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> | |
20 | #include <linux/interrupt.h> | |
21 | #include <linux/virtio.h> | |
22 | #include <linux/virtio_config.h> | |
23 | #include <linux/virtio_ring.h> | |
24 | #include <linux/virtio_pci.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/spinlock.h> | |
27 | ||
28 | MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>"); | |
29 | MODULE_DESCRIPTION("virtio-pci"); | |
30 | MODULE_LICENSE("GPL"); | |
31 | MODULE_VERSION("1"); | |
32 | ||
33 | /* Our device structure */ | |
34 | struct virtio_pci_device | |
35 | { | |
36 | struct virtio_device vdev; | |
37 | struct pci_dev *pci_dev; | |
38 | ||
39 | /* the IO mapping for the PCI config space */ | |
97968358 | 40 | void __iomem *ioaddr; |
3343660d AL |
41 | |
42 | /* a list of queues so we can dispatch IRQs */ | |
43 | spinlock_t lock; | |
44 | struct list_head virtqueues; | |
82af8ce8 MT |
45 | |
46 | /* MSI-X support */ | |
47 | int msix_enabled; | |
48 | int intx_enabled; | |
49 | struct msix_entry *msix_entries; | |
50 | /* Name strings for interrupts. This size should be enough, | |
51 | * and I'm too lazy to allocate each name separately. */ | |
52 | char (*msix_names)[256]; | |
53 | /* Number of available vectors */ | |
54 | unsigned msix_vectors; | |
55 | /* Vectors allocated */ | |
56 | unsigned msix_used_vectors; | |
57 | }; | |
58 | ||
59 | /* Constants for MSI-X */ | |
60 | /* Use first vector for configuration changes, second and the rest for | |
61 | * virtqueues Thus, we need at least 2 vectors for MSI. */ | |
62 | enum { | |
63 | VP_MSIX_CONFIG_VECTOR = 0, | |
64 | VP_MSIX_VQ_VECTOR = 1, | |
3343660d AL |
65 | }; |
66 | ||
67 | struct virtio_pci_vq_info | |
68 | { | |
69 | /* the actual virtqueue */ | |
70 | struct virtqueue *vq; | |
71 | ||
72 | /* the number of entries in the queue */ | |
73 | int num; | |
74 | ||
75 | /* the index of the queue */ | |
76 | int queue_index; | |
77 | ||
78 | /* the virtual address of the ring queue */ | |
79 | void *queue; | |
80 | ||
81 | /* the list node for the virtqueues list */ | |
82 | struct list_head node; | |
82af8ce8 MT |
83 | |
84 | /* MSI-X vector (or none) */ | |
85 | unsigned vector; | |
3343660d AL |
86 | }; |
87 | ||
88 | /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */ | |
89 | static struct pci_device_id virtio_pci_id_table[] = { | |
90 | { 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 }, | |
91 | { 0 }, | |
92 | }; | |
93 | ||
94 | MODULE_DEVICE_TABLE(pci, virtio_pci_id_table); | |
95 | ||
96 | /* A PCI device has it's own struct device and so does a virtio device so | |
97 | * we create a place for the virtio devices to show up in sysfs. I think it | |
98 | * would make more sense for virtio to not insist on having it's own device. */ | |
63d12556 | 99 | static struct device *virtio_pci_root; |
3343660d | 100 | |
3343660d AL |
101 | /* Convert a generic virtio device to our structure */ |
102 | static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev) | |
103 | { | |
104 | return container_of(vdev, struct virtio_pci_device, vdev); | |
105 | } | |
106 | ||
c45a6816 RR |
107 | /* virtio config->get_features() implementation */ |
108 | static u32 vp_get_features(struct virtio_device *vdev) | |
109 | { | |
110 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
111 | ||
112 | /* 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 | } | |
116 | ||
c624896e RR |
117 | /* virtio config->finalize_features() implementation */ |
118 | static void vp_finalize_features(struct virtio_device *vdev) | |
3343660d AL |
119 | { |
120 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
3343660d | 121 | |
e34f8725 RR |
122 | /* Give virtio_ring a chance to accept features. */ |
123 | vring_transport_features(vdev); | |
124 | ||
c624896e RR |
125 | /* 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); | |
3343660d AL |
128 | } |
129 | ||
130 | /* virtio config->get() implementation */ | |
131 | static 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); | |
82af8ce8 MT |
135 | void __iomem *ioaddr = vp_dev->ioaddr + |
136 | VIRTIO_PCI_CONFIG(vp_dev) + offset; | |
3343660d AL |
137 | 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 */ | |
146 | static 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); | |
82af8ce8 MT |
150 | void __iomem *ioaddr = vp_dev->ioaddr + |
151 | VIRTIO_PCI_CONFIG(vp_dev) + offset; | |
3343660d AL |
152 | 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 */ | |
160 | static 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 | ||
166 | static 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); | |
597d56e4 | 171 | iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS); |
3343660d AL |
172 | } |
173 | ||
174 | static void vp_reset(struct virtio_device *vdev) | |
175 | { | |
176 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
177 | /* 0 status means a reset. */ | |
597d56e4 | 178 | iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS); |
3343660d AL |
179 | } |
180 | ||
181 | /* the notify function used when creating a virt queue */ | |
182 | static void vp_notify(struct virtqueue *vq) | |
183 | { | |
184 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | |
185 | struct virtio_pci_vq_info *info = vq->priv; | |
186 | ||
187 | /* we write the queue's selector into the notification register to | |
188 | * signal the other end */ | |
189 | iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY); | |
190 | } | |
191 | ||
77cf5246 MT |
192 | /* Handle a configuration change: Tell driver if it wants to know. */ |
193 | static irqreturn_t vp_config_changed(int irq, void *opaque) | |
194 | { | |
195 | struct virtio_pci_device *vp_dev = opaque; | |
196 | struct virtio_driver *drv; | |
197 | drv = container_of(vp_dev->vdev.dev.driver, | |
198 | struct virtio_driver, driver); | |
199 | ||
200 | if (drv && drv->config_changed) | |
201 | drv->config_changed(&vp_dev->vdev); | |
202 | return IRQ_HANDLED; | |
203 | } | |
204 | ||
205 | /* Notify all virtqueues on an interrupt. */ | |
206 | static irqreturn_t vp_vring_interrupt(int irq, void *opaque) | |
207 | { | |
208 | struct virtio_pci_device *vp_dev = opaque; | |
209 | struct virtio_pci_vq_info *info; | |
210 | irqreturn_t ret = IRQ_NONE; | |
211 | unsigned long flags; | |
212 | ||
213 | spin_lock_irqsave(&vp_dev->lock, flags); | |
214 | list_for_each_entry(info, &vp_dev->virtqueues, node) { | |
215 | if (vring_interrupt(irq, info->vq) == IRQ_HANDLED) | |
216 | ret = IRQ_HANDLED; | |
217 | } | |
218 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
219 | ||
220 | return ret; | |
221 | } | |
222 | ||
3343660d AL |
223 | /* A small wrapper to also acknowledge the interrupt when it's handled. |
224 | * I really need an EIO hook for the vring so I can ack the interrupt once we | |
225 | * know that we'll be handling the IRQ but before we invoke the callback since | |
226 | * the callback may notify the host which results in the host attempting to | |
227 | * raise an interrupt that we would then mask once we acknowledged the | |
228 | * interrupt. */ | |
229 | static irqreturn_t vp_interrupt(int irq, void *opaque) | |
230 | { | |
231 | struct virtio_pci_device *vp_dev = opaque; | |
3343660d AL |
232 | u8 isr; |
233 | ||
234 | /* reading the ISR has the effect of also clearing it so it's very | |
235 | * important to save off the value. */ | |
236 | isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR); | |
237 | ||
238 | /* It's definitely not us if the ISR was not high */ | |
239 | if (!isr) | |
240 | return IRQ_NONE; | |
241 | ||
242 | /* Configuration change? Tell driver if it wants to know. */ | |
77cf5246 MT |
243 | if (isr & VIRTIO_PCI_ISR_CONFIG) |
244 | vp_config_changed(irq, opaque); | |
3343660d | 245 | |
77cf5246 | 246 | return vp_vring_interrupt(irq, opaque); |
3343660d AL |
247 | } |
248 | ||
82af8ce8 MT |
249 | static void vp_free_vectors(struct virtio_device *vdev) |
250 | { | |
251 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
252 | int i; | |
253 | ||
254 | if (vp_dev->intx_enabled) { | |
255 | free_irq(vp_dev->pci_dev->irq, vp_dev); | |
256 | vp_dev->intx_enabled = 0; | |
257 | } | |
258 | ||
259 | for (i = 0; i < vp_dev->msix_used_vectors; ++i) | |
260 | free_irq(vp_dev->msix_entries[i].vector, vp_dev); | |
82af8ce8 MT |
261 | |
262 | if (vp_dev->msix_enabled) { | |
263 | /* Disable the vector used for configuration */ | |
264 | iowrite16(VIRTIO_MSI_NO_VECTOR, | |
265 | vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
266 | /* Flush the write out to device */ | |
267 | ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
268 | ||
82af8ce8 | 269 | pci_disable_msix(vp_dev->pci_dev); |
ff52c3fc MT |
270 | vp_dev->msix_enabled = 0; |
271 | vp_dev->msix_vectors = 0; | |
82af8ce8 | 272 | } |
ff52c3fc MT |
273 | |
274 | vp_dev->msix_used_vectors = 0; | |
275 | kfree(vp_dev->msix_names); | |
276 | vp_dev->msix_names = NULL; | |
277 | kfree(vp_dev->msix_entries); | |
278 | vp_dev->msix_entries = NULL; | |
82af8ce8 MT |
279 | } |
280 | ||
281 | static int vp_enable_msix(struct pci_dev *dev, struct msix_entry *entries, | |
282 | int *options, int noptions) | |
283 | { | |
284 | int i; | |
285 | for (i = 0; i < noptions; ++i) | |
286 | if (!pci_enable_msix(dev, entries, options[i])) | |
287 | return options[i]; | |
288 | return -EBUSY; | |
289 | } | |
290 | ||
291 | static int vp_request_vectors(struct virtio_device *vdev, unsigned max_vqs) | |
292 | { | |
293 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
294 | const char *name = dev_name(&vp_dev->vdev.dev); | |
295 | unsigned i, v; | |
296 | int err = -ENOMEM; | |
297 | /* We want at most one vector per queue and one for config changes. | |
298 | * Fallback to separate vectors for config and a shared for queues. | |
299 | * Finally fall back to regular interrupts. */ | |
300 | int options[] = { max_vqs + 1, 2 }; | |
301 | int nvectors = max(options[0], options[1]); | |
302 | ||
303 | vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries, | |
304 | GFP_KERNEL); | |
305 | if (!vp_dev->msix_entries) | |
ff52c3fc | 306 | goto error; |
82af8ce8 MT |
307 | vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names, |
308 | GFP_KERNEL); | |
309 | if (!vp_dev->msix_names) | |
ff52c3fc | 310 | goto error; |
82af8ce8 MT |
311 | |
312 | for (i = 0; i < nvectors; ++i) | |
313 | vp_dev->msix_entries[i].entry = i; | |
314 | ||
315 | err = vp_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, | |
316 | options, ARRAY_SIZE(options)); | |
317 | if (err < 0) { | |
318 | /* Can't allocate enough MSI-X vectors, use regular interrupt */ | |
319 | vp_dev->msix_vectors = 0; | |
320 | err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, | |
321 | IRQF_SHARED, name, vp_dev); | |
322 | if (err) | |
ff52c3fc | 323 | goto error; |
82af8ce8 MT |
324 | vp_dev->intx_enabled = 1; |
325 | } else { | |
326 | vp_dev->msix_vectors = err; | |
327 | vp_dev->msix_enabled = 1; | |
328 | ||
329 | /* Set the vector used for configuration */ | |
330 | v = vp_dev->msix_used_vectors; | |
331 | snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names, | |
332 | "%s-config", name); | |
333 | err = request_irq(vp_dev->msix_entries[v].vector, | |
334 | vp_config_changed, 0, vp_dev->msix_names[v], | |
335 | vp_dev); | |
336 | if (err) | |
ff52c3fc | 337 | goto error; |
82af8ce8 MT |
338 | ++vp_dev->msix_used_vectors; |
339 | ||
340 | iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
341 | /* Verify we had enough resources to assign the vector */ | |
342 | v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
343 | if (v == VIRTIO_MSI_NO_VECTOR) { | |
344 | err = -EBUSY; | |
ff52c3fc | 345 | goto error; |
82af8ce8 MT |
346 | } |
347 | } | |
348 | ||
349 | if (vp_dev->msix_vectors && vp_dev->msix_vectors != max_vqs + 1) { | |
350 | /* Shared vector for all VQs */ | |
351 | v = vp_dev->msix_used_vectors; | |
352 | snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names, | |
353 | "%s-virtqueues", name); | |
354 | err = request_irq(vp_dev->msix_entries[v].vector, | |
355 | vp_vring_interrupt, 0, vp_dev->msix_names[v], | |
356 | vp_dev); | |
357 | if (err) | |
ff52c3fc | 358 | goto error; |
82af8ce8 MT |
359 | ++vp_dev->msix_used_vectors; |
360 | } | |
361 | return 0; | |
ff52c3fc | 362 | error: |
82af8ce8 | 363 | vp_free_vectors(vdev); |
82af8ce8 MT |
364 | return err; |
365 | } | |
366 | ||
3343660d | 367 | static struct virtqueue *vp_find_vq(struct virtio_device *vdev, unsigned index, |
9499f5e7 RR |
368 | void (*callback)(struct virtqueue *vq), |
369 | const char *name) | |
3343660d AL |
370 | { |
371 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
372 | struct virtio_pci_vq_info *info; | |
373 | struct virtqueue *vq; | |
13b1eb33 | 374 | unsigned long flags, size; |
82af8ce8 | 375 | u16 num, vector; |
3343660d AL |
376 | int err; |
377 | ||
378 | /* Select the queue we're interested in */ | |
379 | iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); | |
380 | ||
381 | /* Check if queue is either not available or already active. */ | |
382 | num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM); | |
383 | if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN)) | |
384 | return ERR_PTR(-ENOENT); | |
385 | ||
386 | /* allocate and fill out our structure the represents an active | |
387 | * queue */ | |
388 | info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL); | |
389 | if (!info) | |
390 | return ERR_PTR(-ENOMEM); | |
391 | ||
392 | info->queue_index = index; | |
393 | info->num = num; | |
82af8ce8 | 394 | info->vector = VIRTIO_MSI_NO_VECTOR; |
3343660d | 395 | |
498af147 | 396 | size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN)); |
13b1eb33 | 397 | info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO); |
3343660d AL |
398 | if (info->queue == NULL) { |
399 | err = -ENOMEM; | |
400 | goto out_info; | |
401 | } | |
402 | ||
403 | /* activate the queue */ | |
480daab4 | 404 | iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT, |
3343660d AL |
405 | vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); |
406 | ||
407 | /* create the vring */ | |
87c7d57c | 408 | vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN, |
9499f5e7 | 409 | vdev, info->queue, vp_notify, callback, name); |
3343660d AL |
410 | if (!vq) { |
411 | err = -ENOMEM; | |
412 | goto out_activate_queue; | |
413 | } | |
414 | ||
415 | vq->priv = info; | |
416 | info->vq = vq; | |
417 | ||
82af8ce8 MT |
418 | /* allocate per-vq vector if available and necessary */ |
419 | if (callback && vp_dev->msix_used_vectors < vp_dev->msix_vectors) { | |
420 | vector = vp_dev->msix_used_vectors; | |
421 | snprintf(vp_dev->msix_names[vector], sizeof *vp_dev->msix_names, | |
422 | "%s-%s", dev_name(&vp_dev->vdev.dev), name); | |
423 | err = request_irq(vp_dev->msix_entries[vector].vector, | |
424 | vring_interrupt, 0, | |
425 | vp_dev->msix_names[vector], vq); | |
426 | if (err) | |
427 | goto out_request_irq; | |
428 | info->vector = vector; | |
429 | ++vp_dev->msix_used_vectors; | |
430 | } else | |
431 | vector = VP_MSIX_VQ_VECTOR; | |
432 | ||
433 | if (callback && vp_dev->msix_enabled) { | |
434 | iowrite16(vector, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
435 | vector = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
436 | if (vector == VIRTIO_MSI_NO_VECTOR) { | |
437 | err = -EBUSY; | |
438 | goto out_assign; | |
439 | } | |
440 | } | |
441 | ||
27ebe308 | 442 | spin_lock_irqsave(&vp_dev->lock, flags); |
3343660d | 443 | list_add(&info->node, &vp_dev->virtqueues); |
27ebe308 | 444 | spin_unlock_irqrestore(&vp_dev->lock, flags); |
3343660d AL |
445 | |
446 | return vq; | |
447 | ||
82af8ce8 MT |
448 | out_assign: |
449 | if (info->vector != VIRTIO_MSI_NO_VECTOR) { | |
450 | free_irq(vp_dev->msix_entries[info->vector].vector, vq); | |
451 | --vp_dev->msix_used_vectors; | |
452 | } | |
453 | out_request_irq: | |
454 | vring_del_virtqueue(vq); | |
3343660d AL |
455 | out_activate_queue: |
456 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); | |
13b1eb33 | 457 | free_pages_exact(info->queue, size); |
3343660d AL |
458 | out_info: |
459 | kfree(info); | |
460 | return ERR_PTR(err); | |
461 | } | |
462 | ||
3343660d AL |
463 | static void vp_del_vq(struct virtqueue *vq) |
464 | { | |
465 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | |
466 | struct virtio_pci_vq_info *info = vq->priv; | |
f6c82507 MT |
467 | unsigned long flags, size; |
468 | ||
469 | spin_lock_irqsave(&vp_dev->lock, flags); | |
470 | list_del(&info->node); | |
471 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
3343660d | 472 | |
82af8ce8 MT |
473 | iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); |
474 | ||
475 | if (info->vector != VIRTIO_MSI_NO_VECTOR) | |
476 | free_irq(vp_dev->msix_entries[info->vector].vector, vq); | |
477 | ||
478 | if (vp_dev->msix_enabled) { | |
479 | iowrite16(VIRTIO_MSI_NO_VECTOR, | |
480 | vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
481 | /* Flush the write out to device */ | |
482 | ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR); | |
483 | } | |
484 | ||
3343660d AL |
485 | vring_del_virtqueue(vq); |
486 | ||
487 | /* Select and deactivate the queue */ | |
3343660d AL |
488 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); |
489 | ||
498af147 | 490 | size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN)); |
13b1eb33 | 491 | free_pages_exact(info->queue, size); |
3343660d AL |
492 | kfree(info); |
493 | } | |
494 | ||
82af8ce8 | 495 | /* the config->del_vqs() implementation */ |
d2a7ddda MT |
496 | static void vp_del_vqs(struct virtio_device *vdev) |
497 | { | |
498 | struct virtqueue *vq, *n; | |
499 | ||
500 | list_for_each_entry_safe(vq, n, &vdev->vqs, list) | |
501 | vp_del_vq(vq); | |
82af8ce8 MT |
502 | |
503 | vp_free_vectors(vdev); | |
d2a7ddda MT |
504 | } |
505 | ||
82af8ce8 | 506 | /* the config->find_vqs() implementation */ |
d2a7ddda MT |
507 | static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs, |
508 | struct virtqueue *vqs[], | |
509 | vq_callback_t *callbacks[], | |
510 | const char *names[]) | |
511 | { | |
82af8ce8 MT |
512 | int vectors = 0; |
513 | int i, err; | |
514 | ||
515 | /* How many vectors would we like? */ | |
516 | for (i = 0; i < nvqs; ++i) | |
517 | if (callbacks[i]) | |
518 | ++vectors; | |
519 | ||
520 | err = vp_request_vectors(vdev, vectors); | |
521 | if (err) | |
522 | goto error_request; | |
d2a7ddda MT |
523 | |
524 | for (i = 0; i < nvqs; ++i) { | |
525 | vqs[i] = vp_find_vq(vdev, i, callbacks[i], names[i]); | |
526 | if (IS_ERR(vqs[i])) | |
82af8ce8 | 527 | goto error_find; |
d2a7ddda MT |
528 | } |
529 | return 0; | |
530 | ||
82af8ce8 | 531 | error_find: |
d2a7ddda | 532 | vp_del_vqs(vdev); |
82af8ce8 MT |
533 | |
534 | error_request: | |
d2a7ddda MT |
535 | return PTR_ERR(vqs[i]); |
536 | } | |
537 | ||
3343660d | 538 | static struct virtio_config_ops virtio_pci_config_ops = { |
3343660d AL |
539 | .get = vp_get, |
540 | .set = vp_set, | |
541 | .get_status = vp_get_status, | |
542 | .set_status = vp_set_status, | |
543 | .reset = vp_reset, | |
d2a7ddda MT |
544 | .find_vqs = vp_find_vqs, |
545 | .del_vqs = vp_del_vqs, | |
c45a6816 | 546 | .get_features = vp_get_features, |
c624896e | 547 | .finalize_features = vp_finalize_features, |
3343660d AL |
548 | }; |
549 | ||
29f9f12e MM |
550 | static void virtio_pci_release_dev(struct device *_d) |
551 | { | |
552 | struct virtio_device *dev = container_of(_d, struct virtio_device, dev); | |
553 | struct virtio_pci_device *vp_dev = to_vp_device(dev); | |
554 | struct pci_dev *pci_dev = vp_dev->pci_dev; | |
555 | ||
82af8ce8 | 556 | vp_del_vqs(dev); |
29f9f12e MM |
557 | pci_set_drvdata(pci_dev, NULL); |
558 | pci_iounmap(pci_dev, vp_dev->ioaddr); | |
559 | pci_release_regions(pci_dev); | |
560 | pci_disable_device(pci_dev); | |
561 | kfree(vp_dev); | |
562 | } | |
563 | ||
3343660d AL |
564 | /* the PCI probing function */ |
565 | static int __devinit virtio_pci_probe(struct pci_dev *pci_dev, | |
566 | const struct pci_device_id *id) | |
567 | { | |
568 | struct virtio_pci_device *vp_dev; | |
569 | int err; | |
570 | ||
571 | /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */ | |
572 | if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f) | |
573 | return -ENODEV; | |
574 | ||
55a7c066 AL |
575 | if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) { |
576 | printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n", | |
577 | VIRTIO_PCI_ABI_VERSION, pci_dev->revision); | |
578 | return -ENODEV; | |
579 | } | |
580 | ||
3343660d AL |
581 | /* allocate our structure and fill it out */ |
582 | vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL); | |
583 | if (vp_dev == NULL) | |
584 | return -ENOMEM; | |
585 | ||
63d12556 | 586 | vp_dev->vdev.dev.parent = virtio_pci_root; |
29f9f12e | 587 | vp_dev->vdev.dev.release = virtio_pci_release_dev; |
3343660d AL |
588 | vp_dev->vdev.config = &virtio_pci_config_ops; |
589 | vp_dev->pci_dev = pci_dev; | |
590 | INIT_LIST_HEAD(&vp_dev->virtqueues); | |
591 | spin_lock_init(&vp_dev->lock); | |
592 | ||
593 | /* enable the device */ | |
594 | err = pci_enable_device(pci_dev); | |
595 | if (err) | |
596 | goto out; | |
597 | ||
598 | err = pci_request_regions(pci_dev, "virtio-pci"); | |
599 | if (err) | |
600 | goto out_enable_device; | |
601 | ||
602 | vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0); | |
603 | if (vp_dev->ioaddr == NULL) | |
604 | goto out_req_regions; | |
605 | ||
606 | pci_set_drvdata(pci_dev, vp_dev); | |
607 | ||
608 | /* we use the subsystem vendor/device id as the virtio vendor/device | |
609 | * id. this allows us to use the same PCI vendor/device id for all | |
610 | * virtio devices and to identify the particular virtio driver by | |
611 | * the subsytem ids */ | |
612 | vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor; | |
613 | vp_dev->vdev.id.device = pci_dev->subsystem_device; | |
614 | ||
3343660d AL |
615 | /* finally register the virtio device */ |
616 | err = register_virtio_device(&vp_dev->vdev); | |
617 | if (err) | |
82af8ce8 | 618 | goto out_set_drvdata; |
3343660d AL |
619 | |
620 | return 0; | |
621 | ||
3343660d AL |
622 | out_set_drvdata: |
623 | pci_set_drvdata(pci_dev, NULL); | |
624 | pci_iounmap(pci_dev, vp_dev->ioaddr); | |
625 | out_req_regions: | |
626 | pci_release_regions(pci_dev); | |
627 | out_enable_device: | |
628 | pci_disable_device(pci_dev); | |
629 | out: | |
630 | kfree(vp_dev); | |
631 | return err; | |
632 | } | |
633 | ||
634 | static void __devexit virtio_pci_remove(struct pci_dev *pci_dev) | |
635 | { | |
636 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
637 | ||
bd6c2690 | 638 | unregister_virtio_device(&vp_dev->vdev); |
3343660d AL |
639 | } |
640 | ||
641 | #ifdef CONFIG_PM | |
642 | static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state) | |
643 | { | |
644 | pci_save_state(pci_dev); | |
645 | pci_set_power_state(pci_dev, PCI_D3hot); | |
646 | return 0; | |
647 | } | |
648 | ||
649 | static int virtio_pci_resume(struct pci_dev *pci_dev) | |
650 | { | |
651 | pci_restore_state(pci_dev); | |
652 | pci_set_power_state(pci_dev, PCI_D0); | |
653 | return 0; | |
654 | } | |
655 | #endif | |
656 | ||
657 | static struct pci_driver virtio_pci_driver = { | |
658 | .name = "virtio-pci", | |
659 | .id_table = virtio_pci_id_table, | |
660 | .probe = virtio_pci_probe, | |
661 | .remove = virtio_pci_remove, | |
662 | #ifdef CONFIG_PM | |
663 | .suspend = virtio_pci_suspend, | |
664 | .resume = virtio_pci_resume, | |
665 | #endif | |
666 | }; | |
667 | ||
668 | static int __init virtio_pci_init(void) | |
669 | { | |
670 | int err; | |
671 | ||
63d12556 MM |
672 | virtio_pci_root = root_device_register("virtio-pci"); |
673 | if (IS_ERR(virtio_pci_root)) | |
674 | return PTR_ERR(virtio_pci_root); | |
3343660d AL |
675 | |
676 | err = pci_register_driver(&virtio_pci_driver); | |
677 | if (err) | |
4b892e65 | 678 | root_device_unregister(virtio_pci_root); |
3343660d AL |
679 | |
680 | return err; | |
681 | } | |
682 | ||
683 | module_init(virtio_pci_init); | |
684 | ||
685 | static void __exit virtio_pci_exit(void) | |
686 | { | |
3343660d | 687 | pci_unregister_driver(&virtio_pci_driver); |
63d12556 | 688 | root_device_unregister(virtio_pci_root); |
3343660d AL |
689 | } |
690 | ||
691 | module_exit(virtio_pci_exit); |