Loading...
1/* 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#include <linux/vhost_iotlb.h>
16#include <linux/irqbypass.h>
17
18struct vhost_work;
19struct vhost_task;
20typedef void (*vhost_work_fn_t)(struct vhost_work *work);
21
22#define VHOST_WORK_QUEUED 1
23struct vhost_work {
24 struct llist_node node;
25 vhost_work_fn_t fn;
26 unsigned long flags;
27};
28
29struct vhost_worker {
30 struct vhost_task *vtsk;
31 struct vhost_dev *dev;
32 /* Used to serialize device wide flushing with worker swapping. */
33 struct mutex mutex;
34 struct llist_head work_list;
35 u64 kcov_handle;
36 u32 id;
37 int attachment_cnt;
38 bool killed;
39};
40
41/* Poll a file (eventfd or socket) */
42/* Note: there's nothing vhost specific about this structure. */
43struct vhost_poll {
44 poll_table table;
45 wait_queue_head_t *wqh;
46 wait_queue_entry_t wait;
47 struct vhost_work work;
48 __poll_t mask;
49 struct vhost_dev *dev;
50 struct vhost_virtqueue *vq;
51};
52
53void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
54 __poll_t mask, struct vhost_dev *dev,
55 struct vhost_virtqueue *vq);
56int vhost_poll_start(struct vhost_poll *poll, struct file *file);
57void vhost_poll_stop(struct vhost_poll *poll);
58void vhost_poll_queue(struct vhost_poll *poll);
59
60void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn);
61void vhost_dev_flush(struct vhost_dev *dev);
62
63struct vhost_log {
64 u64 addr;
65 u64 len;
66};
67
68enum vhost_uaddr_type {
69 VHOST_ADDR_DESC = 0,
70 VHOST_ADDR_AVAIL = 1,
71 VHOST_ADDR_USED = 2,
72 VHOST_NUM_ADDRS = 3,
73};
74
75struct vhost_vring_call {
76 struct eventfd_ctx *ctx;
77 struct irq_bypass_producer producer;
78};
79
80/* The virtqueue structure describes a queue attached to a device. */
81struct vhost_virtqueue {
82 struct vhost_dev *dev;
83 struct vhost_worker __rcu *worker;
84
85 /* The actual ring of buffers. */
86 struct mutex mutex;
87 unsigned int num;
88 vring_desc_t __user *desc;
89 vring_avail_t __user *avail;
90 vring_used_t __user *used;
91 const struct vhost_iotlb_map *meta_iotlb[VHOST_NUM_ADDRS];
92 struct file *kick;
93 struct vhost_vring_call call_ctx;
94 struct eventfd_ctx *error_ctx;
95 struct eventfd_ctx *log_ctx;
96
97 struct vhost_poll poll;
98
99 /* The routine to call when the Guest pings us, or timeout. */
100 vhost_work_fn_t handle_kick;
101
102 /* Last available index we saw.
103 * Values are limited to 0x7fff, and the high bit is used as
104 * a wrap counter when using VIRTIO_F_RING_PACKED. */
105 u16 last_avail_idx;
106
107 /* Caches available index value from user. */
108 u16 avail_idx;
109
110 /* Last index we used.
111 * Values are limited to 0x7fff, and the high bit is used as
112 * a wrap counter when using VIRTIO_F_RING_PACKED. */
113 u16 last_used_idx;
114
115 /* Used flags */
116 u16 used_flags;
117
118 /* Last used index value we have signalled on */
119 u16 signalled_used;
120
121 /* Last used index value we have signalled on */
122 bool signalled_used_valid;
123
124 /* Log writes to used structure. */
125 bool log_used;
126 u64 log_addr;
127
128 struct iovec iov[UIO_MAXIOV];
129 struct iovec iotlb_iov[64];
130 struct iovec *indirect;
131 struct vring_used_elem *heads;
132 /* Protected by virtqueue mutex. */
133 struct vhost_iotlb *umem;
134 struct vhost_iotlb *iotlb;
135 void *private_data;
136 u64 acked_features;
137 u64 acked_backend_features;
138 /* Log write descriptors */
139 void __user *log_base;
140 struct vhost_log *log;
141 struct iovec log_iov[64];
142
143 /* Ring endianness. Defaults to legacy native endianness.
144 * Set to true when starting a modern virtio device. */
145 bool is_le;
146#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
147 /* Ring endianness requested by userspace for cross-endian support. */
148 bool user_be;
149#endif
150 u32 busyloop_timeout;
151};
152
153struct vhost_msg_node {
154 union {
155 struct vhost_msg msg;
156 struct vhost_msg_v2 msg_v2;
157 };
158 struct vhost_virtqueue *vq;
159 struct list_head node;
160};
161
162struct vhost_dev {
163 struct mm_struct *mm;
164 struct mutex mutex;
165 struct vhost_virtqueue **vqs;
166 int nvqs;
167 struct eventfd_ctx *log_ctx;
168 struct vhost_iotlb *umem;
169 struct vhost_iotlb *iotlb;
170 spinlock_t iotlb_lock;
171 struct list_head read_list;
172 struct list_head pending_list;
173 wait_queue_head_t wait;
174 int iov_limit;
175 int weight;
176 int byte_weight;
177 struct xarray worker_xa;
178 bool use_worker;
179 int (*msg_handler)(struct vhost_dev *dev, u32 asid,
180 struct vhost_iotlb_msg *msg);
181};
182
183bool vhost_exceeds_weight(struct vhost_virtqueue *vq, int pkts, int total_len);
184void vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue **vqs,
185 int nvqs, int iov_limit, int weight, int byte_weight,
186 bool use_worker,
187 int (*msg_handler)(struct vhost_dev *dev, u32 asid,
188 struct vhost_iotlb_msg *msg));
189long vhost_dev_set_owner(struct vhost_dev *dev);
190bool vhost_dev_has_owner(struct vhost_dev *dev);
191long vhost_dev_check_owner(struct vhost_dev *);
192struct vhost_iotlb *vhost_dev_reset_owner_prepare(void);
193void vhost_dev_reset_owner(struct vhost_dev *dev, struct vhost_iotlb *iotlb);
194void vhost_dev_cleanup(struct vhost_dev *);
195void vhost_dev_stop(struct vhost_dev *);
196long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, void __user *argp);
197long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *argp);
198long vhost_worker_ioctl(struct vhost_dev *dev, unsigned int ioctl,
199 void __user *argp);
200bool vhost_vq_access_ok(struct vhost_virtqueue *vq);
201bool vhost_log_access_ok(struct vhost_dev *);
202void vhost_clear_msg(struct vhost_dev *dev);
203
204int vhost_get_vq_desc(struct vhost_virtqueue *,
205 struct iovec iov[], unsigned int iov_size,
206 unsigned int *out_num, unsigned int *in_num,
207 struct vhost_log *log, unsigned int *log_num);
208void vhost_discard_vq_desc(struct vhost_virtqueue *, int n);
209
210bool vhost_vq_work_queue(struct vhost_virtqueue *vq, struct vhost_work *work);
211bool vhost_vq_has_work(struct vhost_virtqueue *vq);
212bool vhost_vq_is_setup(struct vhost_virtqueue *vq);
213int vhost_vq_init_access(struct vhost_virtqueue *);
214int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
215int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
216 unsigned count);
217void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *,
218 unsigned int id, int len);
219void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *,
220 struct vring_used_elem *heads, unsigned count);
221void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *);
222void vhost_disable_notify(struct vhost_dev *, struct vhost_virtqueue *);
223bool vhost_vq_avail_empty(struct vhost_dev *, struct vhost_virtqueue *);
224bool vhost_enable_notify(struct vhost_dev *, struct vhost_virtqueue *);
225
226int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
227 unsigned int log_num, u64 len,
228 struct iovec *iov, int count);
229int vq_meta_prefetch(struct vhost_virtqueue *vq);
230
231struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type);
232void vhost_enqueue_msg(struct vhost_dev *dev,
233 struct list_head *head,
234 struct vhost_msg_node *node);
235struct vhost_msg_node *vhost_dequeue_msg(struct vhost_dev *dev,
236 struct list_head *head);
237void vhost_set_backend_features(struct vhost_dev *dev, u64 features);
238
239__poll_t vhost_chr_poll(struct file *file, struct vhost_dev *dev,
240 poll_table *wait);
241ssize_t vhost_chr_read_iter(struct vhost_dev *dev, struct iov_iter *to,
242 int noblock);
243ssize_t vhost_chr_write_iter(struct vhost_dev *dev,
244 struct iov_iter *from);
245int vhost_init_device_iotlb(struct vhost_dev *d);
246
247void vhost_iotlb_map_free(struct vhost_iotlb *iotlb,
248 struct vhost_iotlb_map *map);
249
250#define vq_err(vq, fmt, ...) do { \
251 pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \
252 if ((vq)->error_ctx) \
253 eventfd_signal((vq)->error_ctx);\
254 } while (0)
255
256enum {
257 VHOST_FEATURES = (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) |
258 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) |
259 (1ULL << VIRTIO_RING_F_EVENT_IDX) |
260 (1ULL << VHOST_F_LOG_ALL) |
261 (1ULL << VIRTIO_F_ANY_LAYOUT) |
262 (1ULL << VIRTIO_F_VERSION_1)
263};
264
265/**
266 * vhost_vq_set_backend - Set backend.
267 *
268 * @vq Virtqueue.
269 * @private_data The private data.
270 *
271 * Context: Need to call with vq->mutex acquired.
272 */
273static inline void vhost_vq_set_backend(struct vhost_virtqueue *vq,
274 void *private_data)
275{
276 vq->private_data = private_data;
277}
278
279/**
280 * vhost_vq_get_backend - Get backend.
281 *
282 * @vq Virtqueue.
283 *
284 * Context: Need to call with vq->mutex acquired.
285 * Return: Private data previously set with vhost_vq_set_backend.
286 */
287static inline void *vhost_vq_get_backend(struct vhost_virtqueue *vq)
288{
289 return vq->private_data;
290}
291
292static inline bool vhost_has_feature(struct vhost_virtqueue *vq, int bit)
293{
294 return vq->acked_features & (1ULL << bit);
295}
296
297static inline bool vhost_backend_has_feature(struct vhost_virtqueue *vq, int bit)
298{
299 return vq->acked_backend_features & (1ULL << bit);
300}
301
302#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
303static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
304{
305 return vq->is_le;
306}
307#else
308static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
309{
310 return virtio_legacy_is_little_endian() || vq->is_le;
311}
312#endif
313
314/* Memory accessors */
315static inline u16 vhost16_to_cpu(struct vhost_virtqueue *vq, __virtio16 val)
316{
317 return __virtio16_to_cpu(vhost_is_little_endian(vq), val);
318}
319
320static inline __virtio16 cpu_to_vhost16(struct vhost_virtqueue *vq, u16 val)
321{
322 return __cpu_to_virtio16(vhost_is_little_endian(vq), val);
323}
324
325static inline u32 vhost32_to_cpu(struct vhost_virtqueue *vq, __virtio32 val)
326{
327 return __virtio32_to_cpu(vhost_is_little_endian(vq), val);
328}
329
330static inline __virtio32 cpu_to_vhost32(struct vhost_virtqueue *vq, u32 val)
331{
332 return __cpu_to_virtio32(vhost_is_little_endian(vq), val);
333}
334
335static inline u64 vhost64_to_cpu(struct vhost_virtqueue *vq, __virtio64 val)
336{
337 return __virtio64_to_cpu(vhost_is_little_endian(vq), val);
338}
339
340static inline __virtio64 cpu_to_vhost64(struct vhost_virtqueue *vq, u64 val)
341{
342 return __cpu_to_virtio64(vhost_is_little_endian(vq), val);
343}
344#endif
1#ifndef _VHOST_H
2#define _VHOST_H
3
4#include <linux/eventfd.h>
5#include <linux/vhost.h>
6#include <linux/mm.h>
7#include <linux/mutex.h>
8#include <linux/poll.h>
9#include <linux/file.h>
10#include <linux/uio.h>
11#include <linux/virtio_config.h>
12#include <linux/virtio_ring.h>
13#include <linux/atomic.h>
14
15struct vhost_work;
16typedef void (*vhost_work_fn_t)(struct vhost_work *work);
17
18struct vhost_work {
19 struct list_head node;
20 vhost_work_fn_t fn;
21 wait_queue_head_t done;
22 int flushing;
23 unsigned queue_seq;
24 unsigned done_seq;
25};
26
27/* Poll a file (eventfd or socket) */
28/* Note: there's nothing vhost specific about this structure. */
29struct vhost_poll {
30 poll_table table;
31 wait_queue_head_t *wqh;
32 wait_queue_t wait;
33 struct vhost_work work;
34 unsigned long mask;
35 struct vhost_dev *dev;
36};
37
38void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn);
39void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work);
40bool vhost_has_work(struct vhost_dev *dev);
41
42void vhost_poll_init(struct vhost_poll *poll, vhost_work_fn_t fn,
43 unsigned long mask, struct vhost_dev *dev);
44int vhost_poll_start(struct vhost_poll *poll, struct file *file);
45void vhost_poll_stop(struct vhost_poll *poll);
46void vhost_poll_flush(struct vhost_poll *poll);
47void vhost_poll_queue(struct vhost_poll *poll);
48void vhost_work_flush(struct vhost_dev *dev, struct vhost_work *work);
49long vhost_vring_ioctl(struct vhost_dev *d, int ioctl, void __user *argp);
50
51struct vhost_log {
52 u64 addr;
53 u64 len;
54};
55
56/* The virtqueue structure describes a queue attached to a device. */
57struct vhost_virtqueue {
58 struct vhost_dev *dev;
59
60 /* The actual ring of buffers. */
61 struct mutex mutex;
62 unsigned int num;
63 struct vring_desc __user *desc;
64 struct vring_avail __user *avail;
65 struct vring_used __user *used;
66 struct file *kick;
67 struct file *call;
68 struct file *error;
69 struct eventfd_ctx *call_ctx;
70 struct eventfd_ctx *error_ctx;
71 struct eventfd_ctx *log_ctx;
72
73 struct vhost_poll poll;
74
75 /* The routine to call when the Guest pings us, or timeout. */
76 vhost_work_fn_t handle_kick;
77
78 /* Last available index we saw. */
79 u16 last_avail_idx;
80
81 /* Caches available index value from user. */
82 u16 avail_idx;
83
84 /* Last index we used. */
85 u16 last_used_idx;
86
87 /* Used flags */
88 u16 used_flags;
89
90 /* Last used index value we have signalled on */
91 u16 signalled_used;
92
93 /* Last used index value we have signalled on */
94 bool signalled_used_valid;
95
96 /* Log writes to used structure. */
97 bool log_used;
98 u64 log_addr;
99
100 struct iovec iov[UIO_MAXIOV];
101 struct iovec *indirect;
102 struct vring_used_elem *heads;
103 /* Protected by virtqueue mutex. */
104 struct vhost_memory *memory;
105 void *private_data;
106 u64 acked_features;
107 /* Log write descriptors */
108 void __user *log_base;
109 struct vhost_log *log;
110
111 /* Ring endianness. Defaults to legacy native endianness.
112 * Set to true when starting a modern virtio device. */
113 bool is_le;
114#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
115 /* Ring endianness requested by userspace for cross-endian support. */
116 bool user_be;
117#endif
118 u32 busyloop_timeout;
119};
120
121struct vhost_dev {
122 struct vhost_memory *memory;
123 struct mm_struct *mm;
124 struct mutex mutex;
125 struct vhost_virtqueue **vqs;
126 int nvqs;
127 struct file *log_file;
128 struct eventfd_ctx *log_ctx;
129 spinlock_t work_lock;
130 struct list_head work_list;
131 struct task_struct *worker;
132};
133
134void vhost_dev_init(struct vhost_dev *, struct vhost_virtqueue **vqs, int nvqs);
135long vhost_dev_set_owner(struct vhost_dev *dev);
136bool vhost_dev_has_owner(struct vhost_dev *dev);
137long vhost_dev_check_owner(struct vhost_dev *);
138struct vhost_memory *vhost_dev_reset_owner_prepare(void);
139void vhost_dev_reset_owner(struct vhost_dev *, struct vhost_memory *);
140void vhost_dev_cleanup(struct vhost_dev *, bool locked);
141void vhost_dev_stop(struct vhost_dev *);
142long vhost_dev_ioctl(struct vhost_dev *, unsigned int ioctl, void __user *argp);
143long vhost_vring_ioctl(struct vhost_dev *d, int ioctl, void __user *argp);
144int vhost_vq_access_ok(struct vhost_virtqueue *vq);
145int vhost_log_access_ok(struct vhost_dev *);
146
147int vhost_get_vq_desc(struct vhost_virtqueue *,
148 struct iovec iov[], unsigned int iov_count,
149 unsigned int *out_num, unsigned int *in_num,
150 struct vhost_log *log, unsigned int *log_num);
151void vhost_discard_vq_desc(struct vhost_virtqueue *, int n);
152
153int vhost_vq_init_access(struct vhost_virtqueue *);
154int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
155int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
156 unsigned count);
157void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *,
158 unsigned int id, int len);
159void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *,
160 struct vring_used_elem *heads, unsigned count);
161void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *);
162void vhost_disable_notify(struct vhost_dev *, struct vhost_virtqueue *);
163bool vhost_vq_avail_empty(struct vhost_dev *, struct vhost_virtqueue *);
164bool vhost_enable_notify(struct vhost_dev *, struct vhost_virtqueue *);
165
166int vhost_log_write(struct vhost_virtqueue *vq, struct vhost_log *log,
167 unsigned int log_num, u64 len);
168
169#define vq_err(vq, fmt, ...) do { \
170 pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \
171 if ((vq)->error_ctx) \
172 eventfd_signal((vq)->error_ctx, 1);\
173 } while (0)
174
175enum {
176 VHOST_FEATURES = (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) |
177 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) |
178 (1ULL << VIRTIO_RING_F_EVENT_IDX) |
179 (1ULL << VHOST_F_LOG_ALL) |
180 (1ULL << VIRTIO_F_ANY_LAYOUT) |
181 (1ULL << VIRTIO_F_VERSION_1)
182};
183
184static inline bool vhost_has_feature(struct vhost_virtqueue *vq, int bit)
185{
186 return vq->acked_features & (1ULL << bit);
187}
188
189#ifdef CONFIG_VHOST_CROSS_ENDIAN_LEGACY
190static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
191{
192 return vq->is_le;
193}
194#else
195static inline bool vhost_is_little_endian(struct vhost_virtqueue *vq)
196{
197 return virtio_legacy_is_little_endian() || vq->is_le;
198}
199#endif
200
201/* Memory accessors */
202static inline u16 vhost16_to_cpu(struct vhost_virtqueue *vq, __virtio16 val)
203{
204 return __virtio16_to_cpu(vhost_is_little_endian(vq), val);
205}
206
207static inline __virtio16 cpu_to_vhost16(struct vhost_virtqueue *vq, u16 val)
208{
209 return __cpu_to_virtio16(vhost_is_little_endian(vq), val);
210}
211
212static inline u32 vhost32_to_cpu(struct vhost_virtqueue *vq, __virtio32 val)
213{
214 return __virtio32_to_cpu(vhost_is_little_endian(vq), val);
215}
216
217static inline __virtio32 cpu_to_vhost32(struct vhost_virtqueue *vq, u32 val)
218{
219 return __cpu_to_virtio32(vhost_is_little_endian(vq), val);
220}
221
222static inline u64 vhost64_to_cpu(struct vhost_virtqueue *vq, __virtio64 val)
223{
224 return __virtio64_to_cpu(vhost_is_little_endian(vq), val);
225}
226
227static inline __virtio64 cpu_to_vhost64(struct vhost_virtqueue *vq, u64 val)
228{
229 return __cpu_to_virtio64(vhost_is_little_endian(vq), val);
230}
231#endif