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 | /* Head of free buffer list. */ |
97 | unsigned int free_head; | |
98 | /* Number we've added since last sync. */ | |
99 | unsigned int num_added; | |
100 | ||
101 | /* Last used index we've seen. */ | |
1bc4953e | 102 | u16 last_used_idx; |
0a8a69dd RR |
103 | |
104 | /* How to notify other side. FIXME: commonalize hcalls! */ | |
105 | void (*notify)(struct virtqueue *vq); | |
106 | ||
107 | #ifdef DEBUG | |
108 | /* They're supposed to lock for us. */ | |
109 | unsigned int in_use; | |
e93300b1 RR |
110 | |
111 | /* Figure out if their kicks are too delayed. */ | |
112 | bool last_add_time_valid; | |
113 | ktime_t last_add_time; | |
0a8a69dd RR |
114 | #endif |
115 | ||
116 | /* Tokens for callbacks. */ | |
117 | void *data[]; | |
118 | }; | |
119 | ||
120 | #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq) | |
121 | ||
9fa29b9d MM |
122 | /* Set up an indirect table of descriptors and add it to the queue. */ |
123 | static int vring_add_indirect(struct vring_virtqueue *vq, | |
124 | struct scatterlist sg[], | |
125 | unsigned int out, | |
bbd603ef MT |
126 | unsigned int in, |
127 | gfp_t gfp) | |
9fa29b9d MM |
128 | { |
129 | struct vring_desc *desc; | |
130 | unsigned head; | |
131 | int i; | |
132 | ||
b92b1b89 WD |
133 | /* |
134 | * We require lowmem mappings for the descriptors because | |
135 | * otherwise virt_to_phys will give us bogus addresses in the | |
136 | * virtqueue. | |
137 | */ | |
138 | gfp &= ~(__GFP_HIGHMEM | __GFP_HIGH); | |
139 | ||
bbd603ef | 140 | desc = kmalloc((out + in) * sizeof(struct vring_desc), gfp); |
9fa29b9d | 141 | if (!desc) |
686d3637 | 142 | return -ENOMEM; |
9fa29b9d MM |
143 | |
144 | /* Transfer entries from the sg list into the indirect page */ | |
145 | for (i = 0; i < out; i++) { | |
146 | desc[i].flags = VRING_DESC_F_NEXT; | |
147 | desc[i].addr = sg_phys(sg); | |
148 | desc[i].len = sg->length; | |
149 | desc[i].next = i+1; | |
150 | sg++; | |
151 | } | |
152 | for (; i < (out + in); i++) { | |
153 | desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; | |
154 | desc[i].addr = sg_phys(sg); | |
155 | desc[i].len = sg->length; | |
156 | desc[i].next = i+1; | |
157 | sg++; | |
158 | } | |
159 | ||
160 | /* Last one doesn't continue. */ | |
161 | desc[i-1].flags &= ~VRING_DESC_F_NEXT; | |
162 | desc[i-1].next = 0; | |
163 | ||
164 | /* We're about to use a buffer */ | |
06ca287d | 165 | vq->vq.num_free--; |
9fa29b9d MM |
166 | |
167 | /* Use a single buffer which doesn't continue */ | |
168 | head = vq->free_head; | |
169 | vq->vring.desc[head].flags = VRING_DESC_F_INDIRECT; | |
170 | vq->vring.desc[head].addr = virt_to_phys(desc); | |
171 | vq->vring.desc[head].len = i * sizeof(struct vring_desc); | |
172 | ||
173 | /* Update free pointer */ | |
174 | vq->free_head = vq->vring.desc[head].next; | |
175 | ||
176 | return head; | |
177 | } | |
178 | ||
5dfc1762 | 179 | /** |
f96fde41 | 180 | * virtqueue_add_buf - expose buffer to other end |
5dfc1762 RR |
181 | * @vq: the struct virtqueue we're talking about. |
182 | * @sg: the description of the buffer(s). | |
183 | * @out_num: the number of sg readable by other side | |
184 | * @in_num: the number of sg which are writable (after readable ones) | |
185 | * @data: the token identifying the buffer. | |
186 | * @gfp: how to do memory allocations (if necessary). | |
187 | * | |
188 | * Caller must ensure we don't call this with other virtqueue operations | |
189 | * at the same time (except where noted). | |
190 | * | |
191 | * Returns remaining capacity of queue or a negative error | |
192 | * (ie. ENOSPC). Note that it only really makes sense to treat all | |
193 | * positive return values as "available": indirect buffers mean that | |
194 | * we can put an entire sg[] array inside a single queue entry. | |
195 | */ | |
f96fde41 RR |
196 | int virtqueue_add_buf(struct virtqueue *_vq, |
197 | struct scatterlist sg[], | |
198 | unsigned int out, | |
199 | unsigned int in, | |
200 | void *data, | |
201 | gfp_t gfp) | |
0a8a69dd RR |
202 | { |
203 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1fe9b6fe MT |
204 | unsigned int i, avail, uninitialized_var(prev); |
205 | int head; | |
0a8a69dd | 206 | |
9fa29b9d MM |
207 | START_USE(vq); |
208 | ||
0a8a69dd | 209 | BUG_ON(data == NULL); |
9fa29b9d | 210 | |
e93300b1 RR |
211 | #ifdef DEBUG |
212 | { | |
213 | ktime_t now = ktime_get(); | |
214 | ||
215 | /* No kick or get, with .1 second between? Warn. */ | |
216 | if (vq->last_add_time_valid) | |
217 | WARN_ON(ktime_to_ms(ktime_sub(now, vq->last_add_time)) | |
218 | > 100); | |
219 | vq->last_add_time = now; | |
220 | vq->last_add_time_valid = true; | |
221 | } | |
222 | #endif | |
223 | ||
9fa29b9d MM |
224 | /* If the host supports indirect descriptor tables, and we have multiple |
225 | * buffers, then go indirect. FIXME: tune this threshold */ | |
06ca287d | 226 | if (vq->indirect && (out + in) > 1 && vq->vq.num_free) { |
bbd603ef | 227 | head = vring_add_indirect(vq, sg, out, in, gfp); |
1fe9b6fe | 228 | if (likely(head >= 0)) |
9fa29b9d MM |
229 | goto add_head; |
230 | } | |
231 | ||
0a8a69dd RR |
232 | BUG_ON(out + in > vq->vring.num); |
233 | BUG_ON(out + in == 0); | |
234 | ||
06ca287d | 235 | if (vq->vq.num_free < out + in) { |
0a8a69dd | 236 | pr_debug("Can't add buf len %i - avail = %i\n", |
06ca287d | 237 | out + in, vq->vq.num_free); |
44653eae RR |
238 | /* FIXME: for historical reasons, we force a notify here if |
239 | * there are outgoing parts to the buffer. Presumably the | |
240 | * host should service the ring ASAP. */ | |
241 | if (out) | |
242 | vq->notify(&vq->vq); | |
0a8a69dd RR |
243 | END_USE(vq); |
244 | return -ENOSPC; | |
245 | } | |
246 | ||
247 | /* We're about to use some buffers from the free list. */ | |
06ca287d | 248 | vq->vq.num_free -= out + in; |
0a8a69dd RR |
249 | |
250 | head = vq->free_head; | |
251 | for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) { | |
252 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT; | |
15f9c890 | 253 | vq->vring.desc[i].addr = sg_phys(sg); |
0a8a69dd RR |
254 | vq->vring.desc[i].len = sg->length; |
255 | prev = i; | |
256 | sg++; | |
257 | } | |
258 | for (; in; i = vq->vring.desc[i].next, in--) { | |
259 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; | |
15f9c890 | 260 | vq->vring.desc[i].addr = sg_phys(sg); |
0a8a69dd RR |
261 | vq->vring.desc[i].len = sg->length; |
262 | prev = i; | |
263 | sg++; | |
264 | } | |
265 | /* Last one doesn't continue. */ | |
266 | vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT; | |
267 | ||
268 | /* Update free pointer */ | |
269 | vq->free_head = i; | |
270 | ||
9fa29b9d | 271 | add_head: |
0a8a69dd RR |
272 | /* Set token. */ |
273 | vq->data[head] = data; | |
274 | ||
275 | /* Put entry in available array (but don't update avail->idx until they | |
3b720b8c | 276 | * do sync). */ |
ee7cd898 | 277 | avail = (vq->vring.avail->idx & (vq->vring.num-1)); |
0a8a69dd RR |
278 | vq->vring.avail->ring[avail] = head; |
279 | ||
ee7cd898 RR |
280 | /* Descriptors and available array need to be set before we expose the |
281 | * new available array entries. */ | |
282 | virtio_wmb(vq); | |
283 | vq->vring.avail->idx++; | |
284 | vq->num_added++; | |
285 | ||
286 | /* This is very unlikely, but theoretically possible. Kick | |
287 | * just in case. */ | |
288 | if (unlikely(vq->num_added == (1 << 16) - 1)) | |
289 | virtqueue_kick(_vq); | |
290 | ||
0a8a69dd RR |
291 | pr_debug("Added buffer head %i to %p\n", head, vq); |
292 | END_USE(vq); | |
3c1b27d5 | 293 | |
06ca287d | 294 | return vq->vq.num_free; |
0a8a69dd | 295 | } |
f96fde41 | 296 | EXPORT_SYMBOL_GPL(virtqueue_add_buf); |
0a8a69dd | 297 | |
5dfc1762 | 298 | /** |
41f0377f | 299 | * virtqueue_kick_prepare - first half of split virtqueue_kick call. |
5dfc1762 RR |
300 | * @vq: the struct virtqueue |
301 | * | |
41f0377f RR |
302 | * Instead of virtqueue_kick(), you can do: |
303 | * if (virtqueue_kick_prepare(vq)) | |
304 | * virtqueue_notify(vq); | |
5dfc1762 | 305 | * |
41f0377f RR |
306 | * This is sometimes useful because the virtqueue_kick_prepare() needs |
307 | * to be serialized, but the actual virtqueue_notify() call does not. | |
5dfc1762 | 308 | */ |
41f0377f | 309 | bool virtqueue_kick_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
310 | { |
311 | struct vring_virtqueue *vq = to_vvq(_vq); | |
a5c262c5 | 312 | u16 new, old; |
41f0377f RR |
313 | bool needs_kick; |
314 | ||
0a8a69dd | 315 | START_USE(vq); |
a72caae2 JW |
316 | /* We need to expose available array entries before checking avail |
317 | * event. */ | |
318 | virtio_mb(vq); | |
0a8a69dd | 319 | |
ee7cd898 RR |
320 | old = vq->vring.avail->idx - vq->num_added; |
321 | new = vq->vring.avail->idx; | |
0a8a69dd RR |
322 | vq->num_added = 0; |
323 | ||
e93300b1 RR |
324 | #ifdef DEBUG |
325 | if (vq->last_add_time_valid) { | |
326 | WARN_ON(ktime_to_ms(ktime_sub(ktime_get(), | |
327 | vq->last_add_time)) > 100); | |
328 | } | |
329 | vq->last_add_time_valid = false; | |
330 | #endif | |
331 | ||
41f0377f RR |
332 | if (vq->event) { |
333 | needs_kick = vring_need_event(vring_avail_event(&vq->vring), | |
334 | new, old); | |
335 | } else { | |
336 | needs_kick = !(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY); | |
337 | } | |
0a8a69dd | 338 | END_USE(vq); |
41f0377f RR |
339 | return needs_kick; |
340 | } | |
341 | EXPORT_SYMBOL_GPL(virtqueue_kick_prepare); | |
342 | ||
343 | /** | |
344 | * virtqueue_notify - second half of split virtqueue_kick call. | |
345 | * @vq: the struct virtqueue | |
346 | * | |
347 | * This does not need to be serialized. | |
348 | */ | |
349 | void virtqueue_notify(struct virtqueue *_vq) | |
350 | { | |
351 | struct vring_virtqueue *vq = to_vvq(_vq); | |
352 | ||
353 | /* Prod other side to tell it about changes. */ | |
354 | vq->notify(_vq); | |
355 | } | |
356 | EXPORT_SYMBOL_GPL(virtqueue_notify); | |
357 | ||
358 | /** | |
359 | * virtqueue_kick - update after add_buf | |
360 | * @vq: the struct virtqueue | |
361 | * | |
362 | * After one or more virtqueue_add_buf calls, invoke this to kick | |
363 | * the other side. | |
364 | * | |
365 | * Caller must ensure we don't call this with other virtqueue | |
366 | * operations at the same time (except where noted). | |
367 | */ | |
368 | void virtqueue_kick(struct virtqueue *vq) | |
369 | { | |
370 | if (virtqueue_kick_prepare(vq)) | |
371 | virtqueue_notify(vq); | |
0a8a69dd | 372 | } |
7c5e9ed0 | 373 | EXPORT_SYMBOL_GPL(virtqueue_kick); |
0a8a69dd RR |
374 | |
375 | static void detach_buf(struct vring_virtqueue *vq, unsigned int head) | |
376 | { | |
377 | unsigned int i; | |
378 | ||
379 | /* Clear data ptr. */ | |
380 | vq->data[head] = NULL; | |
381 | ||
382 | /* Put back on free list: find end */ | |
383 | i = head; | |
9fa29b9d MM |
384 | |
385 | /* Free the indirect table */ | |
386 | if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT) | |
387 | kfree(phys_to_virt(vq->vring.desc[i].addr)); | |
388 | ||
0a8a69dd RR |
389 | while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) { |
390 | i = vq->vring.desc[i].next; | |
06ca287d | 391 | vq->vq.num_free++; |
0a8a69dd RR |
392 | } |
393 | ||
394 | vq->vring.desc[i].next = vq->free_head; | |
395 | vq->free_head = head; | |
396 | /* Plus final descriptor */ | |
06ca287d | 397 | vq->vq.num_free++; |
0a8a69dd RR |
398 | } |
399 | ||
0a8a69dd RR |
400 | static inline bool more_used(const struct vring_virtqueue *vq) |
401 | { | |
402 | return vq->last_used_idx != vq->vring.used->idx; | |
403 | } | |
404 | ||
5dfc1762 RR |
405 | /** |
406 | * virtqueue_get_buf - get the next used buffer | |
407 | * @vq: the struct virtqueue we're talking about. | |
408 | * @len: the length written into the buffer | |
409 | * | |
410 | * If the driver wrote data into the buffer, @len will be set to the | |
411 | * amount written. This means you don't need to clear the buffer | |
412 | * beforehand to ensure there's no data leakage in the case of short | |
413 | * writes. | |
414 | * | |
415 | * Caller must ensure we don't call this with other virtqueue | |
416 | * operations at the same time (except where noted). | |
417 | * | |
418 | * Returns NULL if there are no used buffers, or the "data" token | |
f96fde41 | 419 | * handed to virtqueue_add_buf(). |
5dfc1762 | 420 | */ |
7c5e9ed0 | 421 | void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len) |
0a8a69dd RR |
422 | { |
423 | struct vring_virtqueue *vq = to_vvq(_vq); | |
424 | void *ret; | |
425 | unsigned int i; | |
3b720b8c | 426 | u16 last_used; |
0a8a69dd RR |
427 | |
428 | START_USE(vq); | |
429 | ||
5ef82752 RR |
430 | if (unlikely(vq->broken)) { |
431 | END_USE(vq); | |
432 | return NULL; | |
433 | } | |
434 | ||
0a8a69dd RR |
435 | if (!more_used(vq)) { |
436 | pr_debug("No more buffers in queue\n"); | |
437 | END_USE(vq); | |
438 | return NULL; | |
439 | } | |
440 | ||
2d61ba95 | 441 | /* Only get used array entries after they have been exposed by host. */ |
7b21e34f | 442 | virtio_rmb(vq); |
2d61ba95 | 443 | |
3b720b8c RR |
444 | last_used = (vq->last_used_idx & (vq->vring.num - 1)); |
445 | i = vq->vring.used->ring[last_used].id; | |
446 | *len = vq->vring.used->ring[last_used].len; | |
0a8a69dd RR |
447 | |
448 | if (unlikely(i >= vq->vring.num)) { | |
449 | BAD_RING(vq, "id %u out of range\n", i); | |
450 | return NULL; | |
451 | } | |
452 | if (unlikely(!vq->data[i])) { | |
453 | BAD_RING(vq, "id %u is not a head!\n", i); | |
454 | return NULL; | |
455 | } | |
456 | ||
457 | /* detach_buf clears data, so grab it now. */ | |
458 | ret = vq->data[i]; | |
459 | detach_buf(vq, i); | |
460 | vq->last_used_idx++; | |
a5c262c5 MT |
461 | /* If we expect an interrupt for the next entry, tell host |
462 | * by writing event index and flush out the write before | |
463 | * the read in the next get_buf call. */ | |
464 | if (!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) { | |
465 | vring_used_event(&vq->vring) = vq->last_used_idx; | |
7b21e34f | 466 | virtio_mb(vq); |
a5c262c5 MT |
467 | } |
468 | ||
e93300b1 RR |
469 | #ifdef DEBUG |
470 | vq->last_add_time_valid = false; | |
471 | #endif | |
472 | ||
0a8a69dd RR |
473 | END_USE(vq); |
474 | return ret; | |
475 | } | |
7c5e9ed0 | 476 | EXPORT_SYMBOL_GPL(virtqueue_get_buf); |
0a8a69dd | 477 | |
5dfc1762 RR |
478 | /** |
479 | * virtqueue_disable_cb - disable callbacks | |
480 | * @vq: the struct virtqueue we're talking about. | |
481 | * | |
482 | * Note that this is not necessarily synchronous, hence unreliable and only | |
483 | * useful as an optimization. | |
484 | * | |
485 | * Unlike other operations, this need not be serialized. | |
486 | */ | |
7c5e9ed0 | 487 | void virtqueue_disable_cb(struct virtqueue *_vq) |
18445c4d RR |
488 | { |
489 | struct vring_virtqueue *vq = to_vvq(_vq); | |
490 | ||
18445c4d | 491 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; |
18445c4d | 492 | } |
7c5e9ed0 | 493 | EXPORT_SYMBOL_GPL(virtqueue_disable_cb); |
18445c4d | 494 | |
5dfc1762 RR |
495 | /** |
496 | * virtqueue_enable_cb - restart callbacks after disable_cb. | |
497 | * @vq: the struct virtqueue we're talking about. | |
498 | * | |
499 | * This re-enables callbacks; it returns "false" if there are pending | |
500 | * buffers in the queue, to detect a possible race between the driver | |
501 | * checking for more work, and enabling callbacks. | |
502 | * | |
503 | * Caller must ensure we don't call this with other virtqueue | |
504 | * operations at the same time (except where noted). | |
505 | */ | |
7c5e9ed0 | 506 | bool virtqueue_enable_cb(struct virtqueue *_vq) |
0a8a69dd RR |
507 | { |
508 | struct vring_virtqueue *vq = to_vvq(_vq); | |
509 | ||
510 | START_USE(vq); | |
0a8a69dd RR |
511 | |
512 | /* We optimistically turn back on interrupts, then check if there was | |
513 | * more to do. */ | |
a5c262c5 MT |
514 | /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to |
515 | * either clear the flags bit or point the event index at the next | |
516 | * entry. Always do both to keep code simple. */ | |
0a8a69dd | 517 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; |
a5c262c5 | 518 | vring_used_event(&vq->vring) = vq->last_used_idx; |
7b21e34f | 519 | virtio_mb(vq); |
0a8a69dd | 520 | if (unlikely(more_used(vq))) { |
0a8a69dd RR |
521 | END_USE(vq); |
522 | return false; | |
523 | } | |
524 | ||
525 | END_USE(vq); | |
526 | return true; | |
527 | } | |
7c5e9ed0 | 528 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb); |
0a8a69dd | 529 | |
5dfc1762 RR |
530 | /** |
531 | * virtqueue_enable_cb_delayed - restart callbacks after disable_cb. | |
532 | * @vq: the struct virtqueue we're talking about. | |
533 | * | |
534 | * This re-enables callbacks but hints to the other side to delay | |
535 | * interrupts until most of the available buffers have been processed; | |
536 | * it returns "false" if there are many pending buffers in the queue, | |
537 | * to detect a possible race between the driver checking for more work, | |
538 | * and enabling callbacks. | |
539 | * | |
540 | * Caller must ensure we don't call this with other virtqueue | |
541 | * operations at the same time (except where noted). | |
542 | */ | |
7ab358c2 MT |
543 | bool virtqueue_enable_cb_delayed(struct virtqueue *_vq) |
544 | { | |
545 | struct vring_virtqueue *vq = to_vvq(_vq); | |
546 | u16 bufs; | |
547 | ||
548 | START_USE(vq); | |
549 | ||
550 | /* We optimistically turn back on interrupts, then check if there was | |
551 | * more to do. */ | |
552 | /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to | |
553 | * either clear the flags bit or point the event index at the next | |
554 | * entry. Always do both to keep code simple. */ | |
555 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
556 | /* TODO: tune this threshold */ | |
557 | bufs = (u16)(vq->vring.avail->idx - vq->last_used_idx) * 3 / 4; | |
558 | vring_used_event(&vq->vring) = vq->last_used_idx + bufs; | |
7b21e34f | 559 | virtio_mb(vq); |
7ab358c2 MT |
560 | if (unlikely((u16)(vq->vring.used->idx - vq->last_used_idx) > bufs)) { |
561 | END_USE(vq); | |
562 | return false; | |
563 | } | |
564 | ||
565 | END_USE(vq); | |
566 | return true; | |
567 | } | |
568 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed); | |
569 | ||
5dfc1762 RR |
570 | /** |
571 | * virtqueue_detach_unused_buf - detach first unused buffer | |
572 | * @vq: the struct virtqueue we're talking about. | |
573 | * | |
f96fde41 | 574 | * Returns NULL or the "data" token handed to virtqueue_add_buf(). |
5dfc1762 RR |
575 | * This is not valid on an active queue; it is useful only for device |
576 | * shutdown. | |
577 | */ | |
7c5e9ed0 | 578 | void *virtqueue_detach_unused_buf(struct virtqueue *_vq) |
c021eac4 SM |
579 | { |
580 | struct vring_virtqueue *vq = to_vvq(_vq); | |
581 | unsigned int i; | |
582 | void *buf; | |
583 | ||
584 | START_USE(vq); | |
585 | ||
586 | for (i = 0; i < vq->vring.num; i++) { | |
587 | if (!vq->data[i]) | |
588 | continue; | |
589 | /* detach_buf clears data, so grab it now. */ | |
590 | buf = vq->data[i]; | |
591 | detach_buf(vq, i); | |
b3258ff1 | 592 | vq->vring.avail->idx--; |
c021eac4 SM |
593 | END_USE(vq); |
594 | return buf; | |
595 | } | |
596 | /* That should have freed everything. */ | |
06ca287d | 597 | BUG_ON(vq->vq.num_free != vq->vring.num); |
c021eac4 SM |
598 | |
599 | END_USE(vq); | |
600 | return NULL; | |
601 | } | |
7c5e9ed0 | 602 | EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf); |
c021eac4 | 603 | |
0a8a69dd RR |
604 | irqreturn_t vring_interrupt(int irq, void *_vq) |
605 | { | |
606 | struct vring_virtqueue *vq = to_vvq(_vq); | |
607 | ||
608 | if (!more_used(vq)) { | |
609 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
610 | return IRQ_NONE; | |
611 | } | |
612 | ||
613 | if (unlikely(vq->broken)) | |
614 | return IRQ_HANDLED; | |
615 | ||
616 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); | |
18445c4d RR |
617 | if (vq->vq.callback) |
618 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
619 | |
620 | return IRQ_HANDLED; | |
621 | } | |
c6fd4701 | 622 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd | 623 | |
17bb6d40 JW |
624 | struct virtqueue *vring_new_virtqueue(unsigned int index, |
625 | unsigned int num, | |
87c7d57c | 626 | unsigned int vring_align, |
0a8a69dd | 627 | struct virtio_device *vdev, |
7b21e34f | 628 | bool weak_barriers, |
0a8a69dd RR |
629 | void *pages, |
630 | void (*notify)(struct virtqueue *), | |
9499f5e7 RR |
631 | void (*callback)(struct virtqueue *), |
632 | const char *name) | |
0a8a69dd RR |
633 | { |
634 | struct vring_virtqueue *vq; | |
635 | unsigned int i; | |
636 | ||
42b36cc0 RR |
637 | /* We assume num is a power of 2. */ |
638 | if (num & (num - 1)) { | |
639 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
640 | return NULL; | |
641 | } | |
642 | ||
0a8a69dd RR |
643 | vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL); |
644 | if (!vq) | |
645 | return NULL; | |
646 | ||
87c7d57c | 647 | vring_init(&vq->vring, num, pages, vring_align); |
0a8a69dd RR |
648 | vq->vq.callback = callback; |
649 | vq->vq.vdev = vdev; | |
9499f5e7 | 650 | vq->vq.name = name; |
06ca287d RR |
651 | vq->vq.num_free = num; |
652 | vq->vq.index = index; | |
0a8a69dd | 653 | vq->notify = notify; |
7b21e34f | 654 | vq->weak_barriers = weak_barriers; |
0a8a69dd RR |
655 | vq->broken = false; |
656 | vq->last_used_idx = 0; | |
657 | vq->num_added = 0; | |
9499f5e7 | 658 | list_add_tail(&vq->vq.list, &vdev->vqs); |
0a8a69dd RR |
659 | #ifdef DEBUG |
660 | vq->in_use = false; | |
e93300b1 | 661 | vq->last_add_time_valid = false; |
0a8a69dd RR |
662 | #endif |
663 | ||
9fa29b9d | 664 | vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC); |
a5c262c5 | 665 | vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); |
9fa29b9d | 666 | |
0a8a69dd RR |
667 | /* No callback? Tell other side not to bother us. */ |
668 | if (!callback) | |
669 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; | |
670 | ||
671 | /* Put everything in free lists. */ | |
0a8a69dd | 672 | vq->free_head = 0; |
3b870624 | 673 | for (i = 0; i < num-1; i++) { |
0a8a69dd | 674 | vq->vring.desc[i].next = i+1; |
3b870624 AS |
675 | vq->data[i] = NULL; |
676 | } | |
677 | vq->data[i] = NULL; | |
0a8a69dd RR |
678 | |
679 | return &vq->vq; | |
680 | } | |
c6fd4701 | 681 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd RR |
682 | |
683 | void vring_del_virtqueue(struct virtqueue *vq) | |
684 | { | |
9499f5e7 | 685 | list_del(&vq->list); |
0a8a69dd RR |
686 | kfree(to_vvq(vq)); |
687 | } | |
c6fd4701 | 688 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 689 | |
e34f8725 RR |
690 | /* Manipulates transport-specific feature bits. */ |
691 | void vring_transport_features(struct virtio_device *vdev) | |
692 | { | |
693 | unsigned int i; | |
694 | ||
695 | for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { | |
696 | switch (i) { | |
9fa29b9d MM |
697 | case VIRTIO_RING_F_INDIRECT_DESC: |
698 | break; | |
a5c262c5 MT |
699 | case VIRTIO_RING_F_EVENT_IDX: |
700 | break; | |
e34f8725 RR |
701 | default: |
702 | /* We don't understand this bit. */ | |
703 | clear_bit(i, vdev->features); | |
704 | } | |
705 | } | |
706 | } | |
707 | EXPORT_SYMBOL_GPL(vring_transport_features); | |
708 | ||
5dfc1762 RR |
709 | /** |
710 | * virtqueue_get_vring_size - return the size of the virtqueue's vring | |
711 | * @vq: the struct virtqueue containing the vring of interest. | |
712 | * | |
713 | * Returns the size of the vring. This is mainly used for boasting to | |
714 | * userspace. Unlike other operations, this need not be serialized. | |
715 | */ | |
8f9f4668 RJ |
716 | unsigned int virtqueue_get_vring_size(struct virtqueue *_vq) |
717 | { | |
718 | ||
719 | struct vring_virtqueue *vq = to_vvq(_vq); | |
720 | ||
721 | return vq->vring.num; | |
722 | } | |
723 | EXPORT_SYMBOL_GPL(virtqueue_get_vring_size); | |
724 | ||
c6fd4701 | 725 | MODULE_LICENSE("GPL"); |