memstress.c

  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * Copyright (C) 2020, Google LLC.
  4 */
  5#include <inttypes.h>
  6#include <linux/bitmap.h>
  7
  8#include "kvm_util.h"
  9#include "memstress.h"
 10#include "processor.h"
 11#include "ucall_common.h"
 12
 13struct memstress_args memstress_args;
 14
 15/*
 16 * Guest virtual memory offset of the testing memory slot.
 17 * Must not conflict with identity mapped test code.
 18 */
 19static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM;
 20
 21struct vcpu_thread {
 22	/* The index of the vCPU. */
 23	int vcpu_idx;
 24
 25	/* The pthread backing the vCPU. */
 26	pthread_t thread;
 27
 28	/* Set to true once the vCPU thread is up and running. */
 29	bool running;
 30};
 31
 32/* The vCPU threads involved in this test. */
 33static struct vcpu_thread vcpu_threads[KVM_MAX_VCPUS];
 34
 35/* The function run by each vCPU thread, as provided by the test. */
 36static void (*vcpu_thread_fn)(struct memstress_vcpu_args *);
 37
 38/* Set to true once all vCPU threads are up and running. */
 39static bool all_vcpu_threads_running;
 40
 41static struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
 42
 43/*
 44 * Continuously write to the first 8 bytes of each page in the
 45 * specified region.
 46 */
 47void memstress_guest_code(uint32_t vcpu_idx)
 48{
 49	struct memstress_args *args = &memstress_args;
 50	struct memstress_vcpu_args *vcpu_args = &args->vcpu_args[vcpu_idx];
 51	struct guest_random_state rand_state;
 52	uint64_t gva;
 53	uint64_t pages;
 54	uint64_t addr;
 55	uint64_t page;
 56	int i;
 57
 58	rand_state = new_guest_random_state(guest_random_seed + vcpu_idx);
 59
 60	gva = vcpu_args->gva;
 61	pages = vcpu_args->pages;
 62
 63	/* Make sure vCPU args data structure is not corrupt. */
 64	GUEST_ASSERT(vcpu_args->vcpu_idx == vcpu_idx);
 65
 66	while (true) {
 67		for (i = 0; i < sizeof(memstress_args); i += args->guest_page_size)
 68			(void) *((volatile char *)args + i);
 69
 70		for (i = 0; i < pages; i++) {
 71			if (args->random_access)
 72				page = guest_random_u32(&rand_state) % pages;
 73			else
 74				page = i;
 75
 76			addr = gva + (page * args->guest_page_size);
 77
 78			if (__guest_random_bool(&rand_state, args->write_percent))
 79				*(uint64_t *)addr = 0x0123456789ABCDEF;
 80			else
 81				READ_ONCE(*(uint64_t *)addr);
 82		}
 83
 84		GUEST_SYNC(1);
 85	}
 86}
 87
 88void memstress_setup_vcpus(struct kvm_vm *vm, int nr_vcpus,
 89			   struct kvm_vcpu *vcpus[],
 90			   uint64_t vcpu_memory_bytes,
 91			   bool partition_vcpu_memory_access)
 92{
 93	struct memstress_args *args = &memstress_args;
 94	struct memstress_vcpu_args *vcpu_args;
 95	int i;
 96
 97	for (i = 0; i < nr_vcpus; i++) {
 98		vcpu_args = &args->vcpu_args[i];
 99
100		vcpu_args->vcpu = vcpus[i];
101		vcpu_args->vcpu_idx = i;
102
103		if (partition_vcpu_memory_access) {
104			vcpu_args->gva = guest_test_virt_mem +
105					 (i * vcpu_memory_bytes);
106			vcpu_args->pages = vcpu_memory_bytes /
107					   args->guest_page_size;
108			vcpu_args->gpa = args->gpa + (i * vcpu_memory_bytes);
109		} else {
110			vcpu_args->gva = guest_test_virt_mem;
111			vcpu_args->pages = (nr_vcpus * vcpu_memory_bytes) /
112					   args->guest_page_size;
113			vcpu_args->gpa = args->gpa;
114		}
115
116		vcpu_args_set(vcpus[i], 1, i);
117
118		pr_debug("Added VCPU %d with test mem gpa [%lx, %lx)\n",
119			 i, vcpu_args->gpa, vcpu_args->gpa +
120			 (vcpu_args->pages * args->guest_page_size));
121	}
122}
123
124struct kvm_vm *memstress_create_vm(enum vm_guest_mode mode, int nr_vcpus,
125				   uint64_t vcpu_memory_bytes, int slots,
126				   enum vm_mem_backing_src_type backing_src,
127				   bool partition_vcpu_memory_access)
128{
129	struct memstress_args *args = &memstress_args;
130	struct kvm_vm *vm;
131	uint64_t guest_num_pages, slot0_pages = 0;
132	uint64_t backing_src_pagesz = get_backing_src_pagesz(backing_src);
133	uint64_t region_end_gfn;
134	int i;
135
136	pr_info("Testing guest mode: %s\n", vm_guest_mode_string(mode));
137
138	/* By default vCPUs will write to memory. */
139	args->write_percent = 100;
140
141	/*
142	 * Snapshot the non-huge page size.  This is used by the guest code to
143	 * access/dirty pages at the logging granularity.
144	 */
145	args->guest_page_size = vm_guest_mode_params[mode].page_size;
146
147	guest_num_pages = vm_adjust_num_guest_pages(mode,
148				(nr_vcpus * vcpu_memory_bytes) / args->guest_page_size);
149
150	TEST_ASSERT(vcpu_memory_bytes % getpagesize() == 0,
151		    "Guest memory size is not host page size aligned.");
152	TEST_ASSERT(vcpu_memory_bytes % args->guest_page_size == 0,
153		    "Guest memory size is not guest page size aligned.");
154	TEST_ASSERT(guest_num_pages % slots == 0,
155		    "Guest memory cannot be evenly divided into %d slots.",
156		    slots);
157
158	/*
159	 * If using nested, allocate extra pages for the nested page tables and
160	 * in-memory data structures.
161	 */
162	if (args->nested)
163		slot0_pages += memstress_nested_pages(nr_vcpus);
164
165	/*
166	 * Pass guest_num_pages to populate the page tables for test memory.
167	 * The memory is also added to memslot 0, but that's a benign side
168	 * effect as KVM allows aliasing HVAs in meslots.
169	 */
170	vm = __vm_create_with_vcpus(VM_SHAPE(mode), nr_vcpus,
171				    slot0_pages + guest_num_pages,
172				    memstress_guest_code, vcpus);
173
174	args->vm = vm;
175
176	/* Put the test region at the top guest physical memory. */
177	region_end_gfn = vm->max_gfn + 1;
178
179#ifdef __x86_64__
180	/*
181	 * When running vCPUs in L2, restrict the test region to 48 bits to
182	 * avoid needing 5-level page tables to identity map L2.
183	 */
184	if (args->nested)
185		region_end_gfn = min(region_end_gfn, (1UL << 48) / args->guest_page_size);
186#endif
187	/*
188	 * If there should be more memory in the guest test region than there
189	 * can be pages in the guest, it will definitely cause problems.
190	 */
191	TEST_ASSERT(guest_num_pages < region_end_gfn,
192		    "Requested more guest memory than address space allows.\n"
193		    "    guest pages: %" PRIx64 " max gfn: %" PRIx64
194		    " nr_vcpus: %d wss: %" PRIx64 "]",
195		    guest_num_pages, region_end_gfn - 1, nr_vcpus, vcpu_memory_bytes);
196
197	args->gpa = (region_end_gfn - guest_num_pages - 1) * args->guest_page_size;
198	args->gpa = align_down(args->gpa, backing_src_pagesz);
199#ifdef __s390x__
200	/* Align to 1M (segment size) */
201	args->gpa = align_down(args->gpa, 1 << 20);
202#endif
203	args->size = guest_num_pages * args->guest_page_size;
204	pr_info("guest physical test memory: [0x%lx, 0x%lx)\n",
205		args->gpa, args->gpa + args->size);
206
207	/* Add extra memory slots for testing */
208	for (i = 0; i < slots; i++) {
209		uint64_t region_pages = guest_num_pages / slots;
210		vm_paddr_t region_start = args->gpa + region_pages * args->guest_page_size * i;
211
212		vm_userspace_mem_region_add(vm, backing_src, region_start,
213					    MEMSTRESS_MEM_SLOT_INDEX + i,
214					    region_pages, 0);
215	}
216
217	/* Do mapping for the demand paging memory slot */
218	virt_map(vm, guest_test_virt_mem, args->gpa, guest_num_pages);
219
220	memstress_setup_vcpus(vm, nr_vcpus, vcpus, vcpu_memory_bytes,
221			      partition_vcpu_memory_access);
222
223	if (args->nested) {
224		pr_info("Configuring vCPUs to run in L2 (nested).\n");
225		memstress_setup_nested(vm, nr_vcpus, vcpus);
226	}
227
228	/* Export the shared variables to the guest. */
229	sync_global_to_guest(vm, memstress_args);
230
231	return vm;
232}
233
234void memstress_destroy_vm(struct kvm_vm *vm)
235{
236	kvm_vm_free(vm);
237}
238
239void memstress_set_write_percent(struct kvm_vm *vm, uint32_t write_percent)
240{
241	memstress_args.write_percent = write_percent;
242	sync_global_to_guest(vm, memstress_args.write_percent);
243}
244
245void memstress_set_random_access(struct kvm_vm *vm, bool random_access)
246{
247	memstress_args.random_access = random_access;
248	sync_global_to_guest(vm, memstress_args.random_access);
249}
250
251uint64_t __weak memstress_nested_pages(int nr_vcpus)
252{
253	return 0;
254}
255
256void __weak memstress_setup_nested(struct kvm_vm *vm, int nr_vcpus, struct kvm_vcpu **vcpus)
257{
258	pr_info("%s() not support on this architecture, skipping.\n", __func__);
259	exit(KSFT_SKIP);
260}
261
262static void *vcpu_thread_main(void *data)
263{
264	struct vcpu_thread *vcpu = data;
265	int vcpu_idx = vcpu->vcpu_idx;
266
267	if (memstress_args.pin_vcpus)
268		pin_self_to_cpu(memstress_args.vcpu_to_pcpu[vcpu_idx]);
269
270	WRITE_ONCE(vcpu->running, true);
271
272	/*
273	 * Wait for all vCPU threads to be up and running before calling the test-
274	 * provided vCPU thread function. This prevents thread creation (which
275	 * requires taking the mmap_sem in write mode) from interfering with the
276	 * guest faulting in its memory.
277	 */
278	while (!READ_ONCE(all_vcpu_threads_running))
279		;
280
281	vcpu_thread_fn(&memstress_args.vcpu_args[vcpu_idx]);
282
283	return NULL;
284}
285
286void memstress_start_vcpu_threads(int nr_vcpus,
287				  void (*vcpu_fn)(struct memstress_vcpu_args *))
288{
289	int i;
290
291	vcpu_thread_fn = vcpu_fn;
292	WRITE_ONCE(all_vcpu_threads_running, false);
293	WRITE_ONCE(memstress_args.stop_vcpus, false);
294
295	for (i = 0; i < nr_vcpus; i++) {
296		struct vcpu_thread *vcpu = &vcpu_threads[i];
297
298		vcpu->vcpu_idx = i;
299		WRITE_ONCE(vcpu->running, false);
300
301		pthread_create(&vcpu->thread, NULL, vcpu_thread_main, vcpu);
302	}
303
304	for (i = 0; i < nr_vcpus; i++) {
305		while (!READ_ONCE(vcpu_threads[i].running))
306			;
307	}
308
309	WRITE_ONCE(all_vcpu_threads_running, true);
310}
311
312void memstress_join_vcpu_threads(int nr_vcpus)
313{
314	int i;
315
316	WRITE_ONCE(memstress_args.stop_vcpus, true);
317
318	for (i = 0; i < nr_vcpus; i++)
319		pthread_join(vcpu_threads[i].thread, NULL);
320}
321
322static void toggle_dirty_logging(struct kvm_vm *vm, int slots, bool enable)
323{
324	int i;
325
326	for (i = 0; i < slots; i++) {
327		int slot = MEMSTRESS_MEM_SLOT_INDEX + i;
328		int flags = enable ? KVM_MEM_LOG_DIRTY_PAGES : 0;
329
330		vm_mem_region_set_flags(vm, slot, flags);
331	}
332}
333
334void memstress_enable_dirty_logging(struct kvm_vm *vm, int slots)
335{
336	toggle_dirty_logging(vm, slots, true);
337}
338
339void memstress_disable_dirty_logging(struct kvm_vm *vm, int slots)
340{
341	toggle_dirty_logging(vm, slots, false);
342}
343
344void memstress_get_dirty_log(struct kvm_vm *vm, unsigned long *bitmaps[], int slots)
345{
346	int i;
347
348	for (i = 0; i < slots; i++) {
349		int slot = MEMSTRESS_MEM_SLOT_INDEX + i;
350
351		kvm_vm_get_dirty_log(vm, slot, bitmaps[i]);
352	}
353}
354
355void memstress_clear_dirty_log(struct kvm_vm *vm, unsigned long *bitmaps[],
356			       int slots, uint64_t pages_per_slot)
357{
358	int i;
359
360	for (i = 0; i < slots; i++) {
361		int slot = MEMSTRESS_MEM_SLOT_INDEX + i;
362
363		kvm_vm_clear_dirty_log(vm, slot, bitmaps[i], 0, pages_per_slot);
364	}
365}
366
367unsigned long **memstress_alloc_bitmaps(int slots, uint64_t pages_per_slot)
368{
369	unsigned long **bitmaps;
370	int i;
371
372	bitmaps = malloc(slots * sizeof(bitmaps[0]));
373	TEST_ASSERT(bitmaps, "Failed to allocate bitmaps array.");
374
375	for (i = 0; i < slots; i++) {
376		bitmaps[i] = bitmap_zalloc(pages_per_slot);
377		TEST_ASSERT(bitmaps[i], "Failed to allocate slot bitmap.");
378	}
379
380	return bitmaps;
381}
382
383void memstress_free_bitmaps(unsigned long *bitmaps[], int slots)
384{
385	int i;
386
387	for (i = 0; i < slots; i++)
388		free(bitmaps[i]);
389
390	free(bitmaps);
391}