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 {
 115         struct vring_desc *desc;
 116         unsigned head;
 117         int i;
 118
 119         desc = kmalloc((out + in) * sizeof(struct vring_desc), GFP_ATOMIC);
 120         if (!desc)
 121                 return -ENOMEM;
 122
 123         /* Transfer entries from the sg list into the indirect page */
 124         for (i = 0; i < out; i++) {
 125                 desc[i].flags = VRING_DESC_F_NEXT;
 126                 desc[i].addr = sg_phys(sg);
 127                 desc[i].len = sg->length;
 128                 desc[i].next = i+1;
 129                 sg++;
 130         }
 131         for (; i < (out + in); i++) {
 132                 desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
 133                 desc[i].addr = sg_phys(sg);
 134                 desc[i].len = sg->length;
 135                 desc[i].next = i+1;
 136                 sg++;
 137         }
 138
 139         /* Last one doesn't continue. */
 140         desc[i-1].flags &= ~VRING_DESC_F_NEXT;
 141         desc[i-1].next = 0;
 142
 143         /* We're about to use a buffer */
 144         vq->num_free--;
 145
 146         /* Use a single buffer which doesn't continue */
 147         head = vq->free_head;
 148         vq->vring.desc[head].flags = VRING_DESC_F_INDIRECT;
 149         vq->vring.desc[head].addr = virt_to_phys(desc);
 150         vq->vring.desc[head].len = i * sizeof(struct vring_desc);
 151
 152         /* Update free pointer */
 153         vq->free_head = vq->vring.desc[head].next;
 154
 155         return head;
 156 }
 157
 158 static int vring_add_buf(struct virtqueue *_vq,
 159                          struct scatterlist sg[],
 160                          unsigned int out,
 161                          unsigned int in,
 162                          void *data)
 163 {
 164         struct vring_virtqueue *vq = to_vvq(_vq);
 165         unsigned int i, avail, uninitialized_var(prev);
 166         int head;
 167
 168         START_USE(vq);
 169
 170         BUG_ON(data == NULL);
 171
 172         /* If the host supports indirect descriptor tables, and we have multiple
 173          * buffers, then go indirect. FIXME: tune this threshold */
 174         if (vq->indirect && (out + in) > 1 && vq->num_free) {
 175                 head = vring_add_indirect(vq, sg, out, in);
 176                 if (likely(head >= 0))
 177                         goto add_head;
 178         }
 179
 180         BUG_ON(out + in > vq->vring.num);
 181         BUG_ON(out + in == 0);
 182
 183         if (vq->num_free < out + in) {
 184                 pr_debug("Can't add buf len %i - avail = %i\n",
 185                          out + in, vq->num_free);
 186                 /* FIXME: for historical reasons, we force a notify here if
 187                  * there are outgoing parts to the buffer.  Presumably the
 188                  * host should service the ring ASAP. */
 189                 if (out)
 190                         vq->notify(&vq->vq);
 191                 END_USE(vq);
 192                 return -ENOSPC;
 193         }
 194
 195         /* We're about to use some buffers from the free list. */
 196         vq->num_free -= out + in;
 197
 198         head = vq->free_head;
 199         for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) {
 200                 vq->vring.desc[i].flags = VRING_DESC_F_NEXT;
 201                 vq->vring.desc[i].addr = sg_phys(sg);
 202                 vq->vring.desc[i].len = sg->length;
 203                 prev = i;
 204                 sg++;
 205         }
 206         for (; in; i = vq->vring.desc[i].next, in--) {
 207                 vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
 208                 vq->vring.desc[i].addr = sg_phys(sg);
 209                 vq->vring.desc[i].len = sg->length;
 210                 prev = i;
 211                 sg++;
 212         }
 213         /* Last one doesn't continue. */
 214         vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT;
 215
 216         /* Update free pointer */
 217         vq->free_head = i;
 218
 219 add_head:
 220         /* Set token. */
 221         vq->data[head] = data;
 222
 223         /* Put entry in available array (but don't update avail->idx until they
 224          * do sync).  FIXME: avoid modulus here? */
 225         avail = (vq->vring.avail->idx + vq->num_added++) % vq->vring.num;
 226         vq->vring.avail->ring[avail] = head;
 227
 228         pr_debug("Added buffer head %i to %p\n", head, vq);
 229         END_USE(vq);
 230
 231         /* If we're indirect, we can fit many (assuming not OOM). */
 232         if (vq->indirect)
 233                 return vq->num_free ? vq->vring.num : 0;
 234         return vq->num_free;
 235 }
 236
 237 static void vring_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
 258 static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
 259 {
 260         unsigned int i;
 261
 262         /* Clear data ptr. */
 263         vq->data[head] = NULL;
 264
 265         /* Put back on free list: find end */
 266         i = head;
 267
 268         /* Free the indirect table */
 269         if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT)
 270                 kfree(phys_to_virt(vq->vring.desc[i].addr));
 271
 272         while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
 273                 i = vq->vring.desc[i].next;
 274                 vq->num_free++;
 275         }
 276
 277         vq->vring.desc[i].next = vq->free_head;
 278         vq->free_head = head;
 279         /* Plus final descriptor */
 280         vq->num_free++;
 281 }
 282
 283 static inline bool more_used(const struct vring_virtqueue *vq)
 284 {
 285         return vq->last_used_idx != vq->vring.used->idx;
 286 }
 287
 288 static void *vring_get_buf(struct virtqueue *_vq, unsigned int *len)
 289 {
 290         struct vring_virtqueue *vq = to_vvq(_vq);
 291         void *ret;
 292         unsigned int i;
 293
 294         START_USE(vq);
 295
 296         if (unlikely(vq->broken)) {
 297                 END_USE(vq);
 298                 return NULL;
 299         }
 300
 301         if (!more_used(vq)) {
 302                 pr_debug("No more buffers in queue\n");
 303                 END_USE(vq);
 304                 return NULL;
 305         }
 306
 307         /* Only get used array entries after they have been exposed by host. */
 308         virtio_rmb();
 309
 310         i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id;
 311         *len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len;
 312
 313         if (unlikely(i >= vq->vring.num)) {
 314                 BAD_RING(vq, "id %u out of range\n", i);
 315                 return NULL;
 316         }
 317         if (unlikely(!vq->data[i])) {
 318                 BAD_RING(vq, "id %u is not a head!\n", i);
 319                 return NULL;
 320         }
 321
 322         /* detach_buf clears data, so grab it now. */
 323         ret = vq->data[i];
 324         detach_buf(vq, i);
 325         vq->last_used_idx++;
 326         END_USE(vq);
 327         return ret;
 328 }
 329
 330 static void vring_disable_cb(struct virtqueue *_vq)
 331 {
 332         struct vring_virtqueue *vq = to_vvq(_vq);
 333
 334         vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
 335 }
 336
 337 static bool vring_enable_cb(struct virtqueue *_vq)
 338 {
 339         struct vring_virtqueue *vq = to_vvq(_vq);
 340
 341         START_USE(vq);
 342
 343         /* We optimistically turn back on interrupts, then check if there was
 344          * more to do. */
 345         vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
 346         virtio_mb();
 347         if (unlikely(more_used(vq))) {
 348                 END_USE(vq);
 349                 return false;
 350         }
 351
 352         END_USE(vq);
 353         return true;
 354 }
 355
 356 static void *vring_detach_unused_buf(struct virtqueue *_vq)
 357 {
 358         struct vring_virtqueue *vq = to_vvq(_vq);
 359         unsigned int i;
 360         void *buf;
 361
 362         START_USE(vq);
 363
 364         for (i = 0; i < vq->vring.num; i++) {
 365                 if (!vq->data[i])
 366                         continue;
 367                 /* detach_buf clears data, so grab it now. */
 368                 buf = vq->data[i];
 369                 detach_buf(vq, i);
 370                 END_USE(vq);
 371                 return buf;
 372         }
 373         /* That should have freed everything. */
 374         BUG_ON(vq->num_free != vq->vring.num);
 375
 376         END_USE(vq);
 377         return NULL;
 378 }
 379
 380 irqreturn_t vring_interrupt(int irq, void *_vq)
 381 {
 382         struct vring_virtqueue *vq = to_vvq(_vq);
 383
 384         if (!more_used(vq)) {
 385                 pr_debug("virtqueue interrupt with no work for %p\n", vq);
 386                 return IRQ_NONE;
 387         }
 388
 389         if (unlikely(vq->broken))
 390                 return IRQ_HANDLED;
 391
 392         pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
 393         if (vq->vq.callback)
 394                 vq->vq.callback(&vq->vq);
 395
 396         return IRQ_HANDLED;
 397 }
 398 EXPORT_SYMBOL_GPL(vring_interrupt);
 399
 400 static struct virtqueue_ops vring_vq_ops = {
 401         .add_buf = vring_add_buf,
 402         .get_buf = vring_get_buf,
 403         .kick = vring_kick,
 404         .disable_cb = vring_disable_cb,
 405         .enable_cb = vring_enable_cb,
 406         .detach_unused_buf = vring_detach_unused_buf,
 407 };
 408
 409 struct virtqueue *vring_new_virtqueue(unsigned int num,
 410                                       unsigned int vring_align,
 411                                       struct virtio_device *vdev,
 412                                       void *pages,
 413                                       void (*notify)(struct virtqueue *),
 414                                       void (*callback)(struct virtqueue *),
 415                                       const char *name)
 416 {
 417         struct vring_virtqueue *vq;
 418         unsigned int i;
 419
 420         /* We assume num is a power of 2. */
 421         if (num & (num - 1)) {
 422                 dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num);
 423                 return NULL;
 424         }
 425
 426         vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
 427         if (!vq)
 428                 return NULL;
 429
 430         vring_init(&vq->vring, num, pages, vring_align);
 431         vq->vq.callback = callback;
 432         vq->vq.vdev = vdev;
 433         vq->vq.vq_ops = &vring_vq_ops;
 434         vq->vq.name = name;
 435         vq->notify = notify;
 436         vq->broken = false;
 437         vq->last_used_idx = 0;
 438         vq->num_added = 0;
 439         list_add_tail(&vq->vq.list, &vdev->vqs);
 440 #ifdef DEBUG
 441         vq->in_use = false;
 442 #endif
 443
 444         vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC);
 445
 446         /* No callback?  Tell other side not to bother us. */
 447         if (!callback)
 448                 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
 449
 450         /* Put everything in free lists. */
 451         vq->num_free = num;
 452         vq->free_head = 0;
 453         for (i = 0; i < num-1; i++) {
 454                 vq->vring.desc[i].next = i+1;
 455                 vq->data[i] = NULL;
 456         }
 457         vq->data[i] = NULL;
 458
 459         return &vq->vq;
 460 }
 461 EXPORT_SYMBOL_GPL(vring_new_virtqueue);
 462
 463 void vring_del_virtqueue(struct virtqueue *vq)
 464 {
 465         list_del(&vq->list);
 466         kfree(to_vvq(vq));
 467 }
 468 EXPORT_SYMBOL_GPL(vring_del_virtqueue);
 469
 470 /* Manipulates transport-specific feature bits. */
 471 void vring_transport_features(struct virtio_device *vdev)
 472 {
 473         unsigned int i;
 474
 475         for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) {
 476                 switch (i) {
 477                 case VIRTIO_RING_F_INDIRECT_DESC:
 478                         break;
 479                 default:
 480                         /* We don't understand this bit. */
 481                         clear_bit(i, vdev->features);
 482                 }
 483         }
 484 }
 485 EXPORT_SYMBOL_GPL(vring_transport_features);
 486
 487 MODULE_LICENSE("GPL");