1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * tools/testing/selftests/kvm/lib/x86_64/processor.c
4  *
5  * Copyright (C) 2018, Google LLC.
6  */
7 
8 #include "test_util.h"
9 #include "kvm_util.h"
10 #include "processor.h"
11 
12 #ifndef NUM_INTERRUPTS
13 #define NUM_INTERRUPTS 256
14 #endif
15 
16 #define DEFAULT_CODE_SELECTOR 0x8
17 #define DEFAULT_DATA_SELECTOR 0x10
18 
19 #define MAX_NR_CPUID_ENTRIES 100
20 
21 vm_vaddr_t exception_handlers;
22 bool host_cpu_is_amd;
23 bool host_cpu_is_intel;
24 
25 static void regs_dump(FILE *stream, struct kvm_regs *regs, uint8_t indent)
26 {
27 	fprintf(stream, "%*srax: 0x%.16llx rbx: 0x%.16llx "
28 		"rcx: 0x%.16llx rdx: 0x%.16llx\n",
29 		indent, "",
30 		regs->rax, regs->rbx, regs->rcx, regs->rdx);
31 	fprintf(stream, "%*srsi: 0x%.16llx rdi: 0x%.16llx "
32 		"rsp: 0x%.16llx rbp: 0x%.16llx\n",
33 		indent, "",
34 		regs->rsi, regs->rdi, regs->rsp, regs->rbp);
35 	fprintf(stream, "%*sr8:  0x%.16llx r9:  0x%.16llx "
36 		"r10: 0x%.16llx r11: 0x%.16llx\n",
37 		indent, "",
38 		regs->r8, regs->r9, regs->r10, regs->r11);
39 	fprintf(stream, "%*sr12: 0x%.16llx r13: 0x%.16llx "
40 		"r14: 0x%.16llx r15: 0x%.16llx\n",
41 		indent, "",
42 		regs->r12, regs->r13, regs->r14, regs->r15);
43 	fprintf(stream, "%*srip: 0x%.16llx rfl: 0x%.16llx\n",
44 		indent, "",
45 		regs->rip, regs->rflags);
46 }
47 
48 static void segment_dump(FILE *stream, struct kvm_segment *segment,
49 			 uint8_t indent)
50 {
51 	fprintf(stream, "%*sbase: 0x%.16llx limit: 0x%.8x "
52 		"selector: 0x%.4x type: 0x%.2x\n",
53 		indent, "", segment->base, segment->limit,
54 		segment->selector, segment->type);
55 	fprintf(stream, "%*spresent: 0x%.2x dpl: 0x%.2x "
56 		"db: 0x%.2x s: 0x%.2x l: 0x%.2x\n",
57 		indent, "", segment->present, segment->dpl,
58 		segment->db, segment->s, segment->l);
59 	fprintf(stream, "%*sg: 0x%.2x avl: 0x%.2x "
60 		"unusable: 0x%.2x padding: 0x%.2x\n",
61 		indent, "", segment->g, segment->avl,
62 		segment->unusable, segment->padding);
63 }
64 
65 static void dtable_dump(FILE *stream, struct kvm_dtable *dtable,
66 			uint8_t indent)
67 {
68 	fprintf(stream, "%*sbase: 0x%.16llx limit: 0x%.4x "
69 		"padding: 0x%.4x 0x%.4x 0x%.4x\n",
70 		indent, "", dtable->base, dtable->limit,
71 		dtable->padding[0], dtable->padding[1], dtable->padding[2]);
72 }
73 
74 static void sregs_dump(FILE *stream, struct kvm_sregs *sregs, uint8_t indent)
75 {
76 	unsigned int i;
77 
78 	fprintf(stream, "%*scs:\n", indent, "");
79 	segment_dump(stream, &sregs->cs, indent + 2);
80 	fprintf(stream, "%*sds:\n", indent, "");
81 	segment_dump(stream, &sregs->ds, indent + 2);
82 	fprintf(stream, "%*ses:\n", indent, "");
83 	segment_dump(stream, &sregs->es, indent + 2);
84 	fprintf(stream, "%*sfs:\n", indent, "");
85 	segment_dump(stream, &sregs->fs, indent + 2);
86 	fprintf(stream, "%*sgs:\n", indent, "");
87 	segment_dump(stream, &sregs->gs, indent + 2);
88 	fprintf(stream, "%*sss:\n", indent, "");
89 	segment_dump(stream, &sregs->ss, indent + 2);
90 	fprintf(stream, "%*str:\n", indent, "");
91 	segment_dump(stream, &sregs->tr, indent + 2);
92 	fprintf(stream, "%*sldt:\n", indent, "");
93 	segment_dump(stream, &sregs->ldt, indent + 2);
94 
95 	fprintf(stream, "%*sgdt:\n", indent, "");
96 	dtable_dump(stream, &sregs->gdt, indent + 2);
97 	fprintf(stream, "%*sidt:\n", indent, "");
98 	dtable_dump(stream, &sregs->idt, indent + 2);
99 
100 	fprintf(stream, "%*scr0: 0x%.16llx cr2: 0x%.16llx "
101 		"cr3: 0x%.16llx cr4: 0x%.16llx\n",
102 		indent, "",
103 		sregs->cr0, sregs->cr2, sregs->cr3, sregs->cr4);
104 	fprintf(stream, "%*scr8: 0x%.16llx efer: 0x%.16llx "
105 		"apic_base: 0x%.16llx\n",
106 		indent, "",
107 		sregs->cr8, sregs->efer, sregs->apic_base);
108 
109 	fprintf(stream, "%*sinterrupt_bitmap:\n", indent, "");
110 	for (i = 0; i < (KVM_NR_INTERRUPTS + 63) / 64; i++) {
111 		fprintf(stream, "%*s%.16llx\n", indent + 2, "",
112 			sregs->interrupt_bitmap[i]);
113 	}
114 }
115 
116 bool kvm_is_tdp_enabled(void)
117 {
118 	if (host_cpu_is_intel)
119 		return get_kvm_intel_param_bool("ept");
120 	else
121 		return get_kvm_amd_param_bool("npt");
122 }
123 
124 void virt_arch_pgd_alloc(struct kvm_vm *vm)
125 {
126 	TEST_ASSERT(vm->mode == VM_MODE_PXXV48_4K, "Attempt to use "
127 		"unknown or unsupported guest mode, mode: 0x%x", vm->mode);
128 
129 	/* If needed, create page map l4 table. */
130 	if (!vm->pgd_created) {
131 		vm->pgd = vm_alloc_page_table(vm);
132 		vm->pgd_created = true;
133 	}
134 }
135 
136 static void *virt_get_pte(struct kvm_vm *vm, uint64_t *parent_pte,
137 			  uint64_t vaddr, int level)
138 {
139 	uint64_t pt_gpa = PTE_GET_PA(*parent_pte);
140 	uint64_t *page_table = addr_gpa2hva(vm, pt_gpa);
141 	int index = (vaddr >> PG_LEVEL_SHIFT(level)) & 0x1ffu;
142 
143 	TEST_ASSERT((*parent_pte & PTE_PRESENT_MASK) || parent_pte == &vm->pgd,
144 		    "Parent PTE (level %d) not PRESENT for gva: 0x%08lx",
145 		    level + 1, vaddr);
146 
147 	return &page_table[index];
148 }
149 
150 static uint64_t *virt_create_upper_pte(struct kvm_vm *vm,
151 				       uint64_t *parent_pte,
152 				       uint64_t vaddr,
153 				       uint64_t paddr,
154 				       int current_level,
155 				       int target_level)
156 {
157 	uint64_t *pte = virt_get_pte(vm, parent_pte, vaddr, current_level);
158 
159 	if (!(*pte & PTE_PRESENT_MASK)) {
160 		*pte = PTE_PRESENT_MASK | PTE_WRITABLE_MASK;
161 		if (current_level == target_level)
162 			*pte |= PTE_LARGE_MASK | (paddr & PHYSICAL_PAGE_MASK);
163 		else
164 			*pte |= vm_alloc_page_table(vm) & PHYSICAL_PAGE_MASK;
165 	} else {
166 		/*
167 		 * Entry already present.  Assert that the caller doesn't want
168 		 * a hugepage at this level, and that there isn't a hugepage at
169 		 * this level.
170 		 */
171 		TEST_ASSERT(current_level != target_level,
172 			    "Cannot create hugepage at level: %u, vaddr: 0x%lx\n",
173 			    current_level, vaddr);
174 		TEST_ASSERT(!(*pte & PTE_LARGE_MASK),
175 			    "Cannot create page table at level: %u, vaddr: 0x%lx\n",
176 			    current_level, vaddr);
177 	}
178 	return pte;
179 }
180 
181 void __virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, int level)
182 {
183 	const uint64_t pg_size = PG_LEVEL_SIZE(level);
184 	uint64_t *pml4e, *pdpe, *pde;
185 	uint64_t *pte;
186 
187 	TEST_ASSERT(vm->mode == VM_MODE_PXXV48_4K,
188 		    "Unknown or unsupported guest mode, mode: 0x%x", vm->mode);
189 
190 	TEST_ASSERT((vaddr % pg_size) == 0,
191 		    "Virtual address not aligned,\n"
192 		    "vaddr: 0x%lx page size: 0x%lx", vaddr, pg_size);
193 	TEST_ASSERT(sparsebit_is_set(vm->vpages_valid, (vaddr >> vm->page_shift)),
194 		    "Invalid virtual address, vaddr: 0x%lx", vaddr);
195 	TEST_ASSERT((paddr % pg_size) == 0,
196 		    "Physical address not aligned,\n"
197 		    "  paddr: 0x%lx page size: 0x%lx", paddr, pg_size);
198 	TEST_ASSERT((paddr >> vm->page_shift) <= vm->max_gfn,
199 		    "Physical address beyond maximum supported,\n"
200 		    "  paddr: 0x%lx vm->max_gfn: 0x%lx vm->page_size: 0x%x",
201 		    paddr, vm->max_gfn, vm->page_size);
202 
203 	/*
204 	 * Allocate upper level page tables, if not already present.  Return
205 	 * early if a hugepage was created.
206 	 */
207 	pml4e = virt_create_upper_pte(vm, &vm->pgd, vaddr, paddr, PG_LEVEL_512G, level);
208 	if (*pml4e & PTE_LARGE_MASK)
209 		return;
210 
211 	pdpe = virt_create_upper_pte(vm, pml4e, vaddr, paddr, PG_LEVEL_1G, level);
212 	if (*pdpe & PTE_LARGE_MASK)
213 		return;
214 
215 	pde = virt_create_upper_pte(vm, pdpe, vaddr, paddr, PG_LEVEL_2M, level);
216 	if (*pde & PTE_LARGE_MASK)
217 		return;
218 
219 	/* Fill in page table entry. */
220 	pte = virt_get_pte(vm, pde, vaddr, PG_LEVEL_4K);
221 	TEST_ASSERT(!(*pte & PTE_PRESENT_MASK),
222 		    "PTE already present for 4k page at vaddr: 0x%lx\n", vaddr);
223 	*pte = PTE_PRESENT_MASK | PTE_WRITABLE_MASK | (paddr & PHYSICAL_PAGE_MASK);
224 }
225 
226 void virt_arch_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr)
227 {
228 	__virt_pg_map(vm, vaddr, paddr, PG_LEVEL_4K);
229 }
230 
231 void virt_map_level(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr,
232 		    uint64_t nr_bytes, int level)
233 {
234 	uint64_t pg_size = PG_LEVEL_SIZE(level);
235 	uint64_t nr_pages = nr_bytes / pg_size;
236 	int i;
237 
238 	TEST_ASSERT(nr_bytes % pg_size == 0,
239 		    "Region size not aligned: nr_bytes: 0x%lx, page size: 0x%lx",
240 		    nr_bytes, pg_size);
241 
242 	for (i = 0; i < nr_pages; i++) {
243 		__virt_pg_map(vm, vaddr, paddr, level);
244 
245 		vaddr += pg_size;
246 		paddr += pg_size;
247 	}
248 }
249 
250 static bool vm_is_target_pte(uint64_t *pte, int *level, int current_level)
251 {
252 	if (*pte & PTE_LARGE_MASK) {
253 		TEST_ASSERT(*level == PG_LEVEL_NONE ||
254 			    *level == current_level,
255 			    "Unexpected hugepage at level %d\n", current_level);
256 		*level = current_level;
257 	}
258 
259 	return *level == current_level;
260 }
261 
262 uint64_t *__vm_get_page_table_entry(struct kvm_vm *vm, uint64_t vaddr,
263 				    int *level)
264 {
265 	uint64_t *pml4e, *pdpe, *pde;
266 
267 	TEST_ASSERT(*level >= PG_LEVEL_NONE && *level < PG_LEVEL_NUM,
268 		    "Invalid PG_LEVEL_* '%d'", *level);
269 
270 	TEST_ASSERT(vm->mode == VM_MODE_PXXV48_4K, "Attempt to use "
271 		"unknown or unsupported guest mode, mode: 0x%x", vm->mode);
272 	TEST_ASSERT(sparsebit_is_set(vm->vpages_valid,
273 		(vaddr >> vm->page_shift)),
274 		"Invalid virtual address, vaddr: 0x%lx",
275 		vaddr);
276 	/*
277 	 * Based on the mode check above there are 48 bits in the vaddr, so
278 	 * shift 16 to sign extend the last bit (bit-47),
279 	 */
280 	TEST_ASSERT(vaddr == (((int64_t)vaddr << 16) >> 16),
281 		"Canonical check failed.  The virtual address is invalid.");
282 
283 	pml4e = virt_get_pte(vm, &vm->pgd, vaddr, PG_LEVEL_512G);
284 	if (vm_is_target_pte(pml4e, level, PG_LEVEL_512G))
285 		return pml4e;
286 
287 	pdpe = virt_get_pte(vm, pml4e, vaddr, PG_LEVEL_1G);
288 	if (vm_is_target_pte(pdpe, level, PG_LEVEL_1G))
289 		return pdpe;
290 
291 	pde = virt_get_pte(vm, pdpe, vaddr, PG_LEVEL_2M);
292 	if (vm_is_target_pte(pde, level, PG_LEVEL_2M))
293 		return pde;
294 
295 	return virt_get_pte(vm, pde, vaddr, PG_LEVEL_4K);
296 }
297 
298 uint64_t *vm_get_page_table_entry(struct kvm_vm *vm, uint64_t vaddr)
299 {
300 	int level = PG_LEVEL_4K;
301 
302 	return __vm_get_page_table_entry(vm, vaddr, &level);
303 }
304 
305 void virt_arch_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent)
306 {
307 	uint64_t *pml4e, *pml4e_start;
308 	uint64_t *pdpe, *pdpe_start;
309 	uint64_t *pde, *pde_start;
310 	uint64_t *pte, *pte_start;
311 
312 	if (!vm->pgd_created)
313 		return;
314 
315 	fprintf(stream, "%*s                                          "
316 		"                no\n", indent, "");
317 	fprintf(stream, "%*s      index hvaddr         gpaddr         "
318 		"addr         w exec dirty\n",
319 		indent, "");
320 	pml4e_start = (uint64_t *) addr_gpa2hva(vm, vm->pgd);
321 	for (uint16_t n1 = 0; n1 <= 0x1ffu; n1++) {
322 		pml4e = &pml4e_start[n1];
323 		if (!(*pml4e & PTE_PRESENT_MASK))
324 			continue;
325 		fprintf(stream, "%*spml4e 0x%-3zx %p 0x%-12lx 0x%-10llx %u "
326 			" %u\n",
327 			indent, "",
328 			pml4e - pml4e_start, pml4e,
329 			addr_hva2gpa(vm, pml4e), PTE_GET_PFN(*pml4e),
330 			!!(*pml4e & PTE_WRITABLE_MASK), !!(*pml4e & PTE_NX_MASK));
331 
332 		pdpe_start = addr_gpa2hva(vm, *pml4e & PHYSICAL_PAGE_MASK);
333 		for (uint16_t n2 = 0; n2 <= 0x1ffu; n2++) {
334 			pdpe = &pdpe_start[n2];
335 			if (!(*pdpe & PTE_PRESENT_MASK))
336 				continue;
337 			fprintf(stream, "%*spdpe  0x%-3zx %p 0x%-12lx 0x%-10llx "
338 				"%u  %u\n",
339 				indent, "",
340 				pdpe - pdpe_start, pdpe,
341 				addr_hva2gpa(vm, pdpe),
342 				PTE_GET_PFN(*pdpe), !!(*pdpe & PTE_WRITABLE_MASK),
343 				!!(*pdpe & PTE_NX_MASK));
344 
345 			pde_start = addr_gpa2hva(vm, *pdpe & PHYSICAL_PAGE_MASK);
346 			for (uint16_t n3 = 0; n3 <= 0x1ffu; n3++) {
347 				pde = &pde_start[n3];
348 				if (!(*pde & PTE_PRESENT_MASK))
349 					continue;
350 				fprintf(stream, "%*spde   0x%-3zx %p "
351 					"0x%-12lx 0x%-10llx %u  %u\n",
352 					indent, "", pde - pde_start, pde,
353 					addr_hva2gpa(vm, pde),
354 					PTE_GET_PFN(*pde), !!(*pde & PTE_WRITABLE_MASK),
355 					!!(*pde & PTE_NX_MASK));
356 
357 				pte_start = addr_gpa2hva(vm, *pde & PHYSICAL_PAGE_MASK);
358 				for (uint16_t n4 = 0; n4 <= 0x1ffu; n4++) {
359 					pte = &pte_start[n4];
360 					if (!(*pte & PTE_PRESENT_MASK))
361 						continue;
362 					fprintf(stream, "%*spte   0x%-3zx %p "
363 						"0x%-12lx 0x%-10llx %u  %u "
364 						"    %u    0x%-10lx\n",
365 						indent, "",
366 						pte - pte_start, pte,
367 						addr_hva2gpa(vm, pte),
368 						PTE_GET_PFN(*pte),
369 						!!(*pte & PTE_WRITABLE_MASK),
370 						!!(*pte & PTE_NX_MASK),
371 						!!(*pte & PTE_DIRTY_MASK),
372 						((uint64_t) n1 << 27)
373 							| ((uint64_t) n2 << 18)
374 							| ((uint64_t) n3 << 9)
375 							| ((uint64_t) n4));
376 				}
377 			}
378 		}
379 	}
380 }
381 
382 /*
383  * Set Unusable Segment
384  *
385  * Input Args: None
386  *
387  * Output Args:
388  *   segp - Pointer to segment register
389  *
390  * Return: None
391  *
392  * Sets the segment register pointed to by @segp to an unusable state.
393  */
394 static void kvm_seg_set_unusable(struct kvm_segment *segp)
395 {
396 	memset(segp, 0, sizeof(*segp));
397 	segp->unusable = true;
398 }
399 
400 static void kvm_seg_fill_gdt_64bit(struct kvm_vm *vm, struct kvm_segment *segp)
401 {
402 	void *gdt = addr_gva2hva(vm, vm->gdt);
403 	struct desc64 *desc = gdt + (segp->selector >> 3) * 8;
404 
405 	desc->limit0 = segp->limit & 0xFFFF;
406 	desc->base0 = segp->base & 0xFFFF;
407 	desc->base1 = segp->base >> 16;
408 	desc->type = segp->type;
409 	desc->s = segp->s;
410 	desc->dpl = segp->dpl;
411 	desc->p = segp->present;
412 	desc->limit1 = segp->limit >> 16;
413 	desc->avl = segp->avl;
414 	desc->l = segp->l;
415 	desc->db = segp->db;
416 	desc->g = segp->g;
417 	desc->base2 = segp->base >> 24;
418 	if (!segp->s)
419 		desc->base3 = segp->base >> 32;
420 }
421 
422 
423 /*
424  * Set Long Mode Flat Kernel Code Segment
425  *
426  * Input Args:
427  *   vm - VM whose GDT is being filled, or NULL to only write segp
428  *   selector - selector value
429  *
430  * Output Args:
431  *   segp - Pointer to KVM segment
432  *
433  * Return: None
434  *
435  * Sets up the KVM segment pointed to by @segp, to be a code segment
436  * with the selector value given by @selector.
437  */
438 static void kvm_seg_set_kernel_code_64bit(struct kvm_vm *vm, uint16_t selector,
439 	struct kvm_segment *segp)
440 {
441 	memset(segp, 0, sizeof(*segp));
442 	segp->selector = selector;
443 	segp->limit = 0xFFFFFFFFu;
444 	segp->s = 0x1; /* kTypeCodeData */
445 	segp->type = 0x08 | 0x01 | 0x02; /* kFlagCode | kFlagCodeAccessed
446 					  * | kFlagCodeReadable
447 					  */
448 	segp->g = true;
449 	segp->l = true;
450 	segp->present = 1;
451 	if (vm)
452 		kvm_seg_fill_gdt_64bit(vm, segp);
453 }
454 
455 /*
456  * Set Long Mode Flat Kernel Data Segment
457  *
458  * Input Args:
459  *   vm - VM whose GDT is being filled, or NULL to only write segp
460  *   selector - selector value
461  *
462  * Output Args:
463  *   segp - Pointer to KVM segment
464  *
465  * Return: None
466  *
467  * Sets up the KVM segment pointed to by @segp, to be a data segment
468  * with the selector value given by @selector.
469  */
470 static void kvm_seg_set_kernel_data_64bit(struct kvm_vm *vm, uint16_t selector,
471 	struct kvm_segment *segp)
472 {
473 	memset(segp, 0, sizeof(*segp));
474 	segp->selector = selector;
475 	segp->limit = 0xFFFFFFFFu;
476 	segp->s = 0x1; /* kTypeCodeData */
477 	segp->type = 0x00 | 0x01 | 0x02; /* kFlagData | kFlagDataAccessed
478 					  * | kFlagDataWritable
479 					  */
480 	segp->g = true;
481 	segp->present = true;
482 	if (vm)
483 		kvm_seg_fill_gdt_64bit(vm, segp);
484 }
485 
486 vm_paddr_t addr_arch_gva2gpa(struct kvm_vm *vm, vm_vaddr_t gva)
487 {
488 	int level = PG_LEVEL_NONE;
489 	uint64_t *pte = __vm_get_page_table_entry(vm, gva, &level);
490 
491 	TEST_ASSERT(*pte & PTE_PRESENT_MASK,
492 		    "Leaf PTE not PRESENT for gva: 0x%08lx", gva);
493 
494 	/*
495 	 * No need for a hugepage mask on the PTE, x86-64 requires the "unused"
496 	 * address bits to be zero.
497 	 */
498 	return PTE_GET_PA(*pte) | (gva & ~HUGEPAGE_MASK(level));
499 }
500 
501 static void kvm_setup_gdt(struct kvm_vm *vm, struct kvm_dtable *dt)
502 {
503 	if (!vm->gdt)
504 		vm->gdt = __vm_vaddr_alloc_page(vm, MEM_REGION_DATA);
505 
506 	dt->base = vm->gdt;
507 	dt->limit = getpagesize();
508 }
509 
510 static void kvm_setup_tss_64bit(struct kvm_vm *vm, struct kvm_segment *segp,
511 				int selector)
512 {
513 	if (!vm->tss)
514 		vm->tss = __vm_vaddr_alloc_page(vm, MEM_REGION_DATA);
515 
516 	memset(segp, 0, sizeof(*segp));
517 	segp->base = vm->tss;
518 	segp->limit = 0x67;
519 	segp->selector = selector;
520 	segp->type = 0xb;
521 	segp->present = 1;
522 	kvm_seg_fill_gdt_64bit(vm, segp);
523 }
524 
525 static void vcpu_setup(struct kvm_vm *vm, struct kvm_vcpu *vcpu)
526 {
527 	struct kvm_sregs sregs;
528 
529 	/* Set mode specific system register values. */
530 	vcpu_sregs_get(vcpu, &sregs);
531 
532 	sregs.idt.limit = 0;
533 
534 	kvm_setup_gdt(vm, &sregs.gdt);
535 
536 	switch (vm->mode) {
537 	case VM_MODE_PXXV48_4K:
538 		sregs.cr0 = X86_CR0_PE | X86_CR0_NE | X86_CR0_PG;
539 		sregs.cr4 |= X86_CR4_PAE | X86_CR4_OSFXSR;
540 		sregs.efer |= (EFER_LME | EFER_LMA | EFER_NX);
541 
542 		kvm_seg_set_unusable(&sregs.ldt);
543 		kvm_seg_set_kernel_code_64bit(vm, DEFAULT_CODE_SELECTOR, &sregs.cs);
544 		kvm_seg_set_kernel_data_64bit(vm, DEFAULT_DATA_SELECTOR, &sregs.ds);
545 		kvm_seg_set_kernel_data_64bit(vm, DEFAULT_DATA_SELECTOR, &sregs.es);
546 		kvm_setup_tss_64bit(vm, &sregs.tr, 0x18);
547 		break;
548 
549 	default:
550 		TEST_FAIL("Unknown guest mode, mode: 0x%x", vm->mode);
551 	}
552 
553 	sregs.cr3 = vm->pgd;
554 	vcpu_sregs_set(vcpu, &sregs);
555 }
556 
557 void kvm_arch_vm_post_create(struct kvm_vm *vm)
558 {
559 	vm_create_irqchip(vm);
560 	sync_global_to_guest(vm, host_cpu_is_intel);
561 	sync_global_to_guest(vm, host_cpu_is_amd);
562 }
563 
564 struct kvm_vcpu *vm_arch_vcpu_add(struct kvm_vm *vm, uint32_t vcpu_id,
565 				  void *guest_code)
566 {
567 	struct kvm_mp_state mp_state;
568 	struct kvm_regs regs;
569 	vm_vaddr_t stack_vaddr;
570 	struct kvm_vcpu *vcpu;
571 
572 	stack_vaddr = __vm_vaddr_alloc(vm, DEFAULT_STACK_PGS * getpagesize(),
573 				       DEFAULT_GUEST_STACK_VADDR_MIN,
574 				       MEM_REGION_DATA);
575 
576 	vcpu = __vm_vcpu_add(vm, vcpu_id);
577 	vcpu_init_cpuid(vcpu, kvm_get_supported_cpuid());
578 	vcpu_setup(vm, vcpu);
579 
580 	/* Setup guest general purpose registers */
581 	vcpu_regs_get(vcpu, &regs);
582 	regs.rflags = regs.rflags | 0x2;
583 	regs.rsp = stack_vaddr + (DEFAULT_STACK_PGS * getpagesize());
584 	regs.rip = (unsigned long) guest_code;
585 	vcpu_regs_set(vcpu, &regs);
586 
587 	/* Setup the MP state */
588 	mp_state.mp_state = 0;
589 	vcpu_mp_state_set(vcpu, &mp_state);
590 
591 	return vcpu;
592 }
593 
594 struct kvm_vcpu *vm_arch_vcpu_recreate(struct kvm_vm *vm, uint32_t vcpu_id)
595 {
596 	struct kvm_vcpu *vcpu = __vm_vcpu_add(vm, vcpu_id);
597 
598 	vcpu_init_cpuid(vcpu, kvm_get_supported_cpuid());
599 
600 	return vcpu;
601 }
602 
603 void vcpu_arch_free(struct kvm_vcpu *vcpu)
604 {
605 	if (vcpu->cpuid)
606 		free(vcpu->cpuid);
607 }
608 
609 /* Do not use kvm_supported_cpuid directly except for validity checks. */
610 static void *kvm_supported_cpuid;
611 
612 const struct kvm_cpuid2 *kvm_get_supported_cpuid(void)
613 {
614 	int kvm_fd;
615 
616 	if (kvm_supported_cpuid)
617 		return kvm_supported_cpuid;
618 
619 	kvm_supported_cpuid = allocate_kvm_cpuid2(MAX_NR_CPUID_ENTRIES);
620 	kvm_fd = open_kvm_dev_path_or_exit();
621 
622 	kvm_ioctl(kvm_fd, KVM_GET_SUPPORTED_CPUID,
623 		  (struct kvm_cpuid2 *)kvm_supported_cpuid);
624 
625 	close(kvm_fd);
626 	return kvm_supported_cpuid;
627 }
628 
629 static uint32_t __kvm_cpu_has(const struct kvm_cpuid2 *cpuid,
630 			      uint32_t function, uint32_t index,
631 			      uint8_t reg, uint8_t lo, uint8_t hi)
632 {
633 	const struct kvm_cpuid_entry2 *entry;
634 	int i;
635 
636 	for (i = 0; i < cpuid->nent; i++) {
637 		entry = &cpuid->entries[i];
638 
639 		/*
640 		 * The output registers in kvm_cpuid_entry2 are in alphabetical
641 		 * order, but kvm_x86_cpu_feature matches that mess, so yay
642 		 * pointer shenanigans!
643 		 */
644 		if (entry->function == function && entry->index == index)
645 			return ((&entry->eax)[reg] & GENMASK(hi, lo)) >> lo;
646 	}
647 
648 	return 0;
649 }
650 
651 bool kvm_cpuid_has(const struct kvm_cpuid2 *cpuid,
652 		   struct kvm_x86_cpu_feature feature)
653 {
654 	return __kvm_cpu_has(cpuid, feature.function, feature.index,
655 			     feature.reg, feature.bit, feature.bit);
656 }
657 
658 uint32_t kvm_cpuid_property(const struct kvm_cpuid2 *cpuid,
659 			    struct kvm_x86_cpu_property property)
660 {
661 	return __kvm_cpu_has(cpuid, property.function, property.index,
662 			     property.reg, property.lo_bit, property.hi_bit);
663 }
664 
665 uint64_t kvm_get_feature_msr(uint64_t msr_index)
666 {
667 	struct {
668 		struct kvm_msrs header;
669 		struct kvm_msr_entry entry;
670 	} buffer = {};
671 	int r, kvm_fd;
672 
673 	buffer.header.nmsrs = 1;
674 	buffer.entry.index = msr_index;
675 	kvm_fd = open_kvm_dev_path_or_exit();
676 
677 	r = __kvm_ioctl(kvm_fd, KVM_GET_MSRS, &buffer.header);
678 	TEST_ASSERT(r == 1, KVM_IOCTL_ERROR(KVM_GET_MSRS, r));
679 
680 	close(kvm_fd);
681 	return buffer.entry.data;
682 }
683 
684 void __vm_xsave_require_permission(int bit, const char *name)
685 {
686 	int kvm_fd;
687 	u64 bitmask;
688 	long rc;
689 	struct kvm_device_attr attr = {
690 		.group = 0,
691 		.attr = KVM_X86_XCOMP_GUEST_SUPP,
692 		.addr = (unsigned long) &bitmask
693 	};
694 
695 	TEST_ASSERT(!kvm_supported_cpuid,
696 		    "kvm_get_supported_cpuid() cannot be used before ARCH_REQ_XCOMP_GUEST_PERM");
697 
698 	kvm_fd = open_kvm_dev_path_or_exit();
699 	rc = __kvm_ioctl(kvm_fd, KVM_GET_DEVICE_ATTR, &attr);
700 	close(kvm_fd);
701 
702 	if (rc == -1 && (errno == ENXIO || errno == EINVAL))
703 		__TEST_REQUIRE(0, "KVM_X86_XCOMP_GUEST_SUPP not supported");
704 
705 	TEST_ASSERT(rc == 0, "KVM_GET_DEVICE_ATTR(0, KVM_X86_XCOMP_GUEST_SUPP) error: %ld", rc);
706 
707 	__TEST_REQUIRE(bitmask & (1ULL << bit),
708 		       "Required XSAVE feature '%s' not supported", name);
709 
710 	TEST_REQUIRE(!syscall(SYS_arch_prctl, ARCH_REQ_XCOMP_GUEST_PERM, bit));
711 
712 	rc = syscall(SYS_arch_prctl, ARCH_GET_XCOMP_GUEST_PERM, &bitmask);
713 	TEST_ASSERT(rc == 0, "prctl(ARCH_GET_XCOMP_GUEST_PERM) error: %ld", rc);
714 	TEST_ASSERT(bitmask & (1ULL << bit),
715 		    "prctl(ARCH_REQ_XCOMP_GUEST_PERM) failure bitmask=0x%lx",
716 		    bitmask);
717 }
718 
719 void vcpu_init_cpuid(struct kvm_vcpu *vcpu, const struct kvm_cpuid2 *cpuid)
720 {
721 	TEST_ASSERT(cpuid != vcpu->cpuid, "@cpuid can't be the vCPU's CPUID");
722 
723 	/* Allow overriding the default CPUID. */
724 	if (vcpu->cpuid && vcpu->cpuid->nent < cpuid->nent) {
725 		free(vcpu->cpuid);
726 		vcpu->cpuid = NULL;
727 	}
728 
729 	if (!vcpu->cpuid)
730 		vcpu->cpuid = allocate_kvm_cpuid2(cpuid->nent);
731 
732 	memcpy(vcpu->cpuid, cpuid, kvm_cpuid2_size(cpuid->nent));
733 	vcpu_set_cpuid(vcpu);
734 }
735 
736 void vcpu_set_cpuid_maxphyaddr(struct kvm_vcpu *vcpu, uint8_t maxphyaddr)
737 {
738 	struct kvm_cpuid_entry2 *entry = vcpu_get_cpuid_entry(vcpu, 0x80000008);
739 
740 	entry->eax = (entry->eax & ~0xff) | maxphyaddr;
741 	vcpu_set_cpuid(vcpu);
742 }
743 
744 void vcpu_clear_cpuid_entry(struct kvm_vcpu *vcpu, uint32_t function)
745 {
746 	struct kvm_cpuid_entry2 *entry = vcpu_get_cpuid_entry(vcpu, function);
747 
748 	entry->eax = 0;
749 	entry->ebx = 0;
750 	entry->ecx = 0;
751 	entry->edx = 0;
752 	vcpu_set_cpuid(vcpu);
753 }
754 
755 void vcpu_set_or_clear_cpuid_feature(struct kvm_vcpu *vcpu,
756 				     struct kvm_x86_cpu_feature feature,
757 				     bool set)
758 {
759 	struct kvm_cpuid_entry2 *entry;
760 	u32 *reg;
761 
762 	entry = __vcpu_get_cpuid_entry(vcpu, feature.function, feature.index);
763 	reg = (&entry->eax) + feature.reg;
764 
765 	if (set)
766 		*reg |= BIT(feature.bit);
767 	else
768 		*reg &= ~BIT(feature.bit);
769 
770 	vcpu_set_cpuid(vcpu);
771 }
772 
773 uint64_t vcpu_get_msr(struct kvm_vcpu *vcpu, uint64_t msr_index)
774 {
775 	struct {
776 		struct kvm_msrs header;
777 		struct kvm_msr_entry entry;
778 	} buffer = {};
779 
780 	buffer.header.nmsrs = 1;
781 	buffer.entry.index = msr_index;
782 
783 	vcpu_msrs_get(vcpu, &buffer.header);
784 
785 	return buffer.entry.data;
786 }
787 
788 int _vcpu_set_msr(struct kvm_vcpu *vcpu, uint64_t msr_index, uint64_t msr_value)
789 {
790 	struct {
791 		struct kvm_msrs header;
792 		struct kvm_msr_entry entry;
793 	} buffer = {};
794 
795 	memset(&buffer, 0, sizeof(buffer));
796 	buffer.header.nmsrs = 1;
797 	buffer.entry.index = msr_index;
798 	buffer.entry.data = msr_value;
799 
800 	return __vcpu_ioctl(vcpu, KVM_SET_MSRS, &buffer.header);
801 }
802 
803 void vcpu_args_set(struct kvm_vcpu *vcpu, unsigned int num, ...)
804 {
805 	va_list ap;
806 	struct kvm_regs regs;
807 
808 	TEST_ASSERT(num >= 1 && num <= 6, "Unsupported number of args,\n"
809 		    "  num: %u\n",
810 		    num);
811 
812 	va_start(ap, num);
813 	vcpu_regs_get(vcpu, &regs);
814 
815 	if (num >= 1)
816 		regs.rdi = va_arg(ap, uint64_t);
817 
818 	if (num >= 2)
819 		regs.rsi = va_arg(ap, uint64_t);
820 
821 	if (num >= 3)
822 		regs.rdx = va_arg(ap, uint64_t);
823 
824 	if (num >= 4)
825 		regs.rcx = va_arg(ap, uint64_t);
826 
827 	if (num >= 5)
828 		regs.r8 = va_arg(ap, uint64_t);
829 
830 	if (num >= 6)
831 		regs.r9 = va_arg(ap, uint64_t);
832 
833 	vcpu_regs_set(vcpu, &regs);
834 	va_end(ap);
835 }
836 
837 void vcpu_arch_dump(FILE *stream, struct kvm_vcpu *vcpu, uint8_t indent)
838 {
839 	struct kvm_regs regs;
840 	struct kvm_sregs sregs;
841 
842 	fprintf(stream, "%*svCPU ID: %u\n", indent, "", vcpu->id);
843 
844 	fprintf(stream, "%*sregs:\n", indent + 2, "");
845 	vcpu_regs_get(vcpu, &regs);
846 	regs_dump(stream, &regs, indent + 4);
847 
848 	fprintf(stream, "%*ssregs:\n", indent + 2, "");
849 	vcpu_sregs_get(vcpu, &sregs);
850 	sregs_dump(stream, &sregs, indent + 4);
851 }
852 
853 static struct kvm_msr_list *__kvm_get_msr_index_list(bool feature_msrs)
854 {
855 	struct kvm_msr_list *list;
856 	struct kvm_msr_list nmsrs;
857 	int kvm_fd, r;
858 
859 	kvm_fd = open_kvm_dev_path_or_exit();
860 
861 	nmsrs.nmsrs = 0;
862 	if (!feature_msrs)
863 		r = __kvm_ioctl(kvm_fd, KVM_GET_MSR_INDEX_LIST, &nmsrs);
864 	else
865 		r = __kvm_ioctl(kvm_fd, KVM_GET_MSR_FEATURE_INDEX_LIST, &nmsrs);
866 
867 	TEST_ASSERT(r == -1 && errno == E2BIG,
868 		    "Expected -E2BIG, got rc: %i errno: %i (%s)",
869 		    r, errno, strerror(errno));
870 
871 	list = malloc(sizeof(*list) + nmsrs.nmsrs * sizeof(list->indices[0]));
872 	TEST_ASSERT(list, "-ENOMEM when allocating MSR index list");
873 	list->nmsrs = nmsrs.nmsrs;
874 
875 	if (!feature_msrs)
876 		kvm_ioctl(kvm_fd, KVM_GET_MSR_INDEX_LIST, list);
877 	else
878 		kvm_ioctl(kvm_fd, KVM_GET_MSR_FEATURE_INDEX_LIST, list);
879 	close(kvm_fd);
880 
881 	TEST_ASSERT(list->nmsrs == nmsrs.nmsrs,
882 		    "Number of MSRs in list changed, was %d, now %d",
883 		    nmsrs.nmsrs, list->nmsrs);
884 	return list;
885 }
886 
887 const struct kvm_msr_list *kvm_get_msr_index_list(void)
888 {
889 	static const struct kvm_msr_list *list;
890 
891 	if (!list)
892 		list = __kvm_get_msr_index_list(false);
893 	return list;
894 }
895 
896 
897 const struct kvm_msr_list *kvm_get_feature_msr_index_list(void)
898 {
899 	static const struct kvm_msr_list *list;
900 
901 	if (!list)
902 		list = __kvm_get_msr_index_list(true);
903 	return list;
904 }
905 
906 bool kvm_msr_is_in_save_restore_list(uint32_t msr_index)
907 {
908 	const struct kvm_msr_list *list = kvm_get_msr_index_list();
909 	int i;
910 
911 	for (i = 0; i < list->nmsrs; ++i) {
912 		if (list->indices[i] == msr_index)
913 			return true;
914 	}
915 
916 	return false;
917 }
918 
919 static void vcpu_save_xsave_state(struct kvm_vcpu *vcpu,
920 				  struct kvm_x86_state *state)
921 {
922 	int size = vm_check_cap(vcpu->vm, KVM_CAP_XSAVE2);
923 
924 	if (size) {
925 		state->xsave = malloc(size);
926 		vcpu_xsave2_get(vcpu, state->xsave);
927 	} else {
928 		state->xsave = malloc(sizeof(struct kvm_xsave));
929 		vcpu_xsave_get(vcpu, state->xsave);
930 	}
931 }
932 
933 struct kvm_x86_state *vcpu_save_state(struct kvm_vcpu *vcpu)
934 {
935 	const struct kvm_msr_list *msr_list = kvm_get_msr_index_list();
936 	struct kvm_x86_state *state;
937 	int i;
938 
939 	static int nested_size = -1;
940 
941 	if (nested_size == -1) {
942 		nested_size = kvm_check_cap(KVM_CAP_NESTED_STATE);
943 		TEST_ASSERT(nested_size <= sizeof(state->nested_),
944 			    "Nested state size too big, %i > %zi",
945 			    nested_size, sizeof(state->nested_));
946 	}
947 
948 	/*
949 	 * When KVM exits to userspace with KVM_EXIT_IO, KVM guarantees
950 	 * guest state is consistent only after userspace re-enters the
951 	 * kernel with KVM_RUN.  Complete IO prior to migrating state
952 	 * to a new VM.
953 	 */
954 	vcpu_run_complete_io(vcpu);
955 
956 	state = malloc(sizeof(*state) + msr_list->nmsrs * sizeof(state->msrs.entries[0]));
957 
958 	vcpu_events_get(vcpu, &state->events);
959 	vcpu_mp_state_get(vcpu, &state->mp_state);
960 	vcpu_regs_get(vcpu, &state->regs);
961 	vcpu_save_xsave_state(vcpu, state);
962 
963 	if (kvm_has_cap(KVM_CAP_XCRS))
964 		vcpu_xcrs_get(vcpu, &state->xcrs);
965 
966 	vcpu_sregs_get(vcpu, &state->sregs);
967 
968 	if (nested_size) {
969 		state->nested.size = sizeof(state->nested_);
970 
971 		vcpu_nested_state_get(vcpu, &state->nested);
972 		TEST_ASSERT(state->nested.size <= nested_size,
973 			    "Nested state size too big, %i (KVM_CHECK_CAP gave %i)",
974 			    state->nested.size, nested_size);
975 	} else {
976 		state->nested.size = 0;
977 	}
978 
979 	state->msrs.nmsrs = msr_list->nmsrs;
980 	for (i = 0; i < msr_list->nmsrs; i++)
981 		state->msrs.entries[i].index = msr_list->indices[i];
982 	vcpu_msrs_get(vcpu, &state->msrs);
983 
984 	vcpu_debugregs_get(vcpu, &state->debugregs);
985 
986 	return state;
987 }
988 
989 void vcpu_load_state(struct kvm_vcpu *vcpu, struct kvm_x86_state *state)
990 {
991 	vcpu_sregs_set(vcpu, &state->sregs);
992 	vcpu_msrs_set(vcpu, &state->msrs);
993 
994 	if (kvm_has_cap(KVM_CAP_XCRS))
995 		vcpu_xcrs_set(vcpu, &state->xcrs);
996 
997 	vcpu_xsave_set(vcpu,  state->xsave);
998 	vcpu_events_set(vcpu, &state->events);
999 	vcpu_mp_state_set(vcpu, &state->mp_state);
1000 	vcpu_debugregs_set(vcpu, &state->debugregs);
1001 	vcpu_regs_set(vcpu, &state->regs);
1002 
1003 	if (state->nested.size)
1004 		vcpu_nested_state_set(vcpu, &state->nested);
1005 }
1006 
1007 void kvm_x86_state_cleanup(struct kvm_x86_state *state)
1008 {
1009 	free(state->xsave);
1010 	free(state);
1011 }
1012 
1013 void kvm_get_cpu_address_width(unsigned int *pa_bits, unsigned int *va_bits)
1014 {
1015 	if (!kvm_cpu_has_p(X86_PROPERTY_MAX_PHY_ADDR)) {
1016 		*pa_bits = kvm_cpu_has(X86_FEATURE_PAE) ? 36 : 32;
1017 		*va_bits = 32;
1018 	} else {
1019 		*pa_bits = kvm_cpu_property(X86_PROPERTY_MAX_PHY_ADDR);
1020 		*va_bits = kvm_cpu_property(X86_PROPERTY_MAX_VIRT_ADDR);
1021 	}
1022 }
1023 
1024 static void set_idt_entry(struct kvm_vm *vm, int vector, unsigned long addr,
1025 			  int dpl, unsigned short selector)
1026 {
1027 	struct idt_entry *base =
1028 		(struct idt_entry *)addr_gva2hva(vm, vm->idt);
1029 	struct idt_entry *e = &base[vector];
1030 
1031 	memset(e, 0, sizeof(*e));
1032 	e->offset0 = addr;
1033 	e->selector = selector;
1034 	e->ist = 0;
1035 	e->type = 14;
1036 	e->dpl = dpl;
1037 	e->p = 1;
1038 	e->offset1 = addr >> 16;
1039 	e->offset2 = addr >> 32;
1040 }
1041 
1042 
1043 static bool kvm_fixup_exception(struct ex_regs *regs)
1044 {
1045 	if (regs->r9 != KVM_EXCEPTION_MAGIC || regs->rip != regs->r10)
1046 		return false;
1047 
1048 	if (regs->vector == DE_VECTOR)
1049 		return false;
1050 
1051 	regs->rip = regs->r11;
1052 	regs->r9 = regs->vector;
1053 	regs->r10 = regs->error_code;
1054 	return true;
1055 }
1056 
1057 void kvm_exit_unexpected_vector(uint32_t value)
1058 {
1059 	ucall(UCALL_UNHANDLED, 1, value);
1060 }
1061 
1062 void route_exception(struct ex_regs *regs)
1063 {
1064 	typedef void(*handler)(struct ex_regs *);
1065 	handler *handlers = (handler *)exception_handlers;
1066 
1067 	if (handlers && handlers[regs->vector]) {
1068 		handlers[regs->vector](regs);
1069 		return;
1070 	}
1071 
1072 	if (kvm_fixup_exception(regs))
1073 		return;
1074 
1075 	kvm_exit_unexpected_vector(regs->vector);
1076 }
1077 
1078 void vm_init_descriptor_tables(struct kvm_vm *vm)
1079 {
1080 	extern void *idt_handlers;
1081 	int i;
1082 
1083 	vm->idt = __vm_vaddr_alloc_page(vm, MEM_REGION_DATA);
1084 	vm->handlers = __vm_vaddr_alloc_page(vm, MEM_REGION_DATA);
1085 	/* Handlers have the same address in both address spaces.*/
1086 	for (i = 0; i < NUM_INTERRUPTS; i++)
1087 		set_idt_entry(vm, i, (unsigned long)(&idt_handlers)[i], 0,
1088 			DEFAULT_CODE_SELECTOR);
1089 }
1090 
1091 void vcpu_init_descriptor_tables(struct kvm_vcpu *vcpu)
1092 {
1093 	struct kvm_vm *vm = vcpu->vm;
1094 	struct kvm_sregs sregs;
1095 
1096 	vcpu_sregs_get(vcpu, &sregs);
1097 	sregs.idt.base = vm->idt;
1098 	sregs.idt.limit = NUM_INTERRUPTS * sizeof(struct idt_entry) - 1;
1099 	sregs.gdt.base = vm->gdt;
1100 	sregs.gdt.limit = getpagesize() - 1;
1101 	kvm_seg_set_kernel_data_64bit(NULL, DEFAULT_DATA_SELECTOR, &sregs.gs);
1102 	vcpu_sregs_set(vcpu, &sregs);
1103 	*(vm_vaddr_t *)addr_gva2hva(vm, (vm_vaddr_t)(&exception_handlers)) = vm->handlers;
1104 }
1105 
1106 void vm_install_exception_handler(struct kvm_vm *vm, int vector,
1107 			       void (*handler)(struct ex_regs *))
1108 {
1109 	vm_vaddr_t *handlers = (vm_vaddr_t *)addr_gva2hva(vm, vm->handlers);
1110 
1111 	handlers[vector] = (vm_vaddr_t)handler;
1112 }
1113 
1114 void assert_on_unhandled_exception(struct kvm_vcpu *vcpu)
1115 {
1116 	struct ucall uc;
1117 
1118 	if (get_ucall(vcpu, &uc) == UCALL_UNHANDLED) {
1119 		uint64_t vector = uc.args[0];
1120 
1121 		TEST_FAIL("Unexpected vectored event in guest (vector:0x%lx)",
1122 			  vector);
1123 	}
1124 }
1125 
1126 const struct kvm_cpuid_entry2 *get_cpuid_entry(const struct kvm_cpuid2 *cpuid,
1127 					       uint32_t function, uint32_t index)
1128 {
1129 	int i;
1130 
1131 	for (i = 0; i < cpuid->nent; i++) {
1132 		if (cpuid->entries[i].function == function &&
1133 		    cpuid->entries[i].index == index)
1134 			return &cpuid->entries[i];
1135 	}
1136 
1137 	TEST_FAIL("CPUID function 0x%x index 0x%x not found ", function, index);
1138 
1139 	return NULL;
1140 }
1141 
1142 #define X86_HYPERCALL(inputs...)					\
1143 ({									\
1144 	uint64_t r;							\
1145 									\
1146 	asm volatile("test %[use_vmmcall], %[use_vmmcall]\n\t"		\
1147 		     "jnz 1f\n\t"					\
1148 		     "vmcall\n\t"					\
1149 		     "jmp 2f\n\t"					\
1150 		     "1: vmmcall\n\t"					\
1151 		     "2:"						\
1152 		     : "=a"(r)						\
1153 		     : [use_vmmcall] "r" (host_cpu_is_amd), inputs);	\
1154 									\
1155 	r;								\
1156 })
1157 
1158 uint64_t kvm_hypercall(uint64_t nr, uint64_t a0, uint64_t a1, uint64_t a2,
1159 		       uint64_t a3)
1160 {
1161 	return X86_HYPERCALL("a"(nr), "b"(a0), "c"(a1), "d"(a2), "S"(a3));
1162 }
1163 
1164 uint64_t __xen_hypercall(uint64_t nr, uint64_t a0, void *a1)
1165 {
1166 	return X86_HYPERCALL("a"(nr), "D"(a0), "S"(a1));
1167 }
1168 
1169 void xen_hypercall(uint64_t nr, uint64_t a0, void *a1)
1170 {
1171 	GUEST_ASSERT(!__xen_hypercall(nr, a0, a1));
1172 }
1173 
1174 const struct kvm_cpuid2 *kvm_get_supported_hv_cpuid(void)
1175 {
1176 	static struct kvm_cpuid2 *cpuid;
1177 	int kvm_fd;
1178 
1179 	if (cpuid)
1180 		return cpuid;
1181 
1182 	cpuid = allocate_kvm_cpuid2(MAX_NR_CPUID_ENTRIES);
1183 	kvm_fd = open_kvm_dev_path_or_exit();
1184 
1185 	kvm_ioctl(kvm_fd, KVM_GET_SUPPORTED_HV_CPUID, cpuid);
1186 
1187 	close(kvm_fd);
1188 	return cpuid;
1189 }
1190 
1191 void vcpu_set_hv_cpuid(struct kvm_vcpu *vcpu)
1192 {
1193 	static struct kvm_cpuid2 *cpuid_full;
1194 	const struct kvm_cpuid2 *cpuid_sys, *cpuid_hv;
1195 	int i, nent = 0;
1196 
1197 	if (!cpuid_full) {
1198 		cpuid_sys = kvm_get_supported_cpuid();
1199 		cpuid_hv = kvm_get_supported_hv_cpuid();
1200 
1201 		cpuid_full = allocate_kvm_cpuid2(cpuid_sys->nent + cpuid_hv->nent);
1202 		if (!cpuid_full) {
1203 			perror("malloc");
1204 			abort();
1205 		}
1206 
1207 		/* Need to skip KVM CPUID leaves 0x400000xx */
1208 		for (i = 0; i < cpuid_sys->nent; i++) {
1209 			if (cpuid_sys->entries[i].function >= 0x40000000 &&
1210 			    cpuid_sys->entries[i].function < 0x40000100)
1211 				continue;
1212 			cpuid_full->entries[nent] = cpuid_sys->entries[i];
1213 			nent++;
1214 		}
1215 
1216 		memcpy(&cpuid_full->entries[nent], cpuid_hv->entries,
1217 		       cpuid_hv->nent * sizeof(struct kvm_cpuid_entry2));
1218 		cpuid_full->nent = nent + cpuid_hv->nent;
1219 	}
1220 
1221 	vcpu_init_cpuid(vcpu, cpuid_full);
1222 }
1223 
1224 const struct kvm_cpuid2 *vcpu_get_supported_hv_cpuid(struct kvm_vcpu *vcpu)
1225 {
1226 	struct kvm_cpuid2 *cpuid = allocate_kvm_cpuid2(MAX_NR_CPUID_ENTRIES);
1227 
1228 	vcpu_ioctl(vcpu, KVM_GET_SUPPORTED_HV_CPUID, cpuid);
1229 
1230 	return cpuid;
1231 }
1232 
1233 unsigned long vm_compute_max_gfn(struct kvm_vm *vm)
1234 {
1235 	const unsigned long num_ht_pages = 12 << (30 - vm->page_shift); /* 12 GiB */
1236 	unsigned long ht_gfn, max_gfn, max_pfn;
1237 	uint8_t maxphyaddr;
1238 
1239 	max_gfn = (1ULL << (vm->pa_bits - vm->page_shift)) - 1;
1240 
1241 	/* Avoid reserved HyperTransport region on AMD processors.  */
1242 	if (!host_cpu_is_amd)
1243 		return max_gfn;
1244 
1245 	/* On parts with <40 physical address bits, the area is fully hidden */
1246 	if (vm->pa_bits < 40)
1247 		return max_gfn;
1248 
1249 	/* Before family 17h, the HyperTransport area is just below 1T.  */
1250 	ht_gfn = (1 << 28) - num_ht_pages;
1251 	if (this_cpu_family() < 0x17)
1252 		goto done;
1253 
1254 	/*
1255 	 * Otherwise it's at the top of the physical address space, possibly
1256 	 * reduced due to SME by bits 11:6 of CPUID[0x8000001f].EBX.  Use
1257 	 * the old conservative value if MAXPHYADDR is not enumerated.
1258 	 */
1259 	if (!this_cpu_has_p(X86_PROPERTY_MAX_PHY_ADDR))
1260 		goto done;
1261 
1262 	maxphyaddr = this_cpu_property(X86_PROPERTY_MAX_PHY_ADDR);
1263 	max_pfn = (1ULL << (maxphyaddr - vm->page_shift)) - 1;
1264 
1265 	if (this_cpu_has_p(X86_PROPERTY_PHYS_ADDR_REDUCTION))
1266 		max_pfn >>= this_cpu_property(X86_PROPERTY_PHYS_ADDR_REDUCTION);
1267 
1268 	ht_gfn = max_pfn - num_ht_pages;
1269 done:
1270 	return min(max_gfn, ht_gfn - 1);
1271 }
1272 
1273 /* Returns true if kvm_intel was loaded with unrestricted_guest=1. */
1274 bool vm_is_unrestricted_guest(struct kvm_vm *vm)
1275 {
1276 	/* Ensure that a KVM vendor-specific module is loaded. */
1277 	if (vm == NULL)
1278 		close(open_kvm_dev_path_or_exit());
1279 
1280 	return get_kvm_intel_param_bool("unrestricted_guest");
1281 }
1282 
1283 void kvm_selftest_arch_init(void)
1284 {
1285 	host_cpu_is_intel = this_cpu_is_intel();
1286 	host_cpu_is_amd = this_cpu_is_amd();
1287 }
1288