1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * tools/testing/selftests/kvm/lib/kvm_util.c
4  *
5  * Copyright (C) 2018, Google LLC.
6  */
7 
8 #define _GNU_SOURCE /* for program_invocation_name */
9 #include "test_util.h"
10 #include "kvm_util.h"
11 #include "processor.h"
12 
13 #include <assert.h>
14 #include <sched.h>
15 #include <sys/mman.h>
16 #include <sys/types.h>
17 #include <sys/stat.h>
18 #include <unistd.h>
19 #include <linux/kernel.h>
20 
21 #define KVM_UTIL_MIN_PFN	2
22 
23 static int vcpu_mmap_sz(void);
24 
25 int open_path_or_exit(const char *path, int flags)
26 {
27 	int fd;
28 
29 	fd = open(path, flags);
30 	__TEST_REQUIRE(fd >= 0, "%s not available (errno: %d)", path, errno);
31 
32 	return fd;
33 }
34 
35 /*
36  * Open KVM_DEV_PATH if available, otherwise exit the entire program.
37  *
38  * Input Args:
39  *   flags - The flags to pass when opening KVM_DEV_PATH.
40  *
41  * Return:
42  *   The opened file descriptor of /dev/kvm.
43  */
44 static int _open_kvm_dev_path_or_exit(int flags)
45 {
46 	return open_path_or_exit(KVM_DEV_PATH, flags);
47 }
48 
49 int open_kvm_dev_path_or_exit(void)
50 {
51 	return _open_kvm_dev_path_or_exit(O_RDONLY);
52 }
53 
54 static bool get_module_param_bool(const char *module_name, const char *param)
55 {
56 	const int path_size = 128;
57 	char path[path_size];
58 	char value;
59 	ssize_t r;
60 	int fd;
61 
62 	r = snprintf(path, path_size, "/sys/module/%s/parameters/%s",
63 		     module_name, param);
64 	TEST_ASSERT(r < path_size,
65 		    "Failed to construct sysfs path in %d bytes.", path_size);
66 
67 	fd = open_path_or_exit(path, O_RDONLY);
68 
69 	r = read(fd, &value, 1);
70 	TEST_ASSERT(r == 1, "read(%s) failed", path);
71 
72 	r = close(fd);
73 	TEST_ASSERT(!r, "close(%s) failed", path);
74 
75 	if (value == 'Y')
76 		return true;
77 	else if (value == 'N')
78 		return false;
79 
80 	TEST_FAIL("Unrecognized value '%c' for boolean module param", value);
81 }
82 
83 bool get_kvm_intel_param_bool(const char *param)
84 {
85 	return get_module_param_bool("kvm_intel", param);
86 }
87 
88 bool get_kvm_amd_param_bool(const char *param)
89 {
90 	return get_module_param_bool("kvm_amd", param);
91 }
92 
93 /*
94  * Capability
95  *
96  * Input Args:
97  *   cap - Capability
98  *
99  * Output Args: None
100  *
101  * Return:
102  *   On success, the Value corresponding to the capability (KVM_CAP_*)
103  *   specified by the value of cap.  On failure a TEST_ASSERT failure
104  *   is produced.
105  *
106  * Looks up and returns the value corresponding to the capability
107  * (KVM_CAP_*) given by cap.
108  */
109 unsigned int kvm_check_cap(long cap)
110 {
111 	int ret;
112 	int kvm_fd;
113 
114 	kvm_fd = open_kvm_dev_path_or_exit();
115 	ret = __kvm_ioctl(kvm_fd, KVM_CHECK_EXTENSION, (void *)cap);
116 	TEST_ASSERT(ret >= 0, KVM_IOCTL_ERROR(KVM_CHECK_EXTENSION, ret));
117 
118 	close(kvm_fd);
119 
120 	return (unsigned int)ret;
121 }
122 
123 void vm_enable_dirty_ring(struct kvm_vm *vm, uint32_t ring_size)
124 {
125 	if (vm_check_cap(vm, KVM_CAP_DIRTY_LOG_RING_ACQ_REL))
126 		vm_enable_cap(vm, KVM_CAP_DIRTY_LOG_RING_ACQ_REL, ring_size);
127 	else
128 		vm_enable_cap(vm, KVM_CAP_DIRTY_LOG_RING, ring_size);
129 	vm->dirty_ring_size = ring_size;
130 }
131 
132 static void vm_open(struct kvm_vm *vm)
133 {
134 	vm->kvm_fd = _open_kvm_dev_path_or_exit(O_RDWR);
135 
136 	TEST_REQUIRE(kvm_has_cap(KVM_CAP_IMMEDIATE_EXIT));
137 
138 	vm->fd = __kvm_ioctl(vm->kvm_fd, KVM_CREATE_VM, (void *)vm->type);
139 	TEST_ASSERT(vm->fd >= 0, KVM_IOCTL_ERROR(KVM_CREATE_VM, vm->fd));
140 }
141 
142 const char *vm_guest_mode_string(uint32_t i)
143 {
144 	static const char * const strings[] = {
145 		[VM_MODE_P52V48_4K]	= "PA-bits:52,  VA-bits:48,  4K pages",
146 		[VM_MODE_P52V48_64K]	= "PA-bits:52,  VA-bits:48, 64K pages",
147 		[VM_MODE_P48V48_4K]	= "PA-bits:48,  VA-bits:48,  4K pages",
148 		[VM_MODE_P48V48_16K]	= "PA-bits:48,  VA-bits:48, 16K pages",
149 		[VM_MODE_P48V48_64K]	= "PA-bits:48,  VA-bits:48, 64K pages",
150 		[VM_MODE_P40V48_4K]	= "PA-bits:40,  VA-bits:48,  4K pages",
151 		[VM_MODE_P40V48_16K]	= "PA-bits:40,  VA-bits:48, 16K pages",
152 		[VM_MODE_P40V48_64K]	= "PA-bits:40,  VA-bits:48, 64K pages",
153 		[VM_MODE_PXXV48_4K]	= "PA-bits:ANY, VA-bits:48,  4K pages",
154 		[VM_MODE_P47V64_4K]	= "PA-bits:47,  VA-bits:64,  4K pages",
155 		[VM_MODE_P44V64_4K]	= "PA-bits:44,  VA-bits:64,  4K pages",
156 		[VM_MODE_P36V48_4K]	= "PA-bits:36,  VA-bits:48,  4K pages",
157 		[VM_MODE_P36V48_16K]	= "PA-bits:36,  VA-bits:48, 16K pages",
158 		[VM_MODE_P36V48_64K]	= "PA-bits:36,  VA-bits:48, 64K pages",
159 		[VM_MODE_P36V47_16K]	= "PA-bits:36,  VA-bits:47, 16K pages",
160 	};
161 	_Static_assert(sizeof(strings)/sizeof(char *) == NUM_VM_MODES,
162 		       "Missing new mode strings?");
163 
164 	TEST_ASSERT(i < NUM_VM_MODES, "Guest mode ID %d too big", i);
165 
166 	return strings[i];
167 }
168 
169 const struct vm_guest_mode_params vm_guest_mode_params[] = {
170 	[VM_MODE_P52V48_4K]	= { 52, 48,  0x1000, 12 },
171 	[VM_MODE_P52V48_64K]	= { 52, 48, 0x10000, 16 },
172 	[VM_MODE_P48V48_4K]	= { 48, 48,  0x1000, 12 },
173 	[VM_MODE_P48V48_16K]	= { 48, 48,  0x4000, 14 },
174 	[VM_MODE_P48V48_64K]	= { 48, 48, 0x10000, 16 },
175 	[VM_MODE_P40V48_4K]	= { 40, 48,  0x1000, 12 },
176 	[VM_MODE_P40V48_16K]	= { 40, 48,  0x4000, 14 },
177 	[VM_MODE_P40V48_64K]	= { 40, 48, 0x10000, 16 },
178 	[VM_MODE_PXXV48_4K]	= {  0,  0,  0x1000, 12 },
179 	[VM_MODE_P47V64_4K]	= { 47, 64,  0x1000, 12 },
180 	[VM_MODE_P44V64_4K]	= { 44, 64,  0x1000, 12 },
181 	[VM_MODE_P36V48_4K]	= { 36, 48,  0x1000, 12 },
182 	[VM_MODE_P36V48_16K]	= { 36, 48,  0x4000, 14 },
183 	[VM_MODE_P36V48_64K]	= { 36, 48, 0x10000, 16 },
184 	[VM_MODE_P36V47_16K]	= { 36, 47,  0x4000, 14 },
185 };
186 _Static_assert(sizeof(vm_guest_mode_params)/sizeof(struct vm_guest_mode_params) == NUM_VM_MODES,
187 	       "Missing new mode params?");
188 
189 /*
190  * Initializes vm->vpages_valid to match the canonical VA space of the
191  * architecture.
192  *
193  * The default implementation is valid for architectures which split the
194  * range addressed by a single page table into a low and high region
195  * based on the MSB of the VA. On architectures with this behavior
196  * the VA region spans [0, 2^(va_bits - 1)), [-(2^(va_bits - 1), -1].
197  */
198 __weak void vm_vaddr_populate_bitmap(struct kvm_vm *vm)
199 {
200 	sparsebit_set_num(vm->vpages_valid,
201 		0, (1ULL << (vm->va_bits - 1)) >> vm->page_shift);
202 	sparsebit_set_num(vm->vpages_valid,
203 		(~((1ULL << (vm->va_bits - 1)) - 1)) >> vm->page_shift,
204 		(1ULL << (vm->va_bits - 1)) >> vm->page_shift);
205 }
206 
207 struct kvm_vm *____vm_create(enum vm_guest_mode mode)
208 {
209 	struct kvm_vm *vm;
210 
211 	vm = calloc(1, sizeof(*vm));
212 	TEST_ASSERT(vm != NULL, "Insufficient Memory");
213 
214 	INIT_LIST_HEAD(&vm->vcpus);
215 	vm->regions.gpa_tree = RB_ROOT;
216 	vm->regions.hva_tree = RB_ROOT;
217 	hash_init(vm->regions.slot_hash);
218 
219 	vm->mode = mode;
220 	vm->type = 0;
221 
222 	vm->pa_bits = vm_guest_mode_params[mode].pa_bits;
223 	vm->va_bits = vm_guest_mode_params[mode].va_bits;
224 	vm->page_size = vm_guest_mode_params[mode].page_size;
225 	vm->page_shift = vm_guest_mode_params[mode].page_shift;
226 
227 	/* Setup mode specific traits. */
228 	switch (vm->mode) {
229 	case VM_MODE_P52V48_4K:
230 		vm->pgtable_levels = 4;
231 		break;
232 	case VM_MODE_P52V48_64K:
233 		vm->pgtable_levels = 3;
234 		break;
235 	case VM_MODE_P48V48_4K:
236 		vm->pgtable_levels = 4;
237 		break;
238 	case VM_MODE_P48V48_64K:
239 		vm->pgtable_levels = 3;
240 		break;
241 	case VM_MODE_P40V48_4K:
242 	case VM_MODE_P36V48_4K:
243 		vm->pgtable_levels = 4;
244 		break;
245 	case VM_MODE_P40V48_64K:
246 	case VM_MODE_P36V48_64K:
247 		vm->pgtable_levels = 3;
248 		break;
249 	case VM_MODE_P48V48_16K:
250 	case VM_MODE_P40V48_16K:
251 	case VM_MODE_P36V48_16K:
252 		vm->pgtable_levels = 4;
253 		break;
254 	case VM_MODE_P36V47_16K:
255 		vm->pgtable_levels = 3;
256 		break;
257 	case VM_MODE_PXXV48_4K:
258 #ifdef __x86_64__
259 		kvm_get_cpu_address_width(&vm->pa_bits, &vm->va_bits);
260 		/*
261 		 * Ignore KVM support for 5-level paging (vm->va_bits == 57),
262 		 * it doesn't take effect unless a CR4.LA57 is set, which it
263 		 * isn't for this VM_MODE.
264 		 */
265 		TEST_ASSERT(vm->va_bits == 48 || vm->va_bits == 57,
266 			    "Linear address width (%d bits) not supported",
267 			    vm->va_bits);
268 		pr_debug("Guest physical address width detected: %d\n",
269 			 vm->pa_bits);
270 		vm->pgtable_levels = 4;
271 		vm->va_bits = 48;
272 #else
273 		TEST_FAIL("VM_MODE_PXXV48_4K not supported on non-x86 platforms");
274 #endif
275 		break;
276 	case VM_MODE_P47V64_4K:
277 		vm->pgtable_levels = 5;
278 		break;
279 	case VM_MODE_P44V64_4K:
280 		vm->pgtable_levels = 5;
281 		break;
282 	default:
283 		TEST_FAIL("Unknown guest mode, mode: 0x%x", mode);
284 	}
285 
286 #ifdef __aarch64__
287 	if (vm->pa_bits != 40)
288 		vm->type = KVM_VM_TYPE_ARM_IPA_SIZE(vm->pa_bits);
289 #endif
290 
291 	vm_open(vm);
292 
293 	/* Limit to VA-bit canonical virtual addresses. */
294 	vm->vpages_valid = sparsebit_alloc();
295 	vm_vaddr_populate_bitmap(vm);
296 
297 	/* Limit physical addresses to PA-bits. */
298 	vm->max_gfn = vm_compute_max_gfn(vm);
299 
300 	/* Allocate and setup memory for guest. */
301 	vm->vpages_mapped = sparsebit_alloc();
302 
303 	return vm;
304 }
305 
306 static uint64_t vm_nr_pages_required(enum vm_guest_mode mode,
307 				     uint32_t nr_runnable_vcpus,
308 				     uint64_t extra_mem_pages)
309 {
310 	uint64_t nr_pages;
311 
312 	TEST_ASSERT(nr_runnable_vcpus,
313 		    "Use vm_create_barebones() for VMs that _never_ have vCPUs\n");
314 
315 	TEST_ASSERT(nr_runnable_vcpus <= kvm_check_cap(KVM_CAP_MAX_VCPUS),
316 		    "nr_vcpus = %d too large for host, max-vcpus = %d",
317 		    nr_runnable_vcpus, kvm_check_cap(KVM_CAP_MAX_VCPUS));
318 
319 	/*
320 	 * Arbitrarily allocate 512 pages (2mb when page size is 4kb) for the
321 	 * test code and other per-VM assets that will be loaded into memslot0.
322 	 */
323 	nr_pages = 512;
324 
325 	/* Account for the per-vCPU stacks on behalf of the test. */
326 	nr_pages += nr_runnable_vcpus * DEFAULT_STACK_PGS;
327 
328 	/*
329 	 * Account for the number of pages needed for the page tables.  The
330 	 * maximum page table size for a memory region will be when the
331 	 * smallest page size is used. Considering each page contains x page
332 	 * table descriptors, the total extra size for page tables (for extra
333 	 * N pages) will be: N/x+N/x^2+N/x^3+... which is definitely smaller
334 	 * than N/x*2.
335 	 */
336 	nr_pages += (nr_pages + extra_mem_pages) / PTES_PER_MIN_PAGE * 2;
337 
338 	return vm_adjust_num_guest_pages(mode, nr_pages);
339 }
340 
341 struct kvm_vm *__vm_create(enum vm_guest_mode mode, uint32_t nr_runnable_vcpus,
342 			   uint64_t nr_extra_pages)
343 {
344 	uint64_t nr_pages = vm_nr_pages_required(mode, nr_runnable_vcpus,
345 						 nr_extra_pages);
346 	struct userspace_mem_region *slot0;
347 	struct kvm_vm *vm;
348 	int i;
349 
350 	pr_debug("%s: mode='%s' pages='%ld'\n", __func__,
351 		 vm_guest_mode_string(mode), nr_pages);
352 
353 	vm = ____vm_create(mode);
354 
355 	vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS, 0, 0, nr_pages, 0);
356 	for (i = 0; i < NR_MEM_REGIONS; i++)
357 		vm->memslots[i] = 0;
358 
359 	kvm_vm_elf_load(vm, program_invocation_name);
360 
361 	/*
362 	 * TODO: Add proper defines to protect the library's memslots, and then
363 	 * carve out memslot1 for the ucall MMIO address.  KVM treats writes to
364 	 * read-only memslots as MMIO, and creating a read-only memslot for the
365 	 * MMIO region would prevent silently clobbering the MMIO region.
366 	 */
367 	slot0 = memslot2region(vm, 0);
368 	ucall_init(vm, slot0->region.guest_phys_addr + slot0->region.memory_size);
369 
370 	kvm_arch_vm_post_create(vm);
371 
372 	return vm;
373 }
374 
375 /*
376  * VM Create with customized parameters
377  *
378  * Input Args:
379  *   mode - VM Mode (e.g. VM_MODE_P52V48_4K)
380  *   nr_vcpus - VCPU count
381  *   extra_mem_pages - Non-slot0 physical memory total size
382  *   guest_code - Guest entry point
383  *   vcpuids - VCPU IDs
384  *
385  * Output Args: None
386  *
387  * Return:
388  *   Pointer to opaque structure that describes the created VM.
389  *
390  * Creates a VM with the mode specified by mode (e.g. VM_MODE_P52V48_4K).
391  * extra_mem_pages is only used to calculate the maximum page table size,
392  * no real memory allocation for non-slot0 memory in this function.
393  */
394 struct kvm_vm *__vm_create_with_vcpus(enum vm_guest_mode mode, uint32_t nr_vcpus,
395 				      uint64_t extra_mem_pages,
396 				      void *guest_code, struct kvm_vcpu *vcpus[])
397 {
398 	struct kvm_vm *vm;
399 	int i;
400 
401 	TEST_ASSERT(!nr_vcpus || vcpus, "Must provide vCPU array");
402 
403 	vm = __vm_create(mode, nr_vcpus, extra_mem_pages);
404 
405 	for (i = 0; i < nr_vcpus; ++i)
406 		vcpus[i] = vm_vcpu_add(vm, i, guest_code);
407 
408 	return vm;
409 }
410 
411 struct kvm_vm *__vm_create_with_one_vcpu(struct kvm_vcpu **vcpu,
412 					 uint64_t extra_mem_pages,
413 					 void *guest_code)
414 {
415 	struct kvm_vcpu *vcpus[1];
416 	struct kvm_vm *vm;
417 
418 	vm = __vm_create_with_vcpus(VM_MODE_DEFAULT, 1, extra_mem_pages,
419 				    guest_code, vcpus);
420 
421 	*vcpu = vcpus[0];
422 	return vm;
423 }
424 
425 /*
426  * VM Restart
427  *
428  * Input Args:
429  *   vm - VM that has been released before
430  *
431  * Output Args: None
432  *
433  * Reopens the file descriptors associated to the VM and reinstates the
434  * global state, such as the irqchip and the memory regions that are mapped
435  * into the guest.
436  */
437 void kvm_vm_restart(struct kvm_vm *vmp)
438 {
439 	int ctr;
440 	struct userspace_mem_region *region;
441 
442 	vm_open(vmp);
443 	if (vmp->has_irqchip)
444 		vm_create_irqchip(vmp);
445 
446 	hash_for_each(vmp->regions.slot_hash, ctr, region, slot_node) {
447 		int ret = ioctl(vmp->fd, KVM_SET_USER_MEMORY_REGION, &region->region);
448 		TEST_ASSERT(ret == 0, "KVM_SET_USER_MEMORY_REGION IOCTL failed,\n"
449 			    "  rc: %i errno: %i\n"
450 			    "  slot: %u flags: 0x%x\n"
451 			    "  guest_phys_addr: 0x%llx size: 0x%llx",
452 			    ret, errno, region->region.slot,
453 			    region->region.flags,
454 			    region->region.guest_phys_addr,
455 			    region->region.memory_size);
456 	}
457 }
458 
459 __weak struct kvm_vcpu *vm_arch_vcpu_recreate(struct kvm_vm *vm,
460 					      uint32_t vcpu_id)
461 {
462 	return __vm_vcpu_add(vm, vcpu_id);
463 }
464 
465 struct kvm_vcpu *vm_recreate_with_one_vcpu(struct kvm_vm *vm)
466 {
467 	kvm_vm_restart(vm);
468 
469 	return vm_vcpu_recreate(vm, 0);
470 }
471 
472 void kvm_pin_this_task_to_pcpu(uint32_t pcpu)
473 {
474 	cpu_set_t mask;
475 	int r;
476 
477 	CPU_ZERO(&mask);
478 	CPU_SET(pcpu, &mask);
479 	r = sched_setaffinity(0, sizeof(mask), &mask);
480 	TEST_ASSERT(!r, "sched_setaffinity() failed for pCPU '%u'.\n", pcpu);
481 }
482 
483 static uint32_t parse_pcpu(const char *cpu_str, const cpu_set_t *allowed_mask)
484 {
485 	uint32_t pcpu = atoi_non_negative("CPU number", cpu_str);
486 
487 	TEST_ASSERT(CPU_ISSET(pcpu, allowed_mask),
488 		    "Not allowed to run on pCPU '%d', check cgroups?\n", pcpu);
489 	return pcpu;
490 }
491 
492 void kvm_parse_vcpu_pinning(const char *pcpus_string, uint32_t vcpu_to_pcpu[],
493 			    int nr_vcpus)
494 {
495 	cpu_set_t allowed_mask;
496 	char *cpu, *cpu_list;
497 	char delim[2] = ",";
498 	int i, r;
499 
500 	cpu_list = strdup(pcpus_string);
501 	TEST_ASSERT(cpu_list, "strdup() allocation failed.\n");
502 
503 	r = sched_getaffinity(0, sizeof(allowed_mask), &allowed_mask);
504 	TEST_ASSERT(!r, "sched_getaffinity() failed");
505 
506 	cpu = strtok(cpu_list, delim);
507 
508 	/* 1. Get all pcpus for vcpus. */
509 	for (i = 0; i < nr_vcpus; i++) {
510 		TEST_ASSERT(cpu, "pCPU not provided for vCPU '%d'\n", i);
511 		vcpu_to_pcpu[i] = parse_pcpu(cpu, &allowed_mask);
512 		cpu = strtok(NULL, delim);
513 	}
514 
515 	/* 2. Check if the main worker needs to be pinned. */
516 	if (cpu) {
517 		kvm_pin_this_task_to_pcpu(parse_pcpu(cpu, &allowed_mask));
518 		cpu = strtok(NULL, delim);
519 	}
520 
521 	TEST_ASSERT(!cpu, "pCPU list contains trailing garbage characters '%s'", cpu);
522 	free(cpu_list);
523 }
524 
525 /*
526  * Userspace Memory Region Find
527  *
528  * Input Args:
529  *   vm - Virtual Machine
530  *   start - Starting VM physical address
531  *   end - Ending VM physical address, inclusive.
532  *
533  * Output Args: None
534  *
535  * Return:
536  *   Pointer to overlapping region, NULL if no such region.
537  *
538  * Searches for a region with any physical memory that overlaps with
539  * any portion of the guest physical addresses from start to end
540  * inclusive.  If multiple overlapping regions exist, a pointer to any
541  * of the regions is returned.  Null is returned only when no overlapping
542  * region exists.
543  */
544 static struct userspace_mem_region *
545 userspace_mem_region_find(struct kvm_vm *vm, uint64_t start, uint64_t end)
546 {
547 	struct rb_node *node;
548 
549 	for (node = vm->regions.gpa_tree.rb_node; node; ) {
550 		struct userspace_mem_region *region =
551 			container_of(node, struct userspace_mem_region, gpa_node);
552 		uint64_t existing_start = region->region.guest_phys_addr;
553 		uint64_t existing_end = region->region.guest_phys_addr
554 			+ region->region.memory_size - 1;
555 		if (start <= existing_end && end >= existing_start)
556 			return region;
557 
558 		if (start < existing_start)
559 			node = node->rb_left;
560 		else
561 			node = node->rb_right;
562 	}
563 
564 	return NULL;
565 }
566 
567 /*
568  * KVM Userspace Memory Region Find
569  *
570  * Input Args:
571  *   vm - Virtual Machine
572  *   start - Starting VM physical address
573  *   end - Ending VM physical address, inclusive.
574  *
575  * Output Args: None
576  *
577  * Return:
578  *   Pointer to overlapping region, NULL if no such region.
579  *
580  * Public interface to userspace_mem_region_find. Allows tests to look up
581  * the memslot datastructure for a given range of guest physical memory.
582  */
583 struct kvm_userspace_memory_region *
584 kvm_userspace_memory_region_find(struct kvm_vm *vm, uint64_t start,
585 				 uint64_t end)
586 {
587 	struct userspace_mem_region *region;
588 
589 	region = userspace_mem_region_find(vm, start, end);
590 	if (!region)
591 		return NULL;
592 
593 	return &region->region;
594 }
595 
596 __weak void vcpu_arch_free(struct kvm_vcpu *vcpu)
597 {
598 
599 }
600 
601 /*
602  * VM VCPU Remove
603  *
604  * Input Args:
605  *   vcpu - VCPU to remove
606  *
607  * Output Args: None
608  *
609  * Return: None, TEST_ASSERT failures for all error conditions
610  *
611  * Removes a vCPU from a VM and frees its resources.
612  */
613 static void vm_vcpu_rm(struct kvm_vm *vm, struct kvm_vcpu *vcpu)
614 {
615 	int ret;
616 
617 	if (vcpu->dirty_gfns) {
618 		ret = munmap(vcpu->dirty_gfns, vm->dirty_ring_size);
619 		TEST_ASSERT(!ret, __KVM_SYSCALL_ERROR("munmap()", ret));
620 		vcpu->dirty_gfns = NULL;
621 	}
622 
623 	ret = munmap(vcpu->run, vcpu_mmap_sz());
624 	TEST_ASSERT(!ret, __KVM_SYSCALL_ERROR("munmap()", ret));
625 
626 	ret = close(vcpu->fd);
627 	TEST_ASSERT(!ret,  __KVM_SYSCALL_ERROR("close()", ret));
628 
629 	list_del(&vcpu->list);
630 
631 	vcpu_arch_free(vcpu);
632 	free(vcpu);
633 }
634 
635 void kvm_vm_release(struct kvm_vm *vmp)
636 {
637 	struct kvm_vcpu *vcpu, *tmp;
638 	int ret;
639 
640 	list_for_each_entry_safe(vcpu, tmp, &vmp->vcpus, list)
641 		vm_vcpu_rm(vmp, vcpu);
642 
643 	ret = close(vmp->fd);
644 	TEST_ASSERT(!ret,  __KVM_SYSCALL_ERROR("close()", ret));
645 
646 	ret = close(vmp->kvm_fd);
647 	TEST_ASSERT(!ret,  __KVM_SYSCALL_ERROR("close()", ret));
648 }
649 
650 static void __vm_mem_region_delete(struct kvm_vm *vm,
651 				   struct userspace_mem_region *region,
652 				   bool unlink)
653 {
654 	int ret;
655 
656 	if (unlink) {
657 		rb_erase(&region->gpa_node, &vm->regions.gpa_tree);
658 		rb_erase(&region->hva_node, &vm->regions.hva_tree);
659 		hash_del(&region->slot_node);
660 	}
661 
662 	region->region.memory_size = 0;
663 	vm_ioctl(vm, KVM_SET_USER_MEMORY_REGION, &region->region);
664 
665 	sparsebit_free(&region->unused_phy_pages);
666 	ret = munmap(region->mmap_start, region->mmap_size);
667 	TEST_ASSERT(!ret, __KVM_SYSCALL_ERROR("munmap()", ret));
668 	if (region->fd >= 0) {
669 		/* There's an extra map when using shared memory. */
670 		ret = munmap(region->mmap_alias, region->mmap_size);
671 		TEST_ASSERT(!ret, __KVM_SYSCALL_ERROR("munmap()", ret));
672 		close(region->fd);
673 	}
674 
675 	free(region);
676 }
677 
678 /*
679  * Destroys and frees the VM pointed to by vmp.
680  */
681 void kvm_vm_free(struct kvm_vm *vmp)
682 {
683 	int ctr;
684 	struct hlist_node *node;
685 	struct userspace_mem_region *region;
686 
687 	if (vmp == NULL)
688 		return;
689 
690 	/* Free cached stats metadata and close FD */
691 	if (vmp->stats_fd) {
692 		free(vmp->stats_desc);
693 		close(vmp->stats_fd);
694 	}
695 
696 	/* Free userspace_mem_regions. */
697 	hash_for_each_safe(vmp->regions.slot_hash, ctr, node, region, slot_node)
698 		__vm_mem_region_delete(vmp, region, false);
699 
700 	/* Free sparsebit arrays. */
701 	sparsebit_free(&vmp->vpages_valid);
702 	sparsebit_free(&vmp->vpages_mapped);
703 
704 	kvm_vm_release(vmp);
705 
706 	/* Free the structure describing the VM. */
707 	free(vmp);
708 }
709 
710 int kvm_memfd_alloc(size_t size, bool hugepages)
711 {
712 	int memfd_flags = MFD_CLOEXEC;
713 	int fd, r;
714 
715 	if (hugepages)
716 		memfd_flags |= MFD_HUGETLB;
717 
718 	fd = memfd_create("kvm_selftest", memfd_flags);
719 	TEST_ASSERT(fd != -1, __KVM_SYSCALL_ERROR("memfd_create()", fd));
720 
721 	r = ftruncate(fd, size);
722 	TEST_ASSERT(!r, __KVM_SYSCALL_ERROR("ftruncate()", r));
723 
724 	r = fallocate(fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 0, size);
725 	TEST_ASSERT(!r, __KVM_SYSCALL_ERROR("fallocate()", r));
726 
727 	return fd;
728 }
729 
730 /*
731  * Memory Compare, host virtual to guest virtual
732  *
733  * Input Args:
734  *   hva - Starting host virtual address
735  *   vm - Virtual Machine
736  *   gva - Starting guest virtual address
737  *   len - number of bytes to compare
738  *
739  * Output Args: None
740  *
741  * Input/Output Args: None
742  *
743  * Return:
744  *   Returns 0 if the bytes starting at hva for a length of len
745  *   are equal the guest virtual bytes starting at gva.  Returns
746  *   a value < 0, if bytes at hva are less than those at gva.
747  *   Otherwise a value > 0 is returned.
748  *
749  * Compares the bytes starting at the host virtual address hva, for
750  * a length of len, to the guest bytes starting at the guest virtual
751  * address given by gva.
752  */
753 int kvm_memcmp_hva_gva(void *hva, struct kvm_vm *vm, vm_vaddr_t gva, size_t len)
754 {
755 	size_t amt;
756 
757 	/*
758 	 * Compare a batch of bytes until either a match is found
759 	 * or all the bytes have been compared.
760 	 */
761 	for (uintptr_t offset = 0; offset < len; offset += amt) {
762 		uintptr_t ptr1 = (uintptr_t)hva + offset;
763 
764 		/*
765 		 * Determine host address for guest virtual address
766 		 * at offset.
767 		 */
768 		uintptr_t ptr2 = (uintptr_t)addr_gva2hva(vm, gva + offset);
769 
770 		/*
771 		 * Determine amount to compare on this pass.
772 		 * Don't allow the comparsion to cross a page boundary.
773 		 */
774 		amt = len - offset;
775 		if ((ptr1 >> vm->page_shift) != ((ptr1 + amt) >> vm->page_shift))
776 			amt = vm->page_size - (ptr1 % vm->page_size);
777 		if ((ptr2 >> vm->page_shift) != ((ptr2 + amt) >> vm->page_shift))
778 			amt = vm->page_size - (ptr2 % vm->page_size);
779 
780 		assert((ptr1 >> vm->page_shift) == ((ptr1 + amt - 1) >> vm->page_shift));
781 		assert((ptr2 >> vm->page_shift) == ((ptr2 + amt - 1) >> vm->page_shift));
782 
783 		/*
784 		 * Perform the comparison.  If there is a difference
785 		 * return that result to the caller, otherwise need
786 		 * to continue on looking for a mismatch.
787 		 */
788 		int ret = memcmp((void *)ptr1, (void *)ptr2, amt);
789 		if (ret != 0)
790 			return ret;
791 	}
792 
793 	/*
794 	 * No mismatch found.  Let the caller know the two memory
795 	 * areas are equal.
796 	 */
797 	return 0;
798 }
799 
800 static void vm_userspace_mem_region_gpa_insert(struct rb_root *gpa_tree,
801 					       struct userspace_mem_region *region)
802 {
803 	struct rb_node **cur, *parent;
804 
805 	for (cur = &gpa_tree->rb_node, parent = NULL; *cur; ) {
806 		struct userspace_mem_region *cregion;
807 
808 		cregion = container_of(*cur, typeof(*cregion), gpa_node);
809 		parent = *cur;
810 		if (region->region.guest_phys_addr <
811 		    cregion->region.guest_phys_addr)
812 			cur = &(*cur)->rb_left;
813 		else {
814 			TEST_ASSERT(region->region.guest_phys_addr !=
815 				    cregion->region.guest_phys_addr,
816 				    "Duplicate GPA in region tree");
817 
818 			cur = &(*cur)->rb_right;
819 		}
820 	}
821 
822 	rb_link_node(&region->gpa_node, parent, cur);
823 	rb_insert_color(&region->gpa_node, gpa_tree);
824 }
825 
826 static void vm_userspace_mem_region_hva_insert(struct rb_root *hva_tree,
827 					       struct userspace_mem_region *region)
828 {
829 	struct rb_node **cur, *parent;
830 
831 	for (cur = &hva_tree->rb_node, parent = NULL; *cur; ) {
832 		struct userspace_mem_region *cregion;
833 
834 		cregion = container_of(*cur, typeof(*cregion), hva_node);
835 		parent = *cur;
836 		if (region->host_mem < cregion->host_mem)
837 			cur = &(*cur)->rb_left;
838 		else {
839 			TEST_ASSERT(region->host_mem !=
840 				    cregion->host_mem,
841 				    "Duplicate HVA in region tree");
842 
843 			cur = &(*cur)->rb_right;
844 		}
845 	}
846 
847 	rb_link_node(&region->hva_node, parent, cur);
848 	rb_insert_color(&region->hva_node, hva_tree);
849 }
850 
851 
852 int __vm_set_user_memory_region(struct kvm_vm *vm, uint32_t slot, uint32_t flags,
853 				uint64_t gpa, uint64_t size, void *hva)
854 {
855 	struct kvm_userspace_memory_region region = {
856 		.slot = slot,
857 		.flags = flags,
858 		.guest_phys_addr = gpa,
859 		.memory_size = size,
860 		.userspace_addr = (uintptr_t)hva,
861 	};
862 
863 	return ioctl(vm->fd, KVM_SET_USER_MEMORY_REGION, &region);
864 }
865 
866 void vm_set_user_memory_region(struct kvm_vm *vm, uint32_t slot, uint32_t flags,
867 			       uint64_t gpa, uint64_t size, void *hva)
868 {
869 	int ret = __vm_set_user_memory_region(vm, slot, flags, gpa, size, hva);
870 
871 	TEST_ASSERT(!ret, "KVM_SET_USER_MEMORY_REGION failed, errno = %d (%s)",
872 		    errno, strerror(errno));
873 }
874 
875 /*
876  * VM Userspace Memory Region Add
877  *
878  * Input Args:
879  *   vm - Virtual Machine
880  *   src_type - Storage source for this region.
881  *              NULL to use anonymous memory.
882  *   guest_paddr - Starting guest physical address
883  *   slot - KVM region slot
884  *   npages - Number of physical pages
885  *   flags - KVM memory region flags (e.g. KVM_MEM_LOG_DIRTY_PAGES)
886  *
887  * Output Args: None
888  *
889  * Return: None
890  *
891  * Allocates a memory area of the number of pages specified by npages
892  * and maps it to the VM specified by vm, at a starting physical address
893  * given by guest_paddr.  The region is created with a KVM region slot
894  * given by slot, which must be unique and < KVM_MEM_SLOTS_NUM.  The
895  * region is created with the flags given by flags.
896  */
897 void vm_userspace_mem_region_add(struct kvm_vm *vm,
898 	enum vm_mem_backing_src_type src_type,
899 	uint64_t guest_paddr, uint32_t slot, uint64_t npages,
900 	uint32_t flags)
901 {
902 	int ret;
903 	struct userspace_mem_region *region;
904 	size_t backing_src_pagesz = get_backing_src_pagesz(src_type);
905 	size_t alignment;
906 
907 	TEST_ASSERT(vm_adjust_num_guest_pages(vm->mode, npages) == npages,
908 		"Number of guest pages is not compatible with the host. "
909 		"Try npages=%d", vm_adjust_num_guest_pages(vm->mode, npages));
910 
911 	TEST_ASSERT((guest_paddr % vm->page_size) == 0, "Guest physical "
912 		"address not on a page boundary.\n"
913 		"  guest_paddr: 0x%lx vm->page_size: 0x%x",
914 		guest_paddr, vm->page_size);
915 	TEST_ASSERT((((guest_paddr >> vm->page_shift) + npages) - 1)
916 		<= vm->max_gfn, "Physical range beyond maximum "
917 		"supported physical address,\n"
918 		"  guest_paddr: 0x%lx npages: 0x%lx\n"
919 		"  vm->max_gfn: 0x%lx vm->page_size: 0x%x",
920 		guest_paddr, npages, vm->max_gfn, vm->page_size);
921 
922 	/*
923 	 * Confirm a mem region with an overlapping address doesn't
924 	 * already exist.
925 	 */
926 	region = (struct userspace_mem_region *) userspace_mem_region_find(
927 		vm, guest_paddr, (guest_paddr + npages * vm->page_size) - 1);
928 	if (region != NULL)
929 		TEST_FAIL("overlapping userspace_mem_region already "
930 			"exists\n"
931 			"  requested guest_paddr: 0x%lx npages: 0x%lx "
932 			"page_size: 0x%x\n"
933 			"  existing guest_paddr: 0x%lx size: 0x%lx",
934 			guest_paddr, npages, vm->page_size,
935 			(uint64_t) region->region.guest_phys_addr,
936 			(uint64_t) region->region.memory_size);
937 
938 	/* Confirm no region with the requested slot already exists. */
939 	hash_for_each_possible(vm->regions.slot_hash, region, slot_node,
940 			       slot) {
941 		if (region->region.slot != slot)
942 			continue;
943 
944 		TEST_FAIL("A mem region with the requested slot "
945 			"already exists.\n"
946 			"  requested slot: %u paddr: 0x%lx npages: 0x%lx\n"
947 			"  existing slot: %u paddr: 0x%lx size: 0x%lx",
948 			slot, guest_paddr, npages,
949 			region->region.slot,
950 			(uint64_t) region->region.guest_phys_addr,
951 			(uint64_t) region->region.memory_size);
952 	}
953 
954 	/* Allocate and initialize new mem region structure. */
955 	region = calloc(1, sizeof(*region));
956 	TEST_ASSERT(region != NULL, "Insufficient Memory");
957 	region->mmap_size = npages * vm->page_size;
958 
959 #ifdef __s390x__
960 	/* On s390x, the host address must be aligned to 1M (due to PGSTEs) */
961 	alignment = 0x100000;
962 #else
963 	alignment = 1;
964 #endif
965 
966 	/*
967 	 * When using THP mmap is not guaranteed to returned a hugepage aligned
968 	 * address so we have to pad the mmap. Padding is not needed for HugeTLB
969 	 * because mmap will always return an address aligned to the HugeTLB
970 	 * page size.
971 	 */
972 	if (src_type == VM_MEM_SRC_ANONYMOUS_THP)
973 		alignment = max(backing_src_pagesz, alignment);
974 
975 	ASSERT_EQ(guest_paddr, align_up(guest_paddr, backing_src_pagesz));
976 
977 	/* Add enough memory to align up if necessary */
978 	if (alignment > 1)
979 		region->mmap_size += alignment;
980 
981 	region->fd = -1;
982 	if (backing_src_is_shared(src_type))
983 		region->fd = kvm_memfd_alloc(region->mmap_size,
984 					     src_type == VM_MEM_SRC_SHARED_HUGETLB);
985 
986 	region->mmap_start = mmap(NULL, region->mmap_size,
987 				  PROT_READ | PROT_WRITE,
988 				  vm_mem_backing_src_alias(src_type)->flag,
989 				  region->fd, 0);
990 	TEST_ASSERT(region->mmap_start != MAP_FAILED,
991 		    __KVM_SYSCALL_ERROR("mmap()", (int)(unsigned long)MAP_FAILED));
992 
993 	TEST_ASSERT(!is_backing_src_hugetlb(src_type) ||
994 		    region->mmap_start == align_ptr_up(region->mmap_start, backing_src_pagesz),
995 		    "mmap_start %p is not aligned to HugeTLB page size 0x%lx",
996 		    region->mmap_start, backing_src_pagesz);
997 
998 	/* Align host address */
999 	region->host_mem = align_ptr_up(region->mmap_start, alignment);
1000 
1001 	/* As needed perform madvise */
1002 	if ((src_type == VM_MEM_SRC_ANONYMOUS ||
1003 	     src_type == VM_MEM_SRC_ANONYMOUS_THP) && thp_configured()) {
1004 		ret = madvise(region->host_mem, npages * vm->page_size,
1005 			      src_type == VM_MEM_SRC_ANONYMOUS ? MADV_NOHUGEPAGE : MADV_HUGEPAGE);
1006 		TEST_ASSERT(ret == 0, "madvise failed, addr: %p length: 0x%lx src_type: %s",
1007 			    region->host_mem, npages * vm->page_size,
1008 			    vm_mem_backing_src_alias(src_type)->name);
1009 	}
1010 
1011 	region->backing_src_type = src_type;
1012 	region->unused_phy_pages = sparsebit_alloc();
1013 	sparsebit_set_num(region->unused_phy_pages,
1014 		guest_paddr >> vm->page_shift, npages);
1015 	region->region.slot = slot;
1016 	region->region.flags = flags;
1017 	region->region.guest_phys_addr = guest_paddr;
1018 	region->region.memory_size = npages * vm->page_size;
1019 	region->region.userspace_addr = (uintptr_t) region->host_mem;
1020 	ret = __vm_ioctl(vm, KVM_SET_USER_MEMORY_REGION, &region->region);
1021 	TEST_ASSERT(ret == 0, "KVM_SET_USER_MEMORY_REGION IOCTL failed,\n"
1022 		"  rc: %i errno: %i\n"
1023 		"  slot: %u flags: 0x%x\n"
1024 		"  guest_phys_addr: 0x%lx size: 0x%lx",
1025 		ret, errno, slot, flags,
1026 		guest_paddr, (uint64_t) region->region.memory_size);
1027 
1028 	/* Add to quick lookup data structures */
1029 	vm_userspace_mem_region_gpa_insert(&vm->regions.gpa_tree, region);
1030 	vm_userspace_mem_region_hva_insert(&vm->regions.hva_tree, region);
1031 	hash_add(vm->regions.slot_hash, &region->slot_node, slot);
1032 
1033 	/* If shared memory, create an alias. */
1034 	if (region->fd >= 0) {
1035 		region->mmap_alias = mmap(NULL, region->mmap_size,
1036 					  PROT_READ | PROT_WRITE,
1037 					  vm_mem_backing_src_alias(src_type)->flag,
1038 					  region->fd, 0);
1039 		TEST_ASSERT(region->mmap_alias != MAP_FAILED,
1040 			    __KVM_SYSCALL_ERROR("mmap()",  (int)(unsigned long)MAP_FAILED));
1041 
1042 		/* Align host alias address */
1043 		region->host_alias = align_ptr_up(region->mmap_alias, alignment);
1044 	}
1045 }
1046 
1047 /*
1048  * Memslot to region
1049  *
1050  * Input Args:
1051  *   vm - Virtual Machine
1052  *   memslot - KVM memory slot ID
1053  *
1054  * Output Args: None
1055  *
1056  * Return:
1057  *   Pointer to memory region structure that describe memory region
1058  *   using kvm memory slot ID given by memslot.  TEST_ASSERT failure
1059  *   on error (e.g. currently no memory region using memslot as a KVM
1060  *   memory slot ID).
1061  */
1062 struct userspace_mem_region *
1063 memslot2region(struct kvm_vm *vm, uint32_t memslot)
1064 {
1065 	struct userspace_mem_region *region;
1066 
1067 	hash_for_each_possible(vm->regions.slot_hash, region, slot_node,
1068 			       memslot)
1069 		if (region->region.slot == memslot)
1070 			return region;
1071 
1072 	fprintf(stderr, "No mem region with the requested slot found,\n"
1073 		"  requested slot: %u\n", memslot);
1074 	fputs("---- vm dump ----\n", stderr);
1075 	vm_dump(stderr, vm, 2);
1076 	TEST_FAIL("Mem region not found");
1077 	return NULL;
1078 }
1079 
1080 /*
1081  * VM Memory Region Flags Set
1082  *
1083  * Input Args:
1084  *   vm - Virtual Machine
1085  *   flags - Starting guest physical address
1086  *
1087  * Output Args: None
1088  *
1089  * Return: None
1090  *
1091  * Sets the flags of the memory region specified by the value of slot,
1092  * to the values given by flags.
1093  */
1094 void vm_mem_region_set_flags(struct kvm_vm *vm, uint32_t slot, uint32_t flags)
1095 {
1096 	int ret;
1097 	struct userspace_mem_region *region;
1098 
1099 	region = memslot2region(vm, slot);
1100 
1101 	region->region.flags = flags;
1102 
1103 	ret = __vm_ioctl(vm, KVM_SET_USER_MEMORY_REGION, &region->region);
1104 
1105 	TEST_ASSERT(ret == 0, "KVM_SET_USER_MEMORY_REGION IOCTL failed,\n"
1106 		"  rc: %i errno: %i slot: %u flags: 0x%x",
1107 		ret, errno, slot, flags);
1108 }
1109 
1110 /*
1111  * VM Memory Region Move
1112  *
1113  * Input Args:
1114  *   vm - Virtual Machine
1115  *   slot - Slot of the memory region to move
1116  *   new_gpa - Starting guest physical address
1117  *
1118  * Output Args: None
1119  *
1120  * Return: None
1121  *
1122  * Change the gpa of a memory region.
1123  */
1124 void vm_mem_region_move(struct kvm_vm *vm, uint32_t slot, uint64_t new_gpa)
1125 {
1126 	struct userspace_mem_region *region;
1127 	int ret;
1128 
1129 	region = memslot2region(vm, slot);
1130 
1131 	region->region.guest_phys_addr = new_gpa;
1132 
1133 	ret = __vm_ioctl(vm, KVM_SET_USER_MEMORY_REGION, &region->region);
1134 
1135 	TEST_ASSERT(!ret, "KVM_SET_USER_MEMORY_REGION failed\n"
1136 		    "ret: %i errno: %i slot: %u new_gpa: 0x%lx",
1137 		    ret, errno, slot, new_gpa);
1138 }
1139 
1140 /*
1141  * VM Memory Region Delete
1142  *
1143  * Input Args:
1144  *   vm - Virtual Machine
1145  *   slot - Slot of the memory region to delete
1146  *
1147  * Output Args: None
1148  *
1149  * Return: None
1150  *
1151  * Delete a memory region.
1152  */
1153 void vm_mem_region_delete(struct kvm_vm *vm, uint32_t slot)
1154 {
1155 	__vm_mem_region_delete(vm, memslot2region(vm, slot), true);
1156 }
1157 
1158 /* Returns the size of a vCPU's kvm_run structure. */
1159 static int vcpu_mmap_sz(void)
1160 {
1161 	int dev_fd, ret;
1162 
1163 	dev_fd = open_kvm_dev_path_or_exit();
1164 
1165 	ret = ioctl(dev_fd, KVM_GET_VCPU_MMAP_SIZE, NULL);
1166 	TEST_ASSERT(ret >= sizeof(struct kvm_run),
1167 		    KVM_IOCTL_ERROR(KVM_GET_VCPU_MMAP_SIZE, ret));
1168 
1169 	close(dev_fd);
1170 
1171 	return ret;
1172 }
1173 
1174 static bool vcpu_exists(struct kvm_vm *vm, uint32_t vcpu_id)
1175 {
1176 	struct kvm_vcpu *vcpu;
1177 
1178 	list_for_each_entry(vcpu, &vm->vcpus, list) {
1179 		if (vcpu->id == vcpu_id)
1180 			return true;
1181 	}
1182 
1183 	return false;
1184 }
1185 
1186 /*
1187  * Adds a virtual CPU to the VM specified by vm with the ID given by vcpu_id.
1188  * No additional vCPU setup is done.  Returns the vCPU.
1189  */
1190 struct kvm_vcpu *__vm_vcpu_add(struct kvm_vm *vm, uint32_t vcpu_id)
1191 {
1192 	struct kvm_vcpu *vcpu;
1193 
1194 	/* Confirm a vcpu with the specified id doesn't already exist. */
1195 	TEST_ASSERT(!vcpu_exists(vm, vcpu_id), "vCPU%d already exists\n", vcpu_id);
1196 
1197 	/* Allocate and initialize new vcpu structure. */
1198 	vcpu = calloc(1, sizeof(*vcpu));
1199 	TEST_ASSERT(vcpu != NULL, "Insufficient Memory");
1200 
1201 	vcpu->vm = vm;
1202 	vcpu->id = vcpu_id;
1203 	vcpu->fd = __vm_ioctl(vm, KVM_CREATE_VCPU, (void *)(unsigned long)vcpu_id);
1204 	TEST_ASSERT(vcpu->fd >= 0, KVM_IOCTL_ERROR(KVM_CREATE_VCPU, vcpu->fd));
1205 
1206 	TEST_ASSERT(vcpu_mmap_sz() >= sizeof(*vcpu->run), "vcpu mmap size "
1207 		"smaller than expected, vcpu_mmap_sz: %i expected_min: %zi",
1208 		vcpu_mmap_sz(), sizeof(*vcpu->run));
1209 	vcpu->run = (struct kvm_run *) mmap(NULL, vcpu_mmap_sz(),
1210 		PROT_READ | PROT_WRITE, MAP_SHARED, vcpu->fd, 0);
1211 	TEST_ASSERT(vcpu->run != MAP_FAILED,
1212 		    __KVM_SYSCALL_ERROR("mmap()", (int)(unsigned long)MAP_FAILED));
1213 
1214 	/* Add to linked-list of VCPUs. */
1215 	list_add(&vcpu->list, &vm->vcpus);
1216 
1217 	return vcpu;
1218 }
1219 
1220 /*
1221  * VM Virtual Address Unused Gap
1222  *
1223  * Input Args:
1224  *   vm - Virtual Machine
1225  *   sz - Size (bytes)
1226  *   vaddr_min - Minimum Virtual Address
1227  *
1228  * Output Args: None
1229  *
1230  * Return:
1231  *   Lowest virtual address at or below vaddr_min, with at least
1232  *   sz unused bytes.  TEST_ASSERT failure if no area of at least
1233  *   size sz is available.
1234  *
1235  * Within the VM specified by vm, locates the lowest starting virtual
1236  * address >= vaddr_min, that has at least sz unallocated bytes.  A
1237  * TEST_ASSERT failure occurs for invalid input or no area of at least
1238  * sz unallocated bytes >= vaddr_min is available.
1239  */
1240 vm_vaddr_t vm_vaddr_unused_gap(struct kvm_vm *vm, size_t sz,
1241 			       vm_vaddr_t vaddr_min)
1242 {
1243 	uint64_t pages = (sz + vm->page_size - 1) >> vm->page_shift;
1244 
1245 	/* Determine lowest permitted virtual page index. */
1246 	uint64_t pgidx_start = (vaddr_min + vm->page_size - 1) >> vm->page_shift;
1247 	if ((pgidx_start * vm->page_size) < vaddr_min)
1248 		goto no_va_found;
1249 
1250 	/* Loop over section with enough valid virtual page indexes. */
1251 	if (!sparsebit_is_set_num(vm->vpages_valid,
1252 		pgidx_start, pages))
1253 		pgidx_start = sparsebit_next_set_num(vm->vpages_valid,
1254 			pgidx_start, pages);
1255 	do {
1256 		/*
1257 		 * Are there enough unused virtual pages available at
1258 		 * the currently proposed starting virtual page index.
1259 		 * If not, adjust proposed starting index to next
1260 		 * possible.
1261 		 */
1262 		if (sparsebit_is_clear_num(vm->vpages_mapped,
1263 			pgidx_start, pages))
1264 			goto va_found;
1265 		pgidx_start = sparsebit_next_clear_num(vm->vpages_mapped,
1266 			pgidx_start, pages);
1267 		if (pgidx_start == 0)
1268 			goto no_va_found;
1269 
1270 		/*
1271 		 * If needed, adjust proposed starting virtual address,
1272 		 * to next range of valid virtual addresses.
1273 		 */
1274 		if (!sparsebit_is_set_num(vm->vpages_valid,
1275 			pgidx_start, pages)) {
1276 			pgidx_start = sparsebit_next_set_num(
1277 				vm->vpages_valid, pgidx_start, pages);
1278 			if (pgidx_start == 0)
1279 				goto no_va_found;
1280 		}
1281 	} while (pgidx_start != 0);
1282 
1283 no_va_found:
1284 	TEST_FAIL("No vaddr of specified pages available, pages: 0x%lx", pages);
1285 
1286 	/* NOT REACHED */
1287 	return -1;
1288 
1289 va_found:
1290 	TEST_ASSERT(sparsebit_is_set_num(vm->vpages_valid,
1291 		pgidx_start, pages),
1292 		"Unexpected, invalid virtual page index range,\n"
1293 		"  pgidx_start: 0x%lx\n"
1294 		"  pages: 0x%lx",
1295 		pgidx_start, pages);
1296 	TEST_ASSERT(sparsebit_is_clear_num(vm->vpages_mapped,
1297 		pgidx_start, pages),
1298 		"Unexpected, pages already mapped,\n"
1299 		"  pgidx_start: 0x%lx\n"
1300 		"  pages: 0x%lx",
1301 		pgidx_start, pages);
1302 
1303 	return pgidx_start * vm->page_size;
1304 }
1305 
1306 vm_vaddr_t __vm_vaddr_alloc(struct kvm_vm *vm, size_t sz, vm_vaddr_t vaddr_min,
1307 			    enum kvm_mem_region_type type)
1308 {
1309 	uint64_t pages = (sz >> vm->page_shift) + ((sz % vm->page_size) != 0);
1310 
1311 	virt_pgd_alloc(vm);
1312 	vm_paddr_t paddr = vm_phy_pages_alloc(vm, pages,
1313 					      KVM_UTIL_MIN_PFN * vm->page_size,
1314 					      vm->memslots[type]);
1315 
1316 	/*
1317 	 * Find an unused range of virtual page addresses of at least
1318 	 * pages in length.
1319 	 */
1320 	vm_vaddr_t vaddr_start = vm_vaddr_unused_gap(vm, sz, vaddr_min);
1321 
1322 	/* Map the virtual pages. */
1323 	for (vm_vaddr_t vaddr = vaddr_start; pages > 0;
1324 		pages--, vaddr += vm->page_size, paddr += vm->page_size) {
1325 
1326 		virt_pg_map(vm, vaddr, paddr);
1327 
1328 		sparsebit_set(vm->vpages_mapped, vaddr >> vm->page_shift);
1329 	}
1330 
1331 	return vaddr_start;
1332 }
1333 
1334 /*
1335  * VM Virtual Address Allocate
1336  *
1337  * Input Args:
1338  *   vm - Virtual Machine
1339  *   sz - Size in bytes
1340  *   vaddr_min - Minimum starting virtual address
1341  *
1342  * Output Args: None
1343  *
1344  * Return:
1345  *   Starting guest virtual address
1346  *
1347  * Allocates at least sz bytes within the virtual address space of the vm
1348  * given by vm.  The allocated bytes are mapped to a virtual address >=
1349  * the address given by vaddr_min.  Note that each allocation uses a
1350  * a unique set of pages, with the minimum real allocation being at least
1351  * a page. The allocated physical space comes from the TEST_DATA memory region.
1352  */
1353 vm_vaddr_t vm_vaddr_alloc(struct kvm_vm *vm, size_t sz, vm_vaddr_t vaddr_min)
1354 {
1355 	return __vm_vaddr_alloc(vm, sz, vaddr_min, MEM_REGION_TEST_DATA);
1356 }
1357 
1358 /*
1359  * VM Virtual Address Allocate Pages
1360  *
1361  * Input Args:
1362  *   vm - Virtual Machine
1363  *
1364  * Output Args: None
1365  *
1366  * Return:
1367  *   Starting guest virtual address
1368  *
1369  * Allocates at least N system pages worth of bytes within the virtual address
1370  * space of the vm.
1371  */
1372 vm_vaddr_t vm_vaddr_alloc_pages(struct kvm_vm *vm, int nr_pages)
1373 {
1374 	return vm_vaddr_alloc(vm, nr_pages * getpagesize(), KVM_UTIL_MIN_VADDR);
1375 }
1376 
1377 vm_vaddr_t __vm_vaddr_alloc_page(struct kvm_vm *vm, enum kvm_mem_region_type type)
1378 {
1379 	return __vm_vaddr_alloc(vm, getpagesize(), KVM_UTIL_MIN_VADDR, type);
1380 }
1381 
1382 /*
1383  * VM Virtual Address Allocate Page
1384  *
1385  * Input Args:
1386  *   vm - Virtual Machine
1387  *
1388  * Output Args: None
1389  *
1390  * Return:
1391  *   Starting guest virtual address
1392  *
1393  * Allocates at least one system page worth of bytes within the virtual address
1394  * space of the vm.
1395  */
1396 vm_vaddr_t vm_vaddr_alloc_page(struct kvm_vm *vm)
1397 {
1398 	return vm_vaddr_alloc_pages(vm, 1);
1399 }
1400 
1401 /*
1402  * Map a range of VM virtual address to the VM's physical address
1403  *
1404  * Input Args:
1405  *   vm - Virtual Machine
1406  *   vaddr - Virtuall address to map
1407  *   paddr - VM Physical Address
1408  *   npages - The number of pages to map
1409  *
1410  * Output Args: None
1411  *
1412  * Return: None
1413  *
1414  * Within the VM given by @vm, creates a virtual translation for
1415  * @npages starting at @vaddr to the page range starting at @paddr.
1416  */
1417 void virt_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr,
1418 	      unsigned int npages)
1419 {
1420 	size_t page_size = vm->page_size;
1421 	size_t size = npages * page_size;
1422 
1423 	TEST_ASSERT(vaddr + size > vaddr, "Vaddr overflow");
1424 	TEST_ASSERT(paddr + size > paddr, "Paddr overflow");
1425 
1426 	while (npages--) {
1427 		virt_pg_map(vm, vaddr, paddr);
1428 		sparsebit_set(vm->vpages_mapped, vaddr >> vm->page_shift);
1429 
1430 		vaddr += page_size;
1431 		paddr += page_size;
1432 	}
1433 }
1434 
1435 /*
1436  * Address VM Physical to Host Virtual
1437  *
1438  * Input Args:
1439  *   vm - Virtual Machine
1440  *   gpa - VM physical address
1441  *
1442  * Output Args: None
1443  *
1444  * Return:
1445  *   Equivalent host virtual address
1446  *
1447  * Locates the memory region containing the VM physical address given
1448  * by gpa, within the VM given by vm.  When found, the host virtual
1449  * address providing the memory to the vm physical address is returned.
1450  * A TEST_ASSERT failure occurs if no region containing gpa exists.
1451  */
1452 void *addr_gpa2hva(struct kvm_vm *vm, vm_paddr_t gpa)
1453 {
1454 	struct userspace_mem_region *region;
1455 
1456 	region = userspace_mem_region_find(vm, gpa, gpa);
1457 	if (!region) {
1458 		TEST_FAIL("No vm physical memory at 0x%lx", gpa);
1459 		return NULL;
1460 	}
1461 
1462 	return (void *)((uintptr_t)region->host_mem
1463 		+ (gpa - region->region.guest_phys_addr));
1464 }
1465 
1466 /*
1467  * Address Host Virtual to VM Physical
1468  *
1469  * Input Args:
1470  *   vm - Virtual Machine
1471  *   hva - Host virtual address
1472  *
1473  * Output Args: None
1474  *
1475  * Return:
1476  *   Equivalent VM physical address
1477  *
1478  * Locates the memory region containing the host virtual address given
1479  * by hva, within the VM given by vm.  When found, the equivalent
1480  * VM physical address is returned. A TEST_ASSERT failure occurs if no
1481  * region containing hva exists.
1482  */
1483 vm_paddr_t addr_hva2gpa(struct kvm_vm *vm, void *hva)
1484 {
1485 	struct rb_node *node;
1486 
1487 	for (node = vm->regions.hva_tree.rb_node; node; ) {
1488 		struct userspace_mem_region *region =
1489 			container_of(node, struct userspace_mem_region, hva_node);
1490 
1491 		if (hva >= region->host_mem) {
1492 			if (hva <= (region->host_mem
1493 				+ region->region.memory_size - 1))
1494 				return (vm_paddr_t)((uintptr_t)
1495 					region->region.guest_phys_addr
1496 					+ (hva - (uintptr_t)region->host_mem));
1497 
1498 			node = node->rb_right;
1499 		} else
1500 			node = node->rb_left;
1501 	}
1502 
1503 	TEST_FAIL("No mapping to a guest physical address, hva: %p", hva);
1504 	return -1;
1505 }
1506 
1507 /*
1508  * Address VM physical to Host Virtual *alias*.
1509  *
1510  * Input Args:
1511  *   vm - Virtual Machine
1512  *   gpa - VM physical address
1513  *
1514  * Output Args: None
1515  *
1516  * Return:
1517  *   Equivalent address within the host virtual *alias* area, or NULL
1518  *   (without failing the test) if the guest memory is not shared (so
1519  *   no alias exists).
1520  *
1521  * Create a writable, shared virtual=>physical alias for the specific GPA.
1522  * The primary use case is to allow the host selftest to manipulate guest
1523  * memory without mapping said memory in the guest's address space. And, for
1524  * userfaultfd-based demand paging, to do so without triggering userfaults.
1525  */
1526 void *addr_gpa2alias(struct kvm_vm *vm, vm_paddr_t gpa)
1527 {
1528 	struct userspace_mem_region *region;
1529 	uintptr_t offset;
1530 
1531 	region = userspace_mem_region_find(vm, gpa, gpa);
1532 	if (!region)
1533 		return NULL;
1534 
1535 	if (!region->host_alias)
1536 		return NULL;
1537 
1538 	offset = gpa - region->region.guest_phys_addr;
1539 	return (void *) ((uintptr_t) region->host_alias + offset);
1540 }
1541 
1542 /* Create an interrupt controller chip for the specified VM. */
1543 void vm_create_irqchip(struct kvm_vm *vm)
1544 {
1545 	vm_ioctl(vm, KVM_CREATE_IRQCHIP, NULL);
1546 
1547 	vm->has_irqchip = true;
1548 }
1549 
1550 int _vcpu_run(struct kvm_vcpu *vcpu)
1551 {
1552 	int rc;
1553 
1554 	do {
1555 		rc = __vcpu_run(vcpu);
1556 	} while (rc == -1 && errno == EINTR);
1557 
1558 	assert_on_unhandled_exception(vcpu);
1559 
1560 	return rc;
1561 }
1562 
1563 /*
1564  * Invoke KVM_RUN on a vCPU until KVM returns something other than -EINTR.
1565  * Assert if the KVM returns an error (other than -EINTR).
1566  */
1567 void vcpu_run(struct kvm_vcpu *vcpu)
1568 {
1569 	int ret = _vcpu_run(vcpu);
1570 
1571 	TEST_ASSERT(!ret, KVM_IOCTL_ERROR(KVM_RUN, ret));
1572 }
1573 
1574 void vcpu_run_complete_io(struct kvm_vcpu *vcpu)
1575 {
1576 	int ret;
1577 
1578 	vcpu->run->immediate_exit = 1;
1579 	ret = __vcpu_run(vcpu);
1580 	vcpu->run->immediate_exit = 0;
1581 
1582 	TEST_ASSERT(ret == -1 && errno == EINTR,
1583 		    "KVM_RUN IOCTL didn't exit immediately, rc: %i, errno: %i",
1584 		    ret, errno);
1585 }
1586 
1587 /*
1588  * Get the list of guest registers which are supported for
1589  * KVM_GET_ONE_REG/KVM_SET_ONE_REG ioctls.  Returns a kvm_reg_list pointer,
1590  * it is the caller's responsibility to free the list.
1591  */
1592 struct kvm_reg_list *vcpu_get_reg_list(struct kvm_vcpu *vcpu)
1593 {
1594 	struct kvm_reg_list reg_list_n = { .n = 0 }, *reg_list;
1595 	int ret;
1596 
1597 	ret = __vcpu_ioctl(vcpu, KVM_GET_REG_LIST, &reg_list_n);
1598 	TEST_ASSERT(ret == -1 && errno == E2BIG, "KVM_GET_REG_LIST n=0");
1599 
1600 	reg_list = calloc(1, sizeof(*reg_list) + reg_list_n.n * sizeof(__u64));
1601 	reg_list->n = reg_list_n.n;
1602 	vcpu_ioctl(vcpu, KVM_GET_REG_LIST, reg_list);
1603 	return reg_list;
1604 }
1605 
1606 void *vcpu_map_dirty_ring(struct kvm_vcpu *vcpu)
1607 {
1608 	uint32_t page_size = getpagesize();
1609 	uint32_t size = vcpu->vm->dirty_ring_size;
1610 
1611 	TEST_ASSERT(size > 0, "Should enable dirty ring first");
1612 
1613 	if (!vcpu->dirty_gfns) {
1614 		void *addr;
1615 
1616 		addr = mmap(NULL, size, PROT_READ, MAP_PRIVATE, vcpu->fd,
1617 			    page_size * KVM_DIRTY_LOG_PAGE_OFFSET);
1618 		TEST_ASSERT(addr == MAP_FAILED, "Dirty ring mapped private");
1619 
1620 		addr = mmap(NULL, size, PROT_READ | PROT_EXEC, MAP_PRIVATE, vcpu->fd,
1621 			    page_size * KVM_DIRTY_LOG_PAGE_OFFSET);
1622 		TEST_ASSERT(addr == MAP_FAILED, "Dirty ring mapped exec");
1623 
1624 		addr = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, vcpu->fd,
1625 			    page_size * KVM_DIRTY_LOG_PAGE_OFFSET);
1626 		TEST_ASSERT(addr != MAP_FAILED, "Dirty ring map failed");
1627 
1628 		vcpu->dirty_gfns = addr;
1629 		vcpu->dirty_gfns_count = size / sizeof(struct kvm_dirty_gfn);
1630 	}
1631 
1632 	return vcpu->dirty_gfns;
1633 }
1634 
1635 /*
1636  * Device Ioctl
1637  */
1638 
1639 int __kvm_has_device_attr(int dev_fd, uint32_t group, uint64_t attr)
1640 {
1641 	struct kvm_device_attr attribute = {
1642 		.group = group,
1643 		.attr = attr,
1644 		.flags = 0,
1645 	};
1646 
1647 	return ioctl(dev_fd, KVM_HAS_DEVICE_ATTR, &attribute);
1648 }
1649 
1650 int __kvm_test_create_device(struct kvm_vm *vm, uint64_t type)
1651 {
1652 	struct kvm_create_device create_dev = {
1653 		.type = type,
1654 		.flags = KVM_CREATE_DEVICE_TEST,
1655 	};
1656 
1657 	return __vm_ioctl(vm, KVM_CREATE_DEVICE, &create_dev);
1658 }
1659 
1660 int __kvm_create_device(struct kvm_vm *vm, uint64_t type)
1661 {
1662 	struct kvm_create_device create_dev = {
1663 		.type = type,
1664 		.fd = -1,
1665 		.flags = 0,
1666 	};
1667 	int err;
1668 
1669 	err = __vm_ioctl(vm, KVM_CREATE_DEVICE, &create_dev);
1670 	TEST_ASSERT(err <= 0, "KVM_CREATE_DEVICE shouldn't return a positive value");
1671 	return err ? : create_dev.fd;
1672 }
1673 
1674 int __kvm_device_attr_get(int dev_fd, uint32_t group, uint64_t attr, void *val)
1675 {
1676 	struct kvm_device_attr kvmattr = {
1677 		.group = group,
1678 		.attr = attr,
1679 		.flags = 0,
1680 		.addr = (uintptr_t)val,
1681 	};
1682 
1683 	return __kvm_ioctl(dev_fd, KVM_GET_DEVICE_ATTR, &kvmattr);
1684 }
1685 
1686 int __kvm_device_attr_set(int dev_fd, uint32_t group, uint64_t attr, void *val)
1687 {
1688 	struct kvm_device_attr kvmattr = {
1689 		.group = group,
1690 		.attr = attr,
1691 		.flags = 0,
1692 		.addr = (uintptr_t)val,
1693 	};
1694 
1695 	return __kvm_ioctl(dev_fd, KVM_SET_DEVICE_ATTR, &kvmattr);
1696 }
1697 
1698 /*
1699  * IRQ related functions.
1700  */
1701 
1702 int _kvm_irq_line(struct kvm_vm *vm, uint32_t irq, int level)
1703 {
1704 	struct kvm_irq_level irq_level = {
1705 		.irq    = irq,
1706 		.level  = level,
1707 	};
1708 
1709 	return __vm_ioctl(vm, KVM_IRQ_LINE, &irq_level);
1710 }
1711 
1712 void kvm_irq_line(struct kvm_vm *vm, uint32_t irq, int level)
1713 {
1714 	int ret = _kvm_irq_line(vm, irq, level);
1715 
1716 	TEST_ASSERT(ret >= 0, KVM_IOCTL_ERROR(KVM_IRQ_LINE, ret));
1717 }
1718 
1719 struct kvm_irq_routing *kvm_gsi_routing_create(void)
1720 {
1721 	struct kvm_irq_routing *routing;
1722 	size_t size;
1723 
1724 	size = sizeof(struct kvm_irq_routing);
1725 	/* Allocate space for the max number of entries: this wastes 196 KBs. */
1726 	size += KVM_MAX_IRQ_ROUTES * sizeof(struct kvm_irq_routing_entry);
1727 	routing = calloc(1, size);
1728 	assert(routing);
1729 
1730 	return routing;
1731 }
1732 
1733 void kvm_gsi_routing_irqchip_add(struct kvm_irq_routing *routing,
1734 		uint32_t gsi, uint32_t pin)
1735 {
1736 	int i;
1737 
1738 	assert(routing);
1739 	assert(routing->nr < KVM_MAX_IRQ_ROUTES);
1740 
1741 	i = routing->nr;
1742 	routing->entries[i].gsi = gsi;
1743 	routing->entries[i].type = KVM_IRQ_ROUTING_IRQCHIP;
1744 	routing->entries[i].flags = 0;
1745 	routing->entries[i].u.irqchip.irqchip = 0;
1746 	routing->entries[i].u.irqchip.pin = pin;
1747 	routing->nr++;
1748 }
1749 
1750 int _kvm_gsi_routing_write(struct kvm_vm *vm, struct kvm_irq_routing *routing)
1751 {
1752 	int ret;
1753 
1754 	assert(routing);
1755 	ret = __vm_ioctl(vm, KVM_SET_GSI_ROUTING, routing);
1756 	free(routing);
1757 
1758 	return ret;
1759 }
1760 
1761 void kvm_gsi_routing_write(struct kvm_vm *vm, struct kvm_irq_routing *routing)
1762 {
1763 	int ret;
1764 
1765 	ret = _kvm_gsi_routing_write(vm, routing);
1766 	TEST_ASSERT(!ret, KVM_IOCTL_ERROR(KVM_SET_GSI_ROUTING, ret));
1767 }
1768 
1769 /*
1770  * VM Dump
1771  *
1772  * Input Args:
1773  *   vm - Virtual Machine
1774  *   indent - Left margin indent amount
1775  *
1776  * Output Args:
1777  *   stream - Output FILE stream
1778  *
1779  * Return: None
1780  *
1781  * Dumps the current state of the VM given by vm, to the FILE stream
1782  * given by stream.
1783  */
1784 void vm_dump(FILE *stream, struct kvm_vm *vm, uint8_t indent)
1785 {
1786 	int ctr;
1787 	struct userspace_mem_region *region;
1788 	struct kvm_vcpu *vcpu;
1789 
1790 	fprintf(stream, "%*smode: 0x%x\n", indent, "", vm->mode);
1791 	fprintf(stream, "%*sfd: %i\n", indent, "", vm->fd);
1792 	fprintf(stream, "%*spage_size: 0x%x\n", indent, "", vm->page_size);
1793 	fprintf(stream, "%*sMem Regions:\n", indent, "");
1794 	hash_for_each(vm->regions.slot_hash, ctr, region, slot_node) {
1795 		fprintf(stream, "%*sguest_phys: 0x%lx size: 0x%lx "
1796 			"host_virt: %p\n", indent + 2, "",
1797 			(uint64_t) region->region.guest_phys_addr,
1798 			(uint64_t) region->region.memory_size,
1799 			region->host_mem);
1800 		fprintf(stream, "%*sunused_phy_pages: ", indent + 2, "");
1801 		sparsebit_dump(stream, region->unused_phy_pages, 0);
1802 	}
1803 	fprintf(stream, "%*sMapped Virtual Pages:\n", indent, "");
1804 	sparsebit_dump(stream, vm->vpages_mapped, indent + 2);
1805 	fprintf(stream, "%*spgd_created: %u\n", indent, "",
1806 		vm->pgd_created);
1807 	if (vm->pgd_created) {
1808 		fprintf(stream, "%*sVirtual Translation Tables:\n",
1809 			indent + 2, "");
1810 		virt_dump(stream, vm, indent + 4);
1811 	}
1812 	fprintf(stream, "%*sVCPUs:\n", indent, "");
1813 
1814 	list_for_each_entry(vcpu, &vm->vcpus, list)
1815 		vcpu_dump(stream, vcpu, indent + 2);
1816 }
1817 
1818 /* Known KVM exit reasons */
1819 static struct exit_reason {
1820 	unsigned int reason;
1821 	const char *name;
1822 } exit_reasons_known[] = {
1823 	{KVM_EXIT_UNKNOWN, "UNKNOWN"},
1824 	{KVM_EXIT_EXCEPTION, "EXCEPTION"},
1825 	{KVM_EXIT_IO, "IO"},
1826 	{KVM_EXIT_HYPERCALL, "HYPERCALL"},
1827 	{KVM_EXIT_DEBUG, "DEBUG"},
1828 	{KVM_EXIT_HLT, "HLT"},
1829 	{KVM_EXIT_MMIO, "MMIO"},
1830 	{KVM_EXIT_IRQ_WINDOW_OPEN, "IRQ_WINDOW_OPEN"},
1831 	{KVM_EXIT_SHUTDOWN, "SHUTDOWN"},
1832 	{KVM_EXIT_FAIL_ENTRY, "FAIL_ENTRY"},
1833 	{KVM_EXIT_INTR, "INTR"},
1834 	{KVM_EXIT_SET_TPR, "SET_TPR"},
1835 	{KVM_EXIT_TPR_ACCESS, "TPR_ACCESS"},
1836 	{KVM_EXIT_S390_SIEIC, "S390_SIEIC"},
1837 	{KVM_EXIT_S390_RESET, "S390_RESET"},
1838 	{KVM_EXIT_DCR, "DCR"},
1839 	{KVM_EXIT_NMI, "NMI"},
1840 	{KVM_EXIT_INTERNAL_ERROR, "INTERNAL_ERROR"},
1841 	{KVM_EXIT_OSI, "OSI"},
1842 	{KVM_EXIT_PAPR_HCALL, "PAPR_HCALL"},
1843 	{KVM_EXIT_DIRTY_RING_FULL, "DIRTY_RING_FULL"},
1844 	{KVM_EXIT_X86_RDMSR, "RDMSR"},
1845 	{KVM_EXIT_X86_WRMSR, "WRMSR"},
1846 	{KVM_EXIT_XEN, "XEN"},
1847 #ifdef KVM_EXIT_MEMORY_NOT_PRESENT
1848 	{KVM_EXIT_MEMORY_NOT_PRESENT, "MEMORY_NOT_PRESENT"},
1849 #endif
1850 };
1851 
1852 /*
1853  * Exit Reason String
1854  *
1855  * Input Args:
1856  *   exit_reason - Exit reason
1857  *
1858  * Output Args: None
1859  *
1860  * Return:
1861  *   Constant string pointer describing the exit reason.
1862  *
1863  * Locates and returns a constant string that describes the KVM exit
1864  * reason given by exit_reason.  If no such string is found, a constant
1865  * string of "Unknown" is returned.
1866  */
1867 const char *exit_reason_str(unsigned int exit_reason)
1868 {
1869 	unsigned int n1;
1870 
1871 	for (n1 = 0; n1 < ARRAY_SIZE(exit_reasons_known); n1++) {
1872 		if (exit_reason == exit_reasons_known[n1].reason)
1873 			return exit_reasons_known[n1].name;
1874 	}
1875 
1876 	return "Unknown";
1877 }
1878 
1879 /*
1880  * Physical Contiguous Page Allocator
1881  *
1882  * Input Args:
1883  *   vm - Virtual Machine
1884  *   num - number of pages
1885  *   paddr_min - Physical address minimum
1886  *   memslot - Memory region to allocate page from
1887  *
1888  * Output Args: None
1889  *
1890  * Return:
1891  *   Starting physical address
1892  *
1893  * Within the VM specified by vm, locates a range of available physical
1894  * pages at or above paddr_min. If found, the pages are marked as in use
1895  * and their base address is returned. A TEST_ASSERT failure occurs if
1896  * not enough pages are available at or above paddr_min.
1897  */
1898 vm_paddr_t vm_phy_pages_alloc(struct kvm_vm *vm, size_t num,
1899 			      vm_paddr_t paddr_min, uint32_t memslot)
1900 {
1901 	struct userspace_mem_region *region;
1902 	sparsebit_idx_t pg, base;
1903 
1904 	TEST_ASSERT(num > 0, "Must allocate at least one page");
1905 
1906 	TEST_ASSERT((paddr_min % vm->page_size) == 0, "Min physical address "
1907 		"not divisible by page size.\n"
1908 		"  paddr_min: 0x%lx page_size: 0x%x",
1909 		paddr_min, vm->page_size);
1910 
1911 	region = memslot2region(vm, memslot);
1912 	base = pg = paddr_min >> vm->page_shift;
1913 
1914 	do {
1915 		for (; pg < base + num; ++pg) {
1916 			if (!sparsebit_is_set(region->unused_phy_pages, pg)) {
1917 				base = pg = sparsebit_next_set(region->unused_phy_pages, pg);
1918 				break;
1919 			}
1920 		}
1921 	} while (pg && pg != base + num);
1922 
1923 	if (pg == 0) {
1924 		fprintf(stderr, "No guest physical page available, "
1925 			"paddr_min: 0x%lx page_size: 0x%x memslot: %u\n",
1926 			paddr_min, vm->page_size, memslot);
1927 		fputs("---- vm dump ----\n", stderr);
1928 		vm_dump(stderr, vm, 2);
1929 		abort();
1930 	}
1931 
1932 	for (pg = base; pg < base + num; ++pg)
1933 		sparsebit_clear(region->unused_phy_pages, pg);
1934 
1935 	return base * vm->page_size;
1936 }
1937 
1938 vm_paddr_t vm_phy_page_alloc(struct kvm_vm *vm, vm_paddr_t paddr_min,
1939 			     uint32_t memslot)
1940 {
1941 	return vm_phy_pages_alloc(vm, 1, paddr_min, memslot);
1942 }
1943 
1944 /* Arbitrary minimum physical address used for virtual translation tables. */
1945 #define KVM_GUEST_PAGE_TABLE_MIN_PADDR 0x180000
1946 
1947 vm_paddr_t vm_alloc_page_table(struct kvm_vm *vm)
1948 {
1949 	return vm_phy_page_alloc(vm, KVM_GUEST_PAGE_TABLE_MIN_PADDR,
1950 				 vm->memslots[MEM_REGION_PT]);
1951 }
1952 
1953 /*
1954  * Address Guest Virtual to Host Virtual
1955  *
1956  * Input Args:
1957  *   vm - Virtual Machine
1958  *   gva - VM virtual address
1959  *
1960  * Output Args: None
1961  *
1962  * Return:
1963  *   Equivalent host virtual address
1964  */
1965 void *addr_gva2hva(struct kvm_vm *vm, vm_vaddr_t gva)
1966 {
1967 	return addr_gpa2hva(vm, addr_gva2gpa(vm, gva));
1968 }
1969 
1970 unsigned long __weak vm_compute_max_gfn(struct kvm_vm *vm)
1971 {
1972 	return ((1ULL << vm->pa_bits) >> vm->page_shift) - 1;
1973 }
1974 
1975 static unsigned int vm_calc_num_pages(unsigned int num_pages,
1976 				      unsigned int page_shift,
1977 				      unsigned int new_page_shift,
1978 				      bool ceil)
1979 {
1980 	unsigned int n = 1 << (new_page_shift - page_shift);
1981 
1982 	if (page_shift >= new_page_shift)
1983 		return num_pages * (1 << (page_shift - new_page_shift));
1984 
1985 	return num_pages / n + !!(ceil && num_pages % n);
1986 }
1987 
1988 static inline int getpageshift(void)
1989 {
1990 	return __builtin_ffs(getpagesize()) - 1;
1991 }
1992 
1993 unsigned int
1994 vm_num_host_pages(enum vm_guest_mode mode, unsigned int num_guest_pages)
1995 {
1996 	return vm_calc_num_pages(num_guest_pages,
1997 				 vm_guest_mode_params[mode].page_shift,
1998 				 getpageshift(), true);
1999 }
2000 
2001 unsigned int
2002 vm_num_guest_pages(enum vm_guest_mode mode, unsigned int num_host_pages)
2003 {
2004 	return vm_calc_num_pages(num_host_pages, getpageshift(),
2005 				 vm_guest_mode_params[mode].page_shift, false);
2006 }
2007 
2008 unsigned int vm_calc_num_guest_pages(enum vm_guest_mode mode, size_t size)
2009 {
2010 	unsigned int n;
2011 	n = DIV_ROUND_UP(size, vm_guest_mode_params[mode].page_size);
2012 	return vm_adjust_num_guest_pages(mode, n);
2013 }
2014 
2015 /*
2016  * Read binary stats descriptors
2017  *
2018  * Input Args:
2019  *   stats_fd - the file descriptor for the binary stats file from which to read
2020  *   header - the binary stats metadata header corresponding to the given FD
2021  *
2022  * Output Args: None
2023  *
2024  * Return:
2025  *   A pointer to a newly allocated series of stat descriptors.
2026  *   Caller is responsible for freeing the returned kvm_stats_desc.
2027  *
2028  * Read the stats descriptors from the binary stats interface.
2029  */
2030 struct kvm_stats_desc *read_stats_descriptors(int stats_fd,
2031 					      struct kvm_stats_header *header)
2032 {
2033 	struct kvm_stats_desc *stats_desc;
2034 	ssize_t desc_size, total_size, ret;
2035 
2036 	desc_size = get_stats_descriptor_size(header);
2037 	total_size = header->num_desc * desc_size;
2038 
2039 	stats_desc = calloc(header->num_desc, desc_size);
2040 	TEST_ASSERT(stats_desc, "Allocate memory for stats descriptors");
2041 
2042 	ret = pread(stats_fd, stats_desc, total_size, header->desc_offset);
2043 	TEST_ASSERT(ret == total_size, "Read KVM stats descriptors");
2044 
2045 	return stats_desc;
2046 }
2047 
2048 /*
2049  * Read stat data for a particular stat
2050  *
2051  * Input Args:
2052  *   stats_fd - the file descriptor for the binary stats file from which to read
2053  *   header - the binary stats metadata header corresponding to the given FD
2054  *   desc - the binary stat metadata for the particular stat to be read
2055  *   max_elements - the maximum number of 8-byte values to read into data
2056  *
2057  * Output Args:
2058  *   data - the buffer into which stat data should be read
2059  *
2060  * Read the data values of a specified stat from the binary stats interface.
2061  */
2062 void read_stat_data(int stats_fd, struct kvm_stats_header *header,
2063 		    struct kvm_stats_desc *desc, uint64_t *data,
2064 		    size_t max_elements)
2065 {
2066 	size_t nr_elements = min_t(ssize_t, desc->size, max_elements);
2067 	size_t size = nr_elements * sizeof(*data);
2068 	ssize_t ret;
2069 
2070 	TEST_ASSERT(desc->size, "No elements in stat '%s'", desc->name);
2071 	TEST_ASSERT(max_elements, "Zero elements requested for stat '%s'", desc->name);
2072 
2073 	ret = pread(stats_fd, data, size,
2074 		    header->data_offset + desc->offset);
2075 
2076 	TEST_ASSERT(ret >= 0, "pread() failed on stat '%s', errno: %i (%s)",
2077 		    desc->name, errno, strerror(errno));
2078 	TEST_ASSERT(ret == size,
2079 		    "pread() on stat '%s' read %ld bytes, wanted %lu bytes",
2080 		    desc->name, size, ret);
2081 }
2082 
2083 /*
2084  * Read the data of the named stat
2085  *
2086  * Input Args:
2087  *   vm - the VM for which the stat should be read
2088  *   stat_name - the name of the stat to read
2089  *   max_elements - the maximum number of 8-byte values to read into data
2090  *
2091  * Output Args:
2092  *   data - the buffer into which stat data should be read
2093  *
2094  * Read the data values of a specified stat from the binary stats interface.
2095  */
2096 void __vm_get_stat(struct kvm_vm *vm, const char *stat_name, uint64_t *data,
2097 		   size_t max_elements)
2098 {
2099 	struct kvm_stats_desc *desc;
2100 	size_t size_desc;
2101 	int i;
2102 
2103 	if (!vm->stats_fd) {
2104 		vm->stats_fd = vm_get_stats_fd(vm);
2105 		read_stats_header(vm->stats_fd, &vm->stats_header);
2106 		vm->stats_desc = read_stats_descriptors(vm->stats_fd,
2107 							&vm->stats_header);
2108 	}
2109 
2110 	size_desc = get_stats_descriptor_size(&vm->stats_header);
2111 
2112 	for (i = 0; i < vm->stats_header.num_desc; ++i) {
2113 		desc = (void *)vm->stats_desc + (i * size_desc);
2114 
2115 		if (strcmp(desc->name, stat_name))
2116 			continue;
2117 
2118 		read_stat_data(vm->stats_fd, &vm->stats_header, desc,
2119 			       data, max_elements);
2120 
2121 		break;
2122 	}
2123 }
2124 
2125 __weak void kvm_arch_vm_post_create(struct kvm_vm *vm)
2126 {
2127 }
2128 
2129 __weak void kvm_selftest_arch_init(void)
2130 {
2131 }
2132 
2133 void __attribute((constructor)) kvm_selftest_init(void)
2134 {
2135 	/* Tell stdout not to buffer its content. */
2136 	setbuf(stdout, NULL);
2137 
2138 	kvm_selftest_arch_init();
2139 }
2140