drivers/virtio/virtio_ring.c

   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         /* If we're indirect, we can fit many (assuming not OOM). */
 234         if (vq->indirect)
 235                 return vq->num_free ? vq->vring.num : 0;
 236         return vq->num_free;
 237 }
 238 EXPORT_SYMBOL_GPL(virtqueue_add_buf_gfp);
 239
 240 void virtqueue_kick(struct virtqueue *_vq)
 241 {
 242         struct vring_virtqueue *vq = to_vvq(_vq);
 243         START_USE(vq);
 244         /* Descriptors and available array need to be set before we expose the
 245          * new available array entries. */
 246         virtio_wmb();
 247
 248         vq->vring.avail->idx += vq->num_added;
 249         vq->num_added = 0;
 250
 251         /* Need to update avail index before checking if we should notify */
 252         virtio_mb();
 253
 254         if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY))
 255                 /* Prod other side to tell it about changes. */
 256                 vq->notify(&vq->vq);
 257
 258         END_USE(vq);
 259 }
 260 EXPORT_SYMBOL_GPL(virtqueue_kick);
 261
 262 static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
 263 {
 264         unsigned int i;
 265
 266         /* Clear data ptr. */
 267         vq->data[head] = NULL;
 268
 269         /* Put back on free list: find end */
 270         i = head;
 271
 272         /* Free the indirect table */
 273         if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT)
 274                 kfree(phys_to_virt(vq->vring.desc[i].addr));
 275
 276         while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
 277                 i = vq->vring.desc[i].next;
 278                 vq->num_free++;
 279         }
 280
 281         vq->vring.desc[i].next = vq->free_head;
 282         vq->free_head = head;
 283         /* Plus final descriptor */
 284         vq->num_free++;
 285 }
 286
 287 static inline bool more_used(const struct vring_virtqueue *vq)
 288 {
 289         return vq->last_used_idx != vq->vring.used->idx;
 290 }
 291
 292 void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len)
 293 {
 294         struct vring_virtqueue *vq = to_vvq(_vq);
 295         void *ret;
 296         unsigned int i;
 297
 298         START_USE(vq);
 299
 300         if (unlikely(vq->broken)) {
 301                 END_USE(vq);
 302                 return NULL;
 303         }
 304
 305         if (!more_used(vq)) {
 306                 pr_debug("No more buffers in queue\n");
 307                 END_USE(vq);
 308                 return NULL;
 309         }
 310
 311         /* Only get used array entries after they have been exposed by host. */
 312         virtio_rmb();
 313
 314         i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id;
 315         *len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len;
 316
 317         if (unlikely(i >= vq->vring.num)) {
 318                 BAD_RING(vq, "id %u out of range\n", i);
 319                 return NULL;
 320         }
 321         if (unlikely(!vq->data[i])) {
 322                 BAD_RING(vq, "id %u is not a head!\n", i);
 323                 return NULL;
 324         }
 325
 326         /* detach_buf clears data, so grab it now. */
 327         ret = vq->data[i];
 328         detach_buf(vq, i);
 329         vq->last_used_idx++;
 330         END_USE(vq);
 331         return ret;
 332 }
 333 EXPORT_SYMBOL_GPL(virtqueue_get_buf);
 334
 335 void virtqueue_disable_cb(struct virtqueue *_vq)
 336 {
 337         struct vring_virtqueue *vq = to_vvq(_vq);
 338
 339         vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
 340 }
 341 EXPORT_SYMBOL_GPL(virtqueue_disable_cb);
 342
 343 bool virtqueue_enable_cb(struct virtqueue *_vq)
 344 {
 345         struct vring_virtqueue *vq = to_vvq(_vq);
 346
 347         START_USE(vq);
 348
 349         /* We optimistically turn back on interrupts, then check if there was
 350          * more to do. */
 351         vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
 352         virtio_mb();
 353         if (unlikely(more_used(vq))) {
 354                 END_USE(vq);
 355                 return false;
 356         }
 357
 358         END_USE(vq);
 359         return true;
 360 }
 361 EXPORT_SYMBOL_GPL(virtqueue_enable_cb);
 362
 363 void *virtqueue_detach_unused_buf(struct virtqueue *_vq)
 364 {
 365         struct vring_virtqueue *vq = to_vvq(_vq);
 366         unsigned int i;
 367         void *buf;
 368
 369         START_USE(vq);
 370
 371         for (i = 0; i < vq->vring.num; i++) {
 372                 if (!vq->data[i])
 373                         continue;
 374                 /* detach_buf clears data, so grab it now. */
 375                 buf = vq->data[i];
 376                 detach_buf(vq, i);
 377                 END_USE(vq);
 378                 return buf;
 379         }
 380         /* That should have freed everything. */
 381         BUG_ON(vq->num_free != vq->vring.num);
 382
 383         END_USE(vq);
 384         return NULL;
 385 }
 386 EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf);
 387
 388 irqreturn_t vring_interrupt(int irq, void *_vq)
 389 {
 390         struct vring_virtqueue *vq = to_vvq(_vq);
 391
 392         if (!more_used(vq)) {
 393                 pr_debug("virtqueue interrupt with no work for %p\n", vq);
 394                 return IRQ_NONE;
 395         }
 396
 397         if (unlikely(vq->broken))
 398                 return IRQ_HANDLED;
 399
 400         pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
 401         if (vq->vq.callback)
 402                 vq->vq.callback(&vq->vq);
 403
 404         return IRQ_HANDLED;
 405 }
 406 EXPORT_SYMBOL_GPL(vring_interrupt);
 407
 408 struct virtqueue *vring_new_virtqueue(unsigned int num,
 409                                       unsigned int vring_align,
 410                                       struct virtio_device *vdev,
 411                                       void *pages,
 412                                       void (*notify)(struct virtqueue *),
 413                                       void (*callback)(struct virtqueue *),
 414                                       const char *name)
 415 {
 416         struct vring_virtqueue *vq;
 417         unsigned int i;
 418
 419         /* We assume num is a power of 2. */
 420         if (num & (num - 1)) {
 421                 dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num);
 422                 return NULL;
 423         }
 424
 425         vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
 426         if (!vq)
 427                 return NULL;
 428
 429         vring_init(&vq->vring, num, pages, vring_align);
 430         vq->vq.callback = callback;
 431         vq->vq.vdev = vdev;
 432         vq->vq.name = name;
 433         vq->notify = notify;
 434         vq->broken = false;
 435         vq->last_used_idx = 0;
 436         vq->num_added = 0;
 437         list_add_tail(&vq->vq.list, &vdev->vqs);
 438 #ifdef DEBUG
 439         vq->in_use = false;
 440 #endif
 441
 442         vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC);
 443
 444         /* No callback?  Tell other side not to bother us. */
 445         if (!callback)
 446                 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
 447
 448         /* Put everything in free lists. */
 449         vq->num_free = num;
 450         vq->free_head = 0;
 451         for (i = 0; i < num-1; i++) {
 452                 vq->vring.desc[i].next = i+1;
 453                 vq->data[i] = NULL;
 454         }
 455         vq->data[i] = NULL;
 456
 457         return &vq->vq;
 458 }
 459 EXPORT_SYMBOL_GPL(vring_new_virtqueue);
 460
 461 void vring_del_virtqueue(struct virtqueue *vq)
 462 {
 463         list_del(&vq->list);
 464         kfree(to_vvq(vq));
 465 }
 466 EXPORT_SYMBOL_GPL(vring_del_virtqueue);
 467
 468 /* Manipulates transport-specific feature bits. */
 469 void vring_transport_features(struct virtio_device *vdev)
 470 {
 471         unsigned int i;
 472
 473         for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) {
 474                 switch (i) {
 475                 case VIRTIO_RING_F_INDIRECT_DESC:
 476                         break;
 477                 default:
 478                         /* We don't understand this bit. */
 479                         clear_bit(i, vdev->features);
 480                 }
 481         }
 482 }
 483 EXPORT_SYMBOL_GPL(vring_transport_features);
 484
 485 MODULE_LICENSE("GPL");