Linux Audio

Check our new training course

Loading...
v5.9
  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 */
 30static inline u8 vp_ioread8(const u8 __iomem *addr)
 31{
 32	return ioread8(addr);
 33}
 34static inline u16 vp_ioread16 (const __le16 __iomem *addr)
 35{
 36	return ioread16(addr);
 37}
 38
 39static inline u32 vp_ioread32(const __le32 __iomem *addr)
 40{
 41	return ioread32(addr);
 42}
 43
 44static inline void vp_iowrite8(u8 value, u8 __iomem *addr)
 45{
 46	iowrite8(value, addr);
 47}
 48
 49static inline void vp_iowrite16(u16 value, __le16 __iomem *addr)
 50{
 51	iowrite16(value, addr);
 52}
 53
 54static inline void vp_iowrite32(u32 value, __le32 __iomem *addr)
 55{
 56	iowrite32(value, addr);
 57}
 58
 59static 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
 66static 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 */
141static 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
154static 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 */
165static 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 */
191static 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 */
227static 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
261static 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 */
268static 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
274static 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
282static 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
298static 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
307static 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
395err_assign_vector:
396	if (!vp_dev->notify_base)
397		pci_iounmap(vp_dev->pci_dev, (void __iomem __force *)vq->priv);
398err_map_notify:
399	vring_del_virtqueue(vq);
400	return ERR_PTR(err);
401}
402
403static 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
427static 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
447static const struct virtio_config_ops virtio_pci_config_nodev_ops = {
448	.get		= NULL,
449	.set		= NULL,
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	.get_vq_affinity = vp_get_vq_affinity,
 
461};
462
463static const struct virtio_config_ops virtio_pci_config_ops = {
464	.get		= vp_get,
465	.set		= vp_set,
466	.generation	= vp_generation,
467	.get_status	= vp_get_status,
468	.set_status	= vp_set_status,
469	.reset		= vp_reset,
470	.find_vqs	= vp_modern_find_vqs,
471	.del_vqs	= vp_del_vqs,
472	.get_features	= vp_get_features,
473	.finalize_features = vp_finalize_features,
474	.bus_name	= vp_bus_name,
475	.set_vq_affinity = vp_set_vq_affinity,
476	.get_vq_affinity = vp_get_vq_affinity,
 
477};
478
479/**
480 * virtio_pci_find_capability - walk capabilities to find device info.
481 * @dev: the pci device
482 * @cfg_type: the VIRTIO_PCI_CAP_* value we seek
483 * @ioresource_types: IORESOURCE_MEM and/or IORESOURCE_IO.
484 * @bars: the bitmask of BARs
485 *
486 * Returns offset of the capability, or 0.
487 */
488static inline int virtio_pci_find_capability(struct pci_dev *dev, u8 cfg_type,
489					     u32 ioresource_types, int *bars)
490{
491	int pos;
492
493	for (pos = pci_find_capability(dev, PCI_CAP_ID_VNDR);
494	     pos > 0;
495	     pos = pci_find_next_capability(dev, pos, PCI_CAP_ID_VNDR)) {
496		u8 type, bar;
497		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
498							 cfg_type),
499				     &type);
500		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
501							 bar),
502				     &bar);
503
504		/* Ignore structures with reserved BAR values */
505		if (bar > 0x5)
506			continue;
507
508		if (type == cfg_type) {
509			if (pci_resource_len(dev, bar) &&
510			    pci_resource_flags(dev, bar) & ioresource_types) {
511				*bars |= (1 << bar);
512				return pos;
513			}
514		}
515	}
516	return 0;
517}
518
519/* This is part of the ABI.  Don't screw with it. */
520static inline void check_offsets(void)
521{
522	/* Note: disk space was harmed in compilation of this function. */
523	BUILD_BUG_ON(VIRTIO_PCI_CAP_VNDR !=
524		     offsetof(struct virtio_pci_cap, cap_vndr));
525	BUILD_BUG_ON(VIRTIO_PCI_CAP_NEXT !=
526		     offsetof(struct virtio_pci_cap, cap_next));
527	BUILD_BUG_ON(VIRTIO_PCI_CAP_LEN !=
528		     offsetof(struct virtio_pci_cap, cap_len));
529	BUILD_BUG_ON(VIRTIO_PCI_CAP_CFG_TYPE !=
530		     offsetof(struct virtio_pci_cap, cfg_type));
531	BUILD_BUG_ON(VIRTIO_PCI_CAP_BAR !=
532		     offsetof(struct virtio_pci_cap, bar));
533	BUILD_BUG_ON(VIRTIO_PCI_CAP_OFFSET !=
534		     offsetof(struct virtio_pci_cap, offset));
535	BUILD_BUG_ON(VIRTIO_PCI_CAP_LENGTH !=
536		     offsetof(struct virtio_pci_cap, length));
537	BUILD_BUG_ON(VIRTIO_PCI_NOTIFY_CAP_MULT !=
538		     offsetof(struct virtio_pci_notify_cap,
539			      notify_off_multiplier));
540	BUILD_BUG_ON(VIRTIO_PCI_COMMON_DFSELECT !=
541		     offsetof(struct virtio_pci_common_cfg,
542			      device_feature_select));
543	BUILD_BUG_ON(VIRTIO_PCI_COMMON_DF !=
544		     offsetof(struct virtio_pci_common_cfg, device_feature));
545	BUILD_BUG_ON(VIRTIO_PCI_COMMON_GFSELECT !=
546		     offsetof(struct virtio_pci_common_cfg,
547			      guest_feature_select));
548	BUILD_BUG_ON(VIRTIO_PCI_COMMON_GF !=
549		     offsetof(struct virtio_pci_common_cfg, guest_feature));
550	BUILD_BUG_ON(VIRTIO_PCI_COMMON_MSIX !=
551		     offsetof(struct virtio_pci_common_cfg, msix_config));
552	BUILD_BUG_ON(VIRTIO_PCI_COMMON_NUMQ !=
553		     offsetof(struct virtio_pci_common_cfg, num_queues));
554	BUILD_BUG_ON(VIRTIO_PCI_COMMON_STATUS !=
555		     offsetof(struct virtio_pci_common_cfg, device_status));
556	BUILD_BUG_ON(VIRTIO_PCI_COMMON_CFGGENERATION !=
557		     offsetof(struct virtio_pci_common_cfg, config_generation));
558	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SELECT !=
559		     offsetof(struct virtio_pci_common_cfg, queue_select));
560	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SIZE !=
561		     offsetof(struct virtio_pci_common_cfg, queue_size));
562	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_MSIX !=
563		     offsetof(struct virtio_pci_common_cfg, queue_msix_vector));
564	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_ENABLE !=
565		     offsetof(struct virtio_pci_common_cfg, queue_enable));
566	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_NOFF !=
567		     offsetof(struct virtio_pci_common_cfg, queue_notify_off));
568	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCLO !=
569		     offsetof(struct virtio_pci_common_cfg, queue_desc_lo));
570	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCHI !=
571		     offsetof(struct virtio_pci_common_cfg, queue_desc_hi));
572	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILLO !=
573		     offsetof(struct virtio_pci_common_cfg, queue_avail_lo));
574	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILHI !=
575		     offsetof(struct virtio_pci_common_cfg, queue_avail_hi));
576	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDLO !=
577		     offsetof(struct virtio_pci_common_cfg, queue_used_lo));
578	BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDHI !=
579		     offsetof(struct virtio_pci_common_cfg, queue_used_hi));
580}
581
582/* the PCI probing function */
583int virtio_pci_modern_probe(struct virtio_pci_device *vp_dev)
584{
 
585	struct pci_dev *pci_dev = vp_dev->pci_dev;
586	int err, common, isr, notify, device;
587	u32 notify_length;
588	u32 notify_offset;
589
590	check_offsets();
591
592	/* We only own devices >= 0x1000 and <= 0x107f: leave the rest. */
593	if (pci_dev->device < 0x1000 || pci_dev->device > 0x107f)
594		return -ENODEV;
595
596	if (pci_dev->device < 0x1040) {
597		/* Transitional devices: use the PCI subsystem device id as
598		 * virtio device id, same as legacy driver always did.
599		 */
600		vp_dev->vdev.id.device = pci_dev->subsystem_device;
601	} else {
602		/* Modern devices: simply use PCI device id, but start from 0x1040. */
603		vp_dev->vdev.id.device = pci_dev->device - 0x1040;
604	}
605	vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
606
607	/* check for a common config: if not, use legacy mode (bar 0). */
608	common = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_COMMON_CFG,
609					    IORESOURCE_IO | IORESOURCE_MEM,
610					    &vp_dev->modern_bars);
611	if (!common) {
612		dev_info(&pci_dev->dev,
613			 "virtio_pci: leaving for legacy driver\n");
614		return -ENODEV;
615	}
616
617	/* If common is there, these should be too... */
618	isr = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_ISR_CFG,
619					 IORESOURCE_IO | IORESOURCE_MEM,
620					 &vp_dev->modern_bars);
621	notify = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_NOTIFY_CFG,
622					    IORESOURCE_IO | IORESOURCE_MEM,
623					    &vp_dev->modern_bars);
624	if (!isr || !notify) {
625		dev_err(&pci_dev->dev,
626			"virtio_pci: missing capabilities %i/%i/%i\n",
627			common, isr, notify);
628		return -EINVAL;
629	}
630
631	err = dma_set_mask_and_coherent(&pci_dev->dev, DMA_BIT_MASK(64));
632	if (err)
633		err = dma_set_mask_and_coherent(&pci_dev->dev,
634						DMA_BIT_MASK(32));
635	if (err)
636		dev_warn(&pci_dev->dev, "Failed to enable 64-bit or 32-bit DMA.  Trying to continue, but this might not work.\n");
637
638	/* Device capability is only mandatory for devices that have
639	 * device-specific configuration.
640	 */
641	device = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_DEVICE_CFG,
642					    IORESOURCE_IO | IORESOURCE_MEM,
643					    &vp_dev->modern_bars);
644
645	err = pci_request_selected_regions(pci_dev, vp_dev->modern_bars,
646					   "virtio-pci-modern");
647	if (err)
648		return err;
649
650	err = -EINVAL;
651	vp_dev->common = map_capability(pci_dev, common,
652					sizeof(struct virtio_pci_common_cfg), 4,
653					0, sizeof(struct virtio_pci_common_cfg),
654					NULL);
655	if (!vp_dev->common)
656		goto err_map_common;
657	vp_dev->isr = map_capability(pci_dev, isr, sizeof(u8), 1,
658				     0, 1,
659				     NULL);
660	if (!vp_dev->isr)
661		goto err_map_isr;
662
663	/* Read notify_off_multiplier from config space. */
664	pci_read_config_dword(pci_dev,
665			      notify + offsetof(struct virtio_pci_notify_cap,
666						notify_off_multiplier),
667			      &vp_dev->notify_offset_multiplier);
668	/* Read notify length and offset from config space. */
669	pci_read_config_dword(pci_dev,
670			      notify + offsetof(struct virtio_pci_notify_cap,
671						cap.length),
672			      &notify_length);
673
674	pci_read_config_dword(pci_dev,
675			      notify + offsetof(struct virtio_pci_notify_cap,
676						cap.offset),
677			      &notify_offset);
678
679	/* We don't know how many VQs we'll map, ahead of the time.
680	 * If notify length is small, map it all now.
681	 * Otherwise, map each VQ individually later.
682	 */
683	if ((u64)notify_length + (notify_offset % PAGE_SIZE) <= PAGE_SIZE) {
684		vp_dev->notify_base = map_capability(pci_dev, notify, 2, 2,
685						     0, notify_length,
686						     &vp_dev->notify_len);
687		if (!vp_dev->notify_base)
688			goto err_map_notify;
689	} else {
690		vp_dev->notify_map_cap = notify;
691	}
692
693	/* Again, we don't know how much we should map, but PAGE_SIZE
694	 * is more than enough for all existing devices.
695	 */
696	if (device) {
697		vp_dev->device = map_capability(pci_dev, device, 0, 4,
698						0, PAGE_SIZE,
699						&vp_dev->device_len);
700		if (!vp_dev->device)
701			goto err_map_device;
702
703		vp_dev->vdev.config = &virtio_pci_config_ops;
704	} else {
705		vp_dev->vdev.config = &virtio_pci_config_nodev_ops;
706	}
707
708	vp_dev->config_vector = vp_config_vector;
709	vp_dev->setup_vq = setup_vq;
710	vp_dev->del_vq = del_vq;
 
 
711
712	return 0;
713
714err_map_device:
715	if (vp_dev->notify_base)
716		pci_iounmap(pci_dev, vp_dev->notify_base);
717err_map_notify:
718	pci_iounmap(pci_dev, vp_dev->isr);
719err_map_isr:
720	pci_iounmap(pci_dev, vp_dev->common);
721err_map_common:
722	return err;
723}
724
725void virtio_pci_modern_remove(struct virtio_pci_device *vp_dev)
726{
727	struct pci_dev *pci_dev = vp_dev->pci_dev;
728
729	if (vp_dev->device)
730		pci_iounmap(pci_dev, vp_dev->device);
731	if (vp_dev->notify_base)
732		pci_iounmap(pci_dev, vp_dev->notify_base);
733	pci_iounmap(pci_dev, vp_dev->isr);
734	pci_iounmap(pci_dev, vp_dev->common);
735	pci_release_selected_regions(pci_dev, vp_dev->modern_bars);
736}
v5.14.15
  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
 39/* virtio config->finalize_features() implementation */
 40static int vp_finalize_features(struct virtio_device *vdev)
 41{
 42	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 43	u64 features = vdev->features;
 44
 45	/* Give virtio_ring a chance to accept features. */
 46	vring_transport_features(vdev);
 47
 48	/* Give virtio_pci a chance to accept features. */
 49	vp_transport_features(vdev, features);
 50
 51	if (!__virtio_test_bit(vdev, VIRTIO_F_VERSION_1)) {
 52		dev_err(&vdev->dev, "virtio: device uses modern interface "
 53			"but does not have VIRTIO_F_VERSION_1\n");
 54		return -EINVAL;
 55	}
 56
 57	vp_modern_set_features(&vp_dev->mdev, vdev->features);
 
 
 
 58
 59	return 0;
 60}
 61
 62/* virtio config->get() implementation */
 63static void vp_get(struct virtio_device *vdev, unsigned offset,
 64		   void *buf, unsigned len)
 65{
 66	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 67	struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
 68	void __iomem *device = mdev->device;
 69	u8 b;
 70	__le16 w;
 71	__le32 l;
 72
 73	BUG_ON(offset + len > mdev->device_len);
 74
 75	switch (len) {
 76	case 1:
 77		b = ioread8(device + offset);
 78		memcpy(buf, &b, sizeof b);
 79		break;
 80	case 2:
 81		w = cpu_to_le16(ioread16(device + offset));
 82		memcpy(buf, &w, sizeof w);
 83		break;
 84	case 4:
 85		l = cpu_to_le32(ioread32(device + offset));
 86		memcpy(buf, &l, sizeof l);
 87		break;
 88	case 8:
 89		l = cpu_to_le32(ioread32(device + offset));
 90		memcpy(buf, &l, sizeof l);
 91		l = cpu_to_le32(ioread32(device + offset + sizeof l));
 92		memcpy(buf + sizeof l, &l, sizeof l);
 93		break;
 94	default:
 95		BUG();
 96	}
 97}
 98
 99/* the config->set() implementation.  it's symmetric to the config->get()
100 * implementation */
101static void vp_set(struct virtio_device *vdev, unsigned offset,
102		   const void *buf, unsigned len)
103{
104	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
105	struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
106	void __iomem *device = mdev->device;
107	u8 b;
108	__le16 w;
109	__le32 l;
110
111	BUG_ON(offset + len > mdev->device_len);
112
113	switch (len) {
114	case 1:
115		memcpy(&b, buf, sizeof b);
116		iowrite8(b, device + offset);
117		break;
118	case 2:
119		memcpy(&w, buf, sizeof w);
120		iowrite16(le16_to_cpu(w), device + offset);
121		break;
122	case 4:
123		memcpy(&l, buf, sizeof l);
124		iowrite32(le32_to_cpu(l), device + offset);
125		break;
126	case 8:
127		memcpy(&l, buf, sizeof l);
128		iowrite32(le32_to_cpu(l), device + offset);
129		memcpy(&l, buf + sizeof l, sizeof l);
130		iowrite32(le32_to_cpu(l), device + offset + sizeof l);
131		break;
132	default:
133		BUG();
134	}
135}
136
137static u32 vp_generation(struct virtio_device *vdev)
138{
139	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
140
141	return vp_modern_generation(&vp_dev->mdev);
142}
143
144/* config->{get,set}_status() implementations */
145static u8 vp_get_status(struct virtio_device *vdev)
146{
147	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
148
149	return vp_modern_get_status(&vp_dev->mdev);
150}
151
152static void vp_set_status(struct virtio_device *vdev, u8 status)
153{
154	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
155
156	/* We should never be setting status to 0. */
157	BUG_ON(status == 0);
158	vp_modern_set_status(&vp_dev->mdev, status);
159}
160
161static void vp_reset(struct virtio_device *vdev)
162{
163	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
164	struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
165
166	/* 0 status means a reset. */
167	vp_modern_set_status(mdev, 0);
168	/* After writing 0 to device_status, the driver MUST wait for a read of
169	 * device_status to return 0 before reinitializing the device.
170	 * This will flush out the status write, and flush in device writes,
171	 * including MSI-X interrupts, if any.
172	 */
173	while (vp_modern_get_status(mdev))
174		msleep(1);
175	/* Flush pending VQ/configuration callbacks. */
176	vp_synchronize_vectors(vdev);
177}
178
179static u16 vp_config_vector(struct virtio_pci_device *vp_dev, u16 vector)
180{
181	return vp_modern_config_vector(&vp_dev->mdev, vector);
 
 
 
 
182}
183
184static struct virtqueue *setup_vq(struct virtio_pci_device *vp_dev,
185				  struct virtio_pci_vq_info *info,
186				  unsigned index,
187				  void (*callback)(struct virtqueue *vq),
188				  const char *name,
189				  bool ctx,
190				  u16 msix_vec)
191{
192
193	struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
194	struct virtqueue *vq;
195	u16 num;
196	int err;
197
198	if (index >= vp_modern_get_num_queues(mdev))
199		return ERR_PTR(-ENOENT);
200
 
 
 
201	/* Check if queue is either not available or already active. */
202	num = vp_modern_get_queue_size(mdev, index);
203	if (!num || vp_modern_get_queue_enable(mdev, index))
204		return ERR_PTR(-ENOENT);
205
206	if (num & (num - 1)) {
207		dev_warn(&vp_dev->pci_dev->dev, "bad queue size %u", num);
208		return ERR_PTR(-EINVAL);
209	}
210
 
 
 
211	info->msix_vector = msix_vec;
212
213	/* create the vring */
214	vq = vring_create_virtqueue(index, num,
215				    SMP_CACHE_BYTES, &vp_dev->vdev,
216				    true, true, ctx,
217				    vp_notify, callback, name);
218	if (!vq)
219		return ERR_PTR(-ENOMEM);
220
221	/* activate the queue */
222	vp_modern_set_queue_size(mdev, index, virtqueue_get_vring_size(vq));
223	vp_modern_queue_address(mdev, index, virtqueue_get_desc_addr(vq),
224				virtqueue_get_avail_addr(vq),
225				virtqueue_get_used_addr(vq));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
226
227	vq->priv = (void __force *)vp_modern_map_vq_notify(mdev, index, NULL);
228	if (!vq->priv) {
229		err = -ENOMEM;
230		goto err_map_notify;
231	}
232
233	if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
234		msix_vec = vp_modern_queue_vector(mdev, index, msix_vec);
 
235		if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
236			err = -EBUSY;
237			goto err_assign_vector;
238		}
239	}
240
241	return vq;
242
243err_assign_vector:
244	if (!mdev->notify_base)
245		pci_iounmap(mdev->pci_dev, (void __iomem __force *)vq->priv);
246err_map_notify:
247	vring_del_virtqueue(vq);
248	return ERR_PTR(err);
249}
250
251static int vp_modern_find_vqs(struct virtio_device *vdev, unsigned nvqs,
252			      struct virtqueue *vqs[],
253			      vq_callback_t *callbacks[],
254			      const char * const names[], const bool *ctx,
255			      struct irq_affinity *desc)
256{
257	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
258	struct virtqueue *vq;
259	int rc = vp_find_vqs(vdev, nvqs, vqs, callbacks, names, ctx, desc);
260
261	if (rc)
262		return rc;
263
264	/* Select and activate all queues. Has to be done last: once we do
265	 * this, there's no way to go back except reset.
266	 */
267	list_for_each_entry(vq, &vdev->vqs, list)
268		vp_modern_set_queue_enable(&vp_dev->mdev, vq->index, true);
 
 
269
270	return 0;
271}
272
273static void del_vq(struct virtio_pci_vq_info *info)
274{
275	struct virtqueue *vq = info->vq;
276	struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
277	struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
278
279	if (vp_dev->msix_enabled)
280		vp_modern_queue_vector(mdev, vq->index,
281				       VIRTIO_MSI_NO_VECTOR);
282
283	if (!mdev->notify_base)
284		pci_iounmap(mdev->pci_dev, (void __force __iomem *)vq->priv);
285
286	vring_del_virtqueue(vq);
287}
288
289static int virtio_pci_find_shm_cap(struct pci_dev *dev, u8 required_id,
290				   u8 *bar, u64 *offset, u64 *len)
291{
292	int pos;
293
294	for (pos = pci_find_capability(dev, PCI_CAP_ID_VNDR); pos > 0;
295	     pos = pci_find_next_capability(dev, pos, PCI_CAP_ID_VNDR)) {
296		u8 type, cap_len, id;
297		u32 tmp32;
298		u64 res_offset, res_length;
299
300		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
301							 cfg_type), &type);
302		if (type != VIRTIO_PCI_CAP_SHARED_MEMORY_CFG)
303			continue;
304
305		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
306							 cap_len), &cap_len);
307		if (cap_len != sizeof(struct virtio_pci_cap64)) {
308			dev_err(&dev->dev, "%s: shm cap with bad size offset:"
309				" %d size: %d\n", __func__, pos, cap_len);
310			continue;
311		}
312
313		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
314							 id), &id);
315		if (id != required_id)
316			continue;
317
318		/* Type, and ID match, looks good */
319		pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
320							 bar), bar);
321
322		/* Read the lower 32bit of length and offset */
323		pci_read_config_dword(dev, pos + offsetof(struct virtio_pci_cap,
324							  offset), &tmp32);
325		res_offset = tmp32;
326		pci_read_config_dword(dev, pos + offsetof(struct virtio_pci_cap,
327							  length), &tmp32);
328		res_length = tmp32;
329
330		/* and now the top half */
331		pci_read_config_dword(dev,
332				      pos + offsetof(struct virtio_pci_cap64,
333						     offset_hi), &tmp32);
334		res_offset |= ((u64)tmp32) << 32;
335		pci_read_config_dword(dev,
336				      pos + offsetof(struct virtio_pci_cap64,
337						     length_hi), &tmp32);
338		res_length |= ((u64)tmp32) << 32;
339
340		*offset = res_offset;
341		*len = res_length;
342
343		return pos;
 
344	}
345	return 0;
346}
347
348static bool vp_get_shm_region(struct virtio_device *vdev,
349			      struct virtio_shm_region *region, u8 id)
350{
351	struct virtio_pci_device *vp_dev = to_vp_device(vdev);
352	struct pci_dev *pci_dev = vp_dev->pci_dev;
353	u8 bar;
354	u64 offset, len;
355	phys_addr_t phys_addr;
356	size_t bar_len;
357
358	if (!virtio_pci_find_shm_cap(pci_dev, id, &bar, &offset, &len))
359		return false;
360
361	phys_addr = pci_resource_start(pci_dev, bar);
362	bar_len = pci_resource_len(pci_dev, bar);
363
364	if ((offset + len) < offset) {
365		dev_err(&pci_dev->dev, "%s: cap offset+len overflow detected\n",
366			__func__);
367		return false;
368	}
369
370	if (offset + len > bar_len) {
371		dev_err(&pci_dev->dev, "%s: bar shorter than cap offset+len\n",
372			__func__);
373		return false;
374	}
375
376	region->len = len;
377	region->addr = (u64) phys_addr + offset;
378
379	return true;
380}
381
382static const struct virtio_config_ops virtio_pci_config_nodev_ops = {
383	.get		= NULL,
384	.set		= NULL,
385	.generation	= vp_generation,
386	.get_status	= vp_get_status,
387	.set_status	= vp_set_status,
388	.reset		= vp_reset,
389	.find_vqs	= vp_modern_find_vqs,
390	.del_vqs	= vp_del_vqs,
391	.get_features	= vp_get_features,
392	.finalize_features = vp_finalize_features,
393	.bus_name	= vp_bus_name,
394	.set_vq_affinity = vp_set_vq_affinity,
395	.get_vq_affinity = vp_get_vq_affinity,
396	.get_shm_region  = vp_get_shm_region,
397};
398
399static const struct virtio_config_ops virtio_pci_config_ops = {
400	.get		= vp_get,
401	.set		= vp_set,
402	.generation	= vp_generation,
403	.get_status	= vp_get_status,
404	.set_status	= vp_set_status,
405	.reset		= vp_reset,
406	.find_vqs	= vp_modern_find_vqs,
407	.del_vqs	= vp_del_vqs,
408	.get_features	= vp_get_features,
409	.finalize_features = vp_finalize_features,
410	.bus_name	= vp_bus_name,
411	.set_vq_affinity = vp_set_vq_affinity,
412	.get_vq_affinity = vp_get_vq_affinity,
413	.get_shm_region  = vp_get_shm_region,
414};
415
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
416/* the PCI probing function */
417int virtio_pci_modern_probe(struct virtio_pci_device *vp_dev)
418{
419	struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
420	struct pci_dev *pci_dev = vp_dev->pci_dev;
421	int err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
422
423	mdev->pci_dev = pci_dev;
 
 
 
 
 
424
425	err = vp_modern_probe(mdev);
 
426	if (err)
427		return err;
428
429	if (mdev->device)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
430		vp_dev->vdev.config = &virtio_pci_config_ops;
431	else
432		vp_dev->vdev.config = &virtio_pci_config_nodev_ops;
 
433
434	vp_dev->config_vector = vp_config_vector;
435	vp_dev->setup_vq = setup_vq;
436	vp_dev->del_vq = del_vq;
437	vp_dev->isr = mdev->isr;
438	vp_dev->vdev.id = mdev->id;
439
440	return 0;
 
 
 
 
 
 
 
 
 
 
441}
442
443void virtio_pci_modern_remove(struct virtio_pci_device *vp_dev)
444{
445	struct virtio_pci_modern_device *mdev = &vp_dev->mdev;
446
447	vp_modern_remove(mdev);
 
 
 
 
 
 
448}