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/virtio_config.h> 22 #include <linux/device.h> 23 24 #ifdef DEBUG 25 /* For development, we want to crash whenever the ring is screwed. */ 26 #define BAD_RING(_vq, fmt, args...) \ 27 do { \ 28 dev_err(&(_vq)->vq.vdev->dev, \ 29 "%s:"fmt, (_vq)->vq.name, ##args); \ 30 BUG(); \ 31 } while (0) 32 /* Caller is supposed to guarantee no reentry. */ 33 #define START_USE(_vq) \ 34 do { \ 35 if ((_vq)->in_use) \ 36 panic("%s:in_use = %i\n", \ 37 (_vq)->vq.name, (_vq)->in_use); \ 38 (_vq)->in_use = __LINE__; \ 39 mb(); \ 40 } while (0) 41 #define END_USE(_vq) \ 42 do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; mb(); } while(0) 43 #else 44 #define BAD_RING(_vq, fmt, args...) \ 45 do { \ 46 dev_err(&_vq->vq.vdev->dev, \ 47 "%s:"fmt, (_vq)->vq.name, ##args); \ 48 (_vq)->broken = true; \ 49 } while (0) 50 #define START_USE(vq) 51 #define END_USE(vq) 52 #endif 53 54 struct vring_virtqueue 55 { 56 struct virtqueue vq; 57 58 /* Actual memory layout for this queue */ 59 struct vring vring; 60 61 /* Other side has made a mess, don't try any more. */ 62 bool broken; 63 64 /* Host supports indirect buffers */ 65 bool indirect; 66 67 /* Number of free buffers */ 68 unsigned int num_free; 69 /* Head of free buffer list. */ 70 unsigned int free_head; 71 /* Number we've added since last sync. */ 72 unsigned int num_added; 73 74 /* Last used index we've seen. */ 75 u16 last_used_idx; 76 77 /* How to notify other side. FIXME: commonalize hcalls! */ 78 void (*notify)(struct virtqueue *vq); 79 80 #ifdef DEBUG 81 /* They're supposed to lock for us. */ 82 unsigned int in_use; 83 #endif 84 85 /* Tokens for callbacks. */ 86 void *data[]; 87 }; 88 89 #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq) 90 91 /* Set up an indirect table of descriptors and add it to the queue. */ 92 static int vring_add_indirect(struct vring_virtqueue *vq, 93 struct scatterlist sg[], 94 unsigned int out, 95 unsigned int in) 96 { 97 struct vring_desc *desc; 98 unsigned head; 99 int i; 100 101 desc = kmalloc((out + in) * sizeof(struct vring_desc), GFP_ATOMIC); 102 if (!desc) 103 return vq->vring.num; 104 105 /* Transfer entries from the sg list into the indirect page */ 106 for (i = 0; i < out; i++) { 107 desc[i].flags = VRING_DESC_F_NEXT; 108 desc[i].addr = sg_phys(sg); 109 desc[i].len = sg->length; 110 desc[i].next = i+1; 111 sg++; 112 } 113 for (; i < (out + in); i++) { 114 desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; 115 desc[i].addr = sg_phys(sg); 116 desc[i].len = sg->length; 117 desc[i].next = i+1; 118 sg++; 119 } 120 121 /* Last one doesn't continue. */ 122 desc[i-1].flags &= ~VRING_DESC_F_NEXT; 123 desc[i-1].next = 0; 124 125 /* We're about to use a buffer */ 126 vq->num_free--; 127 128 /* Use a single buffer which doesn't continue */ 129 head = vq->free_head; 130 vq->vring.desc[head].flags = VRING_DESC_F_INDIRECT; 131 vq->vring.desc[head].addr = virt_to_phys(desc); 132 vq->vring.desc[head].len = i * sizeof(struct vring_desc); 133 134 /* Update free pointer */ 135 vq->free_head = vq->vring.desc[head].next; 136 137 return head; 138 } 139 140 static int vring_add_buf(struct virtqueue *_vq, 141 struct scatterlist sg[], 142 unsigned int out, 143 unsigned int in, 144 void *data) 145 { 146 struct vring_virtqueue *vq = to_vvq(_vq); 147 unsigned int i, avail, head, uninitialized_var(prev); 148 149 START_USE(vq); 150 151 BUG_ON(data == NULL); 152 153 /* If the host supports indirect descriptor tables, and we have multiple 154 * buffers, then go indirect. FIXME: tune this threshold */ 155 if (vq->indirect && (out + in) > 1 && vq->num_free) { 156 head = vring_add_indirect(vq, sg, out, in); 157 if (head != vq->vring.num) 158 goto add_head; 159 } 160 161 BUG_ON(out + in > vq->vring.num); 162 BUG_ON(out + in == 0); 163 164 if (vq->num_free < out + in) { 165 pr_debug("Can't add buf len %i - avail = %i\n", 166 out + in, vq->num_free); 167 /* FIXME: for historical reasons, we force a notify here if 168 * there are outgoing parts to the buffer. Presumably the 169 * host should service the ring ASAP. */ 170 if (out) 171 vq->notify(&vq->vq); 172 END_USE(vq); 173 return -ENOSPC; 174 } 175 176 /* We're about to use some buffers from the free list. */ 177 vq->num_free -= out + in; 178 179 head = vq->free_head; 180 for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) { 181 vq->vring.desc[i].flags = VRING_DESC_F_NEXT; 182 vq->vring.desc[i].addr = sg_phys(sg); 183 vq->vring.desc[i].len = sg->length; 184 prev = i; 185 sg++; 186 } 187 for (; in; i = vq->vring.desc[i].next, in--) { 188 vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; 189 vq->vring.desc[i].addr = sg_phys(sg); 190 vq->vring.desc[i].len = sg->length; 191 prev = i; 192 sg++; 193 } 194 /* Last one doesn't continue. */ 195 vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT; 196 197 /* Update free pointer */ 198 vq->free_head = i; 199 200 add_head: 201 /* Set token. */ 202 vq->data[head] = data; 203 204 /* Put entry in available array (but don't update avail->idx until they 205 * do sync). FIXME: avoid modulus here? */ 206 avail = (vq->vring.avail->idx + vq->num_added++) % vq->vring.num; 207 vq->vring.avail->ring[avail] = head; 208 209 pr_debug("Added buffer head %i to %p\n", head, vq); 210 END_USE(vq); 211 return 0; 212 } 213 214 static void vring_kick(struct virtqueue *_vq) 215 { 216 struct vring_virtqueue *vq = to_vvq(_vq); 217 START_USE(vq); 218 /* Descriptors and available array need to be set before we expose the 219 * new available array entries. */ 220 wmb(); 221 222 vq->vring.avail->idx += vq->num_added; 223 vq->num_added = 0; 224 225 /* Need to update avail index before checking if we should notify */ 226 mb(); 227 228 if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY)) 229 /* Prod other side to tell it about changes. */ 230 vq->notify(&vq->vq); 231 232 END_USE(vq); 233 } 234 235 static void detach_buf(struct vring_virtqueue *vq, unsigned int head) 236 { 237 unsigned int i; 238 239 /* Clear data ptr. */ 240 vq->data[head] = NULL; 241 242 /* Put back on free list: find end */ 243 i = head; 244 245 /* Free the indirect table */ 246 if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT) 247 kfree(phys_to_virt(vq->vring.desc[i].addr)); 248 249 while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) { 250 i = vq->vring.desc[i].next; 251 vq->num_free++; 252 } 253 254 vq->vring.desc[i].next = vq->free_head; 255 vq->free_head = head; 256 /* Plus final descriptor */ 257 vq->num_free++; 258 } 259 260 static inline bool more_used(const struct vring_virtqueue *vq) 261 { 262 return vq->last_used_idx != vq->vring.used->idx; 263 } 264 265 static void *vring_get_buf(struct virtqueue *_vq, unsigned int *len) 266 { 267 struct vring_virtqueue *vq = to_vvq(_vq); 268 void *ret; 269 unsigned int i; 270 271 START_USE(vq); 272 273 if (unlikely(vq->broken)) { 274 END_USE(vq); 275 return NULL; 276 } 277 278 if (!more_used(vq)) { 279 pr_debug("No more buffers in queue\n"); 280 END_USE(vq); 281 return NULL; 282 } 283 284 i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id; 285 *len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len; 286 287 if (unlikely(i >= vq->vring.num)) { 288 BAD_RING(vq, "id %u out of range\n", i); 289 return NULL; 290 } 291 if (unlikely(!vq->data[i])) { 292 BAD_RING(vq, "id %u is not a head!\n", i); 293 return NULL; 294 } 295 296 /* detach_buf clears data, so grab it now. */ 297 ret = vq->data[i]; 298 detach_buf(vq, i); 299 vq->last_used_idx++; 300 END_USE(vq); 301 return ret; 302 } 303 304 static void vring_disable_cb(struct virtqueue *_vq) 305 { 306 struct vring_virtqueue *vq = to_vvq(_vq); 307 308 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; 309 } 310 311 static bool vring_enable_cb(struct virtqueue *_vq) 312 { 313 struct vring_virtqueue *vq = to_vvq(_vq); 314 315 START_USE(vq); 316 317 /* We optimistically turn back on interrupts, then check if there was 318 * more to do. */ 319 vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; 320 mb(); 321 if (unlikely(more_used(vq))) { 322 END_USE(vq); 323 return false; 324 } 325 326 END_USE(vq); 327 return true; 328 } 329 330 irqreturn_t vring_interrupt(int irq, void *_vq) 331 { 332 struct vring_virtqueue *vq = to_vvq(_vq); 333 334 if (!more_used(vq)) { 335 pr_debug("virtqueue interrupt with no work for %p\n", vq); 336 return IRQ_NONE; 337 } 338 339 if (unlikely(vq->broken)) 340 return IRQ_HANDLED; 341 342 pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); 343 if (vq->vq.callback) 344 vq->vq.callback(&vq->vq); 345 346 return IRQ_HANDLED; 347 } 348 EXPORT_SYMBOL_GPL(vring_interrupt); 349 350 static struct virtqueue_ops vring_vq_ops = { 351 .add_buf = vring_add_buf, 352 .get_buf = vring_get_buf, 353 .kick = vring_kick, 354 .disable_cb = vring_disable_cb, 355 .enable_cb = vring_enable_cb, 356 }; 357 358 struct virtqueue *vring_new_virtqueue(unsigned int num, 359 unsigned int vring_align, 360 struct virtio_device *vdev, 361 void *pages, 362 void (*notify)(struct virtqueue *), 363 void (*callback)(struct virtqueue *), 364 const char *name) 365 { 366 struct vring_virtqueue *vq; 367 unsigned int i; 368 369 /* We assume num is a power of 2. */ 370 if (num & (num - 1)) { 371 dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); 372 return NULL; 373 } 374 375 vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL); 376 if (!vq) 377 return NULL; 378 379 vring_init(&vq->vring, num, pages, vring_align); 380 vq->vq.callback = callback; 381 vq->vq.vdev = vdev; 382 vq->vq.vq_ops = &vring_vq_ops; 383 vq->vq.name = name; 384 vq->notify = notify; 385 vq->broken = false; 386 vq->last_used_idx = 0; 387 vq->num_added = 0; 388 list_add_tail(&vq->vq.list, &vdev->vqs); 389 #ifdef DEBUG 390 vq->in_use = false; 391 #endif 392 393 vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC); 394 395 /* No callback? Tell other side not to bother us. */ 396 if (!callback) 397 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; 398 399 /* Put everything in free lists. */ 400 vq->num_free = num; 401 vq->free_head = 0; 402 for (i = 0; i < num-1; i++) 403 vq->vring.desc[i].next = i+1; 404 405 return &vq->vq; 406 } 407 EXPORT_SYMBOL_GPL(vring_new_virtqueue); 408 409 void vring_del_virtqueue(struct virtqueue *vq) 410 { 411 list_del(&vq->list); 412 kfree(to_vvq(vq)); 413 } 414 EXPORT_SYMBOL_GPL(vring_del_virtqueue); 415 416 /* Manipulates transport-specific feature bits. */ 417 void vring_transport_features(struct virtio_device *vdev) 418 { 419 unsigned int i; 420 421 for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { 422 switch (i) { 423 case VIRTIO_RING_F_INDIRECT_DESC: 424 break; 425 default: 426 /* We don't understand this bit. */ 427 clear_bit(i, vdev->features); 428 } 429 } 430 } 431 EXPORT_SYMBOL_GPL(vring_transport_features); 432 433 MODULE_LICENSE("GPL"); 434