Commit | Line | Data |
---|---|---|
0a8a69dd RR |
1 | /* Virtio ring implementation. |
2 | * | |
3 | * Copyright 2007 Rusty Russell IBM Corporation | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
16 | * along with this program; if not, write to the Free Software | |
17 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
18 | */ | |
19 | #include <linux/virtio.h> | |
20 | #include <linux/virtio_ring.h> | |
21 | #include <linux/device.h> | |
22 | ||
23 | #ifdef DEBUG | |
24 | /* For development, we want to crash whenever the ring is screwed. */ | |
25 | #define BAD_RING(vq, fmt...) \ | |
26 | do { dev_err(&vq->vq.vdev->dev, fmt); BUG(); } while(0) | |
27 | #define START_USE(vq) \ | |
28 | do { if ((vq)->in_use) panic("in_use = %i\n", (vq)->in_use); (vq)->in_use = __LINE__; mb(); } while(0) | |
29 | #define END_USE(vq) \ | |
30 | do { BUG_ON(!(vq)->in_use); (vq)->in_use = 0; mb(); } while(0) | |
31 | #else | |
32 | #define BAD_RING(vq, fmt...) \ | |
33 | do { dev_err(&vq->vq.vdev->dev, fmt); (vq)->broken = true; } while(0) | |
34 | #define START_USE(vq) | |
35 | #define END_USE(vq) | |
36 | #endif | |
37 | ||
38 | struct vring_virtqueue | |
39 | { | |
40 | struct virtqueue vq; | |
41 | ||
42 | /* Actual memory layout for this queue */ | |
43 | struct vring vring; | |
44 | ||
45 | /* Other side has made a mess, don't try any more. */ | |
46 | bool broken; | |
47 | ||
48 | /* Number of free buffers */ | |
49 | unsigned int num_free; | |
50 | /* Head of free buffer list. */ | |
51 | unsigned int free_head; | |
52 | /* Number we've added since last sync. */ | |
53 | unsigned int num_added; | |
54 | ||
55 | /* Last used index we've seen. */ | |
1bc4953e | 56 | u16 last_used_idx; |
0a8a69dd RR |
57 | |
58 | /* How to notify other side. FIXME: commonalize hcalls! */ | |
59 | void (*notify)(struct virtqueue *vq); | |
60 | ||
61 | #ifdef DEBUG | |
62 | /* They're supposed to lock for us. */ | |
63 | unsigned int in_use; | |
64 | #endif | |
65 | ||
66 | /* Tokens for callbacks. */ | |
67 | void *data[]; | |
68 | }; | |
69 | ||
70 | #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq) | |
71 | ||
72 | static int vring_add_buf(struct virtqueue *_vq, | |
73 | struct scatterlist sg[], | |
74 | unsigned int out, | |
75 | unsigned int in, | |
76 | void *data) | |
77 | { | |
78 | struct vring_virtqueue *vq = to_vvq(_vq); | |
79 | unsigned int i, avail, head, uninitialized_var(prev); | |
80 | ||
81 | BUG_ON(data == NULL); | |
82 | BUG_ON(out + in > vq->vring.num); | |
83 | BUG_ON(out + in == 0); | |
84 | ||
85 | START_USE(vq); | |
86 | ||
87 | if (vq->num_free < out + in) { | |
88 | pr_debug("Can't add buf len %i - avail = %i\n", | |
89 | out + in, vq->num_free); | |
426e3e0a RR |
90 | /* We notify *even if* VRING_USED_F_NO_NOTIFY is set here. */ |
91 | vq->notify(&vq->vq); | |
0a8a69dd RR |
92 | END_USE(vq); |
93 | return -ENOSPC; | |
94 | } | |
95 | ||
96 | /* We're about to use some buffers from the free list. */ | |
97 | vq->num_free -= out + in; | |
98 | ||
99 | head = vq->free_head; | |
100 | for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) { | |
101 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT; | |
15f9c890 | 102 | vq->vring.desc[i].addr = sg_phys(sg); |
0a8a69dd RR |
103 | vq->vring.desc[i].len = sg->length; |
104 | prev = i; | |
105 | sg++; | |
106 | } | |
107 | for (; in; i = vq->vring.desc[i].next, in--) { | |
108 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; | |
15f9c890 | 109 | vq->vring.desc[i].addr = sg_phys(sg); |
0a8a69dd RR |
110 | vq->vring.desc[i].len = sg->length; |
111 | prev = i; | |
112 | sg++; | |
113 | } | |
114 | /* Last one doesn't continue. */ | |
115 | vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT; | |
116 | ||
117 | /* Update free pointer */ | |
118 | vq->free_head = i; | |
119 | ||
120 | /* Set token. */ | |
121 | vq->data[head] = data; | |
122 | ||
123 | /* Put entry in available array (but don't update avail->idx until they | |
124 | * do sync). FIXME: avoid modulus here? */ | |
125 | avail = (vq->vring.avail->idx + vq->num_added++) % vq->vring.num; | |
126 | vq->vring.avail->ring[avail] = head; | |
127 | ||
128 | pr_debug("Added buffer head %i to %p\n", head, vq); | |
129 | END_USE(vq); | |
130 | return 0; | |
131 | } | |
132 | ||
133 | static void vring_kick(struct virtqueue *_vq) | |
134 | { | |
135 | struct vring_virtqueue *vq = to_vvq(_vq); | |
136 | START_USE(vq); | |
137 | /* Descriptors and available array need to be set before we expose the | |
138 | * new available array entries. */ | |
139 | wmb(); | |
140 | ||
141 | vq->vring.avail->idx += vq->num_added; | |
142 | vq->num_added = 0; | |
143 | ||
144 | /* Need to update avail index before checking if we should notify */ | |
145 | mb(); | |
146 | ||
147 | if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY)) | |
148 | /* Prod other side to tell it about changes. */ | |
149 | vq->notify(&vq->vq); | |
150 | ||
151 | END_USE(vq); | |
152 | } | |
153 | ||
154 | static void detach_buf(struct vring_virtqueue *vq, unsigned int head) | |
155 | { | |
156 | unsigned int i; | |
157 | ||
158 | /* Clear data ptr. */ | |
159 | vq->data[head] = NULL; | |
160 | ||
161 | /* Put back on free list: find end */ | |
162 | i = head; | |
163 | while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) { | |
164 | i = vq->vring.desc[i].next; | |
165 | vq->num_free++; | |
166 | } | |
167 | ||
168 | vq->vring.desc[i].next = vq->free_head; | |
169 | vq->free_head = head; | |
170 | /* Plus final descriptor */ | |
171 | vq->num_free++; | |
172 | } | |
173 | ||
0a8a69dd RR |
174 | static inline bool more_used(const struct vring_virtqueue *vq) |
175 | { | |
176 | return vq->last_used_idx != vq->vring.used->idx; | |
177 | } | |
178 | ||
179 | static void *vring_get_buf(struct virtqueue *_vq, unsigned int *len) | |
180 | { | |
181 | struct vring_virtqueue *vq = to_vvq(_vq); | |
182 | void *ret; | |
183 | unsigned int i; | |
184 | ||
185 | START_USE(vq); | |
186 | ||
187 | if (!more_used(vq)) { | |
188 | pr_debug("No more buffers in queue\n"); | |
189 | END_USE(vq); | |
190 | return NULL; | |
191 | } | |
192 | ||
193 | i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id; | |
194 | *len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len; | |
195 | ||
196 | if (unlikely(i >= vq->vring.num)) { | |
197 | BAD_RING(vq, "id %u out of range\n", i); | |
198 | return NULL; | |
199 | } | |
200 | if (unlikely(!vq->data[i])) { | |
201 | BAD_RING(vq, "id %u is not a head!\n", i); | |
202 | return NULL; | |
203 | } | |
204 | ||
205 | /* detach_buf clears data, so grab it now. */ | |
206 | ret = vq->data[i]; | |
207 | detach_buf(vq, i); | |
208 | vq->last_used_idx++; | |
209 | END_USE(vq); | |
210 | return ret; | |
211 | } | |
212 | ||
18445c4d RR |
213 | static void vring_disable_cb(struct virtqueue *_vq) |
214 | { | |
215 | struct vring_virtqueue *vq = to_vvq(_vq); | |
216 | ||
217 | START_USE(vq); | |
218 | BUG_ON(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT); | |
219 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; | |
220 | END_USE(vq); | |
221 | } | |
222 | ||
223 | static bool vring_enable_cb(struct virtqueue *_vq) | |
0a8a69dd RR |
224 | { |
225 | struct vring_virtqueue *vq = to_vvq(_vq); | |
226 | ||
227 | START_USE(vq); | |
228 | BUG_ON(!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)); | |
229 | ||
230 | /* We optimistically turn back on interrupts, then check if there was | |
231 | * more to do. */ | |
232 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
233 | mb(); | |
234 | if (unlikely(more_used(vq))) { | |
235 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; | |
236 | END_USE(vq); | |
237 | return false; | |
238 | } | |
239 | ||
240 | END_USE(vq); | |
241 | return true; | |
242 | } | |
243 | ||
244 | irqreturn_t vring_interrupt(int irq, void *_vq) | |
245 | { | |
246 | struct vring_virtqueue *vq = to_vvq(_vq); | |
247 | ||
248 | if (!more_used(vq)) { | |
249 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
250 | return IRQ_NONE; | |
251 | } | |
252 | ||
253 | if (unlikely(vq->broken)) | |
254 | return IRQ_HANDLED; | |
255 | ||
81a8deab RR |
256 | /* Other side may have missed us turning off the interrupt, |
257 | * but we should preserve disable semantic for virtio users. */ | |
258 | if (unlikely(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) { | |
259 | pr_debug("virtqueue interrupt after disable for %p\n", vq); | |
260 | return IRQ_HANDLED; | |
261 | } | |
262 | ||
0a8a69dd | 263 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); |
18445c4d RR |
264 | if (vq->vq.callback) |
265 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
266 | |
267 | return IRQ_HANDLED; | |
268 | } | |
c6fd4701 | 269 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd RR |
270 | |
271 | static struct virtqueue_ops vring_vq_ops = { | |
272 | .add_buf = vring_add_buf, | |
273 | .get_buf = vring_get_buf, | |
274 | .kick = vring_kick, | |
18445c4d RR |
275 | .disable_cb = vring_disable_cb, |
276 | .enable_cb = vring_enable_cb, | |
0a8a69dd RR |
277 | }; |
278 | ||
279 | struct virtqueue *vring_new_virtqueue(unsigned int num, | |
280 | struct virtio_device *vdev, | |
281 | void *pages, | |
282 | void (*notify)(struct virtqueue *), | |
18445c4d | 283 | void (*callback)(struct virtqueue *)) |
0a8a69dd RR |
284 | { |
285 | struct vring_virtqueue *vq; | |
286 | unsigned int i; | |
287 | ||
42b36cc0 RR |
288 | /* We assume num is a power of 2. */ |
289 | if (num & (num - 1)) { | |
290 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
291 | return NULL; | |
292 | } | |
293 | ||
0a8a69dd RR |
294 | vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL); |
295 | if (!vq) | |
296 | return NULL; | |
297 | ||
42b36cc0 | 298 | vring_init(&vq->vring, num, pages, PAGE_SIZE); |
0a8a69dd RR |
299 | vq->vq.callback = callback; |
300 | vq->vq.vdev = vdev; | |
301 | vq->vq.vq_ops = &vring_vq_ops; | |
302 | vq->notify = notify; | |
303 | vq->broken = false; | |
304 | vq->last_used_idx = 0; | |
305 | vq->num_added = 0; | |
306 | #ifdef DEBUG | |
307 | vq->in_use = false; | |
308 | #endif | |
309 | ||
310 | /* No callback? Tell other side not to bother us. */ | |
311 | if (!callback) | |
312 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; | |
313 | ||
314 | /* Put everything in free lists. */ | |
315 | vq->num_free = num; | |
316 | vq->free_head = 0; | |
317 | for (i = 0; i < num-1; i++) | |
318 | vq->vring.desc[i].next = i+1; | |
319 | ||
320 | return &vq->vq; | |
321 | } | |
c6fd4701 | 322 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd RR |
323 | |
324 | void vring_del_virtqueue(struct virtqueue *vq) | |
325 | { | |
326 | kfree(to_vvq(vq)); | |
327 | } | |
c6fd4701 | 328 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 329 | |
c6fd4701 | 330 | MODULE_LICENSE("GPL"); |