virtio_ring.c 7.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319
  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>
  21. #include <linux/device.h>
  22. #ifdef DEBUG
  23. /* For development, we want to crash whenever the ring is screwed. */
  24. #define BAD_RING(vq, fmt...) \
  25. do { dev_err(&vq->vq.vdev->dev, fmt); BUG(); } while(0)
  26. #define START_USE(vq) \
  27. do { if ((vq)->in_use) panic("in_use = %i\n", (vq)->in_use); (vq)->in_use = __LINE__; mb(); } while(0)
  28. #define END_USE(vq) \
  29. do { BUG_ON(!(vq)->in_use); (vq)->in_use = 0; mb(); } while(0)
  30. #else
  31. #define BAD_RING(vq, fmt...) \
  32. do { dev_err(&vq->vq.vdev->dev, fmt); (vq)->broken = true; } while(0)
  33. #define START_USE(vq)
  34. #define END_USE(vq)
  35. #endif
  36. struct vring_virtqueue
  37. {
  38. struct virtqueue vq;
  39. /* Actual memory layout for this queue */
  40. struct vring vring;
  41. /* Other side has made a mess, don't try any more. */
  42. bool broken;
  43. /* Number of free buffers */
  44. unsigned int num_free;
  45. /* Head of free buffer list. */
  46. unsigned int free_head;
  47. /* Number we've added since last sync. */
  48. unsigned int num_added;
  49. /* Last used index we've seen. */
  50. u16 last_used_idx;
  51. /* How to notify other side. FIXME: commonalize hcalls! */
  52. void (*notify)(struct virtqueue *vq);
  53. #ifdef DEBUG
  54. /* They're supposed to lock for us. */
  55. unsigned int in_use;
  56. #endif
  57. /* Tokens for callbacks. */
  58. void *data[];
  59. };
  60. #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq)
  61. static int vring_add_buf(struct virtqueue *_vq,
  62. struct scatterlist sg[],
  63. unsigned int out,
  64. unsigned int in,
  65. void *data)
  66. {
  67. struct vring_virtqueue *vq = to_vvq(_vq);
  68. unsigned int i, avail, head, uninitialized_var(prev);
  69. BUG_ON(data == NULL);
  70. BUG_ON(out + in > vq->vring.num);
  71. BUG_ON(out + in == 0);
  72. START_USE(vq);
  73. if (vq->num_free < out + in) {
  74. pr_debug("Can't add buf len %i - avail = %i\n",
  75. out + in, vq->num_free);
  76. END_USE(vq);
  77. return -ENOSPC;
  78. }
  79. /* We're about to use some buffers from the free list. */
  80. vq->num_free -= out + in;
  81. head = vq->free_head;
  82. for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) {
  83. vq->vring.desc[i].flags = VRING_DESC_F_NEXT;
  84. vq->vring.desc[i].addr = (page_to_pfn(sg_page(sg))<<PAGE_SHIFT)
  85. + sg->offset;
  86. vq->vring.desc[i].len = sg->length;
  87. prev = i;
  88. sg++;
  89. }
  90. for (; in; i = vq->vring.desc[i].next, in--) {
  91. vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
  92. vq->vring.desc[i].addr = (page_to_pfn(sg_page(sg))<<PAGE_SHIFT)
  93. + sg->offset;
  94. vq->vring.desc[i].len = sg->length;
  95. prev = i;
  96. sg++;
  97. }
  98. /* Last one doesn't continue. */
  99. vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT;
  100. /* Update free pointer */
  101. vq->free_head = i;
  102. /* Set token. */
  103. vq->data[head] = data;
  104. /* Put entry in available array (but don't update avail->idx until they
  105. * do sync). FIXME: avoid modulus here? */
  106. avail = (vq->vring.avail->idx + vq->num_added++) % vq->vring.num;
  107. vq->vring.avail->ring[avail] = head;
  108. pr_debug("Added buffer head %i to %p\n", head, vq);
  109. END_USE(vq);
  110. return 0;
  111. }
  112. static void vring_kick(struct virtqueue *_vq)
  113. {
  114. struct vring_virtqueue *vq = to_vvq(_vq);
  115. START_USE(vq);
  116. /* Descriptors and available array need to be set before we expose the
  117. * new available array entries. */
  118. wmb();
  119. vq->vring.avail->idx += vq->num_added;
  120. vq->num_added = 0;
  121. /* Need to update avail index before checking if we should notify */
  122. mb();
  123. if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY))
  124. /* Prod other side to tell it about changes. */
  125. vq->notify(&vq->vq);
  126. END_USE(vq);
  127. }
  128. static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
  129. {
  130. unsigned int i;
  131. /* Clear data ptr. */
  132. vq->data[head] = NULL;
  133. /* Put back on free list: find end */
  134. i = head;
  135. while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
  136. i = vq->vring.desc[i].next;
  137. vq->num_free++;
  138. }
  139. vq->vring.desc[i].next = vq->free_head;
  140. vq->free_head = head;
  141. /* Plus final descriptor */
  142. vq->num_free++;
  143. }
  144. /* FIXME: We need to tell other side about removal, to synchronize. */
  145. static void vring_shutdown(struct virtqueue *_vq)
  146. {
  147. struct vring_virtqueue *vq = to_vvq(_vq);
  148. unsigned int i;
  149. for (i = 0; i < vq->vring.num; i++)
  150. detach_buf(vq, i);
  151. }
  152. static inline bool more_used(const struct vring_virtqueue *vq)
  153. {
  154. return vq->last_used_idx != vq->vring.used->idx;
  155. }
  156. static void *vring_get_buf(struct virtqueue *_vq, unsigned int *len)
  157. {
  158. struct vring_virtqueue *vq = to_vvq(_vq);
  159. void *ret;
  160. unsigned int i;
  161. START_USE(vq);
  162. if (!more_used(vq)) {
  163. pr_debug("No more buffers in queue\n");
  164. END_USE(vq);
  165. return NULL;
  166. }
  167. i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id;
  168. *len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len;
  169. if (unlikely(i >= vq->vring.num)) {
  170. BAD_RING(vq, "id %u out of range\n", i);
  171. return NULL;
  172. }
  173. if (unlikely(!vq->data[i])) {
  174. BAD_RING(vq, "id %u is not a head!\n", i);
  175. return NULL;
  176. }
  177. /* detach_buf clears data, so grab it now. */
  178. ret = vq->data[i];
  179. detach_buf(vq, i);
  180. vq->last_used_idx++;
  181. END_USE(vq);
  182. return ret;
  183. }
  184. static bool vring_restart(struct virtqueue *_vq)
  185. {
  186. struct vring_virtqueue *vq = to_vvq(_vq);
  187. START_USE(vq);
  188. BUG_ON(!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT));
  189. /* We optimistically turn back on interrupts, then check if there was
  190. * more to do. */
  191. vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
  192. mb();
  193. if (unlikely(more_used(vq))) {
  194. vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
  195. END_USE(vq);
  196. return false;
  197. }
  198. END_USE(vq);
  199. return true;
  200. }
  201. irqreturn_t vring_interrupt(int irq, void *_vq)
  202. {
  203. struct vring_virtqueue *vq = to_vvq(_vq);
  204. if (!more_used(vq)) {
  205. pr_debug("virtqueue interrupt with no work for %p\n", vq);
  206. return IRQ_NONE;
  207. }
  208. if (unlikely(vq->broken))
  209. return IRQ_HANDLED;
  210. pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
  211. if (vq->vq.callback && !vq->vq.callback(&vq->vq))
  212. vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
  213. return IRQ_HANDLED;
  214. }
  215. static struct virtqueue_ops vring_vq_ops = {
  216. .add_buf = vring_add_buf,
  217. .get_buf = vring_get_buf,
  218. .kick = vring_kick,
  219. .restart = vring_restart,
  220. .shutdown = vring_shutdown,
  221. };
  222. struct virtqueue *vring_new_virtqueue(unsigned int num,
  223. struct virtio_device *vdev,
  224. void *pages,
  225. void (*notify)(struct virtqueue *),
  226. bool (*callback)(struct virtqueue *))
  227. {
  228. struct vring_virtqueue *vq;
  229. unsigned int i;
  230. /* We assume num is a power of 2. */
  231. if (num & (num - 1)) {
  232. dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num);
  233. return NULL;
  234. }
  235. vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
  236. if (!vq)
  237. return NULL;
  238. vring_init(&vq->vring, num, pages, PAGE_SIZE);
  239. vq->vq.callback = callback;
  240. vq->vq.vdev = vdev;
  241. vq->vq.vq_ops = &vring_vq_ops;
  242. vq->notify = notify;
  243. vq->broken = false;
  244. vq->last_used_idx = 0;
  245. vq->num_added = 0;
  246. #ifdef DEBUG
  247. vq->in_use = false;
  248. #endif
  249. /* No callback? Tell other side not to bother us. */
  250. if (!callback)
  251. vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
  252. /* Put everything in free lists. */
  253. vq->num_free = num;
  254. vq->free_head = 0;
  255. for (i = 0; i < num-1; i++)
  256. vq->vring.desc[i].next = i+1;
  257. return &vq->vq;
  258. }
  259. void vring_del_virtqueue(struct virtqueue *vq)
  260. {
  261. kfree(to_vvq(vq));
  262. }