1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * KVM demand paging test 4 * Adapted from dirty_log_test.c 5 * 6 * Copyright (C) 2018, Red Hat, Inc. 7 * Copyright (C) 2019, Google, Inc. 8 */ 9 #include <inttypes.h> 10 #include <stdio.h> 11 #include <stdlib.h> 12 #include <time.h> 13 #include <poll.h> 14 #include <pthread.h> 15 #include <linux/userfaultfd.h> 16 #include <sys/syscall.h> 17 18 #include "kvm_util.h" 19 #include "test_util.h" 20 #include "memstress.h" 21 #include "guest_modes.h" 22 #include "ucall_common.h" 23 #include "userfaultfd_util.h" 24 25 #ifdef __NR_userfaultfd 26 27 static int nr_vcpus = 1; 28 static uint64_t guest_percpu_mem_size = DEFAULT_PER_VCPU_MEM_SIZE; 29 30 static size_t demand_paging_size; 31 static char *guest_data_prototype; 32 33 static void vcpu_worker(struct memstress_vcpu_args *vcpu_args) 34 { 35 struct kvm_vcpu *vcpu = vcpu_args->vcpu; 36 int vcpu_idx = vcpu_args->vcpu_idx; 37 struct kvm_run *run = vcpu->run; 38 struct timespec start; 39 struct timespec ts_diff; 40 int ret; 41 42 clock_gettime(CLOCK_MONOTONIC, &start); 43 44 /* Let the guest access its memory */ 45 ret = _vcpu_run(vcpu); 46 TEST_ASSERT(ret == 0, "vcpu_run failed: %d", ret); 47 if (get_ucall(vcpu, NULL) != UCALL_SYNC) { 48 TEST_ASSERT(false, 49 "Invalid guest sync status: exit_reason=%s", 50 exit_reason_str(run->exit_reason)); 51 } 52 53 ts_diff = timespec_elapsed(start); 54 PER_VCPU_DEBUG("vCPU %d execution time: %ld.%.9lds\n", vcpu_idx, 55 ts_diff.tv_sec, ts_diff.tv_nsec); 56 } 57 58 static int handle_uffd_page_request(int uffd_mode, int uffd, 59 struct uffd_msg *msg) 60 { 61 pid_t tid = syscall(__NR_gettid); 62 uint64_t addr = msg->arg.pagefault.address; 63 struct timespec start; 64 struct timespec ts_diff; 65 int r; 66 67 clock_gettime(CLOCK_MONOTONIC, &start); 68 69 if (uffd_mode == UFFDIO_REGISTER_MODE_MISSING) { 70 struct uffdio_copy copy; 71 72 copy.src = (uint64_t)guest_data_prototype; 73 copy.dst = addr; 74 copy.len = demand_paging_size; 75 copy.mode = 0; 76 77 r = ioctl(uffd, UFFDIO_COPY, ©); 78 if (r == -1) { 79 pr_info("Failed UFFDIO_COPY in 0x%lx from thread %d with errno: %d\n", 80 addr, tid, errno); 81 return r; 82 } 83 } else if (uffd_mode == UFFDIO_REGISTER_MODE_MINOR) { 84 struct uffdio_continue cont = {0}; 85 86 cont.range.start = addr; 87 cont.range.len = demand_paging_size; 88 89 r = ioctl(uffd, UFFDIO_CONTINUE, &cont); 90 if (r == -1) { 91 pr_info("Failed UFFDIO_CONTINUE in 0x%lx from thread %d with errno: %d\n", 92 addr, tid, errno); 93 return r; 94 } 95 } else { 96 TEST_FAIL("Invalid uffd mode %d", uffd_mode); 97 } 98 99 ts_diff = timespec_elapsed(start); 100 101 PER_PAGE_DEBUG("UFFD page-in %d \t%ld ns\n", tid, 102 timespec_to_ns(ts_diff)); 103 PER_PAGE_DEBUG("Paged in %ld bytes at 0x%lx from thread %d\n", 104 demand_paging_size, addr, tid); 105 106 return 0; 107 } 108 109 struct test_params { 110 int uffd_mode; 111 useconds_t uffd_delay; 112 enum vm_mem_backing_src_type src_type; 113 bool partition_vcpu_memory_access; 114 }; 115 116 static void prefault_mem(void *alias, uint64_t len) 117 { 118 size_t p; 119 120 TEST_ASSERT(alias != NULL, "Alias required for minor faults"); 121 for (p = 0; p < (len / demand_paging_size); ++p) { 122 memcpy(alias + (p * demand_paging_size), 123 guest_data_prototype, demand_paging_size); 124 } 125 } 126 127 static void run_test(enum vm_guest_mode mode, void *arg) 128 { 129 struct memstress_vcpu_args *vcpu_args; 130 struct test_params *p = arg; 131 struct uffd_desc **uffd_descs = NULL; 132 struct timespec start; 133 struct timespec ts_diff; 134 struct kvm_vm *vm; 135 int i; 136 137 vm = memstress_create_vm(mode, nr_vcpus, guest_percpu_mem_size, 1, 138 p->src_type, p->partition_vcpu_memory_access); 139 140 demand_paging_size = get_backing_src_pagesz(p->src_type); 141 142 guest_data_prototype = malloc(demand_paging_size); 143 TEST_ASSERT(guest_data_prototype, 144 "Failed to allocate buffer for guest data pattern"); 145 memset(guest_data_prototype, 0xAB, demand_paging_size); 146 147 if (p->uffd_mode == UFFDIO_REGISTER_MODE_MINOR) { 148 for (i = 0; i < nr_vcpus; i++) { 149 vcpu_args = &memstress_args.vcpu_args[i]; 150 prefault_mem(addr_gpa2alias(vm, vcpu_args->gpa), 151 vcpu_args->pages * memstress_args.guest_page_size); 152 } 153 } 154 155 if (p->uffd_mode) { 156 uffd_descs = malloc(nr_vcpus * sizeof(struct uffd_desc *)); 157 TEST_ASSERT(uffd_descs, "Memory allocation failed"); 158 for (i = 0; i < nr_vcpus; i++) { 159 void *vcpu_hva; 160 161 vcpu_args = &memstress_args.vcpu_args[i]; 162 163 /* Cache the host addresses of the region */ 164 vcpu_hva = addr_gpa2hva(vm, vcpu_args->gpa); 165 /* 166 * Set up user fault fd to handle demand paging 167 * requests. 168 */ 169 uffd_descs[i] = uffd_setup_demand_paging( 170 p->uffd_mode, p->uffd_delay, vcpu_hva, 171 vcpu_args->pages * memstress_args.guest_page_size, 172 &handle_uffd_page_request); 173 } 174 } 175 176 pr_info("Finished creating vCPUs and starting uffd threads\n"); 177 178 clock_gettime(CLOCK_MONOTONIC, &start); 179 memstress_start_vcpu_threads(nr_vcpus, vcpu_worker); 180 pr_info("Started all vCPUs\n"); 181 182 memstress_join_vcpu_threads(nr_vcpus); 183 ts_diff = timespec_elapsed(start); 184 pr_info("All vCPU threads joined\n"); 185 186 if (p->uffd_mode) { 187 /* Tell the user fault fd handler threads to quit */ 188 for (i = 0; i < nr_vcpus; i++) 189 uffd_stop_demand_paging(uffd_descs[i]); 190 } 191 192 pr_info("Total guest execution time: %ld.%.9lds\n", 193 ts_diff.tv_sec, ts_diff.tv_nsec); 194 pr_info("Overall demand paging rate: %f pgs/sec\n", 195 memstress_args.vcpu_args[0].pages * nr_vcpus / 196 ((double)ts_diff.tv_sec + (double)ts_diff.tv_nsec / NSEC_PER_SEC)); 197 198 memstress_destroy_vm(vm); 199 200 free(guest_data_prototype); 201 if (p->uffd_mode) 202 free(uffd_descs); 203 } 204 205 static void help(char *name) 206 { 207 puts(""); 208 printf("usage: %s [-h] [-m vm_mode] [-u uffd_mode] [-d uffd_delay_usec]\n" 209 " [-b memory] [-s type] [-v vcpus] [-c cpu_list] [-o]\n", name); 210 guest_modes_help(); 211 printf(" -u: use userfaultfd to handle vCPU page faults. Mode is a\n" 212 " UFFD registration mode: 'MISSING' or 'MINOR'.\n"); 213 kvm_print_vcpu_pinning_help(); 214 printf(" -d: add a delay in usec to the User Fault\n" 215 " FD handler to simulate demand paging\n" 216 " overheads. Ignored without -u.\n"); 217 printf(" -b: specify the size of the memory region which should be\n" 218 " demand paged by each vCPU. e.g. 10M or 3G.\n" 219 " Default: 1G\n"); 220 backing_src_help("-s"); 221 printf(" -v: specify the number of vCPUs to run.\n"); 222 printf(" -o: Overlap guest memory accesses instead of partitioning\n" 223 " them into a separate region of memory for each vCPU.\n"); 224 puts(""); 225 exit(0); 226 } 227 228 int main(int argc, char *argv[]) 229 { 230 int max_vcpus = kvm_check_cap(KVM_CAP_MAX_VCPUS); 231 const char *cpulist = NULL; 232 struct test_params p = { 233 .src_type = DEFAULT_VM_MEM_SRC, 234 .partition_vcpu_memory_access = true, 235 }; 236 int opt; 237 238 guest_modes_append_default(); 239 240 while ((opt = getopt(argc, argv, "hm:u:d:b:s:v:c:o")) != -1) { 241 switch (opt) { 242 case 'm': 243 guest_modes_cmdline(optarg); 244 break; 245 case 'u': 246 if (!strcmp("MISSING", optarg)) 247 p.uffd_mode = UFFDIO_REGISTER_MODE_MISSING; 248 else if (!strcmp("MINOR", optarg)) 249 p.uffd_mode = UFFDIO_REGISTER_MODE_MINOR; 250 TEST_ASSERT(p.uffd_mode, "UFFD mode must be 'MISSING' or 'MINOR'."); 251 break; 252 case 'd': 253 p.uffd_delay = strtoul(optarg, NULL, 0); 254 TEST_ASSERT(p.uffd_delay >= 0, "A negative UFFD delay is not supported."); 255 break; 256 case 'b': 257 guest_percpu_mem_size = parse_size(optarg); 258 break; 259 case 's': 260 p.src_type = parse_backing_src_type(optarg); 261 break; 262 case 'v': 263 nr_vcpus = atoi_positive("Number of vCPUs", optarg); 264 TEST_ASSERT(nr_vcpus <= max_vcpus, 265 "Invalid number of vcpus, must be between 1 and %d", max_vcpus); 266 break; 267 case 'c': 268 cpulist = optarg; 269 break; 270 case 'o': 271 p.partition_vcpu_memory_access = false; 272 break; 273 case 'h': 274 default: 275 help(argv[0]); 276 break; 277 } 278 } 279 280 if (p.uffd_mode == UFFDIO_REGISTER_MODE_MINOR && 281 !backing_src_is_shared(p.src_type)) { 282 TEST_FAIL("userfaultfd MINOR mode requires shared memory; pick a different -s"); 283 } 284 285 if (cpulist) { 286 kvm_parse_vcpu_pinning(cpulist, memstress_args.vcpu_to_pcpu, 287 nr_vcpus); 288 memstress_args.pin_vcpus = true; 289 } 290 291 for_each_guest_mode(run_test, &p); 292 293 return 0; 294 } 295 296 #else /* __NR_userfaultfd */ 297 298 #warning "missing __NR_userfaultfd definition" 299 300 int main(void) 301 { 302 print_skip("__NR_userfaultfd must be present for userfaultfd test"); 303 return KSFT_SKIP; 304 } 305 306 #endif /* __NR_userfaultfd */ 307