blob: a37c69941d30920649f6f25726ce72b038b752fd [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;
Jason Wang75a0a522012-08-28 13:54:14 +020051 cpumask_var_t *msix_affinity_masks;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +030052 /* Name strings for interrupts. This size should be enough,
53 * and I'm too lazy to allocate each name separately. */
54 char (*msix_names)[256];
55 /* Number of available vectors */
56 unsigned msix_vectors;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +030057 /* Vectors allocated, excluding per-vq vectors if any */
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +030058 unsigned msix_used_vectors;
Amit Shahf0fe6f12011-12-22 16:58:26 +053059
60 /* Status saved during hibernate/restore */
61 u8 saved_status;
62
Michael S. Tsirkine969fed2009-07-26 15:48:08 +030063 /* Whether we have vector per vq */
64 bool per_vq_vectors;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +030065};
66
67/* Constants for MSI-X */
68/* Use first vector for configuration changes, second and the rest for
69 * virtqueues Thus, we need at least 2 vectors for MSI. */
70enum {
71 VP_MSIX_CONFIG_VECTOR = 0,
72 VP_MSIX_VQ_VECTOR = 1,
Anthony Liguori33436602007-11-12 21:30:26 -060073};
74
75struct virtio_pci_vq_info
76{
77 /* the actual virtqueue */
78 struct virtqueue *vq;
79
80 /* the number of entries in the queue */
81 int num;
82
Anthony Liguori33436602007-11-12 21:30:26 -060083 /* the virtual address of the ring queue */
84 void *queue;
85
86 /* the list node for the virtqueues list */
87 struct list_head node;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +030088
89 /* MSI-X vector (or none) */
Rusty Russellf68d2402009-09-23 22:26:29 -060090 unsigned msix_vector;
Anthony Liguori33436602007-11-12 21:30:26 -060091};
92
93/* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
Stephen Hemminger35cdc9e2013-02-10 15:57:39 +103094static DEFINE_PCI_DEVICE_TABLE(virtio_pci_id_table) = {
95 { PCI_DEVICE(0x1af4, PCI_ANY_ID) },
96 { 0 }
Anthony Liguori33436602007-11-12 21:30:26 -060097};
98
99MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
100
Anthony Liguori33436602007-11-12 21:30:26 -0600101/* Convert a generic virtio device to our structure */
102static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
103{
104 return container_of(vdev, struct virtio_pci_device, vdev);
105}
106
Rusty Russellc45a6812008-05-02 21:50:50 -0500107/* virtio config->get_features() implementation */
108static u32 vp_get_features(struct virtio_device *vdev)
Anthony Liguori33436602007-11-12 21:30:26 -0600109{
110 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
Anthony Liguori33436602007-11-12 21:30:26 -0600111
Rusty Russellc45a6812008-05-02 21:50:50 -0500112 /* When someone needs more than 32 feature bits, we'll need to
113 * steal a bit to indicate that the rest are somewhere else. */
114 return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES);
115}
Anthony Liguori33436602007-11-12 21:30:26 -0600116
Rusty Russellc6248962008-07-25 12:06:07 -0500117/* virtio config->finalize_features() implementation */
118static void vp_finalize_features(struct virtio_device *vdev)
Rusty Russellc45a6812008-05-02 21:50:50 -0500119{
120 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
121
Rusty Russelle34f8722008-07-25 12:06:13 -0500122 /* Give virtio_ring a chance to accept features. */
123 vring_transport_features(vdev);
124
Rusty Russellc6248962008-07-25 12:06:07 -0500125 /* We only support 32 feature bits. */
126 BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1);
127 iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES);
Anthony Liguori33436602007-11-12 21:30:26 -0600128}
129
130/* virtio config->get() implementation */
131static void vp_get(struct virtio_device *vdev, unsigned offset,
132 void *buf, unsigned len)
133{
134 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300135 void __iomem *ioaddr = vp_dev->ioaddr +
136 VIRTIO_PCI_CONFIG(vp_dev) + offset;
Anthony Liguori33436602007-11-12 21:30:26 -0600137 u8 *ptr = buf;
138 int i;
139
140 for (i = 0; i < len; i++)
141 ptr[i] = ioread8(ioaddr + i);
142}
143
144/* the config->set() implementation. it's symmetric to the config->get()
145 * implementation */
146static void vp_set(struct virtio_device *vdev, unsigned offset,
147 const void *buf, unsigned len)
148{
149 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300150 void __iomem *ioaddr = vp_dev->ioaddr +
151 VIRTIO_PCI_CONFIG(vp_dev) + offset;
Anthony Liguori33436602007-11-12 21:30:26 -0600152 const u8 *ptr = buf;
153 int i;
154
155 for (i = 0; i < len; i++)
156 iowrite8(ptr[i], ioaddr + i);
157}
158
159/* config->{get,set}_status() implementations */
160static u8 vp_get_status(struct virtio_device *vdev)
161{
162 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
163 return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
164}
165
166static void vp_set_status(struct virtio_device *vdev, u8 status)
167{
168 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
169 /* We should never be setting status to 0. */
170 BUG_ON(status == 0);
Harvey Harrison597d56e2008-03-31 17:53:55 -0700171 iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
Anthony Liguori33436602007-11-12 21:30:26 -0600172}
173
Michael S. Tsirkine6af5782011-11-17 17:41:15 +0200174/* wait for pending irq handlers */
175static void vp_synchronize_vectors(struct virtio_device *vdev)
176{
177 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
178 int i;
179
180 if (vp_dev->intx_enabled)
181 synchronize_irq(vp_dev->pci_dev->irq);
182
183 for (i = 0; i < vp_dev->msix_vectors; ++i)
184 synchronize_irq(vp_dev->msix_entries[i].vector);
185}
186
Anthony Liguori33436602007-11-12 21:30:26 -0600187static void vp_reset(struct virtio_device *vdev)
188{
189 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
190 /* 0 status means a reset. */
Harvey Harrison597d56e2008-03-31 17:53:55 -0700191 iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
Michael S. Tsirkine6af5782011-11-17 17:41:15 +0200192 /* Flush out the status write, and flush in device writes,
193 * including MSi-X interrupts, if any. */
194 ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
195 /* Flush pending VQ/configuration callbacks. */
196 vp_synchronize_vectors(vdev);
Anthony Liguori33436602007-11-12 21:30:26 -0600197}
198
199/* the notify function used when creating a virt queue */
Heinz Graalfs46f9c2b2013-10-29 09:38:50 +1030200static bool vp_notify(struct virtqueue *vq)
Anthony Liguori33436602007-11-12 21:30:26 -0600201{
202 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
Anthony Liguori33436602007-11-12 21:30:26 -0600203
204 /* we write the queue's selector into the notification register to
205 * signal the other end */
Rusty Russell06ca2872012-10-16 23:56:14 +1030206 iowrite16(vq->index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
Heinz Graalfs46f9c2b2013-10-29 09:38:50 +1030207 return true;
Anthony Liguori33436602007-11-12 21:30:26 -0600208}
209
Michael S. Tsirkin77cf52462009-05-14 13:55:31 +0300210/* Handle a configuration change: Tell driver if it wants to know. */
211static irqreturn_t vp_config_changed(int irq, void *opaque)
212{
213 struct virtio_pci_device *vp_dev = opaque;
214 struct virtio_driver *drv;
215 drv = container_of(vp_dev->vdev.dev.driver,
216 struct virtio_driver, driver);
217
218 if (drv && drv->config_changed)
219 drv->config_changed(&vp_dev->vdev);
220 return IRQ_HANDLED;
221}
222
223/* Notify all virtqueues on an interrupt. */
224static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
225{
226 struct virtio_pci_device *vp_dev = opaque;
227 struct virtio_pci_vq_info *info;
228 irqreturn_t ret = IRQ_NONE;
229 unsigned long flags;
230
231 spin_lock_irqsave(&vp_dev->lock, flags);
232 list_for_each_entry(info, &vp_dev->virtqueues, node) {
233 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
234 ret = IRQ_HANDLED;
235 }
236 spin_unlock_irqrestore(&vp_dev->lock, flags);
237
238 return ret;
239}
240
Anthony Liguori33436602007-11-12 21:30:26 -0600241/* A small wrapper to also acknowledge the interrupt when it's handled.
242 * I really need an EIO hook for the vring so I can ack the interrupt once we
243 * know that we'll be handling the IRQ but before we invoke the callback since
244 * the callback may notify the host which results in the host attempting to
245 * raise an interrupt that we would then mask once we acknowledged the
246 * interrupt. */
247static irqreturn_t vp_interrupt(int irq, void *opaque)
248{
249 struct virtio_pci_device *vp_dev = opaque;
Anthony Liguori33436602007-11-12 21:30:26 -0600250 u8 isr;
251
252 /* reading the ISR has the effect of also clearing it so it's very
253 * important to save off the value. */
254 isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
255
256 /* It's definitely not us if the ISR was not high */
257 if (!isr)
258 return IRQ_NONE;
259
260 /* Configuration change? Tell driver if it wants to know. */
Michael S. Tsirkin77cf52462009-05-14 13:55:31 +0300261 if (isr & VIRTIO_PCI_ISR_CONFIG)
262 vp_config_changed(irq, opaque);
Anthony Liguori33436602007-11-12 21:30:26 -0600263
Michael S. Tsirkin77cf52462009-05-14 13:55:31 +0300264 return vp_vring_interrupt(irq, opaque);
Anthony Liguori33436602007-11-12 21:30:26 -0600265}
266
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300267static void vp_free_vectors(struct virtio_device *vdev)
268{
269 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
270 int i;
271
272 if (vp_dev->intx_enabled) {
273 free_irq(vp_dev->pci_dev->irq, vp_dev);
274 vp_dev->intx_enabled = 0;
275 }
276
277 for (i = 0; i < vp_dev->msix_used_vectors; ++i)
278 free_irq(vp_dev->msix_entries[i].vector, vp_dev);
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300279
Jason Wang75a0a522012-08-28 13:54:14 +0200280 for (i = 0; i < vp_dev->msix_vectors; i++)
281 if (vp_dev->msix_affinity_masks[i])
282 free_cpumask_var(vp_dev->msix_affinity_masks[i]);
283
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300284 if (vp_dev->msix_enabled) {
285 /* Disable the vector used for configuration */
286 iowrite16(VIRTIO_MSI_NO_VECTOR,
287 vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
288 /* Flush the write out to device */
289 ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
290
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300291 pci_disable_msix(vp_dev->pci_dev);
Michael S. Tsirkinff52c3f2009-07-23 14:57:37 +0300292 vp_dev->msix_enabled = 0;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300293 }
Michael S. Tsirkinff52c3f2009-07-23 14:57:37 +0300294
Andrew Vaginf11335d2013-07-02 15:35:13 +0930295 vp_dev->msix_vectors = 0;
Michael S. Tsirkinff52c3f2009-07-23 14:57:37 +0300296 vp_dev->msix_used_vectors = 0;
297 kfree(vp_dev->msix_names);
298 vp_dev->msix_names = NULL;
299 kfree(vp_dev->msix_entries);
300 vp_dev->msix_entries = NULL;
Jason Wang75a0a522012-08-28 13:54:14 +0200301 kfree(vp_dev->msix_affinity_masks);
302 vp_dev->msix_affinity_masks = NULL;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300303}
304
Rusty Russellf68d2402009-09-23 22:26:29 -0600305static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
306 bool per_vq_vectors)
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300307{
308 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
309 const char *name = dev_name(&vp_dev->vdev.dev);
310 unsigned i, v;
311 int err = -ENOMEM;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300312
Andrew Vaginf11335d2013-07-02 15:35:13 +0930313 vp_dev->msix_vectors = nvectors;
314
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300315 vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
316 GFP_KERNEL);
317 if (!vp_dev->msix_entries)
Michael S. Tsirkinff52c3f2009-07-23 14:57:37 +0300318 goto error;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300319 vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
320 GFP_KERNEL);
321 if (!vp_dev->msix_names)
Michael S. Tsirkinff52c3f2009-07-23 14:57:37 +0300322 goto error;
Jason Wang75a0a522012-08-28 13:54:14 +0200323 vp_dev->msix_affinity_masks
324 = kzalloc(nvectors * sizeof *vp_dev->msix_affinity_masks,
325 GFP_KERNEL);
326 if (!vp_dev->msix_affinity_masks)
327 goto error;
328 for (i = 0; i < nvectors; ++i)
329 if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
330 GFP_KERNEL))
331 goto error;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300332
333 for (i = 0; i < nvectors; ++i)
334 vp_dev->msix_entries[i].entry = i;
335
Rusty Russellf68d2402009-09-23 22:26:29 -0600336 /* pci_enable_msix returns positive if we can't get this many. */
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300337 err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors);
338 if (err > 0)
339 err = -ENOSPC;
340 if (err)
341 goto error;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300342 vp_dev->msix_enabled = 1;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300343
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300344 /* Set the vector used for configuration */
345 v = vp_dev->msix_used_vectors;
346 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
347 "%s-config", name);
348 err = request_irq(vp_dev->msix_entries[v].vector,
349 vp_config_changed, 0, vp_dev->msix_names[v],
350 vp_dev);
351 if (err)
352 goto error;
353 ++vp_dev->msix_used_vectors;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300354
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300355 iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
356 /* Verify we had enough resources to assign the vector */
357 v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
358 if (v == VIRTIO_MSI_NO_VECTOR) {
359 err = -EBUSY;
360 goto error;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300361 }
362
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300363 if (!per_vq_vectors) {
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300364 /* Shared vector for all VQs */
365 v = vp_dev->msix_used_vectors;
366 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
367 "%s-virtqueues", name);
368 err = request_irq(vp_dev->msix_entries[v].vector,
369 vp_vring_interrupt, 0, vp_dev->msix_names[v],
370 vp_dev);
371 if (err)
Michael S. Tsirkinff52c3f2009-07-23 14:57:37 +0300372 goto error;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300373 ++vp_dev->msix_used_vectors;
374 }
375 return 0;
Michael S. Tsirkinff52c3f2009-07-23 14:57:37 +0300376error:
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300377 vp_free_vectors(vdev);
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300378 return err;
379}
380
Rusty Russellf68d2402009-09-23 22:26:29 -0600381static int vp_request_intx(struct virtio_device *vdev)
382{
383 int err;
384 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
385
386 err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
387 IRQF_SHARED, dev_name(&vdev->dev), vp_dev);
388 if (!err)
389 vp_dev->intx_enabled = 1;
390 return err;
391}
392
393static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index,
394 void (*callback)(struct virtqueue *vq),
395 const char *name,
396 u16 msix_vec)
Anthony Liguori33436602007-11-12 21:30:26 -0600397{
398 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
399 struct virtio_pci_vq_info *info;
400 struct virtqueue *vq;
Hollis Blanchard13b1eb32008-12-02 16:24:40 -0600401 unsigned long flags, size;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300402 u16 num;
Anthony Liguori33436602007-11-12 21:30:26 -0600403 int err;
404
405 /* Select the queue we're interested in */
406 iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
407
408 /* Check if queue is either not available or already active. */
409 num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
410 if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
411 return ERR_PTR(-ENOENT);
412
413 /* allocate and fill out our structure the represents an active
414 * queue */
415 info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
416 if (!info)
417 return ERR_PTR(-ENOMEM);
418
Anthony Liguori33436602007-11-12 21:30:26 -0600419 info->num = num;
Rusty Russellf68d2402009-09-23 22:26:29 -0600420 info->msix_vector = msix_vec;
Anthony Liguori33436602007-11-12 21:30:26 -0600421
Rusty Russell498af142008-12-30 09:25:57 -0600422 size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
Hollis Blanchard13b1eb32008-12-02 16:24:40 -0600423 info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
Anthony Liguori33436602007-11-12 21:30:26 -0600424 if (info->queue == NULL) {
425 err = -ENOMEM;
426 goto out_info;
427 }
428
429 /* activate the queue */
Rusty Russell480daab2008-12-30 09:25:56 -0600430 iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
Anthony Liguori33436602007-11-12 21:30:26 -0600431 vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
432
433 /* create the vring */
Jason Wang17bb6d42012-08-28 13:54:13 +0200434 vq = vring_new_virtqueue(index, info->num, VIRTIO_PCI_VRING_ALIGN, vdev,
Rusty Russell7b21e342012-01-12 15:44:42 +1030435 true, info->queue, vp_notify, callback, name);
Anthony Liguori33436602007-11-12 21:30:26 -0600436 if (!vq) {
437 err = -ENOMEM;
438 goto out_activate_queue;
439 }
440
441 vq->priv = info;
442 info->vq = vq;
443
Rusty Russellf68d2402009-09-23 22:26:29 -0600444 if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
445 iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
446 msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
447 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300448 err = -EBUSY;
449 goto out_assign;
450 }
451 }
452
Krishna Kumar005b20a2011-10-05 11:08:59 +0530453 if (callback) {
454 spin_lock_irqsave(&vp_dev->lock, flags);
455 list_add(&info->node, &vp_dev->virtqueues);
456 spin_unlock_irqrestore(&vp_dev->lock, flags);
457 } else {
458 INIT_LIST_HEAD(&info->node);
459 }
Anthony Liguori33436602007-11-12 21:30:26 -0600460
461 return vq;
462
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300463out_assign:
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300464 vring_del_virtqueue(vq);
Anthony Liguori33436602007-11-12 21:30:26 -0600465out_activate_queue:
466 iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
Hollis Blanchard13b1eb32008-12-02 16:24:40 -0600467 free_pages_exact(info->queue, size);
Anthony Liguori33436602007-11-12 21:30:26 -0600468out_info:
469 kfree(info);
470 return ERR_PTR(err);
471}
472
Anthony Liguori33436602007-11-12 21:30:26 -0600473static void vp_del_vq(struct virtqueue *vq)
474{
475 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
476 struct virtio_pci_vq_info *info = vq->priv;
Michael S. Tsirkinf6c82502009-07-26 15:48:01 +0300477 unsigned long flags, size;
478
479 spin_lock_irqsave(&vp_dev->lock, flags);
480 list_del(&info->node);
481 spin_unlock_irqrestore(&vp_dev->lock, flags);
Anthony Liguori33436602007-11-12 21:30:26 -0600482
Rusty Russell06ca2872012-10-16 23:56:14 +1030483 iowrite16(vq->index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300484
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300485 if (vp_dev->msix_enabled) {
486 iowrite16(VIRTIO_MSI_NO_VECTOR,
487 vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
488 /* Flush the write out to device */
489 ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
490 }
491
Anthony Liguori33436602007-11-12 21:30:26 -0600492 vring_del_virtqueue(vq);
493
494 /* Select and deactivate the queue */
Anthony Liguori33436602007-11-12 21:30:26 -0600495 iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
496
Rusty Russell498af142008-12-30 09:25:57 -0600497 size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
Hollis Blanchard13b1eb32008-12-02 16:24:40 -0600498 free_pages_exact(info->queue, size);
Anthony Liguori33436602007-11-12 21:30:26 -0600499 kfree(info);
500}
501
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300502/* the config->del_vqs() implementation */
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600503static void vp_del_vqs(struct virtio_device *vdev)
504{
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300505 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600506 struct virtqueue *vq, *n;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300507 struct virtio_pci_vq_info *info;
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600508
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300509 list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
510 info = vq->priv;
Michael S. Tsirkin31198152010-02-25 19:08:55 +0200511 if (vp_dev->per_vq_vectors &&
512 info->msix_vector != VIRTIO_MSI_NO_VECTOR)
Rusty Russellf68d2402009-09-23 22:26:29 -0600513 free_irq(vp_dev->msix_entries[info->msix_vector].vector,
514 vq);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600515 vp_del_vq(vq);
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300516 }
517 vp_dev->per_vq_vectors = false;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300518
519 vp_free_vectors(vdev);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600520}
521
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300522static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
523 struct virtqueue *vqs[],
524 vq_callback_t *callbacks[],
525 const char *names[],
Rusty Russellf68d2402009-09-23 22:26:29 -0600526 bool use_msix,
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300527 bool per_vq_vectors)
528{
529 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
Rusty Russellf68d2402009-09-23 22:26:29 -0600530 u16 msix_vec;
531 int i, err, nvectors, allocated_vectors;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300532
Rusty Russellf68d2402009-09-23 22:26:29 -0600533 if (!use_msix) {
534 /* Old style: one normal interrupt for change and all vqs. */
535 err = vp_request_intx(vdev);
536 if (err)
537 goto error_request;
538 } else {
539 if (per_vq_vectors) {
540 /* Best option: one for change interrupt, one per vq. */
541 nvectors = 1;
542 for (i = 0; i < nvqs; ++i)
543 if (callbacks[i])
544 ++nvectors;
545 } else {
546 /* Second best: one for change, shared for all vqs. */
547 nvectors = 2;
548 }
549
550 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
551 if (err)
552 goto error_request;
553 }
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300554
555 vp_dev->per_vq_vectors = per_vq_vectors;
556 allocated_vectors = vp_dev->msix_used_vectors;
557 for (i = 0; i < nvqs; ++i) {
Michael S. Tsirkin6457f122012-09-05 21:47:45 +0300558 if (!names[i]) {
559 vqs[i] = NULL;
560 continue;
561 } else if (!callbacks[i] || !vp_dev->msix_enabled)
Rusty Russellf68d2402009-09-23 22:26:29 -0600562 msix_vec = VIRTIO_MSI_NO_VECTOR;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300563 else if (vp_dev->per_vq_vectors)
Rusty Russellf68d2402009-09-23 22:26:29 -0600564 msix_vec = allocated_vectors++;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300565 else
Rusty Russellf68d2402009-09-23 22:26:29 -0600566 msix_vec = VP_MSIX_VQ_VECTOR;
567 vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300568 if (IS_ERR(vqs[i])) {
569 err = PTR_ERR(vqs[i]);
570 goto error_find;
571 }
Michael S. Tsirkin0b22bd02009-10-22 15:06:06 +0200572
573 if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
574 continue;
575
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300576 /* allocate per-vq irq if available and necessary */
Michael S. Tsirkin0b22bd02009-10-22 15:06:06 +0200577 snprintf(vp_dev->msix_names[msix_vec],
578 sizeof *vp_dev->msix_names,
579 "%s-%s",
580 dev_name(&vp_dev->vdev.dev), names[i]);
581 err = request_irq(vp_dev->msix_entries[msix_vec].vector,
582 vring_interrupt, 0,
583 vp_dev->msix_names[msix_vec],
584 vqs[i]);
585 if (err) {
586 vp_del_vq(vqs[i]);
587 goto error_find;
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300588 }
589 }
590 return 0;
591
592error_find:
593 vp_del_vqs(vdev);
594
595error_request:
596 return err;
597}
598
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300599/* the config->find_vqs() implementation */
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600600static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
601 struct virtqueue *vqs[],
602 vq_callback_t *callbacks[],
603 const char *names[])
604{
Rusty Russellf68d2402009-09-23 22:26:29 -0600605 int err;
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300606
Rusty Russellf68d2402009-09-23 22:26:29 -0600607 /* Try MSI-X with one vector per queue. */
608 err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300609 if (!err)
610 return 0;
Rusty Russellf68d2402009-09-23 22:26:29 -0600611 /* Fallback: MSI-X with one vector for config, one shared for queues. */
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300612 err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
Rusty Russellf68d2402009-09-23 22:26:29 -0600613 true, false);
Michael S. Tsirkine969fed2009-07-26 15:48:08 +0300614 if (!err)
615 return 0;
616 /* Finally fall back to regular interrupts. */
Rusty Russellf68d2402009-09-23 22:26:29 -0600617 return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
618 false, false);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600619}
620
Rick Jones66846042011-11-14 14:17:08 +0000621static const char *vp_bus_name(struct virtio_device *vdev)
622{
623 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
624
625 return pci_name(vp_dev->pci_dev);
626}
627
Jason Wang75a0a522012-08-28 13:54:14 +0200628/* Setup the affinity for a virtqueue:
629 * - force the affinity for per vq vector
630 * - OR over all affinities for shared MSI
631 * - ignore the affinity request if we're using INTX
632 */
633static int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
634{
635 struct virtio_device *vdev = vq->vdev;
636 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
637 struct virtio_pci_vq_info *info = vq->priv;
638 struct cpumask *mask;
639 unsigned int irq;
640
641 if (!vq->callback)
642 return -EINVAL;
643
644 if (vp_dev->msix_enabled) {
645 mask = vp_dev->msix_affinity_masks[info->msix_vector];
646 irq = vp_dev->msix_entries[info->msix_vector].vector;
647 if (cpu == -1)
648 irq_set_affinity_hint(irq, NULL);
649 else {
650 cpumask_set_cpu(cpu, mask);
651 irq_set_affinity_hint(irq, mask);
652 }
653 }
654 return 0;
655}
656
Stephen Hemminger93503932013-02-10 15:57:38 +1030657static const struct virtio_config_ops virtio_pci_config_ops = {
Anthony Liguori33436602007-11-12 21:30:26 -0600658 .get = vp_get,
659 .set = vp_set,
660 .get_status = vp_get_status,
661 .set_status = vp_set_status,
662 .reset = vp_reset,
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600663 .find_vqs = vp_find_vqs,
664 .del_vqs = vp_del_vqs,
Rusty Russellc45a6812008-05-02 21:50:50 -0500665 .get_features = vp_get_features,
Rusty Russellc6248962008-07-25 12:06:07 -0500666 .finalize_features = vp_finalize_features,
Rick Jones66846042011-11-14 14:17:08 +0000667 .bus_name = vp_bus_name,
Jason Wang75a0a522012-08-28 13:54:14 +0200668 .set_vq_affinity = vp_set_vq_affinity,
Anthony Liguori33436602007-11-12 21:30:26 -0600669};
670
Mark McLoughlin29f9f122008-12-10 17:45:34 +0000671static void virtio_pci_release_dev(struct device *_d)
672{
Michael S. Tsirkin72103bd2011-11-07 18:37:05 +0200673 /*
674 * No need for a release method as we allocate/free
675 * all devices together with the pci devices.
676 * Provide an empty one to avoid getting a warning from core.
677 */
Mark McLoughlin29f9f122008-12-10 17:45:34 +0000678}
679
Anthony Liguori33436602007-11-12 21:30:26 -0600680/* the PCI probing function */
Greg Kroah-Hartman8590dbc2012-12-21 13:05:30 -0800681static int virtio_pci_probe(struct pci_dev *pci_dev,
682 const struct pci_device_id *id)
Anthony Liguori33436602007-11-12 21:30:26 -0600683{
684 struct virtio_pci_device *vp_dev;
685 int err;
686
687 /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
688 if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
689 return -ENODEV;
690
Anthony Liguori55a7c062008-01-28 09:59:59 -0600691 if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
692 printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
693 VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
694 return -ENODEV;
695 }
696
Anthony Liguori33436602007-11-12 21:30:26 -0600697 /* allocate our structure and fill it out */
698 vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
699 if (vp_dev == NULL)
700 return -ENOMEM;
701
Milton Miller8b3bb3e2011-01-07 02:55:06 -0600702 vp_dev->vdev.dev.parent = &pci_dev->dev;
Mark McLoughlin29f9f122008-12-10 17:45:34 +0000703 vp_dev->vdev.dev.release = virtio_pci_release_dev;
Anthony Liguori33436602007-11-12 21:30:26 -0600704 vp_dev->vdev.config = &virtio_pci_config_ops;
705 vp_dev->pci_dev = pci_dev;
706 INIT_LIST_HEAD(&vp_dev->virtqueues);
707 spin_lock_init(&vp_dev->lock);
708
Michael S. Tsirkinb03214d2010-06-23 22:49:06 -0600709 /* Disable MSI/MSIX to bring device to a known good state. */
710 pci_msi_off(pci_dev);
711
Anthony Liguori33436602007-11-12 21:30:26 -0600712 /* enable the device */
713 err = pci_enable_device(pci_dev);
714 if (err)
715 goto out;
716
717 err = pci_request_regions(pci_dev, "virtio-pci");
718 if (err)
719 goto out_enable_device;
720
721 vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
Peter Senna Tschudin74a74b32012-09-17 19:31:17 +0200722 if (vp_dev->ioaddr == NULL) {
723 err = -ENOMEM;
Anthony Liguori33436602007-11-12 21:30:26 -0600724 goto out_req_regions;
Peter Senna Tschudin74a74b32012-09-17 19:31:17 +0200725 }
Anthony Liguori33436602007-11-12 21:30:26 -0600726
727 pci_set_drvdata(pci_dev, vp_dev);
Michael S. Tsirkinbc505f32009-11-29 17:52:00 +0200728 pci_set_master(pci_dev);
Anthony Liguori33436602007-11-12 21:30:26 -0600729
730 /* we use the subsystem vendor/device id as the virtio vendor/device
731 * id. this allows us to use the same PCI vendor/device id for all
732 * virtio devices and to identify the particular virtio driver by
Thomas Weber88393162010-03-16 11:47:56 +0100733 * the subsystem ids */
Anthony Liguori33436602007-11-12 21:30:26 -0600734 vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
735 vp_dev->vdev.id.device = pci_dev->subsystem_device;
736
Anthony Liguori33436602007-11-12 21:30:26 -0600737 /* finally register the virtio device */
738 err = register_virtio_device(&vp_dev->vdev);
739 if (err)
Michael S. Tsirkin82af8ce2009-05-14 13:55:41 +0300740 goto out_set_drvdata;
Anthony Liguori33436602007-11-12 21:30:26 -0600741
742 return 0;
743
Anthony Liguori33436602007-11-12 21:30:26 -0600744out_set_drvdata:
745 pci_set_drvdata(pci_dev, NULL);
746 pci_iounmap(pci_dev, vp_dev->ioaddr);
747out_req_regions:
748 pci_release_regions(pci_dev);
749out_enable_device:
750 pci_disable_device(pci_dev);
751out:
752 kfree(vp_dev);
753 return err;
754}
755
Greg Kroah-Hartman8590dbc2012-12-21 13:05:30 -0800756static void virtio_pci_remove(struct pci_dev *pci_dev)
Anthony Liguori33436602007-11-12 21:30:26 -0600757{
758 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
759
Anthony Liguoribd6c2692008-03-19 20:35:04 -0500760 unregister_virtio_device(&vp_dev->vdev);
Amit Shah31a3ddd2011-03-14 17:45:02 +0530761
762 vp_del_vqs(&vp_dev->vdev);
763 pci_set_drvdata(pci_dev, NULL);
764 pci_iounmap(pci_dev, vp_dev->ioaddr);
765 pci_release_regions(pci_dev);
766 pci_disable_device(pci_dev);
Michael S. Tsirkin72103bd2011-11-07 18:37:05 +0200767 kfree(vp_dev);
Anthony Liguori33436602007-11-12 21:30:26 -0600768}
769
Aaron Lu9e266ec2013-09-09 09:57:12 +0930770#ifdef CONFIG_PM_SLEEP
Amit Shahf0fe6f12011-12-22 16:58:26 +0530771static int virtio_pci_freeze(struct device *dev)
772{
773 struct pci_dev *pci_dev = to_pci_dev(dev);
774 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
775 struct virtio_driver *drv;
776 int ret;
777
778 drv = container_of(vp_dev->vdev.dev.driver,
779 struct virtio_driver, driver);
780
781 ret = 0;
782 vp_dev->saved_status = vp_get_status(&vp_dev->vdev);
783 if (drv && drv->freeze)
784 ret = drv->freeze(&vp_dev->vdev);
785
786 if (!ret)
787 pci_disable_device(pci_dev);
788 return ret;
789}
790
Amit Shahf0fe6f12011-12-22 16:58:26 +0530791static int virtio_pci_restore(struct device *dev)
792{
793 struct pci_dev *pci_dev = to_pci_dev(dev);
794 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
795 struct virtio_driver *drv;
796 int ret;
797
798 drv = container_of(vp_dev->vdev.dev.driver,
799 struct virtio_driver, driver);
800
Amit Shah0517fdd2012-03-29 12:54:43 +0530801 ret = pci_enable_device(pci_dev);
802 if (ret)
803 return ret;
804
805 pci_set_master(pci_dev);
806 vp_finalize_features(&vp_dev->vdev);
807
808 if (drv && drv->restore)
Amit Shahf0fe6f12011-12-22 16:58:26 +0530809 ret = drv->restore(&vp_dev->vdev);
810
811 /* Finally, tell the device we're all set */
812 if (!ret)
813 vp_set_status(&vp_dev->vdev, vp_dev->saved_status);
814
815 return ret;
816}
817
Amit Shahd0775362011-12-22 16:58:25 +0530818static const struct dev_pm_ops virtio_pci_pm_ops = {
Amit Shahf878d0b2012-03-29 12:58:05 +0530819 SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore)
Amit Shahd0775362011-12-22 16:58:25 +0530820};
Anthony Liguori33436602007-11-12 21:30:26 -0600821#endif
822
823static struct pci_driver virtio_pci_driver = {
824 .name = "virtio-pci",
825 .id_table = virtio_pci_id_table,
826 .probe = virtio_pci_probe,
Greg Kroah-Hartman8590dbc2012-12-21 13:05:30 -0800827 .remove = virtio_pci_remove,
Aaron Lu9e266ec2013-09-09 09:57:12 +0930828#ifdef CONFIG_PM_SLEEP
Amit Shahd0775362011-12-22 16:58:25 +0530829 .driver.pm = &virtio_pci_pm_ops,
Anthony Liguori33436602007-11-12 21:30:26 -0600830#endif
831};
832
Wei Yongjun1ce68532012-10-16 23:56:13 +1030833module_pci_driver(virtio_pci_driver);