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