xref: /openbmc/linux/tools/testing/selftests/kvm/lib/userfaultfd_util.c (revision 2612e3bbc0386368a850140a6c9b990cd496a5ec)
1a93871d0SRicardo Koller // SPDX-License-Identifier: GPL-2.0
2a93871d0SRicardo Koller /*
3a93871d0SRicardo Koller  * KVM userfaultfd util
4a93871d0SRicardo Koller  * Adapted from demand_paging_test.c
5a93871d0SRicardo Koller  *
6a93871d0SRicardo Koller  * Copyright (C) 2018, Red Hat, Inc.
7a93871d0SRicardo Koller  * Copyright (C) 2019-2022 Google LLC
8a93871d0SRicardo Koller  */
9a93871d0SRicardo Koller 
10a93871d0SRicardo Koller #define _GNU_SOURCE /* for pipe2 */
11a93871d0SRicardo Koller 
12a93871d0SRicardo Koller #include <inttypes.h>
13a93871d0SRicardo Koller #include <stdio.h>
14a93871d0SRicardo Koller #include <stdlib.h>
15a93871d0SRicardo Koller #include <time.h>
16a93871d0SRicardo Koller #include <poll.h>
17a93871d0SRicardo Koller #include <pthread.h>
18a93871d0SRicardo Koller #include <linux/userfaultfd.h>
19a93871d0SRicardo Koller #include <sys/syscall.h>
20a93871d0SRicardo Koller 
21a93871d0SRicardo Koller #include "kvm_util.h"
22a93871d0SRicardo Koller #include "test_util.h"
23eb561891SPaolo Bonzini #include "memstress.h"
24a93871d0SRicardo Koller #include "userfaultfd_util.h"
25a93871d0SRicardo Koller 
26a93871d0SRicardo Koller #ifdef __NR_userfaultfd
27a93871d0SRicardo Koller 
uffd_handler_thread_fn(void * arg)28a93871d0SRicardo Koller static void *uffd_handler_thread_fn(void *arg)
29a93871d0SRicardo Koller {
30a93871d0SRicardo Koller 	struct uffd_desc *uffd_desc = (struct uffd_desc *)arg;
31a93871d0SRicardo Koller 	int uffd = uffd_desc->uffd;
32a93871d0SRicardo Koller 	int pipefd = uffd_desc->pipefds[0];
33a93871d0SRicardo Koller 	useconds_t delay = uffd_desc->delay;
34a93871d0SRicardo Koller 	int64_t pages = 0;
35a93871d0SRicardo Koller 	struct timespec start;
36a93871d0SRicardo Koller 	struct timespec ts_diff;
37a93871d0SRicardo Koller 
38a93871d0SRicardo Koller 	clock_gettime(CLOCK_MONOTONIC, &start);
39a93871d0SRicardo Koller 	while (1) {
40a93871d0SRicardo Koller 		struct uffd_msg msg;
41a93871d0SRicardo Koller 		struct pollfd pollfd[2];
42a93871d0SRicardo Koller 		char tmp_chr;
43a93871d0SRicardo Koller 		int r;
44a93871d0SRicardo Koller 
45a93871d0SRicardo Koller 		pollfd[0].fd = uffd;
46a93871d0SRicardo Koller 		pollfd[0].events = POLLIN;
47a93871d0SRicardo Koller 		pollfd[1].fd = pipefd;
48a93871d0SRicardo Koller 		pollfd[1].events = POLLIN;
49a93871d0SRicardo Koller 
50a93871d0SRicardo Koller 		r = poll(pollfd, 2, -1);
51a93871d0SRicardo Koller 		switch (r) {
52a93871d0SRicardo Koller 		case -1:
53a93871d0SRicardo Koller 			pr_info("poll err");
54a93871d0SRicardo Koller 			continue;
55a93871d0SRicardo Koller 		case 0:
56a93871d0SRicardo Koller 			continue;
57a93871d0SRicardo Koller 		case 1:
58a93871d0SRicardo Koller 			break;
59a93871d0SRicardo Koller 		default:
60a93871d0SRicardo Koller 			pr_info("Polling uffd returned %d", r);
61a93871d0SRicardo Koller 			return NULL;
62a93871d0SRicardo Koller 		}
63a93871d0SRicardo Koller 
64a93871d0SRicardo Koller 		if (pollfd[0].revents & POLLERR) {
65a93871d0SRicardo Koller 			pr_info("uffd revents has POLLERR");
66a93871d0SRicardo Koller 			return NULL;
67a93871d0SRicardo Koller 		}
68a93871d0SRicardo Koller 
69a93871d0SRicardo Koller 		if (pollfd[1].revents & POLLIN) {
70a93871d0SRicardo Koller 			r = read(pollfd[1].fd, &tmp_chr, 1);
71a93871d0SRicardo Koller 			TEST_ASSERT(r == 1,
72a93871d0SRicardo Koller 				    "Error reading pipefd in UFFD thread\n");
73*21912a65SPeter Xu 			break;
74a93871d0SRicardo Koller 		}
75a93871d0SRicardo Koller 
76a93871d0SRicardo Koller 		if (!(pollfd[0].revents & POLLIN))
77a93871d0SRicardo Koller 			continue;
78a93871d0SRicardo Koller 
79a93871d0SRicardo Koller 		r = read(uffd, &msg, sizeof(msg));
80a93871d0SRicardo Koller 		if (r == -1) {
81a93871d0SRicardo Koller 			if (errno == EAGAIN)
82a93871d0SRicardo Koller 				continue;
83a93871d0SRicardo Koller 			pr_info("Read of uffd got errno %d\n", errno);
84a93871d0SRicardo Koller 			return NULL;
85a93871d0SRicardo Koller 		}
86a93871d0SRicardo Koller 
87a93871d0SRicardo Koller 		if (r != sizeof(msg)) {
88a93871d0SRicardo Koller 			pr_info("Read on uffd returned unexpected size: %d bytes", r);
89a93871d0SRicardo Koller 			return NULL;
90a93871d0SRicardo Koller 		}
91a93871d0SRicardo Koller 
92a93871d0SRicardo Koller 		if (!(msg.event & UFFD_EVENT_PAGEFAULT))
93a93871d0SRicardo Koller 			continue;
94a93871d0SRicardo Koller 
95a93871d0SRicardo Koller 		if (delay)
96a93871d0SRicardo Koller 			usleep(delay);
97a93871d0SRicardo Koller 		r = uffd_desc->handler(uffd_desc->uffd_mode, uffd, &msg);
98a93871d0SRicardo Koller 		if (r < 0)
99a93871d0SRicardo Koller 			return NULL;
100a93871d0SRicardo Koller 		pages++;
101a93871d0SRicardo Koller 	}
102a93871d0SRicardo Koller 
103a93871d0SRicardo Koller 	ts_diff = timespec_elapsed(start);
104a93871d0SRicardo Koller 	PER_VCPU_DEBUG("userfaulted %ld pages over %ld.%.9lds. (%f/sec)\n",
105a93871d0SRicardo Koller 		       pages, ts_diff.tv_sec, ts_diff.tv_nsec,
106*21912a65SPeter Xu 		       pages / ((double)ts_diff.tv_sec + (double)ts_diff.tv_nsec / NSEC_PER_SEC));
107a93871d0SRicardo Koller 
108a93871d0SRicardo Koller 	return NULL;
109a93871d0SRicardo Koller }
110a93871d0SRicardo Koller 
uffd_setup_demand_paging(int uffd_mode,useconds_t delay,void * hva,uint64_t len,uffd_handler_t handler)111a93871d0SRicardo Koller struct uffd_desc *uffd_setup_demand_paging(int uffd_mode, useconds_t delay,
112a93871d0SRicardo Koller 					   void *hva, uint64_t len,
113a93871d0SRicardo Koller 					   uffd_handler_t handler)
114a93871d0SRicardo Koller {
115a93871d0SRicardo Koller 	struct uffd_desc *uffd_desc;
116a93871d0SRicardo Koller 	bool is_minor = (uffd_mode == UFFDIO_REGISTER_MODE_MINOR);
117a93871d0SRicardo Koller 	int uffd;
118a93871d0SRicardo Koller 	struct uffdio_api uffdio_api;
119a93871d0SRicardo Koller 	struct uffdio_register uffdio_register;
120a93871d0SRicardo Koller 	uint64_t expected_ioctls = ((uint64_t) 1) << _UFFDIO_COPY;
121a93871d0SRicardo Koller 	int ret;
122a93871d0SRicardo Koller 
123a93871d0SRicardo Koller 	PER_PAGE_DEBUG("Userfaultfd %s mode, faults resolved with %s\n",
124a93871d0SRicardo Koller 		       is_minor ? "MINOR" : "MISSING",
125a93871d0SRicardo Koller 		       is_minor ? "UFFDIO_CONINUE" : "UFFDIO_COPY");
126a93871d0SRicardo Koller 
127a93871d0SRicardo Koller 	uffd_desc = malloc(sizeof(struct uffd_desc));
128a93871d0SRicardo Koller 	TEST_ASSERT(uffd_desc, "malloc failed");
129a93871d0SRicardo Koller 
130a93871d0SRicardo Koller 	/* In order to get minor faults, prefault via the alias. */
131a93871d0SRicardo Koller 	if (is_minor)
132a93871d0SRicardo Koller 		expected_ioctls = ((uint64_t) 1) << _UFFDIO_CONTINUE;
133a93871d0SRicardo Koller 
134a93871d0SRicardo Koller 	uffd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
135a93871d0SRicardo Koller 	TEST_ASSERT(uffd >= 0, "uffd creation failed, errno: %d", errno);
136a93871d0SRicardo Koller 
137a93871d0SRicardo Koller 	uffdio_api.api = UFFD_API;
138a93871d0SRicardo Koller 	uffdio_api.features = 0;
139a93871d0SRicardo Koller 	TEST_ASSERT(ioctl(uffd, UFFDIO_API, &uffdio_api) != -1,
140a93871d0SRicardo Koller 		    "ioctl UFFDIO_API failed: %" PRIu64,
141a93871d0SRicardo Koller 		    (uint64_t)uffdio_api.api);
142a93871d0SRicardo Koller 
143a93871d0SRicardo Koller 	uffdio_register.range.start = (uint64_t)hva;
144a93871d0SRicardo Koller 	uffdio_register.range.len = len;
145a93871d0SRicardo Koller 	uffdio_register.mode = uffd_mode;
146a93871d0SRicardo Koller 	TEST_ASSERT(ioctl(uffd, UFFDIO_REGISTER, &uffdio_register) != -1,
147a93871d0SRicardo Koller 		    "ioctl UFFDIO_REGISTER failed");
148a93871d0SRicardo Koller 	TEST_ASSERT((uffdio_register.ioctls & expected_ioctls) ==
149a93871d0SRicardo Koller 		    expected_ioctls, "missing userfaultfd ioctls");
150a93871d0SRicardo Koller 
151a93871d0SRicardo Koller 	ret = pipe2(uffd_desc->pipefds, O_CLOEXEC | O_NONBLOCK);
152a93871d0SRicardo Koller 	TEST_ASSERT(!ret, "Failed to set up pipefd");
153a93871d0SRicardo Koller 
154a93871d0SRicardo Koller 	uffd_desc->uffd_mode = uffd_mode;
155a93871d0SRicardo Koller 	uffd_desc->uffd = uffd;
156a93871d0SRicardo Koller 	uffd_desc->delay = delay;
157a93871d0SRicardo Koller 	uffd_desc->handler = handler;
158a93871d0SRicardo Koller 	pthread_create(&uffd_desc->thread, NULL, uffd_handler_thread_fn,
159a93871d0SRicardo Koller 		       uffd_desc);
160a93871d0SRicardo Koller 
161a93871d0SRicardo Koller 	PER_VCPU_DEBUG("Created uffd thread for HVA range [%p, %p)\n",
162a93871d0SRicardo Koller 		       hva, hva + len);
163a93871d0SRicardo Koller 
164a93871d0SRicardo Koller 	return uffd_desc;
165a93871d0SRicardo Koller }
166a93871d0SRicardo Koller 
uffd_stop_demand_paging(struct uffd_desc * uffd)167a93871d0SRicardo Koller void uffd_stop_demand_paging(struct uffd_desc *uffd)
168a93871d0SRicardo Koller {
169a93871d0SRicardo Koller 	char c = 0;
170a93871d0SRicardo Koller 	int ret;
171a93871d0SRicardo Koller 
172a93871d0SRicardo Koller 	ret = write(uffd->pipefds[1], &c, 1);
173a93871d0SRicardo Koller 	TEST_ASSERT(ret == 1, "Unable to write to pipefd");
174a93871d0SRicardo Koller 
175a93871d0SRicardo Koller 	ret = pthread_join(uffd->thread, NULL);
176a93871d0SRicardo Koller 	TEST_ASSERT(ret == 0, "Pthread_join failed.");
177a93871d0SRicardo Koller 
178a93871d0SRicardo Koller 	close(uffd->uffd);
179a93871d0SRicardo Koller 
180a93871d0SRicardo Koller 	close(uffd->pipefds[1]);
181a93871d0SRicardo Koller 	close(uffd->pipefds[0]);
182a93871d0SRicardo Koller 
183a93871d0SRicardo Koller 	free(uffd);
184a93871d0SRicardo Koller }
185a93871d0SRicardo Koller 
186a93871d0SRicardo Koller #endif /* __NR_userfaultfd */
187