Commit | Line | Data |
---|---|---|
3343660d AL |
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> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
3343660d AL |
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 | ||
29 | MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>"); | |
30 | MODULE_DESCRIPTION("virtio-pci"); | |
31 | MODULE_LICENSE("GPL"); | |
32 | MODULE_VERSION("1"); | |
33 | ||
34 | /* Our device structure */ | |
35 | struct 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 */ | |
97968358 | 41 | void __iomem *ioaddr; |
3343660d AL |
42 | |
43 | /* a list of queues so we can dispatch IRQs */ | |
44 | spinlock_t lock; | |
45 | struct list_head virtqueues; | |
82af8ce8 MT |
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; | |
e969fed5 | 56 | /* Vectors allocated, excluding per-vq vectors if any */ |
82af8ce8 | 57 | unsigned msix_used_vectors; |
f0fe6f11 AS |
58 | |
59 | /* Status saved during hibernate/restore */ | |
60 | u8 saved_status; | |
61 | ||
e969fed5 MT |
62 | /* Whether we have vector per vq */ |
63 | bool per_vq_vectors; | |
82af8ce8 MT |
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. */ | |
69 | enum { | |
70 | VP_MSIX_CONFIG_VECTOR = 0, | |
71 | VP_MSIX_VQ_VECTOR = 1, | |
3343660d AL |
72 | }; |
73 | ||
74 | struct 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 | ||
3343660d AL |
82 | /* the virtual address of the ring queue */ |
83 | void *queue; | |
84 | ||
85 | /* the list node for the virtqueues list */ | |
86 | struct list_head node; | |
82af8ce8 MT |
87 | |
88 | /* MSI-X vector (or none) */ | |
f68d2408 | 89 | unsigned msix_vector; |
3343660d AL |
90 | }; |
91 | ||
92 | /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */ | |
93 | static struct pci_device_id virtio_pci_id_table[] = { | |
94 | { 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 }, | |
95 | { 0 }, | |
96 | }; | |
97 | ||
98 | MODULE_DEVICE_TABLE(pci, virtio_pci_id_table); | |
99 | ||
3343660d AL |
100 | /* Convert a generic virtio device to our structure */ |
101 | static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev) | |
102 | { | |
103 | return container_of(vdev, struct virtio_pci_device, vdev); | |
104 | } | |
105 | ||
c45a6816 RR |
106 | /* virtio config->get_features() implementation */ |
107 | static u32 vp_get_features(struct virtio_device *vdev) | |
108 | { | |
109 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
110 | ||
111 | /* When someone needs more than 32 feature bits, we'll need to | |
112 | * steal a bit to indicate that the rest are somewhere else. */ | |
113 | return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES); | |
114 | } | |
115 | ||
c624896e RR |
116 | /* virtio config->finalize_features() implementation */ |
117 | static void vp_finalize_features(struct virtio_device *vdev) | |
3343660d AL |
118 | { |
119 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
3343660d | 120 | |
e34f8725 RR |
121 | /* Give virtio_ring a chance to accept features. */ |
122 | vring_transport_features(vdev); | |
123 | ||
c624896e RR |
124 | /* We only support 32 feature bits. */ |
125 | BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1); | |
126 | iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES); | |
3343660d AL |
127 | } |
128 | ||
129 | /* virtio config->get() implementation */ | |
130 | static void vp_get(struct virtio_device *vdev, unsigned offset, | |
131 | void *buf, unsigned len) | |
132 | { | |
133 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
82af8ce8 MT |
134 | void __iomem *ioaddr = vp_dev->ioaddr + |
135 | VIRTIO_PCI_CONFIG(vp_dev) + offset; | |
3343660d AL |
136 | u8 *ptr = buf; |
137 | int i; | |
138 | ||
139 | for (i = 0; i < len; i++) | |
140 | ptr[i] = ioread8(ioaddr + i); | |
141 | } | |
142 | ||
143 | /* the config->set() implementation. it's symmetric to the config->get() | |
144 | * implementation */ | |
145 | static void vp_set(struct virtio_device *vdev, unsigned offset, | |
146 | const void *buf, unsigned len) | |
147 | { | |
148 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
82af8ce8 MT |
149 | void __iomem *ioaddr = vp_dev->ioaddr + |
150 | VIRTIO_PCI_CONFIG(vp_dev) + offset; | |
3343660d AL |
151 | const u8 *ptr = buf; |
152 | int i; | |
153 | ||
154 | for (i = 0; i < len; i++) | |
155 | iowrite8(ptr[i], ioaddr + i); | |
156 | } | |
157 | ||
158 | /* config->{get,set}_status() implementations */ | |
159 | static u8 vp_get_status(struct virtio_device *vdev) | |
160 | { | |
161 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
162 | return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS); | |
163 | } | |
164 | ||
165 | static void vp_set_status(struct virtio_device *vdev, u8 status) | |
166 | { | |
167 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
168 | /* We should never be setting status to 0. */ | |
169 | BUG_ON(status == 0); | |
597d56e4 | 170 | iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS); |
3343660d AL |
171 | } |
172 | ||
e6af578c MT |
173 | /* wait for pending irq handlers */ |
174 | static void vp_synchronize_vectors(struct virtio_device *vdev) | |
175 | { | |
176 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
177 | int i; | |
178 | ||
179 | if (vp_dev->intx_enabled) | |
180 | synchronize_irq(vp_dev->pci_dev->irq); | |
181 | ||
182 | for (i = 0; i < vp_dev->msix_vectors; ++i) | |
183 | synchronize_irq(vp_dev->msix_entries[i].vector); | |
184 | } | |
185 | ||
3343660d AL |
186 | static void vp_reset(struct virtio_device *vdev) |
187 | { | |
188 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
189 | /* 0 status means a reset. */ | |
597d56e4 | 190 | iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS); |
e6af578c MT |
191 | /* Flush out the status write, and flush in device writes, |
192 | * including MSi-X interrupts, if any. */ | |
193 | ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS); | |
194 | /* Flush pending VQ/configuration callbacks. */ | |
195 | vp_synchronize_vectors(vdev); | |
3343660d AL |
196 | } |
197 | ||
198 | /* the notify function used when creating a virt queue */ | |
199 | static void vp_notify(struct virtqueue *vq) | |
200 | { | |
201 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | |
3343660d AL |
202 | |
203 | /* we write the queue's selector into the notification register to | |
204 | * signal the other end */ | |
17bb6d40 JW |
205 | iowrite16(virtqueue_get_queue_index(vq), |
206 | vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY); | |
3343660d AL |
207 | } |
208 | ||
77cf5246 MT |
209 | /* Handle a configuration change: Tell driver if it wants to know. */ |
210 | static irqreturn_t vp_config_changed(int irq, void *opaque) | |
211 | { | |
212 | struct virtio_pci_device *vp_dev = opaque; | |
213 | struct virtio_driver *drv; | |
214 | drv = container_of(vp_dev->vdev.dev.driver, | |
215 | struct virtio_driver, driver); | |
216 | ||
217 | if (drv && drv->config_changed) | |
218 | drv->config_changed(&vp_dev->vdev); | |
219 | return IRQ_HANDLED; | |
220 | } | |
221 | ||
222 | /* Notify all virtqueues on an interrupt. */ | |
223 | static irqreturn_t vp_vring_interrupt(int irq, void *opaque) | |
224 | { | |
225 | struct virtio_pci_device *vp_dev = opaque; | |
226 | struct virtio_pci_vq_info *info; | |
227 | irqreturn_t ret = IRQ_NONE; | |
228 | unsigned long flags; | |
229 | ||
230 | spin_lock_irqsave(&vp_dev->lock, flags); | |
231 | list_for_each_entry(info, &vp_dev->virtqueues, node) { | |
232 | if (vring_interrupt(irq, info->vq) == IRQ_HANDLED) | |
233 | ret = IRQ_HANDLED; | |
234 | } | |
235 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
236 | ||
237 | return ret; | |
238 | } | |
239 | ||
3343660d AL |
240 | /* A small wrapper to also acknowledge the interrupt when it's handled. |
241 | * I really need an EIO hook for the vring so I can ack the interrupt once we | |
242 | * know that we'll be handling the IRQ but before we invoke the callback since | |
243 | * the callback may notify the host which results in the host attempting to | |
244 | * raise an interrupt that we would then mask once we acknowledged the | |
245 | * interrupt. */ | |
246 | static irqreturn_t vp_interrupt(int irq, void *opaque) | |
247 | { | |
248 | struct virtio_pci_device *vp_dev = opaque; | |
3343660d AL |
249 | u8 isr; |
250 | ||
251 | /* reading the ISR has the effect of also clearing it so it's very | |
252 | * important to save off the value. */ | |
253 | isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR); | |
254 | ||
255 | /* It's definitely not us if the ISR was not high */ | |
256 | if (!isr) | |
257 | return IRQ_NONE; | |
258 | ||
259 | /* Configuration change? Tell driver if it wants to know. */ | |
77cf5246 MT |
260 | if (isr & VIRTIO_PCI_ISR_CONFIG) |
261 | vp_config_changed(irq, opaque); | |
3343660d | 262 | |
77cf5246 | 263 | return vp_vring_interrupt(irq, opaque); |
3343660d AL |
264 | } |
265 | ||
82af8ce8 MT |
266 | static void vp_free_vectors(struct virtio_device *vdev) |
267 | { | |
268 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
269 | int i; | |
270 | ||
271 | if (vp_dev->intx_enabled) { | |
272 | free_irq(vp_dev->pci_dev->irq, vp_dev); | |
273 | vp_dev->intx_enabled = 0; | |
274 | } | |
275 | ||
276 | for (i = 0; i < vp_dev->msix_used_vectors; ++i) | |
277 | free_irq(vp_dev->msix_entries[i].vector, vp_dev); | |
82af8ce8 MT |
278 | |
279 | if (vp_dev->msix_enabled) { | |
280 | /* Disable the vector used for configuration */ | |
281 | iowrite16(VIRTIO_MSI_NO_VECTOR, | |
282 | vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
283 | /* Flush the write out to device */ | |
284 | ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
285 | ||
82af8ce8 | 286 | pci_disable_msix(vp_dev->pci_dev); |
ff52c3fc MT |
287 | vp_dev->msix_enabled = 0; |
288 | vp_dev->msix_vectors = 0; | |
82af8ce8 | 289 | } |
ff52c3fc MT |
290 | |
291 | vp_dev->msix_used_vectors = 0; | |
292 | kfree(vp_dev->msix_names); | |
293 | vp_dev->msix_names = NULL; | |
294 | kfree(vp_dev->msix_entries); | |
295 | vp_dev->msix_entries = NULL; | |
82af8ce8 MT |
296 | } |
297 | ||
f68d2408 RR |
298 | static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors, |
299 | bool per_vq_vectors) | |
82af8ce8 MT |
300 | { |
301 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
302 | const char *name = dev_name(&vp_dev->vdev.dev); | |
303 | unsigned i, v; | |
304 | int err = -ENOMEM; | |
e969fed5 | 305 | |
82af8ce8 MT |
306 | vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries, |
307 | GFP_KERNEL); | |
308 | if (!vp_dev->msix_entries) | |
ff52c3fc | 309 | goto error; |
82af8ce8 MT |
310 | vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names, |
311 | GFP_KERNEL); | |
312 | if (!vp_dev->msix_names) | |
ff52c3fc | 313 | goto error; |
82af8ce8 MT |
314 | |
315 | for (i = 0; i < nvectors; ++i) | |
316 | vp_dev->msix_entries[i].entry = i; | |
317 | ||
f68d2408 | 318 | /* pci_enable_msix returns positive if we can't get this many. */ |
e969fed5 MT |
319 | err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors); |
320 | if (err > 0) | |
321 | err = -ENOSPC; | |
322 | if (err) | |
323 | goto error; | |
324 | vp_dev->msix_vectors = nvectors; | |
325 | vp_dev->msix_enabled = 1; | |
326 | ||
327 | /* Set the vector used for configuration */ | |
328 | v = vp_dev->msix_used_vectors; | |
329 | snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names, | |
330 | "%s-config", name); | |
331 | err = request_irq(vp_dev->msix_entries[v].vector, | |
332 | vp_config_changed, 0, vp_dev->msix_names[v], | |
333 | vp_dev); | |
334 | if (err) | |
335 | goto error; | |
336 | ++vp_dev->msix_used_vectors; | |
82af8ce8 | 337 | |
e969fed5 MT |
338 | iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); |
339 | /* Verify we had enough resources to assign the vector */ | |
340 | v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
341 | if (v == VIRTIO_MSI_NO_VECTOR) { | |
342 | err = -EBUSY; | |
343 | goto error; | |
82af8ce8 MT |
344 | } |
345 | ||
e969fed5 | 346 | if (!per_vq_vectors) { |
82af8ce8 MT |
347 | /* Shared vector for all VQs */ |
348 | v = vp_dev->msix_used_vectors; | |
349 | snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names, | |
350 | "%s-virtqueues", name); | |
351 | err = request_irq(vp_dev->msix_entries[v].vector, | |
352 | vp_vring_interrupt, 0, vp_dev->msix_names[v], | |
353 | vp_dev); | |
354 | if (err) | |
ff52c3fc | 355 | goto error; |
82af8ce8 MT |
356 | ++vp_dev->msix_used_vectors; |
357 | } | |
358 | return 0; | |
ff52c3fc | 359 | error: |
82af8ce8 | 360 | vp_free_vectors(vdev); |
82af8ce8 MT |
361 | return err; |
362 | } | |
363 | ||
f68d2408 RR |
364 | static int vp_request_intx(struct virtio_device *vdev) |
365 | { | |
366 | int err; | |
367 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
368 | ||
369 | err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, | |
370 | IRQF_SHARED, dev_name(&vdev->dev), vp_dev); | |
371 | if (!err) | |
372 | vp_dev->intx_enabled = 1; | |
373 | return err; | |
374 | } | |
375 | ||
376 | static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index, | |
377 | void (*callback)(struct virtqueue *vq), | |
378 | const char *name, | |
379 | u16 msix_vec) | |
3343660d AL |
380 | { |
381 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
382 | struct virtio_pci_vq_info *info; | |
383 | struct virtqueue *vq; | |
13b1eb33 | 384 | unsigned long flags, size; |
e969fed5 | 385 | u16 num; |
3343660d AL |
386 | int err; |
387 | ||
388 | /* Select the queue we're interested in */ | |
389 | iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); | |
390 | ||
391 | /* Check if queue is either not available or already active. */ | |
392 | num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM); | |
393 | if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN)) | |
394 | return ERR_PTR(-ENOENT); | |
395 | ||
396 | /* allocate and fill out our structure the represents an active | |
397 | * queue */ | |
398 | info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL); | |
399 | if (!info) | |
400 | return ERR_PTR(-ENOMEM); | |
401 | ||
3343660d | 402 | info->num = num; |
f68d2408 | 403 | info->msix_vector = msix_vec; |
3343660d | 404 | |
498af147 | 405 | size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN)); |
13b1eb33 | 406 | info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO); |
3343660d AL |
407 | if (info->queue == NULL) { |
408 | err = -ENOMEM; | |
409 | goto out_info; | |
410 | } | |
411 | ||
412 | /* activate the queue */ | |
480daab4 | 413 | iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT, |
3343660d AL |
414 | vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); |
415 | ||
416 | /* create the vring */ | |
17bb6d40 | 417 | vq = vring_new_virtqueue(index, info->num, VIRTIO_PCI_VRING_ALIGN, vdev, |
7b21e34f | 418 | true, info->queue, vp_notify, callback, name); |
3343660d AL |
419 | if (!vq) { |
420 | err = -ENOMEM; | |
421 | goto out_activate_queue; | |
422 | } | |
423 | ||
424 | vq->priv = info; | |
425 | info->vq = vq; | |
426 | ||
f68d2408 RR |
427 | if (msix_vec != VIRTIO_MSI_NO_VECTOR) { |
428 | iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
429 | msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
430 | if (msix_vec == VIRTIO_MSI_NO_VECTOR) { | |
82af8ce8 MT |
431 | err = -EBUSY; |
432 | goto out_assign; | |
433 | } | |
434 | } | |
435 | ||
005b20a8 KK |
436 | if (callback) { |
437 | spin_lock_irqsave(&vp_dev->lock, flags); | |
438 | list_add(&info->node, &vp_dev->virtqueues); | |
439 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
440 | } else { | |
441 | INIT_LIST_HEAD(&info->node); | |
442 | } | |
3343660d AL |
443 | |
444 | return vq; | |
445 | ||
82af8ce8 | 446 | out_assign: |
82af8ce8 | 447 | vring_del_virtqueue(vq); |
3343660d AL |
448 | out_activate_queue: |
449 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); | |
13b1eb33 | 450 | free_pages_exact(info->queue, size); |
3343660d AL |
451 | out_info: |
452 | kfree(info); | |
453 | return ERR_PTR(err); | |
454 | } | |
455 | ||
3343660d AL |
456 | static void vp_del_vq(struct virtqueue *vq) |
457 | { | |
458 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | |
459 | struct virtio_pci_vq_info *info = vq->priv; | |
f6c82507 MT |
460 | unsigned long flags, size; |
461 | ||
462 | spin_lock_irqsave(&vp_dev->lock, flags); | |
463 | list_del(&info->node); | |
464 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
3343660d | 465 | |
17bb6d40 JW |
466 | iowrite16(virtqueue_get_queue_index(vq), |
467 | vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); | |
82af8ce8 | 468 | |
82af8ce8 MT |
469 | if (vp_dev->msix_enabled) { |
470 | iowrite16(VIRTIO_MSI_NO_VECTOR, | |
471 | vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
472 | /* Flush the write out to device */ | |
473 | ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR); | |
474 | } | |
475 | ||
3343660d AL |
476 | vring_del_virtqueue(vq); |
477 | ||
478 | /* Select and deactivate the queue */ | |
3343660d AL |
479 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); |
480 | ||
498af147 | 481 | size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN)); |
13b1eb33 | 482 | free_pages_exact(info->queue, size); |
3343660d AL |
483 | kfree(info); |
484 | } | |
485 | ||
82af8ce8 | 486 | /* the config->del_vqs() implementation */ |
d2a7ddda MT |
487 | static void vp_del_vqs(struct virtio_device *vdev) |
488 | { | |
e969fed5 | 489 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); |
d2a7ddda | 490 | struct virtqueue *vq, *n; |
e969fed5 | 491 | struct virtio_pci_vq_info *info; |
d2a7ddda | 492 | |
e969fed5 MT |
493 | list_for_each_entry_safe(vq, n, &vdev->vqs, list) { |
494 | info = vq->priv; | |
31198159 MT |
495 | if (vp_dev->per_vq_vectors && |
496 | info->msix_vector != VIRTIO_MSI_NO_VECTOR) | |
f68d2408 RR |
497 | free_irq(vp_dev->msix_entries[info->msix_vector].vector, |
498 | vq); | |
d2a7ddda | 499 | vp_del_vq(vq); |
e969fed5 MT |
500 | } |
501 | vp_dev->per_vq_vectors = false; | |
82af8ce8 MT |
502 | |
503 | vp_free_vectors(vdev); | |
d2a7ddda MT |
504 | } |
505 | ||
e969fed5 MT |
506 | static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs, |
507 | struct virtqueue *vqs[], | |
508 | vq_callback_t *callbacks[], | |
509 | const char *names[], | |
f68d2408 | 510 | bool use_msix, |
e969fed5 | 511 | bool per_vq_vectors) |
d2a7ddda | 512 | { |
e969fed5 | 513 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); |
f68d2408 RR |
514 | u16 msix_vec; |
515 | int i, err, nvectors, allocated_vectors; | |
82af8ce8 | 516 | |
f68d2408 RR |
517 | if (!use_msix) { |
518 | /* Old style: one normal interrupt for change and all vqs. */ | |
519 | err = vp_request_intx(vdev); | |
520 | if (err) | |
521 | goto error_request; | |
522 | } else { | |
523 | if (per_vq_vectors) { | |
524 | /* Best option: one for change interrupt, one per vq. */ | |
525 | nvectors = 1; | |
526 | for (i = 0; i < nvqs; ++i) | |
527 | if (callbacks[i]) | |
528 | ++nvectors; | |
529 | } else { | |
530 | /* Second best: one for change, shared for all vqs. */ | |
531 | nvectors = 2; | |
532 | } | |
533 | ||
534 | err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors); | |
535 | if (err) | |
536 | goto error_request; | |
537 | } | |
d2a7ddda | 538 | |
e969fed5 MT |
539 | vp_dev->per_vq_vectors = per_vq_vectors; |
540 | allocated_vectors = vp_dev->msix_used_vectors; | |
d2a7ddda | 541 | for (i = 0; i < nvqs; ++i) { |
e969fed5 | 542 | if (!callbacks[i] || !vp_dev->msix_enabled) |
f68d2408 | 543 | msix_vec = VIRTIO_MSI_NO_VECTOR; |
e969fed5 | 544 | else if (vp_dev->per_vq_vectors) |
f68d2408 | 545 | msix_vec = allocated_vectors++; |
e969fed5 | 546 | else |
f68d2408 RR |
547 | msix_vec = VP_MSIX_VQ_VECTOR; |
548 | vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec); | |
e969fed5 MT |
549 | if (IS_ERR(vqs[i])) { |
550 | err = PTR_ERR(vqs[i]); | |
82af8ce8 | 551 | goto error_find; |
e969fed5 | 552 | } |
0b22bd0b MT |
553 | |
554 | if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR) | |
555 | continue; | |
556 | ||
e969fed5 | 557 | /* allocate per-vq irq if available and necessary */ |
0b22bd0b MT |
558 | snprintf(vp_dev->msix_names[msix_vec], |
559 | sizeof *vp_dev->msix_names, | |
560 | "%s-%s", | |
561 | dev_name(&vp_dev->vdev.dev), names[i]); | |
562 | err = request_irq(vp_dev->msix_entries[msix_vec].vector, | |
563 | vring_interrupt, 0, | |
564 | vp_dev->msix_names[msix_vec], | |
565 | vqs[i]); | |
566 | if (err) { | |
567 | vp_del_vq(vqs[i]); | |
568 | goto error_find; | |
e969fed5 | 569 | } |
d2a7ddda MT |
570 | } |
571 | return 0; | |
572 | ||
82af8ce8 | 573 | error_find: |
d2a7ddda | 574 | vp_del_vqs(vdev); |
82af8ce8 MT |
575 | |
576 | error_request: | |
e969fed5 MT |
577 | return err; |
578 | } | |
579 | ||
580 | /* the config->find_vqs() implementation */ | |
581 | static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs, | |
582 | struct virtqueue *vqs[], | |
583 | vq_callback_t *callbacks[], | |
584 | const char *names[]) | |
585 | { | |
f68d2408 | 586 | int err; |
e969fed5 | 587 | |
f68d2408 RR |
588 | /* Try MSI-X with one vector per queue. */ |
589 | err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true); | |
e969fed5 MT |
590 | if (!err) |
591 | return 0; | |
f68d2408 | 592 | /* Fallback: MSI-X with one vector for config, one shared for queues. */ |
e969fed5 | 593 | err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, |
f68d2408 | 594 | true, false); |
e969fed5 MT |
595 | if (!err) |
596 | return 0; | |
597 | /* Finally fall back to regular interrupts. */ | |
f68d2408 RR |
598 | return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, |
599 | false, false); | |
d2a7ddda MT |
600 | } |
601 | ||
66846048 RJ |
602 | static const char *vp_bus_name(struct virtio_device *vdev) |
603 | { | |
604 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
605 | ||
606 | return pci_name(vp_dev->pci_dev); | |
607 | } | |
608 | ||
3343660d | 609 | static struct virtio_config_ops virtio_pci_config_ops = { |
3343660d AL |
610 | .get = vp_get, |
611 | .set = vp_set, | |
612 | .get_status = vp_get_status, | |
613 | .set_status = vp_set_status, | |
614 | .reset = vp_reset, | |
d2a7ddda MT |
615 | .find_vqs = vp_find_vqs, |
616 | .del_vqs = vp_del_vqs, | |
c45a6816 | 617 | .get_features = vp_get_features, |
c624896e | 618 | .finalize_features = vp_finalize_features, |
66846048 | 619 | .bus_name = vp_bus_name, |
3343660d AL |
620 | }; |
621 | ||
29f9f12e MM |
622 | static void virtio_pci_release_dev(struct device *_d) |
623 | { | |
72103bd1 MT |
624 | /* |
625 | * No need for a release method as we allocate/free | |
626 | * all devices together with the pci devices. | |
627 | * Provide an empty one to avoid getting a warning from core. | |
628 | */ | |
29f9f12e MM |
629 | } |
630 | ||
3343660d AL |
631 | /* the PCI probing function */ |
632 | static int __devinit virtio_pci_probe(struct pci_dev *pci_dev, | |
633 | const struct pci_device_id *id) | |
634 | { | |
635 | struct virtio_pci_device *vp_dev; | |
636 | int err; | |
637 | ||
638 | /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */ | |
639 | if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f) | |
640 | return -ENODEV; | |
641 | ||
55a7c066 AL |
642 | if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) { |
643 | printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n", | |
644 | VIRTIO_PCI_ABI_VERSION, pci_dev->revision); | |
645 | return -ENODEV; | |
646 | } | |
647 | ||
3343660d AL |
648 | /* allocate our structure and fill it out */ |
649 | vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL); | |
650 | if (vp_dev == NULL) | |
651 | return -ENOMEM; | |
652 | ||
8b3bb3ec | 653 | vp_dev->vdev.dev.parent = &pci_dev->dev; |
29f9f12e | 654 | vp_dev->vdev.dev.release = virtio_pci_release_dev; |
3343660d AL |
655 | vp_dev->vdev.config = &virtio_pci_config_ops; |
656 | vp_dev->pci_dev = pci_dev; | |
657 | INIT_LIST_HEAD(&vp_dev->virtqueues); | |
658 | spin_lock_init(&vp_dev->lock); | |
659 | ||
b03214d5 MT |
660 | /* Disable MSI/MSIX to bring device to a known good state. */ |
661 | pci_msi_off(pci_dev); | |
662 | ||
3343660d AL |
663 | /* enable the device */ |
664 | err = pci_enable_device(pci_dev); | |
665 | if (err) | |
666 | goto out; | |
667 | ||
668 | err = pci_request_regions(pci_dev, "virtio-pci"); | |
669 | if (err) | |
670 | goto out_enable_device; | |
671 | ||
672 | vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0); | |
673 | if (vp_dev->ioaddr == NULL) | |
674 | goto out_req_regions; | |
675 | ||
676 | pci_set_drvdata(pci_dev, vp_dev); | |
bc505f37 | 677 | pci_set_master(pci_dev); |
3343660d AL |
678 | |
679 | /* we use the subsystem vendor/device id as the virtio vendor/device | |
680 | * id. this allows us to use the same PCI vendor/device id for all | |
681 | * virtio devices and to identify the particular virtio driver by | |
88393161 | 682 | * the subsystem ids */ |
3343660d AL |
683 | vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor; |
684 | vp_dev->vdev.id.device = pci_dev->subsystem_device; | |
685 | ||
3343660d AL |
686 | /* finally register the virtio device */ |
687 | err = register_virtio_device(&vp_dev->vdev); | |
688 | if (err) | |
82af8ce8 | 689 | goto out_set_drvdata; |
3343660d AL |
690 | |
691 | return 0; | |
692 | ||
3343660d AL |
693 | out_set_drvdata: |
694 | pci_set_drvdata(pci_dev, NULL); | |
695 | pci_iounmap(pci_dev, vp_dev->ioaddr); | |
696 | out_req_regions: | |
697 | pci_release_regions(pci_dev); | |
698 | out_enable_device: | |
699 | pci_disable_device(pci_dev); | |
700 | out: | |
701 | kfree(vp_dev); | |
702 | return err; | |
703 | } | |
704 | ||
705 | static void __devexit virtio_pci_remove(struct pci_dev *pci_dev) | |
706 | { | |
707 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
708 | ||
bd6c2690 | 709 | unregister_virtio_device(&vp_dev->vdev); |
31a3ddda AS |
710 | |
711 | vp_del_vqs(&vp_dev->vdev); | |
712 | pci_set_drvdata(pci_dev, NULL); | |
713 | pci_iounmap(pci_dev, vp_dev->ioaddr); | |
714 | pci_release_regions(pci_dev); | |
715 | pci_disable_device(pci_dev); | |
72103bd1 | 716 | kfree(vp_dev); |
3343660d AL |
717 | } |
718 | ||
719 | #ifdef CONFIG_PM | |
f0fe6f11 AS |
720 | static int virtio_pci_freeze(struct device *dev) |
721 | { | |
722 | struct pci_dev *pci_dev = to_pci_dev(dev); | |
723 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
724 | struct virtio_driver *drv; | |
725 | int ret; | |
726 | ||
727 | drv = container_of(vp_dev->vdev.dev.driver, | |
728 | struct virtio_driver, driver); | |
729 | ||
730 | ret = 0; | |
731 | vp_dev->saved_status = vp_get_status(&vp_dev->vdev); | |
732 | if (drv && drv->freeze) | |
733 | ret = drv->freeze(&vp_dev->vdev); | |
734 | ||
735 | if (!ret) | |
736 | pci_disable_device(pci_dev); | |
737 | return ret; | |
738 | } | |
739 | ||
0517fdd1 | 740 | static int virtio_pci_restore(struct device *dev) |
f0fe6f11 AS |
741 | { |
742 | struct pci_dev *pci_dev = to_pci_dev(dev); | |
743 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
0517fdd1 | 744 | struct virtio_driver *drv; |
f0fe6f11 AS |
745 | int ret; |
746 | ||
0517fdd1 AS |
747 | drv = container_of(vp_dev->vdev.dev.driver, |
748 | struct virtio_driver, driver); | |
749 | ||
f0fe6f11 AS |
750 | ret = pci_enable_device(pci_dev); |
751 | if (ret) | |
752 | return ret; | |
0517fdd1 | 753 | |
f0fe6f11 AS |
754 | pci_set_master(pci_dev); |
755 | vp_finalize_features(&vp_dev->vdev); | |
756 | ||
0517fdd1 | 757 | if (drv && drv->restore) |
f0fe6f11 AS |
758 | ret = drv->restore(&vp_dev->vdev); |
759 | ||
760 | /* Finally, tell the device we're all set */ | |
761 | if (!ret) | |
762 | vp_set_status(&vp_dev->vdev, vp_dev->saved_status); | |
763 | ||
764 | return ret; | |
765 | } | |
766 | ||
d0775363 | 767 | static const struct dev_pm_ops virtio_pci_pm_ops = { |
f878d0be | 768 | SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore) |
d0775363 | 769 | }; |
3343660d AL |
770 | #endif |
771 | ||
772 | static struct pci_driver virtio_pci_driver = { | |
773 | .name = "virtio-pci", | |
774 | .id_table = virtio_pci_id_table, | |
775 | .probe = virtio_pci_probe, | |
1f08b833 | 776 | .remove = __devexit_p(virtio_pci_remove), |
3343660d | 777 | #ifdef CONFIG_PM |
d0775363 | 778 | .driver.pm = &virtio_pci_pm_ops, |
3343660d AL |
779 | #endif |
780 | }; | |
781 | ||
782 | static int __init virtio_pci_init(void) | |
783 | { | |
8b3bb3ec | 784 | return pci_register_driver(&virtio_pci_driver); |
3343660d AL |
785 | } |
786 | ||
787 | module_init(virtio_pci_init); | |
788 | ||
789 | static void __exit virtio_pci_exit(void) | |
790 | { | |
3343660d AL |
791 | pci_unregister_driver(&virtio_pci_driver); |
792 | } | |
793 | ||
794 | module_exit(virtio_pci_exit); |