]>
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; | |
45 | }; | |
46 | ||
47 | struct virtio_pci_vq_info | |
48 | { | |
49 | /* the actual virtqueue */ | |
50 | struct virtqueue *vq; | |
51 | ||
52 | /* the number of entries in the queue */ | |
53 | int num; | |
54 | ||
55 | /* the index of the queue */ | |
56 | int queue_index; | |
57 | ||
58 | /* the virtual address of the ring queue */ | |
59 | void *queue; | |
60 | ||
61 | /* the list node for the virtqueues list */ | |
62 | struct list_head node; | |
63 | }; | |
64 | ||
65 | /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */ | |
66 | static struct pci_device_id virtio_pci_id_table[] = { | |
67 | { 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 }, | |
68 | { 0 }, | |
69 | }; | |
70 | ||
71 | MODULE_DEVICE_TABLE(pci, virtio_pci_id_table); | |
72 | ||
73 | /* A PCI device has it's own struct device and so does a virtio device so | |
74 | * we create a place for the virtio devices to show up in sysfs. I think it | |
75 | * would make more sense for virtio to not insist on having it's own device. */ | |
63d12556 | 76 | static struct device *virtio_pci_root; |
3343660d | 77 | |
3343660d AL |
78 | /* Convert a generic virtio device to our structure */ |
79 | static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev) | |
80 | { | |
81 | return container_of(vdev, struct virtio_pci_device, vdev); | |
82 | } | |
83 | ||
c45a6816 RR |
84 | /* virtio config->get_features() implementation */ |
85 | static u32 vp_get_features(struct virtio_device *vdev) | |
86 | { | |
87 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
88 | ||
89 | /* When someone needs more than 32 feature bits, we'll need to | |
90 | * steal a bit to indicate that the rest are somewhere else. */ | |
91 | return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES); | |
92 | } | |
93 | ||
c624896e RR |
94 | /* virtio config->finalize_features() implementation */ |
95 | static void vp_finalize_features(struct virtio_device *vdev) | |
3343660d AL |
96 | { |
97 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
3343660d | 98 | |
e34f8725 RR |
99 | /* Give virtio_ring a chance to accept features. */ |
100 | vring_transport_features(vdev); | |
101 | ||
c624896e RR |
102 | /* We only support 32 feature bits. */ |
103 | BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1); | |
104 | iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES); | |
3343660d AL |
105 | } |
106 | ||
107 | /* virtio config->get() implementation */ | |
108 | static void vp_get(struct virtio_device *vdev, unsigned offset, | |
109 | void *buf, unsigned len) | |
110 | { | |
111 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
97968358 | 112 | void __iomem *ioaddr = vp_dev->ioaddr + VIRTIO_PCI_CONFIG + offset; |
3343660d AL |
113 | u8 *ptr = buf; |
114 | int i; | |
115 | ||
116 | for (i = 0; i < len; i++) | |
117 | ptr[i] = ioread8(ioaddr + i); | |
118 | } | |
119 | ||
120 | /* the config->set() implementation. it's symmetric to the config->get() | |
121 | * implementation */ | |
122 | static void vp_set(struct virtio_device *vdev, unsigned offset, | |
123 | const void *buf, unsigned len) | |
124 | { | |
125 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
97968358 | 126 | void __iomem *ioaddr = vp_dev->ioaddr + VIRTIO_PCI_CONFIG + offset; |
3343660d AL |
127 | const u8 *ptr = buf; |
128 | int i; | |
129 | ||
130 | for (i = 0; i < len; i++) | |
131 | iowrite8(ptr[i], ioaddr + i); | |
132 | } | |
133 | ||
134 | /* config->{get,set}_status() implementations */ | |
135 | static u8 vp_get_status(struct virtio_device *vdev) | |
136 | { | |
137 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
138 | return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS); | |
139 | } | |
140 | ||
141 | static void vp_set_status(struct virtio_device *vdev, u8 status) | |
142 | { | |
143 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
144 | /* We should never be setting status to 0. */ | |
145 | BUG_ON(status == 0); | |
597d56e4 | 146 | iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS); |
3343660d AL |
147 | } |
148 | ||
149 | static void vp_reset(struct virtio_device *vdev) | |
150 | { | |
151 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
152 | /* 0 status means a reset. */ | |
597d56e4 | 153 | iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS); |
3343660d AL |
154 | } |
155 | ||
156 | /* the notify function used when creating a virt queue */ | |
157 | static void vp_notify(struct virtqueue *vq) | |
158 | { | |
159 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | |
160 | struct virtio_pci_vq_info *info = vq->priv; | |
161 | ||
162 | /* we write the queue's selector into the notification register to | |
163 | * signal the other end */ | |
164 | iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY); | |
165 | } | |
166 | ||
167 | /* A small wrapper to also acknowledge the interrupt when it's handled. | |
168 | * I really need an EIO hook for the vring so I can ack the interrupt once we | |
169 | * know that we'll be handling the IRQ but before we invoke the callback since | |
170 | * the callback may notify the host which results in the host attempting to | |
171 | * raise an interrupt that we would then mask once we acknowledged the | |
172 | * interrupt. */ | |
173 | static irqreturn_t vp_interrupt(int irq, void *opaque) | |
174 | { | |
175 | struct virtio_pci_device *vp_dev = opaque; | |
176 | struct virtio_pci_vq_info *info; | |
177 | irqreturn_t ret = IRQ_NONE; | |
27ebe308 | 178 | unsigned long flags; |
3343660d AL |
179 | u8 isr; |
180 | ||
181 | /* reading the ISR has the effect of also clearing it so it's very | |
182 | * important to save off the value. */ | |
183 | isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR); | |
184 | ||
185 | /* It's definitely not us if the ISR was not high */ | |
186 | if (!isr) | |
187 | return IRQ_NONE; | |
188 | ||
189 | /* Configuration change? Tell driver if it wants to know. */ | |
190 | if (isr & VIRTIO_PCI_ISR_CONFIG) { | |
191 | struct virtio_driver *drv; | |
192 | drv = container_of(vp_dev->vdev.dev.driver, | |
193 | struct virtio_driver, driver); | |
194 | ||
3fff0179 | 195 | if (drv && drv->config_changed) |
3343660d AL |
196 | drv->config_changed(&vp_dev->vdev); |
197 | } | |
198 | ||
27ebe308 | 199 | spin_lock_irqsave(&vp_dev->lock, flags); |
3343660d AL |
200 | list_for_each_entry(info, &vp_dev->virtqueues, node) { |
201 | if (vring_interrupt(irq, info->vq) == IRQ_HANDLED) | |
202 | ret = IRQ_HANDLED; | |
203 | } | |
27ebe308 | 204 | spin_unlock_irqrestore(&vp_dev->lock, flags); |
3343660d AL |
205 | |
206 | return ret; | |
207 | } | |
208 | ||
209 | /* the config->find_vq() implementation */ | |
210 | static struct virtqueue *vp_find_vq(struct virtio_device *vdev, unsigned index, | |
9499f5e7 RR |
211 | void (*callback)(struct virtqueue *vq), |
212 | const char *name) | |
3343660d AL |
213 | { |
214 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
215 | struct virtio_pci_vq_info *info; | |
216 | struct virtqueue *vq; | |
13b1eb33 | 217 | unsigned long flags, size; |
3343660d AL |
218 | u16 num; |
219 | int err; | |
220 | ||
221 | /* Select the queue we're interested in */ | |
222 | iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); | |
223 | ||
224 | /* Check if queue is either not available or already active. */ | |
225 | num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM); | |
226 | if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN)) | |
227 | return ERR_PTR(-ENOENT); | |
228 | ||
229 | /* allocate and fill out our structure the represents an active | |
230 | * queue */ | |
231 | info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL); | |
232 | if (!info) | |
233 | return ERR_PTR(-ENOMEM); | |
234 | ||
235 | info->queue_index = index; | |
236 | info->num = num; | |
237 | ||
498af147 | 238 | size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN)); |
13b1eb33 | 239 | info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO); |
3343660d AL |
240 | if (info->queue == NULL) { |
241 | err = -ENOMEM; | |
242 | goto out_info; | |
243 | } | |
244 | ||
245 | /* activate the queue */ | |
480daab4 | 246 | iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT, |
3343660d AL |
247 | vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); |
248 | ||
249 | /* create the vring */ | |
87c7d57c | 250 | vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN, |
9499f5e7 | 251 | vdev, info->queue, vp_notify, callback, name); |
3343660d AL |
252 | if (!vq) { |
253 | err = -ENOMEM; | |
254 | goto out_activate_queue; | |
255 | } | |
256 | ||
257 | vq->priv = info; | |
258 | info->vq = vq; | |
259 | ||
27ebe308 | 260 | spin_lock_irqsave(&vp_dev->lock, flags); |
3343660d | 261 | list_add(&info->node, &vp_dev->virtqueues); |
27ebe308 | 262 | spin_unlock_irqrestore(&vp_dev->lock, flags); |
3343660d AL |
263 | |
264 | return vq; | |
265 | ||
266 | out_activate_queue: | |
267 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); | |
13b1eb33 | 268 | free_pages_exact(info->queue, size); |
3343660d AL |
269 | out_info: |
270 | kfree(info); | |
271 | return ERR_PTR(err); | |
272 | } | |
273 | ||
274 | /* the config->del_vq() implementation */ | |
275 | static void vp_del_vq(struct virtqueue *vq) | |
276 | { | |
277 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | |
278 | struct virtio_pci_vq_info *info = vq->priv; | |
d2a7ddda | 279 | unsigned long size; |
3343660d AL |
280 | |
281 | vring_del_virtqueue(vq); | |
282 | ||
283 | /* Select and deactivate the queue */ | |
284 | iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); | |
285 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); | |
286 | ||
498af147 | 287 | size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN)); |
13b1eb33 | 288 | free_pages_exact(info->queue, size); |
3343660d AL |
289 | kfree(info); |
290 | } | |
291 | ||
d2a7ddda MT |
292 | static void vp_del_vqs(struct virtio_device *vdev) |
293 | { | |
294 | struct virtqueue *vq, *n; | |
295 | ||
296 | list_for_each_entry_safe(vq, n, &vdev->vqs, list) | |
297 | vp_del_vq(vq); | |
298 | } | |
299 | ||
300 | static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs, | |
301 | struct virtqueue *vqs[], | |
302 | vq_callback_t *callbacks[], | |
303 | const char *names[]) | |
304 | { | |
305 | int i; | |
306 | ||
307 | for (i = 0; i < nvqs; ++i) { | |
308 | vqs[i] = vp_find_vq(vdev, i, callbacks[i], names[i]); | |
309 | if (IS_ERR(vqs[i])) | |
310 | goto error; | |
311 | } | |
312 | return 0; | |
313 | ||
314 | error: | |
315 | vp_del_vqs(vdev); | |
316 | return PTR_ERR(vqs[i]); | |
317 | } | |
318 | ||
3343660d | 319 | static struct virtio_config_ops virtio_pci_config_ops = { |
3343660d AL |
320 | .get = vp_get, |
321 | .set = vp_set, | |
322 | .get_status = vp_get_status, | |
323 | .set_status = vp_set_status, | |
324 | .reset = vp_reset, | |
d2a7ddda MT |
325 | .find_vqs = vp_find_vqs, |
326 | .del_vqs = vp_del_vqs, | |
c45a6816 | 327 | .get_features = vp_get_features, |
c624896e | 328 | .finalize_features = vp_finalize_features, |
3343660d AL |
329 | }; |
330 | ||
29f9f12e MM |
331 | static void virtio_pci_release_dev(struct device *_d) |
332 | { | |
333 | struct virtio_device *dev = container_of(_d, struct virtio_device, dev); | |
334 | struct virtio_pci_device *vp_dev = to_vp_device(dev); | |
335 | struct pci_dev *pci_dev = vp_dev->pci_dev; | |
336 | ||
337 | free_irq(pci_dev->irq, vp_dev); | |
338 | pci_set_drvdata(pci_dev, NULL); | |
339 | pci_iounmap(pci_dev, vp_dev->ioaddr); | |
340 | pci_release_regions(pci_dev); | |
341 | pci_disable_device(pci_dev); | |
342 | kfree(vp_dev); | |
343 | } | |
344 | ||
3343660d AL |
345 | /* the PCI probing function */ |
346 | static int __devinit virtio_pci_probe(struct pci_dev *pci_dev, | |
347 | const struct pci_device_id *id) | |
348 | { | |
349 | struct virtio_pci_device *vp_dev; | |
350 | int err; | |
351 | ||
352 | /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */ | |
353 | if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f) | |
354 | return -ENODEV; | |
355 | ||
55a7c066 AL |
356 | if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) { |
357 | printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n", | |
358 | VIRTIO_PCI_ABI_VERSION, pci_dev->revision); | |
359 | return -ENODEV; | |
360 | } | |
361 | ||
3343660d AL |
362 | /* allocate our structure and fill it out */ |
363 | vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL); | |
364 | if (vp_dev == NULL) | |
365 | return -ENOMEM; | |
366 | ||
63d12556 | 367 | vp_dev->vdev.dev.parent = virtio_pci_root; |
29f9f12e | 368 | vp_dev->vdev.dev.release = virtio_pci_release_dev; |
3343660d AL |
369 | vp_dev->vdev.config = &virtio_pci_config_ops; |
370 | vp_dev->pci_dev = pci_dev; | |
371 | INIT_LIST_HEAD(&vp_dev->virtqueues); | |
372 | spin_lock_init(&vp_dev->lock); | |
373 | ||
374 | /* enable the device */ | |
375 | err = pci_enable_device(pci_dev); | |
376 | if (err) | |
377 | goto out; | |
378 | ||
379 | err = pci_request_regions(pci_dev, "virtio-pci"); | |
380 | if (err) | |
381 | goto out_enable_device; | |
382 | ||
383 | vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0); | |
384 | if (vp_dev->ioaddr == NULL) | |
385 | goto out_req_regions; | |
386 | ||
387 | pci_set_drvdata(pci_dev, vp_dev); | |
388 | ||
389 | /* we use the subsystem vendor/device id as the virtio vendor/device | |
390 | * id. this allows us to use the same PCI vendor/device id for all | |
391 | * virtio devices and to identify the particular virtio driver by | |
392 | * the subsytem ids */ | |
393 | vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor; | |
394 | vp_dev->vdev.id.device = pci_dev->subsystem_device; | |
395 | ||
396 | /* register a handler for the queue with the PCI device's interrupt */ | |
397 | err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED, | |
99e0b6c8 | 398 | dev_name(&vp_dev->vdev.dev), vp_dev); |
3343660d AL |
399 | if (err) |
400 | goto out_set_drvdata; | |
401 | ||
402 | /* finally register the virtio device */ | |
403 | err = register_virtio_device(&vp_dev->vdev); | |
404 | if (err) | |
405 | goto out_req_irq; | |
406 | ||
407 | return 0; | |
408 | ||
409 | out_req_irq: | |
410 | free_irq(pci_dev->irq, vp_dev); | |
411 | out_set_drvdata: | |
412 | pci_set_drvdata(pci_dev, NULL); | |
413 | pci_iounmap(pci_dev, vp_dev->ioaddr); | |
414 | out_req_regions: | |
415 | pci_release_regions(pci_dev); | |
416 | out_enable_device: | |
417 | pci_disable_device(pci_dev); | |
418 | out: | |
419 | kfree(vp_dev); | |
420 | return err; | |
421 | } | |
422 | ||
423 | static void __devexit virtio_pci_remove(struct pci_dev *pci_dev) | |
424 | { | |
425 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
426 | ||
bd6c2690 | 427 | unregister_virtio_device(&vp_dev->vdev); |
3343660d AL |
428 | } |
429 | ||
430 | #ifdef CONFIG_PM | |
431 | static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state) | |
432 | { | |
433 | pci_save_state(pci_dev); | |
434 | pci_set_power_state(pci_dev, PCI_D3hot); | |
435 | return 0; | |
436 | } | |
437 | ||
438 | static int virtio_pci_resume(struct pci_dev *pci_dev) | |
439 | { | |
440 | pci_restore_state(pci_dev); | |
441 | pci_set_power_state(pci_dev, PCI_D0); | |
442 | return 0; | |
443 | } | |
444 | #endif | |
445 | ||
446 | static struct pci_driver virtio_pci_driver = { | |
447 | .name = "virtio-pci", | |
448 | .id_table = virtio_pci_id_table, | |
449 | .probe = virtio_pci_probe, | |
450 | .remove = virtio_pci_remove, | |
451 | #ifdef CONFIG_PM | |
452 | .suspend = virtio_pci_suspend, | |
453 | .resume = virtio_pci_resume, | |
454 | #endif | |
455 | }; | |
456 | ||
457 | static int __init virtio_pci_init(void) | |
458 | { | |
459 | int err; | |
460 | ||
63d12556 MM |
461 | virtio_pci_root = root_device_register("virtio-pci"); |
462 | if (IS_ERR(virtio_pci_root)) | |
463 | return PTR_ERR(virtio_pci_root); | |
3343660d AL |
464 | |
465 | err = pci_register_driver(&virtio_pci_driver); | |
466 | if (err) | |
63d12556 | 467 | device_unregister(virtio_pci_root); |
3343660d AL |
468 | |
469 | return err; | |
470 | } | |
471 | ||
472 | module_init(virtio_pci_init); | |
473 | ||
474 | static void __exit virtio_pci_exit(void) | |
475 | { | |
3343660d | 476 | pci_unregister_driver(&virtio_pci_driver); |
63d12556 | 477 | root_device_unregister(virtio_pci_root); |
3343660d AL |
478 | } |
479 | ||
480 | module_exit(virtio_pci_exit); |