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/device.h>
  22
  23 #ifdef DEBUG
  24 /* For development, we want to crash whenever the ring is screwed. */
  25 #define BAD_RING(vq, fmt...)                    \
  26         do { dev_err(&vq->vq.vdev->dev, fmt); BUG(); } while(0)
  27 #define START_USE(vq) \
  28         do { if ((vq)->in_use) panic("in_use = %i\n", (vq)->in_use); (vq)->in_use = __LINE__; mb(); } while(0)
  29 #define END_USE(vq) \
  30         do { BUG_ON(!(vq)->in_use); (vq)->in_use = 0; mb(); } while(0)
  31 #else
  32 #define BAD_RING(vq, fmt...)                    \
  33         do { dev_err(&vq->vq.vdev->dev, fmt); (vq)->broken = true; } while(0)
  34 #define START_USE(vq)
  35 #define END_USE(vq)
  36 #endif
  37
  38 struct vring_virtqueue
  39 {
  40         struct virtqueue vq;
  41
  42         /* Actual memory layout for this queue */
  43         struct vring vring;
  44
  45         /* Other side has made a mess, don't try any more. */
  46         bool broken;
  47
  48         /* Number of free buffers */
  49         unsigned int num_free;
  50         /* Head of free buffer list. */
  51         unsigned int free_head;
  52         /* Number we've added since last sync. */
  53         unsigned int num_added;
  54
  55         /* Last used index we've seen. */
  56         u16 last_used_idx;
  57
  58         /* How to notify other side. FIXME: commonalize hcalls! */
  59         void (*notify)(struct virtqueue *vq);
  60
  61 #ifdef DEBUG
  62         /* They're supposed to lock for us. */
  63         unsigned int in_use;
  64 #endif
  65
  66         /* Tokens for callbacks. */
  67         void *data[];
  68 };
  69
  70 #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq)
  71
  72 static int vring_add_buf(struct virtqueue *_vq,
  73                          struct scatterlist sg[],
  74                          unsigned int out,
  75                          unsigned int in,
  76                          void *data)
  77 {
  78         struct vring_virtqueue *vq = to_vvq(_vq);
  79         unsigned int i, avail, head, uninitialized_var(prev);
  80
  81         BUG_ON(data == NULL);
  82         BUG_ON(out + in > vq->vring.num);
  83         BUG_ON(out + in == 0);
  84
  85         START_USE(vq);
  86
  87         if (vq->num_free < out + in) {
  88                 pr_debug("Can't add buf len %i - avail = %i\n",
  89                          out + in, vq->num_free);
  90                 END_USE(vq);
  91                 return -ENOSPC;
  92         }
  93
  94         /* We're about to use some buffers from the free list. */
  95         vq->num_free -= out + in;
  96
  97         head = vq->free_head;
  98         for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) {
  99                 vq->vring.desc[i].flags = VRING_DESC_F_NEXT;
 100                 vq->vring.desc[i].addr = (page_to_pfn(sg_page(sg))<<PAGE_SHIFT)
 101                         + sg->offset;
 102                 vq->vring.desc[i].len = sg->length;
 103                 prev = i;
 104                 sg++;
 105         }
 106         for (; in; i = vq->vring.desc[i].next, in--) {
 107                 vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
 108                 vq->vring.desc[i].addr = (page_to_pfn(sg_page(sg))<<PAGE_SHIFT)
 109                         + sg->offset;
 110                 vq->vring.desc[i].len = sg->length;
 111                 prev = i;
 112                 sg++;
 113         }
 114         /* Last one doesn't continue. */
 115         vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT;
 116
 117         /* Update free pointer */
 118         vq->free_head = i;
 119
 120         /* Set token. */
 121         vq->data[head] = data;
 122
 123         /* Put entry in available array (but don't update avail->idx until they
 124          * do sync).  FIXME: avoid modulus here? */
 125         avail = (vq->vring.avail->idx + vq->num_added++) % vq->vring.num;
 126         vq->vring.avail->ring[avail] = head;
 127
 128         pr_debug("Added buffer head %i to %p\n", head, vq);
 129         END_USE(vq);
 130         return 0;
 131 }
 132
 133 static void vring_kick(struct virtqueue *_vq)
 134 {
 135         struct vring_virtqueue *vq = to_vvq(_vq);
 136         START_USE(vq);
 137         /* Descriptors and available array need to be set before we expose the
 138          * new available array entries. */
 139         wmb();
 140
 141         vq->vring.avail->idx += vq->num_added;
 142         vq->num_added = 0;
 143
 144         /* Need to update avail index before checking if we should notify */
 145         mb();
 146
 147         if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY))
 148                 /* Prod other side to tell it about changes. */
 149                 vq->notify(&vq->vq);
 150
 151         END_USE(vq);
 152 }
 153
 154 static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
 155 {
 156         unsigned int i;
 157
 158         /* Clear data ptr. */
 159         vq->data[head] = NULL;
 160
 161         /* Put back on free list: find end */
 162         i = head;
 163         while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
 164                 i = vq->vring.desc[i].next;
 165                 vq->num_free++;
 166         }
 167
 168         vq->vring.desc[i].next = vq->free_head;
 169         vq->free_head = head;
 170         /* Plus final descriptor */
 171         vq->num_free++;
 172 }
 173
 174 /* FIXME: We need to tell other side about removal, to synchronize. */
 175 static void vring_shutdown(struct virtqueue *_vq)
 176 {
 177         struct vring_virtqueue *vq = to_vvq(_vq);
 178         unsigned int i;
 179
 180         for (i = 0; i < vq->vring.num; i++)
 181                 detach_buf(vq, i);
 182 }
 183
 184 static inline bool more_used(const struct vring_virtqueue *vq)
 185 {
 186         return vq->last_used_idx != vq->vring.used->idx;
 187 }
 188
 189 static void *vring_get_buf(struct virtqueue *_vq, unsigned int *len)
 190 {
 191         struct vring_virtqueue *vq = to_vvq(_vq);
 192         void *ret;
 193         unsigned int i;
 194
 195         START_USE(vq);
 196
 197         if (!more_used(vq)) {
 198                 pr_debug("No more buffers in queue\n");
 199                 END_USE(vq);
 200                 return NULL;
 201         }
 202
 203         i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id;
 204         *len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len;
 205
 206         if (unlikely(i >= vq->vring.num)) {
 207                 BAD_RING(vq, "id %u out of range\n", i);
 208                 return NULL;
 209         }
 210         if (unlikely(!vq->data[i])) {
 211                 BAD_RING(vq, "id %u is not a head!\n", i);
 212                 return NULL;
 213         }
 214
 215         /* detach_buf clears data, so grab it now. */
 216         ret = vq->data[i];
 217         detach_buf(vq, i);
 218         vq->last_used_idx++;
 219         END_USE(vq);
 220         return ret;
 221 }
 222
 223 static bool vring_restart(struct virtqueue *_vq)
 224 {
 225         struct vring_virtqueue *vq = to_vvq(_vq);
 226
 227         START_USE(vq);
 228         BUG_ON(!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT));
 229
 230         /* We optimistically turn back on interrupts, then check if there was
 231          * more to do. */
 232         vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
 233         mb();
 234         if (unlikely(more_used(vq))) {
 235                 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
 236                 END_USE(vq);
 237                 return false;
 238         }
 239
 240         END_USE(vq);
 241         return true;
 242 }
 243
 244 irqreturn_t vring_interrupt(int irq, void *_vq)
 245 {
 246         struct vring_virtqueue *vq = to_vvq(_vq);
 247
 248         if (!more_used(vq)) {
 249                 pr_debug("virtqueue interrupt with no work for %p\n", vq);
 250                 return IRQ_NONE;
 251         }
 252
 253         if (unlikely(vq->broken))
 254                 return IRQ_HANDLED;
 255
 256         pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
 257         if (vq->vq.callback && !vq->vq.callback(&vq->vq))
 258                 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
 259
 260         return IRQ_HANDLED;
 261 }
 262
 263 static struct virtqueue_ops vring_vq_ops = {
 264         .add_buf = vring_add_buf,
 265         .get_buf = vring_get_buf,
 266         .kick = vring_kick,
 267         .restart = vring_restart,
 268         .shutdown = vring_shutdown,
 269 };
 270
 271 struct virtqueue *vring_new_virtqueue(unsigned int num,
 272                                       struct virtio_device *vdev,
 273                                       void *pages,
 274                                       void (*notify)(struct virtqueue *),
 275                                       bool (*callback)(struct virtqueue *))
 276 {
 277         struct vring_virtqueue *vq;
 278         unsigned int i;
 279
 280         /* We assume num is a power of 2. */
 281         if (num & (num - 1)) {
 282                 dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num);
 283                 return NULL;
 284         }
 285
 286         vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
 287         if (!vq)
 288                 return NULL;
 289
 290         vring_init(&vq->vring, num, pages, PAGE_SIZE);
 291         vq->vq.callback = callback;
 292         vq->vq.vdev = vdev;
 293         vq->vq.vq_ops = &vring_vq_ops;
 294         vq->notify = notify;
 295         vq->broken = false;
 296         vq->last_used_idx = 0;
 297         vq->num_added = 0;
 298 #ifdef DEBUG
 299         vq->in_use = false;
 300 #endif
 301
 302         /* No callback?  Tell other side not to bother us. */
 303         if (!callback)
 304                 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
 305
 306         /* Put everything in free lists. */
 307         vq->num_free = num;
 308         vq->free_head = 0;
 309         for (i = 0; i < num-1; i++)
 310                 vq->vring.desc[i].next = i+1;
 311
 312         return &vq->vq;
 313 }
 314
 315 void vring_del_virtqueue(struct virtqueue *vq)
 316 {
 317         kfree(to_vvq(vq));
 318 }
 319