xref: /linux/tools/testing/selftests/kvm/lib/userfaultfd_util.c (revision 954ea91fb68b771dba6d87cfa61b68e09cc2497f)
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  
10  #define _GNU_SOURCE /* for pipe2 */
11  
12  #include <inttypes.h>
13  #include <stdio.h>
14  #include <stdlib.h>
15  #include <time.h>
16  #include <poll.h>
17  #include <pthread.h>
18  #include <linux/userfaultfd.h>
19  #include <sys/syscall.h>
20  
21  #include "kvm_util.h"
22  #include "test_util.h"
23  #include "memstress.h"
24  #include "userfaultfd_util.h"
25  
26  #ifdef __NR_userfaultfd
27  
28  static void *uffd_handler_thread_fn(void *arg)
29  {
30  	struct uffd_desc *uffd_desc = (struct uffd_desc *)arg;
31  	int uffd = uffd_desc->uffd;
32  	int pipefd = uffd_desc->pipefds[0];
33  	useconds_t delay = uffd_desc->delay;
34  	int64_t pages = 0;
35  	struct timespec start;
36  	struct timespec ts_diff;
37  
38  	clock_gettime(CLOCK_MONOTONIC, &start);
39  	while (1) {
40  		struct uffd_msg msg;
41  		struct pollfd pollfd[2];
42  		char tmp_chr;
43  		int r;
44  
45  		pollfd[0].fd = uffd;
46  		pollfd[0].events = POLLIN;
47  		pollfd[1].fd = pipefd;
48  		pollfd[1].events = POLLIN;
49  
50  		r = poll(pollfd, 2, -1);
51  		switch (r) {
52  		case -1:
53  			pr_info("poll err");
54  			continue;
55  		case 0:
56  			continue;
57  		case 1:
58  			break;
59  		default:
60  			pr_info("Polling uffd returned %d", r);
61  			return NULL;
62  		}
63  
64  		if (pollfd[0].revents & POLLERR) {
65  			pr_info("uffd revents has POLLERR");
66  			return NULL;
67  		}
68  
69  		if (pollfd[1].revents & POLLIN) {
70  			r = read(pollfd[1].fd, &tmp_chr, 1);
71  			TEST_ASSERT(r == 1,
72  				    "Error reading pipefd in UFFD thread\n");
73  			return NULL;
74  		}
75  
76  		if (!(pollfd[0].revents & POLLIN))
77  			continue;
78  
79  		r = read(uffd, &msg, sizeof(msg));
80  		if (r == -1) {
81  			if (errno == EAGAIN)
82  				continue;
83  			pr_info("Read of uffd got errno %d\n", errno);
84  			return NULL;
85  		}
86  
87  		if (r != sizeof(msg)) {
88  			pr_info("Read on uffd returned unexpected size: %d bytes", r);
89  			return NULL;
90  		}
91  
92  		if (!(msg.event & UFFD_EVENT_PAGEFAULT))
93  			continue;
94  
95  		if (delay)
96  			usleep(delay);
97  		r = uffd_desc->handler(uffd_desc->uffd_mode, uffd, &msg);
98  		if (r < 0)
99  			return NULL;
100  		pages++;
101  	}
102  
103  	ts_diff = timespec_elapsed(start);
104  	PER_VCPU_DEBUG("userfaulted %ld pages over %ld.%.9lds. (%f/sec)\n",
105  		       pages, ts_diff.tv_sec, ts_diff.tv_nsec,
106  		       pages / ((double)ts_diff.tv_sec + (double)ts_diff.tv_nsec / 100000000.0));
107  
108  	return NULL;
109  }
110  
111  struct uffd_desc *uffd_setup_demand_paging(int uffd_mode, useconds_t delay,
112  					   void *hva, uint64_t len,
113  					   uffd_handler_t handler)
114  {
115  	struct uffd_desc *uffd_desc;
116  	bool is_minor = (uffd_mode == UFFDIO_REGISTER_MODE_MINOR);
117  	int uffd;
118  	struct uffdio_api uffdio_api;
119  	struct uffdio_register uffdio_register;
120  	uint64_t expected_ioctls = ((uint64_t) 1) << _UFFDIO_COPY;
121  	int ret;
122  
123  	PER_PAGE_DEBUG("Userfaultfd %s mode, faults resolved with %s\n",
124  		       is_minor ? "MINOR" : "MISSING",
125  		       is_minor ? "UFFDIO_CONINUE" : "UFFDIO_COPY");
126  
127  	uffd_desc = malloc(sizeof(struct uffd_desc));
128  	TEST_ASSERT(uffd_desc, "malloc failed");
129  
130  	/* In order to get minor faults, prefault via the alias. */
131  	if (is_minor)
132  		expected_ioctls = ((uint64_t) 1) << _UFFDIO_CONTINUE;
133  
134  	uffd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
135  	TEST_ASSERT(uffd >= 0, "uffd creation failed, errno: %d", errno);
136  
137  	uffdio_api.api = UFFD_API;
138  	uffdio_api.features = 0;
139  	TEST_ASSERT(ioctl(uffd, UFFDIO_API, &uffdio_api) != -1,
140  		    "ioctl UFFDIO_API failed: %" PRIu64,
141  		    (uint64_t)uffdio_api.api);
142  
143  	uffdio_register.range.start = (uint64_t)hva;
144  	uffdio_register.range.len = len;
145  	uffdio_register.mode = uffd_mode;
146  	TEST_ASSERT(ioctl(uffd, UFFDIO_REGISTER, &uffdio_register) != -1,
147  		    "ioctl UFFDIO_REGISTER failed");
148  	TEST_ASSERT((uffdio_register.ioctls & expected_ioctls) ==
149  		    expected_ioctls, "missing userfaultfd ioctls");
150  
151  	ret = pipe2(uffd_desc->pipefds, O_CLOEXEC | O_NONBLOCK);
152  	TEST_ASSERT(!ret, "Failed to set up pipefd");
153  
154  	uffd_desc->uffd_mode = uffd_mode;
155  	uffd_desc->uffd = uffd;
156  	uffd_desc->delay = delay;
157  	uffd_desc->handler = handler;
158  	pthread_create(&uffd_desc->thread, NULL, uffd_handler_thread_fn,
159  		       uffd_desc);
160  
161  	PER_VCPU_DEBUG("Created uffd thread for HVA range [%p, %p)\n",
162  		       hva, hva + len);
163  
164  	return uffd_desc;
165  }
166  
167  void uffd_stop_demand_paging(struct uffd_desc *uffd)
168  {
169  	char c = 0;
170  	int ret;
171  
172  	ret = write(uffd->pipefds[1], &c, 1);
173  	TEST_ASSERT(ret == 1, "Unable to write to pipefd");
174  
175  	ret = pthread_join(uffd->thread, NULL);
176  	TEST_ASSERT(ret == 0, "Pthread_join failed.");
177  
178  	close(uffd->uffd);
179  
180  	close(uffd->pipefds[1]);
181  	close(uffd->pipefds[0]);
182  
183  	free(uffd);
184  }
185  
186  #endif /* __NR_userfaultfd */
187