If the VMRUN/VMSAVE/VMLOAD are attempted by the nested guest,
2538  * the CPU would first check the EAX against host reserved memory
2539  * regions (so far only SMM_ADDR/SMM_MASK are known to cause it),
2540  * and only then signal #VMexit
2541  *
2542  * Try to reproduce this by trying vmsave on each possible 4K aligned memory
2543  * address in the low 4G where the SMM area has to reside.
2544  */
2545 
2546 static void gp_isr(struct ex_regs *r)
2547 {
2548 	svm_errata_reproduced = true;
2549 	/* skip over the vmsave instruction*/
2550 	r->rip += 3;
2551 }
2552 
2553 static void svm_vmrun_errata_test(void)
2554 {
2555 	unsigned long *last_page = NULL;
2556 
2557 	handle_exception(GP_VECTOR, gp_isr);
2558 
2559 	while (!svm_errata_reproduced) {
2560 
2561 		unsigned long *page = alloc_pages(1);
2562 
2563 		if (!page) {
2564 			report_pass("All guest memory tested, no bug found");
2565 			break;
2566 		}
2567 
2568 		physical = virt_to_phys(page);
2569 
2570 		asm volatile (
2571 			      "mov %[_physical], %%rax\n\t"
2572 			      "vmsave %%rax\n\t"
2573 
2574 			      : [_physical] "=m" (physical)
2575 			      : /* no inputs*/
2576 			      : "rax" /*clobbers*/
2577 			      );
2578 
2579 		if (svm_errata_reproduced) {
2580 			report_fail("Got #GP exception - svm errata reproduced at 0x%lx",
2581 				    physical);
2582 			break;
2583 		}
2584 
2585 		*page = (unsigned long)last_page;
2586 		last_page = page;
2587 	}
2588 
2589 	while (last_page) {
2590 		unsigned long *page = last_page;
2591 		last_page = (unsigned long *)*last_page;
2592 		free_pages_by_order(page, 1);
2593 	}
2594 }
2595 
2596 static void vmload_vmsave_guest_main(struct svm_test *test)
2597 {
2598 	u64 vmcb_phys = virt_to_phys(vmcb);
2599 
2600 	asm volatile ("vmload %0" : : "a"(vmcb_phys));
2601 	asm volatile ("vmsave %0" : : "a"(vmcb_phys));
2602 }
2603 
2604 static void svm_vmload_vmsave(void)
2605 {
2606 	u32 intercept_saved = vmcb->control.intercept;
2607 
2608 	test_set_guest(vmload_vmsave_guest_main);
2609 
2610 	/*
2611 	 * Disabling intercept for VMLOAD and VMSAVE doesn't cause
2612 	 * respective #VMEXIT to host
2613 	 */
2614 	vmcb->control.intercept &= ~(1ULL << INTERCEPT_VMLOAD);
2615 	vmcb->control.intercept &= ~(1ULL << INTERCEPT_VMSAVE);
2616 	svm_vmrun();
2617 	report(vmcb->control.exit_code == SVM_EXIT_VMMCALL, "Test "
2618 	       "VMLOAD/VMSAVE intercept: Expected VMMCALL #VMEXIT");
2619 
2620 	/*
2621 	 * Enabling intercept for VMLOAD and VMSAVE causes respective
2622 	 * #VMEXIT to host
2623 	 */
2624 	vmcb->control.intercept |= (1ULL << INTERCEPT_VMLOAD);
2625 	svm_vmrun();
2626 	report(vmcb->control.exit_code == SVM_EXIT_VMLOAD, "Test "
2627 	       "VMLOAD/VMSAVE intercept: Expected VMLOAD #VMEXIT");
2628 	vmcb->control.intercept &= ~(1ULL << INTERCEPT_VMLOAD);
2629 	vmcb->control.intercept |= (1ULL << INTERCEPT_VMSAVE);
2630 	svm_vmrun();
2631 	report(vmcb->control.exit_code == SVM_EXIT_VMSAVE, "Test "
2632 	       "VMLOAD/VMSAVE intercept: Expected VMSAVE #VMEXIT");
2633 	vmcb->control.intercept &= ~(1ULL << INTERCEPT_VMSAVE);
2634 	svm_vmrun();
2635 	report(vmcb->control.exit_code == SVM_EXIT_VMMCALL, "Test "
2636 	       "VMLOAD/VMSAVE intercept: Expected VMMCALL #VMEXIT");
2637 
2638 	vmcb->control.intercept |= (1ULL << INTERCEPT_VMLOAD);
2639 	svm_vmrun();
2640 	report(vmcb->control.exit_code == SVM_EXIT_VMLOAD, "Test "
2641 	       "VMLOAD/VMSAVE intercept: Expected VMLOAD #VMEXIT");
2642 	vmcb->control.intercept &= ~(1ULL << INTERCEPT_VMLOAD);
2643 	svm_vmrun();
2644 	report(vmcb->control.exit_code == SVM_EXIT_VMMCALL, "Test "
2645 	       "VMLOAD/VMSAVE intercept: Expected VMMCALL #VMEXIT");
2646 
2647 	vmcb->control.intercept |= (1ULL << INTERCEPT_VMSAVE);
2648 	svm_vmrun();
2649 	report(vmcb->control.exit_code == SVM_EXIT_VMSAVE, "Test "
2650 	       "VMLOAD/VMSAVE intercept: Expected VMSAVE #VMEXIT");
2651 	vmcb->control.intercept &= ~(1ULL << INTERCEPT_VMSAVE);
2652 	svm_vmrun();
2653 	report(vmcb->control.exit_code == SVM_EXIT_VMMCALL, "Test "
2654 	       "VMLOAD/VMSAVE intercept: Expected VMMCALL #VMEXIT");
2655 
2656 	vmcb->control.intercept = intercept_saved;
2657 }
2658 
2659 static void prepare_vgif_enabled(struct svm_test *test)
2660 {
2661 	default_prepare(test);
2662 }
2663 
2664 static void test_vgif(struct svm_test *test)
2665 {
2666 	asm volatile ("vmmcall\n\tstgi\n\tvmmcall\n\tclgi\n\tvmmcall\n\t");
2667 }
2668 
2669 static bool vgif_finished(struct svm_test *test)
2670 {
2671 	switch (get_test_stage(test))
2672 		{
2673 		case 0:
2674 			if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
2675 				report_fail("VMEXIT not due to vmmcall.");
2676 				return true;
2677 			}
2678 			vmcb->control.int_ctl |= V_GIF_ENABLED_MASK;
2679 			vmcb->save.rip += 3;
2680 			inc_test_stage(test);
2681 			break;
2682 		case 1:
2683 			if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
2684 				report_fail("VMEXIT not due to vmmcall.");
2685 				return true;
2686 			}
2687 			if (!(vmcb->control.int_ctl & V_GIF_MASK)) {
2688 				report_fail("Failed to set VGIF when executing STGI.");
2689 				vmcb->control.int_ctl &= ~V_GIF_ENABLED_MASK;
2690 				return true;
2691 			}
2692 			report_pass("STGI set VGIF bit.");
2693 			vmcb->save.rip += 3;
2694 			inc_test_stage(test);
2695 			break;
2696 		case 2:
2697 			if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
2698 				report_fail("VMEXIT not due to vmmcall.");
2699 				return true;
2700 			}
2701 			if (vmcb->control.int_ctl & V_GIF_MASK) {
2702 				report_fail("Failed to clear VGIF when executing CLGI.");
2703 				vmcb->control.int_ctl &= ~V_GIF_ENABLED_MASK;
2704 				return true;
2705 			}
2706 			report_pass("CLGI cleared VGIF bit.");
2707 			vmcb->save.rip += 3;
2708 			inc_test_stage(test);
2709 			vmcb->control.int_ctl &= ~V_GIF_ENABLED_MASK;
2710 			break;
2711 		default:
2712 			return true;
2713 			break;
2714 		}
2715 
2716 	return get_test_stage(test) == 3;
2717 }
2718 
2719 static bool vgif_check(struct svm_test *test)
2720 {
2721 	return get_test_stage(test) == 3;
2722 }
2723 
2724 
2725 static int pause_test_counter;
2726 static int wait_counter;
2727 
2728 static void pause_filter_test_guest_main(struct svm_test *test)
2729 {
2730 	int i;
2731 	for (i = 0 ; i < pause_test_counter ; i++)
2732 		pause();
2733 
2734 	if (!wait_counter)
2735 		return;
2736 
2737 	for (i = 0; i < wait_counter; i++)
2738 		;
2739 
2740 	for (i = 0 ; i < pause_test_counter ; i++)
2741 		pause();
2742 
2743 }
2744 
2745 static void pause_filter_run_test(int pause_iterations, int filter_value, int wait_iterations, int threshold)
2746 {
2747 	test_set_guest(pause_filter_test_guest_main);
2748 
2749 	pause_test_counter = pause_iterations;
2750 	wait_counter = wait_iterations;
2751 
2752 	vmcb->control.pause_filter_count = filter_value;
2753 	vmcb->control.pause_filter_thresh = threshold;
2754 	svm_vmrun();
2755 
2756 	if (filter_value <= pause_iterations || wait_iterations < threshold)
2757 		report(vmcb->control.exit_code == SVM_EXIT_PAUSE, "expected PAUSE vmexit");
2758 	else
2759 		report(vmcb->control.exit_code == SVM_EXIT_VMMCALL, "no expected PAUSE vmexit");
2760 }
2761 
2762 static void pause_filter_test(void)
2763 {
2764 	if (!pause_filter_supported()) {
2765 		report_skip("PAUSE filter not supported in the guest");
2766 		return;
2767 	}
2768 
2769 	vmcb->control.intercept |= (1 << INTERCEPT_PAUSE);
2770 
2771 	// filter count more that pause count - no VMexit
2772 	pause_filter_run_test(10, 9, 0, 0);
2773 
2774 	// filter count smaller pause count - no VMexit
2775 	pause_filter_run_test(20, 21, 0, 0);
2776 
2777 
2778 	if (pause_threshold_supported()) {
2779 		// filter count smaller pause count - no VMexit +  large enough threshold
2780 		// so that filter counter resets
2781 		pause_filter_run_test(20, 21, 1000, 10);
2782 
2783 		// filter count smaller pause count - no VMexit +  small threshold
2784 		// so that filter doesn't reset
2785 		pause_filter_run_test(20, 21, 10, 1000);
2786 	} else {
2787 		report_skip("PAUSE threshold not supported in the guest");
2788 		return;
2789 	}
2790 }
2791 
2792 /* If CR0.TS and CR0.EM are cleared in L2, no #NM is generated. */
2793 static void svm_no_nm_test(void)
2794 {
2795 	write_cr0(read_cr0() & ~X86_CR0_TS);
2796 	test_set_guest((test_guest_func)fnop);
2797 
2798 	vmcb->save.cr0 = vmcb->save.cr0 & ~(X86_CR0_TS | X86_CR0_EM);
2799 	report(svm_vmrun() == SVM_EXIT_VMMCALL,
2800 	       "fnop with CR0.TS and CR0.EM unset no #NM excpetion");
2801 }
2802 
2803 static u64 amd_get_lbr_rip(u32 msr)
2804 {
2805 	return rdmsr(msr) & ~AMD_LBR_RECORD_MISPREDICT;
2806 }
2807 
2808 #define HOST_CHECK_LBR(from_expected, to_expected)					\
2809 do {											\
2810 	TEST_EXPECT_EQ((u64)from_expected, amd_get_lbr_rip(MSR_IA32_LASTBRANCHFROMIP));	\
2811 	TEST_EXPECT_EQ((u64)to_expected, amd_get_lbr_rip(MSR_IA32_LASTBRANCHTOIP));	\
2812 } while (0)
2813 
2814 /*
2815  * FIXME: Do something other than generate an exception to communicate failure.
2816  * Debugging without expected vs. actual is an absolute nightmare.
2817  */
2818 #define GUEST_CHECK_LBR(from_expected, to_expected)				\
2819 do {										\
2820 	if ((u64)(from_expected) != amd_get_lbr_rip(MSR_IA32_LASTBRANCHFROMIP))	\
2821 		asm volatile("ud2");						\
2822 	if ((u64)(to_expected) != amd_get_lbr_rip(MSR_IA32_LASTBRANCHTOIP))	\
2823 		asm volatile("ud2");						\
2824 } while (0)
2825 
2826 #define REPORT_GUEST_LBR_ERROR(vmcb)						\
2827 	report(false, "LBR guest test failed.  Exit reason 0x%x, RIP = %lx, from = %lx, to = %lx, ex from = %lx, ex to = %lx", \
2828 		       vmcb->control.exit_code, vmcb->save.rip,			\
2829 		       vmcb->save.br_from, vmcb->save.br_to,			\
2830 		       vmcb->save.last_excp_from, vmcb->save.last_excp_to)
2831 
2832 #define DO_BRANCH(branch_name)				\
2833 	asm volatile (					\
2834 		      # branch_name "_from:"		\
2835 		      "jmp " # branch_name  "_to\n"	\
2836 		      "nop\n"				\
2837 		      "nop\n"				\
2838 		      # branch_name  "_to:"		\
2839 		      "nop\n"				\
2840 		       )
2841 
2842 
2843 extern u64 guest_branch0_from, guest_branch0_to;
2844 extern u64 guest_branch2_from, guest_branch2_to;
2845 
2846 extern u64 host_branch0_from, host_branch0_to;
2847 extern u64 host_branch2_from, host_branch2_to;
2848 extern u64 host_branch3_from, host_branch3_to;
2849 extern u64 host_branch4_from, host_branch4_to;
2850 
2851 u64 dbgctl;
2852 
2853 static void svm_lbrv_test_guest1(void)
2854 {
2855 	/*
2856 	 * This guest expects the LBR to be already enabled when it starts,
2857 	 * it does a branch, and then disables the LBR and then checks.
2858 	 */
2859 
2860 	DO_BRANCH(guest_branch0);
2861 
2862 	dbgctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
2863 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0);
2864 
2865 	if (dbgctl != DEBUGCTLMSR_LBR)
2866 		asm volatile("ud2\n");
2867 	if (rdmsr(MSR_IA32_DEBUGCTLMSR) != 0)
2868 		asm volatile("ud2\n");
2869 
2870 	GUEST_CHECK_LBR(&guest_branch0_from, &guest_branch0_to);
2871 	asm volatile ("vmmcall\n");
2872 }
2873 
2874 static void svm_lbrv_test_guest2(void)
2875 {
2876 	/*
2877 	 * This guest expects the LBR to be disabled when it starts,
2878 	 * enables it, does a branch, disables it and then checks.
2879 	 */
2880 
2881 	DO_BRANCH(guest_branch1);
2882 	dbgctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
2883 
2884 	if (dbgctl != 0)
2885 		asm volatile("ud2\n");
2886 
2887 	GUEST_CHECK_LBR(&host_branch2_from, &host_branch2_to);
2888 
2889 	wrmsr(MSR_IA32_DEBUGCTLMSR, DEBUGCTLMSR_LBR);
2890 	dbgctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
2891 	DO_BRANCH(guest_branch2);
2892 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0);
2893 
2894 	if (dbgctl != DEBUGCTLMSR_LBR)
2895 		asm volatile("ud2\n");
2896 	GUEST_CHECK_LBR(&guest_branch2_from, &guest_branch2_to);
2897 
2898 	asm volatile ("vmmcall\n");
2899 }
2900 
2901 static void svm_lbrv_test0(void)
2902 {
2903 	report(true, "Basic LBR test");
2904 	wrmsr(MSR_IA32_DEBUGCTLMSR, DEBUGCTLMSR_LBR);
2905 	DO_BRANCH(host_branch0);
2906 	dbgctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
2907 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0);
2908 
2909 	TEST_EXPECT_EQ(dbgctl, DEBUGCTLMSR_LBR);
2910 	dbgctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
2911 	TEST_EXPECT_EQ(dbgctl, 0);
2912 
2913 	HOST_CHECK_LBR(&host_branch0_from, &host_branch0_to);
2914 }
2915 
2916 static void svm_lbrv_test1(void)
2917 {
2918 	report(true, "Test that without LBRV enabled, guest LBR state does 'leak' to the host(1)");
2919 
2920 	svm_setup_vmrun((u64)svm_lbrv_test_guest1);
2921 	vmcb->control.virt_ext = 0;
2922 
2923 	wrmsr(MSR_IA32_DEBUGCTLMSR, DEBUGCTLMSR_LBR);
2924 	DO_BRANCH(host_branch1);
2925 	SVM_BARE_VMRUN;
2926 	dbgctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
2927 
2928 	if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
2929 		REPORT_GUEST_LBR_ERROR(vmcb);
2930 		return;
2931 	}
2932 
2933 	TEST_EXPECT_EQ(dbgctl, 0);
2934 	HOST_CHECK_LBR(&guest_branch0_from, &guest_branch0_to);
2935 }
2936 
2937 static void svm_lbrv_test2(void)
2938 {
2939 	report(true, "Test that without LBRV enabled, guest LBR state does 'leak' to the host(2)");
2940 
2941 	svm_setup_vmrun((u64)svm_lbrv_test_guest2);
2942 	vmcb->control.virt_ext = 0;
2943 
2944 	wrmsr(MSR_IA32_DEBUGCTLMSR, DEBUGCTLMSR_LBR);
2945 	DO_BRANCH(host_branch2);
2946 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0);
2947 	SVM_BARE_VMRUN;
2948 	dbgctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
2949 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0);
2950 
2951 	if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
2952 		REPORT_GUEST_LBR_ERROR(vmcb);
2953 		return;
2954 	}
2955 
2956 	TEST_EXPECT_EQ(dbgctl, 0);
2957 	HOST_CHECK_LBR(&guest_branch2_from, &guest_branch2_to);
2958 }
2959 
2960 static void svm_lbrv_nested_test1(void)
2961 {
2962 	if (!lbrv_supported()) {
2963 		report_skip("LBRV not supported in the guest");
2964 		return;
2965 	}
2966 
2967 	report(true, "Test that with LBRV enabled, guest LBR state doesn't leak (1)");
2968 	svm_setup_vmrun((u64)svm_lbrv_test_guest1);
2969 	vmcb->control.virt_ext = LBR_CTL_ENABLE_MASK;
2970 	vmcb->save.dbgctl = DEBUGCTLMSR_LBR;
2971 
2972 	wrmsr(MSR_IA32_DEBUGCTLMSR, DEBUGCTLMSR_LBR);
2973 	DO_BRANCH(host_branch3);
2974 	SVM_BARE_VMRUN;
2975 	dbgctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
2976 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0);
2977 
2978 	if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
2979 		REPORT_GUEST_LBR_ERROR(vmcb);
2980 		return;
2981 	}
2982 
2983 	if (vmcb->save.dbgctl != 0) {
2984 		report(false, "unexpected virtual guest MSR_IA32_DEBUGCTLMSR value 0x%lx", vmcb->save.dbgctl);
2985 		return;
2986 	}
2987 
2988 	TEST_EXPECT_EQ(dbgctl, DEBUGCTLMSR_LBR);
2989 	HOST_CHECK_LBR(&host_branch3_from, &host_branch3_to);
2990 }
2991 
2992 static void svm_lbrv_nested_test2(void)
2993 {
2994 	if (!lbrv_supported()) {
2995 		report_skip("LBRV not supported in the guest");
2996 		return;
2997 	}
2998 
2999 	report(true, "Test that with LBRV enabled, guest LBR state doesn't leak (2)");
3000 	svm_setup_vmrun((u64)svm_lbrv_test_guest2);
3001 	vmcb->control.virt_ext = LBR_CTL_ENABLE_MASK;
3002 
3003 	vmcb->save.dbgctl = 0;
3004 	vmcb->save.br_from = (u64)&host_branch2_from;
3005 	vmcb->save.br_to = (u64)&host_branch2_to;
3006 
3007 	wrmsr(MSR_IA32_DEBUGCTLMSR, DEBUGCTLMSR_LBR);
3008 	DO_BRANCH(host_branch4);
3009 	SVM_BARE_VMRUN;
3010 	dbgctl = rdmsr(MSR_IA32_DEBUGCTLMSR);
3011 	wrmsr(MSR_IA32_DEBUGCTLMSR, 0);
3012 
3013 	if (vmcb->control.exit_code != SVM_EXIT_VMMCALL) {
3014 		REPORT_GUEST_LBR_ERROR(vmcb);
3015 		return;
3016 	}
3017 
3018 	TEST_EXPECT_EQ(dbgctl, DEBUGCTLMSR_LBR);
3019 	HOST_CHECK_LBR(&host_branch4_from, &host_branch4_to);
3020 }
3021 
3022 
3023 // test that a nested guest which does enable INTR interception
3024 // but doesn't enable virtual interrupt masking works
3025 
3026 static volatile int dummy_isr_recevied;
3027 static void dummy_isr(isr_regs_t *regs)
3028 {
3029 	dummy_isr_recevied++;
3030 	eoi();
3031 }
3032 
3033 
3034 static volatile int nmi_recevied;
3035 static void dummy_nmi_handler(struct ex_regs *regs)
3036 {
3037 	nmi_recevied++;
3038 }
3039 
3040 
3041 static void svm_intr_intercept_mix_run_guest(volatile int *counter, int expected_vmexit)
3042 {
3043 	if (counter)
3044 		*counter = 0;
3045 
3046 	sti();  // host IF value should not matter
3047 	clgi(); // vmrun will set back GI to 1
3048 
3049 	svm_vmrun();
3050 
3051 	if (counter)
3052 		report(!*counter, "No interrupt expected");
3053 
3054 	stgi();
3055 
3056 	if (counter)
3057 		report(*counter == 1, "Interrupt is expected");
3058 
3059 	report (vmcb->control.exit_code == expected_vmexit, "Test expected VM exit");
3060 	report(vmcb->save.rflags & X86_EFLAGS_IF, "Guest should have EFLAGS.IF set now");
3061 	cli();
3062 }
3063 
3064 
3065 // subtest: test that enabling EFLAGS.IF is enought to trigger an interrupt
3066 static void svm_intr_intercept_mix_if_guest(struct svm_test *test)
3067 {
3068 	asm volatile("nop;nop;nop;nop");
3069 	report(!dummy_isr_recevied, "No interrupt expected");
3070 	sti();
3071 	asm volatile("nop");
3072 	report(0, "must not reach here");
3073 }
3074 
3075 static void svm_intr_intercept_mix_if(void)
3076 {
3077 	// make a physical interrupt to be pending
3078 	handle_irq(0x55, dummy_isr);
3079 
3080 	vmcb->control.intercept |= (1 << INTERCEPT_INTR);
3081 	vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
3082 	vmcb->save.rflags &= ~X86_EFLAGS_IF;
3083 
3084 	test_set_guest(svm_intr_intercept_mix_if_guest);
3085 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL | APIC_DM_FIXED | 0x55, 0);
3086 	svm_intr_intercept_mix_run_guest(&dummy_isr_recevied, SVM_EXIT_INTR);
3087 }
3088 
3089 
3090 // subtest: test that a clever guest can trigger an interrupt by setting GIF
3091 // if GIF is not intercepted
3092 static void svm_intr_intercept_mix_gif_guest(struct svm_test *test)
3093 {
3094 
3095 	asm volatile("nop;nop;nop;nop");
3096 	report(!dummy_isr_recevied, "No interrupt expected");
3097 
3098 	// clear GIF and enable IF
3099 	// that should still not cause VM exit
3100 	clgi();
3101 	sti();
3102 	asm volatile("nop");
3103 	report(!dummy_isr_recevied, "No interrupt expected");
3104 
3105 	stgi();
3106 	asm volatile("nop");
3107 	report(0, "must not reach here");
3108 }
3109 
3110 static void svm_intr_intercept_mix_gif(void)
3111 {
3112 	handle_irq(0x55, dummy_isr);
3113 
3114 	vmcb->control.intercept |= (1 << INTERCEPT_INTR);
3115 	vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
3116 	vmcb->save.rflags &= ~X86_EFLAGS_IF;
3117 
3118 	test_set_guest(svm_intr_intercept_mix_gif_guest);
3119 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL | APIC_DM_FIXED | 0x55, 0);
3120 	svm_intr_intercept_mix_run_guest(&dummy_isr_recevied, SVM_EXIT_INTR);
3121 }
3122 
3123 // subtest: test that a clever guest can trigger an interrupt by setting GIF
3124 // if GIF is not intercepted and interrupt comes after guest
3125 // started running
3126 static void svm_intr_intercept_mix_gif_guest2(struct svm_test *test)
3127 {
3128 	asm volatile("nop;nop;nop;nop");
3129 	report(!dummy_isr_recevied, "No interrupt expected");
3130 
3131 	clgi();
3132 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL | APIC_DM_FIXED | 0x55, 0);
3133 	report(!dummy_isr_recevied, "No interrupt expected");
3134 
3135 	stgi();
3136 	asm volatile("nop");
3137 	report(0, "must not reach here");
3138 }
3139 
3140 static void svm_intr_intercept_mix_gif2(void)
3141 {
3142 	handle_irq(0x55, dummy_isr);
3143 
3144 	vmcb->control.intercept |= (1 << INTERCEPT_INTR);
3145 	vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
3146 	vmcb->save.rflags |= X86_EFLAGS_IF;
3147 
3148 	test_set_guest(svm_intr_intercept_mix_gif_guest2);
3149 	svm_intr_intercept_mix_run_guest(&dummy_isr_recevied, SVM_EXIT_INTR);
3150 }
3151 
3152 
3153 // subtest: test that pending NMI will be handled when guest enables GIF
3154 static void svm_intr_intercept_mix_nmi_guest(struct svm_test *test)
3155 {
3156 	asm volatile("nop;nop;nop;nop");
3157 	report(!nmi_recevied, "No NMI expected");
3158 	cli(); // should have no effect
3159 
3160 	clgi();
3161 	asm volatile("nop");
3162 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL | APIC_DM_NMI, 0);
3163 	sti(); // should have no effect
3164 	asm volatile("nop");
3165 	report(!nmi_recevied, "No NMI expected");
3166 
3167 	stgi();
3168 	asm volatile("nop");
3169 	report(0, "must not reach here");
3170 }
3171 
3172 static void svm_intr_intercept_mix_nmi(void)
3173 {
3174 	handle_exception(2, dummy_nmi_handler);
3175 
3176 	vmcb->control.intercept |= (1 << INTERCEPT_NMI);
3177 	vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
3178 	vmcb->save.rflags |= X86_EFLAGS_IF;
3179 
3180 	test_set_guest(svm_intr_intercept_mix_nmi_guest);
3181 	svm_intr_intercept_mix_run_guest(&nmi_recevied, SVM_EXIT_NMI);
3182 }
3183 
3184 // test that pending SMI will be handled when guest enables GIF
3185 // TODO: can't really count #SMIs so just test that guest doesn't hang
3186 // and VMexits on SMI
3187 static void svm_intr_intercept_mix_smi_guest(struct svm_test *test)
3188 {
3189 	asm volatile("nop;nop;nop;nop");
3190 
3191 	clgi();
3192 	asm volatile("nop");
3193 	apic_icr_write(APIC_DEST_SELF | APIC_DEST_PHYSICAL | APIC_DM_SMI, 0);
3194 	sti(); // should have no effect
3195 	asm volatile("nop");
3196 	stgi();
3197 	asm volatile("nop");
3198 	report(0, "must not reach here");
3199 }
3200 
3201 static void svm_intr_intercept_mix_smi(void)
3202 {
3203 	vmcb->control.intercept |= (1 << INTERCEPT_SMI);
3204 	vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
3205 	test_set_guest(svm_intr_intercept_mix_smi_guest);
3206 	svm_intr_intercept_mix_run_guest(NULL, SVM_EXIT_SMI);
3207 }
3208 
3209 static void svm_l2_ac_test(void)
3210 {
3211 	bool hit_ac = false;
3212 
3213 	write_cr0(read_cr0() | X86_CR0_AM);
3214 	write_rflags(read_rflags() | X86_EFLAGS_AC);
3215 
3216 	run_in_user(generate_usermode_ac, AC_VECTOR, 0, 0, 0, 0, &hit_ac);
3217 	report(hit_ac, "Usermode #AC handled in L2");
3218 	vmmcall();
3219 }
3220 
3221 struct svm_exception_test {
3222 	u8 vector;
3223 	void (*guest_code)(void);
3224 };
3225 
3226 struct svm_exception_test svm_exception_tests[] = {
3227 	{ GP_VECTOR, generate_non_canonical_gp },
3228 	{ UD_VECTOR, generate_ud },
3229 	{ DE_VECTOR, generate_de },
3230 	{ DB_VECTOR, generate_single_step_db },
3231 	{ BP_VECTOR, generate_bp },
3232 	{ AC_VECTOR, svm_l2_ac_test },
3233 	{ OF_VECTOR, generate_of },
3234 	{ NM_VECTOR, generate_cr0_ts_nm },
3235 	{ NM_VECTOR, generate_cr0_em_nm },
3236 };
3237 
3238 static u8 svm_exception_test_vector;
3239 
3240 static void svm_exception_handler(struct ex_regs *regs)
3241 {
3242 	report(regs->vector == svm_exception_test_vector,
3243 		"Handling %s in L2's exception handler",
3244 		exception_mnemonic(svm_exception_test_vector));
3245 	vmmcall();
3246 }
3247 
3248 static void handle_exception_in_l2(u8 vector)
3249 {
3250 	handler old_handler = handle_exception(vector, svm_exception_handler);
3251 	svm_exception_test_vector = vector;
3252 
3253 	report(svm_vmrun() == SVM_EXIT_VMMCALL,
3254 		"%s handled by L2", exception_mnemonic(vector));
3255 
3256 	handle_exception(vector, old_handler);
3257 }
3258 
3259 static void handle_exception_in_l1(u32 vector)
3260 {
3261 	u32 old_ie = vmcb->control.intercept_exceptions;
3262 
3263 	vmcb->control.intercept_exceptions |= (1ULL << vector);
3264 
3265 	report(svm_vmrun() == (SVM_EXIT_EXCP_BASE + vector),
3266 		"%s handled by L1",  exception_mnemonic(vector));
3267 
3268 	vmcb->control.intercept_exceptions = old_ie;
3269 }
3270 
3271 static void svm_exception_test(void)
3272 {
3273 	struct svm_exception_test *t;
3274 	int i;
3275 
3276 	for (i = 0; i < ARRAY_SIZE(svm_exception_tests); i++) {
3277 		t = &svm_exception_tests[i];
3278 		test_set_guest((test_guest_func)t->guest_code);
3279 
3280 		handle_exception_in_l2(t->vector);
3281 		vmcb_ident(vmcb);
3282 
3283 		handle_exception_in_l1(t->vector);
3284 		vmcb_ident(vmcb);
3285 	}
3286 }
3287 
3288 struct svm_test svm_tests[] = {
3289 	{ "null", default_supported, default_prepare,
3290 	  default_prepare_gif_clear, null_test,
3291 	  default_finished, null_check },
3292 	{ "vmrun", default_supported, default_prepare,
3293 	  default_prepare_gif_clear, test_vmrun,
3294 	  default_finished, check_vmrun },
3295 	{ "ioio", default_supported, prepare_ioio,
3296 	  default_prepare_gif_clear, test_ioio,
3297 	  ioio_finished, check_ioio },
3298 	{ "vmrun intercept check", default_supported, prepare_no_vmrun_int,
3299 	  default_prepare_gif_clear, null_test, default_finished,
3300 	  check_no_vmrun_int },
3301 	{ "rsm", default_supported,
3302 	  prepare_rsm_intercept, default_prepare_gif_clear,
3303 	  test_rsm_intercept, finished_rsm_intercept, check_rsm_intercept },
3304 	{ "cr3 read intercept", default_supported,
3305 	  prepare_cr3_intercept, default_prepare_gif_clear,
3306 	  test_cr3_intercept, default_finished, check_cr3_intercept },
3307 	{ "cr3 read nointercept", default_supported, default_prepare,
3308 	  default_prepare_gif_clear, test_cr3_intercept, default_finished,
3309 	  check_cr3_nointercept },
3310 	{ "cr3 read intercept emulate", smp_supported,
3311 	  prepare_cr3_intercept_bypass, default_prepare_gif_clear,
3312 	  test_cr3_intercept_bypass, default_finished, check_cr3_intercept },
3313 	{ "dr intercept check", default_supported, prepare_dr_intercept,
3314 	  default_prepare_gif_clear, test_dr_intercept, dr_intercept_finished,
3315 	  check_dr_intercept },
3316 	{ "next_rip", next_rip_supported, prepare_next_rip,
3317 	  default_prepare_gif_clear, test_next_rip,
3318 	  default_finished, check_next_rip },
3319 	{ "msr intercept check", default_supported, prepare_msr_intercept,
3320 	  default_prepare_gif_clear, test_msr_intercept,
3321 	  msr_intercept_finished, check_msr_intercept },
3322 	{ "mode_switch", default_supported, prepare_mode_switch,
3323 	  default_prepare_gif_clear, test_mode_switch,
3324 	  mode_switch_finished, check_mode_switch },
3325 	{ "asid_zero", default_supported, prepare_asid_zero,
3326 	  default_prepare_gif_clear, test_asid_zero,
3327 	  default_finished, check_asid_zero },
3328 	{ "sel_cr0_bug", default_supported, sel_cr0_bug_prepare,
3329 	  default_prepare_gif_clear, sel_cr0_bug_test,
3330 	  sel_cr0_bug_finished, sel_cr0_bug_check },
3331 	{ "tsc_adjust", tsc_adjust_supported, tsc_adjust_prepare,
3332 	  default_prepare_gif_clear, tsc_adjust_test,
3333 	  default_finished, tsc_adjust_check },
3334 	{ "latency_run_exit", default_supported, latency_prepare,
3335 	  default_prepare_gif_clear, latency_test,
3336 	  latency_finished, latency_check },
3337 	{ "latency_run_exit_clean", default_supported, latency_prepare,
3338 	  default_prepare_gif_clear, latency_test,
3339 	  latency_finished_clean, latency_check },
3340 	{ "latency_svm_insn", default_supported, lat_svm_insn_prepare,
3341 	  default_prepare_gif_clear, null_test,
3342 	  lat_svm_insn_finished, lat_svm_insn_check },
3343 	{ "exc_inject", default_supported, exc_inject_prepare,
3344 	  default_prepare_gif_clear, exc_inject_test,
3345 	  exc_inject_finished, exc_inject_check },
3346 	{ "pending_event", default_supported, pending_event_prepare,
3347 	  default_prepare_gif_clear,
3348 	  pending_event_test, pending_event_finished, pending_event_check },
3349 	{ "pending_event_cli", default_supported, pending_event_cli_prepare,
3350 	  pending_event_cli_prepare_gif_clear,
3351 	  pending_event_cli_test, pending_event_cli_finished,
3352 	  pending_event_cli_check },
3353 	{ "interrupt", default_supported, interrupt_prepare,
3354 	  default_prepare_gif_clear, interrupt_test,
3355 	  interrupt_finished, interrupt_check },
3356 	{ "nmi", default_supported, nmi_prepare,
3357 	  default_prepare_gif_clear, nmi_test,
3358 	  nmi_finished, nmi_check },
3359 	{ "nmi_hlt", smp_supported, nmi_prepare,
3360 	  default_prepare_gif_clear, nmi_hlt_test,
3361 	  nmi_hlt_finished, nmi_hlt_check },
3362         { "vnmi", vnmi_supported, vnmi_prepare,
3363           default_prepare_gif_clear, vnmi_test,
3364           vnmi_finished, vnmi_check },
3365 	{ "virq_inject", default_supported, virq_inject_prepare,
3366 	  default_prepare_gif_clear, virq_inject_test,
3367 	  virq_inject_finished, virq_inject_check },
3368 	{ "reg_corruption", default_supported, reg_corruption_prepare,
3369 	  default_prepare_gif_clear, reg_corruption_test,
3370 	  reg_corruption_finished, reg_corruption_check },
3371 	{ "svm_init_startup_test", smp_supported, init_startup_prepare,
3372 	  default_prepare_gif_clear, null_test,
3373 	  init_startup_finished, init_startup_check },
3374 	{ "svm_init_intercept_test", smp_supported, init_intercept_prepare,
3375 	  default_prepare_gif_clear, init_intercept_test,
3376 	  init_intercept_finished, init_intercept_check, .on_vcpu = 2 },
3377 	{ "host_rflags", default_supported, host_rflags_prepare,
3378 	  host_rflags_prepare_gif_clear, host_rflags_test,
3379 	  host_rflags_finished, host_rflags_check },
3380 	{ "vgif", vgif_supported, prepare_vgif_enabled,
3381 	  default_prepare_gif_clear, test_vgif, vgif_finished,
3382 	  vgif_check },
3383 	TEST(svm_cr4_osxsave_test),
3384 	TEST(svm_guest_state_test),
3385 	TEST(svm_vmrun_errata_test),
3386 	TEST(svm_vmload_vmsave),
3387 	TEST(svm_test_singlestep),
3388 	TEST(svm_no_nm_test),
3389 	TEST(svm_exception_test),
3390 	TEST(svm_lbrv_test0),
3391 	TEST(svm_lbrv_test1),
3392 	TEST(svm_lbrv_test2),
3393 	TEST(svm_lbrv_nested_test1),
3394 	TEST(svm_lbrv_nested_test2),
3395 	TEST(svm_intr_intercept_mix_if),
3396 	TEST(svm_intr_intercept_mix_gif),
3397 	TEST(svm_intr_intercept_mix_gif2),
3398 	TEST(svm_intr_intercept_mix_nmi),
3399 	TEST(svm_intr_intercept_mix_smi),
3400 	TEST(svm_tsc_scale_test),
3401 	TEST(pause_filter_test),
3402 	{ NULL, NULL, NULL, NULL, NULL, NULL, NULL }
3403 };
3404 
3405 int main(int ac, char **av)
3406 {
3407 	setup_vm();
3408 	return run_svm_tests(ac, av, svm_tests);
3409 }
3410