diff options
author | Anthony Liguori <aliguori@us.ibm.com> | 2007-11-12 22:30:26 -0500 |
---|---|---|
committer | Rusty Russell <rusty@rustcorp.com.au> | 2008-02-04 07:50:11 -0500 |
commit | 3343660d8c62c6b00b2f15324ef3fcb6be207bfa (patch) | |
tree | 06d86446da65fc1814edad944e43aeb62f092422 /drivers | |
parent | d50ed907dc3db5bf2dd0a05b4e199a65793a3788 (diff) |
virtio: PCI device
This is a PCI device that implements a transport for virtio. It allows virtio
devices to be used by QEMU based VMMs like KVM or Xen.
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Diffstat (limited to 'drivers')
-rw-r--r-- | drivers/virtio/Kconfig | 17 | ||||
-rw-r--r-- | drivers/virtio/Makefile | 1 | ||||
-rw-r--r-- | drivers/virtio/virtio_pci.c | 440 |
3 files changed, 458 insertions, 0 deletions
diff --git a/drivers/virtio/Kconfig b/drivers/virtio/Kconfig index de0c8c2654e..833db2f36e9 100644 --- a/drivers/virtio/Kconfig +++ b/drivers/virtio/Kconfig | |||
@@ -6,3 +6,20 @@ config VIRTIO | |||
6 | config VIRTIO_RING | 6 | config VIRTIO_RING |
7 | tristate | 7 | tristate |
8 | depends on VIRTIO | 8 | depends on VIRTIO |
9 | |||
10 | config VIRTIO_PCI | ||
11 | tristate "PCI driver for virtio devices (EXPERIMENTAL)" | ||
12 | depends on PCI && EXPERIMENTAL | ||
13 | select VIRTIO | ||
14 | select VIRTIO_RING | ||
15 | ---help--- | ||
16 | This drivers provides support for virtio based paravirtual device | ||
17 | drivers over PCI. This requires that your VMM has appropriate PCI | ||
18 | virtio backends. Most QEMU based VMMs should support these devices | ||
19 | (like KVM or Xen). | ||
20 | |||
21 | Currently, the ABI is not considered stable so there is no guarantee | ||
22 | that this version of the driver will work with your VMM. | ||
23 | |||
24 | If unsure, say M. | ||
25 | |||
diff --git a/drivers/virtio/Makefile b/drivers/virtio/Makefile index f70e40971dd..cc84999f305 100644 --- a/drivers/virtio/Makefile +++ b/drivers/virtio/Makefile | |||
@@ -1,2 +1,3 @@ | |||
1 | obj-$(CONFIG_VIRTIO) += virtio.o | 1 | obj-$(CONFIG_VIRTIO) += virtio.o |
2 | obj-$(CONFIG_VIRTIO_RING) += virtio_ring.o | 2 | obj-$(CONFIG_VIRTIO_RING) += virtio_ring.o |
3 | obj-$(CONFIG_VIRTIO_PCI) += virtio_pci.o | ||
diff --git a/drivers/virtio/virtio_pci.c b/drivers/virtio/virtio_pci.c new file mode 100644 index 00000000000..192687e3a56 --- /dev/null +++ b/drivers/virtio/virtio_pci.c | |||
@@ -0,0 +1,440 @@ | |||
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 */ | ||
40 | void *ioaddr; | ||
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, | ||
78 | .bus_id = "virtio-pci", | ||
79 | }; | ||
80 | |||
81 | /* Unique numbering for devices under the kvm root */ | ||
82 | static unsigned int dev_index; | ||
83 | |||
84 | /* Convert a generic virtio device to our structure */ | ||
85 | static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev) | ||
86 | { | ||
87 | return container_of(vdev, struct virtio_pci_device, vdev); | ||
88 | } | ||
89 | |||
90 | /* virtio config->feature() implementation */ | ||
91 | static bool vp_feature(struct virtio_device *vdev, unsigned bit) | ||
92 | { | ||
93 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | ||
94 | u32 mask; | ||
95 | |||
96 | /* Since this function is supposed to have the side effect of | ||
97 | * enabling a queried feature, we simulate that by doing a read | ||
98 | * from the host feature bitmask and then writing to the guest | ||
99 | * feature bitmask */ | ||
100 | mask = ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES); | ||
101 | if (mask & (1 << bit)) { | ||
102 | mask |= (1 << bit); | ||
103 | iowrite32(mask, vp_dev->ioaddr + VIRTIO_PCI_GUEST_FEATURES); | ||
104 | } | ||
105 | |||
106 | return !!(mask & (1 << bit)); | ||
107 | } | ||
108 | |||
109 | /* virtio config->get() implementation */ | ||
110 | static void vp_get(struct virtio_device *vdev, unsigned offset, | ||
111 | void *buf, unsigned len) | ||
112 | { | ||
113 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | ||
114 | void *ioaddr = vp_dev->ioaddr + VIRTIO_PCI_CONFIG + offset; | ||
115 | u8 *ptr = buf; | ||
116 | int i; | ||
117 | |||
118 | for (i = 0; i < len; i++) | ||
119 | ptr[i] = ioread8(ioaddr + i); | ||
120 | } | ||
121 | |||
122 | /* the config->set() implementation. it's symmetric to the config->get() | ||
123 | * implementation */ | ||
124 | static void vp_set(struct virtio_device *vdev, unsigned offset, | ||
125 | const void *buf, unsigned len) | ||
126 | { | ||
127 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | ||
128 | void *ioaddr = vp_dev->ioaddr + VIRTIO_PCI_CONFIG + offset; | ||
129 | const u8 *ptr = buf; | ||
130 | int i; | ||
131 | |||
132 | for (i = 0; i < len; i++) | ||
133 | iowrite8(ptr[i], ioaddr + i); | ||
134 | } | ||
135 | |||
136 | /* config->{get,set}_status() implementations */ | ||
137 | static u8 vp_get_status(struct virtio_device *vdev) | ||
138 | { | ||
139 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | ||
140 | return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS); | ||
141 | } | ||
142 | |||
143 | static void vp_set_status(struct virtio_device *vdev, u8 status) | ||
144 | { | ||
145 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | ||
146 | /* We should never be setting status to 0. */ | ||
147 | BUG_ON(status == 0); | ||
148 | return iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS); | ||
149 | } | ||
150 | |||
151 | static void vp_reset(struct virtio_device *vdev) | ||
152 | { | ||
153 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | ||
154 | /* 0 status means a reset. */ | ||
155 | return iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS); | ||
156 | } | ||
157 | |||
158 | /* the notify function used when creating a virt queue */ | ||
159 | static void vp_notify(struct virtqueue *vq) | ||
160 | { | ||
161 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | ||
162 | struct virtio_pci_vq_info *info = vq->priv; | ||
163 | |||
164 | /* we write the queue's selector into the notification register to | ||
165 | * signal the other end */ | ||
166 | iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY); | ||
167 | } | ||
168 | |||
169 | /* A small wrapper to also acknowledge the interrupt when it's handled. | ||
170 | * I really need an EIO hook for the vring so I can ack the interrupt once we | ||
171 | * know that we'll be handling the IRQ but before we invoke the callback since | ||
172 | * the callback may notify the host which results in the host attempting to | ||
173 | * raise an interrupt that we would then mask once we acknowledged the | ||
174 | * interrupt. */ | ||
175 | static irqreturn_t vp_interrupt(int irq, void *opaque) | ||
176 | { | ||
177 | struct virtio_pci_device *vp_dev = opaque; | ||
178 | struct virtio_pci_vq_info *info; | ||
179 | irqreturn_t ret = IRQ_NONE; | ||
180 | u8 isr; | ||
181 | |||
182 | /* reading the ISR has the effect of also clearing it so it's very | ||
183 | * important to save off the value. */ | ||
184 | isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR); | ||
185 | |||
186 | /* It's definitely not us if the ISR was not high */ | ||
187 | if (!isr) | ||
188 | return IRQ_NONE; | ||
189 | |||
190 | /* Configuration change? Tell driver if it wants to know. */ | ||
191 | if (isr & VIRTIO_PCI_ISR_CONFIG) { | ||
192 | struct virtio_driver *drv; | ||
193 | drv = container_of(vp_dev->vdev.dev.driver, | ||
194 | struct virtio_driver, driver); | ||
195 | |||
196 | if (drv->config_changed) | ||
197 | drv->config_changed(&vp_dev->vdev); | ||
198 | } | ||
199 | |||
200 | spin_lock(&vp_dev->lock); | ||
201 | list_for_each_entry(info, &vp_dev->virtqueues, node) { | ||
202 | if (vring_interrupt(irq, info->vq) == IRQ_HANDLED) | ||
203 | ret = IRQ_HANDLED; | ||
204 | } | ||
205 | spin_unlock(&vp_dev->lock); | ||
206 | |||
207 | return ret; | ||
208 | } | ||
209 | |||
210 | /* the config->find_vq() implementation */ | ||
211 | static struct virtqueue *vp_find_vq(struct virtio_device *vdev, unsigned index, | ||
212 | void (*callback)(struct virtqueue *vq)) | ||
213 | { | ||
214 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | ||
215 | struct virtio_pci_vq_info *info; | ||
216 | struct virtqueue *vq; | ||
217 | u16 num; | ||
218 | int err; | ||
219 | |||
220 | /* Select the queue we're interested in */ | ||
221 | iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); | ||
222 | |||
223 | /* Check if queue is either not available or already active. */ | ||
224 | num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM); | ||
225 | if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN)) | ||
226 | return ERR_PTR(-ENOENT); | ||
227 | |||
228 | /* allocate and fill out our structure the represents an active | ||
229 | * queue */ | ||
230 | info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL); | ||
231 | if (!info) | ||
232 | return ERR_PTR(-ENOMEM); | ||
233 | |||
234 | info->queue_index = index; | ||
235 | info->num = num; | ||
236 | |||
237 | info->queue = kzalloc(PAGE_ALIGN(vring_size(num,PAGE_SIZE)), GFP_KERNEL); | ||
238 | if (info->queue == NULL) { | ||
239 | err = -ENOMEM; | ||
240 | goto out_info; | ||
241 | } | ||
242 | |||
243 | /* activate the queue */ | ||
244 | iowrite32(virt_to_phys(info->queue) >> PAGE_SHIFT, | ||
245 | vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); | ||
246 | |||
247 | /* create the vring */ | ||
248 | vq = vring_new_virtqueue(info->num, vdev, info->queue, | ||
249 | vp_notify, callback); | ||
250 | if (!vq) { | ||
251 | err = -ENOMEM; | ||
252 | goto out_activate_queue; | ||
253 | } | ||
254 | |||
255 | vq->priv = info; | ||
256 | info->vq = vq; | ||
257 | |||
258 | spin_lock(&vp_dev->lock); | ||
259 | list_add(&info->node, &vp_dev->virtqueues); | ||
260 | spin_unlock(&vp_dev->lock); | ||
261 | |||
262 | return vq; | ||
263 | |||
264 | out_activate_queue: | ||
265 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); | ||
266 | kfree(info->queue); | ||
267 | out_info: | ||
268 | kfree(info); | ||
269 | return ERR_PTR(err); | ||
270 | } | ||
271 | |||
272 | /* the config->del_vq() implementation */ | ||
273 | static void vp_del_vq(struct virtqueue *vq) | ||
274 | { | ||
275 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | ||
276 | struct virtio_pci_vq_info *info = vq->priv; | ||
277 | |||
278 | spin_lock(&vp_dev->lock); | ||
279 | list_del(&info->node); | ||
280 | spin_unlock(&vp_dev->lock); | ||
281 | |||
282 | vring_del_virtqueue(vq); | ||
283 | |||
284 | /* Select and deactivate the queue */ | ||
285 | iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); | ||
286 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); | ||
287 | |||
288 | kfree(info->queue); | ||
289 | kfree(info); | ||
290 | } | ||
291 | |||
292 | static struct virtio_config_ops virtio_pci_config_ops = { | ||
293 | .feature = vp_feature, | ||
294 | .get = vp_get, | ||
295 | .set = vp_set, | ||
296 | .get_status = vp_get_status, | ||
297 | .set_status = vp_set_status, | ||
298 | .reset = vp_reset, | ||
299 | .find_vq = vp_find_vq, | ||
300 | .del_vq = vp_del_vq, | ||
301 | }; | ||
302 | |||
303 | /* the PCI probing function */ | ||
304 | static int __devinit virtio_pci_probe(struct pci_dev *pci_dev, | ||
305 | const struct pci_device_id *id) | ||
306 | { | ||
307 | struct virtio_pci_device *vp_dev; | ||
308 | int err; | ||
309 | |||
310 | /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */ | ||
311 | if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f) | ||
312 | return -ENODEV; | ||
313 | |||
314 | /* allocate our structure and fill it out */ | ||
315 | vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL); | ||
316 | if (vp_dev == NULL) | ||
317 | return -ENOMEM; | ||
318 | |||
319 | snprintf(vp_dev->vdev.dev.bus_id, BUS_ID_SIZE, "virtio%d", dev_index); | ||
320 | vp_dev->vdev.index = dev_index; | ||
321 | dev_index++; | ||
322 | |||
323 | vp_dev->vdev.dev.parent = &virtio_pci_root; | ||
324 | vp_dev->vdev.config = &virtio_pci_config_ops; | ||
325 | vp_dev->pci_dev = pci_dev; | ||
326 | INIT_LIST_HEAD(&vp_dev->virtqueues); | ||
327 | spin_lock_init(&vp_dev->lock); | ||
328 | |||
329 | /* enable the device */ | ||
330 | err = pci_enable_device(pci_dev); | ||
331 | if (err) | ||
332 | goto out; | ||
333 | |||
334 | err = pci_request_regions(pci_dev, "virtio-pci"); | ||
335 | if (err) | ||
336 | goto out_enable_device; | ||
337 | |||
338 | vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0); | ||
339 | if (vp_dev->ioaddr == NULL) | ||
340 | goto out_req_regions; | ||
341 | |||
342 | pci_set_drvdata(pci_dev, vp_dev); | ||
343 | |||
344 | /* we use the subsystem vendor/device id as the virtio vendor/device | ||
345 | * id. this allows us to use the same PCI vendor/device id for all | ||
346 | * virtio devices and to identify the particular virtio driver by | ||
347 | * the subsytem ids */ | ||
348 | vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor; | ||
349 | vp_dev->vdev.id.device = pci_dev->subsystem_device; | ||
350 | |||
351 | /* register a handler for the queue with the PCI device's interrupt */ | ||
352 | err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED, | ||
353 | vp_dev->vdev.dev.bus_id, vp_dev); | ||
354 | if (err) | ||
355 | goto out_set_drvdata; | ||
356 | |||
357 | /* finally register the virtio device */ | ||
358 | err = register_virtio_device(&vp_dev->vdev); | ||
359 | if (err) | ||
360 | goto out_req_irq; | ||
361 | |||
362 | return 0; | ||
363 | |||
364 | out_req_irq: | ||
365 | free_irq(pci_dev->irq, vp_dev); | ||
366 | out_set_drvdata: | ||
367 | pci_set_drvdata(pci_dev, NULL); | ||
368 | pci_iounmap(pci_dev, vp_dev->ioaddr); | ||
369 | out_req_regions: | ||
370 | pci_release_regions(pci_dev); | ||
371 | out_enable_device: | ||
372 | pci_disable_device(pci_dev); | ||
373 | out: | ||
374 | kfree(vp_dev); | ||
375 | return err; | ||
376 | } | ||
377 | |||
378 | static void __devexit virtio_pci_remove(struct pci_dev *pci_dev) | ||
379 | { | ||
380 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | ||
381 | |||
382 | free_irq(pci_dev->irq, vp_dev); | ||
383 | pci_set_drvdata(pci_dev, NULL); | ||
384 | pci_iounmap(pci_dev, vp_dev->ioaddr); | ||
385 | pci_release_regions(pci_dev); | ||
386 | pci_disable_device(pci_dev); | ||
387 | kfree(vp_dev); | ||
388 | } | ||
389 | |||
390 | #ifdef CONFIG_PM | ||
391 | static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state) | ||
392 | { | ||
393 | pci_save_state(pci_dev); | ||
394 | pci_set_power_state(pci_dev, PCI_D3hot); | ||
395 | return 0; | ||
396 | } | ||
397 | |||
398 | static int virtio_pci_resume(struct pci_dev *pci_dev) | ||
399 | { | ||
400 | pci_restore_state(pci_dev); | ||
401 | pci_set_power_state(pci_dev, PCI_D0); | ||
402 | return 0; | ||
403 | } | ||
404 | #endif | ||
405 | |||
406 | static struct pci_driver virtio_pci_driver = { | ||
407 | .name = "virtio-pci", | ||
408 | .id_table = virtio_pci_id_table, | ||
409 | .probe = virtio_pci_probe, | ||
410 | .remove = virtio_pci_remove, | ||
411 | #ifdef CONFIG_PM | ||
412 | .suspend = virtio_pci_suspend, | ||
413 | .resume = virtio_pci_resume, | ||
414 | #endif | ||
415 | }; | ||
416 | |||
417 | static int __init virtio_pci_init(void) | ||
418 | { | ||
419 | int err; | ||
420 | |||
421 | err = device_register(&virtio_pci_root); | ||
422 | if (err) | ||
423 | return err; | ||
424 | |||
425 | err = pci_register_driver(&virtio_pci_driver); | ||
426 | if (err) | ||
427 | device_unregister(&virtio_pci_root); | ||
428 | |||
429 | return err; | ||
430 | } | ||
431 | |||
432 | module_init(virtio_pci_init); | ||
433 | |||
434 | static void __exit virtio_pci_exit(void) | ||
435 | { | ||
436 | device_unregister(&virtio_pci_root); | ||
437 | pci_unregister_driver(&virtio_pci_driver); | ||
438 | } | ||
439 | |||
440 | module_exit(virtio_pci_exit); | ||