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> |
6abb2dd9 | 26 | #include <linux/kmemleak.h> |
780bc790 | 27 | #include <linux/dma-mapping.h> |
78fe3987 | 28 | #include <xen/xen.h> |
0a8a69dd RR |
29 | |
30 | #ifdef DEBUG | |
31 | /* For development, we want to crash whenever the ring is screwed. */ | |
9499f5e7 RR |
32 | #define BAD_RING(_vq, fmt, args...) \ |
33 | do { \ | |
34 | dev_err(&(_vq)->vq.vdev->dev, \ | |
35 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
36 | BUG(); \ | |
37 | } while (0) | |
c5f841f1 RR |
38 | /* Caller is supposed to guarantee no reentry. */ |
39 | #define START_USE(_vq) \ | |
40 | do { \ | |
41 | if ((_vq)->in_use) \ | |
9499f5e7 RR |
42 | panic("%s:in_use = %i\n", \ |
43 | (_vq)->vq.name, (_vq)->in_use); \ | |
c5f841f1 | 44 | (_vq)->in_use = __LINE__; \ |
9499f5e7 | 45 | } while (0) |
3a35ce7d | 46 | #define END_USE(_vq) \ |
97a545ab | 47 | do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; } while(0) |
0a8a69dd | 48 | #else |
9499f5e7 RR |
49 | #define BAD_RING(_vq, fmt, args...) \ |
50 | do { \ | |
51 | dev_err(&_vq->vq.vdev->dev, \ | |
52 | "%s:"fmt, (_vq)->vq.name, ##args); \ | |
53 | (_vq)->broken = true; \ | |
54 | } while (0) | |
0a8a69dd RR |
55 | #define START_USE(vq) |
56 | #define END_USE(vq) | |
57 | #endif | |
58 | ||
780bc790 AL |
59 | struct vring_desc_state { |
60 | void *data; /* Data for callback. */ | |
61 | struct vring_desc *indir_desc; /* Indirect descriptor, if any. */ | |
62 | }; | |
63 | ||
43b4f721 | 64 | struct vring_virtqueue { |
0a8a69dd RR |
65 | struct virtqueue vq; |
66 | ||
67 | /* Actual memory layout for this queue */ | |
68 | struct vring vring; | |
69 | ||
7b21e34f RR |
70 | /* Can we use weak barriers? */ |
71 | bool weak_barriers; | |
72 | ||
0a8a69dd RR |
73 | /* Other side has made a mess, don't try any more. */ |
74 | bool broken; | |
75 | ||
9fa29b9d MM |
76 | /* Host supports indirect buffers */ |
77 | bool indirect; | |
78 | ||
a5c262c5 MT |
79 | /* Host publishes avail event idx */ |
80 | bool event; | |
81 | ||
0a8a69dd RR |
82 | /* Head of free buffer list. */ |
83 | unsigned int free_head; | |
84 | /* Number we've added since last sync. */ | |
85 | unsigned int num_added; | |
86 | ||
87 | /* Last used index we've seen. */ | |
1bc4953e | 88 | u16 last_used_idx; |
0a8a69dd | 89 | |
f277ec42 VS |
90 | /* Last written value to avail->flags */ |
91 | u16 avail_flags_shadow; | |
92 | ||
93 | /* Last written value to avail->idx in guest byte order */ | |
94 | u16 avail_idx_shadow; | |
95 | ||
0a8a69dd | 96 | /* How to notify other side. FIXME: commonalize hcalls! */ |
46f9c2b9 | 97 | bool (*notify)(struct virtqueue *vq); |
0a8a69dd | 98 | |
2a2d1382 AL |
99 | /* DMA, allocation, and size information */ |
100 | bool we_own_ring; | |
101 | size_t queue_size_in_bytes; | |
102 | dma_addr_t queue_dma_addr; | |
103 | ||
0a8a69dd RR |
104 | #ifdef DEBUG |
105 | /* They're supposed to lock for us. */ | |
106 | unsigned int in_use; | |
e93300b1 RR |
107 | |
108 | /* Figure out if their kicks are too delayed. */ | |
109 | bool last_add_time_valid; | |
110 | ktime_t last_add_time; | |
0a8a69dd RR |
111 | #endif |
112 | ||
780bc790 AL |
113 | /* Per-descriptor state. */ |
114 | struct vring_desc_state desc_state[]; | |
0a8a69dd RR |
115 | }; |
116 | ||
117 | #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq) | |
118 | ||
d26c96c8 | 119 | /* |
1a937693 MT |
120 | * Modern virtio devices have feature bits to specify whether they need a |
121 | * quirk and bypass the IOMMU. If not there, just use the DMA API. | |
122 | * | |
123 | * If there, the interaction between virtio and DMA API is messy. | |
d26c96c8 AL |
124 | * |
125 | * On most systems with virtio, physical addresses match bus addresses, | |
126 | * and it doesn't particularly matter whether we use the DMA API. | |
127 | * | |
128 | * On some systems, including Xen and any system with a physical device | |
129 | * that speaks virtio behind a physical IOMMU, we must use the DMA API | |
130 | * for virtio DMA to work at all. | |
131 | * | |
132 | * On other systems, including SPARC and PPC64, virtio-pci devices are | |
133 | * enumerated as though they are behind an IOMMU, but the virtio host | |
134 | * ignores the IOMMU, so we must either pretend that the IOMMU isn't | |
135 | * there or somehow map everything as the identity. | |
136 | * | |
137 | * For the time being, we preserve historic behavior and bypass the DMA | |
138 | * API. | |
1a937693 MT |
139 | * |
140 | * TODO: install a per-device DMA ops structure that does the right thing | |
141 | * taking into account all the above quirks, and use the DMA API | |
142 | * unconditionally on data path. | |
d26c96c8 AL |
143 | */ |
144 | ||
145 | static bool vring_use_dma_api(struct virtio_device *vdev) | |
146 | { | |
1a937693 MT |
147 | if (!virtio_has_iommu_quirk(vdev)) |
148 | return true; | |
149 | ||
150 | /* Otherwise, we are left to guess. */ | |
78fe3987 AL |
151 | /* |
152 | * In theory, it's possible to have a buggy QEMU-supposed | |
153 | * emulated Q35 IOMMU and Xen enabled at the same time. On | |
154 | * such a configuration, virtio has never worked and will | |
155 | * not work without an even larger kludge. Instead, enable | |
156 | * the DMA API if we're a Xen guest, which at least allows | |
157 | * all of the sensible Xen configurations to work correctly. | |
158 | */ | |
159 | if (xen_domain()) | |
160 | return true; | |
161 | ||
d26c96c8 AL |
162 | return false; |
163 | } | |
164 | ||
780bc790 AL |
165 | /* |
166 | * The DMA ops on various arches are rather gnarly right now, and | |
167 | * making all of the arch DMA ops work on the vring device itself | |
168 | * is a mess. For now, we use the parent device for DMA ops. | |
169 | */ | |
170 | struct device *vring_dma_dev(const struct vring_virtqueue *vq) | |
171 | { | |
172 | return vq->vq.vdev->dev.parent; | |
173 | } | |
174 | ||
175 | /* Map one sg entry. */ | |
176 | static dma_addr_t vring_map_one_sg(const struct vring_virtqueue *vq, | |
177 | struct scatterlist *sg, | |
178 | enum dma_data_direction direction) | |
179 | { | |
180 | if (!vring_use_dma_api(vq->vq.vdev)) | |
181 | return (dma_addr_t)sg_phys(sg); | |
182 | ||
183 | /* | |
184 | * We can't use dma_map_sg, because we don't use scatterlists in | |
185 | * the way it expects (we don't guarantee that the scatterlist | |
186 | * will exist for the lifetime of the mapping). | |
187 | */ | |
188 | return dma_map_page(vring_dma_dev(vq), | |
189 | sg_page(sg), sg->offset, sg->length, | |
190 | direction); | |
191 | } | |
192 | ||
193 | static dma_addr_t vring_map_single(const struct vring_virtqueue *vq, | |
194 | void *cpu_addr, size_t size, | |
195 | enum dma_data_direction direction) | |
196 | { | |
197 | if (!vring_use_dma_api(vq->vq.vdev)) | |
198 | return (dma_addr_t)virt_to_phys(cpu_addr); | |
199 | ||
200 | return dma_map_single(vring_dma_dev(vq), | |
201 | cpu_addr, size, direction); | |
202 | } | |
203 | ||
204 | static void vring_unmap_one(const struct vring_virtqueue *vq, | |
205 | struct vring_desc *desc) | |
206 | { | |
207 | u16 flags; | |
208 | ||
209 | if (!vring_use_dma_api(vq->vq.vdev)) | |
210 | return; | |
211 | ||
212 | flags = virtio16_to_cpu(vq->vq.vdev, desc->flags); | |
213 | ||
214 | if (flags & VRING_DESC_F_INDIRECT) { | |
215 | dma_unmap_single(vring_dma_dev(vq), | |
216 | virtio64_to_cpu(vq->vq.vdev, desc->addr), | |
217 | virtio32_to_cpu(vq->vq.vdev, desc->len), | |
218 | (flags & VRING_DESC_F_WRITE) ? | |
219 | DMA_FROM_DEVICE : DMA_TO_DEVICE); | |
220 | } else { | |
221 | dma_unmap_page(vring_dma_dev(vq), | |
222 | virtio64_to_cpu(vq->vq.vdev, desc->addr), | |
223 | virtio32_to_cpu(vq->vq.vdev, desc->len), | |
224 | (flags & VRING_DESC_F_WRITE) ? | |
225 | DMA_FROM_DEVICE : DMA_TO_DEVICE); | |
226 | } | |
227 | } | |
228 | ||
229 | static int vring_mapping_error(const struct vring_virtqueue *vq, | |
230 | dma_addr_t addr) | |
231 | { | |
232 | if (!vring_use_dma_api(vq->vq.vdev)) | |
233 | return 0; | |
234 | ||
235 | return dma_mapping_error(vring_dma_dev(vq), addr); | |
236 | } | |
237 | ||
00e6f3d9 MT |
238 | static struct vring_desc *alloc_indirect(struct virtqueue *_vq, |
239 | unsigned int total_sg, gfp_t gfp) | |
9fa29b9d MM |
240 | { |
241 | struct vring_desc *desc; | |
b25bd251 | 242 | unsigned int i; |
9fa29b9d | 243 | |
b92b1b89 WD |
244 | /* |
245 | * We require lowmem mappings for the descriptors because | |
246 | * otherwise virt_to_phys will give us bogus addresses in the | |
247 | * virtqueue. | |
248 | */ | |
82107539 | 249 | gfp &= ~__GFP_HIGHMEM; |
b92b1b89 | 250 | |
13816c76 | 251 | desc = kmalloc(total_sg * sizeof(struct vring_desc), gfp); |
9fa29b9d | 252 | if (!desc) |
b25bd251 | 253 | return NULL; |
9fa29b9d | 254 | |
b25bd251 | 255 | for (i = 0; i < total_sg; i++) |
00e6f3d9 | 256 | desc[i].next = cpu_to_virtio16(_vq->vdev, i + 1); |
b25bd251 | 257 | return desc; |
9fa29b9d MM |
258 | } |
259 | ||
13816c76 RR |
260 | static inline int virtqueue_add(struct virtqueue *_vq, |
261 | struct scatterlist *sgs[], | |
eeebf9b1 | 262 | unsigned int total_sg, |
13816c76 RR |
263 | unsigned int out_sgs, |
264 | unsigned int in_sgs, | |
265 | void *data, | |
266 | gfp_t gfp) | |
0a8a69dd RR |
267 | { |
268 | struct vring_virtqueue *vq = to_vvq(_vq); | |
13816c76 | 269 | struct scatterlist *sg; |
b25bd251 | 270 | struct vring_desc *desc; |
780bc790 | 271 | unsigned int i, n, avail, descs_used, uninitialized_var(prev), err_idx; |
1fe9b6fe | 272 | int head; |
b25bd251 | 273 | bool indirect; |
0a8a69dd | 274 | |
9fa29b9d MM |
275 | START_USE(vq); |
276 | ||
0a8a69dd | 277 | BUG_ON(data == NULL); |
9fa29b9d | 278 | |
70670444 RR |
279 | if (unlikely(vq->broken)) { |
280 | END_USE(vq); | |
281 | return -EIO; | |
282 | } | |
283 | ||
e93300b1 RR |
284 | #ifdef DEBUG |
285 | { | |
286 | ktime_t now = ktime_get(); | |
287 | ||
288 | /* No kick or get, with .1 second between? Warn. */ | |
289 | if (vq->last_add_time_valid) | |
290 | WARN_ON(ktime_to_ms(ktime_sub(now, vq->last_add_time)) | |
291 | > 100); | |
292 | vq->last_add_time = now; | |
293 | vq->last_add_time_valid = true; | |
294 | } | |
295 | #endif | |
296 | ||
b25bd251 RR |
297 | BUG_ON(total_sg > vq->vring.num); |
298 | BUG_ON(total_sg == 0); | |
299 | ||
300 | head = vq->free_head; | |
301 | ||
9fa29b9d MM |
302 | /* If the host supports indirect descriptor tables, and we have multiple |
303 | * buffers, then go indirect. FIXME: tune this threshold */ | |
b25bd251 | 304 | if (vq->indirect && total_sg > 1 && vq->vq.num_free) |
00e6f3d9 | 305 | desc = alloc_indirect(_vq, total_sg, gfp); |
b25bd251 RR |
306 | else |
307 | desc = NULL; | |
308 | ||
309 | if (desc) { | |
310 | /* Use a single buffer which doesn't continue */ | |
780bc790 | 311 | indirect = true; |
b25bd251 RR |
312 | /* Set up rest to use this indirect table. */ |
313 | i = 0; | |
314 | descs_used = 1; | |
b25bd251 | 315 | } else { |
780bc790 | 316 | indirect = false; |
b25bd251 RR |
317 | desc = vq->vring.desc; |
318 | i = head; | |
319 | descs_used = total_sg; | |
9fa29b9d MM |
320 | } |
321 | ||
b25bd251 | 322 | if (vq->vq.num_free < descs_used) { |
0a8a69dd | 323 | pr_debug("Can't add buf len %i - avail = %i\n", |
b25bd251 | 324 | descs_used, vq->vq.num_free); |
44653eae RR |
325 | /* FIXME: for historical reasons, we force a notify here if |
326 | * there are outgoing parts to the buffer. Presumably the | |
327 | * host should service the ring ASAP. */ | |
13816c76 | 328 | if (out_sgs) |
44653eae | 329 | vq->notify(&vq->vq); |
0a8a69dd RR |
330 | END_USE(vq); |
331 | return -ENOSPC; | |
332 | } | |
333 | ||
13816c76 | 334 | for (n = 0; n < out_sgs; n++) { |
eeebf9b1 | 335 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
780bc790 AL |
336 | dma_addr_t addr = vring_map_one_sg(vq, sg, DMA_TO_DEVICE); |
337 | if (vring_mapping_error(vq, addr)) | |
338 | goto unmap_release; | |
339 | ||
00e6f3d9 | 340 | desc[i].flags = cpu_to_virtio16(_vq->vdev, VRING_DESC_F_NEXT); |
780bc790 | 341 | desc[i].addr = cpu_to_virtio64(_vq->vdev, addr); |
00e6f3d9 | 342 | desc[i].len = cpu_to_virtio32(_vq->vdev, sg->length); |
13816c76 | 343 | prev = i; |
00e6f3d9 | 344 | i = virtio16_to_cpu(_vq->vdev, desc[i].next); |
13816c76 | 345 | } |
0a8a69dd | 346 | } |
13816c76 | 347 | for (; n < (out_sgs + in_sgs); n++) { |
eeebf9b1 | 348 | for (sg = sgs[n]; sg; sg = sg_next(sg)) { |
780bc790 AL |
349 | dma_addr_t addr = vring_map_one_sg(vq, sg, DMA_FROM_DEVICE); |
350 | if (vring_mapping_error(vq, addr)) | |
351 | goto unmap_release; | |
352 | ||
00e6f3d9 | 353 | desc[i].flags = cpu_to_virtio16(_vq->vdev, VRING_DESC_F_NEXT | VRING_DESC_F_WRITE); |
780bc790 | 354 | desc[i].addr = cpu_to_virtio64(_vq->vdev, addr); |
00e6f3d9 | 355 | desc[i].len = cpu_to_virtio32(_vq->vdev, sg->length); |
13816c76 | 356 | prev = i; |
00e6f3d9 | 357 | i = virtio16_to_cpu(_vq->vdev, desc[i].next); |
13816c76 | 358 | } |
0a8a69dd RR |
359 | } |
360 | /* Last one doesn't continue. */ | |
00e6f3d9 | 361 | desc[prev].flags &= cpu_to_virtio16(_vq->vdev, ~VRING_DESC_F_NEXT); |
0a8a69dd | 362 | |
780bc790 AL |
363 | if (indirect) { |
364 | /* Now that the indirect table is filled in, map it. */ | |
365 | dma_addr_t addr = vring_map_single( | |
366 | vq, desc, total_sg * sizeof(struct vring_desc), | |
367 | DMA_TO_DEVICE); | |
368 | if (vring_mapping_error(vq, addr)) | |
369 | goto unmap_release; | |
370 | ||
371 | vq->vring.desc[head].flags = cpu_to_virtio16(_vq->vdev, VRING_DESC_F_INDIRECT); | |
372 | vq->vring.desc[head].addr = cpu_to_virtio64(_vq->vdev, addr); | |
373 | ||
374 | vq->vring.desc[head].len = cpu_to_virtio32(_vq->vdev, total_sg * sizeof(struct vring_desc)); | |
375 | } | |
376 | ||
377 | /* We're using some buffers from the free list. */ | |
378 | vq->vq.num_free -= descs_used; | |
379 | ||
0a8a69dd | 380 | /* Update free pointer */ |
b25bd251 | 381 | if (indirect) |
00e6f3d9 | 382 | vq->free_head = virtio16_to_cpu(_vq->vdev, vq->vring.desc[head].next); |
b25bd251 RR |
383 | else |
384 | vq->free_head = i; | |
0a8a69dd | 385 | |
780bc790 AL |
386 | /* Store token and indirect buffer state. */ |
387 | vq->desc_state[head].data = data; | |
388 | if (indirect) | |
389 | vq->desc_state[head].indir_desc = desc; | |
0a8a69dd RR |
390 | |
391 | /* Put entry in available array (but don't update avail->idx until they | |
3b720b8c | 392 | * do sync). */ |
f277ec42 | 393 | avail = vq->avail_idx_shadow & (vq->vring.num - 1); |
00e6f3d9 | 394 | vq->vring.avail->ring[avail] = cpu_to_virtio16(_vq->vdev, head); |
0a8a69dd | 395 | |
ee7cd898 RR |
396 | /* Descriptors and available array need to be set before we expose the |
397 | * new available array entries. */ | |
a9a0fef7 | 398 | virtio_wmb(vq->weak_barriers); |
f277ec42 VS |
399 | vq->avail_idx_shadow++; |
400 | vq->vring.avail->idx = cpu_to_virtio16(_vq->vdev, vq->avail_idx_shadow); | |
ee7cd898 RR |
401 | vq->num_added++; |
402 | ||
5e05bf58 TH |
403 | pr_debug("Added buffer head %i to %p\n", head, vq); |
404 | END_USE(vq); | |
405 | ||
ee7cd898 RR |
406 | /* This is very unlikely, but theoretically possible. Kick |
407 | * just in case. */ | |
408 | if (unlikely(vq->num_added == (1 << 16) - 1)) | |
409 | virtqueue_kick(_vq); | |
410 | ||
98e8c6bc | 411 | return 0; |
780bc790 AL |
412 | |
413 | unmap_release: | |
414 | err_idx = i; | |
415 | i = head; | |
416 | ||
417 | for (n = 0; n < total_sg; n++) { | |
418 | if (i == err_idx) | |
419 | break; | |
420 | vring_unmap_one(vq, &desc[i]); | |
421 | i = vq->vring.desc[i].next; | |
422 | } | |
423 | ||
424 | vq->vq.num_free += total_sg; | |
425 | ||
426 | if (indirect) | |
427 | kfree(desc); | |
428 | ||
429 | return -EIO; | |
0a8a69dd | 430 | } |
13816c76 | 431 | |
13816c76 RR |
432 | /** |
433 | * virtqueue_add_sgs - expose buffers to other end | |
434 | * @vq: the struct virtqueue we're talking about. | |
435 | * @sgs: array of terminated scatterlists. | |
436 | * @out_num: the number of scatterlists readable by other side | |
437 | * @in_num: the number of scatterlists which are writable (after readable ones) | |
438 | * @data: the token identifying the buffer. | |
439 | * @gfp: how to do memory allocations (if necessary). | |
440 | * | |
441 | * Caller must ensure we don't call this with other virtqueue operations | |
442 | * at the same time (except where noted). | |
443 | * | |
70670444 | 444 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
13816c76 RR |
445 | */ |
446 | int virtqueue_add_sgs(struct virtqueue *_vq, | |
447 | struct scatterlist *sgs[], | |
448 | unsigned int out_sgs, | |
449 | unsigned int in_sgs, | |
450 | void *data, | |
451 | gfp_t gfp) | |
452 | { | |
eeebf9b1 | 453 | unsigned int i, total_sg = 0; |
13816c76 RR |
454 | |
455 | /* Count them first. */ | |
eeebf9b1 | 456 | for (i = 0; i < out_sgs + in_sgs; i++) { |
13816c76 RR |
457 | struct scatterlist *sg; |
458 | for (sg = sgs[i]; sg; sg = sg_next(sg)) | |
eeebf9b1 | 459 | total_sg++; |
13816c76 | 460 | } |
eeebf9b1 | 461 | return virtqueue_add(_vq, sgs, total_sg, out_sgs, in_sgs, data, gfp); |
13816c76 RR |
462 | } |
463 | EXPORT_SYMBOL_GPL(virtqueue_add_sgs); | |
464 | ||
282edb36 RR |
465 | /** |
466 | * virtqueue_add_outbuf - expose output buffers to other end | |
467 | * @vq: the struct virtqueue we're talking about. | |
eeebf9b1 RR |
468 | * @sg: scatterlist (must be well-formed and terminated!) |
469 | * @num: the number of entries in @sg readable by other side | |
282edb36 RR |
470 | * @data: the token identifying the buffer. |
471 | * @gfp: how to do memory allocations (if necessary). | |
472 | * | |
473 | * Caller must ensure we don't call this with other virtqueue operations | |
474 | * at the same time (except where noted). | |
475 | * | |
70670444 | 476 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
282edb36 RR |
477 | */ |
478 | int virtqueue_add_outbuf(struct virtqueue *vq, | |
eeebf9b1 | 479 | struct scatterlist *sg, unsigned int num, |
282edb36 RR |
480 | void *data, |
481 | gfp_t gfp) | |
482 | { | |
eeebf9b1 | 483 | return virtqueue_add(vq, &sg, num, 1, 0, data, gfp); |
282edb36 RR |
484 | } |
485 | EXPORT_SYMBOL_GPL(virtqueue_add_outbuf); | |
486 | ||
487 | /** | |
488 | * virtqueue_add_inbuf - expose input buffers to other end | |
489 | * @vq: the struct virtqueue we're talking about. | |
eeebf9b1 RR |
490 | * @sg: scatterlist (must be well-formed and terminated!) |
491 | * @num: the number of entries in @sg writable by other side | |
282edb36 RR |
492 | * @data: the token identifying the buffer. |
493 | * @gfp: how to do memory allocations (if necessary). | |
494 | * | |
495 | * Caller must ensure we don't call this with other virtqueue operations | |
496 | * at the same time (except where noted). | |
497 | * | |
70670444 | 498 | * Returns zero or a negative error (ie. ENOSPC, ENOMEM, EIO). |
282edb36 RR |
499 | */ |
500 | int virtqueue_add_inbuf(struct virtqueue *vq, | |
eeebf9b1 | 501 | struct scatterlist *sg, unsigned int num, |
282edb36 RR |
502 | void *data, |
503 | gfp_t gfp) | |
504 | { | |
eeebf9b1 | 505 | return virtqueue_add(vq, &sg, num, 0, 1, data, gfp); |
282edb36 RR |
506 | } |
507 | EXPORT_SYMBOL_GPL(virtqueue_add_inbuf); | |
508 | ||
5dfc1762 | 509 | /** |
41f0377f | 510 | * virtqueue_kick_prepare - first half of split virtqueue_kick call. |
5dfc1762 RR |
511 | * @vq: the struct virtqueue |
512 | * | |
41f0377f RR |
513 | * Instead of virtqueue_kick(), you can do: |
514 | * if (virtqueue_kick_prepare(vq)) | |
515 | * virtqueue_notify(vq); | |
5dfc1762 | 516 | * |
41f0377f RR |
517 | * This is sometimes useful because the virtqueue_kick_prepare() needs |
518 | * to be serialized, but the actual virtqueue_notify() call does not. | |
5dfc1762 | 519 | */ |
41f0377f | 520 | bool virtqueue_kick_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
521 | { |
522 | struct vring_virtqueue *vq = to_vvq(_vq); | |
a5c262c5 | 523 | u16 new, old; |
41f0377f RR |
524 | bool needs_kick; |
525 | ||
0a8a69dd | 526 | START_USE(vq); |
a72caae2 JW |
527 | /* We need to expose available array entries before checking avail |
528 | * event. */ | |
a9a0fef7 | 529 | virtio_mb(vq->weak_barriers); |
0a8a69dd | 530 | |
f277ec42 VS |
531 | old = vq->avail_idx_shadow - vq->num_added; |
532 | new = vq->avail_idx_shadow; | |
0a8a69dd RR |
533 | vq->num_added = 0; |
534 | ||
e93300b1 RR |
535 | #ifdef DEBUG |
536 | if (vq->last_add_time_valid) { | |
537 | WARN_ON(ktime_to_ms(ktime_sub(ktime_get(), | |
538 | vq->last_add_time)) > 100); | |
539 | } | |
540 | vq->last_add_time_valid = false; | |
541 | #endif | |
542 | ||
41f0377f | 543 | if (vq->event) { |
00e6f3d9 | 544 | needs_kick = vring_need_event(virtio16_to_cpu(_vq->vdev, vring_avail_event(&vq->vring)), |
41f0377f RR |
545 | new, old); |
546 | } else { | |
00e6f3d9 | 547 | needs_kick = !(vq->vring.used->flags & cpu_to_virtio16(_vq->vdev, VRING_USED_F_NO_NOTIFY)); |
41f0377f | 548 | } |
0a8a69dd | 549 | END_USE(vq); |
41f0377f RR |
550 | return needs_kick; |
551 | } | |
552 | EXPORT_SYMBOL_GPL(virtqueue_kick_prepare); | |
553 | ||
554 | /** | |
555 | * virtqueue_notify - second half of split virtqueue_kick call. | |
556 | * @vq: the struct virtqueue | |
557 | * | |
558 | * This does not need to be serialized. | |
5b1bf7cb HG |
559 | * |
560 | * Returns false if host notify failed or queue is broken, otherwise true. | |
41f0377f | 561 | */ |
5b1bf7cb | 562 | bool virtqueue_notify(struct virtqueue *_vq) |
41f0377f RR |
563 | { |
564 | struct vring_virtqueue *vq = to_vvq(_vq); | |
565 | ||
5b1bf7cb HG |
566 | if (unlikely(vq->broken)) |
567 | return false; | |
568 | ||
41f0377f | 569 | /* Prod other side to tell it about changes. */ |
2342d6a6 | 570 | if (!vq->notify(_vq)) { |
5b1bf7cb HG |
571 | vq->broken = true; |
572 | return false; | |
573 | } | |
574 | return true; | |
41f0377f RR |
575 | } |
576 | EXPORT_SYMBOL_GPL(virtqueue_notify); | |
577 | ||
578 | /** | |
579 | * virtqueue_kick - update after add_buf | |
580 | * @vq: the struct virtqueue | |
581 | * | |
b3087e48 | 582 | * After one or more virtqueue_add_* calls, invoke this to kick |
41f0377f RR |
583 | * the other side. |
584 | * | |
585 | * Caller must ensure we don't call this with other virtqueue | |
586 | * operations at the same time (except where noted). | |
5b1bf7cb HG |
587 | * |
588 | * Returns false if kick failed, otherwise true. | |
41f0377f | 589 | */ |
5b1bf7cb | 590 | bool virtqueue_kick(struct virtqueue *vq) |
41f0377f RR |
591 | { |
592 | if (virtqueue_kick_prepare(vq)) | |
5b1bf7cb HG |
593 | return virtqueue_notify(vq); |
594 | return true; | |
0a8a69dd | 595 | } |
7c5e9ed0 | 596 | EXPORT_SYMBOL_GPL(virtqueue_kick); |
0a8a69dd RR |
597 | |
598 | static void detach_buf(struct vring_virtqueue *vq, unsigned int head) | |
599 | { | |
780bc790 AL |
600 | unsigned int i, j; |
601 | u16 nextflag = cpu_to_virtio16(vq->vq.vdev, VRING_DESC_F_NEXT); | |
0a8a69dd RR |
602 | |
603 | /* Clear data ptr. */ | |
780bc790 | 604 | vq->desc_state[head].data = NULL; |
0a8a69dd | 605 | |
780bc790 | 606 | /* Put back on free list: unmap first-level descriptors and find end */ |
0a8a69dd | 607 | i = head; |
9fa29b9d | 608 | |
780bc790 AL |
609 | while (vq->vring.desc[i].flags & nextflag) { |
610 | vring_unmap_one(vq, &vq->vring.desc[i]); | |
00e6f3d9 | 611 | i = virtio16_to_cpu(vq->vq.vdev, vq->vring.desc[i].next); |
06ca287d | 612 | vq->vq.num_free++; |
0a8a69dd RR |
613 | } |
614 | ||
780bc790 | 615 | vring_unmap_one(vq, &vq->vring.desc[i]); |
00e6f3d9 | 616 | vq->vring.desc[i].next = cpu_to_virtio16(vq->vq.vdev, vq->free_head); |
0a8a69dd | 617 | vq->free_head = head; |
780bc790 | 618 | |
0a8a69dd | 619 | /* Plus final descriptor */ |
06ca287d | 620 | vq->vq.num_free++; |
780bc790 AL |
621 | |
622 | /* Free the indirect table, if any, now that it's unmapped. */ | |
623 | if (vq->desc_state[head].indir_desc) { | |
624 | struct vring_desc *indir_desc = vq->desc_state[head].indir_desc; | |
625 | u32 len = virtio32_to_cpu(vq->vq.vdev, vq->vring.desc[head].len); | |
626 | ||
627 | BUG_ON(!(vq->vring.desc[head].flags & | |
628 | cpu_to_virtio16(vq->vq.vdev, VRING_DESC_F_INDIRECT))); | |
629 | BUG_ON(len == 0 || len % sizeof(struct vring_desc)); | |
630 | ||
631 | for (j = 0; j < len / sizeof(struct vring_desc); j++) | |
632 | vring_unmap_one(vq, &indir_desc[j]); | |
633 | ||
634 | kfree(vq->desc_state[head].indir_desc); | |
635 | vq->desc_state[head].indir_desc = NULL; | |
636 | } | |
0a8a69dd RR |
637 | } |
638 | ||
0a8a69dd RR |
639 | static inline bool more_used(const struct vring_virtqueue *vq) |
640 | { | |
00e6f3d9 | 641 | return vq->last_used_idx != virtio16_to_cpu(vq->vq.vdev, vq->vring.used->idx); |
0a8a69dd RR |
642 | } |
643 | ||
5dfc1762 RR |
644 | /** |
645 | * virtqueue_get_buf - get the next used buffer | |
646 | * @vq: the struct virtqueue we're talking about. | |
647 | * @len: the length written into the buffer | |
648 | * | |
649 | * If the driver wrote data into the buffer, @len will be set to the | |
650 | * amount written. This means you don't need to clear the buffer | |
651 | * beforehand to ensure there's no data leakage in the case of short | |
652 | * writes. | |
653 | * | |
654 | * Caller must ensure we don't call this with other virtqueue | |
655 | * operations at the same time (except where noted). | |
656 | * | |
657 | * Returns NULL if there are no used buffers, or the "data" token | |
b3087e48 | 658 | * handed to virtqueue_add_*(). |
5dfc1762 | 659 | */ |
7c5e9ed0 | 660 | void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len) |
0a8a69dd RR |
661 | { |
662 | struct vring_virtqueue *vq = to_vvq(_vq); | |
663 | void *ret; | |
664 | unsigned int i; | |
3b720b8c | 665 | u16 last_used; |
0a8a69dd RR |
666 | |
667 | START_USE(vq); | |
668 | ||
5ef82752 RR |
669 | if (unlikely(vq->broken)) { |
670 | END_USE(vq); | |
671 | return NULL; | |
672 | } | |
673 | ||
0a8a69dd RR |
674 | if (!more_used(vq)) { |
675 | pr_debug("No more buffers in queue\n"); | |
676 | END_USE(vq); | |
677 | return NULL; | |
678 | } | |
679 | ||
2d61ba95 | 680 | /* Only get used array entries after they have been exposed by host. */ |
a9a0fef7 | 681 | virtio_rmb(vq->weak_barriers); |
2d61ba95 | 682 | |
3b720b8c | 683 | last_used = (vq->last_used_idx & (vq->vring.num - 1)); |
00e6f3d9 MT |
684 | i = virtio32_to_cpu(_vq->vdev, vq->vring.used->ring[last_used].id); |
685 | *len = virtio32_to_cpu(_vq->vdev, vq->vring.used->ring[last_used].len); | |
0a8a69dd RR |
686 | |
687 | if (unlikely(i >= vq->vring.num)) { | |
688 | BAD_RING(vq, "id %u out of range\n", i); | |
689 | return NULL; | |
690 | } | |
780bc790 | 691 | if (unlikely(!vq->desc_state[i].data)) { |
0a8a69dd RR |
692 | BAD_RING(vq, "id %u is not a head!\n", i); |
693 | return NULL; | |
694 | } | |
695 | ||
696 | /* detach_buf clears data, so grab it now. */ | |
780bc790 | 697 | ret = vq->desc_state[i].data; |
0a8a69dd RR |
698 | detach_buf(vq, i); |
699 | vq->last_used_idx++; | |
a5c262c5 MT |
700 | /* If we expect an interrupt for the next entry, tell host |
701 | * by writing event index and flush out the write before | |
702 | * the read in the next get_buf call. */ | |
788e5b3a MT |
703 | if (!(vq->avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT)) |
704 | virtio_store_mb(vq->weak_barriers, | |
705 | &vring_used_event(&vq->vring), | |
706 | cpu_to_virtio16(_vq->vdev, vq->last_used_idx)); | |
a5c262c5 | 707 | |
e93300b1 RR |
708 | #ifdef DEBUG |
709 | vq->last_add_time_valid = false; | |
710 | #endif | |
711 | ||
0a8a69dd RR |
712 | END_USE(vq); |
713 | return ret; | |
714 | } | |
7c5e9ed0 | 715 | EXPORT_SYMBOL_GPL(virtqueue_get_buf); |
0a8a69dd | 716 | |
5dfc1762 RR |
717 | /** |
718 | * virtqueue_disable_cb - disable callbacks | |
719 | * @vq: the struct virtqueue we're talking about. | |
720 | * | |
721 | * Note that this is not necessarily synchronous, hence unreliable and only | |
722 | * useful as an optimization. | |
723 | * | |
724 | * Unlike other operations, this need not be serialized. | |
725 | */ | |
7c5e9ed0 | 726 | void virtqueue_disable_cb(struct virtqueue *_vq) |
18445c4d RR |
727 | { |
728 | struct vring_virtqueue *vq = to_vvq(_vq); | |
729 | ||
f277ec42 VS |
730 | if (!(vq->avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT)) { |
731 | vq->avail_flags_shadow |= VRING_AVAIL_F_NO_INTERRUPT; | |
732 | vq->vring.avail->flags = cpu_to_virtio16(_vq->vdev, vq->avail_flags_shadow); | |
733 | } | |
734 | ||
18445c4d | 735 | } |
7c5e9ed0 | 736 | EXPORT_SYMBOL_GPL(virtqueue_disable_cb); |
18445c4d | 737 | |
5dfc1762 | 738 | /** |
cc229884 | 739 | * virtqueue_enable_cb_prepare - restart callbacks after disable_cb |
5dfc1762 RR |
740 | * @vq: the struct virtqueue we're talking about. |
741 | * | |
cc229884 MT |
742 | * This re-enables callbacks; it returns current queue state |
743 | * in an opaque unsigned value. This value should be later tested by | |
744 | * virtqueue_poll, to detect a possible race between the driver checking for | |
745 | * more work, and enabling callbacks. | |
5dfc1762 RR |
746 | * |
747 | * Caller must ensure we don't call this with other virtqueue | |
748 | * operations at the same time (except where noted). | |
749 | */ | |
cc229884 | 750 | unsigned virtqueue_enable_cb_prepare(struct virtqueue *_vq) |
0a8a69dd RR |
751 | { |
752 | struct vring_virtqueue *vq = to_vvq(_vq); | |
cc229884 | 753 | u16 last_used_idx; |
0a8a69dd RR |
754 | |
755 | START_USE(vq); | |
0a8a69dd RR |
756 | |
757 | /* We optimistically turn back on interrupts, then check if there was | |
758 | * more to do. */ | |
a5c262c5 MT |
759 | /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to |
760 | * either clear the flags bit or point the event index at the next | |
761 | * entry. Always do both to keep code simple. */ | |
f277ec42 VS |
762 | if (vq->avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT) { |
763 | vq->avail_flags_shadow &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
764 | vq->vring.avail->flags = cpu_to_virtio16(_vq->vdev, vq->avail_flags_shadow); | |
765 | } | |
00e6f3d9 | 766 | vring_used_event(&vq->vring) = cpu_to_virtio16(_vq->vdev, last_used_idx = vq->last_used_idx); |
cc229884 MT |
767 | END_USE(vq); |
768 | return last_used_idx; | |
769 | } | |
770 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_prepare); | |
771 | ||
772 | /** | |
773 | * virtqueue_poll - query pending used buffers | |
774 | * @vq: the struct virtqueue we're talking about. | |
775 | * @last_used_idx: virtqueue state (from call to virtqueue_enable_cb_prepare). | |
776 | * | |
777 | * Returns "true" if there are pending used buffers in the queue. | |
778 | * | |
779 | * This does not need to be serialized. | |
780 | */ | |
781 | bool virtqueue_poll(struct virtqueue *_vq, unsigned last_used_idx) | |
782 | { | |
783 | struct vring_virtqueue *vq = to_vvq(_vq); | |
784 | ||
a9a0fef7 | 785 | virtio_mb(vq->weak_barriers); |
00e6f3d9 | 786 | return (u16)last_used_idx != virtio16_to_cpu(_vq->vdev, vq->vring.used->idx); |
cc229884 MT |
787 | } |
788 | EXPORT_SYMBOL_GPL(virtqueue_poll); | |
0a8a69dd | 789 | |
cc229884 MT |
790 | /** |
791 | * virtqueue_enable_cb - restart callbacks after disable_cb. | |
792 | * @vq: the struct virtqueue we're talking about. | |
793 | * | |
794 | * This re-enables callbacks; it returns "false" if there are pending | |
795 | * buffers in the queue, to detect a possible race between the driver | |
796 | * checking for more work, and enabling callbacks. | |
797 | * | |
798 | * Caller must ensure we don't call this with other virtqueue | |
799 | * operations at the same time (except where noted). | |
800 | */ | |
801 | bool virtqueue_enable_cb(struct virtqueue *_vq) | |
802 | { | |
803 | unsigned last_used_idx = virtqueue_enable_cb_prepare(_vq); | |
804 | return !virtqueue_poll(_vq, last_used_idx); | |
0a8a69dd | 805 | } |
7c5e9ed0 | 806 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb); |
0a8a69dd | 807 | |
5dfc1762 RR |
808 | /** |
809 | * virtqueue_enable_cb_delayed - restart callbacks after disable_cb. | |
810 | * @vq: the struct virtqueue we're talking about. | |
811 | * | |
812 | * This re-enables callbacks but hints to the other side to delay | |
813 | * interrupts until most of the available buffers have been processed; | |
814 | * it returns "false" if there are many pending buffers in the queue, | |
815 | * to detect a possible race between the driver checking for more work, | |
816 | * and enabling callbacks. | |
817 | * | |
818 | * Caller must ensure we don't call this with other virtqueue | |
819 | * operations at the same time (except where noted). | |
820 | */ | |
7ab358c2 MT |
821 | bool virtqueue_enable_cb_delayed(struct virtqueue *_vq) |
822 | { | |
823 | struct vring_virtqueue *vq = to_vvq(_vq); | |
824 | u16 bufs; | |
825 | ||
826 | START_USE(vq); | |
827 | ||
828 | /* We optimistically turn back on interrupts, then check if there was | |
829 | * more to do. */ | |
830 | /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to | |
831 | * either clear the flags bit or point the event index at the next | |
832 | * entry. Always do both to keep code simple. */ | |
f277ec42 VS |
833 | if (vq->avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT) { |
834 | vq->avail_flags_shadow &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
835 | vq->vring.avail->flags = cpu_to_virtio16(_vq->vdev, vq->avail_flags_shadow); | |
836 | } | |
7ab358c2 | 837 | /* TODO: tune this threshold */ |
f277ec42 | 838 | bufs = (u16)(vq->avail_idx_shadow - vq->last_used_idx) * 3 / 4; |
788e5b3a MT |
839 | |
840 | virtio_store_mb(vq->weak_barriers, | |
841 | &vring_used_event(&vq->vring), | |
842 | cpu_to_virtio16(_vq->vdev, vq->last_used_idx + bufs)); | |
843 | ||
00e6f3d9 | 844 | if (unlikely((u16)(virtio16_to_cpu(_vq->vdev, vq->vring.used->idx) - vq->last_used_idx) > bufs)) { |
7ab358c2 MT |
845 | END_USE(vq); |
846 | return false; | |
847 | } | |
848 | ||
849 | END_USE(vq); | |
850 | return true; | |
851 | } | |
852 | EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed); | |
853 | ||
5dfc1762 RR |
854 | /** |
855 | * virtqueue_detach_unused_buf - detach first unused buffer | |
856 | * @vq: the struct virtqueue we're talking about. | |
857 | * | |
b3087e48 | 858 | * Returns NULL or the "data" token handed to virtqueue_add_*(). |
5dfc1762 RR |
859 | * This is not valid on an active queue; it is useful only for device |
860 | * shutdown. | |
861 | */ | |
7c5e9ed0 | 862 | void *virtqueue_detach_unused_buf(struct virtqueue *_vq) |
c021eac4 SM |
863 | { |
864 | struct vring_virtqueue *vq = to_vvq(_vq); | |
865 | unsigned int i; | |
866 | void *buf; | |
867 | ||
868 | START_USE(vq); | |
869 | ||
870 | for (i = 0; i < vq->vring.num; i++) { | |
780bc790 | 871 | if (!vq->desc_state[i].data) |
c021eac4 SM |
872 | continue; |
873 | /* detach_buf clears data, so grab it now. */ | |
780bc790 | 874 | buf = vq->desc_state[i].data; |
c021eac4 | 875 | detach_buf(vq, i); |
f277ec42 VS |
876 | vq->avail_idx_shadow--; |
877 | vq->vring.avail->idx = cpu_to_virtio16(_vq->vdev, vq->avail_idx_shadow); | |
c021eac4 SM |
878 | END_USE(vq); |
879 | return buf; | |
880 | } | |
881 | /* That should have freed everything. */ | |
06ca287d | 882 | BUG_ON(vq->vq.num_free != vq->vring.num); |
c021eac4 SM |
883 | |
884 | END_USE(vq); | |
885 | return NULL; | |
886 | } | |
7c5e9ed0 | 887 | EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf); |
c021eac4 | 888 | |
0a8a69dd RR |
889 | irqreturn_t vring_interrupt(int irq, void *_vq) |
890 | { | |
891 | struct vring_virtqueue *vq = to_vvq(_vq); | |
892 | ||
893 | if (!more_used(vq)) { | |
894 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
895 | return IRQ_NONE; | |
896 | } | |
897 | ||
898 | if (unlikely(vq->broken)) | |
899 | return IRQ_HANDLED; | |
900 | ||
901 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); | |
18445c4d RR |
902 | if (vq->vq.callback) |
903 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
904 | |
905 | return IRQ_HANDLED; | |
906 | } | |
c6fd4701 | 907 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd | 908 | |
2a2d1382 AL |
909 | struct virtqueue *__vring_new_virtqueue(unsigned int index, |
910 | struct vring vring, | |
911 | struct virtio_device *vdev, | |
912 | bool weak_barriers, | |
913 | bool (*notify)(struct virtqueue *), | |
914 | void (*callback)(struct virtqueue *), | |
915 | const char *name) | |
0a8a69dd | 916 | { |
0a8a69dd | 917 | unsigned int i; |
2a2d1382 | 918 | struct vring_virtqueue *vq; |
0a8a69dd | 919 | |
2a2d1382 | 920 | vq = kmalloc(sizeof(*vq) + vring.num * sizeof(struct vring_desc_state), |
780bc790 | 921 | GFP_KERNEL); |
0a8a69dd RR |
922 | if (!vq) |
923 | return NULL; | |
924 | ||
2a2d1382 | 925 | vq->vring = vring; |
0a8a69dd RR |
926 | vq->vq.callback = callback; |
927 | vq->vq.vdev = vdev; | |
9499f5e7 | 928 | vq->vq.name = name; |
2a2d1382 | 929 | vq->vq.num_free = vring.num; |
06ca287d | 930 | vq->vq.index = index; |
2a2d1382 AL |
931 | vq->we_own_ring = false; |
932 | vq->queue_dma_addr = 0; | |
933 | vq->queue_size_in_bytes = 0; | |
0a8a69dd | 934 | vq->notify = notify; |
7b21e34f | 935 | vq->weak_barriers = weak_barriers; |
0a8a69dd RR |
936 | vq->broken = false; |
937 | vq->last_used_idx = 0; | |
f277ec42 VS |
938 | vq->avail_flags_shadow = 0; |
939 | vq->avail_idx_shadow = 0; | |
0a8a69dd | 940 | vq->num_added = 0; |
9499f5e7 | 941 | list_add_tail(&vq->vq.list, &vdev->vqs); |
0a8a69dd RR |
942 | #ifdef DEBUG |
943 | vq->in_use = false; | |
e93300b1 | 944 | vq->last_add_time_valid = false; |
0a8a69dd RR |
945 | #endif |
946 | ||
9fa29b9d | 947 | vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC); |
a5c262c5 | 948 | vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); |
9fa29b9d | 949 | |
0a8a69dd | 950 | /* No callback? Tell other side not to bother us. */ |
f277ec42 VS |
951 | if (!callback) { |
952 | vq->avail_flags_shadow |= VRING_AVAIL_F_NO_INTERRUPT; | |
953 | vq->vring.avail->flags = cpu_to_virtio16(vdev, vq->avail_flags_shadow); | |
954 | } | |
0a8a69dd RR |
955 | |
956 | /* Put everything in free lists. */ | |
0a8a69dd | 957 | vq->free_head = 0; |
2a2d1382 | 958 | for (i = 0; i < vring.num-1; i++) |
00e6f3d9 | 959 | vq->vring.desc[i].next = cpu_to_virtio16(vdev, i + 1); |
2a2d1382 | 960 | memset(vq->desc_state, 0, vring.num * sizeof(struct vring_desc_state)); |
0a8a69dd RR |
961 | |
962 | return &vq->vq; | |
963 | } | |
2a2d1382 AL |
964 | EXPORT_SYMBOL_GPL(__vring_new_virtqueue); |
965 | ||
966 | static void *vring_alloc_queue(struct virtio_device *vdev, size_t size, | |
967 | dma_addr_t *dma_handle, gfp_t flag) | |
968 | { | |
969 | if (vring_use_dma_api(vdev)) { | |
970 | return dma_alloc_coherent(vdev->dev.parent, size, | |
971 | dma_handle, flag); | |
972 | } else { | |
973 | void *queue = alloc_pages_exact(PAGE_ALIGN(size), flag); | |
974 | if (queue) { | |
975 | phys_addr_t phys_addr = virt_to_phys(queue); | |
976 | *dma_handle = (dma_addr_t)phys_addr; | |
977 | ||
978 | /* | |
979 | * Sanity check: make sure we dind't truncate | |
980 | * the address. The only arches I can find that | |
981 | * have 64-bit phys_addr_t but 32-bit dma_addr_t | |
982 | * are certain non-highmem MIPS and x86 | |
983 | * configurations, but these configurations | |
984 | * should never allocate physical pages above 32 | |
985 | * bits, so this is fine. Just in case, throw a | |
986 | * warning and abort if we end up with an | |
987 | * unrepresentable address. | |
988 | */ | |
989 | if (WARN_ON_ONCE(*dma_handle != phys_addr)) { | |
990 | free_pages_exact(queue, PAGE_ALIGN(size)); | |
991 | return NULL; | |
992 | } | |
993 | } | |
994 | return queue; | |
995 | } | |
996 | } | |
997 | ||
998 | static void vring_free_queue(struct virtio_device *vdev, size_t size, | |
999 | void *queue, dma_addr_t dma_handle) | |
1000 | { | |
1001 | if (vring_use_dma_api(vdev)) { | |
1002 | dma_free_coherent(vdev->dev.parent, size, queue, dma_handle); | |
1003 | } else { | |
1004 | free_pages_exact(queue, PAGE_ALIGN(size)); | |
1005 | } | |
1006 | } | |
1007 | ||
1008 | struct virtqueue *vring_create_virtqueue( | |
1009 | unsigned int index, | |
1010 | unsigned int num, | |
1011 | unsigned int vring_align, | |
1012 | struct virtio_device *vdev, | |
1013 | bool weak_barriers, | |
1014 | bool may_reduce_num, | |
1015 | bool (*notify)(struct virtqueue *), | |
1016 | void (*callback)(struct virtqueue *), | |
1017 | const char *name) | |
1018 | { | |
1019 | struct virtqueue *vq; | |
e00f7bd2 | 1020 | void *queue = NULL; |
2a2d1382 AL |
1021 | dma_addr_t dma_addr; |
1022 | size_t queue_size_in_bytes; | |
1023 | struct vring vring; | |
1024 | ||
1025 | /* We assume num is a power of 2. */ | |
1026 | if (num & (num - 1)) { | |
1027 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
1028 | return NULL; | |
1029 | } | |
1030 | ||
1031 | /* TODO: allocate each queue chunk individually */ | |
1032 | for (; num && vring_size(num, vring_align) > PAGE_SIZE; num /= 2) { | |
1033 | queue = vring_alloc_queue(vdev, vring_size(num, vring_align), | |
1034 | &dma_addr, | |
1035 | GFP_KERNEL|__GFP_NOWARN|__GFP_ZERO); | |
1036 | if (queue) | |
1037 | break; | |
1038 | } | |
1039 | ||
1040 | if (!num) | |
1041 | return NULL; | |
1042 | ||
1043 | if (!queue) { | |
1044 | /* Try to get a single page. You are my only hope! */ | |
1045 | queue = vring_alloc_queue(vdev, vring_size(num, vring_align), | |
1046 | &dma_addr, GFP_KERNEL|__GFP_ZERO); | |
1047 | } | |
1048 | if (!queue) | |
1049 | return NULL; | |
1050 | ||
1051 | queue_size_in_bytes = vring_size(num, vring_align); | |
1052 | vring_init(&vring, num, queue, vring_align); | |
1053 | ||
1054 | vq = __vring_new_virtqueue(index, vring, vdev, weak_barriers, | |
1055 | notify, callback, name); | |
1056 | if (!vq) { | |
1057 | vring_free_queue(vdev, queue_size_in_bytes, queue, | |
1058 | dma_addr); | |
1059 | return NULL; | |
1060 | } | |
1061 | ||
1062 | to_vvq(vq)->queue_dma_addr = dma_addr; | |
1063 | to_vvq(vq)->queue_size_in_bytes = queue_size_in_bytes; | |
1064 | to_vvq(vq)->we_own_ring = true; | |
1065 | ||
1066 | return vq; | |
1067 | } | |
1068 | EXPORT_SYMBOL_GPL(vring_create_virtqueue); | |
1069 | ||
1070 | struct virtqueue *vring_new_virtqueue(unsigned int index, | |
1071 | unsigned int num, | |
1072 | unsigned int vring_align, | |
1073 | struct virtio_device *vdev, | |
1074 | bool weak_barriers, | |
1075 | void *pages, | |
1076 | bool (*notify)(struct virtqueue *vq), | |
1077 | void (*callback)(struct virtqueue *vq), | |
1078 | const char *name) | |
1079 | { | |
1080 | struct vring vring; | |
1081 | vring_init(&vring, num, pages, vring_align); | |
1082 | return __vring_new_virtqueue(index, vring, vdev, weak_barriers, | |
1083 | notify, callback, name); | |
1084 | } | |
c6fd4701 | 1085 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd | 1086 | |
2a2d1382 | 1087 | void vring_del_virtqueue(struct virtqueue *_vq) |
0a8a69dd | 1088 | { |
2a2d1382 AL |
1089 | struct vring_virtqueue *vq = to_vvq(_vq); |
1090 | ||
1091 | if (vq->we_own_ring) { | |
1092 | vring_free_queue(vq->vq.vdev, vq->queue_size_in_bytes, | |
1093 | vq->vring.desc, vq->queue_dma_addr); | |
1094 | } | |
1095 | list_del(&_vq->list); | |
1096 | kfree(vq); | |
0a8a69dd | 1097 | } |
c6fd4701 | 1098 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 1099 | |
e34f8725 RR |
1100 | /* Manipulates transport-specific feature bits. */ |
1101 | void vring_transport_features(struct virtio_device *vdev) | |
1102 | { | |
1103 | unsigned int i; | |
1104 | ||
1105 | for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { | |
1106 | switch (i) { | |
9fa29b9d MM |
1107 | case VIRTIO_RING_F_INDIRECT_DESC: |
1108 | break; | |
a5c262c5 MT |
1109 | case VIRTIO_RING_F_EVENT_IDX: |
1110 | break; | |
747ae34a MT |
1111 | case VIRTIO_F_VERSION_1: |
1112 | break; | |
1a937693 MT |
1113 | case VIRTIO_F_IOMMU_PLATFORM: |
1114 | break; | |
e34f8725 RR |
1115 | default: |
1116 | /* We don't understand this bit. */ | |
e16e12be | 1117 | __virtio_clear_bit(vdev, i); |
e34f8725 RR |
1118 | } |
1119 | } | |
1120 | } | |
1121 | EXPORT_SYMBOL_GPL(vring_transport_features); | |
1122 | ||
5dfc1762 RR |
1123 | /** |
1124 | * virtqueue_get_vring_size - return the size of the virtqueue's vring | |
1125 | * @vq: the struct virtqueue containing the vring of interest. | |
1126 | * | |
1127 | * Returns the size of the vring. This is mainly used for boasting to | |
1128 | * userspace. Unlike other operations, this need not be serialized. | |
1129 | */ | |
8f9f4668 RJ |
1130 | unsigned int virtqueue_get_vring_size(struct virtqueue *_vq) |
1131 | { | |
1132 | ||
1133 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1134 | ||
1135 | return vq->vring.num; | |
1136 | } | |
1137 | EXPORT_SYMBOL_GPL(virtqueue_get_vring_size); | |
1138 | ||
b3b32c94 HG |
1139 | bool virtqueue_is_broken(struct virtqueue *_vq) |
1140 | { | |
1141 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1142 | ||
1143 | return vq->broken; | |
1144 | } | |
1145 | EXPORT_SYMBOL_GPL(virtqueue_is_broken); | |
1146 | ||
e2dcdfe9 RR |
1147 | /* |
1148 | * This should prevent the device from being used, allowing drivers to | |
1149 | * recover. You may need to grab appropriate locks to flush. | |
1150 | */ | |
1151 | void virtio_break_device(struct virtio_device *dev) | |
1152 | { | |
1153 | struct virtqueue *_vq; | |
1154 | ||
1155 | list_for_each_entry(_vq, &dev->vqs, list) { | |
1156 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1157 | vq->broken = true; | |
1158 | } | |
1159 | } | |
1160 | EXPORT_SYMBOL_GPL(virtio_break_device); | |
1161 | ||
2a2d1382 | 1162 | dma_addr_t virtqueue_get_desc_addr(struct virtqueue *_vq) |
89062652 CH |
1163 | { |
1164 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1165 | ||
2a2d1382 AL |
1166 | BUG_ON(!vq->we_own_ring); |
1167 | ||
1168 | return vq->queue_dma_addr; | |
89062652 | 1169 | } |
2a2d1382 | 1170 | EXPORT_SYMBOL_GPL(virtqueue_get_desc_addr); |
89062652 | 1171 | |
2a2d1382 | 1172 | dma_addr_t virtqueue_get_avail_addr(struct virtqueue *_vq) |
89062652 CH |
1173 | { |
1174 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1175 | ||
2a2d1382 AL |
1176 | BUG_ON(!vq->we_own_ring); |
1177 | ||
1178 | return vq->queue_dma_addr + | |
1179 | ((char *)vq->vring.avail - (char *)vq->vring.desc); | |
1180 | } | |
1181 | EXPORT_SYMBOL_GPL(virtqueue_get_avail_addr); | |
1182 | ||
1183 | dma_addr_t virtqueue_get_used_addr(struct virtqueue *_vq) | |
1184 | { | |
1185 | struct vring_virtqueue *vq = to_vvq(_vq); | |
1186 | ||
1187 | BUG_ON(!vq->we_own_ring); | |
1188 | ||
1189 | return vq->queue_dma_addr + | |
1190 | ((char *)vq->vring.used - (char *)vq->vring.desc); | |
1191 | } | |
1192 | EXPORT_SYMBOL_GPL(virtqueue_get_used_addr); | |
1193 | ||
1194 | const struct vring *virtqueue_get_vring(struct virtqueue *vq) | |
1195 | { | |
1196 | return &to_vvq(vq)->vring; | |
89062652 | 1197 | } |
2a2d1382 | 1198 | EXPORT_SYMBOL_GPL(virtqueue_get_vring); |
89062652 | 1199 | |
c6fd4701 | 1200 | MODULE_LICENSE("GPL"); |