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