]>
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 | ||
82 | /* the index of the queue */ | |
83 | int queue_index; | |
84 | ||
85 | /* the virtual address of the ring queue */ | |
86 | void *queue; | |
87 | ||
88 | /* the list node for the virtqueues list */ | |
89 | struct list_head node; | |
82af8ce8 MT |
90 | |
91 | /* MSI-X vector (or none) */ | |
f68d2408 | 92 | unsigned msix_vector; |
3343660d AL |
93 | }; |
94 | ||
95 | /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */ | |
96 | static struct pci_device_id virtio_pci_id_table[] = { | |
97 | { 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 }, | |
98 | { 0 }, | |
99 | }; | |
100 | ||
101 | MODULE_DEVICE_TABLE(pci, virtio_pci_id_table); | |
102 | ||
3343660d AL |
103 | /* Convert a generic virtio device to our structure */ |
104 | static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev) | |
105 | { | |
106 | return container_of(vdev, struct virtio_pci_device, vdev); | |
107 | } | |
108 | ||
c45a6816 RR |
109 | /* virtio config->get_features() implementation */ |
110 | static u32 vp_get_features(struct virtio_device *vdev) | |
111 | { | |
112 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
113 | ||
114 | /* When someone needs more than 32 feature bits, we'll need to | |
115 | * steal a bit to indicate that the rest are somewhere else. */ | |
116 | return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES); | |
117 | } | |
118 | ||
c624896e RR |
119 | /* virtio config->finalize_features() implementation */ |
120 | static void vp_finalize_features(struct virtio_device *vdev) | |
3343660d AL |
121 | { |
122 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
3343660d | 123 | |
e34f8725 RR |
124 | /* Give virtio_ring a chance to accept features. */ |
125 | vring_transport_features(vdev); | |
126 | ||
c624896e RR |
127 | /* We only support 32 feature bits. */ |
128 | BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1); | |
129 | iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES); | |
3343660d AL |
130 | } |
131 | ||
132 | /* virtio config->get() implementation */ | |
133 | static void vp_get(struct virtio_device *vdev, unsigned offset, | |
134 | void *buf, unsigned len) | |
135 | { | |
136 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
82af8ce8 MT |
137 | void __iomem *ioaddr = vp_dev->ioaddr + |
138 | VIRTIO_PCI_CONFIG(vp_dev) + offset; | |
3343660d AL |
139 | u8 *ptr = buf; |
140 | int i; | |
141 | ||
142 | for (i = 0; i < len; i++) | |
143 | ptr[i] = ioread8(ioaddr + i); | |
144 | } | |
145 | ||
146 | /* the config->set() implementation. it's symmetric to the config->get() | |
147 | * implementation */ | |
148 | static void vp_set(struct virtio_device *vdev, unsigned offset, | |
149 | const void *buf, unsigned len) | |
150 | { | |
151 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
82af8ce8 MT |
152 | void __iomem *ioaddr = vp_dev->ioaddr + |
153 | VIRTIO_PCI_CONFIG(vp_dev) + offset; | |
3343660d AL |
154 | const u8 *ptr = buf; |
155 | int i; | |
156 | ||
157 | for (i = 0; i < len; i++) | |
158 | iowrite8(ptr[i], ioaddr + i); | |
159 | } | |
160 | ||
161 | /* config->{get,set}_status() implementations */ | |
162 | static u8 vp_get_status(struct virtio_device *vdev) | |
163 | { | |
164 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
165 | return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS); | |
166 | } | |
167 | ||
168 | static void vp_set_status(struct virtio_device *vdev, u8 status) | |
169 | { | |
170 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
171 | /* We should never be setting status to 0. */ | |
172 | BUG_ON(status == 0); | |
597d56e4 | 173 | iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS); |
3343660d AL |
174 | } |
175 | ||
e6af578c MT |
176 | /* wait for pending irq handlers */ |
177 | static void vp_synchronize_vectors(struct virtio_device *vdev) | |
178 | { | |
179 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
180 | int i; | |
181 | ||
182 | if (vp_dev->intx_enabled) | |
183 | synchronize_irq(vp_dev->pci_dev->irq); | |
184 | ||
185 | for (i = 0; i < vp_dev->msix_vectors; ++i) | |
186 | synchronize_irq(vp_dev->msix_entries[i].vector); | |
187 | } | |
188 | ||
3343660d AL |
189 | static void vp_reset(struct virtio_device *vdev) |
190 | { | |
191 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
192 | /* 0 status means a reset. */ | |
597d56e4 | 193 | iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS); |
e6af578c MT |
194 | /* Flush out the status write, and flush in device writes, |
195 | * including MSi-X interrupts, if any. */ | |
196 | ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS); | |
197 | /* Flush pending VQ/configuration callbacks. */ | |
198 | vp_synchronize_vectors(vdev); | |
3343660d AL |
199 | } |
200 | ||
201 | /* the notify function used when creating a virt queue */ | |
202 | static void vp_notify(struct virtqueue *vq) | |
203 | { | |
204 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | |
205 | struct virtio_pci_vq_info *info = vq->priv; | |
206 | ||
207 | /* we write the queue's selector into the notification register to | |
208 | * signal the other end */ | |
209 | iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY); | |
210 | } | |
211 | ||
77cf5246 MT |
212 | /* Handle a configuration change: Tell driver if it wants to know. */ |
213 | static irqreturn_t vp_config_changed(int irq, void *opaque) | |
214 | { | |
215 | struct virtio_pci_device *vp_dev = opaque; | |
216 | struct virtio_driver *drv; | |
217 | drv = container_of(vp_dev->vdev.dev.driver, | |
218 | struct virtio_driver, driver); | |
219 | ||
220 | if (drv && drv->config_changed) | |
221 | drv->config_changed(&vp_dev->vdev); | |
222 | return IRQ_HANDLED; | |
223 | } | |
224 | ||
225 | /* Notify all virtqueues on an interrupt. */ | |
226 | static irqreturn_t vp_vring_interrupt(int irq, void *opaque) | |
227 | { | |
228 | struct virtio_pci_device *vp_dev = opaque; | |
229 | struct virtio_pci_vq_info *info; | |
230 | irqreturn_t ret = IRQ_NONE; | |
231 | unsigned long flags; | |
232 | ||
233 | spin_lock_irqsave(&vp_dev->lock, flags); | |
234 | list_for_each_entry(info, &vp_dev->virtqueues, node) { | |
235 | if (vring_interrupt(irq, info->vq) == IRQ_HANDLED) | |
236 | ret = IRQ_HANDLED; | |
237 | } | |
238 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
239 | ||
240 | return ret; | |
241 | } | |
242 | ||
3343660d AL |
243 | /* A small wrapper to also acknowledge the interrupt when it's handled. |
244 | * I really need an EIO hook for the vring so I can ack the interrupt once we | |
245 | * know that we'll be handling the IRQ but before we invoke the callback since | |
246 | * the callback may notify the host which results in the host attempting to | |
247 | * raise an interrupt that we would then mask once we acknowledged the | |
248 | * interrupt. */ | |
249 | static irqreturn_t vp_interrupt(int irq, void *opaque) | |
250 | { | |
251 | struct virtio_pci_device *vp_dev = opaque; | |
3343660d AL |
252 | u8 isr; |
253 | ||
254 | /* reading the ISR has the effect of also clearing it so it's very | |
255 | * important to save off the value. */ | |
256 | isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR); | |
257 | ||
258 | /* It's definitely not us if the ISR was not high */ | |
259 | if (!isr) | |
260 | return IRQ_NONE; | |
261 | ||
262 | /* Configuration change? Tell driver if it wants to know. */ | |
77cf5246 MT |
263 | if (isr & VIRTIO_PCI_ISR_CONFIG) |
264 | vp_config_changed(irq, opaque); | |
3343660d | 265 | |
77cf5246 | 266 | return vp_vring_interrupt(irq, opaque); |
3343660d AL |
267 | } |
268 | ||
82af8ce8 MT |
269 | static void vp_free_vectors(struct virtio_device *vdev) |
270 | { | |
271 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
272 | int i; | |
273 | ||
274 | if (vp_dev->intx_enabled) { | |
275 | free_irq(vp_dev->pci_dev->irq, vp_dev); | |
276 | vp_dev->intx_enabled = 0; | |
277 | } | |
278 | ||
279 | for (i = 0; i < vp_dev->msix_used_vectors; ++i) | |
280 | free_irq(vp_dev->msix_entries[i].vector, vp_dev); | |
82af8ce8 MT |
281 | |
282 | if (vp_dev->msix_enabled) { | |
283 | /* Disable the vector used for configuration */ | |
284 | iowrite16(VIRTIO_MSI_NO_VECTOR, | |
285 | vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
286 | /* Flush the write out to device */ | |
287 | ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
288 | ||
82af8ce8 | 289 | pci_disable_msix(vp_dev->pci_dev); |
ff52c3fc MT |
290 | vp_dev->msix_enabled = 0; |
291 | vp_dev->msix_vectors = 0; | |
82af8ce8 | 292 | } |
ff52c3fc MT |
293 | |
294 | vp_dev->msix_used_vectors = 0; | |
295 | kfree(vp_dev->msix_names); | |
296 | vp_dev->msix_names = NULL; | |
297 | kfree(vp_dev->msix_entries); | |
298 | vp_dev->msix_entries = NULL; | |
82af8ce8 MT |
299 | } |
300 | ||
f68d2408 RR |
301 | static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors, |
302 | bool per_vq_vectors) | |
82af8ce8 MT |
303 | { |
304 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
305 | const char *name = dev_name(&vp_dev->vdev.dev); | |
306 | unsigned i, v; | |
307 | int err = -ENOMEM; | |
e969fed5 | 308 | |
82af8ce8 MT |
309 | vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries, |
310 | GFP_KERNEL); | |
311 | if (!vp_dev->msix_entries) | |
ff52c3fc | 312 | goto error; |
82af8ce8 MT |
313 | vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names, |
314 | GFP_KERNEL); | |
315 | if (!vp_dev->msix_names) | |
ff52c3fc | 316 | goto error; |
82af8ce8 MT |
317 | |
318 | for (i = 0; i < nvectors; ++i) | |
319 | vp_dev->msix_entries[i].entry = i; | |
320 | ||
f68d2408 | 321 | /* pci_enable_msix returns positive if we can't get this many. */ |
e969fed5 MT |
322 | err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors); |
323 | if (err > 0) | |
324 | err = -ENOSPC; | |
325 | if (err) | |
326 | goto error; | |
327 | vp_dev->msix_vectors = nvectors; | |
328 | vp_dev->msix_enabled = 1; | |
329 | ||
330 | /* Set the vector used for configuration */ | |
331 | v = vp_dev->msix_used_vectors; | |
332 | snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names, | |
333 | "%s-config", name); | |
334 | err = request_irq(vp_dev->msix_entries[v].vector, | |
335 | vp_config_changed, 0, vp_dev->msix_names[v], | |
336 | vp_dev); | |
337 | if (err) | |
338 | goto error; | |
339 | ++vp_dev->msix_used_vectors; | |
82af8ce8 | 340 | |
e969fed5 MT |
341 | iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); |
342 | /* Verify we had enough resources to assign the vector */ | |
343 | v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
344 | if (v == VIRTIO_MSI_NO_VECTOR) { | |
345 | err = -EBUSY; | |
346 | goto error; | |
82af8ce8 MT |
347 | } |
348 | ||
e969fed5 | 349 | if (!per_vq_vectors) { |
82af8ce8 MT |
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 | ||
f68d2408 RR |
367 | static int vp_request_intx(struct virtio_device *vdev) |
368 | { | |
369 | int err; | |
370 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
371 | ||
372 | err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, | |
373 | IRQF_SHARED, dev_name(&vdev->dev), vp_dev); | |
374 | if (!err) | |
375 | vp_dev->intx_enabled = 1; | |
376 | return err; | |
377 | } | |
378 | ||
379 | static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index, | |
380 | void (*callback)(struct virtqueue *vq), | |
381 | const char *name, | |
382 | u16 msix_vec) | |
3343660d AL |
383 | { |
384 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
385 | struct virtio_pci_vq_info *info; | |
386 | struct virtqueue *vq; | |
13b1eb33 | 387 | unsigned long flags, size; |
e969fed5 | 388 | u16 num; |
3343660d AL |
389 | int err; |
390 | ||
391 | /* Select the queue we're interested in */ | |
392 | iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); | |
393 | ||
394 | /* Check if queue is either not available or already active. */ | |
395 | num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM); | |
396 | if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN)) | |
397 | return ERR_PTR(-ENOENT); | |
398 | ||
399 | /* allocate and fill out our structure the represents an active | |
400 | * queue */ | |
401 | info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL); | |
402 | if (!info) | |
403 | return ERR_PTR(-ENOMEM); | |
404 | ||
405 | info->queue_index = index; | |
406 | info->num = num; | |
f68d2408 | 407 | info->msix_vector = msix_vec; |
3343660d | 408 | |
498af147 | 409 | size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN)); |
13b1eb33 | 410 | info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO); |
3343660d AL |
411 | if (info->queue == NULL) { |
412 | err = -ENOMEM; | |
413 | goto out_info; | |
414 | } | |
415 | ||
416 | /* activate the queue */ | |
480daab4 | 417 | iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT, |
3343660d AL |
418 | vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); |
419 | ||
420 | /* create the vring */ | |
7b21e34f RR |
421 | vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN, vdev, |
422 | true, info->queue, vp_notify, callback, name); | |
3343660d AL |
423 | if (!vq) { |
424 | err = -ENOMEM; | |
425 | goto out_activate_queue; | |
426 | } | |
427 | ||
428 | vq->priv = info; | |
429 | info->vq = vq; | |
430 | ||
f68d2408 RR |
431 | if (msix_vec != VIRTIO_MSI_NO_VECTOR) { |
432 | iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
433 | msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
434 | if (msix_vec == VIRTIO_MSI_NO_VECTOR) { | |
82af8ce8 MT |
435 | err = -EBUSY; |
436 | goto out_assign; | |
437 | } | |
438 | } | |
439 | ||
005b20a8 KK |
440 | if (callback) { |
441 | spin_lock_irqsave(&vp_dev->lock, flags); | |
442 | list_add(&info->node, &vp_dev->virtqueues); | |
443 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
444 | } else { | |
445 | INIT_LIST_HEAD(&info->node); | |
446 | } | |
3343660d AL |
447 | |
448 | return vq; | |
449 | ||
82af8ce8 | 450 | out_assign: |
82af8ce8 | 451 | vring_del_virtqueue(vq); |
3343660d AL |
452 | out_activate_queue: |
453 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); | |
13b1eb33 | 454 | free_pages_exact(info->queue, size); |
3343660d AL |
455 | out_info: |
456 | kfree(info); | |
457 | return ERR_PTR(err); | |
458 | } | |
459 | ||
3343660d AL |
460 | static void vp_del_vq(struct virtqueue *vq) |
461 | { | |
462 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | |
463 | struct virtio_pci_vq_info *info = vq->priv; | |
f6c82507 MT |
464 | unsigned long flags, size; |
465 | ||
466 | spin_lock_irqsave(&vp_dev->lock, flags); | |
467 | list_del(&info->node); | |
468 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
3343660d | 469 | |
82af8ce8 MT |
470 | iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); |
471 | ||
82af8ce8 MT |
472 | if (vp_dev->msix_enabled) { |
473 | iowrite16(VIRTIO_MSI_NO_VECTOR, | |
474 | vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
475 | /* Flush the write out to device */ | |
476 | ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR); | |
477 | } | |
478 | ||
3343660d AL |
479 | vring_del_virtqueue(vq); |
480 | ||
481 | /* Select and deactivate the queue */ | |
3343660d AL |
482 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); |
483 | ||
498af147 | 484 | size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN)); |
13b1eb33 | 485 | free_pages_exact(info->queue, size); |
3343660d AL |
486 | kfree(info); |
487 | } | |
488 | ||
82af8ce8 | 489 | /* the config->del_vqs() implementation */ |
d2a7ddda MT |
490 | static void vp_del_vqs(struct virtio_device *vdev) |
491 | { | |
e969fed5 | 492 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); |
d2a7ddda | 493 | struct virtqueue *vq, *n; |
e969fed5 | 494 | struct virtio_pci_vq_info *info; |
d2a7ddda | 495 | |
e969fed5 MT |
496 | list_for_each_entry_safe(vq, n, &vdev->vqs, list) { |
497 | info = vq->priv; | |
31198159 MT |
498 | if (vp_dev->per_vq_vectors && |
499 | info->msix_vector != VIRTIO_MSI_NO_VECTOR) | |
f68d2408 RR |
500 | free_irq(vp_dev->msix_entries[info->msix_vector].vector, |
501 | vq); | |
d2a7ddda | 502 | vp_del_vq(vq); |
e969fed5 MT |
503 | } |
504 | vp_dev->per_vq_vectors = false; | |
82af8ce8 MT |
505 | |
506 | vp_free_vectors(vdev); | |
d2a7ddda MT |
507 | } |
508 | ||
e969fed5 MT |
509 | static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs, |
510 | struct virtqueue *vqs[], | |
511 | vq_callback_t *callbacks[], | |
512 | const char *names[], | |
f68d2408 | 513 | bool use_msix, |
e969fed5 | 514 | bool per_vq_vectors) |
d2a7ddda | 515 | { |
e969fed5 | 516 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); |
f68d2408 RR |
517 | u16 msix_vec; |
518 | int i, err, nvectors, allocated_vectors; | |
82af8ce8 | 519 | |
f68d2408 RR |
520 | if (!use_msix) { |
521 | /* Old style: one normal interrupt for change and all vqs. */ | |
522 | err = vp_request_intx(vdev); | |
523 | if (err) | |
524 | goto error_request; | |
525 | } else { | |
526 | if (per_vq_vectors) { | |
527 | /* Best option: one for change interrupt, one per vq. */ | |
528 | nvectors = 1; | |
529 | for (i = 0; i < nvqs; ++i) | |
530 | if (callbacks[i]) | |
531 | ++nvectors; | |
532 | } else { | |
533 | /* Second best: one for change, shared for all vqs. */ | |
534 | nvectors = 2; | |
535 | } | |
536 | ||
537 | err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors); | |
538 | if (err) | |
539 | goto error_request; | |
540 | } | |
d2a7ddda | 541 | |
e969fed5 MT |
542 | vp_dev->per_vq_vectors = per_vq_vectors; |
543 | allocated_vectors = vp_dev->msix_used_vectors; | |
d2a7ddda | 544 | for (i = 0; i < nvqs; ++i) { |
e969fed5 | 545 | if (!callbacks[i] || !vp_dev->msix_enabled) |
f68d2408 | 546 | msix_vec = VIRTIO_MSI_NO_VECTOR; |
e969fed5 | 547 | else if (vp_dev->per_vq_vectors) |
f68d2408 | 548 | msix_vec = allocated_vectors++; |
e969fed5 | 549 | else |
f68d2408 RR |
550 | msix_vec = VP_MSIX_VQ_VECTOR; |
551 | vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec); | |
e969fed5 MT |
552 | if (IS_ERR(vqs[i])) { |
553 | err = PTR_ERR(vqs[i]); | |
82af8ce8 | 554 | goto error_find; |
e969fed5 | 555 | } |
0b22bd0b MT |
556 | |
557 | if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR) | |
558 | continue; | |
559 | ||
e969fed5 | 560 | /* allocate per-vq irq if available and necessary */ |
0b22bd0b MT |
561 | snprintf(vp_dev->msix_names[msix_vec], |
562 | sizeof *vp_dev->msix_names, | |
563 | "%s-%s", | |
564 | dev_name(&vp_dev->vdev.dev), names[i]); | |
565 | err = request_irq(vp_dev->msix_entries[msix_vec].vector, | |
566 | vring_interrupt, 0, | |
567 | vp_dev->msix_names[msix_vec], | |
568 | vqs[i]); | |
569 | if (err) { | |
570 | vp_del_vq(vqs[i]); | |
571 | goto error_find; | |
e969fed5 | 572 | } |
d2a7ddda MT |
573 | } |
574 | return 0; | |
575 | ||
82af8ce8 | 576 | error_find: |
d2a7ddda | 577 | vp_del_vqs(vdev); |
82af8ce8 MT |
578 | |
579 | error_request: | |
e969fed5 MT |
580 | return err; |
581 | } | |
582 | ||
583 | /* the config->find_vqs() implementation */ | |
584 | static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs, | |
585 | struct virtqueue *vqs[], | |
586 | vq_callback_t *callbacks[], | |
587 | const char *names[]) | |
588 | { | |
f68d2408 | 589 | int err; |
e969fed5 | 590 | |
f68d2408 RR |
591 | /* Try MSI-X with one vector per queue. */ |
592 | err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true); | |
e969fed5 MT |
593 | if (!err) |
594 | return 0; | |
f68d2408 | 595 | /* Fallback: MSI-X with one vector for config, one shared for queues. */ |
e969fed5 | 596 | err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, |
f68d2408 | 597 | true, false); |
e969fed5 MT |
598 | if (!err) |
599 | return 0; | |
600 | /* Finally fall back to regular interrupts. */ | |
f68d2408 RR |
601 | return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, |
602 | false, false); | |
d2a7ddda MT |
603 | } |
604 | ||
66846048 RJ |
605 | static const char *vp_bus_name(struct virtio_device *vdev) |
606 | { | |
607 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
608 | ||
609 | return pci_name(vp_dev->pci_dev); | |
610 | } | |
611 | ||
3343660d | 612 | static struct virtio_config_ops virtio_pci_config_ops = { |
3343660d AL |
613 | .get = vp_get, |
614 | .set = vp_set, | |
615 | .get_status = vp_get_status, | |
616 | .set_status = vp_set_status, | |
617 | .reset = vp_reset, | |
d2a7ddda MT |
618 | .find_vqs = vp_find_vqs, |
619 | .del_vqs = vp_del_vqs, | |
c45a6816 | 620 | .get_features = vp_get_features, |
c624896e | 621 | .finalize_features = vp_finalize_features, |
66846048 | 622 | .bus_name = vp_bus_name, |
3343660d AL |
623 | }; |
624 | ||
29f9f12e MM |
625 | static void virtio_pci_release_dev(struct device *_d) |
626 | { | |
72103bd1 MT |
627 | /* |
628 | * No need for a release method as we allocate/free | |
629 | * all devices together with the pci devices. | |
630 | * Provide an empty one to avoid getting a warning from core. | |
631 | */ | |
29f9f12e MM |
632 | } |
633 | ||
3343660d AL |
634 | /* the PCI probing function */ |
635 | static int __devinit virtio_pci_probe(struct pci_dev *pci_dev, | |
636 | const struct pci_device_id *id) | |
637 | { | |
638 | struct virtio_pci_device *vp_dev; | |
639 | int err; | |
640 | ||
641 | /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */ | |
642 | if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f) | |
643 | return -ENODEV; | |
644 | ||
55a7c066 AL |
645 | if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) { |
646 | printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n", | |
647 | VIRTIO_PCI_ABI_VERSION, pci_dev->revision); | |
648 | return -ENODEV; | |
649 | } | |
650 | ||
3343660d AL |
651 | /* allocate our structure and fill it out */ |
652 | vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL); | |
653 | if (vp_dev == NULL) | |
654 | return -ENOMEM; | |
655 | ||
8b3bb3ec | 656 | vp_dev->vdev.dev.parent = &pci_dev->dev; |
29f9f12e | 657 | vp_dev->vdev.dev.release = virtio_pci_release_dev; |
3343660d AL |
658 | vp_dev->vdev.config = &virtio_pci_config_ops; |
659 | vp_dev->pci_dev = pci_dev; | |
660 | INIT_LIST_HEAD(&vp_dev->virtqueues); | |
661 | spin_lock_init(&vp_dev->lock); | |
662 | ||
b03214d5 MT |
663 | /* Disable MSI/MSIX to bring device to a known good state. */ |
664 | pci_msi_off(pci_dev); | |
665 | ||
3343660d AL |
666 | /* enable the device */ |
667 | err = pci_enable_device(pci_dev); | |
668 | if (err) | |
669 | goto out; | |
670 | ||
671 | err = pci_request_regions(pci_dev, "virtio-pci"); | |
672 | if (err) | |
673 | goto out_enable_device; | |
674 | ||
675 | vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0); | |
676 | if (vp_dev->ioaddr == NULL) | |
677 | goto out_req_regions; | |
678 | ||
679 | pci_set_drvdata(pci_dev, vp_dev); | |
bc505f37 | 680 | pci_set_master(pci_dev); |
3343660d AL |
681 | |
682 | /* we use the subsystem vendor/device id as the virtio vendor/device | |
683 | * id. this allows us to use the same PCI vendor/device id for all | |
684 | * virtio devices and to identify the particular virtio driver by | |
88393161 | 685 | * the subsystem ids */ |
3343660d AL |
686 | vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor; |
687 | vp_dev->vdev.id.device = pci_dev->subsystem_device; | |
688 | ||
3343660d AL |
689 | /* finally register the virtio device */ |
690 | err = register_virtio_device(&vp_dev->vdev); | |
691 | if (err) | |
82af8ce8 | 692 | goto out_set_drvdata; |
3343660d AL |
693 | |
694 | return 0; | |
695 | ||
3343660d AL |
696 | out_set_drvdata: |
697 | pci_set_drvdata(pci_dev, NULL); | |
698 | pci_iounmap(pci_dev, vp_dev->ioaddr); | |
699 | out_req_regions: | |
700 | pci_release_regions(pci_dev); | |
701 | out_enable_device: | |
702 | pci_disable_device(pci_dev); | |
703 | out: | |
704 | kfree(vp_dev); | |
705 | return err; | |
706 | } | |
707 | ||
708 | static void __devexit virtio_pci_remove(struct pci_dev *pci_dev) | |
709 | { | |
710 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
711 | ||
bd6c2690 | 712 | unregister_virtio_device(&vp_dev->vdev); |
31a3ddda AS |
713 | |
714 | vp_del_vqs(&vp_dev->vdev); | |
715 | pci_set_drvdata(pci_dev, NULL); | |
716 | pci_iounmap(pci_dev, vp_dev->ioaddr); | |
717 | pci_release_regions(pci_dev); | |
718 | pci_disable_device(pci_dev); | |
72103bd1 | 719 | kfree(vp_dev); |
3343660d AL |
720 | } |
721 | ||
722 | #ifdef CONFIG_PM | |
f0fe6f11 AS |
723 | static int virtio_pci_freeze(struct device *dev) |
724 | { | |
725 | struct pci_dev *pci_dev = to_pci_dev(dev); | |
726 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
727 | struct virtio_driver *drv; | |
728 | int ret; | |
729 | ||
730 | drv = container_of(vp_dev->vdev.dev.driver, | |
731 | struct virtio_driver, driver); | |
732 | ||
733 | ret = 0; | |
734 | vp_dev->saved_status = vp_get_status(&vp_dev->vdev); | |
735 | if (drv && drv->freeze) | |
736 | ret = drv->freeze(&vp_dev->vdev); | |
737 | ||
738 | if (!ret) | |
739 | pci_disable_device(pci_dev); | |
740 | return ret; | |
741 | } | |
742 | ||
0517fdd1 | 743 | static int virtio_pci_restore(struct device *dev) |
f0fe6f11 AS |
744 | { |
745 | struct pci_dev *pci_dev = to_pci_dev(dev); | |
746 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
0517fdd1 | 747 | struct virtio_driver *drv; |
f0fe6f11 AS |
748 | int ret; |
749 | ||
0517fdd1 AS |
750 | drv = container_of(vp_dev->vdev.dev.driver, |
751 | struct virtio_driver, driver); | |
752 | ||
f0fe6f11 AS |
753 | ret = pci_enable_device(pci_dev); |
754 | if (ret) | |
755 | return ret; | |
0517fdd1 | 756 | |
f0fe6f11 AS |
757 | pci_set_master(pci_dev); |
758 | vp_finalize_features(&vp_dev->vdev); | |
759 | ||
0517fdd1 | 760 | if (drv && drv->restore) |
f0fe6f11 AS |
761 | ret = drv->restore(&vp_dev->vdev); |
762 | ||
763 | /* Finally, tell the device we're all set */ | |
764 | if (!ret) | |
765 | vp_set_status(&vp_dev->vdev, vp_dev->saved_status); | |
766 | ||
767 | return ret; | |
768 | } | |
769 | ||
d0775363 | 770 | static const struct dev_pm_ops virtio_pci_pm_ops = { |
04c2322b AS |
771 | .suspend = virtio_pci_freeze, |
772 | .resume = virtio_pci_restore, | |
f0fe6f11 | 773 | .freeze = virtio_pci_freeze, |
f38f8387 | 774 | .thaw = virtio_pci_restore, |
f0fe6f11 | 775 | .restore = virtio_pci_restore, |
04c2322b | 776 | .poweroff = virtio_pci_freeze, |
d0775363 | 777 | }; |
3343660d AL |
778 | #endif |
779 | ||
780 | static struct pci_driver virtio_pci_driver = { | |
781 | .name = "virtio-pci", | |
782 | .id_table = virtio_pci_id_table, | |
783 | .probe = virtio_pci_probe, | |
1f08b833 | 784 | .remove = __devexit_p(virtio_pci_remove), |
3343660d | 785 | #ifdef CONFIG_PM |
d0775363 | 786 | .driver.pm = &virtio_pci_pm_ops, |
3343660d AL |
787 | #endif |
788 | }; | |
789 | ||
790 | static int __init virtio_pci_init(void) | |
791 | { | |
8b3bb3ec | 792 | return pci_register_driver(&virtio_pci_driver); |
3343660d AL |
793 | } |
794 | ||
795 | module_init(virtio_pci_init); | |
796 | ||
797 | static void __exit virtio_pci_exit(void) | |
798 | { | |
3343660d AL |
799 | pci_unregister_driver(&virtio_pci_driver); |
800 | } | |
801 | ||
802 | module_exit(virtio_pci_exit); |