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 
22 /*
23  * Type-safe wrappers for io accesses.
24  * Use these to enforce at compile time the following spec requirement:
25  *
26  * The driver MUST access each field using the “natural” access
27  * method, i.e. 32-bit accesses for 32-bit fields, 16-bit accesses
28  * for 16-bit fields and 8-bit accesses for 8-bit fields.
29  */
30 static inline u8 vp_ioread8(const u8 __iomem *addr)
31 {
32 	return ioread8(addr);
33 }
34 static inline u16 vp_ioread16 (const __le16 __iomem *addr)
35 {
36 	return ioread16(addr);
37 }
38 
39 static inline u32 vp_ioread32(const __le32 __iomem *addr)
40 {
41 	return ioread32(addr);
42 }
43 
44 static inline void vp_iowrite8(u8 value, u8 __iomem *addr)
45 {
46 	iowrite8(value, addr);
47 }
48 
49 static inline void vp_iowrite16(u16 value, __le16 __iomem *addr)
50 {
51 	iowrite16(value, addr);
52 }
53 
54 static inline void vp_iowrite32(u32 value, __le32 __iomem *addr)
55 {
56 	iowrite32(value, addr);
57 }
58 
59 static void vp_iowrite64_twopart(u64 val,
60 				 __le32 __iomem *lo, __le32 __iomem *hi)
61 {
62 	vp_iowrite32((u32)val, lo);
63 	vp_iowrite32(val >> 32, hi);
64 }
65 
66 static void __iomem *map_capability(struct pci_dev *dev, int off,
67 				    size_t minlen,
68 				    u32 align,
69 				    u32 start, u32 size,
70 				    size_t *len)
71 {
72 	u8 bar;
73 	u32 offset, length;
74 	void __iomem *p;
75 
76 	pci_read_config_byte(dev, off + offsetof(struct virtio_pci_cap,
77 						 bar),
78 			     &bar);
79 	pci_read_config_dword(dev, off + offsetof(struct virtio_pci_cap, offset),
80 			     &offset);
81 	pci_read_config_dword(dev, off + offsetof(struct virtio_pci_cap, length),
82 			      &length);
83 
84 	if (length <= start) {
85 		dev_err(&dev->dev,
86 			"virtio_pci: bad capability len %u (>%u expected)\n",
87 			length, start);
88 		return NULL;
89 	}
90 
91 	if (length - start < minlen) {
92 		dev_err(&dev->dev,
93 			"virtio_pci: bad capability len %u (>=%zu expected)\n",
94 			length, minlen);
95 		return NULL;
96 	}
97 
98 	length -= start;
99 
100 	if (start + offset < offset) {
101 		dev_err(&dev->dev,
102 			"virtio_pci: map wrap-around %u+%u\n",
103 			start, offset);
104 		return NULL;
105 	}
106 
107 	offset += start;
108 
109 	if (offset & (align - 1)) {
110 		dev_err(&dev->dev,
111 			"virtio_pci: offset %u not aligned to %u\n",
112 			offset, align);
113 		return NULL;
114 	}
115 
116 	if (length > size)
117 		length = size;
118 
119 	if (len)
120 		*len = length;
121 
122 	if (minlen + offset < minlen ||
123 	    minlen + offset > pci_resource_len(dev, bar)) {
124 		dev_err(&dev->dev,
125 			"virtio_pci: map virtio %zu@%u "
126 			"out of range on bar %i length %lu\n",
127 			minlen, offset,
128 			bar, (unsigned long)pci_resource_len(dev, bar));
129 		return NULL;
130 	}
131 
132 	p = pci_iomap_range(dev, bar, offset, length);
133 	if (!p)
134 		dev_err(&dev->dev,
135 			"virtio_pci: unable to map virtio %u@%u on bar %i\n",
136 			length, offset, bar);
137 	return p;
138 }
139 
140 /* virtio config->get_features() implementation */
141 static u64 vp_get_features(struct virtio_device *vdev)
142 {
143 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
144 	u64 features;
145 
146 	vp_iowrite32(0, &vp_dev->common->device_feature_select);
147 	features = vp_ioread32(&vp_dev->common->device_feature);
148 	vp_iowrite32(1, &vp_dev->common->device_feature_select);
149 	features |= ((u64)vp_ioread32(&vp_dev->common->device_feature) << 32);
150 
151 	return features;
152 }
153 
154 static void vp_transport_features(struct virtio_device *vdev, u64 features)
155 {
156 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
157 	struct pci_dev *pci_dev = vp_dev->pci_dev;
158 
159 	if ((features & BIT_ULL(VIRTIO_F_SR_IOV)) &&
160 			pci_find_ext_capability(pci_dev, PCI_EXT_CAP_ID_SRIOV))
161 		__virtio_set_bit(vdev, VIRTIO_F_SR_IOV);
162 }
163 
164 /* virtio config->finalize_features() implementation */
165 static int vp_finalize_features(struct virtio_device *vdev)
166 {
167 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
168 	u64 features = vdev->features;
169 
170 	/* Give virtio_ring a chance to accept features. */
171 	vring_transport_features(vdev);
172 
173 	/* Give virtio_pci a chance to accept features. */
174 	vp_transport_features(vdev, features);
175 
176 	if (!__virtio_test_bit(vdev, VIRTIO_F_VERSION_1)) {
177 		dev_err(&vdev->dev, "virtio: device uses modern interface "
178 			"but does not have VIRTIO_F_VERSION_1\n");
179 		return -EINVAL;
180 	}
181 
182 	vp_iowrite32(0, &vp_dev->common->guest_feature_select);
183 	vp_iowrite32((u32)vdev->features, &vp_dev->common->guest_feature);
184 	vp_iowrite32(1, &vp_dev->common->guest_feature_select);
185 	vp_iowrite32(vdev->features >> 32, &vp_dev->common->guest_feature);
186 
187 	return 0;
188 }
189 
190 /* virtio config->get() implementation */
191 static void vp_get(struct virtio_device *vdev, unsigned offset,
192 		   void *buf, unsigned len)
193 {
194 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
195 	u8 b;
196 	__le16 w;
197 	__le32 l;
198 
199 	BUG_ON(offset + len > vp_dev->device_len);
200 
201 	switch (len) {
202 	case 1:
203 		b = ioread8(vp_dev->device + offset);
204 		memcpy(buf, &b, sizeof b);
205 		break;
206 	case 2:
207 		w = cpu_to_le16(ioread16(vp_dev->device + offset));
208 		memcpy(buf, &w, sizeof w);
209 		break;
210 	case 4:
211 		l = cpu_to_le32(ioread32(vp_dev->device + offset));
212 		memcpy(buf, &l, sizeof l);
213 		break;
214 	case 8:
215 		l = cpu_to_le32(ioread32(vp_dev->device + offset));
216 		memcpy(buf, &l, sizeof l);
217 		l = cpu_to_le32(ioread32(vp_dev->device + offset + sizeof l));
218 		memcpy(buf + sizeof l, &l, sizeof l);
219 		break;
220 	default:
221 		BUG();
222 	}
223 }
224 
225 /* the config->set() implementation.  it's symmetric to the config->get()
226  * implementation */
227 static void vp_set(struct virtio_device *vdev, unsigned offset,
228 		   const void *buf, unsigned len)
229 {
230 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
231 	u8 b;
232 	__le16 w;
233 	__le32 l;
234 
235 	BUG_ON(offset + len > vp_dev->device_len);
236 
237 	switch (len) {
238 	case 1:
239 		memcpy(&b, buf, sizeof b);
240 		iowrite8(b, vp_dev->device + offset);
241 		break;
242 	case 2:
243 		memcpy(&w, buf, sizeof w);
244 		iowrite16(le16_to_cpu(w), vp_dev->device + offset);
245 		break;
246 	case 4:
247 		memcpy(&l, buf, sizeof l);
248 		iowrite32(le32_to_cpu(l), vp_dev->device + offset);
249 		break;
250 	case 8:
251 		memcpy(&l, buf, sizeof l);
252 		iowrite32(le32_to_cpu(l), vp_dev->device + offset);
253 		memcpy(&l, buf + sizeof l, sizeof l);
254 		iowrite32(le32_to_cpu(l), vp_dev->device + offset + sizeof l);
255 		break;
256 	default:
257 		BUG();
258 	}
259 }
260 
261 static u32 vp_generation(struct virtio_device *vdev)
262 {
263 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
264 	return vp_ioread8(&vp_dev->common->config_generation);
265 }
266 
267 /* config->{get,set}_status() implementations */
268 static u8 vp_get_status(struct virtio_device *vdev)
269 {
270 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
271 	return vp_ioread8(&vp_dev->common->device_status);
272 }
273 
274 static void vp_set_status(struct virtio_device *vdev, u8 status)
275 {
276 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
277 	/* We should never be setting status to 0. */
278 	BUG_ON(status == 0);
279 	vp_iowrite8(status, &vp_dev->common->device_status);
280 }
281 
282 static void vp_reset(struct virtio_device *vdev)
283 {
284 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
285 	/* 0 status means a reset. */
286 	vp_iowrite8(0, &vp_dev->common->device_status);
287 	/* After writing 0 to device_status, the driver MUST wait for a read of
288 	 * device_status to return 0 before reinitializing the device.
289 	 * This will flush out the status write, and flush in device writes,
290 	 * including MSI-X interrupts, if any.
291 	 */
292 	while (vp_ioread8(&vp_dev->common->device_status))
293 		msleep(1);
294 	/* Flush pending VQ/configuration callbacks. */
295 	vp_synchronize_vectors(vdev);
296 }
297 
298 static u16 vp_config_vector(struct virtio_pci_device *vp_dev, u16 vector)
299 {
300 	/* Setup the vector used for configuration events */
301 	vp_iowrite16(vector, &vp_dev->common->msix_config);
302 	/* Verify we had enough resources to assign the vector */
303 	/* Will also flush the write out to device */
304 	return vp_ioread16(&vp_dev->common->msix_config);
305 }
306 
307 static struct virtqueue *setup_vq(struct virtio_pci_device *vp_dev,
308 				  struct virtio_pci_vq_info *info,
309 				  unsigned index,
310 				  void (*callback)(struct virtqueue *vq),
311 				  const char *name,
312 				  bool ctx,
313 				  u16 msix_vec)
314 {
315 	struct virtio_pci_common_cfg __iomem *cfg = vp_dev->common;
316 	struct virtqueue *vq;
317 	u16 num, off;
318 	int err;
319 
320 	if (index >= vp_ioread16(&cfg->num_queues))
321 		return ERR_PTR(-ENOENT);
322 
323 	/* Select the queue we're interested in */
324 	vp_iowrite16(index, &cfg->queue_select);
325 
326 	/* Check if queue is either not available or already active. */
327 	num = vp_ioread16(&cfg->queue_size);
328 	if (!num || vp_ioread16(&cfg->queue_enable))
329 		return ERR_PTR(-ENOENT);
330 
331 	if (num & (num - 1)) {
332 		dev_warn(&vp_dev->pci_dev->dev, "bad queue size %u", num);
333 		return ERR_PTR(-EINVAL);
334 	}
335 
336 	/* get offset of notification word for this vq */
337 	off = vp_ioread16(&cfg->queue_notify_off);
338 
339 	info->msix_vector = msix_vec;
340 
341 	/* create the vring */
342 	vq = vring_create_virtqueue(index, num,
343 				    SMP_CACHE_BYTES, &vp_dev->vdev,
344 				    true, true, ctx,
345 				    vp_notify, callback, name);
346 	if (!vq)
347 		return ERR_PTR(-ENOMEM);
348 
349 	/* activate the queue */
350 	vp_iowrite16(virtqueue_get_vring_size(vq), &cfg->queue_size);
351 	vp_iowrite64_twopart(virtqueue_get_desc_addr(vq),
352 			     &cfg->queue_desc_lo, &cfg->queue_desc_hi);
353 	vp_iowrite64_twopart(virtqueue_get_avail_addr(vq),
354 			     &cfg->queue_avail_lo, &cfg->queue_avail_hi);
355 	vp_iowrite64_twopart(virtqueue_get_used_addr(vq),
356 			     &cfg->queue_used_lo, &cfg->queue_used_hi);
357 
358 	if (vp_dev->notify_base) {
359 		/* offset should not wrap */
360 		if ((u64)off * vp_dev->notify_offset_multiplier + 2
361 		    > vp_dev->notify_len) {
362 			dev_warn(&vp_dev->pci_dev->dev,
363 				 "bad notification offset %u (x %u) "
364 				 "for queue %u > %zd",
365 				 off, vp_dev->notify_offset_multiplier,
366 				 index, vp_dev->notify_len);
367 			err = -EINVAL;
368 			goto err_map_notify;
369 		}
370 		vq->priv = (void __force *)vp_dev->notify_base +
371 			off * vp_dev->notify_offset_multiplier;
372 	} else {
373 		vq->priv = (void __force *)map_capability(vp_dev->pci_dev,
374 					  vp_dev->notify_map_cap, 2, 2,
375 					  off * vp_dev->notify_offset_multiplier, 2,
376 					  NULL);
377 	}
378 
379 	if (!vq->priv) {
380 		err = -ENOMEM;
381 		goto err_map_notify;
382 	}
383 
384 	if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
385 		vp_iowrite16(msix_vec, &cfg->queue_msix_vector);
386 		msix_vec = vp_ioread16(&cfg->queue_msix_vector);
387 		if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
388 			err = -EBUSY;
389 			goto err_assign_vector;
390 		}
391 	}
392 
393 	return vq;
394 
395 err_assign_vector:
396 	if (!vp_dev->notify_base)
397 		pci_iounmap(vp_dev->pci_dev, (void __iomem __force *)vq->priv);
398 err_map_notify:
399 	vring_del_virtqueue(vq);
400 	return ERR_PTR(err);
401 }
402 
403 static int vp_modern_find_vqs(struct virtio_device *vdev, unsigned nvqs,
404 			      struct virtqueue *vqs[],
405 			      vq_callback_t *callbacks[],
406 			      const char * const names[], const bool *ctx,
407 			      struct irq_affinity *desc)
408 {
409 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
410 	struct virtqueue *vq;
411 	int rc = vp_find_vqs(vdev, nvqs, vqs, callbacks, names, ctx, desc);
412 
413 	if (rc)
414 		return rc;
415 
416 	/* Select and activate all queues. Has to be done last: once we do
417 	 * this, there's no way to go back except reset.
418 	 */
419 	list_for_each_entry(vq, &vdev->vqs, list) {
420 		vp_iowrite16(vq->index, &vp_dev->common->queue_select);
421 		vp_iowrite16(1, &vp_dev->common->queue_enable);
422 	}
423 
424 	return 0;
425 }
426 
427 static void del_vq(struct virtio_pci_vq_info *info)
428 {
429 	struct virtqueue *vq = info->vq;
430 	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
431 
432 	vp_iowrite16(vq->index, &vp_dev->common->queue_select);
433 
434 	if (vp_dev->msix_enabled) {
435 		vp_iowrite16(VIRTIO_MSI_NO_VECTOR,
436 			     &vp_dev->common->queue_msix_vector);
437 		/* Flush the write out to device */
438 		vp_ioread16(&vp_dev->common->queue_msix_vector);
439 	}
440 
441 	if (!vp_dev->notify_base)
442 		pci_iounmap(vp_dev->pci_dev, (void __force __iomem *)vq->priv);
443 
444 	vring_del_virtqueue(vq);
445 }
446 
447 static int virtio_pci_find_shm_cap(struct pci_dev *dev, u8 required_id,
448 				   u8 *bar, u64 *offset, u64 *len)
449 {
450 	int pos;
451 
452 	for (pos = pci_find_capability(dev, PCI_CAP_ID_VNDR); pos > 0;
453 	     pos = pci_find_next_capability(dev, pos, PCI_CAP_ID_VNDR)) {
454 		u8 type, cap_len, id;
455 		u32 tmp32;
456 		u64 res_offset, res_length;
457 
458 		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
459 							 cfg_type), &type);
460 		if (type != VIRTIO_PCI_CAP_SHARED_MEMORY_CFG)
461 			continue;
462 
463 		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
464 							 cap_len), &cap_len);
465 		if (cap_len != sizeof(struct virtio_pci_cap64)) {
466 			dev_err(&dev->dev, "%s: shm cap with bad size offset:"
467 				" %d size: %d\n", __func__, pos, cap_len);
468 			continue;
469 		}
470 
471 		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
472 							 id), &id);
473 		if (id != required_id)
474 			continue;
475 
476 		/* Type, and ID match, looks good */
477 		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
478 							 bar), bar);
479 
480 		/* Read the lower 32bit of length and offset */
481 		pci_read_config_dword(dev, pos + offsetof(struct virtio_pci_cap,
482 							  offset), &tmp32);
483 		res_offset = tmp32;
484 		pci_read_config_dword(dev, pos + offsetof(struct virtio_pci_cap,
485 							  length), &tmp32);
486 		res_length = tmp32;
487 
488 		/* and now the top half */
489 		pci_read_config_dword(dev,
490 				      pos + offsetof(struct virtio_pci_cap64,
491 						     offset_hi), &tmp32);
492 		res_offset |= ((u64)tmp32) << 32;
493 		pci_read_config_dword(dev,
494 				      pos + offsetof(struct virtio_pci_cap64,
495 						     length_hi), &tmp32);
496 		res_length |= ((u64)tmp32) << 32;
497 
498 		*offset = res_offset;
499 		*len = res_length;
500 
501 		return pos;
502 	}
503 	return 0;
504 }
505 
506 static bool vp_get_shm_region(struct virtio_device *vdev,
507 			      struct virtio_shm_region *region, u8 id)
508 {
509 	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
510 	struct pci_dev *pci_dev = vp_dev->pci_dev;
511 	u8 bar;
512 	u64 offset, len;
513 	phys_addr_t phys_addr;
514 	size_t bar_len;
515 
516 	if (!virtio_pci_find_shm_cap(pci_dev, id, &bar, &offset, &len))
517 		return false;
518 
519 	phys_addr = pci_resource_start(pci_dev, bar);
520 	bar_len = pci_resource_len(pci_dev, bar);
521 
522 	if ((offset + len) < offset) {
523 		dev_err(&pci_dev->dev, "%s: cap offset+len overflow detected\n",
524 			__func__);
525 		return false;
526 	}
527 
528 	if (offset + len > bar_len) {
529 		dev_err(&pci_dev->dev, "%s: bar shorter than cap offset+len\n",
530 			__func__);
531 		return false;
532 	}
533 
534 	region->len = len;
535 	region->addr = (u64) phys_addr + offset;
536 
537 	return true;
538 }
539 
540 static const struct virtio_config_ops virtio_pci_config_nodev_ops = {
541 	.get		= NULL,
542 	.set		= NULL,
543 	.generation	= vp_generation,
544 	.get_status	= vp_get_status,
545 	.set_status	= vp_set_status,
546 	.reset		= vp_reset,
547 	.find_vqs	= vp_modern_find_vqs,
548 	.del_vqs	= vp_del_vqs,
549 	.get_features	= vp_get_features,
550 	.finalize_features = vp_finalize_features,
551 	.bus_name	= vp_bus_name,
552 	.set_vq_affinity = vp_set_vq_affinity,
553 	.get_vq_affinity = vp_get_vq_affinity,
554 	.get_shm_region  = vp_get_shm_region,
555 };
556 
557 static const struct virtio_config_ops virtio_pci_config_ops = {
558 	.get		= vp_get,
559 	.set		= vp_set,
560 	.generation	= vp_generation,
561 	.get_status	= vp_get_status,
562 	.set_status	= vp_set_status,
563 	.reset		= vp_reset,
564 	.find_vqs	= vp_modern_find_vqs,
565 	.del_vqs	= vp_del_vqs,
566 	.get_features	= vp_get_features,
567 	.finalize_features = vp_finalize_features,
568 	.bus_name	= vp_bus_name,
569 	.set_vq_affinity = vp_set_vq_affinity,
570 	.get_vq_affinity = vp_get_vq_affinity,
571 	.get_shm_region  = vp_get_shm_region,
572 };
573 
574 /**
575  * virtio_pci_find_capability - walk capabilities to find device info.
576  * @dev: the pci device
577  * @cfg_type: the VIRTIO_PCI_CAP_* value we seek
578  * @ioresource_types: IORESOURCE_MEM and/or IORESOURCE_IO.
579  * @bars: the bitmask of BARs
580  *
581  * Returns offset of the capability, or 0.
582  */
583 static inline int virtio_pci_find_capability(struct pci_dev *dev, u8 cfg_type,
584 					     u32 ioresource_types, int *bars)
585 {
586 	int pos;
587 
588 	for (pos = pci_find_capability(dev, PCI_CAP_ID_VNDR);
589 	     pos > 0;
590 	     pos = pci_find_next_capability(dev, pos, PCI_CAP_ID_VNDR)) {
591 		u8 type, bar;
592 		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
593 							 cfg_type),
594 				     &type);
595 		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
596 							 bar),
597 				     &bar);
598 
599 		/* Ignore structures with reserved BAR values */
600 		if (bar > 0x5)
601 			continue;
602 
603 		if (type == cfg_type) {
604 			if (pci_resource_len(dev, bar) &&
605 			    pci_resource_flags(dev, bar) & ioresource_types) {
606 				*bars |= (1 << bar);
607 				return pos;
608 			}
609 		}
610 	}
611 	return 0;
612 }
613 
614 /* This is part of the ABI.  Don't screw with it. */
615 static inline void check_offsets(void)
616 {
617 	/* Note: disk space was harmed in compilation of this function. */
618 	BUILD_BUG_ON(VIRTIO_PCI_CAP_VNDR !=
619 		     offsetof(struct virtio_pci_cap, cap_vndr));
620 	BUILD_BUG_ON(VIRTIO_PCI_CAP_NEXT !=
621 		     offsetof(struct virtio_pci_cap, cap_next));
622 	BUILD_BUG_ON(VIRTIO_PCI_CAP_LEN !=
623 		     offsetof(struct virtio_pci_cap, cap_len));
624 	BUILD_BUG_ON(VIRTIO_PCI_CAP_CFG_TYPE !=
625 		     offsetof(struct virtio_pci_cap, cfg_type));
626 	BUILD_BUG_ON(VIRTIO_PCI_CAP_BAR !=
627 		     offsetof(struct virtio_pci_cap, bar));
628 	BUILD_BUG_ON(VIRTIO_PCI_CAP_OFFSET !=
629 		     offsetof(struct virtio_pci_cap, offset));
630 	BUILD_BUG_ON(VIRTIO_PCI_CAP_LENGTH !=
631 		     offsetof(struct virtio_pci_cap, length));
632 	BUILD_BUG_ON(VIRTIO_PCI_NOTIFY_CAP_MULT !=
633 		     offsetof(struct virtio_pci_notify_cap,
634 			      notify_off_multiplier));
635 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_DFSELECT !=
636 		     offsetof(struct virtio_pci_common_cfg,
637 			      device_feature_select));
638 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_DF !=
639 		     offsetof(struct virtio_pci_common_cfg, device_feature));
640 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_GFSELECT !=
641 		     offsetof(struct virtio_pci_common_cfg,
642 			      guest_feature_select));
643 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_GF !=
644 		     offsetof(struct virtio_pci_common_cfg, guest_feature));
645 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_MSIX !=
646 		     offsetof(struct virtio_pci_common_cfg, msix_config));
647 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_NUMQ !=
648 		     offsetof(struct virtio_pci_common_cfg, num_queues));
649 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_STATUS !=
650 		     offsetof(struct virtio_pci_common_cfg, device_status));
651 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_CFGGENERATION !=
652 		     offsetof(struct virtio_pci_common_cfg, config_generation));
653 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SELECT !=
654 		     offsetof(struct virtio_pci_common_cfg, queue_select));
655 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SIZE !=
656 		     offsetof(struct virtio_pci_common_cfg, queue_size));
657 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_MSIX !=
658 		     offsetof(struct virtio_pci_common_cfg, queue_msix_vector));
659 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_ENABLE !=
660 		     offsetof(struct virtio_pci_common_cfg, queue_enable));
661 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_NOFF !=
662 		     offsetof(struct virtio_pci_common_cfg, queue_notify_off));
663 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCLO !=
664 		     offsetof(struct virtio_pci_common_cfg, queue_desc_lo));
665 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCHI !=
666 		     offsetof(struct virtio_pci_common_cfg, queue_desc_hi));
667 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILLO !=
668 		     offsetof(struct virtio_pci_common_cfg, queue_avail_lo));
669 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILHI !=
670 		     offsetof(struct virtio_pci_common_cfg, queue_avail_hi));
671 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDLO !=
672 		     offsetof(struct virtio_pci_common_cfg, queue_used_lo));
673 	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDHI !=
674 		     offsetof(struct virtio_pci_common_cfg, queue_used_hi));
675 }
676 
677 /* the PCI probing function */
678 int virtio_pci_modern_probe(struct virtio_pci_device *vp_dev)
679 {
680 	struct pci_dev *pci_dev = vp_dev->pci_dev;
681 	int err, common, isr, notify, device;
682 	u32 notify_length;
683 	u32 notify_offset;
684 
685 	check_offsets();
686 
687 	/* We only own devices >= 0x1000 and <= 0x107f: leave the rest. */
688 	if (pci_dev->device < 0x1000 || pci_dev->device > 0x107f)
689 		return -ENODEV;
690 
691 	if (pci_dev->device < 0x1040) {
692 		/* Transitional devices: use the PCI subsystem device id as
693 		 * virtio device id, same as legacy driver always did.
694 		 */
695 		vp_dev->vdev.id.device = pci_dev->subsystem_device;
696 	} else {
697 		/* Modern devices: simply use PCI device id, but start from 0x1040. */
698 		vp_dev->vdev.id.device = pci_dev->device - 0x1040;
699 	}
700 	vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
701 
702 	/* check for a common config: if not, use legacy mode (bar 0). */
703 	common = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_COMMON_CFG,
704 					    IORESOURCE_IO | IORESOURCE_MEM,
705 					    &vp_dev->modern_bars);
706 	if (!common) {
707 		dev_info(&pci_dev->dev,
708 			 "virtio_pci: leaving for legacy driver\n");
709 		return -ENODEV;
710 	}
711 
712 	/* If common is there, these should be too... */
713 	isr = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_ISR_CFG,
714 					 IORESOURCE_IO | IORESOURCE_MEM,
715 					 &vp_dev->modern_bars);
716 	notify = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_NOTIFY_CFG,
717 					    IORESOURCE_IO | IORESOURCE_MEM,
718 					    &vp_dev->modern_bars);
719 	if (!isr || !notify) {
720 		dev_err(&pci_dev->dev,
721 			"virtio_pci: missing capabilities %i/%i/%i\n",
722 			common, isr, notify);
723 		return -EINVAL;
724 	}
725 
726 	err = dma_set_mask_and_coherent(&pci_dev->dev, DMA_BIT_MASK(64));
727 	if (err)
728 		err = dma_set_mask_and_coherent(&pci_dev->dev,
729 						DMA_BIT_MASK(32));
730 	if (err)
731 		dev_warn(&pci_dev->dev, "Failed to enable 64-bit or 32-bit DMA.  Trying to continue, but this might not work.\n");
732 
733 	/* Device capability is only mandatory for devices that have
734 	 * device-specific configuration.
735 	 */
736 	device = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_DEVICE_CFG,
737 					    IORESOURCE_IO | IORESOURCE_MEM,
738 					    &vp_dev->modern_bars);
739 
740 	err = pci_request_selected_regions(pci_dev, vp_dev->modern_bars,
741 					   "virtio-pci-modern");
742 	if (err)
743 		return err;
744 
745 	err = -EINVAL;
746 	vp_dev->common = map_capability(pci_dev, common,
747 					sizeof(struct virtio_pci_common_cfg), 4,
748 					0, sizeof(struct virtio_pci_common_cfg),
749 					NULL);
750 	if (!vp_dev->common)
751 		goto err_map_common;
752 	vp_dev->isr = map_capability(pci_dev, isr, sizeof(u8), 1,
753 				     0, 1,
754 				     NULL);
755 	if (!vp_dev->isr)
756 		goto err_map_isr;
757 
758 	/* Read notify_off_multiplier from config space. */
759 	pci_read_config_dword(pci_dev,
760 			      notify + offsetof(struct virtio_pci_notify_cap,
761 						notify_off_multiplier),
762 			      &vp_dev->notify_offset_multiplier);
763 	/* Read notify length and offset from config space. */
764 	pci_read_config_dword(pci_dev,
765 			      notify + offsetof(struct virtio_pci_notify_cap,
766 						cap.length),
767 			      &notify_length);
768 
769 	pci_read_config_dword(pci_dev,
770 			      notify + offsetof(struct virtio_pci_notify_cap,
771 						cap.offset),
772 			      &notify_offset);
773 
774 	/* We don't know how many VQs we'll map, ahead of the time.
775 	 * If notify length is small, map it all now.
776 	 * Otherwise, map each VQ individually later.
777 	 */
778 	if ((u64)notify_length + (notify_offset % PAGE_SIZE) <= PAGE_SIZE) {
779 		vp_dev->notify_base = map_capability(pci_dev, notify, 2, 2,
780 						     0, notify_length,
781 						     &vp_dev->notify_len);
782 		if (!vp_dev->notify_base)
783 			goto err_map_notify;
784 	} else {
785 		vp_dev->notify_map_cap = notify;
786 	}
787 
788 	/* Again, we don't know how much we should map, but PAGE_SIZE
789 	 * is more than enough for all existing devices.
790 	 */
791 	if (device) {
792 		vp_dev->device = map_capability(pci_dev, device, 0, 4,
793 						0, PAGE_SIZE,
794 						&vp_dev->device_len);
795 		if (!vp_dev->device)
796 			goto err_map_device;
797 
798 		vp_dev->vdev.config = &virtio_pci_config_ops;
799 	} else {
800 		vp_dev->vdev.config = &virtio_pci_config_nodev_ops;
801 	}
802 
803 	vp_dev->config_vector = vp_config_vector;
804 	vp_dev->setup_vq = setup_vq;
805 	vp_dev->del_vq = del_vq;
806 
807 	return 0;
808 
809 err_map_device:
810 	if (vp_dev->notify_base)
811 		pci_iounmap(pci_dev, vp_dev->notify_base);
812 err_map_notify:
813 	pci_iounmap(pci_dev, vp_dev->isr);
814 err_map_isr:
815 	pci_iounmap(pci_dev, vp_dev->common);
816 err_map_common:
817 	return err;
818 }
819 
820 void virtio_pci_modern_remove(struct virtio_pci_device *vp_dev)
821 {
822 	struct pci_dev *pci_dev = vp_dev->pci_dev;
823 
824 	if (vp_dev->device)
825 		pci_iounmap(pci_dev, vp_dev->device);
826 	if (vp_dev->notify_base)
827 		pci_iounmap(pci_dev, vp_dev->notify_base);
828 	pci_iounmap(pci_dev, vp_dev->isr);
829 	pci_iounmap(pci_dev, vp_dev->common);
830 	pci_release_selected_regions(pci_dev, vp_dev->modern_bars);
831 }
832