]>
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> | |
e34f8725 | 21 | #include <linux/virtio_config.h> |
0a8a69dd RR |
22 | #include <linux/device.h> |
23 | ||
24 | #ifdef DEBUG | |
25 | /* For development, we want to crash whenever the ring is screwed. */ | |
9499f5e7 RR |
26 | #define BAD_RING(_vq, fmt, args...) \ |
27 | do { \ | |
28 | dev_err(&(_vq)->vq.vdev->dev, \ | |
29 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
30 | BUG(); \ | |
31 | } while (0) | |
c5f841f1 RR |
32 | /* Caller is supposed to guarantee no reentry. */ |
33 | #define START_USE(_vq) \ | |
34 | do { \ | |
35 | if ((_vq)->in_use) \ | |
9499f5e7 RR |
36 | panic("%s:in_use = %i\n", \ |
37 | (_vq)->vq.name, (_vq)->in_use); \ | |
c5f841f1 | 38 | (_vq)->in_use = __LINE__; \ |
9499f5e7 | 39 | } while (0) |
3a35ce7d | 40 | #define END_USE(_vq) \ |
97a545ab | 41 | do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; } while(0) |
0a8a69dd | 42 | #else |
9499f5e7 RR |
43 | #define BAD_RING(_vq, fmt, args...) \ |
44 | do { \ | |
45 | dev_err(&_vq->vq.vdev->dev, \ | |
46 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
47 | (_vq)->broken = true; \ | |
48 | } while (0) | |
0a8a69dd RR |
49 | #define START_USE(vq) |
50 | #define END_USE(vq) | |
51 | #endif | |
52 | ||
53 | struct vring_virtqueue | |
54 | { | |
55 | struct virtqueue vq; | |
56 | ||
57 | /* Actual memory layout for this queue */ | |
58 | struct vring vring; | |
59 | ||
60 | /* Other side has made a mess, don't try any more. */ | |
61 | bool broken; | |
62 | ||
9fa29b9d MM |
63 | /* Host supports indirect buffers */ |
64 | bool indirect; | |
65 | ||
0a8a69dd RR |
66 | /* Number of free buffers */ |
67 | unsigned int num_free; | |
68 | /* Head of free buffer list. */ | |
69 | unsigned int free_head; | |
70 | /* Number we've added since last sync. */ | |
71 | unsigned int num_added; | |
72 | ||
73 | /* Last used index we've seen. */ | |
1bc4953e | 74 | u16 last_used_idx; |
0a8a69dd RR |
75 | |
76 | /* How to notify other side. FIXME: commonalize hcalls! */ | |
77 | void (*notify)(struct virtqueue *vq); | |
78 | ||
79 | #ifdef DEBUG | |
80 | /* They're supposed to lock for us. */ | |
81 | unsigned int in_use; | |
82 | #endif | |
83 | ||
84 | /* Tokens for callbacks. */ | |
85 | void *data[]; | |
86 | }; | |
87 | ||
88 | #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq) | |
89 | ||
9fa29b9d MM |
90 | /* Set up an indirect table of descriptors and add it to the queue. */ |
91 | static int vring_add_indirect(struct vring_virtqueue *vq, | |
92 | struct scatterlist sg[], | |
93 | unsigned int out, | |
94 | unsigned int in) | |
95 | { | |
96 | struct vring_desc *desc; | |
97 | unsigned head; | |
98 | int i; | |
99 | ||
100 | desc = kmalloc((out + in) * sizeof(struct vring_desc), GFP_ATOMIC); | |
101 | if (!desc) | |
102 | return vq->vring.num; | |
103 | ||
104 | /* Transfer entries from the sg list into the indirect page */ | |
105 | for (i = 0; i < out; i++) { | |
106 | desc[i].flags = VRING_DESC_F_NEXT; | |
107 | desc[i].addr = sg_phys(sg); | |
108 | desc[i].len = sg->length; | |
109 | desc[i].next = i+1; | |
110 | sg++; | |
111 | } | |
112 | for (; i < (out + in); i++) { | |
113 | desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; | |
114 | desc[i].addr = sg_phys(sg); | |
115 | desc[i].len = sg->length; | |
116 | desc[i].next = i+1; | |
117 | sg++; | |
118 | } | |
119 | ||
120 | /* Last one doesn't continue. */ | |
121 | desc[i-1].flags &= ~VRING_DESC_F_NEXT; | |
122 | desc[i-1].next = 0; | |
123 | ||
124 | /* We're about to use a buffer */ | |
125 | vq->num_free--; | |
126 | ||
127 | /* Use a single buffer which doesn't continue */ | |
128 | head = vq->free_head; | |
129 | vq->vring.desc[head].flags = VRING_DESC_F_INDIRECT; | |
130 | vq->vring.desc[head].addr = virt_to_phys(desc); | |
131 | vq->vring.desc[head].len = i * sizeof(struct vring_desc); | |
132 | ||
133 | /* Update free pointer */ | |
134 | vq->free_head = vq->vring.desc[head].next; | |
135 | ||
136 | return head; | |
137 | } | |
138 | ||
0a8a69dd RR |
139 | static int vring_add_buf(struct virtqueue *_vq, |
140 | struct scatterlist sg[], | |
141 | unsigned int out, | |
142 | unsigned int in, | |
143 | void *data) | |
144 | { | |
145 | struct vring_virtqueue *vq = to_vvq(_vq); | |
146 | unsigned int i, avail, head, uninitialized_var(prev); | |
147 | ||
9fa29b9d MM |
148 | START_USE(vq); |
149 | ||
0a8a69dd | 150 | BUG_ON(data == NULL); |
9fa29b9d MM |
151 | |
152 | /* If the host supports indirect descriptor tables, and we have multiple | |
153 | * buffers, then go indirect. FIXME: tune this threshold */ | |
154 | if (vq->indirect && (out + in) > 1 && vq->num_free) { | |
155 | head = vring_add_indirect(vq, sg, out, in); | |
156 | if (head != vq->vring.num) | |
157 | goto add_head; | |
158 | } | |
159 | ||
0a8a69dd RR |
160 | BUG_ON(out + in > vq->vring.num); |
161 | BUG_ON(out + in == 0); | |
162 | ||
0a8a69dd RR |
163 | if (vq->num_free < out + in) { |
164 | pr_debug("Can't add buf len %i - avail = %i\n", | |
165 | out + in, vq->num_free); | |
44653eae RR |
166 | /* FIXME: for historical reasons, we force a notify here if |
167 | * there are outgoing parts to the buffer. Presumably the | |
168 | * host should service the ring ASAP. */ | |
169 | if (out) | |
170 | vq->notify(&vq->vq); | |
0a8a69dd RR |
171 | END_USE(vq); |
172 | return -ENOSPC; | |
173 | } | |
174 | ||
175 | /* We're about to use some buffers from the free list. */ | |
176 | vq->num_free -= out + in; | |
177 | ||
178 | head = vq->free_head; | |
179 | for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) { | |
180 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT; | |
15f9c890 | 181 | vq->vring.desc[i].addr = sg_phys(sg); |
0a8a69dd RR |
182 | vq->vring.desc[i].len = sg->length; |
183 | prev = i; | |
184 | sg++; | |
185 | } | |
186 | for (; in; i = vq->vring.desc[i].next, in--) { | |
187 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; | |
15f9c890 | 188 | vq->vring.desc[i].addr = sg_phys(sg); |
0a8a69dd RR |
189 | vq->vring.desc[i].len = sg->length; |
190 | prev = i; | |
191 | sg++; | |
192 | } | |
193 | /* Last one doesn't continue. */ | |
194 | vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT; | |
195 | ||
196 | /* Update free pointer */ | |
197 | vq->free_head = i; | |
198 | ||
9fa29b9d | 199 | add_head: |
0a8a69dd RR |
200 | /* Set token. */ |
201 | vq->data[head] = data; | |
202 | ||
203 | /* Put entry in available array (but don't update avail->idx until they | |
204 | * do sync). FIXME: avoid modulus here? */ | |
205 | avail = (vq->vring.avail->idx + vq->num_added++) % vq->vring.num; | |
206 | vq->vring.avail->ring[avail] = head; | |
207 | ||
208 | pr_debug("Added buffer head %i to %p\n", head, vq); | |
209 | END_USE(vq); | |
3c1b27d5 RR |
210 | |
211 | /* If we're indirect, we can fit many (assuming not OOM). */ | |
212 | if (vq->indirect) | |
213 | return vq->num_free ? vq->vring.num : 0; | |
214 | return vq->num_free; | |
0a8a69dd RR |
215 | } |
216 | ||
217 | static void vring_kick(struct virtqueue *_vq) | |
218 | { | |
219 | struct vring_virtqueue *vq = to_vvq(_vq); | |
220 | START_USE(vq); | |
221 | /* Descriptors and available array need to be set before we expose the | |
222 | * new available array entries. */ | |
223 | wmb(); | |
224 | ||
225 | vq->vring.avail->idx += vq->num_added; | |
226 | vq->num_added = 0; | |
227 | ||
228 | /* Need to update avail index before checking if we should notify */ | |
229 | mb(); | |
230 | ||
231 | if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY)) | |
232 | /* Prod other side to tell it about changes. */ | |
233 | vq->notify(&vq->vq); | |
234 | ||
235 | END_USE(vq); | |
236 | } | |
237 | ||
238 | static void detach_buf(struct vring_virtqueue *vq, unsigned int head) | |
239 | { | |
240 | unsigned int i; | |
241 | ||
242 | /* Clear data ptr. */ | |
243 | vq->data[head] = NULL; | |
244 | ||
245 | /* Put back on free list: find end */ | |
246 | i = head; | |
9fa29b9d MM |
247 | |
248 | /* Free the indirect table */ | |
249 | if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT) | |
250 | kfree(phys_to_virt(vq->vring.desc[i].addr)); | |
251 | ||
0a8a69dd RR |
252 | while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) { |
253 | i = vq->vring.desc[i].next; | |
254 | vq->num_free++; | |
255 | } | |
256 | ||
257 | vq->vring.desc[i].next = vq->free_head; | |
258 | vq->free_head = head; | |
259 | /* Plus final descriptor */ | |
260 | vq->num_free++; | |
261 | } | |
262 | ||
0a8a69dd RR |
263 | static inline bool more_used(const struct vring_virtqueue *vq) |
264 | { | |
265 | return vq->last_used_idx != vq->vring.used->idx; | |
266 | } | |
267 | ||
268 | static void *vring_get_buf(struct virtqueue *_vq, unsigned int *len) | |
269 | { | |
270 | struct vring_virtqueue *vq = to_vvq(_vq); | |
271 | void *ret; | |
272 | unsigned int i; | |
273 | ||
274 | START_USE(vq); | |
275 | ||
5ef82752 RR |
276 | if (unlikely(vq->broken)) { |
277 | END_USE(vq); | |
278 | return NULL; | |
279 | } | |
280 | ||
0a8a69dd RR |
281 | if (!more_used(vq)) { |
282 | pr_debug("No more buffers in queue\n"); | |
283 | END_USE(vq); | |
284 | return NULL; | |
285 | } | |
286 | ||
2d61ba95 MT |
287 | /* Only get used array entries after they have been exposed by host. */ |
288 | rmb(); | |
289 | ||
0a8a69dd RR |
290 | i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id; |
291 | *len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len; | |
292 | ||
293 | if (unlikely(i >= vq->vring.num)) { | |
294 | BAD_RING(vq, "id %u out of range\n", i); | |
295 | return NULL; | |
296 | } | |
297 | if (unlikely(!vq->data[i])) { | |
298 | BAD_RING(vq, "id %u is not a head!\n", i); | |
299 | return NULL; | |
300 | } | |
301 | ||
302 | /* detach_buf clears data, so grab it now. */ | |
303 | ret = vq->data[i]; | |
304 | detach_buf(vq, i); | |
305 | vq->last_used_idx++; | |
306 | END_USE(vq); | |
307 | return ret; | |
308 | } | |
309 | ||
18445c4d RR |
310 | static void vring_disable_cb(struct virtqueue *_vq) |
311 | { | |
312 | struct vring_virtqueue *vq = to_vvq(_vq); | |
313 | ||
18445c4d | 314 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; |
18445c4d RR |
315 | } |
316 | ||
317 | static bool vring_enable_cb(struct virtqueue *_vq) | |
0a8a69dd RR |
318 | { |
319 | struct vring_virtqueue *vq = to_vvq(_vq); | |
320 | ||
321 | START_USE(vq); | |
0a8a69dd RR |
322 | |
323 | /* We optimistically turn back on interrupts, then check if there was | |
324 | * more to do. */ | |
325 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
326 | mb(); | |
327 | if (unlikely(more_used(vq))) { | |
0a8a69dd RR |
328 | END_USE(vq); |
329 | return false; | |
330 | } | |
331 | ||
332 | END_USE(vq); | |
333 | return true; | |
334 | } | |
335 | ||
336 | irqreturn_t vring_interrupt(int irq, void *_vq) | |
337 | { | |
338 | struct vring_virtqueue *vq = to_vvq(_vq); | |
339 | ||
340 | if (!more_used(vq)) { | |
341 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
342 | return IRQ_NONE; | |
343 | } | |
344 | ||
345 | if (unlikely(vq->broken)) | |
346 | return IRQ_HANDLED; | |
347 | ||
348 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); | |
18445c4d RR |
349 | if (vq->vq.callback) |
350 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
351 | |
352 | return IRQ_HANDLED; | |
353 | } | |
c6fd4701 | 354 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd RR |
355 | |
356 | static struct virtqueue_ops vring_vq_ops = { | |
357 | .add_buf = vring_add_buf, | |
358 | .get_buf = vring_get_buf, | |
359 | .kick = vring_kick, | |
18445c4d RR |
360 | .disable_cb = vring_disable_cb, |
361 | .enable_cb = vring_enable_cb, | |
0a8a69dd RR |
362 | }; |
363 | ||
364 | struct virtqueue *vring_new_virtqueue(unsigned int num, | |
87c7d57c | 365 | unsigned int vring_align, |
0a8a69dd RR |
366 | struct virtio_device *vdev, |
367 | void *pages, | |
368 | void (*notify)(struct virtqueue *), | |
9499f5e7 RR |
369 | void (*callback)(struct virtqueue *), |
370 | const char *name) | |
0a8a69dd RR |
371 | { |
372 | struct vring_virtqueue *vq; | |
373 | unsigned int i; | |
374 | ||
42b36cc0 RR |
375 | /* We assume num is a power of 2. */ |
376 | if (num & (num - 1)) { | |
377 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
378 | return NULL; | |
379 | } | |
380 | ||
0a8a69dd RR |
381 | vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL); |
382 | if (!vq) | |
383 | return NULL; | |
384 | ||
87c7d57c | 385 | vring_init(&vq->vring, num, pages, vring_align); |
0a8a69dd RR |
386 | vq->vq.callback = callback; |
387 | vq->vq.vdev = vdev; | |
388 | vq->vq.vq_ops = &vring_vq_ops; | |
9499f5e7 | 389 | vq->vq.name = name; |
0a8a69dd RR |
390 | vq->notify = notify; |
391 | vq->broken = false; | |
392 | vq->last_used_idx = 0; | |
393 | vq->num_added = 0; | |
9499f5e7 | 394 | list_add_tail(&vq->vq.list, &vdev->vqs); |
0a8a69dd RR |
395 | #ifdef DEBUG |
396 | vq->in_use = false; | |
397 | #endif | |
398 | ||
9fa29b9d MM |
399 | vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC); |
400 | ||
0a8a69dd RR |
401 | /* No callback? Tell other side not to bother us. */ |
402 | if (!callback) | |
403 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; | |
404 | ||
405 | /* Put everything in free lists. */ | |
406 | vq->num_free = num; | |
407 | vq->free_head = 0; | |
408 | for (i = 0; i < num-1; i++) | |
409 | vq->vring.desc[i].next = i+1; | |
410 | ||
411 | return &vq->vq; | |
412 | } | |
c6fd4701 | 413 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd RR |
414 | |
415 | void vring_del_virtqueue(struct virtqueue *vq) | |
416 | { | |
9499f5e7 | 417 | list_del(&vq->list); |
0a8a69dd RR |
418 | kfree(to_vvq(vq)); |
419 | } | |
c6fd4701 | 420 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 421 | |
e34f8725 RR |
422 | /* Manipulates transport-specific feature bits. */ |
423 | void vring_transport_features(struct virtio_device *vdev) | |
424 | { | |
425 | unsigned int i; | |
426 | ||
427 | for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { | |
428 | switch (i) { | |
9fa29b9d MM |
429 | case VIRTIO_RING_F_INDIRECT_DESC: |
430 | break; | |
e34f8725 RR |
431 | default: |
432 | /* We don't understand this bit. */ | |
433 | clear_bit(i, vdev->features); | |
434 | } | |
435 | } | |
436 | } | |
437 | EXPORT_SYMBOL_GPL(vring_transport_features); | |
438 | ||
c6fd4701 | 439 | MODULE_LICENSE("GPL"); |