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; | |
75a0a52b | 51 | cpumask_var_t *msix_affinity_masks; |
82af8ce8 MT |
52 | /* Name strings for interrupts. This size should be enough, |
53 | * and I'm too lazy to allocate each name separately. */ | |
54 | char (*msix_names)[256]; | |
55 | /* Number of available vectors */ | |
56 | unsigned msix_vectors; | |
e969fed5 | 57 | /* Vectors allocated, excluding per-vq vectors if any */ |
82af8ce8 | 58 | unsigned msix_used_vectors; |
f0fe6f11 AS |
59 | |
60 | /* Status saved during hibernate/restore */ | |
61 | u8 saved_status; | |
62 | ||
e969fed5 MT |
63 | /* Whether we have vector per vq */ |
64 | bool per_vq_vectors; | |
82af8ce8 MT |
65 | }; |
66 | ||
67 | /* Constants for MSI-X */ | |
68 | /* Use first vector for configuration changes, second and the rest for | |
69 | * virtqueues Thus, we need at least 2 vectors for MSI. */ | |
70 | enum { | |
71 | VP_MSIX_CONFIG_VECTOR = 0, | |
72 | VP_MSIX_VQ_VECTOR = 1, | |
3343660d AL |
73 | }; |
74 | ||
75 | struct virtio_pci_vq_info | |
76 | { | |
77 | /* the actual virtqueue */ | |
78 | struct virtqueue *vq; | |
79 | ||
80 | /* the number of entries in the queue */ | |
81 | int num; | |
82 | ||
3343660d AL |
83 | /* the virtual address of the ring queue */ |
84 | void *queue; | |
85 | ||
86 | /* the list node for the virtqueues list */ | |
87 | struct list_head node; | |
82af8ce8 MT |
88 | |
89 | /* MSI-X vector (or none) */ | |
f68d2408 | 90 | unsigned msix_vector; |
3343660d AL |
91 | }; |
92 | ||
93 | /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */ | |
94 | static struct pci_device_id virtio_pci_id_table[] = { | |
95 | { 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 }, | |
96 | { 0 }, | |
97 | }; | |
98 | ||
99 | MODULE_DEVICE_TABLE(pci, virtio_pci_id_table); | |
100 | ||
3343660d AL |
101 | /* Convert a generic virtio device to our structure */ |
102 | static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev) | |
103 | { | |
104 | return container_of(vdev, struct virtio_pci_device, vdev); | |
105 | } | |
106 | ||
c45a6816 RR |
107 | /* virtio config->get_features() implementation */ |
108 | static u32 vp_get_features(struct virtio_device *vdev) | |
109 | { | |
110 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
111 | ||
112 | /* When someone needs more than 32 feature bits, we'll need to | |
113 | * steal a bit to indicate that the rest are somewhere else. */ | |
114 | return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES); | |
115 | } | |
116 | ||
c624896e RR |
117 | /* virtio config->finalize_features() implementation */ |
118 | static void vp_finalize_features(struct virtio_device *vdev) | |
3343660d AL |
119 | { |
120 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
3343660d | 121 | |
e34f8725 RR |
122 | /* Give virtio_ring a chance to accept features. */ |
123 | vring_transport_features(vdev); | |
124 | ||
c624896e RR |
125 | /* We only support 32 feature bits. */ |
126 | BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1); | |
127 | iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES); | |
3343660d AL |
128 | } |
129 | ||
130 | /* virtio config->get() implementation */ | |
131 | static void vp_get(struct virtio_device *vdev, unsigned offset, | |
132 | void *buf, unsigned len) | |
133 | { | |
134 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
82af8ce8 MT |
135 | void __iomem *ioaddr = vp_dev->ioaddr + |
136 | VIRTIO_PCI_CONFIG(vp_dev) + offset; | |
3343660d AL |
137 | u8 *ptr = buf; |
138 | int i; | |
139 | ||
140 | for (i = 0; i < len; i++) | |
141 | ptr[i] = ioread8(ioaddr + i); | |
142 | } | |
143 | ||
144 | /* the config->set() implementation. it's symmetric to the config->get() | |
145 | * implementation */ | |
146 | static void vp_set(struct virtio_device *vdev, unsigned offset, | |
147 | const void *buf, unsigned len) | |
148 | { | |
149 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
82af8ce8 MT |
150 | void __iomem *ioaddr = vp_dev->ioaddr + |
151 | VIRTIO_PCI_CONFIG(vp_dev) + offset; | |
3343660d AL |
152 | const u8 *ptr = buf; |
153 | int i; | |
154 | ||
155 | for (i = 0; i < len; i++) | |
156 | iowrite8(ptr[i], ioaddr + i); | |
157 | } | |
158 | ||
159 | /* config->{get,set}_status() implementations */ | |
160 | static u8 vp_get_status(struct virtio_device *vdev) | |
161 | { | |
162 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
163 | return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS); | |
164 | } | |
165 | ||
166 | static void vp_set_status(struct virtio_device *vdev, u8 status) | |
167 | { | |
168 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
169 | /* We should never be setting status to 0. */ | |
170 | BUG_ON(status == 0); | |
597d56e4 | 171 | iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS); |
3343660d AL |
172 | } |
173 | ||
e6af578c MT |
174 | /* wait for pending irq handlers */ |
175 | static void vp_synchronize_vectors(struct virtio_device *vdev) | |
176 | { | |
177 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
178 | int i; | |
179 | ||
180 | if (vp_dev->intx_enabled) | |
181 | synchronize_irq(vp_dev->pci_dev->irq); | |
182 | ||
183 | for (i = 0; i < vp_dev->msix_vectors; ++i) | |
184 | synchronize_irq(vp_dev->msix_entries[i].vector); | |
185 | } | |
186 | ||
3343660d AL |
187 | static void vp_reset(struct virtio_device *vdev) |
188 | { | |
189 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
190 | /* 0 status means a reset. */ | |
597d56e4 | 191 | iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS); |
e6af578c MT |
192 | /* Flush out the status write, and flush in device writes, |
193 | * including MSi-X interrupts, if any. */ | |
194 | ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS); | |
195 | /* Flush pending VQ/configuration callbacks. */ | |
196 | vp_synchronize_vectors(vdev); | |
3343660d AL |
197 | } |
198 | ||
199 | /* the notify function used when creating a virt queue */ | |
200 | static void vp_notify(struct virtqueue *vq) | |
201 | { | |
202 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | |
3343660d AL |
203 | |
204 | /* we write the queue's selector into the notification register to | |
205 | * signal the other end */ | |
17bb6d40 JW |
206 | iowrite16(virtqueue_get_queue_index(vq), |
207 | vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY); | |
3343660d AL |
208 | } |
209 | ||
77cf5246 MT |
210 | /* Handle a configuration change: Tell driver if it wants to know. */ |
211 | static irqreturn_t vp_config_changed(int irq, void *opaque) | |
212 | { | |
213 | struct virtio_pci_device *vp_dev = opaque; | |
214 | struct virtio_driver *drv; | |
215 | drv = container_of(vp_dev->vdev.dev.driver, | |
216 | struct virtio_driver, driver); | |
217 | ||
218 | if (drv && drv->config_changed) | |
219 | drv->config_changed(&vp_dev->vdev); | |
220 | return IRQ_HANDLED; | |
221 | } | |
222 | ||
223 | /* Notify all virtqueues on an interrupt. */ | |
224 | static irqreturn_t vp_vring_interrupt(int irq, void *opaque) | |
225 | { | |
226 | struct virtio_pci_device *vp_dev = opaque; | |
227 | struct virtio_pci_vq_info *info; | |
228 | irqreturn_t ret = IRQ_NONE; | |
229 | unsigned long flags; | |
230 | ||
231 | spin_lock_irqsave(&vp_dev->lock, flags); | |
232 | list_for_each_entry(info, &vp_dev->virtqueues, node) { | |
233 | if (vring_interrupt(irq, info->vq) == IRQ_HANDLED) | |
234 | ret = IRQ_HANDLED; | |
235 | } | |
236 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
237 | ||
238 | return ret; | |
239 | } | |
240 | ||
3343660d AL |
241 | /* A small wrapper to also acknowledge the interrupt when it's handled. |
242 | * I really need an EIO hook for the vring so I can ack the interrupt once we | |
243 | * know that we'll be handling the IRQ but before we invoke the callback since | |
244 | * the callback may notify the host which results in the host attempting to | |
245 | * raise an interrupt that we would then mask once we acknowledged the | |
246 | * interrupt. */ | |
247 | static irqreturn_t vp_interrupt(int irq, void *opaque) | |
248 | { | |
249 | struct virtio_pci_device *vp_dev = opaque; | |
3343660d AL |
250 | u8 isr; |
251 | ||
252 | /* reading the ISR has the effect of also clearing it so it's very | |
253 | * important to save off the value. */ | |
254 | isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR); | |
255 | ||
256 | /* It's definitely not us if the ISR was not high */ | |
257 | if (!isr) | |
258 | return IRQ_NONE; | |
259 | ||
260 | /* Configuration change? Tell driver if it wants to know. */ | |
77cf5246 MT |
261 | if (isr & VIRTIO_PCI_ISR_CONFIG) |
262 | vp_config_changed(irq, opaque); | |
3343660d | 263 | |
77cf5246 | 264 | return vp_vring_interrupt(irq, opaque); |
3343660d AL |
265 | } |
266 | ||
82af8ce8 MT |
267 | static void vp_free_vectors(struct virtio_device *vdev) |
268 | { | |
269 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
270 | int i; | |
271 | ||
272 | if (vp_dev->intx_enabled) { | |
273 | free_irq(vp_dev->pci_dev->irq, vp_dev); | |
274 | vp_dev->intx_enabled = 0; | |
275 | } | |
276 | ||
277 | for (i = 0; i < vp_dev->msix_used_vectors; ++i) | |
278 | free_irq(vp_dev->msix_entries[i].vector, vp_dev); | |
82af8ce8 | 279 | |
75a0a52b JW |
280 | for (i = 0; i < vp_dev->msix_vectors; i++) |
281 | if (vp_dev->msix_affinity_masks[i]) | |
282 | free_cpumask_var(vp_dev->msix_affinity_masks[i]); | |
283 | ||
82af8ce8 MT |
284 | if (vp_dev->msix_enabled) { |
285 | /* Disable the vector used for configuration */ | |
286 | iowrite16(VIRTIO_MSI_NO_VECTOR, | |
287 | vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
288 | /* Flush the write out to device */ | |
289 | ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
290 | ||
82af8ce8 | 291 | pci_disable_msix(vp_dev->pci_dev); |
ff52c3fc MT |
292 | vp_dev->msix_enabled = 0; |
293 | vp_dev->msix_vectors = 0; | |
82af8ce8 | 294 | } |
ff52c3fc MT |
295 | |
296 | vp_dev->msix_used_vectors = 0; | |
297 | kfree(vp_dev->msix_names); | |
298 | vp_dev->msix_names = NULL; | |
299 | kfree(vp_dev->msix_entries); | |
300 | vp_dev->msix_entries = NULL; | |
75a0a52b JW |
301 | kfree(vp_dev->msix_affinity_masks); |
302 | vp_dev->msix_affinity_masks = NULL; | |
82af8ce8 MT |
303 | } |
304 | ||
f68d2408 RR |
305 | static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors, |
306 | bool per_vq_vectors) | |
82af8ce8 MT |
307 | { |
308 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
309 | const char *name = dev_name(&vp_dev->vdev.dev); | |
310 | unsigned i, v; | |
311 | int err = -ENOMEM; | |
e969fed5 | 312 | |
82af8ce8 MT |
313 | vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries, |
314 | GFP_KERNEL); | |
315 | if (!vp_dev->msix_entries) | |
ff52c3fc | 316 | goto error; |
82af8ce8 MT |
317 | vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names, |
318 | GFP_KERNEL); | |
319 | if (!vp_dev->msix_names) | |
ff52c3fc | 320 | goto error; |
75a0a52b JW |
321 | vp_dev->msix_affinity_masks |
322 | = kzalloc(nvectors * sizeof *vp_dev->msix_affinity_masks, | |
323 | GFP_KERNEL); | |
324 | if (!vp_dev->msix_affinity_masks) | |
325 | goto error; | |
326 | for (i = 0; i < nvectors; ++i) | |
327 | if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i], | |
328 | GFP_KERNEL)) | |
329 | goto error; | |
82af8ce8 MT |
330 | |
331 | for (i = 0; i < nvectors; ++i) | |
332 | vp_dev->msix_entries[i].entry = i; | |
333 | ||
f68d2408 | 334 | /* pci_enable_msix returns positive if we can't get this many. */ |
e969fed5 MT |
335 | err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors); |
336 | if (err > 0) | |
337 | err = -ENOSPC; | |
338 | if (err) | |
339 | goto error; | |
340 | vp_dev->msix_vectors = nvectors; | |
341 | vp_dev->msix_enabled = 1; | |
342 | ||
343 | /* Set the vector used for configuration */ | |
344 | v = vp_dev->msix_used_vectors; | |
345 | snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names, | |
346 | "%s-config", name); | |
347 | err = request_irq(vp_dev->msix_entries[v].vector, | |
348 | vp_config_changed, 0, vp_dev->msix_names[v], | |
349 | vp_dev); | |
350 | if (err) | |
351 | goto error; | |
352 | ++vp_dev->msix_used_vectors; | |
82af8ce8 | 353 | |
e969fed5 MT |
354 | iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); |
355 | /* Verify we had enough resources to assign the vector */ | |
356 | v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
357 | if (v == VIRTIO_MSI_NO_VECTOR) { | |
358 | err = -EBUSY; | |
359 | goto error; | |
82af8ce8 MT |
360 | } |
361 | ||
e969fed5 | 362 | if (!per_vq_vectors) { |
82af8ce8 MT |
363 | /* Shared vector for all VQs */ |
364 | v = vp_dev->msix_used_vectors; | |
365 | snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names, | |
366 | "%s-virtqueues", name); | |
367 | err = request_irq(vp_dev->msix_entries[v].vector, | |
368 | vp_vring_interrupt, 0, vp_dev->msix_names[v], | |
369 | vp_dev); | |
370 | if (err) | |
ff52c3fc | 371 | goto error; |
82af8ce8 MT |
372 | ++vp_dev->msix_used_vectors; |
373 | } | |
374 | return 0; | |
ff52c3fc | 375 | error: |
82af8ce8 | 376 | vp_free_vectors(vdev); |
82af8ce8 MT |
377 | return err; |
378 | } | |
379 | ||
f68d2408 RR |
380 | static int vp_request_intx(struct virtio_device *vdev) |
381 | { | |
382 | int err; | |
383 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
384 | ||
385 | err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, | |
386 | IRQF_SHARED, dev_name(&vdev->dev), vp_dev); | |
387 | if (!err) | |
388 | vp_dev->intx_enabled = 1; | |
389 | return err; | |
390 | } | |
391 | ||
392 | static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index, | |
393 | void (*callback)(struct virtqueue *vq), | |
394 | const char *name, | |
395 | u16 msix_vec) | |
3343660d AL |
396 | { |
397 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
398 | struct virtio_pci_vq_info *info; | |
399 | struct virtqueue *vq; | |
13b1eb33 | 400 | unsigned long flags, size; |
e969fed5 | 401 | u16 num; |
3343660d AL |
402 | int err; |
403 | ||
404 | /* Select the queue we're interested in */ | |
405 | iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); | |
406 | ||
407 | /* Check if queue is either not available or already active. */ | |
408 | num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM); | |
409 | if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN)) | |
410 | return ERR_PTR(-ENOENT); | |
411 | ||
412 | /* allocate and fill out our structure the represents an active | |
413 | * queue */ | |
414 | info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL); | |
415 | if (!info) | |
416 | return ERR_PTR(-ENOMEM); | |
417 | ||
3343660d | 418 | info->num = num; |
f68d2408 | 419 | info->msix_vector = msix_vec; |
3343660d | 420 | |
498af147 | 421 | size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN)); |
13b1eb33 | 422 | info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO); |
3343660d AL |
423 | if (info->queue == NULL) { |
424 | err = -ENOMEM; | |
425 | goto out_info; | |
426 | } | |
427 | ||
428 | /* activate the queue */ | |
480daab4 | 429 | iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT, |
3343660d AL |
430 | vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); |
431 | ||
432 | /* create the vring */ | |
17bb6d40 | 433 | vq = vring_new_virtqueue(index, info->num, VIRTIO_PCI_VRING_ALIGN, vdev, |
7b21e34f | 434 | true, info->queue, vp_notify, callback, name); |
3343660d AL |
435 | if (!vq) { |
436 | err = -ENOMEM; | |
437 | goto out_activate_queue; | |
438 | } | |
439 | ||
440 | vq->priv = info; | |
441 | info->vq = vq; | |
442 | ||
f68d2408 RR |
443 | if (msix_vec != VIRTIO_MSI_NO_VECTOR) { |
444 | iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
445 | msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
446 | if (msix_vec == VIRTIO_MSI_NO_VECTOR) { | |
82af8ce8 MT |
447 | err = -EBUSY; |
448 | goto out_assign; | |
449 | } | |
450 | } | |
451 | ||
005b20a8 KK |
452 | if (callback) { |
453 | spin_lock_irqsave(&vp_dev->lock, flags); | |
454 | list_add(&info->node, &vp_dev->virtqueues); | |
455 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
456 | } else { | |
457 | INIT_LIST_HEAD(&info->node); | |
458 | } | |
3343660d AL |
459 | |
460 | return vq; | |
461 | ||
82af8ce8 | 462 | out_assign: |
82af8ce8 | 463 | vring_del_virtqueue(vq); |
3343660d AL |
464 | out_activate_queue: |
465 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); | |
13b1eb33 | 466 | free_pages_exact(info->queue, size); |
3343660d AL |
467 | out_info: |
468 | kfree(info); | |
469 | return ERR_PTR(err); | |
470 | } | |
471 | ||
3343660d AL |
472 | static void vp_del_vq(struct virtqueue *vq) |
473 | { | |
474 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | |
475 | struct virtio_pci_vq_info *info = vq->priv; | |
f6c82507 MT |
476 | unsigned long flags, size; |
477 | ||
478 | spin_lock_irqsave(&vp_dev->lock, flags); | |
479 | list_del(&info->node); | |
480 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
3343660d | 481 | |
17bb6d40 JW |
482 | iowrite16(virtqueue_get_queue_index(vq), |
483 | vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); | |
82af8ce8 | 484 | |
82af8ce8 MT |
485 | if (vp_dev->msix_enabled) { |
486 | iowrite16(VIRTIO_MSI_NO_VECTOR, | |
487 | vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
488 | /* Flush the write out to device */ | |
489 | ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR); | |
490 | } | |
491 | ||
3343660d AL |
492 | vring_del_virtqueue(vq); |
493 | ||
494 | /* Select and deactivate the queue */ | |
3343660d AL |
495 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); |
496 | ||
498af147 | 497 | size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN)); |
13b1eb33 | 498 | free_pages_exact(info->queue, size); |
3343660d AL |
499 | kfree(info); |
500 | } | |
501 | ||
82af8ce8 | 502 | /* the config->del_vqs() implementation */ |
d2a7ddda MT |
503 | static void vp_del_vqs(struct virtio_device *vdev) |
504 | { | |
e969fed5 | 505 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); |
d2a7ddda | 506 | struct virtqueue *vq, *n; |
e969fed5 | 507 | struct virtio_pci_vq_info *info; |
d2a7ddda | 508 | |
e969fed5 MT |
509 | list_for_each_entry_safe(vq, n, &vdev->vqs, list) { |
510 | info = vq->priv; | |
31198159 MT |
511 | if (vp_dev->per_vq_vectors && |
512 | info->msix_vector != VIRTIO_MSI_NO_VECTOR) | |
f68d2408 RR |
513 | free_irq(vp_dev->msix_entries[info->msix_vector].vector, |
514 | vq); | |
d2a7ddda | 515 | vp_del_vq(vq); |
e969fed5 MT |
516 | } |
517 | vp_dev->per_vq_vectors = false; | |
82af8ce8 MT |
518 | |
519 | vp_free_vectors(vdev); | |
d2a7ddda MT |
520 | } |
521 | ||
e969fed5 MT |
522 | static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs, |
523 | struct virtqueue *vqs[], | |
524 | vq_callback_t *callbacks[], | |
525 | const char *names[], | |
f68d2408 | 526 | bool use_msix, |
e969fed5 | 527 | bool per_vq_vectors) |
d2a7ddda | 528 | { |
e969fed5 | 529 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); |
f68d2408 RR |
530 | u16 msix_vec; |
531 | int i, err, nvectors, allocated_vectors; | |
82af8ce8 | 532 | |
f68d2408 RR |
533 | if (!use_msix) { |
534 | /* Old style: one normal interrupt for change and all vqs. */ | |
535 | err = vp_request_intx(vdev); | |
536 | if (err) | |
537 | goto error_request; | |
538 | } else { | |
539 | if (per_vq_vectors) { | |
540 | /* Best option: one for change interrupt, one per vq. */ | |
541 | nvectors = 1; | |
542 | for (i = 0; i < nvqs; ++i) | |
543 | if (callbacks[i]) | |
544 | ++nvectors; | |
545 | } else { | |
546 | /* Second best: one for change, shared for all vqs. */ | |
547 | nvectors = 2; | |
548 | } | |
549 | ||
550 | err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors); | |
551 | if (err) | |
552 | goto error_request; | |
553 | } | |
d2a7ddda | 554 | |
e969fed5 MT |
555 | vp_dev->per_vq_vectors = per_vq_vectors; |
556 | allocated_vectors = vp_dev->msix_used_vectors; | |
d2a7ddda | 557 | for (i = 0; i < nvqs; ++i) { |
6457f126 MT |
558 | if (!names[i]) { |
559 | vqs[i] = NULL; | |
560 | continue; | |
561 | } else if (!callbacks[i] || !vp_dev->msix_enabled) | |
f68d2408 | 562 | msix_vec = VIRTIO_MSI_NO_VECTOR; |
e969fed5 | 563 | else if (vp_dev->per_vq_vectors) |
f68d2408 | 564 | msix_vec = allocated_vectors++; |
e969fed5 | 565 | else |
f68d2408 RR |
566 | msix_vec = VP_MSIX_VQ_VECTOR; |
567 | vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec); | |
e969fed5 MT |
568 | if (IS_ERR(vqs[i])) { |
569 | err = PTR_ERR(vqs[i]); | |
82af8ce8 | 570 | goto error_find; |
e969fed5 | 571 | } |
0b22bd0b MT |
572 | |
573 | if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR) | |
574 | continue; | |
575 | ||
e969fed5 | 576 | /* allocate per-vq irq if available and necessary */ |
0b22bd0b MT |
577 | snprintf(vp_dev->msix_names[msix_vec], |
578 | sizeof *vp_dev->msix_names, | |
579 | "%s-%s", | |
580 | dev_name(&vp_dev->vdev.dev), names[i]); | |
581 | err = request_irq(vp_dev->msix_entries[msix_vec].vector, | |
582 | vring_interrupt, 0, | |
583 | vp_dev->msix_names[msix_vec], | |
584 | vqs[i]); | |
585 | if (err) { | |
586 | vp_del_vq(vqs[i]); | |
587 | goto error_find; | |
e969fed5 | 588 | } |
d2a7ddda MT |
589 | } |
590 | return 0; | |
591 | ||
82af8ce8 | 592 | error_find: |
d2a7ddda | 593 | vp_del_vqs(vdev); |
82af8ce8 MT |
594 | |
595 | error_request: | |
e969fed5 MT |
596 | return err; |
597 | } | |
598 | ||
599 | /* the config->find_vqs() implementation */ | |
600 | static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs, | |
601 | struct virtqueue *vqs[], | |
602 | vq_callback_t *callbacks[], | |
603 | const char *names[]) | |
604 | { | |
f68d2408 | 605 | int err; |
e969fed5 | 606 | |
f68d2408 RR |
607 | /* Try MSI-X with one vector per queue. */ |
608 | err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true); | |
e969fed5 MT |
609 | if (!err) |
610 | return 0; | |
f68d2408 | 611 | /* Fallback: MSI-X with one vector for config, one shared for queues. */ |
e969fed5 | 612 | err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, |
f68d2408 | 613 | true, false); |
e969fed5 MT |
614 | if (!err) |
615 | return 0; | |
616 | /* Finally fall back to regular interrupts. */ | |
f68d2408 RR |
617 | return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, |
618 | false, false); | |
d2a7ddda MT |
619 | } |
620 | ||
66846048 RJ |
621 | static const char *vp_bus_name(struct virtio_device *vdev) |
622 | { | |
623 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
624 | ||
625 | return pci_name(vp_dev->pci_dev); | |
626 | } | |
627 | ||
75a0a52b JW |
628 | /* Setup the affinity for a virtqueue: |
629 | * - force the affinity for per vq vector | |
630 | * - OR over all affinities for shared MSI | |
631 | * - ignore the affinity request if we're using INTX | |
632 | */ | |
633 | static int vp_set_vq_affinity(struct virtqueue *vq, int cpu) | |
634 | { | |
635 | struct virtio_device *vdev = vq->vdev; | |
636 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
637 | struct virtio_pci_vq_info *info = vq->priv; | |
638 | struct cpumask *mask; | |
639 | unsigned int irq; | |
640 | ||
641 | if (!vq->callback) | |
642 | return -EINVAL; | |
643 | ||
644 | if (vp_dev->msix_enabled) { | |
645 | mask = vp_dev->msix_affinity_masks[info->msix_vector]; | |
646 | irq = vp_dev->msix_entries[info->msix_vector].vector; | |
647 | if (cpu == -1) | |
648 | irq_set_affinity_hint(irq, NULL); | |
649 | else { | |
650 | cpumask_set_cpu(cpu, mask); | |
651 | irq_set_affinity_hint(irq, mask); | |
652 | } | |
653 | } | |
654 | return 0; | |
655 | } | |
656 | ||
3343660d | 657 | static struct virtio_config_ops virtio_pci_config_ops = { |
3343660d AL |
658 | .get = vp_get, |
659 | .set = vp_set, | |
660 | .get_status = vp_get_status, | |
661 | .set_status = vp_set_status, | |
662 | .reset = vp_reset, | |
d2a7ddda MT |
663 | .find_vqs = vp_find_vqs, |
664 | .del_vqs = vp_del_vqs, | |
c45a6816 | 665 | .get_features = vp_get_features, |
c624896e | 666 | .finalize_features = vp_finalize_features, |
66846048 | 667 | .bus_name = vp_bus_name, |
75a0a52b | 668 | .set_vq_affinity = vp_set_vq_affinity, |
3343660d AL |
669 | }; |
670 | ||
29f9f12e MM |
671 | static void virtio_pci_release_dev(struct device *_d) |
672 | { | |
72103bd1 MT |
673 | /* |
674 | * No need for a release method as we allocate/free | |
675 | * all devices together with the pci devices. | |
676 | * Provide an empty one to avoid getting a warning from core. | |
677 | */ | |
29f9f12e MM |
678 | } |
679 | ||
3343660d AL |
680 | /* the PCI probing function */ |
681 | static int __devinit virtio_pci_probe(struct pci_dev *pci_dev, | |
682 | const struct pci_device_id *id) | |
683 | { | |
684 | struct virtio_pci_device *vp_dev; | |
685 | int err; | |
686 | ||
687 | /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */ | |
688 | if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f) | |
689 | return -ENODEV; | |
690 | ||
55a7c066 AL |
691 | if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) { |
692 | printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n", | |
693 | VIRTIO_PCI_ABI_VERSION, pci_dev->revision); | |
694 | return -ENODEV; | |
695 | } | |
696 | ||
3343660d AL |
697 | /* allocate our structure and fill it out */ |
698 | vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL); | |
699 | if (vp_dev == NULL) | |
700 | return -ENOMEM; | |
701 | ||
8b3bb3ec | 702 | vp_dev->vdev.dev.parent = &pci_dev->dev; |
29f9f12e | 703 | vp_dev->vdev.dev.release = virtio_pci_release_dev; |
3343660d AL |
704 | vp_dev->vdev.config = &virtio_pci_config_ops; |
705 | vp_dev->pci_dev = pci_dev; | |
706 | INIT_LIST_HEAD(&vp_dev->virtqueues); | |
707 | spin_lock_init(&vp_dev->lock); | |
708 | ||
b03214d5 MT |
709 | /* Disable MSI/MSIX to bring device to a known good state. */ |
710 | pci_msi_off(pci_dev); | |
711 | ||
3343660d AL |
712 | /* enable the device */ |
713 | err = pci_enable_device(pci_dev); | |
714 | if (err) | |
715 | goto out; | |
716 | ||
717 | err = pci_request_regions(pci_dev, "virtio-pci"); | |
718 | if (err) | |
719 | goto out_enable_device; | |
720 | ||
721 | vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0); | |
74a74b37 PST |
722 | if (vp_dev->ioaddr == NULL) { |
723 | err = -ENOMEM; | |
3343660d | 724 | goto out_req_regions; |
74a74b37 | 725 | } |
3343660d AL |
726 | |
727 | pci_set_drvdata(pci_dev, vp_dev); | |
bc505f37 | 728 | pci_set_master(pci_dev); |
3343660d AL |
729 | |
730 | /* we use the subsystem vendor/device id as the virtio vendor/device | |
731 | * id. this allows us to use the same PCI vendor/device id for all | |
732 | * virtio devices and to identify the particular virtio driver by | |
88393161 | 733 | * the subsystem ids */ |
3343660d AL |
734 | vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor; |
735 | vp_dev->vdev.id.device = pci_dev->subsystem_device; | |
736 | ||
3343660d AL |
737 | /* finally register the virtio device */ |
738 | err = register_virtio_device(&vp_dev->vdev); | |
739 | if (err) | |
82af8ce8 | 740 | goto out_set_drvdata; |
3343660d AL |
741 | |
742 | return 0; | |
743 | ||
3343660d AL |
744 | out_set_drvdata: |
745 | pci_set_drvdata(pci_dev, NULL); | |
746 | pci_iounmap(pci_dev, vp_dev->ioaddr); | |
747 | out_req_regions: | |
748 | pci_release_regions(pci_dev); | |
749 | out_enable_device: | |
750 | pci_disable_device(pci_dev); | |
751 | out: | |
752 | kfree(vp_dev); | |
753 | return err; | |
754 | } | |
755 | ||
756 | static void __devexit virtio_pci_remove(struct pci_dev *pci_dev) | |
757 | { | |
758 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
759 | ||
bd6c2690 | 760 | unregister_virtio_device(&vp_dev->vdev); |
31a3ddda AS |
761 | |
762 | vp_del_vqs(&vp_dev->vdev); | |
763 | pci_set_drvdata(pci_dev, NULL); | |
764 | pci_iounmap(pci_dev, vp_dev->ioaddr); | |
765 | pci_release_regions(pci_dev); | |
766 | pci_disable_device(pci_dev); | |
72103bd1 | 767 | kfree(vp_dev); |
3343660d AL |
768 | } |
769 | ||
770 | #ifdef CONFIG_PM | |
f0fe6f11 AS |
771 | static int virtio_pci_freeze(struct device *dev) |
772 | { | |
773 | struct pci_dev *pci_dev = to_pci_dev(dev); | |
774 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
775 | struct virtio_driver *drv; | |
776 | int ret; | |
777 | ||
778 | drv = container_of(vp_dev->vdev.dev.driver, | |
779 | struct virtio_driver, driver); | |
780 | ||
781 | ret = 0; | |
782 | vp_dev->saved_status = vp_get_status(&vp_dev->vdev); | |
783 | if (drv && drv->freeze) | |
784 | ret = drv->freeze(&vp_dev->vdev); | |
785 | ||
786 | if (!ret) | |
787 | pci_disable_device(pci_dev); | |
788 | return ret; | |
789 | } | |
790 | ||
0517fdd1 | 791 | static int virtio_pci_restore(struct device *dev) |
f0fe6f11 AS |
792 | { |
793 | struct pci_dev *pci_dev = to_pci_dev(dev); | |
794 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
0517fdd1 | 795 | struct virtio_driver *drv; |
f0fe6f11 AS |
796 | int ret; |
797 | ||
0517fdd1 AS |
798 | drv = container_of(vp_dev->vdev.dev.driver, |
799 | struct virtio_driver, driver); | |
800 | ||
f0fe6f11 AS |
801 | ret = pci_enable_device(pci_dev); |
802 | if (ret) | |
803 | return ret; | |
0517fdd1 | 804 | |
f0fe6f11 AS |
805 | pci_set_master(pci_dev); |
806 | vp_finalize_features(&vp_dev->vdev); | |
807 | ||
0517fdd1 | 808 | if (drv && drv->restore) |
f0fe6f11 AS |
809 | ret = drv->restore(&vp_dev->vdev); |
810 | ||
811 | /* Finally, tell the device we're all set */ | |
812 | if (!ret) | |
813 | vp_set_status(&vp_dev->vdev, vp_dev->saved_status); | |
814 | ||
815 | return ret; | |
816 | } | |
817 | ||
d0775363 | 818 | static const struct dev_pm_ops virtio_pci_pm_ops = { |
f878d0be | 819 | SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore) |
d0775363 | 820 | }; |
3343660d AL |
821 | #endif |
822 | ||
823 | static struct pci_driver virtio_pci_driver = { | |
824 | .name = "virtio-pci", | |
825 | .id_table = virtio_pci_id_table, | |
826 | .probe = virtio_pci_probe, | |
1f08b833 | 827 | .remove = __devexit_p(virtio_pci_remove), |
3343660d | 828 | #ifdef CONFIG_PM |
d0775363 | 829 | .driver.pm = &virtio_pci_pm_ops, |
3343660d AL |
830 | #endif |
831 | }; | |
832 | ||
1ce6853a | 833 | module_pci_driver(virtio_pci_driver); |