Linux Audio

Check our new training course

Loading...
v3.1
  1/*
  2 * Virtio PCI driver
  3 *
  4 * This module allows virtio devices to be used over a virtual PCI device.
  5 * This can be used with QEMU based VMMs like KVM or Xen.
  6 *
  7 * Copyright IBM Corp. 2007
  8 *
  9 * Authors:
 10 *  Anthony Liguori  <aliguori@us.ibm.com>
 11 *
 12 * This work is licensed under the terms of the GNU GPL, version 2 or later.
 13 * See the COPYING file in the top-level directory.
 14 *
 15 */
 16
 17#include <linux/module.h>
 18#include <linux/list.h>
 19#include <linux/pci.h>
 20#include <linux/slab.h>
 21#include <linux/interrupt.h>
 22#include <linux/virtio.h>
 23#include <linux/virtio_config.h>
 24#include <linux/virtio_ring.h>
 25#include <linux/virtio_pci.h>
 26#include <linux/highmem.h>
 27#include <linux/spinlock.h>
 28
 29MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
 30MODULE_DESCRIPTION("virtio-pci");
 31MODULE_LICENSE("GPL");
 32MODULE_VERSION("1");
 33
 34/* Our device structure */
 35struct virtio_pci_device
 36{
 37	struct virtio_device vdev;
 38	struct pci_dev *pci_dev;
 39
 40	/* the IO mapping for the PCI config space */
 41	void __iomem *ioaddr;
 42
 43	/* a list of queues so we can dispatch IRQs */
 44	spinlock_t lock;
 45	struct list_head virtqueues;
 46
 47	/* MSI-X support */
 48	int msix_enabled;
 49	int intx_enabled;
 50	struct msix_entry *msix_entries;
 51	/* Name strings for interrupts. This size should be enough,
 52	 * and I'm too lazy to allocate each name separately. */
 53	char (*msix_names)[256];
 54	/* Number of available vectors */
 55	unsigned msix_vectors;
 56	/* Vectors allocated, excluding per-vq vectors if any */
 57	unsigned msix_used_vectors;
 
 
 
 
 58	/* Whether we have vector per vq */
 59	bool per_vq_vectors;
 60};
 61
 62/* Constants for MSI-X */
 63/* Use first vector for configuration changes, second and the rest for
 64 * virtqueues Thus, we need at least 2 vectors for MSI. */
 65enum {
 66	VP_MSIX_CONFIG_VECTOR = 0,
 67	VP_MSIX_VQ_VECTOR = 1,
 68};
 69
 70struct virtio_pci_vq_info
 71{
 72	/* the actual virtqueue */
 73	struct virtqueue *vq;
 74
 75	/* the number of entries in the queue */
 76	int num;
 77
 78	/* the index of the queue */
 79	int queue_index;
 80
 81	/* the virtual address of the ring queue */
 82	void *queue;
 83
 84	/* the list node for the virtqueues list */
 85	struct list_head node;
 86
 87	/* MSI-X vector (or none) */
 88	unsigned msix_vector;
 89};
 90
 91/* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
 92static struct pci_device_id virtio_pci_id_table[] = {
 93	{ 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 },
 94	{ 0 },
 95};
 96
 97MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
 98
 99/* Convert a generic virtio device to our structure */
100static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
101{
102	return container_of(vdev, struct virtio_pci_device, vdev);
103}
104
105/* virtio config->get_features() implementation */
106static u32 vp_get_features(struct virtio_device *vdev)
107{
108	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
109
110	/* When someone needs more than 32 feature bits, we'll need to
111	 * steal a bit to indicate that the rest are somewhere else. */
112	return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES);
113}
114
115/* virtio config->finalize_features() implementation */
116static void vp_finalize_features(struct virtio_device *vdev)
117{
118	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
119
120	/* Give virtio_ring a chance to accept features. */
121	vring_transport_features(vdev);
122
123	/* We only support 32 feature bits. */
124	BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1);
125	iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES);
126}
127
128/* virtio config->get() implementation */
129static void vp_get(struct virtio_device *vdev, unsigned offset,
130		   void *buf, unsigned len)
131{
132	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
133	void __iomem *ioaddr = vp_dev->ioaddr +
134				VIRTIO_PCI_CONFIG(vp_dev) + offset;
135	u8 *ptr = buf;
136	int i;
137
138	for (i = 0; i < len; i++)
139		ptr[i] = ioread8(ioaddr + i);
140}
141
142/* the config->set() implementation.  it's symmetric to the config->get()
143 * implementation */
144static void vp_set(struct virtio_device *vdev, unsigned offset,
145		   const void *buf, unsigned len)
146{
147	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
148	void __iomem *ioaddr = vp_dev->ioaddr +
149				VIRTIO_PCI_CONFIG(vp_dev) + offset;
150	const u8 *ptr = buf;
151	int i;
152
153	for (i = 0; i < len; i++)
154		iowrite8(ptr[i], ioaddr + i);
155}
156
157/* config->{get,set}_status() implementations */
158static u8 vp_get_status(struct virtio_device *vdev)
159{
160	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
161	return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
162}
163
164static void vp_set_status(struct virtio_device *vdev, u8 status)
165{
166	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
167	/* We should never be setting status to 0. */
168	BUG_ON(status == 0);
169	iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
170}
171
 
 
 
 
 
 
 
 
 
 
 
 
 
172static void vp_reset(struct virtio_device *vdev)
173{
174	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
175	/* 0 status means a reset. */
176	iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 
 
 
 
 
177}
178
179/* the notify function used when creating a virt queue */
180static void vp_notify(struct virtqueue *vq)
181{
182	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
183	struct virtio_pci_vq_info *info = vq->priv;
184
185	/* we write the queue's selector into the notification register to
186	 * signal the other end */
187	iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
188}
189
190/* Handle a configuration change: Tell driver if it wants to know. */
191static irqreturn_t vp_config_changed(int irq, void *opaque)
192{
193	struct virtio_pci_device *vp_dev = opaque;
194	struct virtio_driver *drv;
195	drv = container_of(vp_dev->vdev.dev.driver,
196			   struct virtio_driver, driver);
197
198	if (drv && drv->config_changed)
199		drv->config_changed(&vp_dev->vdev);
200	return IRQ_HANDLED;
201}
202
203/* Notify all virtqueues on an interrupt. */
204static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
205{
206	struct virtio_pci_device *vp_dev = opaque;
207	struct virtio_pci_vq_info *info;
208	irqreturn_t ret = IRQ_NONE;
209	unsigned long flags;
210
211	spin_lock_irqsave(&vp_dev->lock, flags);
212	list_for_each_entry(info, &vp_dev->virtqueues, node) {
213		if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
214			ret = IRQ_HANDLED;
215	}
216	spin_unlock_irqrestore(&vp_dev->lock, flags);
217
218	return ret;
219}
220
221/* A small wrapper to also acknowledge the interrupt when it's handled.
222 * I really need an EIO hook for the vring so I can ack the interrupt once we
223 * know that we'll be handling the IRQ but before we invoke the callback since
224 * the callback may notify the host which results in the host attempting to
225 * raise an interrupt that we would then mask once we acknowledged the
226 * interrupt. */
227static irqreturn_t vp_interrupt(int irq, void *opaque)
228{
229	struct virtio_pci_device *vp_dev = opaque;
230	u8 isr;
231
232	/* reading the ISR has the effect of also clearing it so it's very
233	 * important to save off the value. */
234	isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
235
236	/* It's definitely not us if the ISR was not high */
237	if (!isr)
238		return IRQ_NONE;
239
240	/* Configuration change?  Tell driver if it wants to know. */
241	if (isr & VIRTIO_PCI_ISR_CONFIG)
242		vp_config_changed(irq, opaque);
243
244	return vp_vring_interrupt(irq, opaque);
245}
246
247static void vp_free_vectors(struct virtio_device *vdev)
248{
249	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
250	int i;
251
252	if (vp_dev->intx_enabled) {
253		free_irq(vp_dev->pci_dev->irq, vp_dev);
254		vp_dev->intx_enabled = 0;
255	}
256
257	for (i = 0; i < vp_dev->msix_used_vectors; ++i)
258		free_irq(vp_dev->msix_entries[i].vector, vp_dev);
259
260	if (vp_dev->msix_enabled) {
261		/* Disable the vector used for configuration */
262		iowrite16(VIRTIO_MSI_NO_VECTOR,
263			  vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
264		/* Flush the write out to device */
265		ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
266
267		pci_disable_msix(vp_dev->pci_dev);
268		vp_dev->msix_enabled = 0;
269		vp_dev->msix_vectors = 0;
270	}
271
272	vp_dev->msix_used_vectors = 0;
273	kfree(vp_dev->msix_names);
274	vp_dev->msix_names = NULL;
275	kfree(vp_dev->msix_entries);
276	vp_dev->msix_entries = NULL;
277}
278
279static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
280				   bool per_vq_vectors)
281{
282	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
283	const char *name = dev_name(&vp_dev->vdev.dev);
284	unsigned i, v;
285	int err = -ENOMEM;
286
287	vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
288				       GFP_KERNEL);
289	if (!vp_dev->msix_entries)
290		goto error;
291	vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
292				     GFP_KERNEL);
293	if (!vp_dev->msix_names)
294		goto error;
295
296	for (i = 0; i < nvectors; ++i)
297		vp_dev->msix_entries[i].entry = i;
298
299	/* pci_enable_msix returns positive if we can't get this many. */
300	err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors);
301	if (err > 0)
302		err = -ENOSPC;
303	if (err)
304		goto error;
305	vp_dev->msix_vectors = nvectors;
306	vp_dev->msix_enabled = 1;
307
308	/* Set the vector used for configuration */
309	v = vp_dev->msix_used_vectors;
310	snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
311		 "%s-config", name);
312	err = request_irq(vp_dev->msix_entries[v].vector,
313			  vp_config_changed, 0, vp_dev->msix_names[v],
314			  vp_dev);
315	if (err)
316		goto error;
317	++vp_dev->msix_used_vectors;
318
319	iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
320	/* Verify we had enough resources to assign the vector */
321	v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
322	if (v == VIRTIO_MSI_NO_VECTOR) {
323		err = -EBUSY;
324		goto error;
325	}
326
327	if (!per_vq_vectors) {
328		/* Shared vector for all VQs */
329		v = vp_dev->msix_used_vectors;
330		snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
331			 "%s-virtqueues", name);
332		err = request_irq(vp_dev->msix_entries[v].vector,
333				  vp_vring_interrupt, 0, vp_dev->msix_names[v],
334				  vp_dev);
335		if (err)
336			goto error;
337		++vp_dev->msix_used_vectors;
338	}
339	return 0;
340error:
341	vp_free_vectors(vdev);
342	return err;
343}
344
345static int vp_request_intx(struct virtio_device *vdev)
346{
347	int err;
348	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
349
350	err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
351			  IRQF_SHARED, dev_name(&vdev->dev), vp_dev);
352	if (!err)
353		vp_dev->intx_enabled = 1;
354	return err;
355}
356
357static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index,
358				  void (*callback)(struct virtqueue *vq),
359				  const char *name,
360				  u16 msix_vec)
361{
362	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
363	struct virtio_pci_vq_info *info;
364	struct virtqueue *vq;
365	unsigned long flags, size;
366	u16 num;
367	int err;
368
369	/* Select the queue we're interested in */
370	iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
371
372	/* Check if queue is either not available or already active. */
373	num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
374	if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
375		return ERR_PTR(-ENOENT);
376
377	/* allocate and fill out our structure the represents an active
378	 * queue */
379	info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
380	if (!info)
381		return ERR_PTR(-ENOMEM);
382
383	info->queue_index = index;
384	info->num = num;
385	info->msix_vector = msix_vec;
386
387	size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
388	info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
389	if (info->queue == NULL) {
390		err = -ENOMEM;
391		goto out_info;
392	}
393
394	/* activate the queue */
395	iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
396		  vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
397
398	/* create the vring */
399	vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN,
400				 vdev, info->queue, vp_notify, callback, name);
401	if (!vq) {
402		err = -ENOMEM;
403		goto out_activate_queue;
404	}
405
406	vq->priv = info;
407	info->vq = vq;
408
409	if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
410		iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
411		msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
412		if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
413			err = -EBUSY;
414			goto out_assign;
415		}
416	}
417
418	spin_lock_irqsave(&vp_dev->lock, flags);
419	list_add(&info->node, &vp_dev->virtqueues);
420	spin_unlock_irqrestore(&vp_dev->lock, flags);
 
 
 
 
421
422	return vq;
423
424out_assign:
425	vring_del_virtqueue(vq);
426out_activate_queue:
427	iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
428	free_pages_exact(info->queue, size);
429out_info:
430	kfree(info);
431	return ERR_PTR(err);
432}
433
434static void vp_del_vq(struct virtqueue *vq)
435{
436	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
437	struct virtio_pci_vq_info *info = vq->priv;
438	unsigned long flags, size;
439
440	spin_lock_irqsave(&vp_dev->lock, flags);
441	list_del(&info->node);
442	spin_unlock_irqrestore(&vp_dev->lock, flags);
443
444	iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
445
446	if (vp_dev->msix_enabled) {
447		iowrite16(VIRTIO_MSI_NO_VECTOR,
448			  vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
449		/* Flush the write out to device */
450		ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
451	}
452
453	vring_del_virtqueue(vq);
454
455	/* Select and deactivate the queue */
456	iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
457
458	size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
459	free_pages_exact(info->queue, size);
460	kfree(info);
461}
462
463/* the config->del_vqs() implementation */
464static void vp_del_vqs(struct virtio_device *vdev)
465{
466	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
467	struct virtqueue *vq, *n;
468	struct virtio_pci_vq_info *info;
469
470	list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
471		info = vq->priv;
472		if (vp_dev->per_vq_vectors &&
473			info->msix_vector != VIRTIO_MSI_NO_VECTOR)
474			free_irq(vp_dev->msix_entries[info->msix_vector].vector,
475				 vq);
476		vp_del_vq(vq);
477	}
478	vp_dev->per_vq_vectors = false;
479
480	vp_free_vectors(vdev);
481}
482
483static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
484			      struct virtqueue *vqs[],
485			      vq_callback_t *callbacks[],
486			      const char *names[],
487			      bool use_msix,
488			      bool per_vq_vectors)
489{
490	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
491	u16 msix_vec;
492	int i, err, nvectors, allocated_vectors;
493
494	if (!use_msix) {
495		/* Old style: one normal interrupt for change and all vqs. */
496		err = vp_request_intx(vdev);
497		if (err)
498			goto error_request;
499	} else {
500		if (per_vq_vectors) {
501			/* Best option: one for change interrupt, one per vq. */
502			nvectors = 1;
503			for (i = 0; i < nvqs; ++i)
504				if (callbacks[i])
505					++nvectors;
506		} else {
507			/* Second best: one for change, shared for all vqs. */
508			nvectors = 2;
509		}
510
511		err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
512		if (err)
513			goto error_request;
514	}
515
516	vp_dev->per_vq_vectors = per_vq_vectors;
517	allocated_vectors = vp_dev->msix_used_vectors;
518	for (i = 0; i < nvqs; ++i) {
519		if (!callbacks[i] || !vp_dev->msix_enabled)
520			msix_vec = VIRTIO_MSI_NO_VECTOR;
521		else if (vp_dev->per_vq_vectors)
522			msix_vec = allocated_vectors++;
523		else
524			msix_vec = VP_MSIX_VQ_VECTOR;
525		vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
526		if (IS_ERR(vqs[i])) {
527			err = PTR_ERR(vqs[i]);
528			goto error_find;
529		}
530
531		if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
532			continue;
533
534		/* allocate per-vq irq if available and necessary */
535		snprintf(vp_dev->msix_names[msix_vec],
536			 sizeof *vp_dev->msix_names,
537			 "%s-%s",
538			 dev_name(&vp_dev->vdev.dev), names[i]);
539		err = request_irq(vp_dev->msix_entries[msix_vec].vector,
540				  vring_interrupt, 0,
541				  vp_dev->msix_names[msix_vec],
542				  vqs[i]);
543		if (err) {
544			vp_del_vq(vqs[i]);
545			goto error_find;
546		}
547	}
548	return 0;
549
550error_find:
551	vp_del_vqs(vdev);
552
553error_request:
554	return err;
555}
556
557/* the config->find_vqs() implementation */
558static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
559		       struct virtqueue *vqs[],
560		       vq_callback_t *callbacks[],
561		       const char *names[])
562{
563	int err;
564
565	/* Try MSI-X with one vector per queue. */
566	err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
567	if (!err)
568		return 0;
569	/* Fallback: MSI-X with one vector for config, one shared for queues. */
570	err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
571				 true, false);
572	if (!err)
573		return 0;
574	/* Finally fall back to regular interrupts. */
575	return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
576				  false, false);
577}
578
 
 
 
 
 
 
 
579static struct virtio_config_ops virtio_pci_config_ops = {
580	.get		= vp_get,
581	.set		= vp_set,
582	.get_status	= vp_get_status,
583	.set_status	= vp_set_status,
584	.reset		= vp_reset,
585	.find_vqs	= vp_find_vqs,
586	.del_vqs	= vp_del_vqs,
587	.get_features	= vp_get_features,
588	.finalize_features = vp_finalize_features,
 
589};
590
591static void virtio_pci_release_dev(struct device *_d)
592{
593	struct virtio_device *dev = container_of(_d, struct virtio_device,
594						 dev);
595	struct virtio_pci_device *vp_dev = to_vp_device(dev);
596
597	kfree(vp_dev);
598}
599
600/* the PCI probing function */
601static int __devinit virtio_pci_probe(struct pci_dev *pci_dev,
602				      const struct pci_device_id *id)
603{
604	struct virtio_pci_device *vp_dev;
605	int err;
606
607	/* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
608	if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
609		return -ENODEV;
610
611	if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
612		printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
613		       VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
614		return -ENODEV;
615	}
616
617	/* allocate our structure and fill it out */
618	vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
619	if (vp_dev == NULL)
620		return -ENOMEM;
621
622	vp_dev->vdev.dev.parent = &pci_dev->dev;
623	vp_dev->vdev.dev.release = virtio_pci_release_dev;
624	vp_dev->vdev.config = &virtio_pci_config_ops;
625	vp_dev->pci_dev = pci_dev;
626	INIT_LIST_HEAD(&vp_dev->virtqueues);
627	spin_lock_init(&vp_dev->lock);
628
629	/* Disable MSI/MSIX to bring device to a known good state. */
630	pci_msi_off(pci_dev);
631
632	/* enable the device */
633	err = pci_enable_device(pci_dev);
634	if (err)
635		goto out;
636
637	err = pci_request_regions(pci_dev, "virtio-pci");
638	if (err)
639		goto out_enable_device;
640
641	vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
642	if (vp_dev->ioaddr == NULL)
643		goto out_req_regions;
644
645	pci_set_drvdata(pci_dev, vp_dev);
646	pci_set_master(pci_dev);
647
648	/* we use the subsystem vendor/device id as the virtio vendor/device
649	 * id.  this allows us to use the same PCI vendor/device id for all
650	 * virtio devices and to identify the particular virtio driver by
651	 * the subsystem ids */
652	vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
653	vp_dev->vdev.id.device = pci_dev->subsystem_device;
654
655	/* finally register the virtio device */
656	err = register_virtio_device(&vp_dev->vdev);
657	if (err)
658		goto out_set_drvdata;
659
660	return 0;
661
662out_set_drvdata:
663	pci_set_drvdata(pci_dev, NULL);
664	pci_iounmap(pci_dev, vp_dev->ioaddr);
665out_req_regions:
666	pci_release_regions(pci_dev);
667out_enable_device:
668	pci_disable_device(pci_dev);
669out:
670	kfree(vp_dev);
671	return err;
672}
673
674static void __devexit virtio_pci_remove(struct pci_dev *pci_dev)
675{
676	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
677
678	unregister_virtio_device(&vp_dev->vdev);
679
680	vp_del_vqs(&vp_dev->vdev);
681	pci_set_drvdata(pci_dev, NULL);
682	pci_iounmap(pci_dev, vp_dev->ioaddr);
683	pci_release_regions(pci_dev);
684	pci_disable_device(pci_dev);
 
685}
686
687#ifdef CONFIG_PM
688static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state)
689{
690	pci_save_state(pci_dev);
691	pci_set_power_state(pci_dev, PCI_D3hot);
692	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
693}
694
695static int virtio_pci_resume(struct pci_dev *pci_dev)
696{
697	pci_restore_state(pci_dev);
698	pci_set_power_state(pci_dev, PCI_D0);
699	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
700}
 
 
 
 
701#endif
702
703static struct pci_driver virtio_pci_driver = {
704	.name		= "virtio-pci",
705	.id_table	= virtio_pci_id_table,
706	.probe		= virtio_pci_probe,
707	.remove		= __devexit_p(virtio_pci_remove),
708#ifdef CONFIG_PM
709	.suspend	= virtio_pci_suspend,
710	.resume		= virtio_pci_resume,
711#endif
712};
713
714static int __init virtio_pci_init(void)
715{
716	return pci_register_driver(&virtio_pci_driver);
717}
718
719module_init(virtio_pci_init);
720
721static void __exit virtio_pci_exit(void)
722{
723	pci_unregister_driver(&virtio_pci_driver);
724}
725
726module_exit(virtio_pci_exit);
v3.5.6
  1/*
  2 * Virtio PCI driver
  3 *
  4 * This module allows virtio devices to be used over a virtual PCI device.
  5 * This can be used with QEMU based VMMs like KVM or Xen.
  6 *
  7 * Copyright IBM Corp. 2007
  8 *
  9 * Authors:
 10 *  Anthony Liguori  <aliguori@us.ibm.com>
 11 *
 12 * This work is licensed under the terms of the GNU GPL, version 2 or later.
 13 * See the COPYING file in the top-level directory.
 14 *
 15 */
 16
 17#include <linux/module.h>
 18#include <linux/list.h>
 19#include <linux/pci.h>
 20#include <linux/slab.h>
 21#include <linux/interrupt.h>
 22#include <linux/virtio.h>
 23#include <linux/virtio_config.h>
 24#include <linux/virtio_ring.h>
 25#include <linux/virtio_pci.h>
 26#include <linux/highmem.h>
 27#include <linux/spinlock.h>
 28
 29MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
 30MODULE_DESCRIPTION("virtio-pci");
 31MODULE_LICENSE("GPL");
 32MODULE_VERSION("1");
 33
 34/* Our device structure */
 35struct virtio_pci_device
 36{
 37	struct virtio_device vdev;
 38	struct pci_dev *pci_dev;
 39
 40	/* the IO mapping for the PCI config space */
 41	void __iomem *ioaddr;
 42
 43	/* a list of queues so we can dispatch IRQs */
 44	spinlock_t lock;
 45	struct list_head virtqueues;
 46
 47	/* MSI-X support */
 48	int msix_enabled;
 49	int intx_enabled;
 50	struct msix_entry *msix_entries;
 51	/* Name strings for interrupts. This size should be enough,
 52	 * and I'm too lazy to allocate each name separately. */
 53	char (*msix_names)[256];
 54	/* Number of available vectors */
 55	unsigned msix_vectors;
 56	/* Vectors allocated, excluding per-vq vectors if any */
 57	unsigned msix_used_vectors;
 58
 59	/* Status saved during hibernate/restore */
 60	u8 saved_status;
 61
 62	/* Whether we have vector per vq */
 63	bool per_vq_vectors;
 64};
 65
 66/* Constants for MSI-X */
 67/* Use first vector for configuration changes, second and the rest for
 68 * virtqueues Thus, we need at least 2 vectors for MSI. */
 69enum {
 70	VP_MSIX_CONFIG_VECTOR = 0,
 71	VP_MSIX_VQ_VECTOR = 1,
 72};
 73
 74struct virtio_pci_vq_info
 75{
 76	/* the actual virtqueue */
 77	struct virtqueue *vq;
 78
 79	/* the number of entries in the queue */
 80	int num;
 81
 82	/* the index of the queue */
 83	int queue_index;
 84
 85	/* the virtual address of the ring queue */
 86	void *queue;
 87
 88	/* the list node for the virtqueues list */
 89	struct list_head node;
 90
 91	/* MSI-X vector (or none) */
 92	unsigned msix_vector;
 93};
 94
 95/* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
 96static struct pci_device_id virtio_pci_id_table[] = {
 97	{ 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 },
 98	{ 0 },
 99};
100
101MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
102
103/* Convert a generic virtio device to our structure */
104static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
105{
106	return container_of(vdev, struct virtio_pci_device, vdev);
107}
108
109/* virtio config->get_features() implementation */
110static u32 vp_get_features(struct virtio_device *vdev)
111{
112	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
113
114	/* When someone needs more than 32 feature bits, we'll need to
115	 * steal a bit to indicate that the rest are somewhere else. */
116	return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES);
117}
118
119/* virtio config->finalize_features() implementation */
120static void vp_finalize_features(struct virtio_device *vdev)
121{
122	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
123
124	/* Give virtio_ring a chance to accept features. */
125	vring_transport_features(vdev);
126
127	/* We only support 32 feature bits. */
128	BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1);
129	iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES);
130}
131
132/* virtio config->get() implementation */
133static void vp_get(struct virtio_device *vdev, unsigned offset,
134		   void *buf, unsigned len)
135{
136	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
137	void __iomem *ioaddr = vp_dev->ioaddr +
138				VIRTIO_PCI_CONFIG(vp_dev) + offset;
139	u8 *ptr = buf;
140	int i;
141
142	for (i = 0; i < len; i++)
143		ptr[i] = ioread8(ioaddr + i);
144}
145
146/* the config->set() implementation.  it's symmetric to the config->get()
147 * implementation */
148static void vp_set(struct virtio_device *vdev, unsigned offset,
149		   const void *buf, unsigned len)
150{
151	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
152	void __iomem *ioaddr = vp_dev->ioaddr +
153				VIRTIO_PCI_CONFIG(vp_dev) + offset;
154	const u8 *ptr = buf;
155	int i;
156
157	for (i = 0; i < len; i++)
158		iowrite8(ptr[i], ioaddr + i);
159}
160
161/* config->{get,set}_status() implementations */
162static u8 vp_get_status(struct virtio_device *vdev)
163{
164	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
165	return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
166}
167
168static void vp_set_status(struct virtio_device *vdev, u8 status)
169{
170	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
171	/* We should never be setting status to 0. */
172	BUG_ON(status == 0);
173	iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
174}
175
176/* wait for pending irq handlers */
177static void vp_synchronize_vectors(struct virtio_device *vdev)
178{
179	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
180	int i;
181
182	if (vp_dev->intx_enabled)
183		synchronize_irq(vp_dev->pci_dev->irq);
184
185	for (i = 0; i < vp_dev->msix_vectors; ++i)
186		synchronize_irq(vp_dev->msix_entries[i].vector);
187}
188
189static void vp_reset(struct virtio_device *vdev)
190{
191	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
192	/* 0 status means a reset. */
193	iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
194	/* Flush out the status write, and flush in device writes,
195	 * including MSi-X interrupts, if any. */
196	ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
197	/* Flush pending VQ/configuration callbacks. */
198	vp_synchronize_vectors(vdev);
199}
200
201/* the notify function used when creating a virt queue */
202static void vp_notify(struct virtqueue *vq)
203{
204	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
205	struct virtio_pci_vq_info *info = vq->priv;
206
207	/* we write the queue's selector into the notification register to
208	 * signal the other end */
209	iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
210}
211
212/* Handle a configuration change: Tell driver if it wants to know. */
213static irqreturn_t vp_config_changed(int irq, void *opaque)
214{
215	struct virtio_pci_device *vp_dev = opaque;
216	struct virtio_driver *drv;
217	drv = container_of(vp_dev->vdev.dev.driver,
218			   struct virtio_driver, driver);
219
220	if (drv && drv->config_changed)
221		drv->config_changed(&vp_dev->vdev);
222	return IRQ_HANDLED;
223}
224
225/* Notify all virtqueues on an interrupt. */
226static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
227{
228	struct virtio_pci_device *vp_dev = opaque;
229	struct virtio_pci_vq_info *info;
230	irqreturn_t ret = IRQ_NONE;
231	unsigned long flags;
232
233	spin_lock_irqsave(&vp_dev->lock, flags);
234	list_for_each_entry(info, &vp_dev->virtqueues, node) {
235		if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
236			ret = IRQ_HANDLED;
237	}
238	spin_unlock_irqrestore(&vp_dev->lock, flags);
239
240	return ret;
241}
242
243/* A small wrapper to also acknowledge the interrupt when it's handled.
244 * I really need an EIO hook for the vring so I can ack the interrupt once we
245 * know that we'll be handling the IRQ but before we invoke the callback since
246 * the callback may notify the host which results in the host attempting to
247 * raise an interrupt that we would then mask once we acknowledged the
248 * interrupt. */
249static irqreturn_t vp_interrupt(int irq, void *opaque)
250{
251	struct virtio_pci_device *vp_dev = opaque;
252	u8 isr;
253
254	/* reading the ISR has the effect of also clearing it so it's very
255	 * important to save off the value. */
256	isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
257
258	/* It's definitely not us if the ISR was not high */
259	if (!isr)
260		return IRQ_NONE;
261
262	/* Configuration change?  Tell driver if it wants to know. */
263	if (isr & VIRTIO_PCI_ISR_CONFIG)
264		vp_config_changed(irq, opaque);
265
266	return vp_vring_interrupt(irq, opaque);
267}
268
269static void vp_free_vectors(struct virtio_device *vdev)
270{
271	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
272	int i;
273
274	if (vp_dev->intx_enabled) {
275		free_irq(vp_dev->pci_dev->irq, vp_dev);
276		vp_dev->intx_enabled = 0;
277	}
278
279	for (i = 0; i < vp_dev->msix_used_vectors; ++i)
280		free_irq(vp_dev->msix_entries[i].vector, vp_dev);
281
282	if (vp_dev->msix_enabled) {
283		/* Disable the vector used for configuration */
284		iowrite16(VIRTIO_MSI_NO_VECTOR,
285			  vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
286		/* Flush the write out to device */
287		ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
288
289		pci_disable_msix(vp_dev->pci_dev);
290		vp_dev->msix_enabled = 0;
291		vp_dev->msix_vectors = 0;
292	}
293
294	vp_dev->msix_used_vectors = 0;
295	kfree(vp_dev->msix_names);
296	vp_dev->msix_names = NULL;
297	kfree(vp_dev->msix_entries);
298	vp_dev->msix_entries = NULL;
299}
300
301static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
302				   bool per_vq_vectors)
303{
304	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
305	const char *name = dev_name(&vp_dev->vdev.dev);
306	unsigned i, v;
307	int err = -ENOMEM;
308
309	vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
310				       GFP_KERNEL);
311	if (!vp_dev->msix_entries)
312		goto error;
313	vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
314				     GFP_KERNEL);
315	if (!vp_dev->msix_names)
316		goto error;
317
318	for (i = 0; i < nvectors; ++i)
319		vp_dev->msix_entries[i].entry = i;
320
321	/* pci_enable_msix returns positive if we can't get this many. */
322	err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors);
323	if (err > 0)
324		err = -ENOSPC;
325	if (err)
326		goto error;
327	vp_dev->msix_vectors = nvectors;
328	vp_dev->msix_enabled = 1;
329
330	/* Set the vector used for configuration */
331	v = vp_dev->msix_used_vectors;
332	snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
333		 "%s-config", name);
334	err = request_irq(vp_dev->msix_entries[v].vector,
335			  vp_config_changed, 0, vp_dev->msix_names[v],
336			  vp_dev);
337	if (err)
338		goto error;
339	++vp_dev->msix_used_vectors;
340
341	iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
342	/* Verify we had enough resources to assign the vector */
343	v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
344	if (v == VIRTIO_MSI_NO_VECTOR) {
345		err = -EBUSY;
346		goto error;
347	}
348
349	if (!per_vq_vectors) {
350		/* Shared vector for all VQs */
351		v = vp_dev->msix_used_vectors;
352		snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
353			 "%s-virtqueues", name);
354		err = request_irq(vp_dev->msix_entries[v].vector,
355				  vp_vring_interrupt, 0, vp_dev->msix_names[v],
356				  vp_dev);
357		if (err)
358			goto error;
359		++vp_dev->msix_used_vectors;
360	}
361	return 0;
362error:
363	vp_free_vectors(vdev);
364	return err;
365}
366
367static int vp_request_intx(struct virtio_device *vdev)
368{
369	int err;
370	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
371
372	err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
373			  IRQF_SHARED, dev_name(&vdev->dev), vp_dev);
374	if (!err)
375		vp_dev->intx_enabled = 1;
376	return err;
377}
378
379static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index,
380				  void (*callback)(struct virtqueue *vq),
381				  const char *name,
382				  u16 msix_vec)
383{
384	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
385	struct virtio_pci_vq_info *info;
386	struct virtqueue *vq;
387	unsigned long flags, size;
388	u16 num;
389	int err;
390
391	/* Select the queue we're interested in */
392	iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
393
394	/* Check if queue is either not available or already active. */
395	num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
396	if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
397		return ERR_PTR(-ENOENT);
398
399	/* allocate and fill out our structure the represents an active
400	 * queue */
401	info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
402	if (!info)
403		return ERR_PTR(-ENOMEM);
404
405	info->queue_index = index;
406	info->num = num;
407	info->msix_vector = msix_vec;
408
409	size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
410	info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
411	if (info->queue == NULL) {
412		err = -ENOMEM;
413		goto out_info;
414	}
415
416	/* activate the queue */
417	iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
418		  vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
419
420	/* create the vring */
421	vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN, vdev,
422				 true, info->queue, vp_notify, callback, name);
423	if (!vq) {
424		err = -ENOMEM;
425		goto out_activate_queue;
426	}
427
428	vq->priv = info;
429	info->vq = vq;
430
431	if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
432		iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
433		msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
434		if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
435			err = -EBUSY;
436			goto out_assign;
437		}
438	}
439
440	if (callback) {
441		spin_lock_irqsave(&vp_dev->lock, flags);
442		list_add(&info->node, &vp_dev->virtqueues);
443		spin_unlock_irqrestore(&vp_dev->lock, flags);
444	} else {
445		INIT_LIST_HEAD(&info->node);
446	}
447
448	return vq;
449
450out_assign:
451	vring_del_virtqueue(vq);
452out_activate_queue:
453	iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
454	free_pages_exact(info->queue, size);
455out_info:
456	kfree(info);
457	return ERR_PTR(err);
458}
459
460static void vp_del_vq(struct virtqueue *vq)
461{
462	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
463	struct virtio_pci_vq_info *info = vq->priv;
464	unsigned long flags, size;
465
466	spin_lock_irqsave(&vp_dev->lock, flags);
467	list_del(&info->node);
468	spin_unlock_irqrestore(&vp_dev->lock, flags);
469
470	iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
471
472	if (vp_dev->msix_enabled) {
473		iowrite16(VIRTIO_MSI_NO_VECTOR,
474			  vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
475		/* Flush the write out to device */
476		ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
477	}
478
479	vring_del_virtqueue(vq);
480
481	/* Select and deactivate the queue */
482	iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
483
484	size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
485	free_pages_exact(info->queue, size);
486	kfree(info);
487}
488
489/* the config->del_vqs() implementation */
490static void vp_del_vqs(struct virtio_device *vdev)
491{
492	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
493	struct virtqueue *vq, *n;
494	struct virtio_pci_vq_info *info;
495
496	list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
497		info = vq->priv;
498		if (vp_dev->per_vq_vectors &&
499			info->msix_vector != VIRTIO_MSI_NO_VECTOR)
500			free_irq(vp_dev->msix_entries[info->msix_vector].vector,
501				 vq);
502		vp_del_vq(vq);
503	}
504	vp_dev->per_vq_vectors = false;
505
506	vp_free_vectors(vdev);
507}
508
509static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
510			      struct virtqueue *vqs[],
511			      vq_callback_t *callbacks[],
512			      const char *names[],
513			      bool use_msix,
514			      bool per_vq_vectors)
515{
516	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
517	u16 msix_vec;
518	int i, err, nvectors, allocated_vectors;
519
520	if (!use_msix) {
521		/* Old style: one normal interrupt for change and all vqs. */
522		err = vp_request_intx(vdev);
523		if (err)
524			goto error_request;
525	} else {
526		if (per_vq_vectors) {
527			/* Best option: one for change interrupt, one per vq. */
528			nvectors = 1;
529			for (i = 0; i < nvqs; ++i)
530				if (callbacks[i])
531					++nvectors;
532		} else {
533			/* Second best: one for change, shared for all vqs. */
534			nvectors = 2;
535		}
536
537		err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
538		if (err)
539			goto error_request;
540	}
541
542	vp_dev->per_vq_vectors = per_vq_vectors;
543	allocated_vectors = vp_dev->msix_used_vectors;
544	for (i = 0; i < nvqs; ++i) {
545		if (!callbacks[i] || !vp_dev->msix_enabled)
546			msix_vec = VIRTIO_MSI_NO_VECTOR;
547		else if (vp_dev->per_vq_vectors)
548			msix_vec = allocated_vectors++;
549		else
550			msix_vec = VP_MSIX_VQ_VECTOR;
551		vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
552		if (IS_ERR(vqs[i])) {
553			err = PTR_ERR(vqs[i]);
554			goto error_find;
555		}
556
557		if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
558			continue;
559
560		/* allocate per-vq irq if available and necessary */
561		snprintf(vp_dev->msix_names[msix_vec],
562			 sizeof *vp_dev->msix_names,
563			 "%s-%s",
564			 dev_name(&vp_dev->vdev.dev), names[i]);
565		err = request_irq(vp_dev->msix_entries[msix_vec].vector,
566				  vring_interrupt, 0,
567				  vp_dev->msix_names[msix_vec],
568				  vqs[i]);
569		if (err) {
570			vp_del_vq(vqs[i]);
571			goto error_find;
572		}
573	}
574	return 0;
575
576error_find:
577	vp_del_vqs(vdev);
578
579error_request:
580	return err;
581}
582
583/* the config->find_vqs() implementation */
584static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
585		       struct virtqueue *vqs[],
586		       vq_callback_t *callbacks[],
587		       const char *names[])
588{
589	int err;
590
591	/* Try MSI-X with one vector per queue. */
592	err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
593	if (!err)
594		return 0;
595	/* Fallback: MSI-X with one vector for config, one shared for queues. */
596	err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
597				 true, false);
598	if (!err)
599		return 0;
600	/* Finally fall back to regular interrupts. */
601	return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
602				  false, false);
603}
604
605static const char *vp_bus_name(struct virtio_device *vdev)
606{
607	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
608
609	return pci_name(vp_dev->pci_dev);
610}
611
612static struct virtio_config_ops virtio_pci_config_ops = {
613	.get		= vp_get,
614	.set		= vp_set,
615	.get_status	= vp_get_status,
616	.set_status	= vp_set_status,
617	.reset		= vp_reset,
618	.find_vqs	= vp_find_vqs,
619	.del_vqs	= vp_del_vqs,
620	.get_features	= vp_get_features,
621	.finalize_features = vp_finalize_features,
622	.bus_name	= vp_bus_name,
623};
624
625static void virtio_pci_release_dev(struct device *_d)
626{
627	/*
628	 * No need for a release method as we allocate/free
629	 * all devices together with the pci devices.
630	 * Provide an empty one to avoid getting a warning from core.
631	 */
632}
633
634/* the PCI probing function */
635static int __devinit virtio_pci_probe(struct pci_dev *pci_dev,
636				      const struct pci_device_id *id)
637{
638	struct virtio_pci_device *vp_dev;
639	int err;
640
641	/* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
642	if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
643		return -ENODEV;
644
645	if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
646		printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
647		       VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
648		return -ENODEV;
649	}
650
651	/* allocate our structure and fill it out */
652	vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
653	if (vp_dev == NULL)
654		return -ENOMEM;
655
656	vp_dev->vdev.dev.parent = &pci_dev->dev;
657	vp_dev->vdev.dev.release = virtio_pci_release_dev;
658	vp_dev->vdev.config = &virtio_pci_config_ops;
659	vp_dev->pci_dev = pci_dev;
660	INIT_LIST_HEAD(&vp_dev->virtqueues);
661	spin_lock_init(&vp_dev->lock);
662
663	/* Disable MSI/MSIX to bring device to a known good state. */
664	pci_msi_off(pci_dev);
665
666	/* enable the device */
667	err = pci_enable_device(pci_dev);
668	if (err)
669		goto out;
670
671	err = pci_request_regions(pci_dev, "virtio-pci");
672	if (err)
673		goto out_enable_device;
674
675	vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
676	if (vp_dev->ioaddr == NULL)
677		goto out_req_regions;
678
679	pci_set_drvdata(pci_dev, vp_dev);
680	pci_set_master(pci_dev);
681
682	/* we use the subsystem vendor/device id as the virtio vendor/device
683	 * id.  this allows us to use the same PCI vendor/device id for all
684	 * virtio devices and to identify the particular virtio driver by
685	 * the subsystem ids */
686	vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
687	vp_dev->vdev.id.device = pci_dev->subsystem_device;
688
689	/* finally register the virtio device */
690	err = register_virtio_device(&vp_dev->vdev);
691	if (err)
692		goto out_set_drvdata;
693
694	return 0;
695
696out_set_drvdata:
697	pci_set_drvdata(pci_dev, NULL);
698	pci_iounmap(pci_dev, vp_dev->ioaddr);
699out_req_regions:
700	pci_release_regions(pci_dev);
701out_enable_device:
702	pci_disable_device(pci_dev);
703out:
704	kfree(vp_dev);
705	return err;
706}
707
708static void __devexit virtio_pci_remove(struct pci_dev *pci_dev)
709{
710	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
711
712	unregister_virtio_device(&vp_dev->vdev);
713
714	vp_del_vqs(&vp_dev->vdev);
715	pci_set_drvdata(pci_dev, NULL);
716	pci_iounmap(pci_dev, vp_dev->ioaddr);
717	pci_release_regions(pci_dev);
718	pci_disable_device(pci_dev);
719	kfree(vp_dev);
720}
721
722#ifdef CONFIG_PM
723static int virtio_pci_freeze(struct device *dev)
724{
725	struct pci_dev *pci_dev = to_pci_dev(dev);
726	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
727	struct virtio_driver *drv;
728	int ret;
729
730	drv = container_of(vp_dev->vdev.dev.driver,
731			   struct virtio_driver, driver);
732
733	ret = 0;
734	vp_dev->saved_status = vp_get_status(&vp_dev->vdev);
735	if (drv && drv->freeze)
736		ret = drv->freeze(&vp_dev->vdev);
737
738	if (!ret)
739		pci_disable_device(pci_dev);
740	return ret;
741}
742
743static int virtio_pci_restore(struct device *dev)
744{
745	struct pci_dev *pci_dev = to_pci_dev(dev);
746	struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
747	struct virtio_driver *drv;
748	int ret;
749
750	drv = container_of(vp_dev->vdev.dev.driver,
751			   struct virtio_driver, driver);
752
753	ret = pci_enable_device(pci_dev);
754	if (ret)
755		return ret;
756
757	pci_set_master(pci_dev);
758	vp_finalize_features(&vp_dev->vdev);
759
760	if (drv && drv->restore)
761		ret = drv->restore(&vp_dev->vdev);
762
763	/* Finally, tell the device we're all set */
764	if (!ret)
765		vp_set_status(&vp_dev->vdev, vp_dev->saved_status);
766
767	return ret;
768}
769
770static const struct dev_pm_ops virtio_pci_pm_ops = {
771	SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore)
772};
773#endif
774
775static struct pci_driver virtio_pci_driver = {
776	.name		= "virtio-pci",
777	.id_table	= virtio_pci_id_table,
778	.probe		= virtio_pci_probe,
779	.remove		= __devexit_p(virtio_pci_remove),
780#ifdef CONFIG_PM
781	.driver.pm	= &virtio_pci_pm_ops,
 
782#endif
783};
784
785static int __init virtio_pci_init(void)
786{
787	return pci_register_driver(&virtio_pci_driver);
788}
789
790module_init(virtio_pci_init);
791
792static void __exit virtio_pci_exit(void)
793{
794	pci_unregister_driver(&virtio_pci_driver);
795}
796
797module_exit(virtio_pci_exit);