linux/drivers/virtio/virtio_ring.c
<<
>>
Prefs
   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#include <linux/module.h>
  25#include <linux/hrtimer.h>
  26
  27#ifdef DEBUG
  28/* For development, we want to crash whenever the ring is screwed. */
  29#define BAD_RING(_vq, fmt, args...)                             \
  30        do {                                                    \
  31                dev_err(&(_vq)->vq.vdev->dev,                   \
  32                        "%s:"fmt, (_vq)->vq.name, ##args);      \
  33                BUG();                                          \
  34        } while (0)
  35/* Caller is supposed to guarantee no reentry. */
  36#define START_USE(_vq)                                          \
  37        do {                                                    \
  38                if ((_vq)->in_use)                              \
  39                        panic("%s:in_use = %i\n",               \
  40                              (_vq)->vq.name, (_vq)->in_use);   \
  41                (_vq)->in_use = __LINE__;                       \
  42        } while (0)
  43#define END_USE(_vq) \
  44        do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; } while(0)
  45#else
  46#define BAD_RING(_vq, fmt, args...)                             \
  47        do {                                                    \
  48                dev_err(&_vq->vq.vdev->dev,                     \
  49                        "%s:"fmt, (_vq)->vq.name, ##args);      \
  50                (_vq)->broken = true;                           \
  51        } while (0)
  52#define START_USE(vq)
  53#define END_USE(vq)
  54#endif
  55
  56struct vring_virtqueue
  57{
  58        struct virtqueue vq;
  59
  60        /* Actual memory layout for this queue */
  61        struct vring vring;
  62
  63        /* Can we use weak barriers? */
  64        bool weak_barriers;
  65
  66        /* Other side has made a mess, don't try any more. */
  67        bool broken;
  68
  69        /* Host supports indirect buffers */
  70        bool indirect;
  71
  72        /* Host publishes avail event idx */
  73        bool event;
  74
  75        /* Head of free buffer list. */
  76        unsigned int free_head;
  77        /* Number we've added since last sync. */
  78        unsigned int num_added;
  79
  80        /* Last used index we've seen. */
  81        u16 last_used_idx;
  82
  83        /* How to notify other side. FIXME: commonalize hcalls! */
  84        void (*notify)(struct virtqueue *vq);
  85
  86#ifdef DEBUG
  87        /* They're supposed to lock for us. */
  88        unsigned int in_use;
  89
  90        /* Figure out if their kicks are too delayed. */
  91        bool last_add_time_valid;
  92        ktime_t last_add_time;
  93#endif
  94
  95        /* Tokens for callbacks. */
  96        void *data[];
  97};
  98
  99#define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq)
 100
 101static inline struct scatterlist *sg_next_chained(struct scatterlist *sg,
 102                                                  unsigned int *count)
 103{
 104        return sg_next(sg);
 105}
 106
 107static inline struct scatterlist *sg_next_arr(struct scatterlist *sg,
 108                                              unsigned int *count)
 109{
 110        if (--(*count) == 0)
 111                return NULL;
 112        return sg + 1;
 113}
 114
 115/* Set up an indirect table of descriptors and add it to the queue. */
 116static inline int vring_add_indirect(struct vring_virtqueue *vq,
 117                                     struct scatterlist *sgs[],
 118                                     struct scatterlist *(*next)
 119                                       (struct scatterlist *, unsigned int *),
 120                                     unsigned int total_sg,
 121                                     unsigned int total_out,
 122                                     unsigned int total_in,
 123                                     unsigned int out_sgs,
 124                                     unsigned int in_sgs,
 125                                     gfp_t gfp)
 126{
 127        struct vring_desc *desc;
 128        unsigned head;
 129        struct scatterlist *sg;
 130        int i, n;
 131
 132        /*
 133         * We require lowmem mappings for the descriptors because
 134         * otherwise virt_to_phys will give us bogus addresses in the
 135         * virtqueue.
 136         */
 137        gfp &= ~(__GFP_HIGHMEM | __GFP_HIGH);
 138
 139        desc = kmalloc(total_sg * sizeof(struct vring_desc), gfp);
 140        if (!desc)
 141                return -ENOMEM;
 142
 143        /* Transfer entries from the sg lists into the indirect page */
 144        i = 0;
 145        for (n = 0; n < out_sgs; n++) {
 146                for (sg = sgs[n]; sg; sg = next(sg, &total_out)) {
 147                        desc[i].flags = VRING_DESC_F_NEXT;
 148                        desc[i].addr = sg_phys(sg);
 149                        desc[i].len = sg->length;
 150                        desc[i].next = i+1;
 151                        i++;
 152                }
 153        }
 154        for (; n < (out_sgs + in_sgs); n++) {
 155                for (sg = sgs[n]; sg; sg = next(sg, &total_in)) {
 156                        desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
 157                        desc[i].addr = sg_phys(sg);
 158                        desc[i].len = sg->length;
 159                        desc[i].next = i+1;
 160                        i++;
 161                }
 162        }
 163        BUG_ON(i != total_sg);
 164
 165        /* Last one doesn't continue. */
 166        desc[i-1].flags &= ~VRING_DESC_F_NEXT;
 167        desc[i-1].next = 0;
 168
 169        /* We're about to use a buffer */
 170        vq->vq.num_free--;
 171
 172        /* Use a single buffer which doesn't continue */
 173        head = vq->free_head;
 174        vq->vring.desc[head].flags = VRING_DESC_F_INDIRECT;
 175        vq->vring.desc[head].addr = virt_to_phys(desc);
 176        vq->vring.desc[head].len = i * sizeof(struct vring_desc);
 177
 178        /* Update free pointer */
 179        vq->free_head = vq->vring.desc[head].next;
 180
 181        return head;
 182}
 183
 184static inline int virtqueue_add(struct virtqueue *_vq,
 185                                struct scatterlist *sgs[],
 186                                struct scatterlist *(*next)
 187                                  (struct scatterlist *, unsigned int *),
 188                                unsigned int total_out,
 189                                unsigned int total_in,
 190                                unsigned int out_sgs,
 191                                unsigned int in_sgs,
 192                                void *data,
 193                                gfp_t gfp)
 194{
 195        struct vring_virtqueue *vq = to_vvq(_vq);
 196        struct scatterlist *sg;
 197        unsigned int i, n, avail, uninitialized_var(prev), total_sg;
 198        int head;
 199
 200        START_USE(vq);
 201
 202        BUG_ON(data == NULL);
 203
 204#ifdef DEBUG
 205        {
 206                ktime_t now = ktime_get();
 207
 208                /* No kick or get, with .1 second between?  Warn. */
 209                if (vq->last_add_time_valid)
 210                        WARN_ON(ktime_to_ms(ktime_sub(now, vq->last_add_time))
 211                                            > 100);
 212                vq->last_add_time = now;
 213                vq->last_add_time_valid = true;
 214        }
 215#endif
 216
 217        total_sg = total_in + total_out;
 218
 219        /* If the host supports indirect descriptor tables, and we have multiple
 220         * buffers, then go indirect. FIXME: tune this threshold */
 221        if (vq->indirect && total_sg > 1 && vq->vq.num_free) {
 222                head = vring_add_indirect(vq, sgs, next, total_sg, total_out,
 223                                          total_in,
 224                                          out_sgs, in_sgs, gfp);
 225                if (likely(head >= 0))
 226                        goto add_head;
 227        }
 228
 229        BUG_ON(total_sg > vq->vring.num);
 230        BUG_ON(total_sg == 0);
 231
 232        if (vq->vq.num_free < total_sg) {
 233                pr_debug("Can't add buf len %i - avail = %i\n",
 234                         total_sg, vq->vq.num_free);
 235                /* FIXME: for historical reasons, we force a notify here if
 236                 * there are outgoing parts to the buffer.  Presumably the
 237                 * host should service the ring ASAP. */
 238                if (out_sgs)
 239                        vq->notify(&vq->vq);
 240                END_USE(vq);
 241                return -ENOSPC;
 242        }
 243
 244        /* We're about to use some buffers from the free list. */
 245        vq->vq.num_free -= total_sg;
 246
 247        head = i = vq->free_head;
 248        for (n = 0; n < out_sgs; n++) {
 249                for (sg = sgs[n]; sg; sg = next(sg, &total_out)) {
 250                        vq->vring.desc[i].flags = VRING_DESC_F_NEXT;
 251                        vq->vring.desc[i].addr = sg_phys(sg);
 252                        vq->vring.desc[i].len = sg->length;
 253                        prev = i;
 254                        i = vq->vring.desc[i].next;
 255                }
 256        }
 257        for (; n < (out_sgs + in_sgs); n++) {
 258                for (sg = sgs[n]; sg; sg = next(sg, &total_in)) {
 259                        vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
 260                        vq->vring.desc[i].addr = sg_phys(sg);
 261                        vq->vring.desc[i].len = sg->length;
 262                        prev = i;
 263                        i = vq->vring.desc[i].next;
 264                }
 265        }
 266        /* Last one doesn't continue. */
 267        vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT;
 268
 269        /* Update free pointer */
 270        vq->free_head = i;
 271
 272add_head:
 273        /* Set token. */
 274        vq->data[head] = data;
 275
 276        /* Put entry in available array (but don't update avail->idx until they
 277         * do sync). */
 278        avail = (vq->vring.avail->idx & (vq->vring.num-1));
 279        vq->vring.avail->ring[avail] = head;
 280
 281        /* Descriptors and available array need to be set before we expose the
 282         * new available array entries. */
 283        virtio_wmb(vq->weak_barriers);
 284        vq->vring.avail->idx++;
 285        vq->num_added++;
 286
 287        /* This is very unlikely, but theoretically possible.  Kick
 288         * just in case. */
 289        if (unlikely(vq->num_added == (1 << 16) - 1))
 290                virtqueue_kick(_vq);
 291
 292        pr_debug("Added buffer head %i to %p\n", head, vq);
 293        END_USE(vq);
 294
 295        return 0;
 296}
 297
 298/**
 299 * virtqueue_add_buf - expose buffer to other end
 300 * @vq: the struct virtqueue we're talking about.
 301 * @sg: the description of the buffer(s).
 302 * @out_num: the number of sg readable by other side
 303 * @in_num: the number of sg which are writable (after readable ones)
 304 * @data: the token identifying the buffer.
 305 * @gfp: how to do memory allocations (if necessary).
 306 *
 307 * Caller must ensure we don't call this with other virtqueue operations
 308 * at the same time (except where noted).
 309 *
 310 * Returns zero or a negative error (ie. ENOSPC, ENOMEM).
 311 */
 312int virtqueue_add_buf(struct virtqueue *_vq,
 313                      struct scatterlist sg[],
 314                      unsigned int out,
 315                      unsigned int in,
 316                      void *data,
 317                      gfp_t gfp)
 318{
 319        struct scatterlist *sgs[2];
 320
 321        sgs[0] = sg;
 322        sgs[1] = sg + out;
 323
 324        return virtqueue_add(_vq, sgs, sg_next_arr,
 325                             out, in, out ? 1 : 0, in ? 1 : 0, data, gfp);
 326}
 327EXPORT_SYMBOL_GPL(virtqueue_add_buf);
 328
 329/**
 330 * virtqueue_add_sgs - expose buffers to other end
 331 * @vq: the struct virtqueue we're talking about.
 332 * @sgs: array of terminated scatterlists.
 333 * @out_num: the number of scatterlists readable by other side
 334 * @in_num: the number of scatterlists which are writable (after readable ones)
 335 * @data: the token identifying the buffer.
 336 * @gfp: how to do memory allocations (if necessary).
 337 *
 338 * Caller must ensure we don't call this with other virtqueue operations
 339 * at the same time (except where noted).
 340 *
 341 * Returns zero or a negative error (ie. ENOSPC, ENOMEM).
 342 */
 343int virtqueue_add_sgs(struct virtqueue *_vq,
 344                      struct scatterlist *sgs[],
 345                      unsigned int out_sgs,
 346                      unsigned int in_sgs,
 347                      void *data,
 348                      gfp_t gfp)
 349{
 350        unsigned int i, total_out, total_in;
 351
 352        /* Count them first. */
 353        for (i = total_out = total_in = 0; i < out_sgs; i++) {
 354                struct scatterlist *sg;
 355                for (sg = sgs[i]; sg; sg = sg_next(sg))
 356                        total_out++;
 357        }
 358        for (; i < out_sgs + in_sgs; i++) {
 359                struct scatterlist *sg;
 360                for (sg = sgs[i]; sg; sg = sg_next(sg))
 361                        total_in++;
 362        }
 363        return virtqueue_add(_vq, sgs, sg_next_chained,
 364                             total_out, total_in, out_sgs, in_sgs, data, gfp);
 365}
 366EXPORT_SYMBOL_GPL(virtqueue_add_sgs);
 367
 368/**
 369 * virtqueue_add_outbuf - expose output buffers to other end
 370 * @vq: the struct virtqueue we're talking about.
 371 * @sgs: array of scatterlists (need not be terminated!)
 372 * @num: the number of scatterlists readable by other side
 373 * @data: the token identifying the buffer.
 374 * @gfp: how to do memory allocations (if necessary).
 375 *
 376 * Caller must ensure we don't call this with other virtqueue operations
 377 * at the same time (except where noted).
 378 *
 379 * Returns zero or a negative error (ie. ENOSPC, ENOMEM).
 380 */
 381int virtqueue_add_outbuf(struct virtqueue *vq,
 382                         struct scatterlist sg[], unsigned int num,
 383                         void *data,
 384                         gfp_t gfp)
 385{
 386        return virtqueue_add(vq, &sg, sg_next_arr, num, 0, 1, 0, data, gfp);
 387}
 388EXPORT_SYMBOL_GPL(virtqueue_add_outbuf);
 389
 390/**
 391 * virtqueue_add_inbuf - expose input buffers to other end
 392 * @vq: the struct virtqueue we're talking about.
 393 * @sgs: array of scatterlists (need not be terminated!)
 394 * @num: the number of scatterlists writable by other side
 395 * @data: the token identifying the buffer.
 396 * @gfp: how to do memory allocations (if necessary).
 397 *
 398 * Caller must ensure we don't call this with other virtqueue operations
 399 * at the same time (except where noted).
 400 *
 401 * Returns zero or a negative error (ie. ENOSPC, ENOMEM).
 402 */
 403int virtqueue_add_inbuf(struct virtqueue *vq,
 404                        struct scatterlist sg[], unsigned int num,
 405                        void *data,
 406                        gfp_t gfp)
 407{
 408        return virtqueue_add(vq, &sg, sg_next_arr, 0, num, 0, 1, data, gfp);
 409}
 410EXPORT_SYMBOL_GPL(virtqueue_add_inbuf);
 411
 412/**
 413 * virtqueue_kick_prepare - first half of split virtqueue_kick call.
 414 * @vq: the struct virtqueue
 415 *
 416 * Instead of virtqueue_kick(), you can do:
 417 *      if (virtqueue_kick_prepare(vq))
 418 *              virtqueue_notify(vq);
 419 *
 420 * This is sometimes useful because the virtqueue_kick_prepare() needs
 421 * to be serialized, but the actual virtqueue_notify() call does not.
 422 */
 423bool virtqueue_kick_prepare(struct virtqueue *_vq)
 424{
 425        struct vring_virtqueue *vq = to_vvq(_vq);
 426        u16 new, old;
 427        bool needs_kick;
 428
 429        START_USE(vq);
 430        /* We need to expose available array entries before checking avail
 431         * event. */
 432        virtio_mb(vq->weak_barriers);
 433
 434        old = vq->vring.avail->idx - vq->num_added;
 435        new = vq->vring.avail->idx;
 436        vq->num_added = 0;
 437
 438#ifdef DEBUG
 439        if (vq->last_add_time_valid) {
 440                WARN_ON(ktime_to_ms(ktime_sub(ktime_get(),
 441                                              vq->last_add_time)) > 100);
 442        }
 443        vq->last_add_time_valid = false;
 444#endif
 445
 446        if (vq->event) {
 447                needs_kick = vring_need_event(vring_avail_event(&vq->vring),
 448                                              new, old);
 449        } else {
 450                needs_kick = !(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY);
 451        }
 452        END_USE(vq);
 453        return needs_kick;
 454}
 455EXPORT_SYMBOL_GPL(virtqueue_kick_prepare);
 456
 457/**
 458 * virtqueue_notify - second half of split virtqueue_kick call.
 459 * @vq: the struct virtqueue
 460 *
 461 * This does not need to be serialized.
 462 */
 463void virtqueue_notify(struct virtqueue *_vq)
 464{
 465        struct vring_virtqueue *vq = to_vvq(_vq);
 466
 467        /* Prod other side to tell it about changes. */
 468        vq->notify(_vq);
 469}
 470EXPORT_SYMBOL_GPL(virtqueue_notify);
 471
 472/**
 473 * virtqueue_kick - update after add_buf
 474 * @vq: the struct virtqueue
 475 *
 476 * After one or more virtqueue_add_buf calls, invoke this to kick
 477 * the other side.
 478 *
 479 * Caller must ensure we don't call this with other virtqueue
 480 * operations at the same time (except where noted).
 481 */
 482void virtqueue_kick(struct virtqueue *vq)
 483{
 484        if (virtqueue_kick_prepare(vq))
 485                virtqueue_notify(vq);
 486}
 487EXPORT_SYMBOL_GPL(virtqueue_kick);
 488
 489static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
 490{
 491        unsigned int i;
 492
 493        /* Clear data ptr. */
 494        vq->data[head] = NULL;
 495
 496        /* Put back on free list: find end */
 497        i = head;
 498
 499        /* Free the indirect table */
 500        if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT)
 501                kfree(phys_to_virt(vq->vring.desc[i].addr));
 502
 503        while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
 504                i = vq->vring.desc[i].next;
 505                vq->vq.num_free++;
 506        }
 507
 508        vq->vring.desc[i].next = vq->free_head;
 509        vq->free_head = head;
 510        /* Plus final descriptor */
 511        vq->vq.num_free++;
 512}
 513
 514static inline bool more_used(const struct vring_virtqueue *vq)
 515{
 516        return vq->last_used_idx != vq->vring.used->idx;
 517}
 518
 519/**
 520 * virtqueue_get_buf - get the next used buffer
 521 * @vq: the struct virtqueue we're talking about.
 522 * @len: the length written into the buffer
 523 *
 524 * If the driver wrote data into the buffer, @len will be set to the
 525 * amount written.  This means you don't need to clear the buffer
 526 * beforehand to ensure there's no data leakage in the case of short
 527 * writes.
 528 *
 529 * Caller must ensure we don't call this with other virtqueue
 530 * operations at the same time (except where noted).
 531 *
 532 * Returns NULL if there are no used buffers, or the "data" token
 533 * handed to virtqueue_add_buf().
 534 */
 535void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len)
 536{
 537        struct vring_virtqueue *vq = to_vvq(_vq);
 538        void *ret;
 539        unsigned int i;
 540        u16 last_used;
 541
 542        START_USE(vq);
 543
 544        if (unlikely(vq->broken)) {
 545                END_USE(vq);
 546                return NULL;
 547        }
 548
 549        if (!more_used(vq)) {
 550                pr_debug("No more buffers in queue\n");
 551                END_USE(vq);
 552                return NULL;
 553        }
 554
 555        /* Only get used array entries after they have been exposed by host. */
 556        virtio_rmb(vq->weak_barriers);
 557
 558        last_used = (vq->last_used_idx & (vq->vring.num - 1));
 559        i = vq->vring.used->ring[last_used].id;
 560        *len = vq->vring.used->ring[last_used].len;
 561
 562        if (unlikely(i >= vq->vring.num)) {
 563                BAD_RING(vq, "id %u out of range\n", i);
 564                return NULL;
 565        }
 566        if (unlikely(!vq->data[i])) {
 567                BAD_RING(vq, "id %u is not a head!\n", i);
 568                return NULL;
 569        }
 570
 571        /* detach_buf clears data, so grab it now. */
 572        ret = vq->data[i];
 573        detach_buf(vq, i);
 574        vq->last_used_idx++;
 575        /* If we expect an interrupt for the next entry, tell host
 576         * by writing event index and flush out the write before
 577         * the read in the next get_buf call. */
 578        if (!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) {
 579                vring_used_event(&vq->vring) = vq->last_used_idx;
 580                virtio_mb(vq->weak_barriers);
 581        }
 582
 583#ifdef DEBUG
 584        vq->last_add_time_valid = false;
 585#endif
 586
 587        END_USE(vq);
 588        return ret;
 589}
 590EXPORT_SYMBOL_GPL(virtqueue_get_buf);
 591
 592/**
 593 * virtqueue_disable_cb - disable callbacks
 594 * @vq: the struct virtqueue we're talking about.
 595 *
 596 * Note that this is not necessarily synchronous, hence unreliable and only
 597 * useful as an optimization.
 598 *
 599 * Unlike other operations, this need not be serialized.
 600 */
 601void virtqueue_disable_cb(struct virtqueue *_vq)
 602{
 603        struct vring_virtqueue *vq = to_vvq(_vq);
 604
 605        vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
 606}
 607EXPORT_SYMBOL_GPL(virtqueue_disable_cb);
 608
 609/**
 610 * virtqueue_enable_cb - restart callbacks after disable_cb.
 611 * @vq: the struct virtqueue we're talking about.
 612 *
 613 * This re-enables callbacks; it returns "false" if there are pending
 614 * buffers in the queue, to detect a possible race between the driver
 615 * checking for more work, and enabling callbacks.
 616 *
 617 * Caller must ensure we don't call this with other virtqueue
 618 * operations at the same time (except where noted).
 619 */
 620bool virtqueue_enable_cb(struct virtqueue *_vq)
 621{
 622        struct vring_virtqueue *vq = to_vvq(_vq);
 623
 624        START_USE(vq);
 625
 626        /* We optimistically turn back on interrupts, then check if there was
 627         * more to do. */
 628        /* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to
 629         * either clear the flags bit or point the event index at the next
 630         * entry. Always do both to keep code simple. */
 631        vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
 632        vring_used_event(&vq->vring) = vq->last_used_idx;
 633        virtio_mb(vq->weak_barriers);
 634        if (unlikely(more_used(vq))) {
 635                END_USE(vq);
 636                return false;
 637        }
 638
 639        END_USE(vq);
 640        return true;
 641}
 642EXPORT_SYMBOL_GPL(virtqueue_enable_cb);
 643
 644/**
 645 * virtqueue_enable_cb_delayed - restart callbacks after disable_cb.
 646 * @vq: the struct virtqueue we're talking about.
 647 *
 648 * This re-enables callbacks but hints to the other side to delay
 649 * interrupts until most of the available buffers have been processed;
 650 * it returns "false" if there are many pending buffers in the queue,
 651 * to detect a possible race between the driver checking for more work,
 652 * and enabling callbacks.
 653 *
 654 * Caller must ensure we don't call this with other virtqueue
 655 * operations at the same time (except where noted).
 656 */
 657bool virtqueue_enable_cb_delayed(struct virtqueue *_vq)
 658{
 659        struct vring_virtqueue *vq = to_vvq(_vq);
 660        u16 bufs;
 661
 662        START_USE(vq);
 663
 664        /* We optimistically turn back on interrupts, then check if there was
 665         * more to do. */
 666        /* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to
 667         * either clear the flags bit or point the event index at the next
 668         * entry. Always do both to keep code simple. */
 669        vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
 670        /* TODO: tune this threshold */
 671        bufs = (u16)(vq->vring.avail->idx - vq->last_used_idx) * 3 / 4;
 672        vring_used_event(&vq->vring) = vq->last_used_idx + bufs;
 673        virtio_mb(vq->weak_barriers);
 674        if (unlikely((u16)(vq->vring.used->idx - vq->last_used_idx) > bufs)) {
 675                END_USE(vq);
 676                return false;
 677        }
 678
 679        END_USE(vq);
 680        return true;
 681}
 682EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed);
 683
 684/**
 685 * virtqueue_detach_unused_buf - detach first unused buffer
 686 * @vq: the struct virtqueue we're talking about.
 687 *
 688 * Returns NULL or the "data" token handed to virtqueue_add_buf().
 689 * This is not valid on an active queue; it is useful only for device
 690 * shutdown.
 691 */
 692void *virtqueue_detach_unused_buf(struct virtqueue *_vq)
 693{
 694        struct vring_virtqueue *vq = to_vvq(_vq);
 695        unsigned int i;
 696        void *buf;
 697
 698        START_USE(vq);
 699
 700        for (i = 0; i < vq->vring.num; i++) {
 701                if (!vq->data[i])
 702                        continue;
 703                /* detach_buf clears data, so grab it now. */
 704                buf = vq->data[i];
 705                detach_buf(vq, i);
 706                vq->vring.avail->idx--;
 707                END_USE(vq);
 708                return buf;
 709        }
 710        /* That should have freed everything. */
 711        BUG_ON(vq->vq.num_free != vq->vring.num);
 712
 713        END_USE(vq);
 714        return NULL;
 715}
 716EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf);
 717
 718irqreturn_t vring_interrupt(int irq, void *_vq)
 719{
 720        struct vring_virtqueue *vq = to_vvq(_vq);
 721
 722        if (!more_used(vq)) {
 723                pr_debug("virtqueue interrupt with no work for %p\n", vq);
 724                return IRQ_NONE;
 725        }
 726
 727        if (unlikely(vq->broken))
 728                return IRQ_HANDLED;
 729
 730        pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
 731        if (vq->vq.callback)
 732                vq->vq.callback(&vq->vq);
 733
 734        return IRQ_HANDLED;
 735}
 736EXPORT_SYMBOL_GPL(vring_interrupt);
 737
 738struct virtqueue *vring_new_virtqueue(unsigned int index,
 739                                      unsigned int num,
 740                                      unsigned int vring_align,
 741                                      struct virtio_device *vdev,
 742                                      bool weak_barriers,
 743                                      void *pages,
 744                                      void (*notify)(struct virtqueue *),
 745                                      void (*callback)(struct virtqueue *),
 746                                      const char *name)
 747{
 748        struct vring_virtqueue *vq;
 749        unsigned int i;
 750
 751        /* We assume num is a power of 2. */
 752        if (num & (num - 1)) {
 753                dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num);
 754                return NULL;
 755        }
 756
 757        vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
 758        if (!vq)
 759                return NULL;
 760
 761        vring_init(&vq->vring, num, pages, vring_align);
 762        vq->vq.callback = callback;
 763        vq->vq.vdev = vdev;
 764        vq->vq.name = name;
 765        vq->vq.num_free = num;
 766        vq->vq.index = index;
 767        vq->notify = notify;
 768        vq->weak_barriers = weak_barriers;
 769        vq->broken = false;
 770        vq->last_used_idx = 0;
 771        vq->num_added = 0;
 772        list_add_tail(&vq->vq.list, &vdev->vqs);
 773#ifdef DEBUG
 774        vq->in_use = false;
 775        vq->last_add_time_valid = false;
 776#endif
 777
 778        vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC);
 779        vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
 780
 781        /* No callback?  Tell other side not to bother us. */
 782        if (!callback)
 783                vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
 784
 785        /* Put everything in free lists. */
 786        vq->free_head = 0;
 787        for (i = 0; i < num-1; i++) {
 788                vq->vring.desc[i].next = i+1;
 789                vq->data[i] = NULL;
 790        }
 791        vq->data[i] = NULL;
 792
 793        return &vq->vq;
 794}
 795EXPORT_SYMBOL_GPL(vring_new_virtqueue);
 796
 797void vring_del_virtqueue(struct virtqueue *vq)
 798{
 799        list_del(&vq->list);
 800        kfree(to_vvq(vq));
 801}
 802EXPORT_SYMBOL_GPL(vring_del_virtqueue);
 803
 804/* Manipulates transport-specific feature bits. */
 805void vring_transport_features(struct virtio_device *vdev)
 806{
 807        unsigned int i;
 808
 809        for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) {
 810                switch (i) {
 811                case VIRTIO_RING_F_INDIRECT_DESC:
 812                        break;
 813                case VIRTIO_RING_F_EVENT_IDX:
 814                        break;
 815                default:
 816                        /* We don't understand this bit. */
 817                        clear_bit(i, vdev->features);
 818                }
 819        }
 820}
 821EXPORT_SYMBOL_GPL(vring_transport_features);
 822
 823/**
 824 * virtqueue_get_vring_size - return the size of the virtqueue's vring
 825 * @vq: the struct virtqueue containing the vring of interest.
 826 *
 827 * Returns the size of the vring.  This is mainly used for boasting to
 828 * userspace.  Unlike other operations, this need not be serialized.
 829 */
 830unsigned int virtqueue_get_vring_size(struct virtqueue *_vq)
 831{
 832
 833        struct vring_virtqueue *vq = to_vvq(_vq);
 834
 835        return vq->vring.num;
 836}
 837EXPORT_SYMBOL_GPL(virtqueue_get_vring_size);
 838
 839MODULE_LICENSE("GPL");
 840