userfaultfd_util.c 5.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206
  1. // SPDX-License-Identifier: GPL-2.0
  2. /*
  3. * KVM userfaultfd util
  4. * Adapted from demand_paging_test.c
  5. *
  6. * Copyright (C) 2018, Red Hat, Inc.
  7. * Copyright (C) 2019-2022 Google LLC
  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/epoll.h>
  17. #include <sys/syscall.h>
  18. #include "kvm_util.h"
  19. #include "test_util.h"
  20. #include "memstress.h"
  21. #include "userfaultfd_util.h"
  22. #ifdef __NR_userfaultfd
  23. static void *uffd_handler_thread_fn(void *arg)
  24. {
  25. struct uffd_reader_args *reader_args = (struct uffd_reader_args *)arg;
  26. int uffd = reader_args->uffd;
  27. int64_t pages = 0;
  28. struct timespec start;
  29. struct timespec ts_diff;
  30. struct epoll_event evt;
  31. int epollfd;
  32. epollfd = epoll_create(1);
  33. TEST_ASSERT(epollfd >= 0, "Failed to create epollfd.");
  34. evt.events = EPOLLIN | EPOLLEXCLUSIVE;
  35. evt.data.u32 = 0;
  36. TEST_ASSERT(!epoll_ctl(epollfd, EPOLL_CTL_ADD, uffd, &evt),
  37. "Failed to add uffd to epollfd");
  38. evt.events = EPOLLIN;
  39. evt.data.u32 = 1;
  40. TEST_ASSERT(!epoll_ctl(epollfd, EPOLL_CTL_ADD, reader_args->pipe, &evt),
  41. "Failed to add pipe to epollfd");
  42. clock_gettime(CLOCK_MONOTONIC, &start);
  43. while (1) {
  44. struct uffd_msg msg;
  45. int r;
  46. r = epoll_wait(epollfd, &evt, 1, -1);
  47. TEST_ASSERT(r == 1,
  48. "Unexpected number of events (%d) from epoll, errno = %d",
  49. r, errno);
  50. if (evt.data.u32 == 1) {
  51. char tmp_chr;
  52. TEST_ASSERT(!(evt.events & (EPOLLERR | EPOLLHUP)),
  53. "Reader thread received EPOLLERR or EPOLLHUP on pipe.");
  54. r = read(reader_args->pipe, &tmp_chr, 1);
  55. TEST_ASSERT(r == 1,
  56. "Error reading pipefd in uffd reader thread");
  57. break;
  58. }
  59. TEST_ASSERT(!(evt.events & (EPOLLERR | EPOLLHUP)),
  60. "Reader thread received EPOLLERR or EPOLLHUP on uffd.");
  61. r = read(uffd, &msg, sizeof(msg));
  62. if (r == -1) {
  63. TEST_ASSERT(errno == EAGAIN,
  64. "Error reading from UFFD: errno = %d", errno);
  65. continue;
  66. }
  67. TEST_ASSERT(r == sizeof(msg),
  68. "Read on uffd returned unexpected number of bytes (%d)", r);
  69. if (!(msg.event & UFFD_EVENT_PAGEFAULT))
  70. continue;
  71. if (reader_args->delay)
  72. usleep(reader_args->delay);
  73. r = reader_args->handler(reader_args->uffd_mode, uffd, &msg);
  74. TEST_ASSERT(r >= 0,
  75. "Reader thread handler fn returned negative value %d", r);
  76. pages++;
  77. }
  78. ts_diff = timespec_elapsed(start);
  79. PER_VCPU_DEBUG("userfaulted %ld pages over %ld.%.9lds. (%f/sec)\n",
  80. pages, ts_diff.tv_sec, ts_diff.tv_nsec,
  81. pages / ((double)ts_diff.tv_sec + (double)ts_diff.tv_nsec / NSEC_PER_SEC));
  82. return NULL;
  83. }
  84. struct uffd_desc *uffd_setup_demand_paging(int uffd_mode, useconds_t delay,
  85. void *hva, uint64_t len,
  86. uint64_t num_readers,
  87. uffd_handler_t handler)
  88. {
  89. struct uffd_desc *uffd_desc;
  90. bool is_minor = (uffd_mode == UFFDIO_REGISTER_MODE_MINOR);
  91. int uffd;
  92. struct uffdio_api uffdio_api;
  93. struct uffdio_register uffdio_register;
  94. uint64_t expected_ioctls = ((uint64_t) 1) << _UFFDIO_COPY;
  95. int ret, i;
  96. PER_PAGE_DEBUG("Userfaultfd %s mode, faults resolved with %s\n",
  97. is_minor ? "MINOR" : "MISSING",
  98. is_minor ? "UFFDIO_CONTINUE" : "UFFDIO_COPY");
  99. uffd_desc = malloc(sizeof(struct uffd_desc));
  100. TEST_ASSERT(uffd_desc, "Failed to malloc uffd descriptor");
  101. uffd_desc->pipefds = calloc(sizeof(int), num_readers);
  102. TEST_ASSERT(uffd_desc->pipefds, "Failed to alloc pipes");
  103. uffd_desc->readers = calloc(sizeof(pthread_t), num_readers);
  104. TEST_ASSERT(uffd_desc->readers, "Failed to alloc reader threads");
  105. uffd_desc->reader_args = calloc(sizeof(struct uffd_reader_args), num_readers);
  106. TEST_ASSERT(uffd_desc->reader_args, "Failed to alloc reader_args");
  107. uffd_desc->num_readers = num_readers;
  108. /* In order to get minor faults, prefault via the alias. */
  109. if (is_minor)
  110. expected_ioctls = ((uint64_t) 1) << _UFFDIO_CONTINUE;
  111. uffd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
  112. TEST_ASSERT(uffd >= 0, "uffd creation failed, errno: %d", errno);
  113. uffdio_api.api = UFFD_API;
  114. uffdio_api.features = 0;
  115. TEST_ASSERT(ioctl(uffd, UFFDIO_API, &uffdio_api) != -1,
  116. "ioctl UFFDIO_API failed: %" PRIu64,
  117. (uint64_t)uffdio_api.api);
  118. uffdio_register.range.start = (uint64_t)hva;
  119. uffdio_register.range.len = len;
  120. uffdio_register.mode = uffd_mode;
  121. TEST_ASSERT(ioctl(uffd, UFFDIO_REGISTER, &uffdio_register) != -1,
  122. "ioctl UFFDIO_REGISTER failed");
  123. TEST_ASSERT((uffdio_register.ioctls & expected_ioctls) ==
  124. expected_ioctls, "missing userfaultfd ioctls");
  125. uffd_desc->uffd = uffd;
  126. for (i = 0; i < uffd_desc->num_readers; ++i) {
  127. int pipes[2];
  128. ret = pipe2((int *) &pipes, O_CLOEXEC | O_NONBLOCK);
  129. TEST_ASSERT(!ret, "Failed to set up pipefd %i for uffd_desc %p",
  130. i, uffd_desc);
  131. uffd_desc->pipefds[i] = pipes[1];
  132. uffd_desc->reader_args[i].uffd_mode = uffd_mode;
  133. uffd_desc->reader_args[i].uffd = uffd;
  134. uffd_desc->reader_args[i].delay = delay;
  135. uffd_desc->reader_args[i].handler = handler;
  136. uffd_desc->reader_args[i].pipe = pipes[0];
  137. pthread_create(&uffd_desc->readers[i], NULL, uffd_handler_thread_fn,
  138. &uffd_desc->reader_args[i]);
  139. PER_VCPU_DEBUG("Created uffd thread %i for HVA range [%p, %p)\n",
  140. i, hva, hva + len);
  141. }
  142. return uffd_desc;
  143. }
  144. void uffd_stop_demand_paging(struct uffd_desc *uffd)
  145. {
  146. char c = 0;
  147. int i;
  148. for (i = 0; i < uffd->num_readers; ++i)
  149. TEST_ASSERT(write(uffd->pipefds[i], &c, 1) == 1,
  150. "Unable to write to pipefd %i for uffd_desc %p", i, uffd);
  151. for (i = 0; i < uffd->num_readers; ++i)
  152. TEST_ASSERT(!pthread_join(uffd->readers[i], NULL),
  153. "Pthread_join failed on reader %i for uffd_desc %p", i, uffd);
  154. close(uffd->uffd);
  155. for (i = 0; i < uffd->num_readers; ++i) {
  156. close(uffd->pipefds[i]);
  157. close(uffd->reader_args[i].pipe);
  158. }
  159. free(uffd->pipefds);
  160. free(uffd->readers);
  161. free(uffd->reader_args);
  162. free(uffd);
  163. }
  164. #endif /* __NR_userfaultfd */