Commit | Line | Data |
---|---|---|
0a8a69dd RR |
1 | /* Virtio ring implementation. |
2 | * | |
3 | * Copyright 2007 Rusty Russell IBM Corporation | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
16 | * along with this program; if not, write to the Free Software | |
17 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
18 | */ | |
19 | #include <linux/virtio.h> | |
20 | #include <linux/virtio_ring.h> | |
e34f8725 | 21 | #include <linux/virtio_config.h> |
0a8a69dd RR |
22 | #include <linux/device.h> |
23 | ||
24 | #ifdef DEBUG | |
25 | /* For development, we want to crash whenever the ring is screwed. */ | |
3a35ce7d | 26 | #define BAD_RING(_vq, fmt...) \ |
c5f841f1 RR |
27 | do { dev_err(&(_vq)->vq.vdev->dev, fmt); BUG(); } while(0) |
28 | /* Caller is supposed to guarantee no reentry. */ | |
29 | #define START_USE(_vq) \ | |
30 | do { \ | |
31 | if ((_vq)->in_use) \ | |
32 | panic("in_use = %i\n", (_vq)->in_use); \ | |
33 | (_vq)->in_use = __LINE__; \ | |
34 | mb(); \ | |
35 | } while(0) | |
3a35ce7d RK |
36 | #define END_USE(_vq) \ |
37 | do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; mb(); } while(0) | |
0a8a69dd | 38 | #else |
3a35ce7d RK |
39 | #define BAD_RING(_vq, fmt...) \ |
40 | do { dev_err(&_vq->vq.vdev->dev, fmt); (_vq)->broken = true; } while(0) | |
0a8a69dd RR |
41 | #define START_USE(vq) |
42 | #define END_USE(vq) | |
43 | #endif | |
44 | ||
45 | struct vring_virtqueue | |
46 | { | |
47 | struct virtqueue vq; | |
48 | ||
49 | /* Actual memory layout for this queue */ | |
50 | struct vring vring; | |
51 | ||
52 | /* Other side has made a mess, don't try any more. */ | |
53 | bool broken; | |
54 | ||
55 | /* Number of free buffers */ | |
56 | unsigned int num_free; | |
57 | /* Head of free buffer list. */ | |
58 | unsigned int free_head; | |
59 | /* Number we've added since last sync. */ | |
60 | unsigned int num_added; | |
61 | ||
62 | /* Last used index we've seen. */ | |
1bc4953e | 63 | u16 last_used_idx; |
0a8a69dd RR |
64 | |
65 | /* How to notify other side. FIXME: commonalize hcalls! */ | |
66 | void (*notify)(struct virtqueue *vq); | |
67 | ||
68 | #ifdef DEBUG | |
69 | /* They're supposed to lock for us. */ | |
70 | unsigned int in_use; | |
71 | #endif | |
72 | ||
73 | /* Tokens for callbacks. */ | |
74 | void *data[]; | |
75 | }; | |
76 | ||
77 | #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq) | |
78 | ||
79 | static int vring_add_buf(struct virtqueue *_vq, | |
80 | struct scatterlist sg[], | |
81 | unsigned int out, | |
82 | unsigned int in, | |
83 | void *data) | |
84 | { | |
85 | struct vring_virtqueue *vq = to_vvq(_vq); | |
86 | unsigned int i, avail, head, uninitialized_var(prev); | |
87 | ||
88 | BUG_ON(data == NULL); | |
89 | BUG_ON(out + in > vq->vring.num); | |
90 | BUG_ON(out + in == 0); | |
91 | ||
92 | START_USE(vq); | |
93 | ||
94 | if (vq->num_free < out + in) { | |
95 | pr_debug("Can't add buf len %i - avail = %i\n", | |
96 | out + in, vq->num_free); | |
44653eae RR |
97 | /* FIXME: for historical reasons, we force a notify here if |
98 | * there are outgoing parts to the buffer. Presumably the | |
99 | * host should service the ring ASAP. */ | |
100 | if (out) | |
101 | vq->notify(&vq->vq); | |
0a8a69dd RR |
102 | END_USE(vq); |
103 | return -ENOSPC; | |
104 | } | |
105 | ||
106 | /* We're about to use some buffers from the free list. */ | |
107 | vq->num_free -= out + in; | |
108 | ||
109 | head = vq->free_head; | |
110 | for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) { | |
111 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT; | |
15f9c890 | 112 | vq->vring.desc[i].addr = sg_phys(sg); |
0a8a69dd RR |
113 | vq->vring.desc[i].len = sg->length; |
114 | prev = i; | |
115 | sg++; | |
116 | } | |
117 | for (; in; i = vq->vring.desc[i].next, in--) { | |
118 | vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; | |
15f9c890 | 119 | vq->vring.desc[i].addr = sg_phys(sg); |
0a8a69dd RR |
120 | vq->vring.desc[i].len = sg->length; |
121 | prev = i; | |
122 | sg++; | |
123 | } | |
124 | /* Last one doesn't continue. */ | |
125 | vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT; | |
126 | ||
127 | /* Update free pointer */ | |
128 | vq->free_head = i; | |
129 | ||
130 | /* Set token. */ | |
131 | vq->data[head] = data; | |
132 | ||
133 | /* Put entry in available array (but don't update avail->idx until they | |
134 | * do sync). FIXME: avoid modulus here? */ | |
135 | avail = (vq->vring.avail->idx + vq->num_added++) % vq->vring.num; | |
136 | vq->vring.avail->ring[avail] = head; | |
137 | ||
138 | pr_debug("Added buffer head %i to %p\n", head, vq); | |
139 | END_USE(vq); | |
140 | return 0; | |
141 | } | |
142 | ||
143 | static void vring_kick(struct virtqueue *_vq) | |
144 | { | |
145 | struct vring_virtqueue *vq = to_vvq(_vq); | |
146 | START_USE(vq); | |
147 | /* Descriptors and available array need to be set before we expose the | |
148 | * new available array entries. */ | |
149 | wmb(); | |
150 | ||
151 | vq->vring.avail->idx += vq->num_added; | |
152 | vq->num_added = 0; | |
153 | ||
154 | /* Need to update avail index before checking if we should notify */ | |
155 | mb(); | |
156 | ||
157 | if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY)) | |
158 | /* Prod other side to tell it about changes. */ | |
159 | vq->notify(&vq->vq); | |
160 | ||
161 | END_USE(vq); | |
162 | } | |
163 | ||
164 | static void detach_buf(struct vring_virtqueue *vq, unsigned int head) | |
165 | { | |
166 | unsigned int i; | |
167 | ||
168 | /* Clear data ptr. */ | |
169 | vq->data[head] = NULL; | |
170 | ||
171 | /* Put back on free list: find end */ | |
172 | i = head; | |
173 | while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) { | |
174 | i = vq->vring.desc[i].next; | |
175 | vq->num_free++; | |
176 | } | |
177 | ||
178 | vq->vring.desc[i].next = vq->free_head; | |
179 | vq->free_head = head; | |
180 | /* Plus final descriptor */ | |
181 | vq->num_free++; | |
182 | } | |
183 | ||
0a8a69dd RR |
184 | static inline bool more_used(const struct vring_virtqueue *vq) |
185 | { | |
186 | return vq->last_used_idx != vq->vring.used->idx; | |
187 | } | |
188 | ||
189 | static void *vring_get_buf(struct virtqueue *_vq, unsigned int *len) | |
190 | { | |
191 | struct vring_virtqueue *vq = to_vvq(_vq); | |
192 | void *ret; | |
193 | unsigned int i; | |
194 | ||
195 | START_USE(vq); | |
196 | ||
5ef82752 RR |
197 | if (unlikely(vq->broken)) { |
198 | END_USE(vq); | |
199 | return NULL; | |
200 | } | |
201 | ||
0a8a69dd RR |
202 | if (!more_used(vq)) { |
203 | pr_debug("No more buffers in queue\n"); | |
204 | END_USE(vq); | |
205 | return NULL; | |
206 | } | |
207 | ||
208 | i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id; | |
209 | *len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len; | |
210 | ||
211 | if (unlikely(i >= vq->vring.num)) { | |
212 | BAD_RING(vq, "id %u out of range\n", i); | |
213 | return NULL; | |
214 | } | |
215 | if (unlikely(!vq->data[i])) { | |
216 | BAD_RING(vq, "id %u is not a head!\n", i); | |
217 | return NULL; | |
218 | } | |
219 | ||
220 | /* detach_buf clears data, so grab it now. */ | |
221 | ret = vq->data[i]; | |
222 | detach_buf(vq, i); | |
223 | vq->last_used_idx++; | |
224 | END_USE(vq); | |
225 | return ret; | |
226 | } | |
227 | ||
18445c4d RR |
228 | static void vring_disable_cb(struct virtqueue *_vq) |
229 | { | |
230 | struct vring_virtqueue *vq = to_vvq(_vq); | |
231 | ||
18445c4d | 232 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; |
18445c4d RR |
233 | } |
234 | ||
235 | static bool vring_enable_cb(struct virtqueue *_vq) | |
0a8a69dd RR |
236 | { |
237 | struct vring_virtqueue *vq = to_vvq(_vq); | |
238 | ||
239 | START_USE(vq); | |
0a8a69dd RR |
240 | |
241 | /* We optimistically turn back on interrupts, then check if there was | |
242 | * more to do. */ | |
243 | vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; | |
244 | mb(); | |
245 | if (unlikely(more_used(vq))) { | |
0a8a69dd RR |
246 | END_USE(vq); |
247 | return false; | |
248 | } | |
249 | ||
250 | END_USE(vq); | |
251 | return true; | |
252 | } | |
253 | ||
254 | irqreturn_t vring_interrupt(int irq, void *_vq) | |
255 | { | |
256 | struct vring_virtqueue *vq = to_vvq(_vq); | |
257 | ||
258 | if (!more_used(vq)) { | |
259 | pr_debug("virtqueue interrupt with no work for %p\n", vq); | |
260 | return IRQ_NONE; | |
261 | } | |
262 | ||
263 | if (unlikely(vq->broken)) | |
264 | return IRQ_HANDLED; | |
265 | ||
266 | pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); | |
18445c4d RR |
267 | if (vq->vq.callback) |
268 | vq->vq.callback(&vq->vq); | |
0a8a69dd RR |
269 | |
270 | return IRQ_HANDLED; | |
271 | } | |
c6fd4701 | 272 | EXPORT_SYMBOL_GPL(vring_interrupt); |
0a8a69dd RR |
273 | |
274 | static struct virtqueue_ops vring_vq_ops = { | |
275 | .add_buf = vring_add_buf, | |
276 | .get_buf = vring_get_buf, | |
277 | .kick = vring_kick, | |
18445c4d RR |
278 | .disable_cb = vring_disable_cb, |
279 | .enable_cb = vring_enable_cb, | |
0a8a69dd RR |
280 | }; |
281 | ||
282 | struct virtqueue *vring_new_virtqueue(unsigned int num, | |
87c7d57c | 283 | unsigned int vring_align, |
0a8a69dd RR |
284 | struct virtio_device *vdev, |
285 | void *pages, | |
286 | void (*notify)(struct virtqueue *), | |
18445c4d | 287 | void (*callback)(struct virtqueue *)) |
0a8a69dd RR |
288 | { |
289 | struct vring_virtqueue *vq; | |
290 | unsigned int i; | |
291 | ||
42b36cc0 RR |
292 | /* We assume num is a power of 2. */ |
293 | if (num & (num - 1)) { | |
294 | dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); | |
295 | return NULL; | |
296 | } | |
297 | ||
0a8a69dd RR |
298 | vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL); |
299 | if (!vq) | |
300 | return NULL; | |
301 | ||
87c7d57c | 302 | vring_init(&vq->vring, num, pages, vring_align); |
0a8a69dd RR |
303 | vq->vq.callback = callback; |
304 | vq->vq.vdev = vdev; | |
305 | vq->vq.vq_ops = &vring_vq_ops; | |
306 | vq->notify = notify; | |
307 | vq->broken = false; | |
308 | vq->last_used_idx = 0; | |
309 | vq->num_added = 0; | |
310 | #ifdef DEBUG | |
311 | vq->in_use = false; | |
312 | #endif | |
313 | ||
314 | /* No callback? Tell other side not to bother us. */ | |
315 | if (!callback) | |
316 | vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; | |
317 | ||
318 | /* Put everything in free lists. */ | |
319 | vq->num_free = num; | |
320 | vq->free_head = 0; | |
321 | for (i = 0; i < num-1; i++) | |
322 | vq->vring.desc[i].next = i+1; | |
323 | ||
324 | return &vq->vq; | |
325 | } | |
c6fd4701 | 326 | EXPORT_SYMBOL_GPL(vring_new_virtqueue); |
0a8a69dd RR |
327 | |
328 | void vring_del_virtqueue(struct virtqueue *vq) | |
329 | { | |
330 | kfree(to_vvq(vq)); | |
331 | } | |
c6fd4701 | 332 | EXPORT_SYMBOL_GPL(vring_del_virtqueue); |
0a8a69dd | 333 | |
e34f8725 RR |
334 | /* Manipulates transport-specific feature bits. */ |
335 | void vring_transport_features(struct virtio_device *vdev) | |
336 | { | |
337 | unsigned int i; | |
338 | ||
339 | for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { | |
340 | switch (i) { | |
341 | default: | |
342 | /* We don't understand this bit. */ | |
343 | clear_bit(i, vdev->features); | |
344 | } | |
345 | } | |
346 | } | |
347 | EXPORT_SYMBOL_GPL(vring_transport_features); | |
348 | ||
c6fd4701 | 349 | MODULE_LICENSE("GPL"); |