1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * KVM dirty logging page splitting test
4 *
5 * Based on dirty_log_perf.c
6 *
7 * Copyright (C) 2018, Red Hat, Inc.
8 * Copyright (C) 2023, Google, Inc.
9 */
10
11 #include <stdio.h>
12 #include <stdlib.h>
13 #include <pthread.h>
14 #include <linux/bitmap.h>
15
16 #include "kvm_util.h"
17 #include "test_util.h"
18 #include "memstress.h"
19 #include "guest_modes.h"
20 #include "ucall_common.h"
21
22 #define VCPUS 2
23 #define SLOTS 2
24 #define ITERATIONS 2
25
26 static uint64_t guest_percpu_mem_size = DEFAULT_PER_VCPU_MEM_SIZE;
27
28 static enum vm_mem_backing_src_type backing_src = VM_MEM_SRC_ANONYMOUS_HUGETLB;
29
30 static u64 dirty_log_manual_caps;
31 static bool host_quit;
32 static int iteration;
33 static int vcpu_last_completed_iteration[KVM_MAX_VCPUS];
34
35 struct kvm_page_stats {
36 uint64_t pages_4k;
37 uint64_t pages_2m;
38 uint64_t pages_1g;
39 uint64_t hugepages;
40 };
41
get_page_stats(struct kvm_vm * vm,struct kvm_page_stats * stats,const char * stage)42 static void get_page_stats(struct kvm_vm *vm, struct kvm_page_stats *stats, const char *stage)
43 {
44 stats->pages_4k = vm_get_stat(vm, "pages_4k");
45 stats->pages_2m = vm_get_stat(vm, "pages_2m");
46 stats->pages_1g = vm_get_stat(vm, "pages_1g");
47 stats->hugepages = stats->pages_2m + stats->pages_1g;
48
49 pr_debug("\nPage stats after %s: 4K: %ld 2M: %ld 1G: %ld huge: %ld\n",
50 stage, stats->pages_4k, stats->pages_2m, stats->pages_1g,
51 stats->hugepages);
52 }
53
run_vcpu_iteration(struct kvm_vm * vm)54 static void run_vcpu_iteration(struct kvm_vm *vm)
55 {
56 int i;
57
58 iteration++;
59 for (i = 0; i < VCPUS; i++) {
60 while (READ_ONCE(vcpu_last_completed_iteration[i]) !=
61 iteration)
62 ;
63 }
64 }
65
vcpu_worker(struct memstress_vcpu_args * vcpu_args)66 static void vcpu_worker(struct memstress_vcpu_args *vcpu_args)
67 {
68 struct kvm_vcpu *vcpu = vcpu_args->vcpu;
69 int vcpu_idx = vcpu_args->vcpu_idx;
70
71 while (!READ_ONCE(host_quit)) {
72 int current_iteration = READ_ONCE(iteration);
73
74 vcpu_run(vcpu);
75
76 TEST_ASSERT_EQ(get_ucall(vcpu, NULL), UCALL_SYNC);
77
78 vcpu_last_completed_iteration[vcpu_idx] = current_iteration;
79
80 /* Wait for the start of the next iteration to be signaled. */
81 while (current_iteration == READ_ONCE(iteration) &&
82 READ_ONCE(iteration) >= 0 &&
83 !READ_ONCE(host_quit))
84 ;
85 }
86 }
87
run_test(enum vm_guest_mode mode,void * unused)88 static void run_test(enum vm_guest_mode mode, void *unused)
89 {
90 struct kvm_vm *vm;
91 unsigned long **bitmaps;
92 uint64_t guest_num_pages;
93 uint64_t host_num_pages;
94 uint64_t pages_per_slot;
95 int i;
96 struct kvm_page_stats stats_populated;
97 struct kvm_page_stats stats_dirty_logging_enabled;
98 struct kvm_page_stats stats_dirty_pass[ITERATIONS];
99 struct kvm_page_stats stats_clear_pass[ITERATIONS];
100 struct kvm_page_stats stats_dirty_logging_disabled;
101 struct kvm_page_stats stats_repopulated;
102
103 vm = memstress_create_vm(mode, VCPUS, guest_percpu_mem_size,
104 SLOTS, backing_src, false);
105
106 guest_num_pages = (VCPUS * guest_percpu_mem_size) >> vm->page_shift;
107 guest_num_pages = vm_adjust_num_guest_pages(mode, guest_num_pages);
108 host_num_pages = vm_num_host_pages(mode, guest_num_pages);
109 pages_per_slot = host_num_pages / SLOTS;
110 TEST_ASSERT_EQ(host_num_pages, pages_per_slot * SLOTS);
111 TEST_ASSERT(!(host_num_pages % 512),
112 "Number of pages, '%lu' not a multiple of 2MiB", host_num_pages);
113
114 bitmaps = memstress_alloc_bitmaps(SLOTS, pages_per_slot);
115
116 if (dirty_log_manual_caps)
117 vm_enable_cap(vm, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2,
118 dirty_log_manual_caps);
119
120 /* Start the iterations */
121 iteration = -1;
122 host_quit = false;
123
124 for (i = 0; i < VCPUS; i++)
125 vcpu_last_completed_iteration[i] = -1;
126
127 memstress_start_vcpu_threads(VCPUS, vcpu_worker);
128
129 run_vcpu_iteration(vm);
130 get_page_stats(vm, &stats_populated, "populating memory");
131
132 /* Enable dirty logging */
133 memstress_enable_dirty_logging(vm, SLOTS);
134
135 get_page_stats(vm, &stats_dirty_logging_enabled, "enabling dirty logging");
136
137 while (iteration < ITERATIONS) {
138 run_vcpu_iteration(vm);
139 get_page_stats(vm, &stats_dirty_pass[iteration - 1],
140 "dirtying memory");
141
142 memstress_get_dirty_log(vm, bitmaps, SLOTS);
143
144 if (dirty_log_manual_caps) {
145 memstress_clear_dirty_log(vm, bitmaps, SLOTS, pages_per_slot);
146
147 get_page_stats(vm, &stats_clear_pass[iteration - 1], "clearing dirty log");
148 }
149 }
150
151 /* Disable dirty logging */
152 memstress_disable_dirty_logging(vm, SLOTS);
153
154 get_page_stats(vm, &stats_dirty_logging_disabled, "disabling dirty logging");
155
156 /* Run vCPUs again to fault pages back in. */
157 run_vcpu_iteration(vm);
158 get_page_stats(vm, &stats_repopulated, "repopulating memory");
159
160 /*
161 * Tell the vCPU threads to quit. No need to manually check that vCPUs
162 * have stopped running after disabling dirty logging, the join will
163 * wait for them to exit.
164 */
165 host_quit = true;
166 memstress_join_vcpu_threads(VCPUS);
167
168 memstress_free_bitmaps(bitmaps, SLOTS);
169 memstress_destroy_vm(vm);
170
171 TEST_ASSERT_EQ((stats_populated.pages_2m * 512 +
172 stats_populated.pages_1g * 512 * 512), host_num_pages);
173
174 /*
175 * Check that all huge pages were split. Since large pages can only
176 * exist in the data slot, and the vCPUs should have dirtied all pages
177 * in the data slot, there should be no huge pages left after splitting.
178 * Splitting happens at dirty log enable time without
179 * KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 and after the first clear pass
180 * with that capability.
181 */
182 if (dirty_log_manual_caps) {
183 TEST_ASSERT_EQ(stats_clear_pass[0].hugepages, 0);
184 TEST_ASSERT(stats_clear_pass[0].pages_4k >= host_num_pages,
185 "Expected at least '%lu' 4KiB pages, found only '%lu'",
186 host_num_pages, stats_clear_pass[0].pages_4k);
187 TEST_ASSERT_EQ(stats_dirty_logging_enabled.hugepages, stats_populated.hugepages);
188 } else {
189 TEST_ASSERT_EQ(stats_dirty_logging_enabled.hugepages, 0);
190 TEST_ASSERT(stats_dirty_logging_enabled.pages_4k >= host_num_pages,
191 "Expected at least '%lu' 4KiB pages, found only '%lu'",
192 host_num_pages, stats_dirty_logging_enabled.pages_4k);
193 }
194
195 /*
196 * Once dirty logging is disabled and the vCPUs have touched all their
197 * memory again, the hugepage counts should be the same as they were
198 * right after initial population of memory.
199 */
200 TEST_ASSERT_EQ(stats_populated.pages_2m, stats_repopulated.pages_2m);
201 TEST_ASSERT_EQ(stats_populated.pages_1g, stats_repopulated.pages_1g);
202 }
203
help(char * name)204 static void help(char *name)
205 {
206 puts("");
207 printf("usage: %s [-h] [-b vcpu bytes] [-s mem type]\n",
208 name);
209 puts("");
210 printf(" -b: specify the size of the memory region which should be\n"
211 " dirtied by each vCPU. e.g. 10M or 3G.\n"
212 " (default: 1G)\n");
213 backing_src_help("-s");
214 puts("");
215 }
216
main(int argc,char * argv[])217 int main(int argc, char *argv[])
218 {
219 int opt;
220
221 TEST_REQUIRE(get_kvm_param_bool("eager_page_split"));
222 TEST_REQUIRE(get_kvm_param_bool("tdp_mmu"));
223
224 while ((opt = getopt(argc, argv, "b:hs:")) != -1) {
225 switch (opt) {
226 case 'b':
227 guest_percpu_mem_size = parse_size(optarg);
228 break;
229 case 'h':
230 help(argv[0]);
231 exit(0);
232 case 's':
233 backing_src = parse_backing_src_type(optarg);
234 break;
235 default:
236 help(argv[0]);
237 exit(1);
238 }
239 }
240
241 if (!is_backing_src_hugetlb(backing_src)) {
242 pr_info("This test will only work reliably with HugeTLB memory. "
243 "It can work with THP, but that is best effort.\n");
244 }
245
246 guest_modes_append_default();
247
248 dirty_log_manual_caps = 0;
249 for_each_guest_mode(run_test, NULL);
250
251 dirty_log_manual_caps =
252 kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2);
253
254 if (dirty_log_manual_caps) {
255 dirty_log_manual_caps &= (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE |
256 KVM_DIRTY_LOG_INITIALLY_SET);
257 for_each_guest_mode(run_test, NULL);
258 } else {
259 pr_info("Skipping testing with MANUAL_PROTECT as it is not supported");
260 }
261
262 return 0;
263 }
264