blob: 3d1bf41e8892b8e2150a51403726ebb52ea9d56c [file] [log] [blame]
Anthony Liguori33436602007-11-12 21:30:26 -06001/*
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090020#include <linux/slab.h>
Anthony Liguori33436602007-11-12 21:30:26 -060021#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 */
Al Viro97968352008-03-29 03:09:48 +000041 void __iomem *ioaddr;
Anthony Liguori33436602007-11-12 21:30:26 -060042
43 /* a list of queues so we can dispatch IRQs */
44 spinlock_t lock;
45 struct list_head virtqueues;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +030046
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;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +030056 /* Vectors allocated, excluding per-vq vectors if any */
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +030057 unsigned msix_used_vectors;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +030058 /* Whether we have vector per vq */
59 bool per_vq_vectors;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +030060};
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,
Anthony Liguori33436602007-11-12 21:30:26 -060068};
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;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +030086
87 /* MSI-X vector (or none) */
Rusty Russellf68d2402009-09-23 22:26:29 -060088 unsigned msix_vector;
Anthony Liguori33436602007-11-12 21:30:26 -060089};
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
Anthony Liguori33436602007-11-12 21:30:26 -060099/* 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
Rusty Russellc45a6812008-05-02 21:50:50 -0500105/* virtio config->get_features() implementation */
106static u32 vp_get_features(struct virtio_device *vdev)
Anthony Liguori33436602007-11-12 21:30:26 -0600107{
108 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
Anthony Liguori33436602007-11-12 21:30:26 -0600109
Rusty Russellc45a6812008-05-02 21:50:50 -0500110 /* 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}
Anthony Liguori33436602007-11-12 21:30:26 -0600114
Rusty Russellc6248962008-07-25 12:06:07 -0500115/* virtio config->finalize_features() implementation */
116static void vp_finalize_features(struct virtio_device *vdev)
Rusty Russellc45a6812008-05-02 21:50:50 -0500117{
118 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
119
Rusty Russelle34f8722008-07-25 12:06:13 -0500120 /* Give virtio_ring a chance to accept features. */
121 vring_transport_features(vdev);
122
Rusty Russellc6248962008-07-25 12:06:07 -0500123 /* 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);
Anthony Liguori33436602007-11-12 21:30:26 -0600126}
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);
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300133 void __iomem *ioaddr = vp_dev->ioaddr +
134 VIRTIO_PCI_CONFIG(vp_dev) + offset;
Anthony Liguori33436602007-11-12 21:30:26 -0600135 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);
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300148 void __iomem *ioaddr = vp_dev->ioaddr +
149 VIRTIO_PCI_CONFIG(vp_dev) + offset;
Anthony Liguori33436602007-11-12 21:30:26 -0600150 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);
Harvey Harrison597d56e2008-03-31 17:53:55 -0700169 iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
Anthony Liguori33436602007-11-12 21:30:26 -0600170}
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. */
Harvey Harrison597d56e2008-03-31 17:53:55 -0700176 iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
Anthony Liguori33436602007-11-12 21:30:26 -0600177}
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
Michael S. Tsirkin77cf52462009-05-14 13:55:31 +0300190/* 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
Anthony Liguori33436602007-11-12 21:30:26 -0600221/* 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;
Anthony Liguori33436602007-11-12 21:30:26 -0600230 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. */
Michael S. Tsirkin77cf52462009-05-14 13:55:31 +0300241 if (isr & VIRTIO_PCI_ISR_CONFIG)
242 vp_config_changed(irq, opaque);
Anthony Liguori33436602007-11-12 21:30:26 -0600243
Michael S. Tsirkin77cf52462009-05-14 13:55:31 +0300244 return vp_vring_interrupt(irq, opaque);
Anthony Liguori33436602007-11-12 21:30:26 -0600245}
246
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300247static 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);
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300259
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
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300267 pci_disable_msix(vp_dev->pci_dev);
Michael S. Tsirkinff52c3f2009-07-23 14:57:37 +0300268 vp_dev->msix_enabled = 0;
269 vp_dev->msix_vectors = 0;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300270 }
Michael S. Tsirkinff52c3f2009-07-23 14:57:37 +0300271
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;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300277}
278
Rusty Russellf68d2402009-09-23 22:26:29 -0600279static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
280 bool per_vq_vectors)
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300281{
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;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300286
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300287 vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
288 GFP_KERNEL);
289 if (!vp_dev->msix_entries)
Michael S. Tsirkinff52c3f2009-07-23 14:57:37 +0300290 goto error;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300291 vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
292 GFP_KERNEL);
293 if (!vp_dev->msix_names)
Michael S. Tsirkinff52c3f2009-07-23 14:57:37 +0300294 goto error;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300295
296 for (i = 0; i < nvectors; ++i)
297 vp_dev->msix_entries[i].entry = i;
298
Rusty Russellf68d2402009-09-23 22:26:29 -0600299 /* pci_enable_msix returns positive if we can't get this many. */
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300300 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;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300307
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300308 /* 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;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300318
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300319 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;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300325 }
326
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300327 if (!per_vq_vectors) {
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300328 /* 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)
Michael S. Tsirkinff52c3f2009-07-23 14:57:37 +0300336 goto error;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300337 ++vp_dev->msix_used_vectors;
338 }
339 return 0;
Michael S. Tsirkinff52c3f2009-07-23 14:57:37 +0300340error:
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300341 vp_free_vectors(vdev);
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300342 return err;
343}
344
Rusty Russellf68d2402009-09-23 22:26:29 -0600345static 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)
Anthony Liguori33436602007-11-12 21:30:26 -0600361{
362 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
363 struct virtio_pci_vq_info *info;
364 struct virtqueue *vq;
Hollis Blanchard13b1eb32008-12-02 16:24:40 -0600365 unsigned long flags, size;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300366 u16 num;
Anthony Liguori33436602007-11-12 21:30:26 -0600367 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;
Rusty Russellf68d2402009-09-23 22:26:29 -0600385 info->msix_vector = msix_vec;
Anthony Liguori33436602007-11-12 21:30:26 -0600386
Rusty Russell498af142008-12-30 09:25:57 -0600387 size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
Hollis Blanchard13b1eb32008-12-02 16:24:40 -0600388 info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
Anthony Liguori33436602007-11-12 21:30:26 -0600389 if (info->queue == NULL) {
390 err = -ENOMEM;
391 goto out_info;
392 }
393
394 /* activate the queue */
Rusty Russell480daab2008-12-30 09:25:56 -0600395 iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
Anthony Liguori33436602007-11-12 21:30:26 -0600396 vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
397
398 /* create the vring */
Rusty Russell87c7d572008-12-30 09:26:03 -0600399 vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN,
Rusty Russell9499f5e2009-06-12 22:16:35 -0600400 vdev, info->queue, vp_notify, callback, name);
Anthony Liguori33436602007-11-12 21:30:26 -0600401 if (!vq) {
402 err = -ENOMEM;
403 goto out_activate_queue;
404 }
405
406 vq->priv = info;
407 info->vq = vq;
408
Rusty Russellf68d2402009-09-23 22:26:29 -0600409 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) {
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300413 err = -EBUSY;
414 goto out_assign;
415 }
416 }
417
Krishna Kumar005b20a2011-10-05 11:08:59 +0530418 if (callback) {
419 spin_lock_irqsave(&vp_dev->lock, flags);
420 list_add(&info->node, &vp_dev->virtqueues);
421 spin_unlock_irqrestore(&vp_dev->lock, flags);
422 } else {
423 INIT_LIST_HEAD(&info->node);
424 }
Anthony Liguori33436602007-11-12 21:30:26 -0600425
426 return vq;
427
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300428out_assign:
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300429 vring_del_virtqueue(vq);
Anthony Liguori33436602007-11-12 21:30:26 -0600430out_activate_queue:
431 iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
Hollis Blanchard13b1eb32008-12-02 16:24:40 -0600432 free_pages_exact(info->queue, size);
Anthony Liguori33436602007-11-12 21:30:26 -0600433out_info:
434 kfree(info);
435 return ERR_PTR(err);
436}
437
Anthony Liguori33436602007-11-12 21:30:26 -0600438static void vp_del_vq(struct virtqueue *vq)
439{
440 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
441 struct virtio_pci_vq_info *info = vq->priv;
Michael S. Tsirkinf6c82502009-07-26 15:48:01 +0300442 unsigned long flags, size;
443
444 spin_lock_irqsave(&vp_dev->lock, flags);
445 list_del(&info->node);
446 spin_unlock_irqrestore(&vp_dev->lock, flags);
Anthony Liguori33436602007-11-12 21:30:26 -0600447
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300448 iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
449
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300450 if (vp_dev->msix_enabled) {
451 iowrite16(VIRTIO_MSI_NO_VECTOR,
452 vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
453 /* Flush the write out to device */
454 ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
455 }
456
Anthony Liguori33436602007-11-12 21:30:26 -0600457 vring_del_virtqueue(vq);
458
459 /* Select and deactivate the queue */
Anthony Liguori33436602007-11-12 21:30:26 -0600460 iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
461
Rusty Russell498af142008-12-30 09:25:57 -0600462 size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
Hollis Blanchard13b1eb32008-12-02 16:24:40 -0600463 free_pages_exact(info->queue, size);
Anthony Liguori33436602007-11-12 21:30:26 -0600464 kfree(info);
465}
466
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300467/* the config->del_vqs() implementation */
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600468static void vp_del_vqs(struct virtio_device *vdev)
469{
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300470 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600471 struct virtqueue *vq, *n;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300472 struct virtio_pci_vq_info *info;
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600473
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300474 list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
475 info = vq->priv;
Michael S. Tsirkin31198152010-02-25 19:08:55 +0200476 if (vp_dev->per_vq_vectors &&
477 info->msix_vector != VIRTIO_MSI_NO_VECTOR)
Rusty Russellf68d2402009-09-23 22:26:29 -0600478 free_irq(vp_dev->msix_entries[info->msix_vector].vector,
479 vq);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600480 vp_del_vq(vq);
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300481 }
482 vp_dev->per_vq_vectors = false;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300483
484 vp_free_vectors(vdev);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600485}
486
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300487static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
488 struct virtqueue *vqs[],
489 vq_callback_t *callbacks[],
490 const char *names[],
Rusty Russellf68d2402009-09-23 22:26:29 -0600491 bool use_msix,
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300492 bool per_vq_vectors)
493{
494 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
Rusty Russellf68d2402009-09-23 22:26:29 -0600495 u16 msix_vec;
496 int i, err, nvectors, allocated_vectors;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300497
Rusty Russellf68d2402009-09-23 22:26:29 -0600498 if (!use_msix) {
499 /* Old style: one normal interrupt for change and all vqs. */
500 err = vp_request_intx(vdev);
501 if (err)
502 goto error_request;
503 } else {
504 if (per_vq_vectors) {
505 /* Best option: one for change interrupt, one per vq. */
506 nvectors = 1;
507 for (i = 0; i < nvqs; ++i)
508 if (callbacks[i])
509 ++nvectors;
510 } else {
511 /* Second best: one for change, shared for all vqs. */
512 nvectors = 2;
513 }
514
515 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
516 if (err)
517 goto error_request;
518 }
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300519
520 vp_dev->per_vq_vectors = per_vq_vectors;
521 allocated_vectors = vp_dev->msix_used_vectors;
522 for (i = 0; i < nvqs; ++i) {
523 if (!callbacks[i] || !vp_dev->msix_enabled)
Rusty Russellf68d2402009-09-23 22:26:29 -0600524 msix_vec = VIRTIO_MSI_NO_VECTOR;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300525 else if (vp_dev->per_vq_vectors)
Rusty Russellf68d2402009-09-23 22:26:29 -0600526 msix_vec = allocated_vectors++;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300527 else
Rusty Russellf68d2402009-09-23 22:26:29 -0600528 msix_vec = VP_MSIX_VQ_VECTOR;
529 vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300530 if (IS_ERR(vqs[i])) {
531 err = PTR_ERR(vqs[i]);
532 goto error_find;
533 }
Michael S. Tsirkin0b22bd02009-10-22 15:06:06 +0200534
535 if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
536 continue;
537
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300538 /* allocate per-vq irq if available and necessary */
Michael S. Tsirkin0b22bd02009-10-22 15:06:06 +0200539 snprintf(vp_dev->msix_names[msix_vec],
540 sizeof *vp_dev->msix_names,
541 "%s-%s",
542 dev_name(&vp_dev->vdev.dev), names[i]);
543 err = request_irq(vp_dev->msix_entries[msix_vec].vector,
544 vring_interrupt, 0,
545 vp_dev->msix_names[msix_vec],
546 vqs[i]);
547 if (err) {
548 vp_del_vq(vqs[i]);
549 goto error_find;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300550 }
551 }
552 return 0;
553
554error_find:
555 vp_del_vqs(vdev);
556
557error_request:
558 return err;
559}
560
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300561/* the config->find_vqs() implementation */
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600562static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
563 struct virtqueue *vqs[],
564 vq_callback_t *callbacks[],
565 const char *names[])
566{
Rusty Russellf68d2402009-09-23 22:26:29 -0600567 int err;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300568
Rusty Russellf68d2402009-09-23 22:26:29 -0600569 /* Try MSI-X with one vector per queue. */
570 err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300571 if (!err)
572 return 0;
Rusty Russellf68d2402009-09-23 22:26:29 -0600573 /* Fallback: MSI-X with one vector for config, one shared for queues. */
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300574 err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
Rusty Russellf68d2402009-09-23 22:26:29 -0600575 true, false);
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300576 if (!err)
577 return 0;
578 /* Finally fall back to regular interrupts. */
Rusty Russellf68d2402009-09-23 22:26:29 -0600579 return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
580 false, false);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600581}
582
Anthony Liguori33436602007-11-12 21:30:26 -0600583static struct virtio_config_ops virtio_pci_config_ops = {
Anthony Liguori33436602007-11-12 21:30:26 -0600584 .get = vp_get,
585 .set = vp_set,
586 .get_status = vp_get_status,
587 .set_status = vp_set_status,
588 .reset = vp_reset,
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600589 .find_vqs = vp_find_vqs,
590 .del_vqs = vp_del_vqs,
Rusty Russellc45a6812008-05-02 21:50:50 -0500591 .get_features = vp_get_features,
Rusty Russellc6248962008-07-25 12:06:07 -0500592 .finalize_features = vp_finalize_features,
Anthony Liguori33436602007-11-12 21:30:26 -0600593};
594
Mark McLoughlin29f9f122008-12-10 17:45:34 +0000595static void virtio_pci_release_dev(struct device *_d)
596{
Michael S. Tsirkin72103bd2011-11-07 18:37:05 +0200597 /*
598 * No need for a release method as we allocate/free
599 * all devices together with the pci devices.
600 * Provide an empty one to avoid getting a warning from core.
601 */
Mark McLoughlin29f9f122008-12-10 17:45:34 +0000602}
603
Anthony Liguori33436602007-11-12 21:30:26 -0600604/* the PCI probing function */
605static int __devinit virtio_pci_probe(struct pci_dev *pci_dev,
606 const struct pci_device_id *id)
607{
608 struct virtio_pci_device *vp_dev;
609 int err;
610
611 /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
612 if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
613 return -ENODEV;
614
Anthony Liguori55a7c062008-01-28 09:59:59 -0600615 if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
616 printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
617 VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
618 return -ENODEV;
619 }
620
Anthony Liguori33436602007-11-12 21:30:26 -0600621 /* allocate our structure and fill it out */
622 vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
623 if (vp_dev == NULL)
624 return -ENOMEM;
625
Milton Miller8b3bb3e2011-01-07 02:55:06 -0600626 vp_dev->vdev.dev.parent = &pci_dev->dev;
Mark McLoughlin29f9f122008-12-10 17:45:34 +0000627 vp_dev->vdev.dev.release = virtio_pci_release_dev;
Anthony Liguori33436602007-11-12 21:30:26 -0600628 vp_dev->vdev.config = &virtio_pci_config_ops;
629 vp_dev->pci_dev = pci_dev;
630 INIT_LIST_HEAD(&vp_dev->virtqueues);
631 spin_lock_init(&vp_dev->lock);
632
Michael S. Tsirkinb03214d2010-06-23 22:49:06 -0600633 /* Disable MSI/MSIX to bring device to a known good state. */
634 pci_msi_off(pci_dev);
635
Anthony Liguori33436602007-11-12 21:30:26 -0600636 /* enable the device */
637 err = pci_enable_device(pci_dev);
638 if (err)
639 goto out;
640
641 err = pci_request_regions(pci_dev, "virtio-pci");
642 if (err)
643 goto out_enable_device;
644
645 vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
646 if (vp_dev->ioaddr == NULL)
647 goto out_req_regions;
648
649 pci_set_drvdata(pci_dev, vp_dev);
Michael S. Tsirkinbc505f32009-11-29 17:52:00 +0200650 pci_set_master(pci_dev);
Anthony Liguori33436602007-11-12 21:30:26 -0600651
652 /* we use the subsystem vendor/device id as the virtio vendor/device
653 * id. this allows us to use the same PCI vendor/device id for all
654 * virtio devices and to identify the particular virtio driver by
Thomas Weber88393162010-03-16 11:47:56 +0100655 * the subsystem ids */
Anthony Liguori33436602007-11-12 21:30:26 -0600656 vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
657 vp_dev->vdev.id.device = pci_dev->subsystem_device;
658
Anthony Liguori33436602007-11-12 21:30:26 -0600659 /* finally register the virtio device */
660 err = register_virtio_device(&vp_dev->vdev);
661 if (err)
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300662 goto out_set_drvdata;
Anthony Liguori33436602007-11-12 21:30:26 -0600663
664 return 0;
665
Anthony Liguori33436602007-11-12 21:30:26 -0600666out_set_drvdata:
667 pci_set_drvdata(pci_dev, NULL);
668 pci_iounmap(pci_dev, vp_dev->ioaddr);
669out_req_regions:
670 pci_release_regions(pci_dev);
671out_enable_device:
672 pci_disable_device(pci_dev);
673out:
674 kfree(vp_dev);
675 return err;
676}
677
678static void __devexit virtio_pci_remove(struct pci_dev *pci_dev)
679{
680 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
681
Anthony Liguoribd6c2692008-03-19 20:35:04 -0500682 unregister_virtio_device(&vp_dev->vdev);
Amit Shah31a3ddd2011-03-14 17:45:02 +0530683
684 vp_del_vqs(&vp_dev->vdev);
685 pci_set_drvdata(pci_dev, NULL);
686 pci_iounmap(pci_dev, vp_dev->ioaddr);
687 pci_release_regions(pci_dev);
688 pci_disable_device(pci_dev);
Michael S. Tsirkin72103bd2011-11-07 18:37:05 +0200689 kfree(vp_dev);
Anthony Liguori33436602007-11-12 21:30:26 -0600690}
691
692#ifdef CONFIG_PM
693static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state)
694{
695 pci_save_state(pci_dev);
696 pci_set_power_state(pci_dev, PCI_D3hot);
697 return 0;
698}
699
700static int virtio_pci_resume(struct pci_dev *pci_dev)
701{
702 pci_restore_state(pci_dev);
703 pci_set_power_state(pci_dev, PCI_D0);
704 return 0;
705}
706#endif
707
708static struct pci_driver virtio_pci_driver = {
709 .name = "virtio-pci",
710 .id_table = virtio_pci_id_table,
711 .probe = virtio_pci_probe,
Jamie Lokier1f08b832010-01-08 22:01:43 +0000712 .remove = __devexit_p(virtio_pci_remove),
Anthony Liguori33436602007-11-12 21:30:26 -0600713#ifdef CONFIG_PM
714 .suspend = virtio_pci_suspend,
715 .resume = virtio_pci_resume,
716#endif
717};
718
719static int __init virtio_pci_init(void)
720{
Milton Miller8b3bb3e2011-01-07 02:55:06 -0600721 return pci_register_driver(&virtio_pci_driver);
Anthony Liguori33436602007-11-12 21:30:26 -0600722}
723
724module_init(virtio_pci_init);
725
726static void __exit virtio_pci_exit(void)
727{
Anthony Liguori33436602007-11-12 21:30:26 -0600728 pci_unregister_driver(&virtio_pci_driver);
Anthony Liguori33436602007-11-12 21:30:26 -0600729}
730
731module_exit(virtio_pci_exit);