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; | |
e969fed5 | 55 | /* Vectors allocated, excluding per-vq vectors if any */ |
82af8ce8 | 56 | unsigned msix_used_vectors; |
e969fed5 MT |
57 | /* Whether we have vector per vq */ |
58 | bool per_vq_vectors; | |
82af8ce8 MT |
59 | }; |
60 | ||
61 | /* Constants for MSI-X */ | |
62 | /* Use first vector for configuration changes, second and the rest for | |
63 | * virtqueues Thus, we need at least 2 vectors for MSI. */ | |
64 | enum { | |
65 | VP_MSIX_CONFIG_VECTOR = 0, | |
66 | VP_MSIX_VQ_VECTOR = 1, | |
3343660d AL |
67 | }; |
68 | ||
69 | struct virtio_pci_vq_info | |
70 | { | |
71 | /* the actual virtqueue */ | |
72 | struct virtqueue *vq; | |
73 | ||
74 | /* the number of entries in the queue */ | |
75 | int num; | |
76 | ||
77 | /* the index of the queue */ | |
78 | int queue_index; | |
79 | ||
80 | /* the virtual address of the ring queue */ | |
81 | void *queue; | |
82 | ||
83 | /* the list node for the virtqueues list */ | |
84 | struct list_head node; | |
82af8ce8 MT |
85 | |
86 | /* MSI-X vector (or none) */ | |
f68d2408 | 87 | unsigned msix_vector; |
3343660d AL |
88 | }; |
89 | ||
90 | /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */ | |
91 | static struct pci_device_id virtio_pci_id_table[] = { | |
92 | { 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 }, | |
93 | { 0 }, | |
94 | }; | |
95 | ||
96 | MODULE_DEVICE_TABLE(pci, virtio_pci_id_table); | |
97 | ||
98 | /* A PCI device has it's own struct device and so does a virtio device so | |
99 | * we create a place for the virtio devices to show up in sysfs. I think it | |
100 | * would make more sense for virtio to not insist on having it's own device. */ | |
63d12556 | 101 | static struct device *virtio_pci_root; |
3343660d | 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 | ||
176 | static void vp_reset(struct virtio_device *vdev) | |
177 | { | |
178 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
179 | /* 0 status means a reset. */ | |
597d56e4 | 180 | iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS); |
3343660d AL |
181 | } |
182 | ||
183 | /* the notify function used when creating a virt queue */ | |
184 | static void vp_notify(struct virtqueue *vq) | |
185 | { | |
186 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | |
187 | struct virtio_pci_vq_info *info = vq->priv; | |
188 | ||
189 | /* we write the queue's selector into the notification register to | |
190 | * signal the other end */ | |
191 | iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY); | |
192 | } | |
193 | ||
77cf5246 MT |
194 | /* Handle a configuration change: Tell driver if it wants to know. */ |
195 | static irqreturn_t vp_config_changed(int irq, void *opaque) | |
196 | { | |
197 | struct virtio_pci_device *vp_dev = opaque; | |
198 | struct virtio_driver *drv; | |
199 | drv = container_of(vp_dev->vdev.dev.driver, | |
200 | struct virtio_driver, driver); | |
201 | ||
202 | if (drv && drv->config_changed) | |
203 | drv->config_changed(&vp_dev->vdev); | |
204 | return IRQ_HANDLED; | |
205 | } | |
206 | ||
207 | /* Notify all virtqueues on an interrupt. */ | |
208 | static irqreturn_t vp_vring_interrupt(int irq, void *opaque) | |
209 | { | |
210 | struct virtio_pci_device *vp_dev = opaque; | |
211 | struct virtio_pci_vq_info *info; | |
212 | irqreturn_t ret = IRQ_NONE; | |
213 | unsigned long flags; | |
214 | ||
215 | spin_lock_irqsave(&vp_dev->lock, flags); | |
216 | list_for_each_entry(info, &vp_dev->virtqueues, node) { | |
217 | if (vring_interrupt(irq, info->vq) == IRQ_HANDLED) | |
218 | ret = IRQ_HANDLED; | |
219 | } | |
220 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
221 | ||
222 | return ret; | |
223 | } | |
224 | ||
3343660d AL |
225 | /* A small wrapper to also acknowledge the interrupt when it's handled. |
226 | * I really need an EIO hook for the vring so I can ack the interrupt once we | |
227 | * know that we'll be handling the IRQ but before we invoke the callback since | |
228 | * the callback may notify the host which results in the host attempting to | |
229 | * raise an interrupt that we would then mask once we acknowledged the | |
230 | * interrupt. */ | |
231 | static irqreturn_t vp_interrupt(int irq, void *opaque) | |
232 | { | |
233 | struct virtio_pci_device *vp_dev = opaque; | |
3343660d AL |
234 | u8 isr; |
235 | ||
236 | /* reading the ISR has the effect of also clearing it so it's very | |
237 | * important to save off the value. */ | |
238 | isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR); | |
239 | ||
240 | /* It's definitely not us if the ISR was not high */ | |
241 | if (!isr) | |
242 | return IRQ_NONE; | |
243 | ||
244 | /* Configuration change? Tell driver if it wants to know. */ | |
77cf5246 MT |
245 | if (isr & VIRTIO_PCI_ISR_CONFIG) |
246 | vp_config_changed(irq, opaque); | |
3343660d | 247 | |
77cf5246 | 248 | return vp_vring_interrupt(irq, opaque); |
3343660d AL |
249 | } |
250 | ||
82af8ce8 MT |
251 | static void vp_free_vectors(struct virtio_device *vdev) |
252 | { | |
253 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
254 | int i; | |
255 | ||
256 | if (vp_dev->intx_enabled) { | |
257 | free_irq(vp_dev->pci_dev->irq, vp_dev); | |
258 | vp_dev->intx_enabled = 0; | |
259 | } | |
260 | ||
261 | for (i = 0; i < vp_dev->msix_used_vectors; ++i) | |
262 | free_irq(vp_dev->msix_entries[i].vector, vp_dev); | |
82af8ce8 MT |
263 | |
264 | if (vp_dev->msix_enabled) { | |
265 | /* Disable the vector used for configuration */ | |
266 | iowrite16(VIRTIO_MSI_NO_VECTOR, | |
267 | vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
268 | /* Flush the write out to device */ | |
269 | ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
270 | ||
82af8ce8 | 271 | pci_disable_msix(vp_dev->pci_dev); |
ff52c3fc MT |
272 | vp_dev->msix_enabled = 0; |
273 | vp_dev->msix_vectors = 0; | |
82af8ce8 | 274 | } |
ff52c3fc MT |
275 | |
276 | vp_dev->msix_used_vectors = 0; | |
277 | kfree(vp_dev->msix_names); | |
278 | vp_dev->msix_names = NULL; | |
279 | kfree(vp_dev->msix_entries); | |
280 | vp_dev->msix_entries = NULL; | |
82af8ce8 MT |
281 | } |
282 | ||
f68d2408 RR |
283 | static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors, |
284 | bool per_vq_vectors) | |
82af8ce8 MT |
285 | { |
286 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
287 | const char *name = dev_name(&vp_dev->vdev.dev); | |
288 | unsigned i, v; | |
289 | int err = -ENOMEM; | |
e969fed5 | 290 | |
82af8ce8 MT |
291 | vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries, |
292 | GFP_KERNEL); | |
293 | if (!vp_dev->msix_entries) | |
ff52c3fc | 294 | goto error; |
82af8ce8 MT |
295 | vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names, |
296 | GFP_KERNEL); | |
297 | if (!vp_dev->msix_names) | |
ff52c3fc | 298 | goto error; |
82af8ce8 MT |
299 | |
300 | for (i = 0; i < nvectors; ++i) | |
301 | vp_dev->msix_entries[i].entry = i; | |
302 | ||
f68d2408 | 303 | /* pci_enable_msix returns positive if we can't get this many. */ |
e969fed5 MT |
304 | err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors); |
305 | if (err > 0) | |
306 | err = -ENOSPC; | |
307 | if (err) | |
308 | goto error; | |
309 | vp_dev->msix_vectors = nvectors; | |
310 | vp_dev->msix_enabled = 1; | |
311 | ||
312 | /* Set the vector used for configuration */ | |
313 | v = vp_dev->msix_used_vectors; | |
314 | snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names, | |
315 | "%s-config", name); | |
316 | err = request_irq(vp_dev->msix_entries[v].vector, | |
317 | vp_config_changed, 0, vp_dev->msix_names[v], | |
318 | vp_dev); | |
319 | if (err) | |
320 | goto error; | |
321 | ++vp_dev->msix_used_vectors; | |
82af8ce8 | 322 | |
e969fed5 MT |
323 | iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); |
324 | /* Verify we had enough resources to assign the vector */ | |
325 | v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
326 | if (v == VIRTIO_MSI_NO_VECTOR) { | |
327 | err = -EBUSY; | |
328 | goto error; | |
82af8ce8 MT |
329 | } |
330 | ||
e969fed5 | 331 | if (!per_vq_vectors) { |
82af8ce8 MT |
332 | /* Shared vector for all VQs */ |
333 | v = vp_dev->msix_used_vectors; | |
334 | snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names, | |
335 | "%s-virtqueues", name); | |
336 | err = request_irq(vp_dev->msix_entries[v].vector, | |
337 | vp_vring_interrupt, 0, vp_dev->msix_names[v], | |
338 | vp_dev); | |
339 | if (err) | |
ff52c3fc | 340 | goto error; |
82af8ce8 MT |
341 | ++vp_dev->msix_used_vectors; |
342 | } | |
343 | return 0; | |
ff52c3fc | 344 | error: |
82af8ce8 | 345 | vp_free_vectors(vdev); |
82af8ce8 MT |
346 | return err; |
347 | } | |
348 | ||
f68d2408 RR |
349 | static int vp_request_intx(struct virtio_device *vdev) |
350 | { | |
351 | int err; | |
352 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
353 | ||
354 | err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, | |
355 | IRQF_SHARED, dev_name(&vdev->dev), vp_dev); | |
356 | if (!err) | |
357 | vp_dev->intx_enabled = 1; | |
358 | return err; | |
359 | } | |
360 | ||
361 | static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index, | |
362 | void (*callback)(struct virtqueue *vq), | |
363 | const char *name, | |
364 | u16 msix_vec) | |
3343660d AL |
365 | { |
366 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
367 | struct virtio_pci_vq_info *info; | |
368 | struct virtqueue *vq; | |
13b1eb33 | 369 | unsigned long flags, size; |
e969fed5 | 370 | u16 num; |
3343660d AL |
371 | int err; |
372 | ||
373 | /* Select the queue we're interested in */ | |
374 | iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); | |
375 | ||
376 | /* Check if queue is either not available or already active. */ | |
377 | num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM); | |
378 | if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN)) | |
379 | return ERR_PTR(-ENOENT); | |
380 | ||
381 | /* allocate and fill out our structure the represents an active | |
382 | * queue */ | |
383 | info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL); | |
384 | if (!info) | |
385 | return ERR_PTR(-ENOMEM); | |
386 | ||
387 | info->queue_index = index; | |
388 | info->num = num; | |
f68d2408 | 389 | info->msix_vector = msix_vec; |
3343660d | 390 | |
498af147 | 391 | size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN)); |
13b1eb33 | 392 | info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO); |
3343660d AL |
393 | if (info->queue == NULL) { |
394 | err = -ENOMEM; | |
395 | goto out_info; | |
396 | } | |
397 | ||
398 | /* activate the queue */ | |
480daab4 | 399 | iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT, |
3343660d AL |
400 | vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); |
401 | ||
402 | /* create the vring */ | |
87c7d57c | 403 | vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN, |
9499f5e7 | 404 | vdev, info->queue, vp_notify, callback, name); |
3343660d AL |
405 | if (!vq) { |
406 | err = -ENOMEM; | |
407 | goto out_activate_queue; | |
408 | } | |
409 | ||
410 | vq->priv = info; | |
411 | info->vq = vq; | |
412 | ||
f68d2408 RR |
413 | if (msix_vec != VIRTIO_MSI_NO_VECTOR) { |
414 | iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
415 | msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
416 | if (msix_vec == VIRTIO_MSI_NO_VECTOR) { | |
82af8ce8 MT |
417 | err = -EBUSY; |
418 | goto out_assign; | |
419 | } | |
420 | } | |
421 | ||
27ebe308 | 422 | spin_lock_irqsave(&vp_dev->lock, flags); |
3343660d | 423 | list_add(&info->node, &vp_dev->virtqueues); |
27ebe308 | 424 | spin_unlock_irqrestore(&vp_dev->lock, flags); |
3343660d AL |
425 | |
426 | return vq; | |
427 | ||
82af8ce8 | 428 | out_assign: |
82af8ce8 | 429 | vring_del_virtqueue(vq); |
3343660d AL |
430 | out_activate_queue: |
431 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); | |
13b1eb33 | 432 | free_pages_exact(info->queue, size); |
3343660d AL |
433 | out_info: |
434 | kfree(info); | |
435 | return ERR_PTR(err); | |
436 | } | |
437 | ||
3343660d AL |
438 | static void vp_del_vq(struct virtqueue *vq) |
439 | { | |
440 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | |
441 | struct virtio_pci_vq_info *info = vq->priv; | |
f6c82507 MT |
442 | unsigned long flags, size; |
443 | ||
444 | spin_lock_irqsave(&vp_dev->lock, flags); | |
445 | list_del(&info->node); | |
446 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
3343660d | 447 | |
82af8ce8 MT |
448 | iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); |
449 | ||
82af8ce8 MT |
450 | if (vp_dev->msix_enabled) { |
451 | iowrite16(VIRTIO_MSI_NO_VECTOR, | |
452 | vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
453 | /* Flush the write out to device */ | |
454 | ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR); | |
455 | } | |
456 | ||
3343660d AL |
457 | vring_del_virtqueue(vq); |
458 | ||
459 | /* Select and deactivate the queue */ | |
3343660d AL |
460 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); |
461 | ||
498af147 | 462 | size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN)); |
13b1eb33 | 463 | free_pages_exact(info->queue, size); |
3343660d AL |
464 | kfree(info); |
465 | } | |
466 | ||
82af8ce8 | 467 | /* the config->del_vqs() implementation */ |
d2a7ddda MT |
468 | static void vp_del_vqs(struct virtio_device *vdev) |
469 | { | |
e969fed5 | 470 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); |
d2a7ddda | 471 | struct virtqueue *vq, *n; |
e969fed5 | 472 | struct virtio_pci_vq_info *info; |
d2a7ddda | 473 | |
e969fed5 MT |
474 | list_for_each_entry_safe(vq, n, &vdev->vqs, list) { |
475 | info = vq->priv; | |
31198159 MT |
476 | if (vp_dev->per_vq_vectors && |
477 | info->msix_vector != VIRTIO_MSI_NO_VECTOR) | |
f68d2408 RR |
478 | free_irq(vp_dev->msix_entries[info->msix_vector].vector, |
479 | vq); | |
d2a7ddda | 480 | vp_del_vq(vq); |
e969fed5 MT |
481 | } |
482 | vp_dev->per_vq_vectors = false; | |
82af8ce8 MT |
483 | |
484 | vp_free_vectors(vdev); | |
d2a7ddda MT |
485 | } |
486 | ||
e969fed5 MT |
487 | static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs, |
488 | struct virtqueue *vqs[], | |
489 | vq_callback_t *callbacks[], | |
490 | const char *names[], | |
f68d2408 | 491 | bool use_msix, |
e969fed5 | 492 | bool per_vq_vectors) |
d2a7ddda | 493 | { |
e969fed5 | 494 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); |
f68d2408 RR |
495 | u16 msix_vec; |
496 | int i, err, nvectors, allocated_vectors; | |
82af8ce8 | 497 | |
f68d2408 RR |
498 | if (!use_msix) { |
499 | /* Old style: one normal interrupt for change and all vqs. */ | |
500 | err = vp_request_intx(vdev); | |
501 | if (err) | |
502 | goto error_request; | |
503 | } else { | |
504 | if (per_vq_vectors) { | |
505 | /* Best option: one for change interrupt, one per vq. */ | |
506 | nvectors = 1; | |
507 | for (i = 0; i < nvqs; ++i) | |
508 | if (callbacks[i]) | |
509 | ++nvectors; | |
510 | } else { | |
511 | /* Second best: one for change, shared for all vqs. */ | |
512 | nvectors = 2; | |
513 | } | |
514 | ||
515 | err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors); | |
516 | if (err) | |
517 | goto error_request; | |
518 | } | |
d2a7ddda | 519 | |
e969fed5 MT |
520 | vp_dev->per_vq_vectors = per_vq_vectors; |
521 | allocated_vectors = vp_dev->msix_used_vectors; | |
d2a7ddda | 522 | for (i = 0; i < nvqs; ++i) { |
e969fed5 | 523 | if (!callbacks[i] || !vp_dev->msix_enabled) |
f68d2408 | 524 | msix_vec = VIRTIO_MSI_NO_VECTOR; |
e969fed5 | 525 | else if (vp_dev->per_vq_vectors) |
f68d2408 | 526 | msix_vec = allocated_vectors++; |
e969fed5 | 527 | else |
f68d2408 RR |
528 | msix_vec = VP_MSIX_VQ_VECTOR; |
529 | vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec); | |
e969fed5 MT |
530 | if (IS_ERR(vqs[i])) { |
531 | err = PTR_ERR(vqs[i]); | |
82af8ce8 | 532 | goto error_find; |
e969fed5 | 533 | } |
0b22bd0b MT |
534 | |
535 | if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR) | |
536 | continue; | |
537 | ||
e969fed5 | 538 | /* allocate per-vq irq if available and necessary */ |
0b22bd0b MT |
539 | snprintf(vp_dev->msix_names[msix_vec], |
540 | sizeof *vp_dev->msix_names, | |
541 | "%s-%s", | |
542 | dev_name(&vp_dev->vdev.dev), names[i]); | |
543 | err = request_irq(vp_dev->msix_entries[msix_vec].vector, | |
544 | vring_interrupt, 0, | |
545 | vp_dev->msix_names[msix_vec], | |
546 | vqs[i]); | |
547 | if (err) { | |
548 | vp_del_vq(vqs[i]); | |
549 | goto error_find; | |
e969fed5 | 550 | } |
d2a7ddda MT |
551 | } |
552 | return 0; | |
553 | ||
82af8ce8 | 554 | error_find: |
d2a7ddda | 555 | vp_del_vqs(vdev); |
82af8ce8 MT |
556 | |
557 | error_request: | |
e969fed5 MT |
558 | return err; |
559 | } | |
560 | ||
561 | /* the config->find_vqs() implementation */ | |
562 | static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs, | |
563 | struct virtqueue *vqs[], | |
564 | vq_callback_t *callbacks[], | |
565 | const char *names[]) | |
566 | { | |
f68d2408 | 567 | int err; |
e969fed5 | 568 | |
f68d2408 RR |
569 | /* Try MSI-X with one vector per queue. */ |
570 | err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true); | |
e969fed5 MT |
571 | if (!err) |
572 | return 0; | |
f68d2408 | 573 | /* Fallback: MSI-X with one vector for config, one shared for queues. */ |
e969fed5 | 574 | err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, |
f68d2408 | 575 | true, false); |
e969fed5 MT |
576 | if (!err) |
577 | return 0; | |
578 | /* Finally fall back to regular interrupts. */ | |
f68d2408 RR |
579 | return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, |
580 | false, false); | |
d2a7ddda MT |
581 | } |
582 | ||
3343660d | 583 | static struct virtio_config_ops virtio_pci_config_ops = { |
3343660d AL |
584 | .get = vp_get, |
585 | .set = vp_set, | |
586 | .get_status = vp_get_status, | |
587 | .set_status = vp_set_status, | |
588 | .reset = vp_reset, | |
d2a7ddda MT |
589 | .find_vqs = vp_find_vqs, |
590 | .del_vqs = vp_del_vqs, | |
c45a6816 | 591 | .get_features = vp_get_features, |
c624896e | 592 | .finalize_features = vp_finalize_features, |
3343660d AL |
593 | }; |
594 | ||
29f9f12e MM |
595 | static void virtio_pci_release_dev(struct device *_d) |
596 | { | |
597 | struct virtio_device *dev = container_of(_d, struct virtio_device, dev); | |
598 | struct virtio_pci_device *vp_dev = to_vp_device(dev); | |
599 | struct pci_dev *pci_dev = vp_dev->pci_dev; | |
600 | ||
82af8ce8 | 601 | vp_del_vqs(dev); |
29f9f12e MM |
602 | pci_set_drvdata(pci_dev, NULL); |
603 | pci_iounmap(pci_dev, vp_dev->ioaddr); | |
604 | pci_release_regions(pci_dev); | |
605 | pci_disable_device(pci_dev); | |
606 | kfree(vp_dev); | |
607 | } | |
608 | ||
3343660d AL |
609 | /* the PCI probing function */ |
610 | static int __devinit virtio_pci_probe(struct pci_dev *pci_dev, | |
611 | const struct pci_device_id *id) | |
612 | { | |
613 | struct virtio_pci_device *vp_dev; | |
614 | int err; | |
615 | ||
616 | /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */ | |
617 | if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f) | |
618 | return -ENODEV; | |
619 | ||
55a7c066 AL |
620 | if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) { |
621 | printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n", | |
622 | VIRTIO_PCI_ABI_VERSION, pci_dev->revision); | |
623 | return -ENODEV; | |
624 | } | |
625 | ||
3343660d AL |
626 | /* allocate our structure and fill it out */ |
627 | vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL); | |
628 | if (vp_dev == NULL) | |
629 | return -ENOMEM; | |
630 | ||
63d12556 | 631 | vp_dev->vdev.dev.parent = virtio_pci_root; |
29f9f12e | 632 | vp_dev->vdev.dev.release = virtio_pci_release_dev; |
3343660d AL |
633 | vp_dev->vdev.config = &virtio_pci_config_ops; |
634 | vp_dev->pci_dev = pci_dev; | |
635 | INIT_LIST_HEAD(&vp_dev->virtqueues); | |
636 | spin_lock_init(&vp_dev->lock); | |
637 | ||
638 | /* enable the device */ | |
639 | err = pci_enable_device(pci_dev); | |
640 | if (err) | |
641 | goto out; | |
642 | ||
643 | err = pci_request_regions(pci_dev, "virtio-pci"); | |
644 | if (err) | |
645 | goto out_enable_device; | |
646 | ||
647 | vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0); | |
648 | if (vp_dev->ioaddr == NULL) | |
649 | goto out_req_regions; | |
650 | ||
651 | pci_set_drvdata(pci_dev, vp_dev); | |
652 | ||
653 | /* we use the subsystem vendor/device id as the virtio vendor/device | |
654 | * id. this allows us to use the same PCI vendor/device id for all | |
655 | * virtio devices and to identify the particular virtio driver by | |
656 | * the subsytem ids */ | |
657 | vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor; | |
658 | vp_dev->vdev.id.device = pci_dev->subsystem_device; | |
659 | ||
3343660d AL |
660 | /* finally register the virtio device */ |
661 | err = register_virtio_device(&vp_dev->vdev); | |
662 | if (err) | |
82af8ce8 | 663 | goto out_set_drvdata; |
3343660d AL |
664 | |
665 | return 0; | |
666 | ||
3343660d AL |
667 | out_set_drvdata: |
668 | pci_set_drvdata(pci_dev, NULL); | |
669 | pci_iounmap(pci_dev, vp_dev->ioaddr); | |
670 | out_req_regions: | |
671 | pci_release_regions(pci_dev); | |
672 | out_enable_device: | |
673 | pci_disable_device(pci_dev); | |
674 | out: | |
675 | kfree(vp_dev); | |
676 | return err; | |
677 | } | |
678 | ||
679 | static void __devexit virtio_pci_remove(struct pci_dev *pci_dev) | |
680 | { | |
681 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
682 | ||
bd6c2690 | 683 | unregister_virtio_device(&vp_dev->vdev); |
3343660d AL |
684 | } |
685 | ||
686 | #ifdef CONFIG_PM | |
687 | static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state) | |
688 | { | |
689 | pci_save_state(pci_dev); | |
690 | pci_set_power_state(pci_dev, PCI_D3hot); | |
691 | return 0; | |
692 | } | |
693 | ||
694 | static int virtio_pci_resume(struct pci_dev *pci_dev) | |
695 | { | |
696 | pci_restore_state(pci_dev); | |
697 | pci_set_power_state(pci_dev, PCI_D0); | |
698 | return 0; | |
699 | } | |
700 | #endif | |
701 | ||
702 | static struct pci_driver virtio_pci_driver = { | |
703 | .name = "virtio-pci", | |
704 | .id_table = virtio_pci_id_table, | |
705 | .probe = virtio_pci_probe, | |
1f08b833 | 706 | .remove = __devexit_p(virtio_pci_remove), |
3343660d AL |
707 | #ifdef CONFIG_PM |
708 | .suspend = virtio_pci_suspend, | |
709 | .resume = virtio_pci_resume, | |
710 | #endif | |
711 | }; | |
712 | ||
713 | static int __init virtio_pci_init(void) | |
714 | { | |
715 | int err; | |
716 | ||
63d12556 MM |
717 | virtio_pci_root = root_device_register("virtio-pci"); |
718 | if (IS_ERR(virtio_pci_root)) | |
719 | return PTR_ERR(virtio_pci_root); | |
3343660d AL |
720 | |
721 | err = pci_register_driver(&virtio_pci_driver); | |
722 | if (err) | |
4b892e65 | 723 | root_device_unregister(virtio_pci_root); |
3343660d AL |
724 | |
725 | return err; | |
726 | } | |
727 | ||
728 | module_init(virtio_pci_init); | |
729 | ||
730 | static void __exit virtio_pci_exit(void) | |
731 | { | |
3343660d | 732 | pci_unregister_driver(&virtio_pci_driver); |
63d12556 | 733 | root_device_unregister(virtio_pci_root); |
3343660d AL |
734 | } |
735 | ||
736 | module_exit(virtio_pci_exit); |