xref: /linux/drivers/virtio/virtio_ring.c (revision a33f32244d8550da8b4a26e277ce07d5c6d158b5)
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 
25 /* virtio guest is communicating with a virtual "device" that actually runs on
26  * a host processor.  Memory barriers are used to control SMP effects. */
27 #ifdef CONFIG_SMP
28 /* Where possible, use SMP barriers which are more lightweight than mandatory
29  * barriers, because mandatory barriers control MMIO effects on accesses
30  * through relaxed memory I/O windows (which virtio does not use). */
31 #define virtio_mb() smp_mb()
32 #define virtio_rmb() smp_rmb()
33 #define virtio_wmb() smp_wmb()
34 #else
35 /* We must force memory ordering even if guest is UP since host could be
36  * running on another CPU, but SMP barriers are defined to barrier() in that
37  * configuration. So fall back to mandatory barriers instead. */
38 #define virtio_mb() mb()
39 #define virtio_rmb() rmb()
40 #define virtio_wmb() wmb()
41 #endif
42 
43 #ifdef DEBUG
44 /* For development, we want to crash whenever the ring is screwed. */
45 #define BAD_RING(_vq, fmt, args...)				\
46 	do {							\
47 		dev_err(&(_vq)->vq.vdev->dev,			\
48 			"%s:"fmt, (_vq)->vq.name, ##args);	\
49 		BUG();						\
50 	} while (0)
51 /* Caller is supposed to guarantee no reentry. */
52 #define START_USE(_vq)						\
53 	do {							\
54 		if ((_vq)->in_use)				\
55 			panic("%s:in_use = %i\n",		\
56 			      (_vq)->vq.name, (_vq)->in_use);	\
57 		(_vq)->in_use = __LINE__;			\
58 	} while (0)
59 #define END_USE(_vq) \
60 	do { BUG_ON(!(_vq)->in_use); (_vq)->in_use = 0; } while(0)
61 #else
62 #define BAD_RING(_vq, fmt, args...)				\
63 	do {							\
64 		dev_err(&_vq->vq.vdev->dev,			\
65 			"%s:"fmt, (_vq)->vq.name, ##args);	\
66 		(_vq)->broken = true;				\
67 	} while (0)
68 #define START_USE(vq)
69 #define END_USE(vq)
70 #endif
71 
72 struct vring_virtqueue
73 {
74 	struct virtqueue vq;
75 
76 	/* Actual memory layout for this queue */
77 	struct vring vring;
78 
79 	/* Other side has made a mess, don't try any more. */
80 	bool broken;
81 
82 	/* Host supports indirect buffers */
83 	bool indirect;
84 
85 	/* Number of free buffers */
86 	unsigned int num_free;
87 	/* Head of free buffer list. */
88 	unsigned int free_head;
89 	/* Number we've added since last sync. */
90 	unsigned int num_added;
91 
92 	/* Last used index we've seen. */
93 	u16 last_used_idx;
94 
95 	/* How to notify other side. FIXME: commonalize hcalls! */
96 	void (*notify)(struct virtqueue *vq);
97 
98 #ifdef DEBUG
99 	/* They're supposed to lock for us. */
100 	unsigned int in_use;
101 #endif
102 
103 	/* Tokens for callbacks. */
104 	void *data[];
105 };
106 
107 #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq)
108 
109 /* Set up an indirect table of descriptors and add it to the queue. */
110 static int vring_add_indirect(struct vring_virtqueue *vq,
111 			      struct scatterlist sg[],
112 			      unsigned int out,
113 			      unsigned int in)
114 {
115 	struct vring_desc *desc;
116 	unsigned head;
117 	int i;
118 
119 	desc = kmalloc((out + in) * sizeof(struct vring_desc), GFP_ATOMIC);
120 	if (!desc)
121 		return vq->vring.num;
122 
123 	/* Transfer entries from the sg list into the indirect page */
124 	for (i = 0; i < out; i++) {
125 		desc[i].flags = VRING_DESC_F_NEXT;
126 		desc[i].addr = sg_phys(sg);
127 		desc[i].len = sg->length;
128 		desc[i].next = i+1;
129 		sg++;
130 	}
131 	for (; i < (out + in); i++) {
132 		desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
133 		desc[i].addr = sg_phys(sg);
134 		desc[i].len = sg->length;
135 		desc[i].next = i+1;
136 		sg++;
137 	}
138 
139 	/* Last one doesn't continue. */
140 	desc[i-1].flags &= ~VRING_DESC_F_NEXT;
141 	desc[i-1].next = 0;
142 
143 	/* We're about to use a buffer */
144 	vq->num_free--;
145 
146 	/* Use a single buffer which doesn't continue */
147 	head = vq->free_head;
148 	vq->vring.desc[head].flags = VRING_DESC_F_INDIRECT;
149 	vq->vring.desc[head].addr = virt_to_phys(desc);
150 	vq->vring.desc[head].len = i * sizeof(struct vring_desc);
151 
152 	/* Update free pointer */
153 	vq->free_head = vq->vring.desc[head].next;
154 
155 	return head;
156 }
157 
158 static int vring_add_buf(struct virtqueue *_vq,
159 			 struct scatterlist sg[],
160 			 unsigned int out,
161 			 unsigned int in,
162 			 void *data)
163 {
164 	struct vring_virtqueue *vq = to_vvq(_vq);
165 	unsigned int i, avail, head, uninitialized_var(prev);
166 
167 	START_USE(vq);
168 
169 	BUG_ON(data == NULL);
170 
171 	/* If the host supports indirect descriptor tables, and we have multiple
172 	 * buffers, then go indirect. FIXME: tune this threshold */
173 	if (vq->indirect && (out + in) > 1 && vq->num_free) {
174 		head = vring_add_indirect(vq, sg, out, in);
175 		if (head != vq->vring.num)
176 			goto add_head;
177 	}
178 
179 	BUG_ON(out + in > vq->vring.num);
180 	BUG_ON(out + in == 0);
181 
182 	if (vq->num_free < out + in) {
183 		pr_debug("Can't add buf len %i - avail = %i\n",
184 			 out + in, vq->num_free);
185 		/* FIXME: for historical reasons, we force a notify here if
186 		 * there are outgoing parts to the buffer.  Presumably the
187 		 * host should service the ring ASAP. */
188 		if (out)
189 			vq->notify(&vq->vq);
190 		END_USE(vq);
191 		return -ENOSPC;
192 	}
193 
194 	/* We're about to use some buffers from the free list. */
195 	vq->num_free -= out + in;
196 
197 	head = vq->free_head;
198 	for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) {
199 		vq->vring.desc[i].flags = VRING_DESC_F_NEXT;
200 		vq->vring.desc[i].addr = sg_phys(sg);
201 		vq->vring.desc[i].len = sg->length;
202 		prev = i;
203 		sg++;
204 	}
205 	for (; in; i = vq->vring.desc[i].next, in--) {
206 		vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
207 		vq->vring.desc[i].addr = sg_phys(sg);
208 		vq->vring.desc[i].len = sg->length;
209 		prev = i;
210 		sg++;
211 	}
212 	/* Last one doesn't continue. */
213 	vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT;
214 
215 	/* Update free pointer */
216 	vq->free_head = i;
217 
218 add_head:
219 	/* Set token. */
220 	vq->data[head] = data;
221 
222 	/* Put entry in available array (but don't update avail->idx until they
223 	 * do sync).  FIXME: avoid modulus here? */
224 	avail = (vq->vring.avail->idx + vq->num_added++) % vq->vring.num;
225 	vq->vring.avail->ring[avail] = head;
226 
227 	pr_debug("Added buffer head %i to %p\n", head, vq);
228 	END_USE(vq);
229 
230 	/* If we're indirect, we can fit many (assuming not OOM). */
231 	if (vq->indirect)
232 		return vq->num_free ? vq->vring.num : 0;
233 	return vq->num_free;
234 }
235 
236 static void vring_kick(struct virtqueue *_vq)
237 {
238 	struct vring_virtqueue *vq = to_vvq(_vq);
239 	START_USE(vq);
240 	/* Descriptors and available array need to be set before we expose the
241 	 * new available array entries. */
242 	virtio_wmb();
243 
244 	vq->vring.avail->idx += vq->num_added;
245 	vq->num_added = 0;
246 
247 	/* Need to update avail index before checking if we should notify */
248 	virtio_mb();
249 
250 	if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY))
251 		/* Prod other side to tell it about changes. */
252 		vq->notify(&vq->vq);
253 
254 	END_USE(vq);
255 }
256 
257 static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
258 {
259 	unsigned int i;
260 
261 	/* Clear data ptr. */
262 	vq->data[head] = NULL;
263 
264 	/* Put back on free list: find end */
265 	i = head;
266 
267 	/* Free the indirect table */
268 	if (vq->vring.desc[i].flags & VRING_DESC_F_INDIRECT)
269 		kfree(phys_to_virt(vq->vring.desc[i].addr));
270 
271 	while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
272 		i = vq->vring.desc[i].next;
273 		vq->num_free++;
274 	}
275 
276 	vq->vring.desc[i].next = vq->free_head;
277 	vq->free_head = head;
278 	/* Plus final descriptor */
279 	vq->num_free++;
280 }
281 
282 static inline bool more_used(const struct vring_virtqueue *vq)
283 {
284 	return vq->last_used_idx != vq->vring.used->idx;
285 }
286 
287 static void *vring_get_buf(struct virtqueue *_vq, unsigned int *len)
288 {
289 	struct vring_virtqueue *vq = to_vvq(_vq);
290 	void *ret;
291 	unsigned int i;
292 
293 	START_USE(vq);
294 
295 	if (unlikely(vq->broken)) {
296 		END_USE(vq);
297 		return NULL;
298 	}
299 
300 	if (!more_used(vq)) {
301 		pr_debug("No more buffers in queue\n");
302 		END_USE(vq);
303 		return NULL;
304 	}
305 
306 	/* Only get used array entries after they have been exposed by host. */
307 	virtio_rmb();
308 
309 	i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id;
310 	*len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len;
311 
312 	if (unlikely(i >= vq->vring.num)) {
313 		BAD_RING(vq, "id %u out of range\n", i);
314 		return NULL;
315 	}
316 	if (unlikely(!vq->data[i])) {
317 		BAD_RING(vq, "id %u is not a head!\n", i);
318 		return NULL;
319 	}
320 
321 	/* detach_buf clears data, so grab it now. */
322 	ret = vq->data[i];
323 	detach_buf(vq, i);
324 	vq->last_used_idx++;
325 	END_USE(vq);
326 	return ret;
327 }
328 
329 static void vring_disable_cb(struct virtqueue *_vq)
330 {
331 	struct vring_virtqueue *vq = to_vvq(_vq);
332 
333 	vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
334 }
335 
336 static bool vring_enable_cb(struct virtqueue *_vq)
337 {
338 	struct vring_virtqueue *vq = to_vvq(_vq);
339 
340 	START_USE(vq);
341 
342 	/* We optimistically turn back on interrupts, then check if there was
343 	 * more to do. */
344 	vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
345 	virtio_mb();
346 	if (unlikely(more_used(vq))) {
347 		END_USE(vq);
348 		return false;
349 	}
350 
351 	END_USE(vq);
352 	return true;
353 }
354 
355 static void *vring_detach_unused_buf(struct virtqueue *_vq)
356 {
357 	struct vring_virtqueue *vq = to_vvq(_vq);
358 	unsigned int i;
359 	void *buf;
360 
361 	START_USE(vq);
362 
363 	for (i = 0; i < vq->vring.num; i++) {
364 		if (!vq->data[i])
365 			continue;
366 		/* detach_buf clears data, so grab it now. */
367 		buf = vq->data[i];
368 		detach_buf(vq, i);
369 		END_USE(vq);
370 		return buf;
371 	}
372 	/* That should have freed everything. */
373 	BUG_ON(vq->num_free != vq->vring.num);
374 
375 	END_USE(vq);
376 	return NULL;
377 }
378 
379 irqreturn_t vring_interrupt(int irq, void *_vq)
380 {
381 	struct vring_virtqueue *vq = to_vvq(_vq);
382 
383 	if (!more_used(vq)) {
384 		pr_debug("virtqueue interrupt with no work for %p\n", vq);
385 		return IRQ_NONE;
386 	}
387 
388 	if (unlikely(vq->broken))
389 		return IRQ_HANDLED;
390 
391 	pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
392 	if (vq->vq.callback)
393 		vq->vq.callback(&vq->vq);
394 
395 	return IRQ_HANDLED;
396 }
397 EXPORT_SYMBOL_GPL(vring_interrupt);
398 
399 static struct virtqueue_ops vring_vq_ops = {
400 	.add_buf = vring_add_buf,
401 	.get_buf = vring_get_buf,
402 	.kick = vring_kick,
403 	.disable_cb = vring_disable_cb,
404 	.enable_cb = vring_enable_cb,
405 	.detach_unused_buf = vring_detach_unused_buf,
406 };
407 
408 struct virtqueue *vring_new_virtqueue(unsigned int num,
409 				      unsigned int vring_align,
410 				      struct virtio_device *vdev,
411 				      void *pages,
412 				      void (*notify)(struct virtqueue *),
413 				      void (*callback)(struct virtqueue *),
414 				      const char *name)
415 {
416 	struct vring_virtqueue *vq;
417 	unsigned int i;
418 
419 	/* We assume num is a power of 2. */
420 	if (num & (num - 1)) {
421 		dev_warn(&vdev->dev, "Bad virtqueue length %u\n", num);
422 		return NULL;
423 	}
424 
425 	vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
426 	if (!vq)
427 		return NULL;
428 
429 	vring_init(&vq->vring, num, pages, vring_align);
430 	vq->vq.callback = callback;
431 	vq->vq.vdev = vdev;
432 	vq->vq.vq_ops = &vring_vq_ops;
433 	vq->vq.name = name;
434 	vq->notify = notify;
435 	vq->broken = false;
436 	vq->last_used_idx = 0;
437 	vq->num_added = 0;
438 	list_add_tail(&vq->vq.list, &vdev->vqs);
439 #ifdef DEBUG
440 	vq->in_use = false;
441 #endif
442 
443 	vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC);
444 
445 	/* No callback?  Tell other side not to bother us. */
446 	if (!callback)
447 		vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
448 
449 	/* Put everything in free lists. */
450 	vq->num_free = num;
451 	vq->free_head = 0;
452 	for (i = 0; i < num-1; i++) {
453 		vq->vring.desc[i].next = i+1;
454 		vq->data[i] = NULL;
455 	}
456 	vq->data[i] = NULL;
457 
458 	return &vq->vq;
459 }
460 EXPORT_SYMBOL_GPL(vring_new_virtqueue);
461 
462 void vring_del_virtqueue(struct virtqueue *vq)
463 {
464 	list_del(&vq->list);
465 	kfree(to_vvq(vq));
466 }
467 EXPORT_SYMBOL_GPL(vring_del_virtqueue);
468 
469 /* Manipulates transport-specific feature bits. */
470 void vring_transport_features(struct virtio_device *vdev)
471 {
472 	unsigned int i;
473 
474 	for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++) {
475 		switch (i) {
476 		case VIRTIO_RING_F_INDIRECT_DESC:
477 			break;
478 		default:
479 			/* We don't understand this bit. */
480 			clear_bit(i, vdev->features);
481 		}
482 	}
483 }
484 EXPORT_SYMBOL_GPL(vring_transport_features);
485 
486 MODULE_LICENSE("GPL");
487