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 #include <linux/slab.h> 24 25 /* virtio guest is communicating with a virtual "device" that actually runs on 26 * a host processor. Memory barriers are used to control SMP effects. */ 27 #ifdef CONFIG_SMP 28 /* Where possible, use SMP barriers which are more lightweight than mandatory 29 * barriers, because mandatory barriers control MMIO effects on accesses 30 * through relaxed memory I/O windows (which virtio does not use). */ 31 #define virtio_mb() smp_mb() 32 #define virtio_rmb() smp_rmb() 33 #define virtio_wmb() smp_wmb() 34 #else 35 /* We must force memory ordering even if guest is UP since host could be 36 * running on another CPU, but SMP barriers are defined to barrier() in that 37 * configuration. So fall back to mandatory barriers instead. */ 38 #define virtio_mb() mb() 39 #define virtio_rmb() rmb() 40 #define virtio_wmb() wmb() 41 #endif 42 43 #ifdef DEBUG 44 /* For development, we want to crash whenever the ring is screwed. */ 45 #define BAD_RING(_vq, fmt, args...) \ 46 do { \ 47 dev_err(&(_vq)->vq.vdev->dev, \ 48 "%s:"fmt, (_vq)->vq.name, ##args); \ 49 BUG(); \ 50 } while (0) 51 /* Caller is supposed to guarantee no reentry. */ 52 #define START_USE(_vq) \ 53 do { \ 54 if ((_vq)->in_use) \ 55 panic("%s:in_use = %i\n", \ 56 (_vq)->vq.name, (_vq)->in_use); \ 57 (_vq)->in_use = __LINE__; \ 58 } while (0) 59 #define END_USE(_vq) \ 60 do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; } while(0) 61 #else 62 #define BAD_RING(_vq, fmt, args...) \ 63 do { \ 64 dev_err(&_vq->vq.vdev->dev, \ 65 "%s:"fmt, (_vq)->vq.name, ##args); \ 66 (_vq)->broken = true; \ 67 } while (0) 68 #define START_USE(vq) 69 #define END_USE(vq) 70 #endif 71 72 struct vring_virtqueue 73 { 74 struct virtqueue vq; 75 76 /* Actual memory layout for this queue */ 77 struct vring vring; 78 79 /* Other side has made a mess, don't try any more. */ 80 bool broken; 81 82 /* Host supports indirect buffers */ 83 bool indirect; 84 85 /* Number of free buffers */ 86 unsigned int num_free; 87 /* Head of free buffer list. */ 88 unsigned int free_head; 89 /* Number we've added since last sync. */ 90 unsigned int num_added; 91 92 /* Last used index we've seen. */ 93 u16 last_used_idx; 94 95 /* How to notify other side. FIXME: commonalize hcalls! */ 96 void (*notify)(struct virtqueue *vq); 97 98 #ifdef DEBUG 99 /* They're supposed to lock for us. */ 100 unsigned int in_use; 101 #endif 102 103 /* Tokens for callbacks. */ 104 void *data[]; 105 }; 106 107 #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq) 108 109 /* Set up an indirect table of descriptors and add it to the queue. */ 110 static int vring_add_indirect(struct vring_virtqueue *vq, 111 struct scatterlist sg[], 112 unsigned int out, 113 unsigned int in, 114 gfp_t gfp) 115 { 116 struct vring_desc *desc; 117 unsigned head; 118 int i; 119 120 desc = kmalloc((out + in) * sizeof(struct vring_desc), gfp); 121 if (!desc) 122 return -ENOMEM; 123 124 /* Transfer entries from the sg list into the indirect page */ 125 for (i = 0; i < out; i++) { 126 desc[i].flags = VRING_DESC_F_NEXT; 127 desc[i].addr = sg_phys(sg); 128 desc[i].len = sg->length; 129 desc[i].next = i+1; 130 sg++; 131 } 132 for (; i < (out + in); i++) { 133 desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; 134 desc[i].addr = sg_phys(sg); 135 desc[i].len = sg->length; 136 desc[i].next = i+1; 137 sg++; 138 } 139 140 /* Last one doesn't continue. */ 141 desc[i-1].flags &= ~VRING_DESC_F_NEXT; 142 desc[i-1].next = 0; 143 144 /* We're about to use a buffer */ 145 vq->num_free--; 146 147 /* Use a single buffer which doesn't continue */ 148 head = vq->free_head; 149 vq->vring.desc[head].flags = VRING_DESC_F_INDIRECT; 150 vq->vring.desc[head].addr = virt_to_phys(desc); 151 vq->vring.desc[head].len = i * sizeof(struct vring_desc); 152 153 /* Update free pointer */ 154 vq->free_head = vq->vring.desc[head].next; 155 156 return head; 157 } 158 159 int virtqueue_add_buf_gfp(struct virtqueue *_vq, 160 struct scatterlist sg[], 161 unsigned int out, 162 unsigned int in, 163 void *data, 164 gfp_t gfp) 165 { 166 struct vring_virtqueue *vq = to_vvq(_vq); 167 unsigned int i, avail, uninitialized_var(prev); 168 int head; 169 170 START_USE(vq); 171 172 BUG_ON(data == NULL); 173 174 /* If the host supports indirect descriptor tables, and we have multiple 175 * buffers, then go indirect. FIXME: tune this threshold */ 176 if (vq->indirect && (out + in) > 1 && vq->num_free) { 177 head = vring_add_indirect(vq, sg, out, in, gfp); 178 if (likely(head >= 0)) 179 goto add_head; 180 } 181 182 BUG_ON(out + in > vq->vring.num); 183 BUG_ON(out + in == 0); 184 185 if (vq->num_free < out + in) { 186 pr_debug("Can't add buf len %i - avail = %i\n", 187 out + in, vq->num_free); 188 /* FIXME: for historical reasons, we force a notify here if 189 * there are outgoing parts to the buffer. Presumably the 190 * host should service the ring ASAP. */ 191 if (out) 192 vq->notify(&vq->vq); 193 END_USE(vq); 194 return -ENOSPC; 195 } 196 197 /* We're about to use some buffers from the free list. */ 198 vq->num_free -= out + in; 199 200 head = vq->free_head; 201 for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) { 202 vq->vring.desc[i].flags = VRING_DESC_F_NEXT; 203 vq->vring.desc[i].addr = sg_phys(sg); 204 vq->vring.desc[i].len = sg->length; 205 prev = i; 206 sg++; 207 } 208 for (; in; i = vq->vring.desc[i].next, in--) { 209 vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE; 210 vq->vring.desc[i].addr = sg_phys(sg); 211 vq->vring.desc[i].len = sg->length; 212 prev = i; 213 sg++; 214 } 215 /* Last one doesn't continue. */ 216 vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT; 217 218 /* Update free pointer */ 219 vq->free_head = i; 220 221 add_head: 222 /* Set token. */ 223 vq->data[head] = data; 224 225 /* Put entry in available array (but don't update avail->idx until they 226 * do sync). FIXME: avoid modulus here? */ 227 avail = (vq->vring.avail->idx + vq->num_added++) % vq->vring.num; 228 vq->vring.avail->ring[avail] = head; 229 230 pr_debug("Added buffer head %i to %p\n", head, vq); 231 END_USE(vq); 232 233 return vq->num_free; 234 } 235 EXPORT_SYMBOL_GPL(virtqueue_add_buf_gfp); 236 237 void virtqueue_kick(struct virtqueue *_vq) 238 { 239 struct vring_virtqueue *vq = to_vvq(_vq); 240 START_USE(vq); 241 /* Descriptors and available array need to be set before we expose the 242 * new available array entries. */ 243 virtio_wmb(); 244 245 vq->vring.avail->idx += vq->num_added; 246 vq->num_added = 0; 247 248 /* Need to update avail index before checking if we should notify */ 249 virtio_mb(); 250 251 if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY)) 252 /* Prod other side to tell it about changes. */ 253 vq->notify(&vq->vq); 254 255 END_USE(vq); 256 } 257 EXPORT_SYMBOL_GPL(virtqueue_kick); 258 259 static void detach_buf(struct vring_virtqueue *vq, unsigned int head) 260 { 261 unsigned int i; 262 263 /* Clear data ptr. */ 264 vq->data[head] = NULL; 265 266 /* Put back on free list: find end */ 267 i = head; 268 269 /* Free the indirect table */ 270 if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT) 271 kfree(phys_to_virt(vq->vring.desc[i].addr)); 272 273 while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) { 274 i = vq->vring.desc[i].next; 275 vq->num_free++; 276 } 277 278 vq->vring.desc[i].next = vq->free_head; 279 vq->free_head = head; 280 /* Plus final descriptor */ 281 vq->num_free++; 282 } 283 284 static inline bool more_used(const struct vring_virtqueue *vq) 285 { 286 return vq->last_used_idx != vq->vring.used->idx; 287 } 288 289 void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len) 290 { 291 struct vring_virtqueue *vq = to_vvq(_vq); 292 void *ret; 293 unsigned int i; 294 295 START_USE(vq); 296 297 if (unlikely(vq->broken)) { 298 END_USE(vq); 299 return NULL; 300 } 301 302 if (!more_used(vq)) { 303 pr_debug("No more buffers in queue\n"); 304 END_USE(vq); 305 return NULL; 306 } 307 308 /* Only get used array entries after they have been exposed by host. */ 309 virtio_rmb(); 310 311 i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id; 312 *len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len; 313 314 if (unlikely(i >= vq->vring.num)) { 315 BAD_RING(vq, "id %u out of range\n", i); 316 return NULL; 317 } 318 if (unlikely(!vq->data[i])) { 319 BAD_RING(vq, "id %u is not a head!\n", i); 320 return NULL; 321 } 322 323 /* detach_buf clears data, so grab it now. */ 324 ret = vq->data[i]; 325 detach_buf(vq, i); 326 vq->last_used_idx++; 327 END_USE(vq); 328 return ret; 329 } 330 EXPORT_SYMBOL_GPL(virtqueue_get_buf); 331 332 void virtqueue_disable_cb(struct virtqueue *_vq) 333 { 334 struct vring_virtqueue *vq = to_vvq(_vq); 335 336 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; 337 } 338 EXPORT_SYMBOL_GPL(virtqueue_disable_cb); 339 340 bool virtqueue_enable_cb(struct virtqueue *_vq) 341 { 342 struct vring_virtqueue *vq = to_vvq(_vq); 343 344 START_USE(vq); 345 346 /* We optimistically turn back on interrupts, then check if there was 347 * more to do. */ 348 vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT; 349 virtio_mb(); 350 if (unlikely(more_used(vq))) { 351 END_USE(vq); 352 return false; 353 } 354 355 END_USE(vq); 356 return true; 357 } 358 EXPORT_SYMBOL_GPL(virtqueue_enable_cb); 359 360 void *virtqueue_detach_unused_buf(struct virtqueue *_vq) 361 { 362 struct vring_virtqueue *vq = to_vvq(_vq); 363 unsigned int i; 364 void *buf; 365 366 START_USE(vq); 367 368 for (i = 0; i < vq->vring.num; i++) { 369 if (!vq->data[i]) 370 continue; 371 /* detach_buf clears data, so grab it now. */ 372 buf = vq->data[i]; 373 detach_buf(vq, i); 374 END_USE(vq); 375 return buf; 376 } 377 /* That should have freed everything. */ 378 BUG_ON(vq->num_free != vq->vring.num); 379 380 END_USE(vq); 381 return NULL; 382 } 383 EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf); 384 385 irqreturn_t vring_interrupt(int irq, void *_vq) 386 { 387 struct vring_virtqueue *vq = to_vvq(_vq); 388 389 if (!more_used(vq)) { 390 pr_debug("virtqueue interrupt with no work for %p\n", vq); 391 return IRQ_NONE; 392 } 393 394 if (unlikely(vq->broken)) 395 return IRQ_HANDLED; 396 397 pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback); 398 if (vq->vq.callback) 399 vq->vq.callback(&vq->vq); 400 401 return IRQ_HANDLED; 402 } 403 EXPORT_SYMBOL_GPL(vring_interrupt); 404 405 struct virtqueue *vring_new_virtqueue(unsigned int num, 406 unsigned int vring_align, 407 struct virtio_device *vdev, 408 void *pages, 409 void (*notify)(struct virtqueue *), 410 void (*callback)(struct virtqueue *), 411 const char *name) 412 { 413 struct vring_virtqueue *vq; 414 unsigned int i; 415 416 /* We assume num is a power of 2. */ 417 if (num & (num - 1)) { 418 dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num); 419 return NULL; 420 } 421 422 vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL); 423 if (!vq) 424 return NULL; 425 426 vring_init(&vq->vring, num, pages, vring_align); 427 vq->vq.callback = callback; 428 vq->vq.vdev = vdev; 429 vq->vq.name = name; 430 vq->notify = notify; 431 vq->broken = false; 432 vq->last_used_idx = 0; 433 vq->num_added = 0; 434 list_add_tail(&vq->vq.list, &vdev->vqs); 435 #ifdef DEBUG 436 vq->in_use = false; 437 #endif 438 439 vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC); 440 441 /* No callback? Tell other side not to bother us. */ 442 if (!callback) 443 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT; 444 445 /* Put everything in free lists. */ 446 vq->num_free = num; 447 vq->free_head = 0; 448 for (i = 0; i < num-1; i++) { 449 vq->vring.desc[i].next = i+1; 450 vq->data[i] = NULL; 451 } 452 vq->data[i] = NULL; 453 454 return &vq->vq; 455 } 456 EXPORT_SYMBOL_GPL(vring_new_virtqueue); 457 458 void vring_del_virtqueue(struct virtqueue *vq) 459 { 460 list_del(&vq->list); 461 kfree(to_vvq(vq)); 462 } 463 EXPORT_SYMBOL_GPL(vring_del_virtqueue); 464 465 /* Manipulates transport-specific feature bits. */ 466 void vring_transport_features(struct virtio_device *vdev) 467 { 468 unsigned int i; 469 470 for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) { 471 switch (i) { 472 case VIRTIO_RING_F_INDIRECT_DESC: 473 break; 474 default: 475 /* We don't understand this bit. */ 476 clear_bit(i, vdev->features); 477 } 478 } 479 } 480 EXPORT_SYMBOL_GPL(vring_transport_features); 481 482 MODULE_LICENSE("GPL"); 483