xref: /linux/drivers/vfio/virqfd.c (revision 0e685c3e7158d35626d6d76b9f859eae806d87fa)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * VFIO generic eventfd code for IRQFD support.
4  * Derived from drivers/vfio/pci/vfio_pci_intrs.c
5  *
6  * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
7  *     Author: Alex Williamson <alex.williamson@redhat.com>
8  */
9 
10 #include <linux/vfio.h>
11 #include <linux/eventfd.h>
12 #include <linux/file.h>
13 #include <linux/module.h>
14 #include <linux/slab.h>
15 #include "vfio.h"
16 
17 static struct workqueue_struct *vfio_irqfd_cleanup_wq;
18 static DEFINE_SPINLOCK(virqfd_lock);
19 
20 int __init vfio_virqfd_init(void)
21 {
22 	vfio_irqfd_cleanup_wq =
23 		create_singlethread_workqueue("vfio-irqfd-cleanup");
24 	if (!vfio_irqfd_cleanup_wq)
25 		return -ENOMEM;
26 
27 	return 0;
28 }
29 
30 void vfio_virqfd_exit(void)
31 {
32 	destroy_workqueue(vfio_irqfd_cleanup_wq);
33 }
34 
35 static void virqfd_deactivate(struct virqfd *virqfd)
36 {
37 	queue_work(vfio_irqfd_cleanup_wq, &virqfd->shutdown);
38 }
39 
40 static int virqfd_wakeup(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
41 {
42 	struct virqfd *virqfd = container_of(wait, struct virqfd, wait);
43 	__poll_t flags = key_to_poll(key);
44 
45 	if (flags & EPOLLIN) {
46 		u64 cnt;
47 		eventfd_ctx_do_read(virqfd->eventfd, &cnt);
48 
49 		/* An event has been signaled, call function */
50 		if ((!virqfd->handler ||
51 		     virqfd->handler(virqfd->opaque, virqfd->data)) &&
52 		    virqfd->thread)
53 			schedule_work(&virqfd->inject);
54 	}
55 
56 	if (flags & EPOLLHUP) {
57 		unsigned long flags;
58 		spin_lock_irqsave(&virqfd_lock, flags);
59 
60 		/*
61 		 * The eventfd is closing, if the virqfd has not yet been
62 		 * queued for release, as determined by testing whether the
63 		 * virqfd pointer to it is still valid, queue it now.  As
64 		 * with kvm irqfds, we know we won't race against the virqfd
65 		 * going away because we hold the lock to get here.
66 		 */
67 		if (*(virqfd->pvirqfd) == virqfd) {
68 			*(virqfd->pvirqfd) = NULL;
69 			virqfd_deactivate(virqfd);
70 		}
71 
72 		spin_unlock_irqrestore(&virqfd_lock, flags);
73 	}
74 
75 	return 0;
76 }
77 
78 static void virqfd_ptable_queue_proc(struct file *file,
79 				     wait_queue_head_t *wqh, poll_table *pt)
80 {
81 	struct virqfd *virqfd = container_of(pt, struct virqfd, pt);
82 	add_wait_queue(wqh, &virqfd->wait);
83 }
84 
85 static void virqfd_shutdown(struct work_struct *work)
86 {
87 	struct virqfd *virqfd = container_of(work, struct virqfd, shutdown);
88 	u64 cnt;
89 
90 	eventfd_ctx_remove_wait_queue(virqfd->eventfd, &virqfd->wait, &cnt);
91 	flush_work(&virqfd->inject);
92 	eventfd_ctx_put(virqfd->eventfd);
93 
94 	kfree(virqfd);
95 }
96 
97 static void virqfd_inject(struct work_struct *work)
98 {
99 	struct virqfd *virqfd = container_of(work, struct virqfd, inject);
100 	if (virqfd->thread)
101 		virqfd->thread(virqfd->opaque, virqfd->data);
102 }
103 
104 int vfio_virqfd_enable(void *opaque,
105 		       int (*handler)(void *, void *),
106 		       void (*thread)(void *, void *),
107 		       void *data, struct virqfd **pvirqfd, int fd)
108 {
109 	struct fd irqfd;
110 	struct eventfd_ctx *ctx;
111 	struct virqfd *virqfd;
112 	int ret = 0;
113 	__poll_t events;
114 
115 	virqfd = kzalloc(sizeof(*virqfd), GFP_KERNEL_ACCOUNT);
116 	if (!virqfd)
117 		return -ENOMEM;
118 
119 	virqfd->pvirqfd = pvirqfd;
120 	virqfd->opaque = opaque;
121 	virqfd->handler = handler;
122 	virqfd->thread = thread;
123 	virqfd->data = data;
124 
125 	INIT_WORK(&virqfd->shutdown, virqfd_shutdown);
126 	INIT_WORK(&virqfd->inject, virqfd_inject);
127 
128 	irqfd = fdget(fd);
129 	if (!irqfd.file) {
130 		ret = -EBADF;
131 		goto err_fd;
132 	}
133 
134 	ctx = eventfd_ctx_fileget(irqfd.file);
135 	if (IS_ERR(ctx)) {
136 		ret = PTR_ERR(ctx);
137 		goto err_ctx;
138 	}
139 
140 	virqfd->eventfd = ctx;
141 
142 	/*
143 	 * virqfds can be released by closing the eventfd or directly
144 	 * through ioctl.  These are both done through a workqueue, so
145 	 * we update the pointer to the virqfd under lock to avoid
146 	 * pushing multiple jobs to release the same virqfd.
147 	 */
148 	spin_lock_irq(&virqfd_lock);
149 
150 	if (*pvirqfd) {
151 		spin_unlock_irq(&virqfd_lock);
152 		ret = -EBUSY;
153 		goto err_busy;
154 	}
155 	*pvirqfd = virqfd;
156 
157 	spin_unlock_irq(&virqfd_lock);
158 
159 	/*
160 	 * Install our own custom wake-up handling so we are notified via
161 	 * a callback whenever someone signals the underlying eventfd.
162 	 */
163 	init_waitqueue_func_entry(&virqfd->wait, virqfd_wakeup);
164 	init_poll_funcptr(&virqfd->pt, virqfd_ptable_queue_proc);
165 
166 	events = vfs_poll(irqfd.file, &virqfd->pt);
167 
168 	/*
169 	 * Check if there was an event already pending on the eventfd
170 	 * before we registered and trigger it as if we didn't miss it.
171 	 */
172 	if (events & EPOLLIN) {
173 		if ((!handler || handler(opaque, data)) && thread)
174 			schedule_work(&virqfd->inject);
175 	}
176 
177 	/*
178 	 * Do not drop the file until the irqfd is fully initialized,
179 	 * otherwise we might race against the EPOLLHUP.
180 	 */
181 	fdput(irqfd);
182 
183 	return 0;
184 err_busy:
185 	eventfd_ctx_put(ctx);
186 err_ctx:
187 	fdput(irqfd);
188 err_fd:
189 	kfree(virqfd);
190 
191 	return ret;
192 }
193 EXPORT_SYMBOL_GPL(vfio_virqfd_enable);
194 
195 void vfio_virqfd_disable(struct virqfd **pvirqfd)
196 {
197 	unsigned long flags;
198 
199 	spin_lock_irqsave(&virqfd_lock, flags);
200 
201 	if (*pvirqfd) {
202 		virqfd_deactivate(*pvirqfd);
203 		*pvirqfd = NULL;
204 	}
205 
206 	spin_unlock_irqrestore(&virqfd_lock, flags);
207 
208 	/*
209 	 * Block until we know all outstanding shutdown jobs have completed.
210 	 * Even if we don't queue the job, flush the wq to be sure it's
211 	 * been released.
212 	 */
213 	flush_workqueue(vfio_irqfd_cleanup_wq);
214 }
215 EXPORT_SYMBOL_GPL(vfio_virqfd_disable);
216