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
19#include "kvm_util.h"
20#include "test_util.h"
21#include "memstress.h"
22#include "userfaultfd_util.h"
23
24#ifdef __NR_userfaultfd
25
26static void *uffd_handler_thread_fn(void *arg)
27{
28	struct uffd_reader_args *reader_args = (struct uffd_reader_args *)arg;
29	int uffd = reader_args->uffd;
30	int64_t pages = 0;
31	struct timespec start;
32	struct timespec ts_diff;
33	struct epoll_event evt;
34	int epollfd;
35
36	epollfd = epoll_create(1);
37	TEST_ASSERT(epollfd >= 0, "Failed to create epollfd.");
38
39	evt.events = EPOLLIN | EPOLLEXCLUSIVE;
40	evt.data.u32 = 0;
41	TEST_ASSERT(!epoll_ctl(epollfd, EPOLL_CTL_ADD, uffd, &evt),
42		    "Failed to add uffd to epollfd");
43
44	evt.events = EPOLLIN;
45	evt.data.u32 = 1;
46	TEST_ASSERT(!epoll_ctl(epollfd, EPOLL_CTL_ADD, reader_args->pipe, &evt),
47		    "Failed to add pipe to epollfd");
48
49	clock_gettime(CLOCK_MONOTONIC, &start);
50	while (1) {
51		struct uffd_msg msg;
52		int r;
53
54		r = epoll_wait(epollfd, &evt, 1, -1);
55		TEST_ASSERT(r == 1,
56			    "Unexpected number of events (%d) from epoll, errno = %d",
57			    r, errno);
58
59		if (evt.data.u32 == 1) {
60			char tmp_chr;
61
62			TEST_ASSERT(!(evt.events & (EPOLLERR | EPOLLHUP)),
63				    "Reader thread received EPOLLERR or EPOLLHUP on pipe.");
64			r = read(reader_args->pipe, &tmp_chr, 1);
65			TEST_ASSERT(r == 1,
66				    "Error reading pipefd in uffd reader thread");
67			break;
68		}
69
70		TEST_ASSERT(!(evt.events & (EPOLLERR | EPOLLHUP)),
71			    "Reader thread received EPOLLERR or EPOLLHUP on uffd.");
72
73		r = read(uffd, &msg, sizeof(msg));
74		if (r == -1) {
75			TEST_ASSERT(errno == EAGAIN,
76				    "Error reading from UFFD: errno = %d", errno);
77			continue;
78		}
79
80		TEST_ASSERT(r == sizeof(msg),
81			    "Read on uffd returned unexpected number of bytes (%d)", r);
82
83		if (!(msg.event & UFFD_EVENT_PAGEFAULT))
84			continue;
85
86		if (reader_args->delay)
87			usleep(reader_args->delay);
88		r = reader_args->handler(reader_args->uffd_mode, uffd, &msg);
89		TEST_ASSERT(r >= 0,
90			    "Reader thread handler fn returned negative value %d", r);
91		pages++;
92	}
93
94	ts_diff = timespec_elapsed(start);
95	PER_VCPU_DEBUG("userfaulted %ld pages over %ld.%.9lds. (%f/sec)\n",
96		       pages, ts_diff.tv_sec, ts_diff.tv_nsec,
97		       pages / ((double)ts_diff.tv_sec + (double)ts_diff.tv_nsec / NSEC_PER_SEC));
98
99	return NULL;
100}
101
102struct uffd_desc *uffd_setup_demand_paging(int uffd_mode, useconds_t delay,
103					   void *hva, uint64_t len,
104					   uint64_t num_readers,
105					   uffd_handler_t handler)
106{
107	struct uffd_desc *uffd_desc;
108	bool is_minor = (uffd_mode == UFFDIO_REGISTER_MODE_MINOR);
109	int uffd;
110	struct uffdio_api uffdio_api;
111	struct uffdio_register uffdio_register;
112	uint64_t expected_ioctls = ((uint64_t) 1) << _UFFDIO_COPY;
113	int ret, i;
114
115	PER_PAGE_DEBUG("Userfaultfd %s mode, faults resolved with %s\n",
116		       is_minor ? "MINOR" : "MISSING",
117		       is_minor ? "UFFDIO_CONINUE" : "UFFDIO_COPY");
118
119	uffd_desc = malloc(sizeof(struct uffd_desc));
120	TEST_ASSERT(uffd_desc, "Failed to malloc uffd descriptor");
121
122	uffd_desc->pipefds = calloc(sizeof(int), num_readers);
123	TEST_ASSERT(uffd_desc->pipefds, "Failed to alloc pipes");
124
125	uffd_desc->readers = calloc(sizeof(pthread_t), num_readers);
126	TEST_ASSERT(uffd_desc->readers, "Failed to alloc reader threads");
127
128	uffd_desc->reader_args = calloc(sizeof(struct uffd_reader_args), num_readers);
129	TEST_ASSERT(uffd_desc->reader_args, "Failed to alloc reader_args");
130
131	uffd_desc->num_readers = num_readers;
132
133	/* In order to get minor faults, prefault via the alias. */
134	if (is_minor)
135		expected_ioctls = ((uint64_t) 1) << _UFFDIO_CONTINUE;
136
137	uffd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
138	TEST_ASSERT(uffd >= 0, "uffd creation failed, errno: %d", errno);
139
140	uffdio_api.api = UFFD_API;
141	uffdio_api.features = 0;
142	TEST_ASSERT(ioctl(uffd, UFFDIO_API, &uffdio_api) != -1,
143		    "ioctl UFFDIO_API failed: %" PRIu64,
144		    (uint64_t)uffdio_api.api);
145
146	uffdio_register.range.start = (uint64_t)hva;
147	uffdio_register.range.len = len;
148	uffdio_register.mode = uffd_mode;
149	TEST_ASSERT(ioctl(uffd, UFFDIO_REGISTER, &uffdio_register) != -1,
150		    "ioctl UFFDIO_REGISTER failed");
151	TEST_ASSERT((uffdio_register.ioctls & expected_ioctls) ==
152		    expected_ioctls, "missing userfaultfd ioctls");
153
154	uffd_desc->uffd = uffd;
155	for (i = 0; i < uffd_desc->num_readers; ++i) {
156		int pipes[2];
157
158		ret = pipe2((int *) &pipes, O_CLOEXEC | O_NONBLOCK);
159		TEST_ASSERT(!ret, "Failed to set up pipefd %i for uffd_desc %p",
160			    i, uffd_desc);
161
162		uffd_desc->pipefds[i] = pipes[1];
163
164		uffd_desc->reader_args[i].uffd_mode = uffd_mode;
165		uffd_desc->reader_args[i].uffd = uffd;
166		uffd_desc->reader_args[i].delay = delay;
167		uffd_desc->reader_args[i].handler = handler;
168		uffd_desc->reader_args[i].pipe = pipes[0];
169
170		pthread_create(&uffd_desc->readers[i], NULL, uffd_handler_thread_fn,
171			       &uffd_desc->reader_args[i]);
172
173		PER_VCPU_DEBUG("Created uffd thread %i for HVA range [%p, %p)\n",
174			       i, hva, hva + len);
175	}
176
177	return uffd_desc;
178}
179
180void uffd_stop_demand_paging(struct uffd_desc *uffd)
181{
182	char c = 0;
183	int i;
184
185	for (i = 0; i < uffd->num_readers; ++i)
186		TEST_ASSERT(write(uffd->pipefds[i], &c, 1) == 1,
187			    "Unable to write to pipefd %i for uffd_desc %p", i, uffd);
188
189	for (i = 0; i < uffd->num_readers; ++i)
190		TEST_ASSERT(!pthread_join(uffd->readers[i], NULL),
191			    "Pthread_join failed on reader %i for uffd_desc %p", i, uffd);
192
193	close(uffd->uffd);
194
195	for (i = 0; i < uffd->num_readers; ++i) {
196		close(uffd->pipefds[i]);
197		close(uffd->reader_args[i].pipe);
198	}
199
200	free(uffd->pipefds);
201	free(uffd->readers);
202	free(uffd->reader_args);
203	free(uffd);
204}
205
206#endif /* __NR_userfaultfd */
207