blob: 950c5c4e4ee3d97cd91ee5454bd36c5e1510d28b [file] [log] [blame]
rjw1f884582022-01-06 17:20:42 +08001/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _VHOST_H
3#define _VHOST_H
4
5#include <linux/eventfd.h>
6#include <linux/vhost.h>
7#include <linux/mm.h>
8#include <linux/mutex.h>
9#include <linux/poll.h>
10#include <linux/file.h>
11#include <linux/uio.h>
12#include <linux/virtio_config.h>
13#include <linux/virtio_ring.h>
14#include <linux/atomic.h>
15
16struct vhost_work;
17typedef void (*vhost_work_fn_t)(struct vhost_work *work);
18
19#define VHOST_WORK_QUEUED 1
20struct vhost_work {
21 struct llist_node node;
22 vhost_work_fn_t fn;
23 wait_queue_head_t done;
24 int flushing;
25 unsigned queue_seq;
26 unsigned done_seq;
27 unsigned long flags;
28};
29
30/* Poll a file (eventfd or socket) */
31/* Note: there's nothing vhost specific about this structure. */
32struct vhost_poll {
33 poll_table table;
34 wait_queue_head_t *wqh;
35 wait_queue_entry_t wait;
36 struct vhost_work work;
37 unsigned long mask;
38 struct vhost_dev *dev;
39};
40
41void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn);
42void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work);
43bool vhost_has_work(struct vhost_dev *dev);
44
45void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
46 unsigned long mask, struct vhost_dev *dev);
47int vhost_poll_start(struct vhost_poll *poll, struct file *file);
48void vhost_poll_stop(struct vhost_poll *poll);
49void vhost_poll_flush(struct vhost_poll *poll);
50void vhost_poll_queue(struct vhost_poll *poll);
51void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work);
52long vhost_vring_ioctl(struct vhost_dev *d, int ioctl, void __user *argp);
53
54struct vhost_log {
55 u64 addr;
56 u64 len;
57};
58
59#define START(node) ((node)->start)
60#define LAST(node) ((node)->last)
61
62struct vhost_umem_node {
63 struct rb_node rb;
64 struct list_head link;
65 __u64 start;
66 __u64 last;
67 __u64 size;
68 __u64 userspace_addr;
69 __u32 perm;
70 __u32 flags_padding;
71 __u64 __subtree_last;
72};
73
74struct vhost_umem {
75 struct rb_root_cached umem_tree;
76 struct list_head umem_list;
77 int numem;
78};
79
80enum vhost_uaddr_type {
81 VHOST_ADDR_DESC = 0,
82 VHOST_ADDR_AVAIL = 1,
83 VHOST_ADDR_USED = 2,
84 VHOST_NUM_ADDRS = 3,
85};
86
87/* The virtqueue structure describes a queue attached to a device. */
88struct vhost_virtqueue {
89 struct vhost_dev *dev;
90
91 /* The actual ring of buffers. */
92 struct mutex mutex;
93 unsigned int num;
94 struct vring_desc __user *desc;
95 struct vring_avail __user *avail;
96 struct vring_used __user *used;
97 const struct vhost_umem_node *meta_iotlb[VHOST_NUM_ADDRS];
98 struct file *kick;
99 struct file *call;
100 struct file *error;
101 struct eventfd_ctx *call_ctx;
102 struct eventfd_ctx *error_ctx;
103 struct eventfd_ctx *log_ctx;
104
105 struct vhost_poll poll;
106
107 /* The routine to call when the Guest pings us, or timeout. */
108 vhost_work_fn_t handle_kick;
109
110 /* Last available index we saw. */
111 u16 last_avail_idx;
112
113 /* Caches available index value from user. */
114 u16 avail_idx;
115
116 /* Last index we used. */
117 u16 last_used_idx;
118
119 /* Used flags */
120 u16 used_flags;
121
122 /* Last used index value we have signalled on */
123 u16 signalled_used;
124
125 /* Last used index value we have signalled on */
126 bool signalled_used_valid;
127
128 /* Log writes to used structure. */
129 bool log_used;
130 u64 log_addr;
131
132 struct iovec iov[UIO_MAXIOV];
133 struct iovec iotlb_iov[64];
134 struct iovec *indirect;
135 struct vring_used_elem *heads;
136 /* Protected by virtqueue mutex. */
137 struct vhost_umem *umem;
138 struct vhost_umem *iotlb;
139 void *private_data;
140 u64 acked_features;
141 /* Log write descriptors */
142 void __user *log_base;
143 struct vhost_log *log;
144
145 /* Ring endianness. Defaults to legacy native endianness.
146 * Set to true when starting a modern virtio device. */
147 bool is_le;
148#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
149 /* Ring endianness requested by userspace for cross-endian support. */
150 bool user_be;
151#endif
152 u32 busyloop_timeout;
153};
154
155struct vhost_msg_node {
156 struct vhost_msg msg;
157 struct vhost_virtqueue *vq;
158 struct list_head node;
159};
160
161struct vhost_dev {
162 struct mm_struct *mm;
163 struct mutex mutex;
164 struct vhost_virtqueue **vqs;
165 int nvqs;
166 struct file *log_file;
167 struct eventfd_ctx *log_ctx;
168 struct llist_head work_list;
169 struct task_struct *worker;
170 struct vhost_umem *umem;
171 struct vhost_umem *iotlb;
172 spinlock_t iotlb_lock;
173 struct list_head read_list;
174 struct list_head pending_list;
175 wait_queue_head_t wait;
176 int weight;
177 int byte_weight;
178};
179
180bool vhost_exceeds_weight(struct vhost_virtqueue *vq, int pkts, int total_len);
181void vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue **vqs,
182 int nvqs, int weight, int byte_weight);
183long vhost_dev_set_owner(struct vhost_dev *dev);
184bool vhost_dev_has_owner(struct vhost_dev *dev);
185long vhost_dev_check_owner(struct vhost_dev *);
186struct vhost_umem *vhost_dev_reset_owner_prepare(void);
187void vhost_dev_reset_owner(struct vhost_dev *, struct vhost_umem *);
188void vhost_dev_cleanup(struct vhost_dev *, bool locked);
189void vhost_dev_stop(struct vhost_dev *);
190long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, void __user *argp);
191long vhost_vring_ioctl(struct vhost_dev *d, int ioctl, void __user *argp);
192int vhost_vq_access_ok(struct vhost_virtqueue *vq);
193int vhost_log_access_ok(struct vhost_dev *);
194
195int vhost_get_vq_desc(struct vhost_virtqueue *,
196 struct iovec iov[], unsigned int iov_count,
197 unsigned int *out_num, unsigned int *in_num,
198 struct vhost_log *log, unsigned int *log_num);
199void vhost_discard_vq_desc(struct vhost_virtqueue *, int n);
200
201int vhost_vq_init_access(struct vhost_virtqueue *);
202int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
203int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
204 unsigned count);
205void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *,
206 unsigned int id, int len);
207void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *,
208 struct vring_used_elem *heads, unsigned count);
209void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *);
210void vhost_disable_notify(struct vhost_dev *, struct vhost_virtqueue *);
211bool vhost_vq_avail_empty(struct vhost_dev *, struct vhost_virtqueue *);
212bool vhost_enable_notify(struct vhost_dev *, struct vhost_virtqueue *);
213
214int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
215 unsigned int log_num, u64 len,
216 struct iovec *iov, int count);
217int vq_iotlb_prefetch(struct vhost_virtqueue *vq);
218
219struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type);
220void vhost_enqueue_msg(struct vhost_dev *dev,
221 struct list_head *head,
222 struct vhost_msg_node *node);
223struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
224 struct list_head *head);
225unsigned int vhost_chr_poll(struct file *file, struct vhost_dev *dev,
226 poll_table *wait);
227ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
228 int noblock);
229ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
230 struct iov_iter *from);
231int vhost_init_device_iotlb(struct vhost_dev *d, bool enabled);
232
233#define vq_err(vq, fmt, ...) do { \
234 pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \
235 if ((vq)->error_ctx) \
236 eventfd_signal((vq)->error_ctx, 1);\
237 } while (0)
238
239enum {
240 VHOST_FEATURES = (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) |
241 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) |
242 (1ULL << VIRTIO_RING_F_EVENT_IDX) |
243 (1ULL << VHOST_F_LOG_ALL) |
244 (1ULL << VIRTIO_F_ANY_LAYOUT) |
245 (1ULL << VIRTIO_F_VERSION_1)
246};
247
248static inline bool vhost_has_feature(struct vhost_virtqueue *vq, int bit)
249{
250 return vq->acked_features & (1ULL << bit);
251}
252
253#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
254static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
255{
256 return vq->is_le;
257}
258#else
259static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
260{
261 return virtio_legacy_is_little_endian() || vq->is_le;
262}
263#endif
264
265/* Memory accessors */
266static inline u16 vhost16_to_cpu(struct vhost_virtqueue *vq, __virtio16 val)
267{
268 return __virtio16_to_cpu(vhost_is_little_endian(vq), val);
269}
270
271static inline __virtio16 cpu_to_vhost16(struct vhost_virtqueue *vq, u16 val)
272{
273 return __cpu_to_virtio16(vhost_is_little_endian(vq), val);
274}
275
276static inline u32 vhost32_to_cpu(struct vhost_virtqueue *vq, __virtio32 val)
277{
278 return __virtio32_to_cpu(vhost_is_little_endian(vq), val);
279}
280
281static inline __virtio32 cpu_to_vhost32(struct vhost_virtqueue *vq, u32 val)
282{
283 return __cpu_to_virtio32(vhost_is_little_endian(vq), val);
284}
285
286static inline u64 vhost64_to_cpu(struct vhost_virtqueue *vq, __virtio64 val)
287{
288 return __virtio64_to_cpu(vhost_is_little_endian(vq), val);
289}
290
291static inline __virtio64 cpu_to_vhost64(struct vhost_virtqueue *vq, u64 val)
292{
293 return __cpu_to_virtio64(vhost_is_little_endian(vq), val);
294}
295#endif