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 | 22 | #include <linux/device.h> |
5a0e3ad6 | 23 | #include <linux/slab.h> |
b5a2c4f1 | 24 | #include <linux/module.h> |
e93300b1 | 25 | #include <linux/hrtimer.h> |
0a8a69dd | 26 | |
d57ed95d MT |
27 | /* virtio guest is communicating with a virtual "device" that actually runs on |
28 | * a host processor. Memory barriers are used to control SMP effects. */ | |
29 | #ifdef CONFIG_SMP | |
30 | /* Where possible, use SMP barriers which are more lightweight than mandatory | |
31 | * barriers, because mandatory barriers control MMIO effects on accesses | |
7b21e34f RR |
32 | * through relaxed memory I/O windows (which virtio-pci does not use). */ |
33 | #define virtio_mb(vq) \ | |
34 | do { if ((vq)->weak_barriers) smp_mb(); else mb(); } while(0) | |
35 | #define virtio_rmb(vq) \ | |
36 | do { if ((vq)->weak_barriers) smp_rmb(); else rmb(); } while(0) | |
37 | #define virtio_wmb(vq) \ | |
4dbc5d9f | 38 | do { if ((vq)->weak_barriers) smp_wmb(); else wmb(); } while(0) |
d57ed95d MT |
39 | #else |
40 | /* We must force memory ordering even if guest is UP since host could be | |
41 | * running on another CPU, but SMP barriers are defined to barrier() in that | |
42 | * configuration. So fall back to mandatory barriers instead. */ | |
7b21e34f RR |
43 | #define virtio_mb(vq) mb() |
44 | #define virtio_rmb(vq) rmb() | |
45 | #define virtio_wmb(vq) wmb() | |
d57ed95d MT |
46 | #endif |
47 | ||
0a8a69dd RR |
48 | #ifdef DEBUG |
49 | /* For development, we want to crash whenever the ring is screwed. */ | |
9499f5e7 RR |
50 | #define BAD_RING(_vq, fmt, args...) \ |
51 | do { \ | |
52 | dev_err(&(_vq)->vq.vdev->dev, \ | |
53 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
54 | BUG(); \ | |
55 | } while (0) | |
c5f841f1 RR |
56 | /* Caller is supposed to guarantee no reentry. */ |
57 | #define START_USE(_vq) \ | |
58 | do { \ | |
59 | if ((_vq)->in_use) \ | |
9499f5e7 RR |
60 | panic("%s:in_use = %i\n", \ |
61 | (_vq)->vq.name, (_vq)->in_use); \ | |
c5f841f1 | 62 | (_vq)->in_use = __LINE__; \ |
9499f5e7 | 63 | } while (0) |
3a35ce7d | 64 | #define END_USE(_vq) \ |
97a545ab | 65 | do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; } while(0) |
0a8a69dd | 66 | #else |
9499f5e7 RR |
67 | #define BAD_RING(_vq, fmt, args...) \ |
68 | do { \ | |
69 | dev_err(&_vq->vq.vdev->dev, \ | |
70 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
71 | (_vq)->broken = true; \ | |
72 | } while (0) | |
0a8a69dd RR |
73 | #define START_USE(vq) |
74 | #define END_USE(vq) | |
75 | #endif | |
76 | ||
77 | struct vring_virtqueue | |
78 | { | |
79 | struct virtqueue vq; | |
80 | ||
81 | /* Actual memory layout for this queue */ | |
82 | struct vring vring; | |
83 | ||
7b21e34f RR |
84 | /* Can we use weak barriers? */ |
85 | bool weak_barriers; | |
86 | ||
0a8a69dd RR |
87 | /* Other side has made a mess, don't try any more. */ |
88 | bool broken; | |
89 | ||
9fa29b9d MM |
90 | /* Host supports indirect buffers */ |
91 | bool indirect; | |
92 | ||
a5c262c5 MT |
93 | /* Host publishes avail event idx */ |
94 | bool event; | |
95 | ||
0a8a69dd RR |
96 | /* Number of free buffers */ |
97 | unsigned int num_free; | |
98 | /* Head of free buffer list. */ | |
99 | unsigned int free_head; | |
100 | /* Number we've added since last sync. */ | |
101 | unsigned int num_added; | |
102 | ||
103 | /* Last used index we've seen. */ | |
1bc4953e | 104 | u16 last_used_idx; |
0a8a69dd RR |
105 | |
106 | /* How to notify other side. FIXME: commonalize hcalls! */ | |
107 | void (*notify)(struct virtqueue *vq); | |
108 | ||
17bb6d40 JW |
109 | /* Index of the queue */ |
110 | int queue_index; | |
111 | ||
0a8a69dd RR |
112 | #ifdef DEBUG |
113 | /* They're supposed to lock for us. */ | |
114 | unsigned int in_use; | |
e93300b1 RR |
115 | |
116 | /* Figure out if their kicks are too delayed. */ | |
117 | bool last_add_time_valid; | |
118 | ktime_t last_add_time; | |
0a8a69dd RR |
119 | #endif |
120 | ||
121 | /* Tokens for callbacks. */ | |
122 | void *data[]; | |
123 | }; | |
124 | ||
125 | #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq) | |
126 | ||
9fa29b9d MM |
127 | /* Set up an indirect table of descriptors and add it to the queue. */ |
128 | static int vring_add_indirect(struct vring_virtqueue *vq, | |
129 | struct scatterlist sg[], | |
130 | unsigned int out, | |
bbd603ef MT |
131 | unsigned int in, |
132 | gfp_t gfp) | |
9fa29b9d MM |
133 | { |
134 | struct vring_desc *desc; | |
135 | unsigned head; | |
136 | int i; | |
137 | ||
bbd603ef | 138 | desc = kmalloc((out + in) * sizeof(struct vring_desc), gfp); |
9fa29b9d | 139 | if (!desc) |
686d3637 | 140 | return -ENOMEM; |
9fa29b9d MM |
141 | |
142 | /* Transfer entries from the sg list into the indirect page */ | |
143 | for (i = 0; i < out; i++) { | |
144 | desc[i].flags = VRING_DESC_F_NEXT; | |
145 | desc[i].addr = sg_phys(sg); | |
146 | desc[i].len = sg->length; | |
147 | desc[i].next = i+1; | |
148 | sg++; | |
149 | } | |
150 | for (; i < (out + in); i++) { | |
151 | desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; | |
152 | desc[i].addr = sg_phys(sg); | |
153 | desc[i].len = sg->length; | |
154 | desc[i].next = i+1; | |
155 | sg++; | |
156 | } | |
157 | ||
158 | /* Last one doesn't continue. */ | |
159 | desc[i-1].flags &= ~VRING_DESC_F_NEXT; | |
160 | desc[i-1].next = 0; | |
161 | ||
162 | /* We're about to use a buffer */ | |
163 | vq->num_free--; | |
164 | ||
165 | /* Use a single buffer which doesn't continue */ | |
166 | head = vq->free_head; | |
167 | vq->vring.desc[head].flags = VRING_DESC_F_INDIRECT; | |
168 | vq->vring.desc[head].addr = virt_to_phys(desc); | |
169 | vq->vring.desc[head].len = i * sizeof(struct vring_desc); | |
170 | ||
171 | /* Update free pointer */ | |
172 | vq->free_head = vq->vring.desc[head].next; | |
173 | ||
174 | return head; | |
175 | } | |
176 | ||
17bb6d40 JW |
177 | int virtqueue_get_queue_index(struct virtqueue *_vq) |
178 | { | |
179 | struct vring_virtqueue *vq = to_vvq(_vq); | |
180 | return vq->queue_index; | |
181 | } | |
182 | EXPORT_SYMBOL_GPL(virtqueue_get_queue_index); | |
183 | ||
5dfc1762 | 184 | /** |
f96fde41 | 185 | * virtqueue_add_buf - expose buffer to other end |
5dfc1762 RR |
186 | * @vq: the struct virtqueue we're talking about. |
187 | * @sg: the description of the buffer(s). | |
188 | * @out_num: the number of sg readable by other side | |
189 | * @in_num: the number of sg which are writable (after readable ones) | |
190 | * @data: the token identifying the buffer. | |
191 | * @gfp: how to do memory allocations (if necessary). | |
192 | * | |
193 | * Caller must ensure we don't call this with other virtqueue operations | |
194 | * at the same time (except where noted). | |
195 | * | |
196 | * Returns remaining capacity of queue or a negative error | |
197 | * (ie. ENOSPC). Note that it only really makes sense to treat all | |
198 | * positive return values as "available": indirect buffers mean that | |
199 | * we can put an entire sg[] array inside a single queue entry. | |
200 | */ | |
f96fde41 RR |
201 | int virtqueue_add_buf(struct virtqueue *_vq, |
202 | struct scatterlist sg[], | |
203 | unsigned int out, | |
204 | unsigned int in, | |
205 | void *data, | |
206 | gfp_t gfp) | |
0a8a69dd RR |
207 | { |
208 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1fe9b6fe MT |
209 | unsigned int i, avail, uninitialized_var(prev); |
210 | int head; | |
0a8a69dd | 211 | |
9fa29b9d MM |
212 | START_USE(vq); |
213 | ||
0a8a69dd | 214 | BUG_ON(data == NULL); |
9fa29b9d | 215 | |
e93300b1 RR |
216 | #ifdef DEBUG |
217 | { | |
218 | ktime_t now = ktime_get(); | |
219 | ||
220 | /* No kick or get, with .1 second between? Warn. */ | |
221 | if (vq->last_add_time_valid) | |
222 | WARN_ON(ktime_to_ms(ktime_sub(now, vq->last_add_time)) | |
223 | > 100); | |
224 | vq->last_add_time = now; | |
225 | vq->last_add_time_valid = true; | |
226 | } | |
227 | #endif | |
228 | ||
9fa29b9d MM |
229 | /* If the host supports indirect descriptor tables, and we have multiple |
230 | * buffers, then go indirect. FIXME: tune this threshold */ | |
231 | if (vq->indirect && (out + in) > 1 && vq->num_free) { | |
bbd603ef | 232 | head = vring_add_indirect(vq, sg, out, in, gfp); |
1fe9b6fe | 233 | if (likely(head >= 0)) |
9fa29b9d MM |
234 | goto add_head; |
235 | } | |
236 | ||
0a8a69dd RR |
237 | BUG_ON(out + in > vq->vring.num); |
238 | BUG_ON(out + in == 0); | |
239 | ||
0a8a69dd RR |
240 | if (vq->num_free < out + in) { |
241 | pr_debug("Can't add buf len %i - avail = %i\n", | |
242 | out + in, vq->num_free); | |
44653eae RR |
243 | /* FIXME: for historical reasons, we force a notify here if |
244 | * there are outgoing parts to the buffer. Presumably the | |
245 | * host should service the ring ASAP. */ | |
246 | if (out) | |
247 | vq->notify(&vq->vq); | |
0a8a69dd RR |
248 | END_USE(vq); |
249 | return -ENOSPC; | |
250 | } | |
251 | ||
252 | /* We're about to use some buffers from the free list. */ | |
253 | vq->num_free -= out + in; | |
254 | ||
255 | head = vq->free_head; | |
256 | for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) { | |
257 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT; | |
15f9c890 | 258 | vq->vring.desc[i].addr = sg_phys(sg); |
0a8a69dd RR |
259 | vq->vring.desc[i].len = sg->length; |
260 | prev = i; | |
261 | sg++; | |
262 | } | |
263 | for (; in; i = vq->vring.desc[i].next, in--) { | |
264 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; | |
15f9c890 | 265 | vq->vring.desc[i].addr = sg_phys(sg); |
0a8a69dd RR |
266 | vq->vring.desc[i].len = sg->length; |
267 | prev = i; | |
268 | sg++; | |
269 | } | |
270 | /* Last one doesn't continue. */ | |
271 | vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT; | |
272 | ||
273 | /* Update free pointer */ | |
274 | vq->free_head = i; | |
275 | ||
9fa29b9d | 276 | add_head: |
0a8a69dd RR |
277 | /* Set token. */ |
278 | vq->data[head] = data; | |
279 | ||
280 | /* Put entry in available array (but don't update avail->idx until they | |
3b720b8c | 281 | * do sync). */ |
ee7cd898 | 282 | avail = (vq->vring.avail->idx & (vq->vring.num-1)); |
0a8a69dd RR |
283 | vq->vring.avail->ring[avail] = head; |
284 | ||
ee7cd898 RR |
285 | /* Descriptors and available array need to be set before we expose the |
286 | * new available array entries. */ | |
287 | virtio_wmb(vq); | |
288 | vq->vring.avail->idx++; | |
289 | vq->num_added++; | |
290 | ||
291 | /* This is very unlikely, but theoretically possible. Kick | |
292 | * just in case. */ | |
293 | if (unlikely(vq->num_added == (1 << 16) - 1)) | |
294 | virtqueue_kick(_vq); | |
295 | ||
0a8a69dd RR |
296 | pr_debug("Added buffer head %i to %p\n", head, vq); |
297 | END_USE(vq); | |
3c1b27d5 | 298 | |
3c1b27d5 | 299 | return vq->num_free; |
0a8a69dd | 300 | } |
f96fde41 | 301 | EXPORT_SYMBOL_GPL(virtqueue_add_buf); |
0a8a69dd | 302 | |
5dfc1762 | 303 | /** |
41f0377f | 304 | * virtqueue_kick_prepare - first half of split virtqueue_kick call. |
5dfc1762 RR |
305 | * @vq: the struct virtqueue |
306 | * | |
41f0377f RR |
307 | * Instead of virtqueue_kick(), you can do: |
308 | * if (virtqueue_kick_prepare(vq)) | |
309 | * virtqueue_notify(vq); | |
5dfc1762 | 310 | * |
41f0377f RR |
311 | * This is sometimes useful because the virtqueue_kick_prepare() needs |
312 | * to be serialized, but the actual virtqueue_notify() call does not. | |
5dfc1762 | 313 | */ |
41f0377f | 314 | bool virtqueue_kick_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
315 | { |
316 | struct vring_virtqueue *vq = to_vvq(_vq); | |
a5c262c5 | 317 | u16 new, old; |
41f0377f RR |
318 | bool needs_kick; |
319 | ||
0a8a69dd | 320 | START_USE(vq); |
a72caae2 JW |
321 | /* We need to expose available array entries before checking avail |
322 | * event. */ | |
323 | virtio_mb(vq); | |
0a8a69dd | 324 | |
ee7cd898 RR |
325 | old = vq->vring.avail->idx - vq->num_added; |
326 | new = vq->vring.avail->idx; | |
0a8a69dd RR |
327 | vq->num_added = 0; |
328 | ||
e93300b1 RR |
329 | #ifdef DEBUG |
330 | if (vq->last_add_time_valid) { | |
331 | WARN_ON(ktime_to_ms(ktime_sub(ktime_get(), | |
332 | vq->last_add_time)) > 100); | |
333 | } | |
334 | vq->last_add_time_valid = false; | |
335 | #endif | |
336 | ||
41f0377f RR |
337 | if (vq->event) { |
338 | needs_kick = vring_need_event(vring_avail_event(&vq->vring), | |
339 | new, old); | |
340 | } else { | |
341 | needs_kick = !(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY); | |
342 | } | |
0a8a69dd | 343 | END_USE(vq); |
41f0377f RR |
344 | return needs_kick; |
345 | } | |
346 | EXPORT_SYMBOL_GPL(virtqueue_kick_prepare); | |
347 | ||
348 | /** | |
349 | * virtqueue_notify - second half of split virtqueue_kick call. | |
350 | * @vq: the struct virtqueue | |
351 | * | |
352 | * This does not need to be serialized. | |
353 | */ | |
354 | void virtqueue_notify(struct virtqueue *_vq) | |
355 | { | |
356 | struct vring_virtqueue *vq = to_vvq(_vq); | |
357 | ||
358 | /* Prod other side to tell it about changes. */ | |
359 | vq->notify(_vq); | |
360 | } | |
361 | EXPORT_SYMBOL_GPL(virtqueue_notify); | |
362 | ||
363 | /** | |
364 | * virtqueue_kick - update after add_buf | |
365 | * @vq: the struct virtqueue | |
366 | * | |
367 | * After one or more virtqueue_add_buf calls, invoke this to kick | |
368 | * the other side. | |
369 | * | |
370 | * Caller must ensure we don't call this with other virtqueue | |
371 | * operations at the same time (except where noted). | |
372 | */ | |
373 | void virtqueue_kick(struct virtqueue *vq) | |
374 | { | |
375 | if (virtqueue_kick_prepare(vq)) | |
376 | virtqueue_notify(vq); | |
0a8a69dd | 377 | } |
7c5e9ed0 | 378 | EXPORT_SYMBOL_GPL(virtqueue_kick); |
0a8a69dd RR |
379 | |
380 | static void detach_buf(struct vring_virtqueue *vq, unsigned int head) | |
381 | { | |
382 | unsigned int i; | |
383 | ||
384 | /* Clear data ptr. */ | |
385 | vq->data[head] = NULL; | |
386 | ||
387 | /* Put back on free list: find end */ | |
388 | i = head; | |
9fa29b9d MM |
389 | |
390 | /* Free the indirect table */ | |
391 | if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT) | |
392 | kfree(phys_to_virt(vq->vring.desc[i].addr)); | |
393 | ||
0a8a69dd RR |
394 | while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) { |
395 | i = vq->vring.desc[i].next; | |
396 | vq->num_free++; | |
397 | } | |
398 | ||
399 | vq->vring.desc[i].next = vq->free_head; | |
400 | vq->free_head = head; | |
401 | /* Plus final descriptor */ | |
402 | vq->num_free++; | |
403 | } | |
404 | ||
0a8a69dd RR |
405 | static inline bool more_used(const struct vring_virtqueue *vq) |
406 | { | |
407 | return vq->last_used_idx != vq->vring.used->idx; | |
408 | } | |
409 | ||
5dfc1762 RR |
410 | /** |
411 | * virtqueue_get_buf - get the next used buffer | |
412 | * @vq: the struct virtqueue we're talking about. | |
413 | * @len: the length written into the buffer | |
414 | * | |
415 | * If the driver wrote data into the buffer, @len will be set to the | |
416 | * amount written. This means you don't need to clear the buffer | |
417 | * beforehand to ensure there's no data leakage in the case of short | |
418 | * writes. | |
419 | * | |
420 | * Caller must ensure we don't call this with other virtqueue | |
421 | * operations at the same time (except where noted). | |
422 | * | |
423 | * Returns NULL if there are no used buffers, or the "data" token | |
f96fde41 | 424 | * handed to virtqueue_add_buf(). |
5dfc1762 | 425 | */ |
7c5e9ed0 | 426 | void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len) |
0a8a69dd RR |
427 | { |
428 | struct vring_virtqueue *vq = to_vvq(_vq); | |
429 | void *ret; | |
430 | unsigned int i; | |
3b720b8c | 431 | u16 last_used; |
0a8a69dd RR |
432 | |
433 | START_USE(vq); | |
434 | ||
5ef82752 RR |
435 | if (unlikely(vq->broken)) { |
436 | END_USE(vq); | |
437 | return NULL; | |
438 | } | |
439 | ||
0a8a69dd RR |
440 | if (!more_used(vq)) { |
441 | pr_debug("No more buffers in queue\n"); | |
442 | END_USE(vq); | |
443 | return NULL; | |
444 | } | |
445 | ||
2d61ba95 | 446 | /* Only get used array entries after they have been exposed by host. */ |
7b21e34f | 447 | virtio_rmb(vq); |
2d61ba95 | 448 | |
3b720b8c RR |
449 | last_used = (vq->last_used_idx & (vq->vring.num - 1)); |
450 | i = vq->vring.used->ring[last_used].id; | |
451 | *len = vq->vring.used->ring[last_used].len; | |
0a8a69dd RR |
452 | |
453 | if (unlikely(i >= vq->vring.num)) { | |
454 | BAD_RING(vq, "id %u out of range\n", i); | |
455 | return NULL; | |
456 | } | |
457 | if (unlikely(!vq->data[i])) { | |
458 | BAD_RING(vq, "id %u is not a head!\n", i); | |
459 | return NULL; | |
460 | } | |
461 | ||
462 | /* detach_buf clears data, so grab it now. */ | |
463 | ret = vq->data[i]; | |
464 | detach_buf(vq, i); | |
465 | vq->last_used_idx++; | |
a5c262c5 MT |
466 | /* If we expect an interrupt for the next entry, tell host |
467 | * by writing event index and flush out the write before | |
468 | * the read in the next get_buf call. */ | |
469 | if (!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) { | |
470 | vring_used_event(&vq->vring) = vq->last_used_idx; | |
7b21e34f | 471 | virtio_mb(vq); |
a5c262c5 MT |
472 | } |
473 | ||
e93300b1 RR |
474 | #ifdef DEBUG |
475 | vq->last_add_time_valid = false; | |
476 | #endif | |
477 | ||
0a8a69dd RR |
478 | END_USE(vq); |
479 | return ret; | |
480 | } | |
7c5e9ed0 | 481 | EXPORT_SYMBOL_GPL(virtqueue_get_buf); |
0a8a69dd | 482 | |
5dfc1762 RR |
483 | /** |
484 | * virtqueue_disable_cb - disable callbacks | |
485 | * @vq: the struct virtqueue we're talking about. | |
486 | * | |
487 | * Note that this is not necessarily synchronous, hence unreliable and only | |
488 | * useful as an optimization. | |
489 | * | |
490 | * Unlike other operations, this need not be serialized. | |
491 | */ | |
7c5e9ed0 | 492 | void virtqueue_disable_cb(struct virtqueue *_vq) |
18445c4d RR |
493 | { |
494 | struct vring_virtqueue *vq = to_vvq(_vq); | |
495 | ||
18445c4d | 496 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; |
18445c4d | 497 | } |
7c5e9ed0 | 498 | EXPORT_SYMBOL_GPL(virtqueue_disable_cb); |
18445c4d | 499 | |
5dfc1762 RR |
500 | /** |
501 | * virtqueue_enable_cb - restart callbacks after disable_cb. | |
502 | * @vq: the struct virtqueue we're talking about. | |
503 | * | |
504 | * This re-enables callbacks; it returns "false" if there are pending | |
505 | * buffers in the queue, to detect a possible race between the driver | |
506 | * checking for more work, and enabling callbacks. | |
507 | * | |
508 | * Caller must ensure we don't call this with other virtqueue | |
509 | * operations at the same time (except where noted). | |
510 | */ | |
7c5e9ed0 | 511 | bool virtqueue_enable_cb(struct virtqueue *_vq) |
0a8a69dd RR |
512 | { |
513 | struct vring_virtqueue *vq = to_vvq(_vq); | |
514 | ||
515 | START_USE(vq); | |
0a8a69dd RR |
516 | |
517 | /* We optimistically turn back on interrupts, then check if there was | |
518 | * more to do. */ | |
a5c262c5 MT |
519 | /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to |
520 | * either clear the flags bit or point the event index at the next | |
521 | * entry. Always do both to keep code simple. */ | |
0a8a69dd | 522 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; |
a5c262c5 | 523 | vring_used_event(&vq->vring) = vq->last_used_idx; |
7b21e34f | 524 | virtio_mb(vq); |
0a8a69dd | 525 | if (unlikely(more_used(vq))) { |
0a8a69dd RR |
526 | END_USE(vq); |
527 | return false; | |
528 | } | |
529 | ||
530 | END_USE(vq); | |
531 | return true; | |
532 | } | |
7c5e9ed0 | 533 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb); |
0a8a69dd | 534 | |
5dfc1762 RR |
535 | /** |
536 | * virtqueue_enable_cb_delayed - restart callbacks after disable_cb. | |
537 | * @vq: the struct virtqueue we're talking about. | |
538 | * | |
539 | * This re-enables callbacks but hints to the other side to delay | |
540 | * interrupts until most of the available buffers have been processed; | |
541 | * it returns "false" if there are many pending buffers in the queue, | |
542 | * to detect a possible race between the driver checking for more work, | |
543 | * and enabling callbacks. | |
544 | * | |
545 | * Caller must ensure we don't call this with other virtqueue | |
546 | * operations at the same time (except where noted). | |
547 | */ | |
7ab358c2 MT |
548 | bool virtqueue_enable_cb_delayed(struct virtqueue *_vq) |
549 | { | |
550 | struct vring_virtqueue *vq = to_vvq(_vq); | |
551 | u16 bufs; | |
552 | ||
553 | START_USE(vq); | |
554 | ||
555 | /* We optimistically turn back on interrupts, then check if there was | |
556 | * more to do. */ | |
557 | /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to | |
558 | * either clear the flags bit or point the event index at the next | |
559 | * entry. Always do both to keep code simple. */ | |
560 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
561 | /* TODO: tune this threshold */ | |
562 | bufs = (u16)(vq->vring.avail->idx - vq->last_used_idx) * 3 / 4; | |
563 | vring_used_event(&vq->vring) = vq->last_used_idx + bufs; | |
7b21e34f | 564 | virtio_mb(vq); |
7ab358c2 MT |
565 | if (unlikely((u16)(vq->vring.used->idx - vq->last_used_idx) > bufs)) { |
566 | END_USE(vq); | |
567 | return false; | |
568 | } | |
569 | ||
570 | END_USE(vq); | |
571 | return true; | |
572 | } | |
573 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed); | |
574 | ||
5dfc1762 RR |
575 | /** |
576 | * virtqueue_detach_unused_buf - detach first unused buffer | |
577 | * @vq: the struct virtqueue we're talking about. | |
578 | * | |
f96fde41 | 579 | * Returns NULL or the "data" token handed to virtqueue_add_buf(). |
5dfc1762 RR |
580 | * This is not valid on an active queue; it is useful only for device |
581 | * shutdown. | |
582 | */ | |
7c5e9ed0 | 583 | void *virtqueue_detach_unused_buf(struct virtqueue *_vq) |
c021eac4 SM |
584 | { |
585 | struct vring_virtqueue *vq = to_vvq(_vq); | |
586 | unsigned int i; | |
587 | void *buf; | |
588 | ||
589 | START_USE(vq); | |
590 | ||
591 | for (i = 0; i < vq->vring.num; i++) { | |
592 | if (!vq->data[i]) | |
593 | continue; | |
594 | /* detach_buf clears data, so grab it now. */ | |
595 | buf = vq->data[i]; | |
596 | detach_buf(vq, i); | |
b3258ff1 | 597 | vq->vring.avail->idx--; |
c021eac4 SM |
598 | END_USE(vq); |
599 | return buf; | |
600 | } | |
601 | /* That should have freed everything. */ | |
602 | BUG_ON(vq->num_free != vq->vring.num); | |
603 | ||
604 | END_USE(vq); | |
605 | return NULL; | |
606 | } | |
7c5e9ed0 | 607 | EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf); |
c021eac4 | 608 | |
0a8a69dd RR |
609 | irqreturn_t vring_interrupt(int irq, void *_vq) |
610 | { | |
611 | struct vring_virtqueue *vq = to_vvq(_vq); | |
612 | ||
613 | if (!more_used(vq)) { | |
614 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
615 | return IRQ_NONE; | |
616 | } | |
617 | ||
618 | if (unlikely(vq->broken)) | |
619 | return IRQ_HANDLED; | |
620 | ||
621 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); | |
18445c4d RR |
622 | if (vq->vq.callback) |
623 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
624 | |
625 | return IRQ_HANDLED; | |
626 | } | |
c6fd4701 | 627 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd | 628 | |
17bb6d40 JW |
629 | struct virtqueue *vring_new_virtqueue(unsigned int index, |
630 | unsigned int num, | |
87c7d57c | 631 | unsigned int vring_align, |
0a8a69dd | 632 | struct virtio_device *vdev, |
7b21e34f | 633 | bool weak_barriers, |
0a8a69dd RR |
634 | void *pages, |
635 | void (*notify)(struct virtqueue *), | |
9499f5e7 RR |
636 | void (*callback)(struct virtqueue *), |
637 | const char *name) | |
0a8a69dd RR |
638 | { |
639 | struct vring_virtqueue *vq; | |
640 | unsigned int i; | |
641 | ||
42b36cc0 RR |
642 | /* We assume num is a power of 2. */ |
643 | if (num & (num - 1)) { | |
644 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
645 | return NULL; | |
646 | } | |
647 | ||
0a8a69dd RR |
648 | vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL); |
649 | if (!vq) | |
650 | return NULL; | |
651 | ||
87c7d57c | 652 | vring_init(&vq->vring, num, pages, vring_align); |
0a8a69dd RR |
653 | vq->vq.callback = callback; |
654 | vq->vq.vdev = vdev; | |
9499f5e7 | 655 | vq->vq.name = name; |
0a8a69dd | 656 | vq->notify = notify; |
7b21e34f | 657 | vq->weak_barriers = weak_barriers; |
0a8a69dd RR |
658 | vq->broken = false; |
659 | vq->last_used_idx = 0; | |
660 | vq->num_added = 0; | |
17bb6d40 | 661 | vq->queue_index = index; |
9499f5e7 | 662 | list_add_tail(&vq->vq.list, &vdev->vqs); |
0a8a69dd RR |
663 | #ifdef DEBUG |
664 | vq->in_use = false; | |
e93300b1 | 665 | vq->last_add_time_valid = false; |
0a8a69dd RR |
666 | #endif |
667 | ||
9fa29b9d | 668 | vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC); |
a5c262c5 | 669 | vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); |
9fa29b9d | 670 | |
0a8a69dd RR |
671 | /* No callback? Tell other side not to bother us. */ |
672 | if (!callback) | |
673 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; | |
674 | ||
675 | /* Put everything in free lists. */ | |
676 | vq->num_free = num; | |
677 | vq->free_head = 0; | |
3b870624 | 678 | for (i = 0; i < num-1; i++) { |
0a8a69dd | 679 | vq->vring.desc[i].next = i+1; |
3b870624 AS |
680 | vq->data[i] = NULL; |
681 | } | |
682 | vq->data[i] = NULL; | |
0a8a69dd RR |
683 | |
684 | return &vq->vq; | |
685 | } | |
c6fd4701 | 686 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd RR |
687 | |
688 | void vring_del_virtqueue(struct virtqueue *vq) | |
689 | { | |
9499f5e7 | 690 | list_del(&vq->list); |
0a8a69dd RR |
691 | kfree(to_vvq(vq)); |
692 | } | |
c6fd4701 | 693 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 694 | |
e34f8725 RR |
695 | /* Manipulates transport-specific feature bits. */ |
696 | void vring_transport_features(struct virtio_device *vdev) | |
697 | { | |
698 | unsigned int i; | |
699 | ||
700 | for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { | |
701 | switch (i) { | |
9fa29b9d MM |
702 | case VIRTIO_RING_F_INDIRECT_DESC: |
703 | break; | |
a5c262c5 MT |
704 | case VIRTIO_RING_F_EVENT_IDX: |
705 | break; | |
e34f8725 RR |
706 | default: |
707 | /* We don't understand this bit. */ | |
708 | clear_bit(i, vdev->features); | |
709 | } | |
710 | } | |
711 | } | |
712 | EXPORT_SYMBOL_GPL(vring_transport_features); | |
713 | ||
5dfc1762 RR |
714 | /** |
715 | * virtqueue_get_vring_size - return the size of the virtqueue's vring | |
716 | * @vq: the struct virtqueue containing the vring of interest. | |
717 | * | |
718 | * Returns the size of the vring. This is mainly used for boasting to | |
719 | * userspace. Unlike other operations, this need not be serialized. | |
720 | */ | |
8f9f4668 RJ |
721 | unsigned int virtqueue_get_vring_size(struct virtqueue *_vq) |
722 | { | |
723 | ||
724 | struct vring_virtqueue *vq = to_vvq(_vq); | |
725 | ||
726 | return vq->vring.num; | |
727 | } | |
728 | EXPORT_SYMBOL_GPL(virtqueue_get_vring_size); | |
729 | ||
c6fd4701 | 730 | MODULE_LICENSE("GPL"); |