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 | ||
3ec7a77b MT |
34 | struct virtio_pci_vq_info { |
35 | /* the actual virtqueue */ | |
36 | struct virtqueue *vq; | |
37 | ||
38 | /* the number of entries in the queue */ | |
39 | int num; | |
40 | ||
41 | /* the virtual address of the ring queue */ | |
42 | void *queue; | |
43 | ||
44 | /* the list node for the virtqueues list */ | |
45 | struct list_head node; | |
46 | ||
47 | /* MSI-X vector (or none) */ | |
48 | unsigned msix_vector; | |
49 | }; | |
50 | ||
3343660d | 51 | /* Our device structure */ |
f913dd45 | 52 | struct virtio_pci_device { |
3343660d AL |
53 | struct virtio_device vdev; |
54 | struct pci_dev *pci_dev; | |
55 | ||
56 | /* the IO mapping for the PCI config space */ | |
97968358 | 57 | void __iomem *ioaddr; |
3343660d | 58 | |
af535722 MT |
59 | /* the IO mapping for ISR operation */ |
60 | void __iomem *isr; | |
61 | ||
3343660d AL |
62 | /* a list of queues so we can dispatch IRQs */ |
63 | spinlock_t lock; | |
64 | struct list_head virtqueues; | |
82af8ce8 | 65 | |
3ec7a77b MT |
66 | /* array of all queues for house-keeping */ |
67 | struct virtio_pci_vq_info **vqs; | |
68 | ||
82af8ce8 MT |
69 | /* MSI-X support */ |
70 | int msix_enabled; | |
71 | int intx_enabled; | |
72 | struct msix_entry *msix_entries; | |
75a0a52b | 73 | cpumask_var_t *msix_affinity_masks; |
82af8ce8 MT |
74 | /* Name strings for interrupts. This size should be enough, |
75 | * and I'm too lazy to allocate each name separately. */ | |
76 | char (*msix_names)[256]; | |
77 | /* Number of available vectors */ | |
78 | unsigned msix_vectors; | |
e969fed5 | 79 | /* Vectors allocated, excluding per-vq vectors if any */ |
82af8ce8 | 80 | unsigned msix_used_vectors; |
f0fe6f11 | 81 | |
e969fed5 MT |
82 | /* Whether we have vector per vq */ |
83 | bool per_vq_vectors; | |
5386cef2 | 84 | |
b09f00bb MT |
85 | struct virtqueue *(*setup_vq)(struct virtio_pci_device *vp_dev, |
86 | struct virtio_pci_vq_info *info, | |
87 | unsigned idx, | |
88 | void (*callback)(struct virtqueue *vq), | |
89 | const char *name, | |
90 | u16 msix_vec); | |
5386cef2 | 91 | void (*del_vq)(struct virtio_pci_vq_info *info); |
82af8ce8 MT |
92 | }; |
93 | ||
94 | /* Constants for MSI-X */ | |
95 | /* Use first vector for configuration changes, second and the rest for | |
96 | * virtqueues Thus, we need at least 2 vectors for MSI. */ | |
97 | enum { | |
98 | VP_MSIX_CONFIG_VECTOR = 0, | |
99 | VP_MSIX_VQ_VECTOR = 1, | |
3343660d AL |
100 | }; |
101 | ||
3343660d | 102 | /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */ |
cef340e6 | 103 | static const struct pci_device_id virtio_pci_id_table[] = { |
35cdc9eb SH |
104 | { PCI_DEVICE(0x1af4, PCI_ANY_ID) }, |
105 | { 0 } | |
3343660d AL |
106 | }; |
107 | ||
108 | MODULE_DEVICE_TABLE(pci, virtio_pci_id_table); | |
109 | ||
3343660d AL |
110 | /* Convert a generic virtio device to our structure */ |
111 | static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev) | |
112 | { | |
113 | return container_of(vdev, struct virtio_pci_device, vdev); | |
114 | } | |
115 | ||
c45a6816 | 116 | /* virtio config->get_features() implementation */ |
d0254773 | 117 | static u64 vp_get_features(struct virtio_device *vdev) |
c45a6816 RR |
118 | { |
119 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
120 | ||
121 | /* When someone needs more than 32 feature bits, we'll need to | |
122 | * steal a bit to indicate that the rest are somewhere else. */ | |
123 | return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES); | |
124 | } | |
125 | ||
c624896e | 126 | /* virtio config->finalize_features() implementation */ |
5c609a5e | 127 | static int vp_finalize_features(struct virtio_device *vdev) |
3343660d AL |
128 | { |
129 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
3343660d | 130 | |
e34f8725 RR |
131 | /* Give virtio_ring a chance to accept features. */ |
132 | vring_transport_features(vdev); | |
133 | ||
93d389f8 MT |
134 | /* Make sure we don't have any features > 32 bits! */ |
135 | BUG_ON((u32)vdev->features != vdev->features); | |
136 | ||
c624896e | 137 | /* We only support 32 feature bits. */ |
e16e12be | 138 | iowrite32(vdev->features, vp_dev->ioaddr + VIRTIO_PCI_GUEST_FEATURES); |
5c609a5e MT |
139 | |
140 | return 0; | |
3343660d AL |
141 | } |
142 | ||
143 | /* virtio config->get() implementation */ | |
144 | static void vp_get(struct virtio_device *vdev, unsigned offset, | |
145 | void *buf, unsigned len) | |
146 | { | |
147 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
82af8ce8 MT |
148 | void __iomem *ioaddr = vp_dev->ioaddr + |
149 | VIRTIO_PCI_CONFIG(vp_dev) + offset; | |
3343660d AL |
150 | u8 *ptr = buf; |
151 | int i; | |
152 | ||
153 | for (i = 0; i < len; i++) | |
154 | ptr[i] = ioread8(ioaddr + i); | |
155 | } | |
156 | ||
157 | /* the config->set() implementation. it's symmetric to the config->get() | |
158 | * implementation */ | |
159 | static void vp_set(struct virtio_device *vdev, unsigned offset, | |
160 | const void *buf, unsigned len) | |
161 | { | |
162 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
82af8ce8 MT |
163 | void __iomem *ioaddr = vp_dev->ioaddr + |
164 | VIRTIO_PCI_CONFIG(vp_dev) + offset; | |
3343660d AL |
165 | const u8 *ptr = buf; |
166 | int i; | |
167 | ||
168 | for (i = 0; i < len; i++) | |
169 | iowrite8(ptr[i], ioaddr + i); | |
170 | } | |
171 | ||
172 | /* config->{get,set}_status() implementations */ | |
173 | static u8 vp_get_status(struct virtio_device *vdev) | |
174 | { | |
175 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
176 | return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS); | |
177 | } | |
178 | ||
179 | static void vp_set_status(struct virtio_device *vdev, u8 status) | |
180 | { | |
181 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
182 | /* We should never be setting status to 0. */ | |
183 | BUG_ON(status == 0); | |
597d56e4 | 184 | iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS); |
3343660d AL |
185 | } |
186 | ||
e6af578c MT |
187 | /* wait for pending irq handlers */ |
188 | static void vp_synchronize_vectors(struct virtio_device *vdev) | |
189 | { | |
190 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
191 | int i; | |
192 | ||
193 | if (vp_dev->intx_enabled) | |
194 | synchronize_irq(vp_dev->pci_dev->irq); | |
195 | ||
196 | for (i = 0; i < vp_dev->msix_vectors; ++i) | |
197 | synchronize_irq(vp_dev->msix_entries[i].vector); | |
198 | } | |
199 | ||
3343660d AL |
200 | static void vp_reset(struct virtio_device *vdev) |
201 | { | |
202 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
203 | /* 0 status means a reset. */ | |
597d56e4 | 204 | iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS); |
e6af578c MT |
205 | /* Flush out the status write, and flush in device writes, |
206 | * including MSi-X interrupts, if any. */ | |
207 | ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS); | |
208 | /* Flush pending VQ/configuration callbacks. */ | |
209 | vp_synchronize_vectors(vdev); | |
3343660d AL |
210 | } |
211 | ||
212 | /* the notify function used when creating a virt queue */ | |
46f9c2b9 | 213 | static bool vp_notify(struct virtqueue *vq) |
3343660d | 214 | { |
3343660d AL |
215 | /* we write the queue's selector into the notification register to |
216 | * signal the other end */ | |
f30eaf4a | 217 | iowrite16(vq->index, (void __iomem *)vq->priv); |
46f9c2b9 | 218 | return true; |
3343660d AL |
219 | } |
220 | ||
77cf5246 MT |
221 | /* Handle a configuration change: Tell driver if it wants to know. */ |
222 | static irqreturn_t vp_config_changed(int irq, void *opaque) | |
223 | { | |
224 | struct virtio_pci_device *vp_dev = opaque; | |
77cf5246 | 225 | |
016c98c6 | 226 | virtio_config_changed(&vp_dev->vdev); |
77cf5246 MT |
227 | return IRQ_HANDLED; |
228 | } | |
229 | ||
230 | /* Notify all virtqueues on an interrupt. */ | |
231 | static irqreturn_t vp_vring_interrupt(int irq, void *opaque) | |
232 | { | |
233 | struct virtio_pci_device *vp_dev = opaque; | |
234 | struct virtio_pci_vq_info *info; | |
235 | irqreturn_t ret = IRQ_NONE; | |
236 | unsigned long flags; | |
237 | ||
238 | spin_lock_irqsave(&vp_dev->lock, flags); | |
239 | list_for_each_entry(info, &vp_dev->virtqueues, node) { | |
240 | if (vring_interrupt(irq, info->vq) == IRQ_HANDLED) | |
241 | ret = IRQ_HANDLED; | |
242 | } | |
243 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
244 | ||
245 | return ret; | |
246 | } | |
247 | ||
3343660d AL |
248 | /* A small wrapper to also acknowledge the interrupt when it's handled. |
249 | * I really need an EIO hook for the vring so I can ack the interrupt once we | |
250 | * know that we'll be handling the IRQ but before we invoke the callback since | |
251 | * the callback may notify the host which results in the host attempting to | |
252 | * raise an interrupt that we would then mask once we acknowledged the | |
253 | * interrupt. */ | |
254 | static irqreturn_t vp_interrupt(int irq, void *opaque) | |
255 | { | |
256 | struct virtio_pci_device *vp_dev = opaque; | |
3343660d AL |
257 | u8 isr; |
258 | ||
259 | /* reading the ISR has the effect of also clearing it so it's very | |
260 | * important to save off the value. */ | |
af535722 | 261 | isr = ioread8(vp_dev->isr); |
3343660d AL |
262 | |
263 | /* It's definitely not us if the ISR was not high */ | |
264 | if (!isr) | |
265 | return IRQ_NONE; | |
266 | ||
267 | /* Configuration change? Tell driver if it wants to know. */ | |
77cf5246 MT |
268 | if (isr & VIRTIO_PCI_ISR_CONFIG) |
269 | vp_config_changed(irq, opaque); | |
3343660d | 270 | |
77cf5246 | 271 | return vp_vring_interrupt(irq, opaque); |
3343660d AL |
272 | } |
273 | ||
82af8ce8 MT |
274 | static void vp_free_vectors(struct virtio_device *vdev) |
275 | { | |
276 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
277 | int i; | |
278 | ||
279 | if (vp_dev->intx_enabled) { | |
280 | free_irq(vp_dev->pci_dev->irq, vp_dev); | |
281 | vp_dev->intx_enabled = 0; | |
282 | } | |
283 | ||
284 | for (i = 0; i < vp_dev->msix_used_vectors; ++i) | |
285 | free_irq(vp_dev->msix_entries[i].vector, vp_dev); | |
82af8ce8 | 286 | |
75a0a52b JW |
287 | for (i = 0; i < vp_dev->msix_vectors; i++) |
288 | if (vp_dev->msix_affinity_masks[i]) | |
289 | free_cpumask_var(vp_dev->msix_affinity_masks[i]); | |
290 | ||
82af8ce8 MT |
291 | if (vp_dev->msix_enabled) { |
292 | /* Disable the vector used for configuration */ | |
293 | iowrite16(VIRTIO_MSI_NO_VECTOR, | |
294 | vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
295 | /* Flush the write out to device */ | |
296 | ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
297 | ||
82af8ce8 | 298 | pci_disable_msix(vp_dev->pci_dev); |
ff52c3fc | 299 | vp_dev->msix_enabled = 0; |
82af8ce8 | 300 | } |
ff52c3fc | 301 | |
f11335db | 302 | vp_dev->msix_vectors = 0; |
ff52c3fc MT |
303 | vp_dev->msix_used_vectors = 0; |
304 | kfree(vp_dev->msix_names); | |
305 | vp_dev->msix_names = NULL; | |
306 | kfree(vp_dev->msix_entries); | |
307 | vp_dev->msix_entries = NULL; | |
75a0a52b JW |
308 | kfree(vp_dev->msix_affinity_masks); |
309 | vp_dev->msix_affinity_masks = NULL; | |
82af8ce8 MT |
310 | } |
311 | ||
f68d2408 RR |
312 | static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors, |
313 | bool per_vq_vectors) | |
82af8ce8 MT |
314 | { |
315 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
316 | const char *name = dev_name(&vp_dev->vdev.dev); | |
317 | unsigned i, v; | |
318 | int err = -ENOMEM; | |
e969fed5 | 319 | |
f11335db AV |
320 | vp_dev->msix_vectors = nvectors; |
321 | ||
82af8ce8 MT |
322 | vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries, |
323 | GFP_KERNEL); | |
324 | if (!vp_dev->msix_entries) | |
ff52c3fc | 325 | goto error; |
82af8ce8 MT |
326 | vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names, |
327 | GFP_KERNEL); | |
328 | if (!vp_dev->msix_names) | |
ff52c3fc | 329 | goto error; |
75a0a52b JW |
330 | vp_dev->msix_affinity_masks |
331 | = kzalloc(nvectors * sizeof *vp_dev->msix_affinity_masks, | |
332 | GFP_KERNEL); | |
333 | if (!vp_dev->msix_affinity_masks) | |
334 | goto error; | |
335 | for (i = 0; i < nvectors; ++i) | |
336 | if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i], | |
337 | GFP_KERNEL)) | |
338 | goto error; | |
82af8ce8 MT |
339 | |
340 | for (i = 0; i < nvectors; ++i) | |
341 | vp_dev->msix_entries[i].entry = i; | |
342 | ||
5e37f670 AG |
343 | err = pci_enable_msix_exact(vp_dev->pci_dev, |
344 | vp_dev->msix_entries, nvectors); | |
e969fed5 MT |
345 | if (err) |
346 | goto error; | |
e969fed5 MT |
347 | vp_dev->msix_enabled = 1; |
348 | ||
349 | /* Set the vector used for configuration */ | |
350 | v = vp_dev->msix_used_vectors; | |
351 | snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names, | |
352 | "%s-config", name); | |
353 | err = request_irq(vp_dev->msix_entries[v].vector, | |
354 | vp_config_changed, 0, vp_dev->msix_names[v], | |
355 | vp_dev); | |
356 | if (err) | |
357 | goto error; | |
358 | ++vp_dev->msix_used_vectors; | |
82af8ce8 | 359 | |
e969fed5 MT |
360 | iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); |
361 | /* Verify we had enough resources to assign the vector */ | |
362 | v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR); | |
363 | if (v == VIRTIO_MSI_NO_VECTOR) { | |
364 | err = -EBUSY; | |
365 | goto error; | |
82af8ce8 MT |
366 | } |
367 | ||
e969fed5 | 368 | if (!per_vq_vectors) { |
82af8ce8 MT |
369 | /* Shared vector for all VQs */ |
370 | v = vp_dev->msix_used_vectors; | |
371 | snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names, | |
372 | "%s-virtqueues", name); | |
373 | err = request_irq(vp_dev->msix_entries[v].vector, | |
374 | vp_vring_interrupt, 0, vp_dev->msix_names[v], | |
375 | vp_dev); | |
376 | if (err) | |
ff52c3fc | 377 | goto error; |
82af8ce8 MT |
378 | ++vp_dev->msix_used_vectors; |
379 | } | |
380 | return 0; | |
ff52c3fc | 381 | error: |
82af8ce8 | 382 | vp_free_vectors(vdev); |
82af8ce8 MT |
383 | return err; |
384 | } | |
385 | ||
f68d2408 RR |
386 | static int vp_request_intx(struct virtio_device *vdev) |
387 | { | |
388 | int err; | |
389 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
390 | ||
391 | err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, | |
392 | IRQF_SHARED, dev_name(&vdev->dev), vp_dev); | |
393 | if (!err) | |
394 | vp_dev->intx_enabled = 1; | |
395 | return err; | |
396 | } | |
397 | ||
b09f00bb MT |
398 | static struct virtqueue *setup_vq(struct virtio_pci_device *vp_dev, |
399 | struct virtio_pci_vq_info *info, | |
400 | unsigned index, | |
f68d2408 RR |
401 | void (*callback)(struct virtqueue *vq), |
402 | const char *name, | |
403 | u16 msix_vec) | |
3343660d | 404 | { |
3343660d | 405 | struct virtqueue *vq; |
b09f00bb | 406 | unsigned long size; |
e969fed5 | 407 | u16 num; |
3343660d AL |
408 | int err; |
409 | ||
410 | /* Select the queue we're interested in */ | |
411 | iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); | |
412 | ||
413 | /* Check if queue is either not available or already active. */ | |
414 | num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM); | |
415 | if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN)) | |
416 | return ERR_PTR(-ENOENT); | |
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); |
b09f00bb MT |
423 | if (info->queue == NULL) |
424 | return ERR_PTR(-ENOMEM); | |
3343660d AL |
425 | |
426 | /* activate the queue */ | |
480daab4 | 427 | iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT, |
3343660d AL |
428 | vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); |
429 | ||
430 | /* create the vring */ | |
b09f00bb MT |
431 | vq = vring_new_virtqueue(index, info->num, |
432 | VIRTIO_PCI_VRING_ALIGN, &vp_dev->vdev, | |
7b21e34f | 433 | true, info->queue, vp_notify, callback, name); |
3343660d AL |
434 | if (!vq) { |
435 | err = -ENOMEM; | |
436 | goto out_activate_queue; | |
437 | } | |
438 | ||
f30eaf4a | 439 | vq->priv = (void __force *)vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY; |
3343660d | 440 | |
f68d2408 RR |
441 | if (msix_vec != VIRTIO_MSI_NO_VECTOR) { |
442 | iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
443 | msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
444 | if (msix_vec == VIRTIO_MSI_NO_VECTOR) { | |
82af8ce8 MT |
445 | err = -EBUSY; |
446 | goto out_assign; | |
447 | } | |
448 | } | |
449 | ||
b09f00bb MT |
450 | return vq; |
451 | ||
452 | out_assign: | |
453 | vring_del_virtqueue(vq); | |
454 | out_activate_queue: | |
455 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); | |
456 | free_pages_exact(info->queue, size); | |
457 | return ERR_PTR(err); | |
458 | } | |
459 | ||
460 | static struct virtqueue *vp_setup_vq(struct virtio_device *vdev, unsigned index, | |
461 | void (*callback)(struct virtqueue *vq), | |
462 | const char *name, | |
463 | u16 msix_vec) | |
464 | { | |
465 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
466 | struct virtio_pci_vq_info *info = kmalloc(sizeof *info, GFP_KERNEL); | |
467 | struct virtqueue *vq; | |
468 | unsigned long flags; | |
469 | ||
470 | /* fill out our structure that represents an active queue */ | |
471 | if (!info) | |
472 | return ERR_PTR(-ENOMEM); | |
473 | ||
474 | vq = vp_dev->setup_vq(vp_dev, info, index, callback, name, msix_vec); | |
475 | if (IS_ERR(vq)) | |
476 | goto out_info; | |
477 | ||
478 | info->vq = vq; | |
005b20a8 KK |
479 | if (callback) { |
480 | spin_lock_irqsave(&vp_dev->lock, flags); | |
481 | list_add(&info->node, &vp_dev->virtqueues); | |
482 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
483 | } else { | |
484 | INIT_LIST_HEAD(&info->node); | |
485 | } | |
3343660d | 486 | |
3ec7a77b | 487 | vp_dev->vqs[index] = info; |
3343660d AL |
488 | return vq; |
489 | ||
3343660d AL |
490 | out_info: |
491 | kfree(info); | |
b09f00bb | 492 | return vq; |
3343660d AL |
493 | } |
494 | ||
5386cef2 | 495 | static void del_vq(struct virtio_pci_vq_info *info) |
3343660d | 496 | { |
5386cef2 | 497 | struct virtqueue *vq = info->vq; |
3343660d | 498 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); |
5386cef2 | 499 | unsigned long size; |
3343660d | 500 | |
06ca287d | 501 | iowrite16(vq->index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL); |
82af8ce8 | 502 | |
82af8ce8 MT |
503 | if (vp_dev->msix_enabled) { |
504 | iowrite16(VIRTIO_MSI_NO_VECTOR, | |
505 | vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR); | |
506 | /* Flush the write out to device */ | |
507 | ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR); | |
508 | } | |
509 | ||
3343660d AL |
510 | vring_del_virtqueue(vq); |
511 | ||
512 | /* Select and deactivate the queue */ | |
3343660d AL |
513 | iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN); |
514 | ||
498af147 | 515 | size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN)); |
13b1eb33 | 516 | free_pages_exact(info->queue, size); |
5386cef2 MT |
517 | } |
518 | ||
519 | static void vp_del_vq(struct virtqueue *vq) | |
520 | { | |
521 | struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev); | |
522 | struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index]; | |
523 | unsigned long flags; | |
524 | ||
525 | spin_lock_irqsave(&vp_dev->lock, flags); | |
526 | list_del(&info->node); | |
527 | spin_unlock_irqrestore(&vp_dev->lock, flags); | |
528 | ||
529 | vp_dev->del_vq(info); | |
3343660d AL |
530 | kfree(info); |
531 | } | |
532 | ||
82af8ce8 | 533 | /* the config->del_vqs() implementation */ |
d2a7ddda MT |
534 | static void vp_del_vqs(struct virtio_device *vdev) |
535 | { | |
e969fed5 | 536 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); |
d2a7ddda | 537 | struct virtqueue *vq, *n; |
e969fed5 | 538 | struct virtio_pci_vq_info *info; |
d2a7ddda | 539 | |
e969fed5 | 540 | list_for_each_entry_safe(vq, n, &vdev->vqs, list) { |
3ec7a77b | 541 | info = vp_dev->vqs[vq->index]; |
31198159 MT |
542 | if (vp_dev->per_vq_vectors && |
543 | info->msix_vector != VIRTIO_MSI_NO_VECTOR) | |
f68d2408 RR |
544 | free_irq(vp_dev->msix_entries[info->msix_vector].vector, |
545 | vq); | |
d2a7ddda | 546 | vp_del_vq(vq); |
e969fed5 MT |
547 | } |
548 | vp_dev->per_vq_vectors = false; | |
82af8ce8 MT |
549 | |
550 | vp_free_vectors(vdev); | |
3ec7a77b | 551 | kfree(vp_dev->vqs); |
d2a7ddda MT |
552 | } |
553 | ||
e969fed5 MT |
554 | static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs, |
555 | struct virtqueue *vqs[], | |
556 | vq_callback_t *callbacks[], | |
557 | const char *names[], | |
f68d2408 | 558 | bool use_msix, |
e969fed5 | 559 | bool per_vq_vectors) |
d2a7ddda | 560 | { |
e969fed5 | 561 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); |
f68d2408 RR |
562 | u16 msix_vec; |
563 | int i, err, nvectors, allocated_vectors; | |
82af8ce8 | 564 | |
3ec7a77b MT |
565 | vp_dev->vqs = kmalloc(nvqs * sizeof *vp_dev->vqs, GFP_KERNEL); |
566 | if (!vp_dev->vqs) | |
567 | return -ENOMEM; | |
568 | ||
f68d2408 RR |
569 | if (!use_msix) { |
570 | /* Old style: one normal interrupt for change and all vqs. */ | |
571 | err = vp_request_intx(vdev); | |
572 | if (err) | |
3ec7a77b | 573 | goto error_find; |
f68d2408 RR |
574 | } else { |
575 | if (per_vq_vectors) { | |
576 | /* Best option: one for change interrupt, one per vq. */ | |
577 | nvectors = 1; | |
578 | for (i = 0; i < nvqs; ++i) | |
579 | if (callbacks[i]) | |
580 | ++nvectors; | |
581 | } else { | |
582 | /* Second best: one for change, shared for all vqs. */ | |
583 | nvectors = 2; | |
584 | } | |
585 | ||
586 | err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors); | |
587 | if (err) | |
3ec7a77b | 588 | goto error_find; |
f68d2408 | 589 | } |
d2a7ddda | 590 | |
e969fed5 MT |
591 | vp_dev->per_vq_vectors = per_vq_vectors; |
592 | allocated_vectors = vp_dev->msix_used_vectors; | |
d2a7ddda | 593 | for (i = 0; i < nvqs; ++i) { |
6457f126 MT |
594 | if (!names[i]) { |
595 | vqs[i] = NULL; | |
596 | continue; | |
597 | } else if (!callbacks[i] || !vp_dev->msix_enabled) | |
f68d2408 | 598 | msix_vec = VIRTIO_MSI_NO_VECTOR; |
e969fed5 | 599 | else if (vp_dev->per_vq_vectors) |
f68d2408 | 600 | msix_vec = allocated_vectors++; |
e969fed5 | 601 | else |
f68d2408 | 602 | msix_vec = VP_MSIX_VQ_VECTOR; |
b09f00bb | 603 | vqs[i] = vp_setup_vq(vdev, i, callbacks[i], names[i], msix_vec); |
e969fed5 MT |
604 | if (IS_ERR(vqs[i])) { |
605 | err = PTR_ERR(vqs[i]); | |
82af8ce8 | 606 | goto error_find; |
e969fed5 | 607 | } |
0b22bd0b MT |
608 | |
609 | if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR) | |
610 | continue; | |
611 | ||
e969fed5 | 612 | /* allocate per-vq irq if available and necessary */ |
0b22bd0b MT |
613 | snprintf(vp_dev->msix_names[msix_vec], |
614 | sizeof *vp_dev->msix_names, | |
615 | "%s-%s", | |
616 | dev_name(&vp_dev->vdev.dev), names[i]); | |
617 | err = request_irq(vp_dev->msix_entries[msix_vec].vector, | |
618 | vring_interrupt, 0, | |
619 | vp_dev->msix_names[msix_vec], | |
620 | vqs[i]); | |
621 | if (err) { | |
622 | vp_del_vq(vqs[i]); | |
623 | goto error_find; | |
e969fed5 | 624 | } |
d2a7ddda MT |
625 | } |
626 | return 0; | |
627 | ||
82af8ce8 | 628 | error_find: |
d2a7ddda | 629 | vp_del_vqs(vdev); |
e969fed5 MT |
630 | return err; |
631 | } | |
632 | ||
633 | /* the config->find_vqs() implementation */ | |
634 | static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs, | |
635 | struct virtqueue *vqs[], | |
636 | vq_callback_t *callbacks[], | |
637 | const char *names[]) | |
638 | { | |
f68d2408 | 639 | int err; |
e969fed5 | 640 | |
f68d2408 RR |
641 | /* Try MSI-X with one vector per queue. */ |
642 | err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true); | |
e969fed5 MT |
643 | if (!err) |
644 | return 0; | |
f68d2408 | 645 | /* Fallback: MSI-X with one vector for config, one shared for queues. */ |
e969fed5 | 646 | err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, |
f68d2408 | 647 | true, false); |
e969fed5 MT |
648 | if (!err) |
649 | return 0; | |
650 | /* Finally fall back to regular interrupts. */ | |
f68d2408 RR |
651 | return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, |
652 | false, false); | |
d2a7ddda MT |
653 | } |
654 | ||
66846048 RJ |
655 | static const char *vp_bus_name(struct virtio_device *vdev) |
656 | { | |
657 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
658 | ||
659 | return pci_name(vp_dev->pci_dev); | |
660 | } | |
661 | ||
75a0a52b JW |
662 | /* Setup the affinity for a virtqueue: |
663 | * - force the affinity for per vq vector | |
664 | * - OR over all affinities for shared MSI | |
665 | * - ignore the affinity request if we're using INTX | |
666 | */ | |
667 | static int vp_set_vq_affinity(struct virtqueue *vq, int cpu) | |
668 | { | |
669 | struct virtio_device *vdev = vq->vdev; | |
670 | struct virtio_pci_device *vp_dev = to_vp_device(vdev); | |
3ec7a77b | 671 | struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index]; |
75a0a52b JW |
672 | struct cpumask *mask; |
673 | unsigned int irq; | |
674 | ||
675 | if (!vq->callback) | |
676 | return -EINVAL; | |
677 | ||
678 | if (vp_dev->msix_enabled) { | |
679 | mask = vp_dev->msix_affinity_masks[info->msix_vector]; | |
680 | irq = vp_dev->msix_entries[info->msix_vector].vector; | |
681 | if (cpu == -1) | |
682 | irq_set_affinity_hint(irq, NULL); | |
683 | else { | |
684 | cpumask_set_cpu(cpu, mask); | |
685 | irq_set_affinity_hint(irq, mask); | |
686 | } | |
687 | } | |
688 | return 0; | |
689 | } | |
690 | ||
93503932 | 691 | static const struct virtio_config_ops virtio_pci_config_ops = { |
3343660d AL |
692 | .get = vp_get, |
693 | .set = vp_set, | |
694 | .get_status = vp_get_status, | |
695 | .set_status = vp_set_status, | |
696 | .reset = vp_reset, | |
d2a7ddda MT |
697 | .find_vqs = vp_find_vqs, |
698 | .del_vqs = vp_del_vqs, | |
c45a6816 | 699 | .get_features = vp_get_features, |
c624896e | 700 | .finalize_features = vp_finalize_features, |
66846048 | 701 | .bus_name = vp_bus_name, |
75a0a52b | 702 | .set_vq_affinity = vp_set_vq_affinity, |
3343660d AL |
703 | }; |
704 | ||
29f9f12e MM |
705 | static void virtio_pci_release_dev(struct device *_d) |
706 | { | |
72103bd1 MT |
707 | /* |
708 | * No need for a release method as we allocate/free | |
709 | * all devices together with the pci devices. | |
710 | * Provide an empty one to avoid getting a warning from core. | |
711 | */ | |
29f9f12e MM |
712 | } |
713 | ||
3343660d | 714 | /* the PCI probing function */ |
8590dbc7 GKH |
715 | static int virtio_pci_probe(struct pci_dev *pci_dev, |
716 | const struct pci_device_id *id) | |
3343660d AL |
717 | { |
718 | struct virtio_pci_device *vp_dev; | |
719 | int err; | |
720 | ||
721 | /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */ | |
722 | if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f) | |
723 | return -ENODEV; | |
724 | ||
55a7c066 AL |
725 | if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) { |
726 | printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n", | |
727 | VIRTIO_PCI_ABI_VERSION, pci_dev->revision); | |
728 | return -ENODEV; | |
729 | } | |
730 | ||
3343660d AL |
731 | /* allocate our structure and fill it out */ |
732 | vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL); | |
733 | if (vp_dev == NULL) | |
734 | return -ENOMEM; | |
735 | ||
8b3bb3ec | 736 | vp_dev->vdev.dev.parent = &pci_dev->dev; |
29f9f12e | 737 | vp_dev->vdev.dev.release = virtio_pci_release_dev; |
3343660d AL |
738 | vp_dev->vdev.config = &virtio_pci_config_ops; |
739 | vp_dev->pci_dev = pci_dev; | |
740 | INIT_LIST_HEAD(&vp_dev->virtqueues); | |
741 | spin_lock_init(&vp_dev->lock); | |
742 | ||
b03214d5 MT |
743 | /* Disable MSI/MSIX to bring device to a known good state. */ |
744 | pci_msi_off(pci_dev); | |
745 | ||
3343660d AL |
746 | /* enable the device */ |
747 | err = pci_enable_device(pci_dev); | |
748 | if (err) | |
749 | goto out; | |
750 | ||
751 | err = pci_request_regions(pci_dev, "virtio-pci"); | |
752 | if (err) | |
753 | goto out_enable_device; | |
754 | ||
755 | vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0); | |
74a74b37 PST |
756 | if (vp_dev->ioaddr == NULL) { |
757 | err = -ENOMEM; | |
3343660d | 758 | goto out_req_regions; |
74a74b37 | 759 | } |
3343660d | 760 | |
af535722 MT |
761 | vp_dev->isr = vp_dev->ioaddr + VIRTIO_PCI_ISR; |
762 | ||
3343660d | 763 | pci_set_drvdata(pci_dev, vp_dev); |
bc505f37 | 764 | pci_set_master(pci_dev); |
3343660d AL |
765 | |
766 | /* we use the subsystem vendor/device id as the virtio vendor/device | |
767 | * id. this allows us to use the same PCI vendor/device id for all | |
768 | * virtio devices and to identify the particular virtio driver by | |
88393161 | 769 | * the subsystem ids */ |
3343660d AL |
770 | vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor; |
771 | vp_dev->vdev.id.device = pci_dev->subsystem_device; | |
772 | ||
b09f00bb | 773 | vp_dev->setup_vq = setup_vq; |
5386cef2 MT |
774 | vp_dev->del_vq = del_vq; |
775 | ||
3343660d AL |
776 | /* finally register the virtio device */ |
777 | err = register_virtio_device(&vp_dev->vdev); | |
778 | if (err) | |
82af8ce8 | 779 | goto out_set_drvdata; |
3343660d AL |
780 | |
781 | return 0; | |
782 | ||
3343660d | 783 | out_set_drvdata: |
3343660d AL |
784 | pci_iounmap(pci_dev, vp_dev->ioaddr); |
785 | out_req_regions: | |
786 | pci_release_regions(pci_dev); | |
787 | out_enable_device: | |
788 | pci_disable_device(pci_dev); | |
789 | out: | |
790 | kfree(vp_dev); | |
791 | return err; | |
792 | } | |
793 | ||
8590dbc7 | 794 | static void virtio_pci_remove(struct pci_dev *pci_dev) |
3343660d AL |
795 | { |
796 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
797 | ||
bd6c2690 | 798 | unregister_virtio_device(&vp_dev->vdev); |
31a3ddda AS |
799 | |
800 | vp_del_vqs(&vp_dev->vdev); | |
31a3ddda AS |
801 | pci_iounmap(pci_dev, vp_dev->ioaddr); |
802 | pci_release_regions(pci_dev); | |
803 | pci_disable_device(pci_dev); | |
72103bd1 | 804 | kfree(vp_dev); |
3343660d AL |
805 | } |
806 | ||
9e266ece | 807 | #ifdef CONFIG_PM_SLEEP |
f0fe6f11 AS |
808 | static int virtio_pci_freeze(struct device *dev) |
809 | { | |
810 | struct pci_dev *pci_dev = to_pci_dev(dev); | |
811 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
f0fe6f11 AS |
812 | int ret; |
813 | ||
c6716bae | 814 | ret = virtio_device_freeze(&vp_dev->vdev); |
f0fe6f11 AS |
815 | |
816 | if (!ret) | |
817 | pci_disable_device(pci_dev); | |
818 | return ret; | |
819 | } | |
820 | ||
0517fdd1 | 821 | static int virtio_pci_restore(struct device *dev) |
f0fe6f11 AS |
822 | { |
823 | struct pci_dev *pci_dev = to_pci_dev(dev); | |
824 | struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev); | |
825 | int ret; | |
826 | ||
827 | ret = pci_enable_device(pci_dev); | |
828 | if (ret) | |
829 | return ret; | |
0517fdd1 | 830 | |
f0fe6f11 | 831 | pci_set_master(pci_dev); |
c6716bae | 832 | return virtio_device_restore(&vp_dev->vdev); |
f0fe6f11 AS |
833 | } |
834 | ||
d0775363 | 835 | static const struct dev_pm_ops virtio_pci_pm_ops = { |
f878d0be | 836 | SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore) |
d0775363 | 837 | }; |
3343660d AL |
838 | #endif |
839 | ||
840 | static struct pci_driver virtio_pci_driver = { | |
841 | .name = "virtio-pci", | |
842 | .id_table = virtio_pci_id_table, | |
843 | .probe = virtio_pci_probe, | |
8590dbc7 | 844 | .remove = virtio_pci_remove, |
9e266ece | 845 | #ifdef CONFIG_PM_SLEEP |
d0775363 | 846 | .driver.pm = &virtio_pci_pm_ops, |
3343660d AL |
847 | #endif |
848 | }; | |
849 | ||
1ce6853a | 850 | module_pci_driver(virtio_pci_driver); |