xref: /openbmc/linux/drivers/virtio/virtio_ring.c (revision 7b6d864b)
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 
56 struct 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 
101 static inline struct scatterlist *sg_next_chained(struct scatterlist *sg,
102 						  unsigned int *count)
103 {
104 	return sg_next(sg);
105 }
106 
107 static 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. */
116 static 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 
184 static 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 
272 add_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_sgs - expose buffers to other end
300  * @vq: the struct virtqueue we're talking about.
301  * @sgs: array of terminated scatterlists.
302  * @out_num: the number of scatterlists readable by other side
303  * @in_num: the number of scatterlists 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  */
312 int virtqueue_add_sgs(struct virtqueue *_vq,
313 		      struct scatterlist *sgs[],
314 		      unsigned int out_sgs,
315 		      unsigned int in_sgs,
316 		      void *data,
317 		      gfp_t gfp)
318 {
319 	unsigned int i, total_out, total_in;
320 
321 	/* Count them first. */
322 	for (i = total_out = total_in = 0; i < out_sgs; i++) {
323 		struct scatterlist *sg;
324 		for (sg = sgs[i]; sg; sg = sg_next(sg))
325 			total_out++;
326 	}
327 	for (; i < out_sgs + in_sgs; i++) {
328 		struct scatterlist *sg;
329 		for (sg = sgs[i]; sg; sg = sg_next(sg))
330 			total_in++;
331 	}
332 	return virtqueue_add(_vq, sgs, sg_next_chained,
333 			     total_out, total_in, out_sgs, in_sgs, data, gfp);
334 }
335 EXPORT_SYMBOL_GPL(virtqueue_add_sgs);
336 
337 /**
338  * virtqueue_add_outbuf - expose output buffers to other end
339  * @vq: the struct virtqueue we're talking about.
340  * @sgs: array of scatterlists (need not be terminated!)
341  * @num: the number of scatterlists readable by other side
342  * @data: the token identifying the buffer.
343  * @gfp: how to do memory allocations (if necessary).
344  *
345  * Caller must ensure we don't call this with other virtqueue operations
346  * at the same time (except where noted).
347  *
348  * Returns zero or a negative error (ie. ENOSPC, ENOMEM).
349  */
350 int virtqueue_add_outbuf(struct virtqueue *vq,
351 			 struct scatterlist sg[], unsigned int num,
352 			 void *data,
353 			 gfp_t gfp)
354 {
355 	return virtqueue_add(vq, &sg, sg_next_arr, num, 0, 1, 0, data, gfp);
356 }
357 EXPORT_SYMBOL_GPL(virtqueue_add_outbuf);
358 
359 /**
360  * virtqueue_add_inbuf - expose input buffers to other end
361  * @vq: the struct virtqueue we're talking about.
362  * @sgs: array of scatterlists (need not be terminated!)
363  * @num: the number of scatterlists writable by other side
364  * @data: the token identifying the buffer.
365  * @gfp: how to do memory allocations (if necessary).
366  *
367  * Caller must ensure we don't call this with other virtqueue operations
368  * at the same time (except where noted).
369  *
370  * Returns zero or a negative error (ie. ENOSPC, ENOMEM).
371  */
372 int virtqueue_add_inbuf(struct virtqueue *vq,
373 			struct scatterlist sg[], unsigned int num,
374 			void *data,
375 			gfp_t gfp)
376 {
377 	return virtqueue_add(vq, &sg, sg_next_arr, 0, num, 0, 1, data, gfp);
378 }
379 EXPORT_SYMBOL_GPL(virtqueue_add_inbuf);
380 
381 /**
382  * virtqueue_kick_prepare - first half of split virtqueue_kick call.
383  * @vq: the struct virtqueue
384  *
385  * Instead of virtqueue_kick(), you can do:
386  *	if (virtqueue_kick_prepare(vq))
387  *		virtqueue_notify(vq);
388  *
389  * This is sometimes useful because the virtqueue_kick_prepare() needs
390  * to be serialized, but the actual virtqueue_notify() call does not.
391  */
392 bool virtqueue_kick_prepare(struct virtqueue *_vq)
393 {
394 	struct vring_virtqueue *vq = to_vvq(_vq);
395 	u16 new, old;
396 	bool needs_kick;
397 
398 	START_USE(vq);
399 	/* We need to expose available array entries before checking avail
400 	 * event. */
401 	virtio_mb(vq->weak_barriers);
402 
403 	old = vq->vring.avail->idx - vq->num_added;
404 	new = vq->vring.avail->idx;
405 	vq->num_added = 0;
406 
407 #ifdef DEBUG
408 	if (vq->last_add_time_valid) {
409 		WARN_ON(ktime_to_ms(ktime_sub(ktime_get(),
410 					      vq->last_add_time)) > 100);
411 	}
412 	vq->last_add_time_valid = false;
413 #endif
414 
415 	if (vq->event) {
416 		needs_kick = vring_need_event(vring_avail_event(&vq->vring),
417 					      new, old);
418 	} else {
419 		needs_kick = !(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY);
420 	}
421 	END_USE(vq);
422 	return needs_kick;
423 }
424 EXPORT_SYMBOL_GPL(virtqueue_kick_prepare);
425 
426 /**
427  * virtqueue_notify - second half of split virtqueue_kick call.
428  * @vq: the struct virtqueue
429  *
430  * This does not need to be serialized.
431  */
432 void virtqueue_notify(struct virtqueue *_vq)
433 {
434 	struct vring_virtqueue *vq = to_vvq(_vq);
435 
436 	/* Prod other side to tell it about changes. */
437 	vq->notify(_vq);
438 }
439 EXPORT_SYMBOL_GPL(virtqueue_notify);
440 
441 /**
442  * virtqueue_kick - update after add_buf
443  * @vq: the struct virtqueue
444  *
445  * After one or more virtqueue_add_* calls, invoke this to kick
446  * the other side.
447  *
448  * Caller must ensure we don't call this with other virtqueue
449  * operations at the same time (except where noted).
450  */
451 void virtqueue_kick(struct virtqueue *vq)
452 {
453 	if (virtqueue_kick_prepare(vq))
454 		virtqueue_notify(vq);
455 }
456 EXPORT_SYMBOL_GPL(virtqueue_kick);
457 
458 static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
459 {
460 	unsigned int i;
461 
462 	/* Clear data ptr. */
463 	vq->data[head] = NULL;
464 
465 	/* Put back on free list: find end */
466 	i = head;
467 
468 	/* Free the indirect table */
469 	if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT)
470 		kfree(phys_to_virt(vq->vring.desc[i].addr));
471 
472 	while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
473 		i = vq->vring.desc[i].next;
474 		vq->vq.num_free++;
475 	}
476 
477 	vq->vring.desc[i].next = vq->free_head;
478 	vq->free_head = head;
479 	/* Plus final descriptor */
480 	vq->vq.num_free++;
481 }
482 
483 static inline bool more_used(const struct vring_virtqueue *vq)
484 {
485 	return vq->last_used_idx != vq->vring.used->idx;
486 }
487 
488 /**
489  * virtqueue_get_buf - get the next used buffer
490  * @vq: the struct virtqueue we're talking about.
491  * @len: the length written into the buffer
492  *
493  * If the driver wrote data into the buffer, @len will be set to the
494  * amount written.  This means you don't need to clear the buffer
495  * beforehand to ensure there's no data leakage in the case of short
496  * writes.
497  *
498  * Caller must ensure we don't call this with other virtqueue
499  * operations at the same time (except where noted).
500  *
501  * Returns NULL if there are no used buffers, or the "data" token
502  * handed to virtqueue_add_*().
503  */
504 void *virtqueue_get_buf(struct virtqueue *_vq, unsigned int *len)
505 {
506 	struct vring_virtqueue *vq = to_vvq(_vq);
507 	void *ret;
508 	unsigned int i;
509 	u16 last_used;
510 
511 	START_USE(vq);
512 
513 	if (unlikely(vq->broken)) {
514 		END_USE(vq);
515 		return NULL;
516 	}
517 
518 	if (!more_used(vq)) {
519 		pr_debug("No more buffers in queue\n");
520 		END_USE(vq);
521 		return NULL;
522 	}
523 
524 	/* Only get used array entries after they have been exposed by host. */
525 	virtio_rmb(vq->weak_barriers);
526 
527 	last_used = (vq->last_used_idx & (vq->vring.num - 1));
528 	i = vq->vring.used->ring[last_used].id;
529 	*len = vq->vring.used->ring[last_used].len;
530 
531 	if (unlikely(i >= vq->vring.num)) {
532 		BAD_RING(vq, "id %u out of range\n", i);
533 		return NULL;
534 	}
535 	if (unlikely(!vq->data[i])) {
536 		BAD_RING(vq, "id %u is not a head!\n", i);
537 		return NULL;
538 	}
539 
540 	/* detach_buf clears data, so grab it now. */
541 	ret = vq->data[i];
542 	detach_buf(vq, i);
543 	vq->last_used_idx++;
544 	/* If we expect an interrupt for the next entry, tell host
545 	 * by writing event index and flush out the write before
546 	 * the read in the next get_buf call. */
547 	if (!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) {
548 		vring_used_event(&vq->vring) = vq->last_used_idx;
549 		virtio_mb(vq->weak_barriers);
550 	}
551 
552 #ifdef DEBUG
553 	vq->last_add_time_valid = false;
554 #endif
555 
556 	END_USE(vq);
557 	return ret;
558 }
559 EXPORT_SYMBOL_GPL(virtqueue_get_buf);
560 
561 /**
562  * virtqueue_disable_cb - disable callbacks
563  * @vq: the struct virtqueue we're talking about.
564  *
565  * Note that this is not necessarily synchronous, hence unreliable and only
566  * useful as an optimization.
567  *
568  * Unlike other operations, this need not be serialized.
569  */
570 void virtqueue_disable_cb(struct virtqueue *_vq)
571 {
572 	struct vring_virtqueue *vq = to_vvq(_vq);
573 
574 	vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
575 }
576 EXPORT_SYMBOL_GPL(virtqueue_disable_cb);
577 
578 /**
579  * virtqueue_enable_cb - restart callbacks after disable_cb.
580  * @vq: the struct virtqueue we're talking about.
581  *
582  * This re-enables callbacks; it returns "false" if there are pending
583  * buffers in the queue, to detect a possible race between the driver
584  * checking for more work, and enabling callbacks.
585  *
586  * Caller must ensure we don't call this with other virtqueue
587  * operations at the same time (except where noted).
588  */
589 bool virtqueue_enable_cb(struct virtqueue *_vq)
590 {
591 	struct vring_virtqueue *vq = to_vvq(_vq);
592 
593 	START_USE(vq);
594 
595 	/* We optimistically turn back on interrupts, then check if there was
596 	 * more to do. */
597 	/* Depending on the VIRTIO_RING_F_EVENT_IDX feature, we need to
598 	 * either clear the flags bit or point the event index at the next
599 	 * entry. Always do both to keep code simple. */
600 	vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
601 	vring_used_event(&vq->vring) = vq->last_used_idx;
602 	virtio_mb(vq->weak_barriers);
603 	if (unlikely(more_used(vq))) {
604 		END_USE(vq);
605 		return false;
606 	}
607 
608 	END_USE(vq);
609 	return true;
610 }
611 EXPORT_SYMBOL_GPL(virtqueue_enable_cb);
612 
613 /**
614  * virtqueue_enable_cb_delayed - restart callbacks after disable_cb.
615  * @vq: the struct virtqueue we're talking about.
616  *
617  * This re-enables callbacks but hints to the other side to delay
618  * interrupts until most of the available buffers have been processed;
619  * it returns "false" if there are many pending buffers in the queue,
620  * to detect a possible race between the driver checking for more work,
621  * and enabling callbacks.
622  *
623  * Caller must ensure we don't call this with other virtqueue
624  * operations at the same time (except where noted).
625  */
626 bool virtqueue_enable_cb_delayed(struct virtqueue *_vq)
627 {
628 	struct vring_virtqueue *vq = to_vvq(_vq);
629 	u16 bufs;
630 
631 	START_USE(vq);
632 
633 	/* We optimistically turn back on interrupts, then check if there was
634 	 * more to do. */
635 	/* Depending on the VIRTIO_RING_F_USED_EVENT_IDX feature, we need to
636 	 * either clear the flags bit or point the event index at the next
637 	 * entry. Always do both to keep code simple. */
638 	vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
639 	/* TODO: tune this threshold */
640 	bufs = (u16)(vq->vring.avail->idx - vq->last_used_idx) * 3 / 4;
641 	vring_used_event(&vq->vring) = vq->last_used_idx + bufs;
642 	virtio_mb(vq->weak_barriers);
643 	if (unlikely((u16)(vq->vring.used->idx - vq->last_used_idx) > bufs)) {
644 		END_USE(vq);
645 		return false;
646 	}
647 
648 	END_USE(vq);
649 	return true;
650 }
651 EXPORT_SYMBOL_GPL(virtqueue_enable_cb_delayed);
652 
653 /**
654  * virtqueue_detach_unused_buf - detach first unused buffer
655  * @vq: the struct virtqueue we're talking about.
656  *
657  * Returns NULL or the "data" token handed to virtqueue_add_*().
658  * This is not valid on an active queue; it is useful only for device
659  * shutdown.
660  */
661 void *virtqueue_detach_unused_buf(struct virtqueue *_vq)
662 {
663 	struct vring_virtqueue *vq = to_vvq(_vq);
664 	unsigned int i;
665 	void *buf;
666 
667 	START_USE(vq);
668 
669 	for (i = 0; i < vq->vring.num; i++) {
670 		if (!vq->data[i])
671 			continue;
672 		/* detach_buf clears data, so grab it now. */
673 		buf = vq->data[i];
674 		detach_buf(vq, i);
675 		vq->vring.avail->idx--;
676 		END_USE(vq);
677 		return buf;
678 	}
679 	/* That should have freed everything. */
680 	BUG_ON(vq->vq.num_free != vq->vring.num);
681 
682 	END_USE(vq);
683 	return NULL;
684 }
685 EXPORT_SYMBOL_GPL(virtqueue_detach_unused_buf);
686 
687 irqreturn_t vring_interrupt(int irq, void *_vq)
688 {
689 	struct vring_virtqueue *vq = to_vvq(_vq);
690 
691 	if (!more_used(vq)) {
692 		pr_debug("virtqueue interrupt with no work for %p\n", vq);
693 		return IRQ_NONE;
694 	}
695 
696 	if (unlikely(vq->broken))
697 		return IRQ_HANDLED;
698 
699 	pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
700 	if (vq->vq.callback)
701 		vq->vq.callback(&vq->vq);
702 
703 	return IRQ_HANDLED;
704 }
705 EXPORT_SYMBOL_GPL(vring_interrupt);
706 
707 struct virtqueue *vring_new_virtqueue(unsigned int index,
708 				      unsigned int num,
709 				      unsigned int vring_align,
710 				      struct virtio_device *vdev,
711 				      bool weak_barriers,
712 				      void *pages,
713 				      void (*notify)(struct virtqueue *),
714 				      void (*callback)(struct virtqueue *),
715 				      const char *name)
716 {
717 	struct vring_virtqueue *vq;
718 	unsigned int i;
719 
720 	/* We assume num is a power of 2. */
721 	if (num & (num - 1)) {
722 		dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num);
723 		return NULL;
724 	}
725 
726 	vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
727 	if (!vq)
728 		return NULL;
729 
730 	vring_init(&vq->vring, num, pages, vring_align);
731 	vq->vq.callback = callback;
732 	vq->vq.vdev = vdev;
733 	vq->vq.name = name;
734 	vq->vq.num_free = num;
735 	vq->vq.index = index;
736 	vq->notify = notify;
737 	vq->weak_barriers = weak_barriers;
738 	vq->broken = false;
739 	vq->last_used_idx = 0;
740 	vq->num_added = 0;
741 	list_add_tail(&vq->vq.list, &vdev->vqs);
742 #ifdef DEBUG
743 	vq->in_use = false;
744 	vq->last_add_time_valid = false;
745 #endif
746 
747 	vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC);
748 	vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
749 
750 	/* No callback?  Tell other side not to bother us. */
751 	if (!callback)
752 		vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
753 
754 	/* Put everything in free lists. */
755 	vq->free_head = 0;
756 	for (i = 0; i < num-1; i++) {
757 		vq->vring.desc[i].next = i+1;
758 		vq->data[i] = NULL;
759 	}
760 	vq->data[i] = NULL;
761 
762 	return &vq->vq;
763 }
764 EXPORT_SYMBOL_GPL(vring_new_virtqueue);
765 
766 void vring_del_virtqueue(struct virtqueue *vq)
767 {
768 	list_del(&vq->list);
769 	kfree(to_vvq(vq));
770 }
771 EXPORT_SYMBOL_GPL(vring_del_virtqueue);
772 
773 /* Manipulates transport-specific feature bits. */
774 void vring_transport_features(struct virtio_device *vdev)
775 {
776 	unsigned int i;
777 
778 	for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) {
779 		switch (i) {
780 		case VIRTIO_RING_F_INDIRECT_DESC:
781 			break;
782 		case VIRTIO_RING_F_EVENT_IDX:
783 			break;
784 		default:
785 			/* We don't understand this bit. */
786 			clear_bit(i, vdev->features);
787 		}
788 	}
789 }
790 EXPORT_SYMBOL_GPL(vring_transport_features);
791 
792 /**
793  * virtqueue_get_vring_size - return the size of the virtqueue's vring
794  * @vq: the struct virtqueue containing the vring of interest.
795  *
796  * Returns the size of the vring.  This is mainly used for boasting to
797  * userspace.  Unlike other operations, this need not be serialized.
798  */
799 unsigned int virtqueue_get_vring_size(struct virtqueue *_vq)
800 {
801 
802 	struct vring_virtqueue *vq = to_vvq(_vq);
803 
804 	return vq->vring.num;
805 }
806 EXPORT_SYMBOL_GPL(virtqueue_get_vring_size);
807 
808 MODULE_LICENSE("GPL");
809