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
  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
 212         /* If we're indirect, we can fit many (assuming not OOM). */
 213         if (vq->indirect)
 214                 return vq->num_free ? vq->vring.num : 0;
 215         return vq->num_free;
 216 }
 217
 218 static void vring_kick(struct virtqueue *_vq)
 219 {
 220         struct vring_virtqueue *vq = to_vvq(_vq);
 221         START_USE(vq);
 222         /* Descriptors and available array need to be set before we expose the
 223          * new available array entries. */
 224         wmb();
 225
 226         vq->vring.avail->idx += vq->num_added;
 227         vq->num_added = 0;
 228
 229         /* Need to update avail index before checking if we should notify */
 230         mb();
 231
 232         if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY))
 233                 /* Prod other side to tell it about changes. */
 234                 vq->notify(&vq->vq);
 235
 236         END_USE(vq);
 237 }
 238
 239 static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
 240 {
 241         unsigned int i;
 242
 243         /* Clear data ptr. */
 244         vq->data[head] = NULL;
 245
 246         /* Put back on free list: find end */
 247         i = head;
 248
 249         /* Free the indirect table */
 250         if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT)
 251                 kfree(phys_to_virt(vq->vring.desc[i].addr));
 252
 253         while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
 254                 i = vq->vring.desc[i].next;
 255                 vq->num_free++;
 256         }
 257
 258         vq->vring.desc[i].next = vq->free_head;
 259         vq->free_head = head;
 260         /* Plus final descriptor */
 261         vq->num_free++;
 262 }
 263
 264 static inline bool more_used(const struct vring_virtqueue *vq)
 265 {
 266         return vq->last_used_idx != vq->vring.used->idx;
 267 }
 268
 269 static void *vring_get_buf(struct virtqueue *_vq, unsigned int *len)
 270 {
 271         struct vring_virtqueue *vq = to_vvq(_vq);
 272         void *ret;
 273         unsigned int i;
 274
 275         START_USE(vq);
 276
 277         if (unlikely(vq->broken)) {
 278                 END_USE(vq);
 279                 return NULL;
 280         }
 281
 282         if (!more_used(vq)) {
 283                 pr_debug("No more buffers in queue\n");
 284                 END_USE(vq);
 285                 return NULL;
 286         }
 287
 288         i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id;
 289         *len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len;
 290
 291         if (unlikely(i >= vq->vring.num)) {
 292                 BAD_RING(vq, "id %u out of range\n", i);
 293                 return NULL;
 294         }
 295         if (unlikely(!vq->data[i])) {
 296                 BAD_RING(vq, "id %u is not a head!\n", i);
 297                 return NULL;
 298         }
 299
 300         /* detach_buf clears data, so grab it now. */
 301         ret = vq->data[i];
 302         detach_buf(vq, i);
 303         vq->last_used_idx++;
 304         END_USE(vq);
 305         return ret;
 306 }
 307
 308 static void vring_disable_cb(struct virtqueue *_vq)
 309 {
 310         struct vring_virtqueue *vq = to_vvq(_vq);
 311
 312         vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
 313 }
 314
 315 static bool vring_enable_cb(struct virtqueue *_vq)
 316 {
 317         struct vring_virtqueue *vq = to_vvq(_vq);
 318
 319         START_USE(vq);
 320
 321         /* We optimistically turn back on interrupts, then check if there was
 322          * more to do. */
 323         vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
 324         mb();
 325         if (unlikely(more_used(vq))) {
 326                 END_USE(vq);
 327                 return false;
 328         }
 329
 330         END_USE(vq);
 331         return true;
 332 }
 333
 334 irqreturn_t vring_interrupt(int irq, void *_vq)
 335 {
 336         struct vring_virtqueue *vq = to_vvq(_vq);
 337
 338         if (!more_used(vq)) {
 339                 pr_debug("virtqueue interrupt with no work for %p\n", vq);
 340                 return IRQ_NONE;
 341         }
 342
 343         if (unlikely(vq->broken))
 344                 return IRQ_HANDLED;
 345
 346         pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
 347         if (vq->vq.callback)
 348                 vq->vq.callback(&vq->vq);
 349
 350         return IRQ_HANDLED;
 351 }
 352 EXPORT_SYMBOL_GPL(vring_interrupt);
 353
 354 static struct virtqueue_ops vring_vq_ops = {
 355         .add_buf = vring_add_buf,
 356         .get_buf = vring_get_buf,
 357         .kick = vring_kick,
 358         .disable_cb = vring_disable_cb,
 359         .enable_cb = vring_enable_cb,
 360 };
 361
 362 struct virtqueue *vring_new_virtqueue(unsigned int num,
 363                                       unsigned int vring_align,
 364                                       struct virtio_device *vdev,
 365                                       void *pages,
 366                                       void (*notify)(struct virtqueue *),
 367                                       void (*callback)(struct virtqueue *),
 368                                       const char *name)
 369 {
 370         struct vring_virtqueue *vq;
 371         unsigned int i;
 372
 373         /* We assume num is a power of 2. */
 374         if (num & (num - 1)) {
 375                 dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num);
 376                 return NULL;
 377         }
 378
 379         vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
 380         if (!vq)
 381                 return NULL;
 382
 383         vring_init(&vq->vring, num, pages, vring_align);
 384         vq->vq.callback = callback;
 385         vq->vq.vdev = vdev;
 386         vq->vq.vq_ops = &vring_vq_ops;
 387         vq->vq.name = name;
 388         vq->notify = notify;
 389         vq->broken = false;
 390         vq->last_used_idx = 0;
 391         vq->num_added = 0;
 392         list_add_tail(&vq->vq.list, &vdev->vqs);
 393 #ifdef DEBUG
 394         vq->in_use = false;
 395 #endif
 396
 397         vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC);
 398
 399         /* No callback?  Tell other side not to bother us. */
 400         if (!callback)
 401                 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
 402
 403         /* Put everything in free lists. */
 404         vq->num_free = num;
 405         vq->free_head = 0;
 406         for (i = 0; i < num-1; i++)
 407                 vq->vring.desc[i].next = i+1;
 408
 409         return &vq->vq;
 410 }
 411 EXPORT_SYMBOL_GPL(vring_new_virtqueue);
 412
 413 void vring_del_virtqueue(struct virtqueue *vq)
 414 {
 415         list_del(&vq->list);
 416         kfree(to_vvq(vq));
 417 }
 418 EXPORT_SYMBOL_GPL(vring_del_virtqueue);
 419
 420 /* Manipulates transport-specific feature bits. */
 421 void vring_transport_features(struct virtio_device *vdev)
 422 {
 423         unsigned int i;
 424
 425         for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) {
 426                 switch (i) {
 427                 case VIRTIO_RING_F_INDIRECT_DESC:
 428                         break;
 429                 default:
 430                         /* We don't understand this bit. */
 431                         clear_bit(i, vdev->features);
 432                 }
 433         }
 434 }
 435 EXPORT_SYMBOL_GPL(vring_transport_features);
 436
 437 MODULE_LICENSE("GPL");