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...)                    \
  27         do { dev_err(&vq->vq.vdev->dev, fmt); BUG(); } while(0)
  28 #define START_USE(vq) \
  29         do { if ((vq)->in_use) panic("in_use = %i\n", (vq)->in_use); (vq)->in_use = __LINE__; mb(); } while(0)
  30 #define END_USE(vq) \
  31         do { BUG_ON(!(vq)->in_use); (vq)->in_use = 0; mb(); } while(0)
  32 #else
  33 #define BAD_RING(vq, fmt...)                    \
  34         do { dev_err(&vq->vq.vdev->dev, fmt); (vq)->broken = true; } while(0)
  35 #define START_USE(vq)
  36 #define END_USE(vq)
  37 #endif
  38
  39 struct vring_virtqueue
  40 {
  41         struct virtqueue vq;
  42
  43         /* Actual memory layout for this queue */
  44         struct vring vring;
  45
  46         /* Other side has made a mess, don't try any more. */
  47         bool broken;
  48
  49         /* Number of free buffers */
  50         unsigned int num_free;
  51         /* Head of free buffer list. */
  52         unsigned int free_head;
  53         /* Number we've added since last sync. */
  54         unsigned int num_added;
  55
  56         /* Last used index we've seen. */
  57         u16 last_used_idx;
  58
  59         /* How to notify other side. FIXME: commonalize hcalls! */
  60         void (*notify)(struct virtqueue *vq);
  61
  62 #ifdef DEBUG
  63         /* They're supposed to lock for us. */
  64         unsigned int in_use;
  65 #endif
  66
  67         /* Tokens for callbacks. */
  68         void *data[];
  69 };
  70
  71 #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq)
  72
  73 static int vring_add_buf(struct virtqueue *_vq,
  74                          struct scatterlist sg[],
  75                          unsigned int out,
  76                          unsigned int in,
  77                          void *data)
  78 {
  79         struct vring_virtqueue *vq = to_vvq(_vq);
  80         unsigned int i, avail, head, uninitialized_var(prev);
  81
  82         BUG_ON(data == NULL);
  83         BUG_ON(out + in > vq->vring.num);
  84         BUG_ON(out + in == 0);
  85
  86         START_USE(vq);
  87
  88         if (vq->num_free < out + in) {
  89                 pr_debug("Can't add buf len %i - avail = %i\n",
  90                          out + in, vq->num_free);
  91                 /* FIXME: for historical reasons, we force a notify here if
  92                  * there are outgoing parts to the buffer.  Presumably the
  93                  * host should service the ring ASAP. */
  94                 if (out)
  95                         vq->notify(&vq->vq);
  96                 END_USE(vq);
  97                 return -ENOSPC;
  98         }
  99
 100         /* We're about to use some buffers from the free list. */
 101         vq->num_free -= out + in;
 102
 103         head = vq->free_head;
 104         for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) {
 105                 vq->vring.desc[i].flags = VRING_DESC_F_NEXT;
 106                 vq->vring.desc[i].addr = sg_phys(sg);
 107                 vq->vring.desc[i].len = sg->length;
 108                 prev = i;
 109                 sg++;
 110         }
 111         for (; in; i = vq->vring.desc[i].next, in--) {
 112                 vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
 113                 vq->vring.desc[i].addr = sg_phys(sg);
 114                 vq->vring.desc[i].len = sg->length;
 115                 prev = i;
 116                 sg++;
 117         }
 118         /* Last one doesn't continue. */
 119         vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT;
 120
 121         /* Update free pointer */
 122         vq->free_head = i;
 123
 124         /* Set token. */
 125         vq->data[head] = data;
 126
 127         /* Put entry in available array (but don't update avail->idx until they
 128          * do sync).  FIXME: avoid modulus here? */
 129         avail = (vq->vring.avail->idx + vq->num_added++) % vq->vring.num;
 130         vq->vring.avail->ring[avail] = head;
 131
 132         pr_debug("Added buffer head %i to %p\n", head, vq);
 133         END_USE(vq);
 134         return 0;
 135 }
 136
 137 static void vring_kick(struct virtqueue *_vq)
 138 {
 139         struct vring_virtqueue *vq = to_vvq(_vq);
 140         START_USE(vq);
 141         /* Descriptors and available array need to be set before we expose the
 142          * new available array entries. */
 143         wmb();
 144
 145         vq->vring.avail->idx += vq->num_added;
 146         vq->num_added = 0;
 147
 148         /* Need to update avail index before checking if we should notify */
 149         mb();
 150
 151         if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY))
 152                 /* Prod other side to tell it about changes. */
 153                 vq->notify(&vq->vq);
 154
 155         END_USE(vq);
 156 }
 157
 158 static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
 159 {
 160         unsigned int i;
 161
 162         /* Clear data ptr. */
 163         vq->data[head] = NULL;
 164
 165         /* Put back on free list: find end */
 166         i = head;
 167         while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
 168                 i = vq->vring.desc[i].next;
 169                 vq->num_free++;
 170         }
 171
 172         vq->vring.desc[i].next = vq->free_head;
 173         vq->free_head = head;
 174         /* Plus final descriptor */
 175         vq->num_free++;
 176 }
 177
 178 static inline bool more_used(const struct vring_virtqueue *vq)
 179 {
 180         return vq->last_used_idx != vq->vring.used->idx;
 181 }
 182
 183 static void *vring_get_buf(struct virtqueue *_vq, unsigned int *len)
 184 {
 185         struct vring_virtqueue *vq = to_vvq(_vq);
 186         void *ret;
 187         unsigned int i;
 188
 189         START_USE(vq);
 190
 191         if (unlikely(vq->broken)) {
 192                 END_USE(vq);
 193                 return NULL;
 194         }
 195
 196         if (!more_used(vq)) {
 197                 pr_debug("No more buffers in queue\n");
 198                 END_USE(vq);
 199                 return NULL;
 200         }
 201
 202         i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id;
 203         *len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len;
 204
 205         if (unlikely(i >= vq->vring.num)) {
 206                 BAD_RING(vq, "id %u out of range\n", i);
 207                 return NULL;
 208         }
 209         if (unlikely(!vq->data[i])) {
 210                 BAD_RING(vq, "id %u is not a head!\n", i);
 211                 return NULL;
 212         }
 213
 214         /* detach_buf clears data, so grab it now. */
 215         ret = vq->data[i];
 216         detach_buf(vq, i);
 217         vq->last_used_idx++;
 218         END_USE(vq);
 219         return ret;
 220 }
 221
 222 static void vring_disable_cb(struct virtqueue *_vq)
 223 {
 224         struct vring_virtqueue *vq = to_vvq(_vq);
 225
 226         vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
 227 }
 228
 229 static bool vring_enable_cb(struct virtqueue *_vq)
 230 {
 231         struct vring_virtqueue *vq = to_vvq(_vq);
 232
 233         START_USE(vq);
 234
 235         /* We optimistically turn back on interrupts, then check if there was
 236          * more to do. */
 237         vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
 238         mb();
 239         if (unlikely(more_used(vq))) {
 240                 END_USE(vq);
 241                 return false;
 242         }
 243
 244         END_USE(vq);
 245         return true;
 246 }
 247
 248 irqreturn_t vring_interrupt(int irq, void *_vq)
 249 {
 250         struct vring_virtqueue *vq = to_vvq(_vq);
 251
 252         if (!more_used(vq)) {
 253                 pr_debug("virtqueue interrupt with no work for %p\n", vq);
 254                 return IRQ_NONE;
 255         }
 256
 257         if (unlikely(vq->broken))
 258                 return IRQ_HANDLED;
 259
 260         pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
 261         if (vq->vq.callback)
 262                 vq->vq.callback(&vq->vq);
 263
 264         return IRQ_HANDLED;
 265 }
 266 EXPORT_SYMBOL_GPL(vring_interrupt);
 267
 268 static struct virtqueue_ops vring_vq_ops = {
 269         .add_buf = vring_add_buf,
 270         .get_buf = vring_get_buf,
 271         .kick = vring_kick,
 272         .disable_cb = vring_disable_cb,
 273         .enable_cb = vring_enable_cb,
 274 };
 275
 276 struct virtqueue *vring_new_virtqueue(unsigned int num,
 277                                       unsigned int vring_align,
 278                                       struct virtio_device *vdev,
 279                                       void *pages,
 280                                       void (*notify)(struct virtqueue *),
 281                                       void (*callback)(struct virtqueue *))
 282 {
 283         struct vring_virtqueue *vq;
 284         unsigned int i;
 285
 286         /* We assume num is a power of 2. */
 287         if (num & (num - 1)) {
 288                 dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num);
 289                 return NULL;
 290         }
 291
 292         vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
 293         if (!vq)
 294                 return NULL;
 295
 296         vring_init(&vq->vring, num, pages, vring_align);
 297         vq->vq.callback = callback;
 298         vq->vq.vdev = vdev;
 299         vq->vq.vq_ops = &vring_vq_ops;
 300         vq->notify = notify;
 301         vq->broken = false;
 302         vq->last_used_idx = 0;
 303         vq->num_added = 0;
 304 #ifdef DEBUG
 305         vq->in_use = false;
 306 #endif
 307
 308         /* No callback?  Tell other side not to bother us. */
 309         if (!callback)
 310                 vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
 311
 312         /* Put everything in free lists. */
 313         vq->num_free = num;
 314         vq->free_head = 0;
 315         for (i = 0; i < num-1; i++)
 316                 vq->vring.desc[i].next = i+1;
 317
 318         return &vq->vq;
 319 }
 320 EXPORT_SYMBOL_GPL(vring_new_virtqueue);
 321
 322 void vring_del_virtqueue(struct virtqueue *vq)
 323 {
 324         kfree(to_vvq(vq));
 325 }
 326 EXPORT_SYMBOL_GPL(vring_del_virtqueue);
 327
 328 /* Manipulates transport-specific feature bits. */
 329 void vring_transport_features(struct virtio_device *vdev)
 330 {
 331         unsigned int i;
 332
 333         for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) {
 334                 switch (i) {
 335                 default:
 336                         /* We don't understand this bit. */
 337                         clear_bit(i, vdev->features);
 338                 }
 339         }
 340 }
 341 EXPORT_SYMBOL_GPL(vring_transport_features);
 342
 343 MODULE_LICENSE("GPL");