Loading...
1/*
2 * Virtio PCI driver - modern (virtio 1.0) device support
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 * Copyright Red Hat, Inc. 2014
9 *
10 * Authors:
11 * Anthony Liguori <aliguori@us.ibm.com>
12 * Rusty Russell <rusty@rustcorp.com.au>
13 * Michael S. Tsirkin <mst@redhat.com>
14 *
15 * This work is licensed under the terms of the GNU GPL, version 2 or later.
16 * See the COPYING file in the top-level directory.
17 *
18 */
19
20#include <linux/delay.h>
21#define VIRTIO_PCI_NO_LEGACY
22#include "virtio_pci_common.h"
23
24/*
25 * Type-safe wrappers for io accesses.
26 * Use these to enforce at compile time the following spec requirement:
27 *
28 * The driver MUST access each field using the “natural” access
29 * method, i.e. 32-bit accesses for 32-bit fields, 16-bit accesses
30 * for 16-bit fields and 8-bit accesses for 8-bit fields.
31 */
32static inline u8 vp_ioread8(u8 __iomem *addr)
33{
34 return ioread8(addr);
35}
36static inline u16 vp_ioread16 (u16 __iomem *addr)
37{
38 return ioread16(addr);
39}
40
41static inline u32 vp_ioread32(u32 __iomem *addr)
42{
43 return ioread32(addr);
44}
45
46static inline void vp_iowrite8(u8 value, u8 __iomem *addr)
47{
48 iowrite8(value, addr);
49}
50
51static inline void vp_iowrite16(u16 value, u16 __iomem *addr)
52{
53 iowrite16(value, addr);
54}
55
56static inline void vp_iowrite32(u32 value, u32 __iomem *addr)
57{
58 iowrite32(value, addr);
59}
60
61static void vp_iowrite64_twopart(u64 val,
62 __le32 __iomem *lo, __le32 __iomem *hi)
63{
64 vp_iowrite32((u32)val, lo);
65 vp_iowrite32(val >> 32, hi);
66}
67
68static void __iomem *map_capability(struct pci_dev *dev, int off,
69 size_t minlen,
70 u32 align,
71 u32 start, u32 size,
72 size_t *len)
73{
74 u8 bar;
75 u32 offset, length;
76 void __iomem *p;
77
78 pci_read_config_byte(dev, off + offsetof(struct virtio_pci_cap,
79 bar),
80 &bar);
81 pci_read_config_dword(dev, off + offsetof(struct virtio_pci_cap, offset),
82 &offset);
83 pci_read_config_dword(dev, off + offsetof(struct virtio_pci_cap, length),
84 &length);
85
86 if (length <= start) {
87 dev_err(&dev->dev,
88 "virtio_pci: bad capability len %u (>%u expected)\n",
89 length, start);
90 return NULL;
91 }
92
93 if (length - start < minlen) {
94 dev_err(&dev->dev,
95 "virtio_pci: bad capability len %u (>=%zu expected)\n",
96 length, minlen);
97 return NULL;
98 }
99
100 length -= start;
101
102 if (start + offset < offset) {
103 dev_err(&dev->dev,
104 "virtio_pci: map wrap-around %u+%u\n",
105 start, offset);
106 return NULL;
107 }
108
109 offset += start;
110
111 if (offset & (align - 1)) {
112 dev_err(&dev->dev,
113 "virtio_pci: offset %u not aligned to %u\n",
114 offset, align);
115 return NULL;
116 }
117
118 if (length > size)
119 length = size;
120
121 if (len)
122 *len = length;
123
124 if (minlen + offset < minlen ||
125 minlen + offset > pci_resource_len(dev, bar)) {
126 dev_err(&dev->dev,
127 "virtio_pci: map virtio %zu@%u "
128 "out of range on bar %i length %lu\n",
129 minlen, offset,
130 bar, (unsigned long)pci_resource_len(dev, bar));
131 return NULL;
132 }
133
134 p = pci_iomap_range(dev, bar, offset, length);
135 if (!p)
136 dev_err(&dev->dev,
137 "virtio_pci: unable to map virtio %u@%u on bar %i\n",
138 length, offset, bar);
139 return p;
140}
141
142/* virtio config->get_features() implementation */
143static u64 vp_get_features(struct virtio_device *vdev)
144{
145 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
146 u64 features;
147
148 vp_iowrite32(0, &vp_dev->common->device_feature_select);
149 features = vp_ioread32(&vp_dev->common->device_feature);
150 vp_iowrite32(1, &vp_dev->common->device_feature_select);
151 features |= ((u64)vp_ioread32(&vp_dev->common->device_feature) << 32);
152
153 return features;
154}
155
156/* virtio config->finalize_features() implementation */
157static int vp_finalize_features(struct virtio_device *vdev)
158{
159 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
160
161 /* Give virtio_ring a chance to accept features. */
162 vring_transport_features(vdev);
163
164 if (!__virtio_test_bit(vdev, VIRTIO_F_VERSION_1)) {
165 dev_err(&vdev->dev, "virtio: device uses modern interface "
166 "but does not have VIRTIO_F_VERSION_1\n");
167 return -EINVAL;
168 }
169
170 vp_iowrite32(0, &vp_dev->common->guest_feature_select);
171 vp_iowrite32((u32)vdev->features, &vp_dev->common->guest_feature);
172 vp_iowrite32(1, &vp_dev->common->guest_feature_select);
173 vp_iowrite32(vdev->features >> 32, &vp_dev->common->guest_feature);
174
175 return 0;
176}
177
178/* virtio config->get() implementation */
179static void vp_get(struct virtio_device *vdev, unsigned offset,
180 void *buf, unsigned len)
181{
182 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
183 u8 b;
184 __le16 w;
185 __le32 l;
186
187 BUG_ON(offset + len > vp_dev->device_len);
188
189 switch (len) {
190 case 1:
191 b = ioread8(vp_dev->device + offset);
192 memcpy(buf, &b, sizeof b);
193 break;
194 case 2:
195 w = cpu_to_le16(ioread16(vp_dev->device + offset));
196 memcpy(buf, &w, sizeof w);
197 break;
198 case 4:
199 l = cpu_to_le32(ioread32(vp_dev->device + offset));
200 memcpy(buf, &l, sizeof l);
201 break;
202 case 8:
203 l = cpu_to_le32(ioread32(vp_dev->device + offset));
204 memcpy(buf, &l, sizeof l);
205 l = cpu_to_le32(ioread32(vp_dev->device + offset + sizeof l));
206 memcpy(buf + sizeof l, &l, sizeof l);
207 break;
208 default:
209 BUG();
210 }
211}
212
213/* the config->set() implementation. it's symmetric to the config->get()
214 * implementation */
215static void vp_set(struct virtio_device *vdev, unsigned offset,
216 const void *buf, unsigned len)
217{
218 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
219 u8 b;
220 __le16 w;
221 __le32 l;
222
223 BUG_ON(offset + len > vp_dev->device_len);
224
225 switch (len) {
226 case 1:
227 memcpy(&b, buf, sizeof b);
228 iowrite8(b, vp_dev->device + offset);
229 break;
230 case 2:
231 memcpy(&w, buf, sizeof w);
232 iowrite16(le16_to_cpu(w), vp_dev->device + offset);
233 break;
234 case 4:
235 memcpy(&l, buf, sizeof l);
236 iowrite32(le32_to_cpu(l), vp_dev->device + offset);
237 break;
238 case 8:
239 memcpy(&l, buf, sizeof l);
240 iowrite32(le32_to_cpu(l), vp_dev->device + offset);
241 memcpy(&l, buf + sizeof l, sizeof l);
242 iowrite32(le32_to_cpu(l), vp_dev->device + offset + sizeof l);
243 break;
244 default:
245 BUG();
246 }
247}
248
249static u32 vp_generation(struct virtio_device *vdev)
250{
251 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
252 return vp_ioread8(&vp_dev->common->config_generation);
253}
254
255/* config->{get,set}_status() implementations */
256static u8 vp_get_status(struct virtio_device *vdev)
257{
258 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
259 return vp_ioread8(&vp_dev->common->device_status);
260}
261
262static void vp_set_status(struct virtio_device *vdev, u8 status)
263{
264 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
265 /* We should never be setting status to 0. */
266 BUG_ON(status == 0);
267 vp_iowrite8(status, &vp_dev->common->device_status);
268}
269
270static void vp_reset(struct virtio_device *vdev)
271{
272 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
273 /* 0 status means a reset. */
274 vp_iowrite8(0, &vp_dev->common->device_status);
275 /* After writing 0 to device_status, the driver MUST wait for a read of
276 * device_status to return 0 before reinitializing the device.
277 * This will flush out the status write, and flush in device writes,
278 * including MSI-X interrupts, if any.
279 */
280 while (vp_ioread8(&vp_dev->common->device_status))
281 msleep(1);
282 /* Flush pending VQ/configuration callbacks. */
283 vp_synchronize_vectors(vdev);
284}
285
286static u16 vp_config_vector(struct virtio_pci_device *vp_dev, u16 vector)
287{
288 /* Setup the vector used for configuration events */
289 vp_iowrite16(vector, &vp_dev->common->msix_config);
290 /* Verify we had enough resources to assign the vector */
291 /* Will also flush the write out to device */
292 return vp_ioread16(&vp_dev->common->msix_config);
293}
294
295static struct virtqueue *setup_vq(struct virtio_pci_device *vp_dev,
296 struct virtio_pci_vq_info *info,
297 unsigned index,
298 void (*callback)(struct virtqueue *vq),
299 const char *name,
300 u16 msix_vec)
301{
302 struct virtio_pci_common_cfg __iomem *cfg = vp_dev->common;
303 struct virtqueue *vq;
304 u16 num, off;
305 int err;
306
307 if (index >= vp_ioread16(&cfg->num_queues))
308 return ERR_PTR(-ENOENT);
309
310 /* Select the queue we're interested in */
311 vp_iowrite16(index, &cfg->queue_select);
312
313 /* Check if queue is either not available or already active. */
314 num = vp_ioread16(&cfg->queue_size);
315 if (!num || vp_ioread16(&cfg->queue_enable))
316 return ERR_PTR(-ENOENT);
317
318 if (num & (num - 1)) {
319 dev_warn(&vp_dev->pci_dev->dev, "bad queue size %u", num);
320 return ERR_PTR(-EINVAL);
321 }
322
323 /* get offset of notification word for this vq */
324 off = vp_ioread16(&cfg->queue_notify_off);
325
326 info->msix_vector = msix_vec;
327
328 /* create the vring */
329 vq = vring_create_virtqueue(index, num,
330 SMP_CACHE_BYTES, &vp_dev->vdev,
331 true, true, vp_notify, callback, name);
332 if (!vq)
333 return ERR_PTR(-ENOMEM);
334
335 /* activate the queue */
336 vp_iowrite16(virtqueue_get_vring_size(vq), &cfg->queue_size);
337 vp_iowrite64_twopart(virtqueue_get_desc_addr(vq),
338 &cfg->queue_desc_lo, &cfg->queue_desc_hi);
339 vp_iowrite64_twopart(virtqueue_get_avail_addr(vq),
340 &cfg->queue_avail_lo, &cfg->queue_avail_hi);
341 vp_iowrite64_twopart(virtqueue_get_used_addr(vq),
342 &cfg->queue_used_lo, &cfg->queue_used_hi);
343
344 if (vp_dev->notify_base) {
345 /* offset should not wrap */
346 if ((u64)off * vp_dev->notify_offset_multiplier + 2
347 > vp_dev->notify_len) {
348 dev_warn(&vp_dev->pci_dev->dev,
349 "bad notification offset %u (x %u) "
350 "for queue %u > %zd",
351 off, vp_dev->notify_offset_multiplier,
352 index, vp_dev->notify_len);
353 err = -EINVAL;
354 goto err_map_notify;
355 }
356 vq->priv = (void __force *)vp_dev->notify_base +
357 off * vp_dev->notify_offset_multiplier;
358 } else {
359 vq->priv = (void __force *)map_capability(vp_dev->pci_dev,
360 vp_dev->notify_map_cap, 2, 2,
361 off * vp_dev->notify_offset_multiplier, 2,
362 NULL);
363 }
364
365 if (!vq->priv) {
366 err = -ENOMEM;
367 goto err_map_notify;
368 }
369
370 if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
371 vp_iowrite16(msix_vec, &cfg->queue_msix_vector);
372 msix_vec = vp_ioread16(&cfg->queue_msix_vector);
373 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
374 err = -EBUSY;
375 goto err_assign_vector;
376 }
377 }
378
379 return vq;
380
381err_assign_vector:
382 if (!vp_dev->notify_base)
383 pci_iounmap(vp_dev->pci_dev, (void __iomem __force *)vq->priv);
384err_map_notify:
385 vring_del_virtqueue(vq);
386 return ERR_PTR(err);
387}
388
389static int vp_modern_find_vqs(struct virtio_device *vdev, unsigned nvqs,
390 struct virtqueue *vqs[],
391 vq_callback_t *callbacks[],
392 const char * const names[])
393{
394 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
395 struct virtqueue *vq;
396 int rc = vp_find_vqs(vdev, nvqs, vqs, callbacks, names);
397
398 if (rc)
399 return rc;
400
401 /* Select and activate all queues. Has to be done last: once we do
402 * this, there's no way to go back except reset.
403 */
404 list_for_each_entry(vq, &vdev->vqs, list) {
405 vp_iowrite16(vq->index, &vp_dev->common->queue_select);
406 vp_iowrite16(1, &vp_dev->common->queue_enable);
407 }
408
409 return 0;
410}
411
412static void del_vq(struct virtio_pci_vq_info *info)
413{
414 struct virtqueue *vq = info->vq;
415 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
416
417 vp_iowrite16(vq->index, &vp_dev->common->queue_select);
418
419 if (vp_dev->msix_enabled) {
420 vp_iowrite16(VIRTIO_MSI_NO_VECTOR,
421 &vp_dev->common->queue_msix_vector);
422 /* Flush the write out to device */
423 vp_ioread16(&vp_dev->common->queue_msix_vector);
424 }
425
426 if (!vp_dev->notify_base)
427 pci_iounmap(vp_dev->pci_dev, (void __force __iomem *)vq->priv);
428
429 vring_del_virtqueue(vq);
430}
431
432static const struct virtio_config_ops virtio_pci_config_nodev_ops = {
433 .get = NULL,
434 .set = NULL,
435 .generation = vp_generation,
436 .get_status = vp_get_status,
437 .set_status = vp_set_status,
438 .reset = vp_reset,
439 .find_vqs = vp_modern_find_vqs,
440 .del_vqs = vp_del_vqs,
441 .get_features = vp_get_features,
442 .finalize_features = vp_finalize_features,
443 .bus_name = vp_bus_name,
444 .set_vq_affinity = vp_set_vq_affinity,
445};
446
447static const struct virtio_config_ops virtio_pci_config_ops = {
448 .get = vp_get,
449 .set = vp_set,
450 .generation = vp_generation,
451 .get_status = vp_get_status,
452 .set_status = vp_set_status,
453 .reset = vp_reset,
454 .find_vqs = vp_modern_find_vqs,
455 .del_vqs = vp_del_vqs,
456 .get_features = vp_get_features,
457 .finalize_features = vp_finalize_features,
458 .bus_name = vp_bus_name,
459 .set_vq_affinity = vp_set_vq_affinity,
460};
461
462/**
463 * virtio_pci_find_capability - walk capabilities to find device info.
464 * @dev: the pci device
465 * @cfg_type: the VIRTIO_PCI_CAP_* value we seek
466 * @ioresource_types: IORESOURCE_MEM and/or IORESOURCE_IO.
467 *
468 * Returns offset of the capability, or 0.
469 */
470static inline int virtio_pci_find_capability(struct pci_dev *dev, u8 cfg_type,
471 u32 ioresource_types, int *bars)
472{
473 int pos;
474
475 for (pos = pci_find_capability(dev, PCI_CAP_ID_VNDR);
476 pos > 0;
477 pos = pci_find_next_capability(dev, pos, PCI_CAP_ID_VNDR)) {
478 u8 type, bar;
479 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
480 cfg_type),
481 &type);
482 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
483 bar),
484 &bar);
485
486 /* Ignore structures with reserved BAR values */
487 if (bar > 0x5)
488 continue;
489
490 if (type == cfg_type) {
491 if (pci_resource_len(dev, bar) &&
492 pci_resource_flags(dev, bar) & ioresource_types) {
493 *bars |= (1 << bar);
494 return pos;
495 }
496 }
497 }
498 return 0;
499}
500
501/* This is part of the ABI. Don't screw with it. */
502static inline void check_offsets(void)
503{
504 /* Note: disk space was harmed in compilation of this function. */
505 BUILD_BUG_ON(VIRTIO_PCI_CAP_VNDR !=
506 offsetof(struct virtio_pci_cap, cap_vndr));
507 BUILD_BUG_ON(VIRTIO_PCI_CAP_NEXT !=
508 offsetof(struct virtio_pci_cap, cap_next));
509 BUILD_BUG_ON(VIRTIO_PCI_CAP_LEN !=
510 offsetof(struct virtio_pci_cap, cap_len));
511 BUILD_BUG_ON(VIRTIO_PCI_CAP_CFG_TYPE !=
512 offsetof(struct virtio_pci_cap, cfg_type));
513 BUILD_BUG_ON(VIRTIO_PCI_CAP_BAR !=
514 offsetof(struct virtio_pci_cap, bar));
515 BUILD_BUG_ON(VIRTIO_PCI_CAP_OFFSET !=
516 offsetof(struct virtio_pci_cap, offset));
517 BUILD_BUG_ON(VIRTIO_PCI_CAP_LENGTH !=
518 offsetof(struct virtio_pci_cap, length));
519 BUILD_BUG_ON(VIRTIO_PCI_NOTIFY_CAP_MULT !=
520 offsetof(struct virtio_pci_notify_cap,
521 notify_off_multiplier));
522 BUILD_BUG_ON(VIRTIO_PCI_COMMON_DFSELECT !=
523 offsetof(struct virtio_pci_common_cfg,
524 device_feature_select));
525 BUILD_BUG_ON(VIRTIO_PCI_COMMON_DF !=
526 offsetof(struct virtio_pci_common_cfg, device_feature));
527 BUILD_BUG_ON(VIRTIO_PCI_COMMON_GFSELECT !=
528 offsetof(struct virtio_pci_common_cfg,
529 guest_feature_select));
530 BUILD_BUG_ON(VIRTIO_PCI_COMMON_GF !=
531 offsetof(struct virtio_pci_common_cfg, guest_feature));
532 BUILD_BUG_ON(VIRTIO_PCI_COMMON_MSIX !=
533 offsetof(struct virtio_pci_common_cfg, msix_config));
534 BUILD_BUG_ON(VIRTIO_PCI_COMMON_NUMQ !=
535 offsetof(struct virtio_pci_common_cfg, num_queues));
536 BUILD_BUG_ON(VIRTIO_PCI_COMMON_STATUS !=
537 offsetof(struct virtio_pci_common_cfg, device_status));
538 BUILD_BUG_ON(VIRTIO_PCI_COMMON_CFGGENERATION !=
539 offsetof(struct virtio_pci_common_cfg, config_generation));
540 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SELECT !=
541 offsetof(struct virtio_pci_common_cfg, queue_select));
542 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SIZE !=
543 offsetof(struct virtio_pci_common_cfg, queue_size));
544 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_MSIX !=
545 offsetof(struct virtio_pci_common_cfg, queue_msix_vector));
546 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_ENABLE !=
547 offsetof(struct virtio_pci_common_cfg, queue_enable));
548 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_NOFF !=
549 offsetof(struct virtio_pci_common_cfg, queue_notify_off));
550 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCLO !=
551 offsetof(struct virtio_pci_common_cfg, queue_desc_lo));
552 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCHI !=
553 offsetof(struct virtio_pci_common_cfg, queue_desc_hi));
554 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILLO !=
555 offsetof(struct virtio_pci_common_cfg, queue_avail_lo));
556 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILHI !=
557 offsetof(struct virtio_pci_common_cfg, queue_avail_hi));
558 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDLO !=
559 offsetof(struct virtio_pci_common_cfg, queue_used_lo));
560 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDHI !=
561 offsetof(struct virtio_pci_common_cfg, queue_used_hi));
562}
563
564/* the PCI probing function */
565int virtio_pci_modern_probe(struct virtio_pci_device *vp_dev)
566{
567 struct pci_dev *pci_dev = vp_dev->pci_dev;
568 int err, common, isr, notify, device;
569 u32 notify_length;
570 u32 notify_offset;
571
572 check_offsets();
573
574 /* We only own devices >= 0x1000 and <= 0x107f: leave the rest. */
575 if (pci_dev->device < 0x1000 || pci_dev->device > 0x107f)
576 return -ENODEV;
577
578 if (pci_dev->device < 0x1040) {
579 /* Transitional devices: use the PCI subsystem device id as
580 * virtio device id, same as legacy driver always did.
581 */
582 vp_dev->vdev.id.device = pci_dev->subsystem_device;
583 } else {
584 /* Modern devices: simply use PCI device id, but start from 0x1040. */
585 vp_dev->vdev.id.device = pci_dev->device - 0x1040;
586 }
587 vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
588
589 /* check for a common config: if not, use legacy mode (bar 0). */
590 common = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_COMMON_CFG,
591 IORESOURCE_IO | IORESOURCE_MEM,
592 &vp_dev->modern_bars);
593 if (!common) {
594 dev_info(&pci_dev->dev,
595 "virtio_pci: leaving for legacy driver\n");
596 return -ENODEV;
597 }
598
599 /* If common is there, these should be too... */
600 isr = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_ISR_CFG,
601 IORESOURCE_IO | IORESOURCE_MEM,
602 &vp_dev->modern_bars);
603 notify = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_NOTIFY_CFG,
604 IORESOURCE_IO | IORESOURCE_MEM,
605 &vp_dev->modern_bars);
606 if (!isr || !notify) {
607 dev_err(&pci_dev->dev,
608 "virtio_pci: missing capabilities %i/%i/%i\n",
609 common, isr, notify);
610 return -EINVAL;
611 }
612
613 err = dma_set_mask_and_coherent(&pci_dev->dev, DMA_BIT_MASK(64));
614 if (err)
615 err = dma_set_mask_and_coherent(&pci_dev->dev,
616 DMA_BIT_MASK(32));
617 if (err)
618 dev_warn(&pci_dev->dev, "Failed to enable 64-bit or 32-bit DMA. Trying to continue, but this might not work.\n");
619
620 /* Device capability is only mandatory for devices that have
621 * device-specific configuration.
622 */
623 device = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_DEVICE_CFG,
624 IORESOURCE_IO | IORESOURCE_MEM,
625 &vp_dev->modern_bars);
626
627 err = pci_request_selected_regions(pci_dev, vp_dev->modern_bars,
628 "virtio-pci-modern");
629 if (err)
630 return err;
631
632 err = -EINVAL;
633 vp_dev->common = map_capability(pci_dev, common,
634 sizeof(struct virtio_pci_common_cfg), 4,
635 0, sizeof(struct virtio_pci_common_cfg),
636 NULL);
637 if (!vp_dev->common)
638 goto err_map_common;
639 vp_dev->isr = map_capability(pci_dev, isr, sizeof(u8), 1,
640 0, 1,
641 NULL);
642 if (!vp_dev->isr)
643 goto err_map_isr;
644
645 /* Read notify_off_multiplier from config space. */
646 pci_read_config_dword(pci_dev,
647 notify + offsetof(struct virtio_pci_notify_cap,
648 notify_off_multiplier),
649 &vp_dev->notify_offset_multiplier);
650 /* Read notify length and offset from config space. */
651 pci_read_config_dword(pci_dev,
652 notify + offsetof(struct virtio_pci_notify_cap,
653 cap.length),
654 ¬ify_length);
655
656 pci_read_config_dword(pci_dev,
657 notify + offsetof(struct virtio_pci_notify_cap,
658 cap.offset),
659 ¬ify_offset);
660
661 /* We don't know how many VQs we'll map, ahead of the time.
662 * If notify length is small, map it all now.
663 * Otherwise, map each VQ individually later.
664 */
665 if ((u64)notify_length + (notify_offset % PAGE_SIZE) <= PAGE_SIZE) {
666 vp_dev->notify_base = map_capability(pci_dev, notify, 2, 2,
667 0, notify_length,
668 &vp_dev->notify_len);
669 if (!vp_dev->notify_base)
670 goto err_map_notify;
671 } else {
672 vp_dev->notify_map_cap = notify;
673 }
674
675 /* Again, we don't know how much we should map, but PAGE_SIZE
676 * is more than enough for all existing devices.
677 */
678 if (device) {
679 vp_dev->device = map_capability(pci_dev, device, 0, 4,
680 0, PAGE_SIZE,
681 &vp_dev->device_len);
682 if (!vp_dev->device)
683 goto err_map_device;
684
685 vp_dev->vdev.config = &virtio_pci_config_ops;
686 } else {
687 vp_dev->vdev.config = &virtio_pci_config_nodev_ops;
688 }
689
690 vp_dev->config_vector = vp_config_vector;
691 vp_dev->setup_vq = setup_vq;
692 vp_dev->del_vq = del_vq;
693
694 return 0;
695
696err_map_device:
697 if (vp_dev->notify_base)
698 pci_iounmap(pci_dev, vp_dev->notify_base);
699err_map_notify:
700 pci_iounmap(pci_dev, vp_dev->isr);
701err_map_isr:
702 pci_iounmap(pci_dev, vp_dev->common);
703err_map_common:
704 return err;
705}
706
707void virtio_pci_modern_remove(struct virtio_pci_device *vp_dev)
708{
709 struct pci_dev *pci_dev = vp_dev->pci_dev;
710
711 if (vp_dev->device)
712 pci_iounmap(pci_dev, vp_dev->device);
713 if (vp_dev->notify_base)
714 pci_iounmap(pci_dev, vp_dev->notify_base);
715 pci_iounmap(pci_dev, vp_dev->isr);
716 pci_iounmap(pci_dev, vp_dev->common);
717 pci_release_selected_regions(pci_dev, vp_dev->modern_bars);
718}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Virtio PCI driver - modern (virtio 1.0) device support
4 *
5 * This module allows virtio devices to be used over a virtual PCI device.
6 * This can be used with QEMU based VMMs like KVM or Xen.
7 *
8 * Copyright IBM Corp. 2007
9 * Copyright Red Hat, Inc. 2014
10 *
11 * Authors:
12 * Anthony Liguori <aliguori@us.ibm.com>
13 * Rusty Russell <rusty@rustcorp.com.au>
14 * Michael S. Tsirkin <mst@redhat.com>
15 */
16
17#include <linux/delay.h>
18#define VIRTIO_PCI_NO_LEGACY
19#define VIRTIO_RING_NO_LEGACY
20#include "virtio_pci_common.h"
21
22static u64 vp_get_features(struct virtio_device *vdev)
23{
24 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
25
26 return vp_modern_get_features(&vp_dev->mdev);
27}
28
29static void vp_transport_features(struct virtio_device *vdev, u64 features)
30{
31 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
32 struct pci_dev *pci_dev = vp_dev->pci_dev;
33
34 if ((features & BIT_ULL(VIRTIO_F_SR_IOV)) &&
35 pci_find_ext_capability(pci_dev, PCI_EXT_CAP_ID_SRIOV))
36 __virtio_set_bit(vdev, VIRTIO_F_SR_IOV);
37
38 if (features & BIT_ULL(VIRTIO_F_RING_RESET))
39 __virtio_set_bit(vdev, VIRTIO_F_RING_RESET);
40}
41
42/* virtio config->finalize_features() implementation */
43static int vp_finalize_features(struct virtio_device *vdev)
44{
45 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
46 u64 features = vdev->features;
47
48 /* Give virtio_ring a chance to accept features. */
49 vring_transport_features(vdev);
50
51 /* Give virtio_pci a chance to accept features. */
52 vp_transport_features(vdev, features);
53
54 if (!__virtio_test_bit(vdev, VIRTIO_F_VERSION_1)) {
55 dev_err(&vdev->dev, "virtio: device uses modern interface "
56 "but does not have VIRTIO_F_VERSION_1\n");
57 return -EINVAL;
58 }
59
60 vp_modern_set_features(&vp_dev->mdev, vdev->features);
61
62 return 0;
63}
64
65/* virtio config->get() implementation */
66static void vp_get(struct virtio_device *vdev, unsigned int offset,
67 void *buf, unsigned int len)
68{
69 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
70 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
71 void __iomem *device = mdev->device;
72 u8 b;
73 __le16 w;
74 __le32 l;
75
76 BUG_ON(offset + len > mdev->device_len);
77
78 switch (len) {
79 case 1:
80 b = ioread8(device + offset);
81 memcpy(buf, &b, sizeof b);
82 break;
83 case 2:
84 w = cpu_to_le16(ioread16(device + offset));
85 memcpy(buf, &w, sizeof w);
86 break;
87 case 4:
88 l = cpu_to_le32(ioread32(device + offset));
89 memcpy(buf, &l, sizeof l);
90 break;
91 case 8:
92 l = cpu_to_le32(ioread32(device + offset));
93 memcpy(buf, &l, sizeof l);
94 l = cpu_to_le32(ioread32(device + offset + sizeof l));
95 memcpy(buf + sizeof l, &l, sizeof l);
96 break;
97 default:
98 BUG();
99 }
100}
101
102/* the config->set() implementation. it's symmetric to the config->get()
103 * implementation */
104static void vp_set(struct virtio_device *vdev, unsigned int offset,
105 const void *buf, unsigned int len)
106{
107 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
108 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
109 void __iomem *device = mdev->device;
110 u8 b;
111 __le16 w;
112 __le32 l;
113
114 BUG_ON(offset + len > mdev->device_len);
115
116 switch (len) {
117 case 1:
118 memcpy(&b, buf, sizeof b);
119 iowrite8(b, device + offset);
120 break;
121 case 2:
122 memcpy(&w, buf, sizeof w);
123 iowrite16(le16_to_cpu(w), device + offset);
124 break;
125 case 4:
126 memcpy(&l, buf, sizeof l);
127 iowrite32(le32_to_cpu(l), device + offset);
128 break;
129 case 8:
130 memcpy(&l, buf, sizeof l);
131 iowrite32(le32_to_cpu(l), device + offset);
132 memcpy(&l, buf + sizeof l, sizeof l);
133 iowrite32(le32_to_cpu(l), device + offset + sizeof l);
134 break;
135 default:
136 BUG();
137 }
138}
139
140static u32 vp_generation(struct virtio_device *vdev)
141{
142 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
143
144 return vp_modern_generation(&vp_dev->mdev);
145}
146
147/* config->{get,set}_status() implementations */
148static u8 vp_get_status(struct virtio_device *vdev)
149{
150 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
151
152 return vp_modern_get_status(&vp_dev->mdev);
153}
154
155static void vp_set_status(struct virtio_device *vdev, u8 status)
156{
157 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
158
159 /* We should never be setting status to 0. */
160 BUG_ON(status == 0);
161 vp_modern_set_status(&vp_dev->mdev, status);
162}
163
164static void vp_reset(struct virtio_device *vdev)
165{
166 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
167 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
168
169 /* 0 status means a reset. */
170 vp_modern_set_status(mdev, 0);
171 /* After writing 0 to device_status, the driver MUST wait for a read of
172 * device_status to return 0 before reinitializing the device.
173 * This will flush out the status write, and flush in device writes,
174 * including MSI-X interrupts, if any.
175 */
176 while (vp_modern_get_status(mdev))
177 msleep(1);
178 /* Flush pending VQ/configuration callbacks. */
179 vp_synchronize_vectors(vdev);
180}
181
182static int vp_active_vq(struct virtqueue *vq, u16 msix_vec)
183{
184 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
185 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
186 unsigned long index;
187
188 index = vq->index;
189
190 /* activate the queue */
191 vp_modern_set_queue_size(mdev, index, virtqueue_get_vring_size(vq));
192 vp_modern_queue_address(mdev, index, virtqueue_get_desc_addr(vq),
193 virtqueue_get_avail_addr(vq),
194 virtqueue_get_used_addr(vq));
195
196 if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
197 msix_vec = vp_modern_queue_vector(mdev, index, msix_vec);
198 if (msix_vec == VIRTIO_MSI_NO_VECTOR)
199 return -EBUSY;
200 }
201
202 return 0;
203}
204
205static int vp_modern_disable_vq_and_reset(struct virtqueue *vq)
206{
207 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
208 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
209 struct virtio_pci_vq_info *info;
210 unsigned long flags;
211
212 if (!virtio_has_feature(vq->vdev, VIRTIO_F_RING_RESET))
213 return -ENOENT;
214
215 vp_modern_set_queue_reset(mdev, vq->index);
216
217 info = vp_dev->vqs[vq->index];
218
219 /* delete vq from irq handler */
220 spin_lock_irqsave(&vp_dev->lock, flags);
221 list_del(&info->node);
222 spin_unlock_irqrestore(&vp_dev->lock, flags);
223
224 INIT_LIST_HEAD(&info->node);
225
226#ifdef CONFIG_VIRTIO_HARDEN_NOTIFICATION
227 __virtqueue_break(vq);
228#endif
229
230 /* For the case where vq has an exclusive irq, call synchronize_irq() to
231 * wait for completion.
232 *
233 * note: We can't use disable_irq() since it conflicts with the affinity
234 * managed IRQ that is used by some drivers.
235 */
236 if (vp_dev->per_vq_vectors && info->msix_vector != VIRTIO_MSI_NO_VECTOR)
237 synchronize_irq(pci_irq_vector(vp_dev->pci_dev, info->msix_vector));
238
239 vq->reset = true;
240
241 return 0;
242}
243
244static int vp_modern_enable_vq_after_reset(struct virtqueue *vq)
245{
246 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
247 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
248 struct virtio_pci_vq_info *info;
249 unsigned long flags, index;
250 int err;
251
252 if (!vq->reset)
253 return -EBUSY;
254
255 index = vq->index;
256 info = vp_dev->vqs[index];
257
258 if (vp_modern_get_queue_reset(mdev, index))
259 return -EBUSY;
260
261 if (vp_modern_get_queue_enable(mdev, index))
262 return -EBUSY;
263
264 err = vp_active_vq(vq, info->msix_vector);
265 if (err)
266 return err;
267
268 if (vq->callback) {
269 spin_lock_irqsave(&vp_dev->lock, flags);
270 list_add(&info->node, &vp_dev->virtqueues);
271 spin_unlock_irqrestore(&vp_dev->lock, flags);
272 } else {
273 INIT_LIST_HEAD(&info->node);
274 }
275
276#ifdef CONFIG_VIRTIO_HARDEN_NOTIFICATION
277 __virtqueue_unbreak(vq);
278#endif
279
280 vp_modern_set_queue_enable(&vp_dev->mdev, index, true);
281 vq->reset = false;
282
283 return 0;
284}
285
286static u16 vp_config_vector(struct virtio_pci_device *vp_dev, u16 vector)
287{
288 return vp_modern_config_vector(&vp_dev->mdev, vector);
289}
290
291static struct virtqueue *setup_vq(struct virtio_pci_device *vp_dev,
292 struct virtio_pci_vq_info *info,
293 unsigned int index,
294 void (*callback)(struct virtqueue *vq),
295 const char *name,
296 bool ctx,
297 u16 msix_vec)
298{
299
300 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
301 struct virtqueue *vq;
302 u16 num;
303 int err;
304
305 if (index >= vp_modern_get_num_queues(mdev))
306 return ERR_PTR(-EINVAL);
307
308 /* Check if queue is either not available or already active. */
309 num = vp_modern_get_queue_size(mdev, index);
310 if (!num || vp_modern_get_queue_enable(mdev, index))
311 return ERR_PTR(-ENOENT);
312
313 if (!is_power_of_2(num)) {
314 dev_warn(&vp_dev->pci_dev->dev, "bad queue size %u", num);
315 return ERR_PTR(-EINVAL);
316 }
317
318 info->msix_vector = msix_vec;
319
320 /* create the vring */
321 vq = vring_create_virtqueue(index, num,
322 SMP_CACHE_BYTES, &vp_dev->vdev,
323 true, true, ctx,
324 vp_notify, callback, name);
325 if (!vq)
326 return ERR_PTR(-ENOMEM);
327
328 vq->num_max = num;
329
330 err = vp_active_vq(vq, msix_vec);
331 if (err)
332 goto err;
333
334 vq->priv = (void __force *)vp_modern_map_vq_notify(mdev, index, NULL);
335 if (!vq->priv) {
336 err = -ENOMEM;
337 goto err;
338 }
339
340 return vq;
341
342err:
343 vring_del_virtqueue(vq);
344 return ERR_PTR(err);
345}
346
347static int vp_modern_find_vqs(struct virtio_device *vdev, unsigned int nvqs,
348 struct virtqueue *vqs[],
349 vq_callback_t *callbacks[],
350 const char * const names[], const bool *ctx,
351 struct irq_affinity *desc)
352{
353 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
354 struct virtqueue *vq;
355 int rc = vp_find_vqs(vdev, nvqs, vqs, callbacks, names, ctx, desc);
356
357 if (rc)
358 return rc;
359
360 /* Select and activate all queues. Has to be done last: once we do
361 * this, there's no way to go back except reset.
362 */
363 list_for_each_entry(vq, &vdev->vqs, list)
364 vp_modern_set_queue_enable(&vp_dev->mdev, vq->index, true);
365
366 return 0;
367}
368
369static void del_vq(struct virtio_pci_vq_info *info)
370{
371 struct virtqueue *vq = info->vq;
372 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
373 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
374
375 if (vp_dev->msix_enabled)
376 vp_modern_queue_vector(mdev, vq->index,
377 VIRTIO_MSI_NO_VECTOR);
378
379 if (!mdev->notify_base)
380 pci_iounmap(mdev->pci_dev, (void __force __iomem *)vq->priv);
381
382 vring_del_virtqueue(vq);
383}
384
385static int virtio_pci_find_shm_cap(struct pci_dev *dev, u8 required_id,
386 u8 *bar, u64 *offset, u64 *len)
387{
388 int pos;
389
390 for (pos = pci_find_capability(dev, PCI_CAP_ID_VNDR); pos > 0;
391 pos = pci_find_next_capability(dev, pos, PCI_CAP_ID_VNDR)) {
392 u8 type, cap_len, id, res_bar;
393 u32 tmp32;
394 u64 res_offset, res_length;
395
396 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
397 cfg_type), &type);
398 if (type != VIRTIO_PCI_CAP_SHARED_MEMORY_CFG)
399 continue;
400
401 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
402 cap_len), &cap_len);
403 if (cap_len != sizeof(struct virtio_pci_cap64)) {
404 dev_err(&dev->dev, "%s: shm cap with bad size offset:"
405 " %d size: %d\n", __func__, pos, cap_len);
406 continue;
407 }
408
409 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
410 id), &id);
411 if (id != required_id)
412 continue;
413
414 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
415 bar), &res_bar);
416 if (res_bar >= PCI_STD_NUM_BARS)
417 continue;
418
419 /* Type and ID match, and the BAR value isn't reserved.
420 * Looks good.
421 */
422
423 /* Read the lower 32bit of length and offset */
424 pci_read_config_dword(dev, pos + offsetof(struct virtio_pci_cap,
425 offset), &tmp32);
426 res_offset = tmp32;
427 pci_read_config_dword(dev, pos + offsetof(struct virtio_pci_cap,
428 length), &tmp32);
429 res_length = tmp32;
430
431 /* and now the top half */
432 pci_read_config_dword(dev,
433 pos + offsetof(struct virtio_pci_cap64,
434 offset_hi), &tmp32);
435 res_offset |= ((u64)tmp32) << 32;
436 pci_read_config_dword(dev,
437 pos + offsetof(struct virtio_pci_cap64,
438 length_hi), &tmp32);
439 res_length |= ((u64)tmp32) << 32;
440
441 *bar = res_bar;
442 *offset = res_offset;
443 *len = res_length;
444
445 return pos;
446 }
447 return 0;
448}
449
450static bool vp_get_shm_region(struct virtio_device *vdev,
451 struct virtio_shm_region *region, u8 id)
452{
453 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
454 struct pci_dev *pci_dev = vp_dev->pci_dev;
455 u8 bar;
456 u64 offset, len;
457 phys_addr_t phys_addr;
458 size_t bar_len;
459
460 if (!virtio_pci_find_shm_cap(pci_dev, id, &bar, &offset, &len))
461 return false;
462
463 phys_addr = pci_resource_start(pci_dev, bar);
464 bar_len = pci_resource_len(pci_dev, bar);
465
466 if ((offset + len) < offset) {
467 dev_err(&pci_dev->dev, "%s: cap offset+len overflow detected\n",
468 __func__);
469 return false;
470 }
471
472 if (offset + len > bar_len) {
473 dev_err(&pci_dev->dev, "%s: bar shorter than cap offset+len\n",
474 __func__);
475 return false;
476 }
477
478 region->len = len;
479 region->addr = (u64) phys_addr + offset;
480
481 return true;
482}
483
484static const struct virtio_config_ops virtio_pci_config_nodev_ops = {
485 .get = NULL,
486 .set = NULL,
487 .generation = vp_generation,
488 .get_status = vp_get_status,
489 .set_status = vp_set_status,
490 .reset = vp_reset,
491 .find_vqs = vp_modern_find_vqs,
492 .del_vqs = vp_del_vqs,
493 .synchronize_cbs = vp_synchronize_vectors,
494 .get_features = vp_get_features,
495 .finalize_features = vp_finalize_features,
496 .bus_name = vp_bus_name,
497 .set_vq_affinity = vp_set_vq_affinity,
498 .get_vq_affinity = vp_get_vq_affinity,
499 .get_shm_region = vp_get_shm_region,
500 .disable_vq_and_reset = vp_modern_disable_vq_and_reset,
501 .enable_vq_after_reset = vp_modern_enable_vq_after_reset,
502};
503
504static const struct virtio_config_ops virtio_pci_config_ops = {
505 .get = vp_get,
506 .set = vp_set,
507 .generation = vp_generation,
508 .get_status = vp_get_status,
509 .set_status = vp_set_status,
510 .reset = vp_reset,
511 .find_vqs = vp_modern_find_vqs,
512 .del_vqs = vp_del_vqs,
513 .synchronize_cbs = vp_synchronize_vectors,
514 .get_features = vp_get_features,
515 .finalize_features = vp_finalize_features,
516 .bus_name = vp_bus_name,
517 .set_vq_affinity = vp_set_vq_affinity,
518 .get_vq_affinity = vp_get_vq_affinity,
519 .get_shm_region = vp_get_shm_region,
520 .disable_vq_and_reset = vp_modern_disable_vq_and_reset,
521 .enable_vq_after_reset = vp_modern_enable_vq_after_reset,
522};
523
524/* the PCI probing function */
525int virtio_pci_modern_probe(struct virtio_pci_device *vp_dev)
526{
527 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
528 struct pci_dev *pci_dev = vp_dev->pci_dev;
529 int err;
530
531 mdev->pci_dev = pci_dev;
532
533 err = vp_modern_probe(mdev);
534 if (err)
535 return err;
536
537 if (mdev->device)
538 vp_dev->vdev.config = &virtio_pci_config_ops;
539 else
540 vp_dev->vdev.config = &virtio_pci_config_nodev_ops;
541
542 vp_dev->config_vector = vp_config_vector;
543 vp_dev->setup_vq = setup_vq;
544 vp_dev->del_vq = del_vq;
545 vp_dev->isr = mdev->isr;
546 vp_dev->vdev.id = mdev->id;
547
548 return 0;
549}
550
551void virtio_pci_modern_remove(struct virtio_pci_device *vp_dev)
552{
553 struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
554
555 vp_modern_remove(mdev);
556}