xref: /linux/drivers/virtio/virtio_pci_modern.c (revision cc622420798c4bcf093785d872525087a7798db9)
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 #define VIRTIO_PCI_NO_LEGACY
21 #include "virtio_pci_common.h"
22 
23 /*
24  * Type-safe wrappers for io accesses.
25  * Use these to enforce at compile time the following spec requirement:
26  *
27  * The driver MUST access each field using the “natural” access
28  * method, i.e. 32-bit accesses for 32-bit fields, 16-bit accesses
29  * for 16-bit fields and 8-bit accesses for 8-bit fields.
30  */
31 static inline u8 vp_ioread8(u8 __iomem *addr)
32 {
33 	return ioread8(addr);
34 }
35 static inline u16 vp_ioread16 (u16 __iomem *addr)
36 {
37 	return ioread16(addr);
38 }
39 
40 static inline u32 vp_ioread32(u32 __iomem *addr)
41 {
42 	return ioread32(addr);
43 }
44 
45 static inline void vp_iowrite8(u8 value, u8 __iomem *addr)
46 {
47 	iowrite8(value, addr);
48 }
49 
50 static inline void vp_iowrite16(u16 value, u16 __iomem *addr)
51 {
52 	iowrite16(value, addr);
53 }
54 
55 static inline void vp_iowrite32(u32 value, u32 __iomem *addr)
56 {
57 	iowrite32(value, addr);
58 }
59 
60 static void vp_iowrite64_twopart(u64 val,
61 				 __le32 __iomem *lo, __le32 __iomem *hi)
62 {
63 	vp_iowrite32((u32)val, lo);
64 	vp_iowrite32(val >> 32, hi);
65 }
66 
67 static void __iomem *map_capability(struct pci_dev *dev, int off,
68 				    size_t minlen,
69 				    u32 align,
70 				    u32 start, u32 size,
71 				    size_t *len)
72 {
73 	u8 bar;
74 	u32 offset, length;
75 	void __iomem *p;
76 
77 	pci_read_config_byte(dev, off + offsetof(struct virtio_pci_cap,
78 						 bar),
79 			     &bar);
80 	pci_read_config_dword(dev, off + offsetof(struct virtio_pci_cap, offset),
81 			     &offset);
82 	pci_read_config_dword(dev, off + offsetof(struct virtio_pci_cap, length),
83 			      &length);
84 
85 	if (length <= start) {
86 		dev_err(&dev->dev,
87 			"virtio_pci: bad capability len %u (>%u expected)\n",
88 			length, start);
89 		return NULL;
90 	}
91 
92 	if (length - start < minlen) {
93 		dev_err(&dev->dev,
94 			"virtio_pci: bad capability len %u (>=%zu expected)\n",
95 			length, minlen);
96 		return NULL;
97 	}
98 
99 	length -= start;
100 
101 	if (start + offset < offset) {
102 		dev_err(&dev->dev,
103 			"virtio_pci: map wrap-around %u+%u\n",
104 			start, offset);
105 		return NULL;
106 	}
107 
108 	offset += start;
109 
110 	if (offset & (align - 1)) {
111 		dev_err(&dev->dev,
112 			"virtio_pci: offset %u not aligned to %u\n",
113 			offset, align);
114 		return NULL;
115 	}
116 
117 	if (length > size)
118 		length = size;
119 
120 	if (len)
121 		*len = length;
122 
123 	if (minlen + offset < minlen ||
124 	    minlen + offset > pci_resource_len(dev, bar)) {
125 		dev_err(&dev->dev,
126 			"virtio_pci: map virtio %zu@%u "
127 			"out of range on bar %i length %lu\n",
128 			minlen, offset,
129 			bar, (unsigned long)pci_resource_len(dev, bar));
130 		return NULL;
131 	}
132 
133 	p = pci_iomap_range(dev, bar, offset, length);
134 	if (!p)
135 		dev_err(&dev->dev,
136 			"virtio_pci: unable to map virtio %u@%u on bar %i\n",
137 			length, offset, bar);
138 	return p;
139 }
140 
141 /* virtio config->get_features() implementation */
142 static u64 vp_get_features(struct virtio_device *vdev)
143 {
144 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
145 	u64 features;
146 
147 	vp_iowrite32(0, &vp_dev->common->device_feature_select);
148 	features = vp_ioread32(&vp_dev->common->device_feature);
149 	vp_iowrite32(1, &vp_dev->common->device_feature_select);
150 	features |= ((u64)vp_ioread32(&vp_dev->common->device_feature) << 32);
151 
152 	return features;
153 }
154 
155 /* virtio config->finalize_features() implementation */
156 static int vp_finalize_features(struct virtio_device *vdev)
157 {
158 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
159 
160 	/* Give virtio_ring a chance to accept features. */
161 	vring_transport_features(vdev);
162 
163 	if (!__virtio_test_bit(vdev, VIRTIO_F_VERSION_1)) {
164 		dev_err(&vdev->dev, "virtio: device uses modern interface "
165 			"but does not have VIRTIO_F_VERSION_1\n");
166 		return -EINVAL;
167 	}
168 
169 	vp_iowrite32(0, &vp_dev->common->guest_feature_select);
170 	vp_iowrite32((u32)vdev->features, &vp_dev->common->guest_feature);
171 	vp_iowrite32(1, &vp_dev->common->guest_feature_select);
172 	vp_iowrite32(vdev->features >> 32, &vp_dev->common->guest_feature);
173 
174 	return 0;
175 }
176 
177 /* virtio config->get() implementation */
178 static void vp_get(struct virtio_device *vdev, unsigned offset,
179 		   void *buf, unsigned len)
180 {
181 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
182 	u8 b;
183 	__le16 w;
184 	__le32 l;
185 
186 	BUG_ON(offset + len > vp_dev->device_len);
187 
188 	switch (len) {
189 	case 1:
190 		b = ioread8(vp_dev->device + offset);
191 		memcpy(buf, &b, sizeof b);
192 		break;
193 	case 2:
194 		w = cpu_to_le16(ioread16(vp_dev->device + offset));
195 		memcpy(buf, &w, sizeof w);
196 		break;
197 	case 4:
198 		l = cpu_to_le32(ioread32(vp_dev->device + offset));
199 		memcpy(buf, &l, sizeof l);
200 		break;
201 	case 8:
202 		l = cpu_to_le32(ioread32(vp_dev->device + offset));
203 		memcpy(buf, &l, sizeof l);
204 		l = cpu_to_le32(ioread32(vp_dev->device + offset + sizeof l));
205 		memcpy(buf + sizeof l, &l, sizeof l);
206 		break;
207 	default:
208 		BUG();
209 	}
210 }
211 
212 /* the config->set() implementation.  it's symmetric to the config->get()
213  * implementation */
214 static void vp_set(struct virtio_device *vdev, unsigned offset,
215 		   const void *buf, unsigned len)
216 {
217 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
218 	u8 b;
219 	__le16 w;
220 	__le32 l;
221 
222 	BUG_ON(offset + len > vp_dev->device_len);
223 
224 	switch (len) {
225 	case 1:
226 		memcpy(&b, buf, sizeof b);
227 		iowrite8(b, vp_dev->device + offset);
228 		break;
229 	case 2:
230 		memcpy(&w, buf, sizeof w);
231 		iowrite16(le16_to_cpu(w), vp_dev->device + offset);
232 		break;
233 	case 4:
234 		memcpy(&l, buf, sizeof l);
235 		iowrite32(le32_to_cpu(l), vp_dev->device + offset);
236 		break;
237 	case 8:
238 		memcpy(&l, buf, sizeof l);
239 		iowrite32(le32_to_cpu(l), vp_dev->device + offset);
240 		memcpy(&l, buf + sizeof l, sizeof l);
241 		iowrite32(le32_to_cpu(l), vp_dev->device + offset + sizeof l);
242 		break;
243 	default:
244 		BUG();
245 	}
246 }
247 
248 static u32 vp_generation(struct virtio_device *vdev)
249 {
250 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
251 	return vp_ioread8(&vp_dev->common->config_generation);
252 }
253 
254 /* config->{get,set}_status() implementations */
255 static u8 vp_get_status(struct virtio_device *vdev)
256 {
257 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
258 	return vp_ioread8(&vp_dev->common->device_status);
259 }
260 
261 static void vp_set_status(struct virtio_device *vdev, u8 status)
262 {
263 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
264 	/* We should never be setting status to 0. */
265 	BUG_ON(status == 0);
266 	vp_iowrite8(status, &vp_dev->common->device_status);
267 }
268 
269 static void vp_reset(struct virtio_device *vdev)
270 {
271 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
272 	/* 0 status means a reset. */
273 	vp_iowrite8(0, &vp_dev->common->device_status);
274 	/* Flush out the status write, and flush in device writes,
275 	 * including MSI-X interrupts, if any. */
276 	vp_ioread8(&vp_dev->common->device_status);
277 	/* Flush pending VQ/configuration callbacks. */
278 	vp_synchronize_vectors(vdev);
279 }
280 
281 static u16 vp_config_vector(struct virtio_pci_device *vp_dev, u16 vector)
282 {
283 	/* Setup the vector used for configuration events */
284 	vp_iowrite16(vector, &vp_dev->common->msix_config);
285 	/* Verify we had enough resources to assign the vector */
286 	/* Will also flush the write out to device */
287 	return vp_ioread16(&vp_dev->common->msix_config);
288 }
289 
290 static struct virtqueue *setup_vq(struct virtio_pci_device *vp_dev,
291 				  struct virtio_pci_vq_info *info,
292 				  unsigned index,
293 				  void (*callback)(struct virtqueue *vq),
294 				  const char *name,
295 				  u16 msix_vec)
296 {
297 	struct virtio_pci_common_cfg __iomem *cfg = vp_dev->common;
298 	struct virtqueue *vq;
299 	u16 num, off;
300 	int err;
301 
302 	if (index >= vp_ioread16(&cfg->num_queues))
303 		return ERR_PTR(-ENOENT);
304 
305 	/* Select the queue we're interested in */
306 	vp_iowrite16(index, &cfg->queue_select);
307 
308 	/* Check if queue is either not available or already active. */
309 	num = vp_ioread16(&cfg->queue_size);
310 	if (!num || vp_ioread16(&cfg->queue_enable))
311 		return ERR_PTR(-ENOENT);
312 
313 	if (num & (num - 1)) {
314 		dev_warn(&vp_dev->pci_dev->dev, "bad queue size %u", num);
315 		return ERR_PTR(-EINVAL);
316 	}
317 
318 	/* get offset of notification word for this vq */
319 	off = vp_ioread16(&cfg->queue_notify_off);
320 
321 	info->msix_vector = msix_vec;
322 
323 	/* create the vring */
324 	vq = vring_create_virtqueue(index, num,
325 				    SMP_CACHE_BYTES, &vp_dev->vdev,
326 				    true, true, vp_notify, callback, name);
327 	if (!vq)
328 		return ERR_PTR(-ENOMEM);
329 
330 	/* activate the queue */
331 	vp_iowrite16(virtqueue_get_vring_size(vq), &cfg->queue_size);
332 	vp_iowrite64_twopart(virtqueue_get_desc_addr(vq),
333 			     &cfg->queue_desc_lo, &cfg->queue_desc_hi);
334 	vp_iowrite64_twopart(virtqueue_get_avail_addr(vq),
335 			     &cfg->queue_avail_lo, &cfg->queue_avail_hi);
336 	vp_iowrite64_twopart(virtqueue_get_used_addr(vq),
337 			     &cfg->queue_used_lo, &cfg->queue_used_hi);
338 
339 	if (vp_dev->notify_base) {
340 		/* offset should not wrap */
341 		if ((u64)off * vp_dev->notify_offset_multiplier + 2
342 		    > vp_dev->notify_len) {
343 			dev_warn(&vp_dev->pci_dev->dev,
344 				 "bad notification offset %u (x %u) "
345 				 "for queue %u > %zd",
346 				 off, vp_dev->notify_offset_multiplier,
347 				 index, vp_dev->notify_len);
348 			err = -EINVAL;
349 			goto err_map_notify;
350 		}
351 		vq->priv = (void __force *)vp_dev->notify_base +
352 			off * vp_dev->notify_offset_multiplier;
353 	} else {
354 		vq->priv = (void __force *)map_capability(vp_dev->pci_dev,
355 					  vp_dev->notify_map_cap, 2, 2,
356 					  off * vp_dev->notify_offset_multiplier, 2,
357 					  NULL);
358 	}
359 
360 	if (!vq->priv) {
361 		err = -ENOMEM;
362 		goto err_map_notify;
363 	}
364 
365 	if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
366 		vp_iowrite16(msix_vec, &cfg->queue_msix_vector);
367 		msix_vec = vp_ioread16(&cfg->queue_msix_vector);
368 		if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
369 			err = -EBUSY;
370 			goto err_assign_vector;
371 		}
372 	}
373 
374 	return vq;
375 
376 err_assign_vector:
377 	if (!vp_dev->notify_base)
378 		pci_iounmap(vp_dev->pci_dev, (void __iomem __force *)vq->priv);
379 err_map_notify:
380 	vring_del_virtqueue(vq);
381 	return ERR_PTR(err);
382 }
383 
384 static int vp_modern_find_vqs(struct virtio_device *vdev, unsigned nvqs,
385 			      struct virtqueue *vqs[],
386 			      vq_callback_t *callbacks[],
387 			      const char * const names[])
388 {
389 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
390 	struct virtqueue *vq;
391 	int rc = vp_find_vqs(vdev, nvqs, vqs, callbacks, names);
392 
393 	if (rc)
394 		return rc;
395 
396 	/* Select and activate all queues. Has to be done last: once we do
397 	 * this, there's no way to go back except reset.
398 	 */
399 	list_for_each_entry(vq, &vdev->vqs, list) {
400 		vp_iowrite16(vq->index, &vp_dev->common->queue_select);
401 		vp_iowrite16(1, &vp_dev->common->queue_enable);
402 	}
403 
404 	return 0;
405 }
406 
407 static void del_vq(struct virtio_pci_vq_info *info)
408 {
409 	struct virtqueue *vq = info->vq;
410 	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
411 
412 	vp_iowrite16(vq->index, &vp_dev->common->queue_select);
413 
414 	if (vp_dev->msix_enabled) {
415 		vp_iowrite16(VIRTIO_MSI_NO_VECTOR,
416 			     &vp_dev->common->queue_msix_vector);
417 		/* Flush the write out to device */
418 		vp_ioread16(&vp_dev->common->queue_msix_vector);
419 	}
420 
421 	if (!vp_dev->notify_base)
422 		pci_iounmap(vp_dev->pci_dev, (void __force __iomem *)vq->priv);
423 
424 	vring_del_virtqueue(vq);
425 }
426 
427 static const struct virtio_config_ops virtio_pci_config_nodev_ops = {
428 	.get		= NULL,
429 	.set		= NULL,
430 	.generation	= vp_generation,
431 	.get_status	= vp_get_status,
432 	.set_status	= vp_set_status,
433 	.reset		= vp_reset,
434 	.find_vqs	= vp_modern_find_vqs,
435 	.del_vqs	= vp_del_vqs,
436 	.get_features	= vp_get_features,
437 	.finalize_features = vp_finalize_features,
438 	.bus_name	= vp_bus_name,
439 	.set_vq_affinity = vp_set_vq_affinity,
440 };
441 
442 static const struct virtio_config_ops virtio_pci_config_ops = {
443 	.get		= vp_get,
444 	.set		= vp_set,
445 	.generation	= vp_generation,
446 	.get_status	= vp_get_status,
447 	.set_status	= vp_set_status,
448 	.reset		= vp_reset,
449 	.find_vqs	= vp_modern_find_vqs,
450 	.del_vqs	= vp_del_vqs,
451 	.get_features	= vp_get_features,
452 	.finalize_features = vp_finalize_features,
453 	.bus_name	= vp_bus_name,
454 	.set_vq_affinity = vp_set_vq_affinity,
455 };
456 
457 /**
458  * virtio_pci_find_capability - walk capabilities to find device info.
459  * @dev: the pci device
460  * @cfg_type: the VIRTIO_PCI_CAP_* value we seek
461  * @ioresource_types: IORESOURCE_MEM and/or IORESOURCE_IO.
462  *
463  * Returns offset of the capability, or 0.
464  */
465 static inline int virtio_pci_find_capability(struct pci_dev *dev, u8 cfg_type,
466 					     u32 ioresource_types, int *bars)
467 {
468 	int pos;
469 
470 	for (pos = pci_find_capability(dev, PCI_CAP_ID_VNDR);
471 	     pos > 0;
472 	     pos = pci_find_next_capability(dev, pos, PCI_CAP_ID_VNDR)) {
473 		u8 type, bar;
474 		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
475 							 cfg_type),
476 				     &type);
477 		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
478 							 bar),
479 				     &bar);
480 
481 		/* Ignore structures with reserved BAR values */
482 		if (bar > 0x5)
483 			continue;
484 
485 		if (type == cfg_type) {
486 			if (pci_resource_len(dev, bar) &&
487 			    pci_resource_flags(dev, bar) & ioresource_types) {
488 				*bars |= (1 << bar);
489 				return pos;
490 			}
491 		}
492 	}
493 	return 0;
494 }
495 
496 /* This is part of the ABI.  Don't screw with it. */
497 static inline void check_offsets(void)
498 {
499 	/* Note: disk space was harmed in compilation of this function. */
500 	BUILD_BUG_ON(VIRTIO_PCI_CAP_VNDR !=
501 		     offsetof(struct virtio_pci_cap, cap_vndr));
502 	BUILD_BUG_ON(VIRTIO_PCI_CAP_NEXT !=
503 		     offsetof(struct virtio_pci_cap, cap_next));
504 	BUILD_BUG_ON(VIRTIO_PCI_CAP_LEN !=
505 		     offsetof(struct virtio_pci_cap, cap_len));
506 	BUILD_BUG_ON(VIRTIO_PCI_CAP_CFG_TYPE !=
507 		     offsetof(struct virtio_pci_cap, cfg_type));
508 	BUILD_BUG_ON(VIRTIO_PCI_CAP_BAR !=
509 		     offsetof(struct virtio_pci_cap, bar));
510 	BUILD_BUG_ON(VIRTIO_PCI_CAP_OFFSET !=
511 		     offsetof(struct virtio_pci_cap, offset));
512 	BUILD_BUG_ON(VIRTIO_PCI_CAP_LENGTH !=
513 		     offsetof(struct virtio_pci_cap, length));
514 	BUILD_BUG_ON(VIRTIO_PCI_NOTIFY_CAP_MULT !=
515 		     offsetof(struct virtio_pci_notify_cap,
516 			      notify_off_multiplier));
517 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_DFSELECT !=
518 		     offsetof(struct virtio_pci_common_cfg,
519 			      device_feature_select));
520 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_DF !=
521 		     offsetof(struct virtio_pci_common_cfg, device_feature));
522 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_GFSELECT !=
523 		     offsetof(struct virtio_pci_common_cfg,
524 			      guest_feature_select));
525 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_GF !=
526 		     offsetof(struct virtio_pci_common_cfg, guest_feature));
527 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_MSIX !=
528 		     offsetof(struct virtio_pci_common_cfg, msix_config));
529 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_NUMQ !=
530 		     offsetof(struct virtio_pci_common_cfg, num_queues));
531 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_STATUS !=
532 		     offsetof(struct virtio_pci_common_cfg, device_status));
533 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_CFGGENERATION !=
534 		     offsetof(struct virtio_pci_common_cfg, config_generation));
535 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SELECT !=
536 		     offsetof(struct virtio_pci_common_cfg, queue_select));
537 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SIZE !=
538 		     offsetof(struct virtio_pci_common_cfg, queue_size));
539 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_MSIX !=
540 		     offsetof(struct virtio_pci_common_cfg, queue_msix_vector));
541 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_ENABLE !=
542 		     offsetof(struct virtio_pci_common_cfg, queue_enable));
543 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_NOFF !=
544 		     offsetof(struct virtio_pci_common_cfg, queue_notify_off));
545 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCLO !=
546 		     offsetof(struct virtio_pci_common_cfg, queue_desc_lo));
547 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCHI !=
548 		     offsetof(struct virtio_pci_common_cfg, queue_desc_hi));
549 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILLO !=
550 		     offsetof(struct virtio_pci_common_cfg, queue_avail_lo));
551 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILHI !=
552 		     offsetof(struct virtio_pci_common_cfg, queue_avail_hi));
553 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDLO !=
554 		     offsetof(struct virtio_pci_common_cfg, queue_used_lo));
555 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDHI !=
556 		     offsetof(struct virtio_pci_common_cfg, queue_used_hi));
557 }
558 
559 /* the PCI probing function */
560 int virtio_pci_modern_probe(struct virtio_pci_device *vp_dev)
561 {
562 	struct pci_dev *pci_dev = vp_dev->pci_dev;
563 	int err, common, isr, notify, device;
564 	u32 notify_length;
565 	u32 notify_offset;
566 
567 	check_offsets();
568 
569 	/* We only own devices >= 0x1000 and <= 0x107f: leave the rest. */
570 	if (pci_dev->device < 0x1000 || pci_dev->device > 0x107f)
571 		return -ENODEV;
572 
573 	if (pci_dev->device < 0x1040) {
574 		/* Transitional devices: use the PCI subsystem device id as
575 		 * virtio device id, same as legacy driver always did.
576 		 */
577 		vp_dev->vdev.id.device = pci_dev->subsystem_device;
578 	} else {
579 		/* Modern devices: simply use PCI device id, but start from 0x1040. */
580 		vp_dev->vdev.id.device = pci_dev->device - 0x1040;
581 	}
582 	vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
583 
584 	/* check for a common config: if not, use legacy mode (bar 0). */
585 	common = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_COMMON_CFG,
586 					    IORESOURCE_IO | IORESOURCE_MEM,
587 					    &vp_dev->modern_bars);
588 	if (!common) {
589 		dev_info(&pci_dev->dev,
590 			 "virtio_pci: leaving for legacy driver\n");
591 		return -ENODEV;
592 	}
593 
594 	/* If common is there, these should be too... */
595 	isr = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_ISR_CFG,
596 					 IORESOURCE_IO | IORESOURCE_MEM,
597 					 &vp_dev->modern_bars);
598 	notify = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_NOTIFY_CFG,
599 					    IORESOURCE_IO | IORESOURCE_MEM,
600 					    &vp_dev->modern_bars);
601 	if (!isr || !notify) {
602 		dev_err(&pci_dev->dev,
603 			"virtio_pci: missing capabilities %i/%i/%i\n",
604 			common, isr, notify);
605 		return -EINVAL;
606 	}
607 
608 	err = dma_set_mask_and_coherent(&pci_dev->dev, DMA_BIT_MASK(64));
609 	if (err)
610 		err = dma_set_mask_and_coherent(&pci_dev->dev,
611 						DMA_BIT_MASK(32));
612 	if (err)
613 		dev_warn(&pci_dev->dev, "Failed to enable 64-bit or 32-bit DMA.  Trying to continue, but this might not work.\n");
614 
615 	/* Device capability is only mandatory for devices that have
616 	 * device-specific configuration.
617 	 */
618 	device = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_DEVICE_CFG,
619 					    IORESOURCE_IO | IORESOURCE_MEM,
620 					    &vp_dev->modern_bars);
621 
622 	err = pci_request_selected_regions(pci_dev, vp_dev->modern_bars,
623 					   "virtio-pci-modern");
624 	if (err)
625 		return err;
626 
627 	err = -EINVAL;
628 	vp_dev->common = map_capability(pci_dev, common,
629 					sizeof(struct virtio_pci_common_cfg), 4,
630 					0, sizeof(struct virtio_pci_common_cfg),
631 					NULL);
632 	if (!vp_dev->common)
633 		goto err_map_common;
634 	vp_dev->isr = map_capability(pci_dev, isr, sizeof(u8), 1,
635 				     0, 1,
636 				     NULL);
637 	if (!vp_dev->isr)
638 		goto err_map_isr;
639 
640 	/* Read notify_off_multiplier from config space. */
641 	pci_read_config_dword(pci_dev,
642 			      notify + offsetof(struct virtio_pci_notify_cap,
643 						notify_off_multiplier),
644 			      &vp_dev->notify_offset_multiplier);
645 	/* Read notify length and offset from config space. */
646 	pci_read_config_dword(pci_dev,
647 			      notify + offsetof(struct virtio_pci_notify_cap,
648 						cap.length),
649 			      &notify_length);
650 
651 	pci_read_config_dword(pci_dev,
652 			      notify + offsetof(struct virtio_pci_notify_cap,
653 						cap.offset),
654 			      &notify_offset);
655 
656 	/* We don't know how many VQs we'll map, ahead of the time.
657 	 * If notify length is small, map it all now.
658 	 * Otherwise, map each VQ individually later.
659 	 */
660 	if ((u64)notify_length + (notify_offset % PAGE_SIZE) <= PAGE_SIZE) {
661 		vp_dev->notify_base = map_capability(pci_dev, notify, 2, 2,
662 						     0, notify_length,
663 						     &vp_dev->notify_len);
664 		if (!vp_dev->notify_base)
665 			goto err_map_notify;
666 	} else {
667 		vp_dev->notify_map_cap = notify;
668 	}
669 
670 	/* Again, we don't know how much we should map, but PAGE_SIZE
671 	 * is more than enough for all existing devices.
672 	 */
673 	if (device) {
674 		vp_dev->device = map_capability(pci_dev, device, 0, 4,
675 						0, PAGE_SIZE,
676 						&vp_dev->device_len);
677 		if (!vp_dev->device)
678 			goto err_map_device;
679 
680 		vp_dev->vdev.config = &virtio_pci_config_ops;
681 	} else {
682 		vp_dev->vdev.config = &virtio_pci_config_nodev_ops;
683 	}
684 
685 	vp_dev->config_vector = vp_config_vector;
686 	vp_dev->setup_vq = setup_vq;
687 	vp_dev->del_vq = del_vq;
688 
689 	return 0;
690 
691 err_map_device:
692 	if (vp_dev->notify_base)
693 		pci_iounmap(pci_dev, vp_dev->notify_base);
694 err_map_notify:
695 	pci_iounmap(pci_dev, vp_dev->isr);
696 err_map_isr:
697 	pci_iounmap(pci_dev, vp_dev->common);
698 err_map_common:
699 	return err;
700 }
701 
702 void virtio_pci_modern_remove(struct virtio_pci_device *vp_dev)
703 {
704 	struct pci_dev *pci_dev = vp_dev->pci_dev;
705 
706 	if (vp_dev->device)
707 		pci_iounmap(pci_dev, vp_dev->device);
708 	if (vp_dev->notify_base)
709 		pci_iounmap(pci_dev, vp_dev->notify_base);
710 	pci_iounmap(pci_dev, vp_dev->isr);
711 	pci_iounmap(pci_dev, vp_dev->common);
712 	pci_release_selected_regions(pci_dev, vp_dev->modern_bars);
713 }
714